1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 2f46b5a66SChristoph Hellwig /* 3f46b5a66SChristoph Hellwig * Copyright (C) 2007 Oracle. All rights reserved. 4f46b5a66SChristoph Hellwig */ 5f46b5a66SChristoph Hellwig 6f46b5a66SChristoph Hellwig #include <linux/kernel.h> 7f46b5a66SChristoph Hellwig #include <linux/bio.h> 8f46b5a66SChristoph Hellwig #include <linux/file.h> 9f46b5a66SChristoph Hellwig #include <linux/fs.h> 10cb8e7090SChristoph Hellwig #include <linux/fsnotify.h> 11f46b5a66SChristoph Hellwig #include <linux/pagemap.h> 12f46b5a66SChristoph Hellwig #include <linux/highmem.h> 13f46b5a66SChristoph Hellwig #include <linux/time.h> 14f46b5a66SChristoph Hellwig #include <linux/string.h> 15f46b5a66SChristoph Hellwig #include <linux/backing-dev.h> 16cb8e7090SChristoph Hellwig #include <linux/mount.h> 17cb8e7090SChristoph Hellwig #include <linux/namei.h> 18f46b5a66SChristoph Hellwig #include <linux/writeback.h> 19f46b5a66SChristoph Hellwig #include <linux/compat.h> 20cb8e7090SChristoph Hellwig #include <linux/security.h> 21f46b5a66SChristoph Hellwig #include <linux/xattr.h> 22f54de068SDavid Sterba #include <linux/mm.h> 235a0e3ad6STejun Heo #include <linux/slab.h> 24f7039b1dSLi Dongyang #include <linux/blkdev.h> 258ea05e3aSAlexander Block #include <linux/uuid.h> 2655e301fdSFilipe Brandenburger #include <linux/btrfs.h> 27416161dbSMark Fasheh #include <linux/uaccess.h> 28ae5e165dSJeff Layton #include <linux/iversion.h> 2997fc2977SMiklos Szeredi #include <linux/fileattr.h> 3014605409SBoris Burkov #include <linux/fsverity.h> 311881fba8SOmar Sandoval #include <linux/sched/xacct.h> 32f46b5a66SChristoph Hellwig #include "ctree.h" 33f46b5a66SChristoph Hellwig #include "disk-io.h" 34949964c9SMarcos Paulo de Souza #include "export.h" 35f46b5a66SChristoph Hellwig #include "transaction.h" 36f46b5a66SChristoph Hellwig #include "btrfs_inode.h" 37f46b5a66SChristoph Hellwig #include "print-tree.h" 38f46b5a66SChristoph Hellwig #include "volumes.h" 39925baeddSChris Mason #include "locking.h" 40d7728c96SJan Schmidt #include "backref.h" 41606686eeSJosef Bacik #include "rcu-string.h" 4231db9f7cSAlexander Block #include "send.h" 433f6bcfbdSStefan Behrens #include "dev-replace.h" 4463541927SFilipe David Borba Manana #include "props.h" 453b02a68aSJeff Mahoney #include "sysfs.h" 46fcebe456SJosef Bacik #include "qgroup.h" 471ec9a1aeSFilipe Manana #include "tree-log.h" 48ebb8765bSAnand Jain #include "compression.h" 498719aaaeSJosef Bacik #include "space-info.h" 5086736342SJosef Bacik #include "delalloc-space.h" 51aac0023cSJosef Bacik #include "block-group.h" 5222b398eeSQu Wenruo #include "subpage.h" 53f46b5a66SChristoph Hellwig 54abccd00fSHugo Mills #ifdef CONFIG_64BIT 55abccd00fSHugo Mills /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI 56abccd00fSHugo Mills * structures are incorrect, as the timespec structure from userspace 57abccd00fSHugo Mills * is 4 bytes too small. We define these alternatives here to teach 58abccd00fSHugo Mills * the kernel about the 32-bit struct packing. 59abccd00fSHugo Mills */ 60abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 { 61abccd00fSHugo Mills __u64 sec; 62abccd00fSHugo Mills __u32 nsec; 63abccd00fSHugo Mills } __attribute__ ((__packed__)); 64abccd00fSHugo Mills 65abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 { 66abccd00fSHugo Mills char uuid[BTRFS_UUID_SIZE]; /* in */ 67abccd00fSHugo Mills __u64 stransid; /* in */ 68abccd00fSHugo Mills __u64 rtransid; /* out */ 69abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 stime; /* in */ 70abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 rtime; /* out */ 71abccd00fSHugo Mills __u64 flags; /* in */ 72abccd00fSHugo Mills __u64 reserved[16]; /* in */ 73abccd00fSHugo Mills } __attribute__ ((__packed__)); 74abccd00fSHugo Mills 75abccd00fSHugo Mills #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \ 76abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32) 77abccd00fSHugo Mills #endif 78abccd00fSHugo Mills 792351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 802351f431SJosef Bacik struct btrfs_ioctl_send_args_32 { 812351f431SJosef Bacik __s64 send_fd; /* in */ 822351f431SJosef Bacik __u64 clone_sources_count; /* in */ 832351f431SJosef Bacik compat_uptr_t clone_sources; /* in */ 842351f431SJosef Bacik __u64 parent_root; /* in */ 852351f431SJosef Bacik __u64 flags; /* in */ 86e77fbf99SDavid Sterba __u32 version; /* in */ 87e77fbf99SDavid Sterba __u8 reserved[28]; /* in */ 882351f431SJosef Bacik } __attribute__ ((__packed__)); 892351f431SJosef Bacik 902351f431SJosef Bacik #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \ 912351f431SJosef Bacik struct btrfs_ioctl_send_args_32) 921881fba8SOmar Sandoval 931881fba8SOmar Sandoval struct btrfs_ioctl_encoded_io_args_32 { 941881fba8SOmar Sandoval compat_uptr_t iov; 951881fba8SOmar Sandoval compat_ulong_t iovcnt; 961881fba8SOmar Sandoval __s64 offset; 971881fba8SOmar Sandoval __u64 flags; 981881fba8SOmar Sandoval __u64 len; 991881fba8SOmar Sandoval __u64 unencoded_len; 1001881fba8SOmar Sandoval __u64 unencoded_offset; 1011881fba8SOmar Sandoval __u32 compression; 1021881fba8SOmar Sandoval __u32 encryption; 1031881fba8SOmar Sandoval __u8 reserved[64]; 1041881fba8SOmar Sandoval }; 1051881fba8SOmar Sandoval 1061881fba8SOmar Sandoval #define BTRFS_IOC_ENCODED_READ_32 _IOR(BTRFS_IOCTL_MAGIC, 64, \ 1071881fba8SOmar Sandoval struct btrfs_ioctl_encoded_io_args_32) 1087c0c7269SOmar Sandoval #define BTRFS_IOC_ENCODED_WRITE_32 _IOW(BTRFS_IOCTL_MAGIC, 64, \ 1097c0c7269SOmar Sandoval struct btrfs_ioctl_encoded_io_args_32) 1102351f431SJosef Bacik #endif 111abccd00fSHugo Mills 1126cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 1131905a0f7SDavid Sterba static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode, 1141905a0f7SDavid Sterba unsigned int flags) 1156cbff00fSChristoph Hellwig { 1161905a0f7SDavid Sterba if (S_ISDIR(inode->i_mode)) 1176cbff00fSChristoph Hellwig return flags; 1181905a0f7SDavid Sterba else if (S_ISREG(inode->i_mode)) 1196cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 1206cbff00fSChristoph Hellwig else 1216cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 1226cbff00fSChristoph Hellwig } 123f46b5a66SChristoph Hellwig 1246cbff00fSChristoph Hellwig /* 125a157d4fdSDavid Sterba * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS 126a157d4fdSDavid Sterba * ioctl. 1276cbff00fSChristoph Hellwig */ 12877eea05eSBoris Burkov static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode) 1296cbff00fSChristoph Hellwig { 1306cbff00fSChristoph Hellwig unsigned int iflags = 0; 13177eea05eSBoris Burkov u32 flags = binode->flags; 13214605409SBoris Burkov u32 ro_flags = binode->ro_flags; 1336cbff00fSChristoph Hellwig 1346cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 1356cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 1366cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 1376cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 1386cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 1396cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 1406cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 1416cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 1426cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 1436cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 1446cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 1456cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 146d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 147d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 14814605409SBoris Burkov if (ro_flags & BTRFS_INODE_RO_VERITY) 14914605409SBoris Burkov iflags |= FS_VERITY_FL; 150d0092bddSLi Zefan 15113f48dc9SSatoru Takeuchi if (flags & BTRFS_INODE_NOCOMPRESS) 152d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 15313f48dc9SSatoru Takeuchi else if (flags & BTRFS_INODE_COMPRESS) 15413f48dc9SSatoru Takeuchi iflags |= FS_COMPR_FL; 1556cbff00fSChristoph Hellwig 1566cbff00fSChristoph Hellwig return iflags; 1576cbff00fSChristoph Hellwig } 1586cbff00fSChristoph Hellwig 1596cbff00fSChristoph Hellwig /* 1606cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 1616cbff00fSChristoph Hellwig */ 1627b6a221eSDavid Sterba void btrfs_sync_inode_flags_to_i_flags(struct inode *inode) 1636cbff00fSChristoph Hellwig { 1645c57b8b6SDavid Sterba struct btrfs_inode *binode = BTRFS_I(inode); 1653cc79392SFilipe Manana unsigned int new_fl = 0; 1666cbff00fSChristoph Hellwig 1675c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_SYNC) 1683cc79392SFilipe Manana new_fl |= S_SYNC; 1695c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_IMMUTABLE) 1703cc79392SFilipe Manana new_fl |= S_IMMUTABLE; 1715c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_APPEND) 1723cc79392SFilipe Manana new_fl |= S_APPEND; 1735c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_NOATIME) 1743cc79392SFilipe Manana new_fl |= S_NOATIME; 1755c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_DIRSYNC) 1763cc79392SFilipe Manana new_fl |= S_DIRSYNC; 17714605409SBoris Burkov if (binode->ro_flags & BTRFS_INODE_RO_VERITY) 17814605409SBoris Burkov new_fl |= S_VERITY; 1793cc79392SFilipe Manana 1803cc79392SFilipe Manana set_mask_bits(&inode->i_flags, 18114605409SBoris Burkov S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC | 18214605409SBoris Burkov S_VERITY, new_fl); 1836cbff00fSChristoph Hellwig } 1846cbff00fSChristoph Hellwig 185f37c563bSDavid Sterba /* 186f37c563bSDavid Sterba * Check if @flags are a supported and valid set of FS_*_FL flags and that 187f37c563bSDavid Sterba * the old and new flags are not conflicting 188f37c563bSDavid Sterba */ 189f37c563bSDavid Sterba static int check_fsflags(unsigned int old_flags, unsigned int flags) 19075e7cb7fSLiu Bo { 19175e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 19275e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 19375e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 194e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 195e1e8fb6aSLi Zefan FS_NOCOW_FL)) 19675e7cb7fSLiu Bo return -EOPNOTSUPP; 19775e7cb7fSLiu Bo 198f37c563bSDavid Sterba /* COMPR and NOCOMP on new/old are valid */ 19975e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 20075e7cb7fSLiu Bo return -EINVAL; 20175e7cb7fSLiu Bo 202f37c563bSDavid Sterba if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL)) 203f37c563bSDavid Sterba return -EINVAL; 204f37c563bSDavid Sterba 205f37c563bSDavid Sterba /* NOCOW and compression options are mutually exclusive */ 206f37c563bSDavid Sterba if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL))) 207f37c563bSDavid Sterba return -EINVAL; 208f37c563bSDavid Sterba if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL))) 209f37c563bSDavid Sterba return -EINVAL; 210f37c563bSDavid Sterba 21175e7cb7fSLiu Bo return 0; 21275e7cb7fSLiu Bo } 21375e7cb7fSLiu Bo 214d206e9c9SNaohiro Aota static int check_fsflags_compatible(struct btrfs_fs_info *fs_info, 215d206e9c9SNaohiro Aota unsigned int flags) 216d206e9c9SNaohiro Aota { 217d206e9c9SNaohiro Aota if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL)) 218d206e9c9SNaohiro Aota return -EPERM; 219d206e9c9SNaohiro Aota 220d206e9c9SNaohiro Aota return 0; 221d206e9c9SNaohiro Aota } 222d206e9c9SNaohiro Aota 22397fc2977SMiklos Szeredi /* 22497fc2977SMiklos Szeredi * Set flags/xflags from the internal inode flags. The remaining items of 22597fc2977SMiklos Szeredi * fsxattr are zeroed. 22697fc2977SMiklos Szeredi */ 22797fc2977SMiklos Szeredi int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa) 2286cbff00fSChristoph Hellwig { 22997fc2977SMiklos Szeredi struct btrfs_inode *binode = BTRFS_I(d_inode(dentry)); 23097fc2977SMiklos Szeredi 23177eea05eSBoris Burkov fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode)); 23297fc2977SMiklos Szeredi return 0; 23397fc2977SMiklos Szeredi } 23497fc2977SMiklos Szeredi 23597fc2977SMiklos Szeredi int btrfs_fileattr_set(struct user_namespace *mnt_userns, 23697fc2977SMiklos Szeredi struct dentry *dentry, struct fileattr *fa) 23797fc2977SMiklos Szeredi { 23897fc2977SMiklos Szeredi struct inode *inode = d_inode(dentry); 2390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2405c57b8b6SDavid Sterba struct btrfs_inode *binode = BTRFS_I(inode); 2415c57b8b6SDavid Sterba struct btrfs_root *root = binode->root; 2426cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 2435aca2842SDarrick J. Wong unsigned int fsflags, old_fsflags; 2446cbff00fSChristoph Hellwig int ret; 245ff9fef55SAnand Jain const char *comp = NULL; 246f37c563bSDavid Sterba u32 binode_flags; 2476cbff00fSChristoph Hellwig 248b83cc969SLi Zefan if (btrfs_root_readonly(root)) 249b83cc969SLi Zefan return -EROFS; 250b83cc969SLi Zefan 25197fc2977SMiklos Szeredi if (fileattr_has_fsx(fa)) 25297fc2977SMiklos Szeredi return -EOPNOTSUPP; 2536cbff00fSChristoph Hellwig 25497fc2977SMiklos Szeredi fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags); 25577eea05eSBoris Burkov old_fsflags = btrfs_inode_flags_to_fsflags(binode); 25697fc2977SMiklos Szeredi ret = check_fsflags(old_fsflags, fsflags); 257e7848683SJan Kara if (ret) 258e7848683SJan Kara return ret; 259e7848683SJan Kara 260d206e9c9SNaohiro Aota ret = check_fsflags_compatible(fs_info, fsflags); 261d206e9c9SNaohiro Aota if (ret) 26297fc2977SMiklos Szeredi return ret; 263d206e9c9SNaohiro Aota 264f37c563bSDavid Sterba binode_flags = binode->flags; 2655c57b8b6SDavid Sterba if (fsflags & FS_SYNC_FL) 266d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_SYNC; 2676cbff00fSChristoph Hellwig else 268d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_SYNC; 2695c57b8b6SDavid Sterba if (fsflags & FS_IMMUTABLE_FL) 270d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_IMMUTABLE; 2716cbff00fSChristoph Hellwig else 272d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_IMMUTABLE; 2735c57b8b6SDavid Sterba if (fsflags & FS_APPEND_FL) 274d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_APPEND; 2756cbff00fSChristoph Hellwig else 276d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_APPEND; 2775c57b8b6SDavid Sterba if (fsflags & FS_NODUMP_FL) 278d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODUMP; 2796cbff00fSChristoph Hellwig else 280d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NODUMP; 2815c57b8b6SDavid Sterba if (fsflags & FS_NOATIME_FL) 282d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NOATIME; 2836cbff00fSChristoph Hellwig else 284d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NOATIME; 28597fc2977SMiklos Szeredi 28697fc2977SMiklos Szeredi /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */ 28797fc2977SMiklos Szeredi if (!fa->flags_valid) { 28897fc2977SMiklos Szeredi /* 1 item for the inode */ 28997fc2977SMiklos Szeredi trans = btrfs_start_transaction(root, 1); 2909b8a233bSRitesh Harjani if (IS_ERR(trans)) 2919b8a233bSRitesh Harjani return PTR_ERR(trans); 29297fc2977SMiklos Szeredi goto update_flags; 29397fc2977SMiklos Szeredi } 29497fc2977SMiklos Szeredi 2955c57b8b6SDavid Sterba if (fsflags & FS_DIRSYNC_FL) 296d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_DIRSYNC; 2976cbff00fSChristoph Hellwig else 298d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_DIRSYNC; 2995c57b8b6SDavid Sterba if (fsflags & FS_NOCOW_FL) { 30044e5194bSAnand Jain if (S_ISREG(inode->i_mode)) { 3017e97b8daSDavid Sterba /* 3027e97b8daSDavid Sterba * It's safe to turn csums off here, no extents exist. 3037e97b8daSDavid Sterba * Otherwise we want the flag to reflect the real COW 3047e97b8daSDavid Sterba * status of the file and will not set it. 3057e97b8daSDavid Sterba */ 3067e97b8daSDavid Sterba if (inode->i_size == 0) 307d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODATACOW | 308d2b8fcfeSAnand Jain BTRFS_INODE_NODATASUM; 3097e97b8daSDavid Sterba } else { 310d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODATACOW; 3117e97b8daSDavid Sterba } 3127e97b8daSDavid Sterba } else { 3137e97b8daSDavid Sterba /* 31401327610SNicholas D Steeves * Revert back under same assumptions as above 3157e97b8daSDavid Sterba */ 31644e5194bSAnand Jain if (S_ISREG(inode->i_mode)) { 3177e97b8daSDavid Sterba if (inode->i_size == 0) 318d2b8fcfeSAnand Jain binode_flags &= ~(BTRFS_INODE_NODATACOW | 319d2b8fcfeSAnand Jain BTRFS_INODE_NODATASUM); 3207e97b8daSDavid Sterba } else { 321d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NODATACOW; 3227e97b8daSDavid Sterba } 3237e97b8daSDavid Sterba } 3246cbff00fSChristoph Hellwig 32575e7cb7fSLiu Bo /* 32675e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 32775e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 32875e7cb7fSLiu Bo * things smaller. 32975e7cb7fSLiu Bo */ 3305c57b8b6SDavid Sterba if (fsflags & FS_NOCOMP_FL) { 331d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_COMPRESS; 332d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NOCOMPRESS; 3335c57b8b6SDavid Sterba } else if (fsflags & FS_COMPR_FL) { 33463541927SFilipe David Borba Manana 33597fc2977SMiklos Szeredi if (IS_SWAPFILE(inode)) 33697fc2977SMiklos Szeredi return -ETXTBSY; 337eede2bf3SOmar Sandoval 338d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_COMPRESS; 339d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NOCOMPRESS; 34063541927SFilipe David Borba Manana 34193370509SDavid Sterba comp = btrfs_compress_type2str(fs_info->compress_type); 34293370509SDavid Sterba if (!comp || comp[0] == 0) 34393370509SDavid Sterba comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB); 344ebcb904dSLi Zefan } else { 345d2b8fcfeSAnand Jain binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 34675e7cb7fSLiu Bo } 3476cbff00fSChristoph Hellwig 348ff9fef55SAnand Jain /* 349ff9fef55SAnand Jain * 1 for inode item 350ff9fef55SAnand Jain * 2 for properties 351ff9fef55SAnand Jain */ 352ff9fef55SAnand Jain trans = btrfs_start_transaction(root, 3); 35397fc2977SMiklos Szeredi if (IS_ERR(trans)) 35497fc2977SMiklos Szeredi return PTR_ERR(trans); 3556cbff00fSChristoph Hellwig 356ff9fef55SAnand Jain if (comp) { 357ff9fef55SAnand Jain ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp, 358ff9fef55SAnand Jain strlen(comp), 0); 359ff9fef55SAnand Jain if (ret) { 360ff9fef55SAnand Jain btrfs_abort_transaction(trans, ret); 361ff9fef55SAnand Jain goto out_end_trans; 362ff9fef55SAnand Jain } 363ff9fef55SAnand Jain } else { 364ff9fef55SAnand Jain ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL, 365ff9fef55SAnand Jain 0, 0); 366ff9fef55SAnand Jain if (ret && ret != -ENODATA) { 367ff9fef55SAnand Jain btrfs_abort_transaction(trans, ret); 368ff9fef55SAnand Jain goto out_end_trans; 369ff9fef55SAnand Jain } 370ff9fef55SAnand Jain } 371ff9fef55SAnand Jain 37297fc2977SMiklos Szeredi update_flags: 373d2b8fcfeSAnand Jain binode->flags = binode_flags; 3747b6a221eSDavid Sterba btrfs_sync_inode_flags_to_i_flags(inode); 3750c4d2d95SJosef Bacik inode_inc_iversion(inode); 376c2050a45SDeepa Dinamani inode->i_ctime = current_time(inode); 3779a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 3786cbff00fSChristoph Hellwig 379ff9fef55SAnand Jain out_end_trans: 3803a45bb20SJeff Mahoney btrfs_end_transaction(trans); 3812d4e6f6aSliubo return ret; 3826cbff00fSChristoph Hellwig } 3836cbff00fSChristoph Hellwig 3840d7ed32cSDavid Sterba /* 3850d7ed32cSDavid Sterba * Start exclusive operation @type, return true on success 3860d7ed32cSDavid Sterba */ 387c3e1f96cSGoldwyn Rodrigues bool btrfs_exclop_start(struct btrfs_fs_info *fs_info, 388c3e1f96cSGoldwyn Rodrigues enum btrfs_exclusive_operation type) 389c3e1f96cSGoldwyn Rodrigues { 3900d7ed32cSDavid Sterba bool ret = false; 3910d7ed32cSDavid Sterba 3920d7ed32cSDavid Sterba spin_lock(&fs_info->super_lock); 3930d7ed32cSDavid Sterba if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) { 3940d7ed32cSDavid Sterba fs_info->exclusive_operation = type; 3950d7ed32cSDavid Sterba ret = true; 3960d7ed32cSDavid Sterba } 3970d7ed32cSDavid Sterba spin_unlock(&fs_info->super_lock); 3980d7ed32cSDavid Sterba 3990d7ed32cSDavid Sterba return ret; 400c3e1f96cSGoldwyn Rodrigues } 401c3e1f96cSGoldwyn Rodrigues 402578bda9eSDavid Sterba /* 403578bda9eSDavid Sterba * Conditionally allow to enter the exclusive operation in case it's compatible 404578bda9eSDavid Sterba * with the running one. This must be paired with btrfs_exclop_start_unlock and 405578bda9eSDavid Sterba * btrfs_exclop_finish. 406578bda9eSDavid Sterba * 407578bda9eSDavid Sterba * Compatibility: 408578bda9eSDavid Sterba * - the same type is already running 409621a1ee1SNikolay Borisov * - when trying to add a device and balance has been paused 410578bda9eSDavid Sterba * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller 411578bda9eSDavid Sterba * must check the condition first that would allow none -> @type 412578bda9eSDavid Sterba */ 413578bda9eSDavid Sterba bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info, 414578bda9eSDavid Sterba enum btrfs_exclusive_operation type) 415578bda9eSDavid Sterba { 416578bda9eSDavid Sterba spin_lock(&fs_info->super_lock); 417621a1ee1SNikolay Borisov if (fs_info->exclusive_operation == type || 418621a1ee1SNikolay Borisov (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED && 419621a1ee1SNikolay Borisov type == BTRFS_EXCLOP_DEV_ADD)) 420578bda9eSDavid Sterba return true; 421578bda9eSDavid Sterba 422578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 423578bda9eSDavid Sterba return false; 424578bda9eSDavid Sterba } 425578bda9eSDavid Sterba 426578bda9eSDavid Sterba void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info) 427578bda9eSDavid Sterba { 428578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 429578bda9eSDavid Sterba } 430578bda9eSDavid Sterba 431c3e1f96cSGoldwyn Rodrigues void btrfs_exclop_finish(struct btrfs_fs_info *fs_info) 432c3e1f96cSGoldwyn Rodrigues { 4330d7ed32cSDavid Sterba spin_lock(&fs_info->super_lock); 434c3e1f96cSGoldwyn Rodrigues WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE); 4350d7ed32cSDavid Sterba spin_unlock(&fs_info->super_lock); 43666a2823cSGoldwyn Rodrigues sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation"); 437c3e1f96cSGoldwyn Rodrigues } 438c3e1f96cSGoldwyn Rodrigues 439efc0e69cSNikolay Borisov void btrfs_exclop_balance(struct btrfs_fs_info *fs_info, 440efc0e69cSNikolay Borisov enum btrfs_exclusive_operation op) 441efc0e69cSNikolay Borisov { 442efc0e69cSNikolay Borisov switch (op) { 443efc0e69cSNikolay Borisov case BTRFS_EXCLOP_BALANCE_PAUSED: 444efc0e69cSNikolay Borisov spin_lock(&fs_info->super_lock); 445efc0e69cSNikolay Borisov ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE || 446efc0e69cSNikolay Borisov fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD); 447efc0e69cSNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED; 448efc0e69cSNikolay Borisov spin_unlock(&fs_info->super_lock); 449efc0e69cSNikolay Borisov break; 450efc0e69cSNikolay Borisov case BTRFS_EXCLOP_BALANCE: 451efc0e69cSNikolay Borisov spin_lock(&fs_info->super_lock); 452efc0e69cSNikolay Borisov ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 453efc0e69cSNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 454efc0e69cSNikolay Borisov spin_unlock(&fs_info->super_lock); 455efc0e69cSNikolay Borisov break; 456efc0e69cSNikolay Borisov default: 457efc0e69cSNikolay Borisov btrfs_warn(fs_info, 458efc0e69cSNikolay Borisov "invalid exclop balance operation %d requested", op); 459efc0e69cSNikolay Borisov } 460efc0e69cSNikolay Borisov } 461efc0e69cSNikolay Borisov 462dc408ccdSSahil Kang static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg) 4636cbff00fSChristoph Hellwig { 4646cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 4656cbff00fSChristoph Hellwig } 466f46b5a66SChristoph Hellwig 467b929c1d8SMarcos Paulo de Souza static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info, 468b929c1d8SMarcos Paulo de Souza void __user *arg) 469f7039b1dSLi Dongyang { 470f7039b1dSLi Dongyang struct btrfs_device *device; 471f7039b1dSLi Dongyang struct request_queue *q; 472f7039b1dSLi Dongyang struct fstrim_range range; 473f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 474f7039b1dSLi Dongyang u64 num_devices = 0; 475f7039b1dSLi Dongyang int ret; 476f7039b1dSLi Dongyang 477f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 478f7039b1dSLi Dongyang return -EPERM; 479f7039b1dSLi Dongyang 480f35f06c3SFilipe Manana /* 4811cb3dc3fSNaohiro Aota * btrfs_trim_block_group() depends on space cache, which is not 4821cb3dc3fSNaohiro Aota * available in zoned filesystem. So, disallow fitrim on a zoned 4831cb3dc3fSNaohiro Aota * filesystem for now. 4841cb3dc3fSNaohiro Aota */ 4851cb3dc3fSNaohiro Aota if (btrfs_is_zoned(fs_info)) 4861cb3dc3fSNaohiro Aota return -EOPNOTSUPP; 4871cb3dc3fSNaohiro Aota 4881cb3dc3fSNaohiro Aota /* 489f35f06c3SFilipe Manana * If the fs is mounted with nologreplay, which requires it to be 490f35f06c3SFilipe Manana * mounted in RO mode as well, we can not allow discard on free space 491f35f06c3SFilipe Manana * inside block groups, because log trees refer to extents that are not 492f35f06c3SFilipe Manana * pinned in a block group's free space cache (pinning the extents is 493f35f06c3SFilipe Manana * precisely the first phase of replaying a log tree). 494f35f06c3SFilipe Manana */ 495f35f06c3SFilipe Manana if (btrfs_test_opt(fs_info, NOLOGREPLAY)) 496f35f06c3SFilipe Manana return -EROFS; 497f35f06c3SFilipe Manana 4981f78160cSXiao Guangrong rcu_read_lock(); 4991f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 5001f78160cSXiao Guangrong dev_list) { 501f7039b1dSLi Dongyang if (!device->bdev) 502f7039b1dSLi Dongyang continue; 503f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 504f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 505f7039b1dSLi Dongyang num_devices++; 50650d0446eSSeraphime Kirkovski minlen = min_t(u64, q->limits.discard_granularity, 507f7039b1dSLi Dongyang minlen); 508f7039b1dSLi Dongyang } 509f7039b1dSLi Dongyang } 5101f78160cSXiao Guangrong rcu_read_unlock(); 511f4c697e6SLukas Czerner 512f7039b1dSLi Dongyang if (!num_devices) 513f7039b1dSLi Dongyang return -EOPNOTSUPP; 514f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 515f7039b1dSLi Dongyang return -EFAULT; 5166ba9fc8eSQu Wenruo 5176ba9fc8eSQu Wenruo /* 5186ba9fc8eSQu Wenruo * NOTE: Don't truncate the range using super->total_bytes. Bytenr of 5196ba9fc8eSQu Wenruo * block group is in the logical address space, which can be any 5206ba9fc8eSQu Wenruo * sectorsize aligned bytenr in the range [0, U64_MAX]. 5216ba9fc8eSQu Wenruo */ 5226ba9fc8eSQu Wenruo if (range.len < fs_info->sb->s_blocksize) 523f4c697e6SLukas Czerner return -EINVAL; 524f7039b1dSLi Dongyang 525f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 5262ff7e61eSJeff Mahoney ret = btrfs_trim_fs(fs_info, &range); 527f7039b1dSLi Dongyang if (ret < 0) 528f7039b1dSLi Dongyang return ret; 529f7039b1dSLi Dongyang 530f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 531f7039b1dSLi Dongyang return -EFAULT; 532f7039b1dSLi Dongyang 533f7039b1dSLi Dongyang return 0; 534f7039b1dSLi Dongyang } 535f7039b1dSLi Dongyang 536e1f60a65SDavid Sterba int __pure btrfs_is_empty_uuid(u8 *uuid) 537dd5f9615SStefan Behrens { 53846e0f66aSChris Mason int i; 53946e0f66aSChris Mason 54046e0f66aSChris Mason for (i = 0; i < BTRFS_UUID_SIZE; i++) { 54146e0f66aSChris Mason if (uuid[i]) 54246e0f66aSChris Mason return 0; 54346e0f66aSChris Mason } 54446e0f66aSChris Mason return 1; 545dd5f9615SStefan Behrens } 546dd5f9615SStefan Behrens 5473538d68dSOmar Sandoval /* 5483538d68dSOmar Sandoval * Calculate the number of transaction items to reserve for creating a subvolume 5493538d68dSOmar Sandoval * or snapshot, not including the inode, directory entries, or parent directory. 5503538d68dSOmar Sandoval */ 5513538d68dSOmar Sandoval static unsigned int create_subvol_num_items(struct btrfs_qgroup_inherit *inherit) 5523538d68dSOmar Sandoval { 5533538d68dSOmar Sandoval /* 5543538d68dSOmar Sandoval * 1 to add root block 5553538d68dSOmar Sandoval * 1 to add root item 5563538d68dSOmar Sandoval * 1 to add root ref 5573538d68dSOmar Sandoval * 1 to add root backref 5583538d68dSOmar Sandoval * 1 to add UUID item 5593538d68dSOmar Sandoval * 1 to add qgroup info 5603538d68dSOmar Sandoval * 1 to add qgroup limit 5613538d68dSOmar Sandoval * 5623538d68dSOmar Sandoval * Ideally the last two would only be accounted if qgroups are enabled, 5633538d68dSOmar Sandoval * but that can change between now and the time we would insert them. 5643538d68dSOmar Sandoval */ 5653538d68dSOmar Sandoval unsigned int num_items = 7; 5663538d68dSOmar Sandoval 5673538d68dSOmar Sandoval if (inherit) { 5683538d68dSOmar Sandoval /* 2 to add qgroup relations for each inherited qgroup */ 5693538d68dSOmar Sandoval num_items += 2 * inherit->num_qgroups; 5703538d68dSOmar Sandoval } 5713538d68dSOmar Sandoval return num_items; 5723538d68dSOmar Sandoval } 5733538d68dSOmar Sandoval 5744d4340c9SChristian Brauner static noinline int create_subvol(struct user_namespace *mnt_userns, 5754d4340c9SChristian Brauner struct inode *dir, struct dentry *dentry, 5768696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 577f46b5a66SChristoph Hellwig { 5780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 579f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 580f46b5a66SChristoph Hellwig struct btrfs_key key; 58149a3c4d9SDavid Sterba struct btrfs_root_item *root_item; 582f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 583f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 584d5c12070SMiao Xie struct btrfs_root *root = BTRFS_I(dir)->root; 58576dda93cSYan, Zheng struct btrfs_root *new_root; 586d5c12070SMiao Xie struct btrfs_block_rsv block_rsv; 58795582b00SDeepa Dinamani struct timespec64 cur_time = current_time(dir); 5883538d68dSOmar Sandoval struct btrfs_new_inode_args new_inode_args = { 5893538d68dSOmar Sandoval .dir = dir, 5903538d68dSOmar Sandoval .dentry = dentry, 5913538d68dSOmar Sandoval .subvol = true, 5923538d68dSOmar Sandoval }; 5933538d68dSOmar Sandoval unsigned int trans_num_items; 594f46b5a66SChristoph Hellwig int ret; 5952256e901SOmar Sandoval dev_t anon_dev; 596f46b5a66SChristoph Hellwig u64 objectid; 597f46b5a66SChristoph Hellwig 59849a3c4d9SDavid Sterba root_item = kzalloc(sizeof(*root_item), GFP_KERNEL); 59949a3c4d9SDavid Sterba if (!root_item) 60049a3c4d9SDavid Sterba return -ENOMEM; 60149a3c4d9SDavid Sterba 602543068a2SNikolay Borisov ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid); 6032fbe8c8aSAl Viro if (ret) 6042256e901SOmar Sandoval goto out_root_item; 6052dfb1e43SQu Wenruo 606e09fe2d2SQu Wenruo /* 607e09fe2d2SQu Wenruo * Don't create subvolume whose level is not zero. Or qgroup will be 60801327610SNicholas D Steeves * screwed up since it assumes subvolume qgroup's level to be 0. 609e09fe2d2SQu Wenruo */ 61049a3c4d9SDavid Sterba if (btrfs_qgroup_level(objectid)) { 61149a3c4d9SDavid Sterba ret = -ENOSPC; 6122256e901SOmar Sandoval goto out_root_item; 61349a3c4d9SDavid Sterba } 614e09fe2d2SQu Wenruo 6152256e901SOmar Sandoval ret = get_anon_bdev(&anon_dev); 6162256e901SOmar Sandoval if (ret < 0) 6172256e901SOmar Sandoval goto out_root_item; 6182256e901SOmar Sandoval 6193538d68dSOmar Sandoval new_inode_args.inode = btrfs_new_subvol_inode(mnt_userns, dir); 6203538d68dSOmar Sandoval if (!new_inode_args.inode) { 621a1fd0c35SOmar Sandoval ret = -ENOMEM; 622a1fd0c35SOmar Sandoval goto out_anon_dev; 623a1fd0c35SOmar Sandoval } 6243538d68dSOmar Sandoval ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items); 625d5c12070SMiao Xie if (ret) 626a1fd0c35SOmar Sandoval goto out_inode; 6273538d68dSOmar Sandoval trans_num_items += create_subvol_num_items(inherit); 6283538d68dSOmar Sandoval 6293538d68dSOmar Sandoval btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 6303538d68dSOmar Sandoval ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 6313538d68dSOmar Sandoval trans_num_items, false); 6323538d68dSOmar Sandoval if (ret) 6333538d68dSOmar Sandoval goto out_new_inode_args; 634f46b5a66SChristoph Hellwig 635d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 636d5c12070SMiao Xie if (IS_ERR(trans)) { 637d5c12070SMiao Xie ret = PTR_ERR(trans); 638e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 6393538d68dSOmar Sandoval goto out_new_inode_args; 640d5c12070SMiao Xie } 641d5c12070SMiao Xie trans->block_rsv = &block_rsv; 642d5c12070SMiao Xie trans->bytes_reserved = block_rsv.size; 643f46b5a66SChristoph Hellwig 644a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit); 6456f72c7e2SArne Jansen if (ret) 6462256e901SOmar Sandoval goto out; 6476f72c7e2SArne Jansen 6489631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 6499631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 6508e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 6518e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 6522256e901SOmar Sandoval goto out; 6538e8a1e31SJosef Bacik } 654f46b5a66SChristoph Hellwig 655f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 656f46b5a66SChristoph Hellwig 65749a3c4d9SDavid Sterba inode_item = &root_item->inode; 6583cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 6593cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 6603cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 661da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 6620b246afaSJeff Mahoney fs_info->nodesize); 6633cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 664f46b5a66SChristoph Hellwig 66549a3c4d9SDavid Sterba btrfs_set_root_flags(root_item, 0); 66649a3c4d9SDavid Sterba btrfs_set_root_limit(root_item, 0); 6673cae210fSQu Wenruo btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 66808fe4db1SLi Zefan 66949a3c4d9SDavid Sterba btrfs_set_root_bytenr(root_item, leaf->start); 67049a3c4d9SDavid Sterba btrfs_set_root_generation(root_item, trans->transid); 67149a3c4d9SDavid Sterba btrfs_set_root_level(root_item, 0); 67249a3c4d9SDavid Sterba btrfs_set_root_refs(root_item, 1); 67349a3c4d9SDavid Sterba btrfs_set_root_used(root_item, leaf->len); 67449a3c4d9SDavid Sterba btrfs_set_root_last_snapshot(root_item, 0); 675f46b5a66SChristoph Hellwig 67649a3c4d9SDavid Sterba btrfs_set_root_generation_v2(root_item, 67749a3c4d9SDavid Sterba btrfs_root_generation(root_item)); 678807fc790SAndy Shevchenko generate_random_guid(root_item->uuid); 67949a3c4d9SDavid Sterba btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec); 68049a3c4d9SDavid Sterba btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec); 68149a3c4d9SDavid Sterba root_item->ctime = root_item->otime; 68249a3c4d9SDavid Sterba btrfs_set_root_ctransid(root_item, trans->transid); 68349a3c4d9SDavid Sterba btrfs_set_root_otransid(root_item, trans->transid); 684f46b5a66SChristoph Hellwig 685925baeddSChris Mason btrfs_tree_unlock(leaf); 686f46b5a66SChristoph Hellwig 68769948022SNikolay Borisov btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID); 688f46b5a66SChristoph Hellwig 689f46b5a66SChristoph Hellwig key.objectid = objectid; 6905d4f98a2SYan Zheng key.offset = 0; 691962a298fSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 6920b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, &key, 69349a3c4d9SDavid Sterba root_item); 69467addf29SFilipe Manana if (ret) { 69567addf29SFilipe Manana /* 69667addf29SFilipe Manana * Since we don't abort the transaction in this case, free the 69767addf29SFilipe Manana * tree block so that we don't leak space and leave the 69867addf29SFilipe Manana * filesystem in an inconsistent state (an extent item in the 6997a163608SFilipe Manana * extent tree with a backreference for a root that does not 700212a58fdSFilipe Manana * exists). 70167addf29SFilipe Manana */ 702212a58fdSFilipe Manana btrfs_tree_lock(leaf); 703212a58fdSFilipe Manana btrfs_clean_tree_block(leaf); 704212a58fdSFilipe Manana btrfs_tree_unlock(leaf); 7057a163608SFilipe Manana btrfs_free_tree_block(trans, objectid, leaf, 0, 1); 70667addf29SFilipe Manana free_extent_buffer(leaf); 7072256e901SOmar Sandoval goto out; 70867addf29SFilipe Manana } 70967addf29SFilipe Manana 71067addf29SFilipe Manana free_extent_buffer(leaf); 71167addf29SFilipe Manana leaf = NULL; 712f46b5a66SChristoph Hellwig 7132dfb1e43SQu Wenruo new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev); 71479787eaaSJeff Mahoney if (IS_ERR(new_root)) { 71579787eaaSJeff Mahoney ret = PTR_ERR(new_root); 71666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 7172256e901SOmar Sandoval goto out; 71879787eaaSJeff Mahoney } 7192256e901SOmar Sandoval /* anon_dev is owned by new_root now. */ 7202dfb1e43SQu Wenruo anon_dev = 0; 7213538d68dSOmar Sandoval BTRFS_I(new_inode_args.inode)->root = new_root; 7223538d68dSOmar Sandoval /* ... and new_root is owned by new_inode_args.inode now. */ 72376dda93cSYan, Zheng 724221581e4SJosef Bacik ret = btrfs_record_root_in_trans(trans, new_root); 725221581e4SJosef Bacik if (ret) { 726221581e4SJosef Bacik btrfs_abort_transaction(trans, ret); 7272256e901SOmar Sandoval goto out; 728221581e4SJosef Bacik } 72976dda93cSYan, Zheng 730cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, root_item->uuid, 7316bccf3abSJeff Mahoney BTRFS_UUID_KEY_SUBVOL, objectid); 732caae78e0SOmar Sandoval if (ret) { 73366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 734caae78e0SOmar Sandoval goto out; 735caae78e0SOmar Sandoval } 736caae78e0SOmar Sandoval 737caae78e0SOmar Sandoval ret = btrfs_create_new_inode(trans, &new_inode_args); 738caae78e0SOmar Sandoval if (ret) { 739caae78e0SOmar Sandoval btrfs_abort_transaction(trans, ret); 740caae78e0SOmar Sandoval goto out; 741caae78e0SOmar Sandoval } 742caae78e0SOmar Sandoval 743caae78e0SOmar Sandoval d_instantiate_new(dentry, new_inode_args.inode); 744caae78e0SOmar Sandoval new_inode_args.inode = NULL; 745dd5f9615SStefan Behrens 7462256e901SOmar Sandoval out: 747d5c12070SMiao Xie trans->block_rsv = NULL; 748d5c12070SMiao Xie trans->bytes_reserved = 0; 749e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 750de6e8200SLiu Bo 7511b58ae0eSFilipe Manana if (ret) 7521b58ae0eSFilipe Manana btrfs_end_transaction(trans); 7531b58ae0eSFilipe Manana else 7541b58ae0eSFilipe Manana ret = btrfs_commit_transaction(trans); 7553538d68dSOmar Sandoval out_new_inode_args: 7563538d68dSOmar Sandoval btrfs_new_inode_args_destroy(&new_inode_args); 757a1fd0c35SOmar Sandoval out_inode: 7583538d68dSOmar Sandoval iput(new_inode_args.inode); 7592256e901SOmar Sandoval out_anon_dev: 7602dfb1e43SQu Wenruo if (anon_dev) 7612dfb1e43SQu Wenruo free_anon_bdev(anon_dev); 7622256e901SOmar Sandoval out_root_item: 76349a3c4d9SDavid Sterba kfree(root_item); 76449a3c4d9SDavid Sterba return ret; 765f46b5a66SChristoph Hellwig } 766f46b5a66SChristoph Hellwig 767e9662f70SMiao Xie static int create_snapshot(struct btrfs_root *root, struct inode *dir, 7689babda9fSNikolay Borisov struct dentry *dentry, bool readonly, 769e9662f70SMiao Xie struct btrfs_qgroup_inherit *inherit) 770f46b5a66SChristoph Hellwig { 7710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 7722e4bfab9SYan, Zheng struct inode *inode; 773f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 7743538d68dSOmar Sandoval unsigned int trans_num_items; 775f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 7762e4bfab9SYan, Zheng int ret; 777f46b5a66SChristoph Hellwig 778813febdbSJosef Bacik /* We do not support snapshotting right now. */ 779813febdbSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 780813febdbSJosef Bacik btrfs_warn(fs_info, 781813febdbSJosef Bacik "extent tree v2 doesn't support snapshotting yet"); 782813febdbSJosef Bacik return -EOPNOTSUPP; 783813febdbSJosef Bacik } 784813febdbSJosef Bacik 78592a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 786f46b5a66SChristoph Hellwig return -EINVAL; 787f46b5a66SChristoph Hellwig 788eede2bf3SOmar Sandoval if (atomic_read(&root->nr_swapfiles)) { 789eede2bf3SOmar Sandoval btrfs_warn(fs_info, 790eede2bf3SOmar Sandoval "cannot snapshot subvolume with active swapfile"); 791eede2bf3SOmar Sandoval return -ETXTBSY; 792eede2bf3SOmar Sandoval } 793eede2bf3SOmar Sandoval 79423269bf5SDavid Sterba pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL); 795a1ee7362SDavid Sterba if (!pending_snapshot) 796a1ee7362SDavid Sterba return -ENOMEM; 797a1ee7362SDavid Sterba 7982dfb1e43SQu Wenruo ret = get_anon_bdev(&pending_snapshot->anon_dev); 7992dfb1e43SQu Wenruo if (ret < 0) 8002dfb1e43SQu Wenruo goto free_pending; 801b0c0ea63SDavid Sterba pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item), 80223269bf5SDavid Sterba GFP_KERNEL); 8038546b570SDavid Sterba pending_snapshot->path = btrfs_alloc_path(); 8048546b570SDavid Sterba if (!pending_snapshot->root_item || !pending_snapshot->path) { 805b0c0ea63SDavid Sterba ret = -ENOMEM; 806b0c0ea63SDavid Sterba goto free_pending; 807b0c0ea63SDavid Sterba } 808b0c0ea63SDavid Sterba 80966d8f3ddSMiao Xie btrfs_init_block_rsv(&pending_snapshot->block_rsv, 81066d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 811d5c12070SMiao Xie /* 8123538d68dSOmar Sandoval * 1 to add dir item 8133538d68dSOmar Sandoval * 1 to add dir index 8143538d68dSOmar Sandoval * 1 to update parent inode item 815d5c12070SMiao Xie */ 8163538d68dSOmar Sandoval trans_num_items = create_subvol_num_items(inherit) + 3; 817d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 8183538d68dSOmar Sandoval &pending_snapshot->block_rsv, 8193538d68dSOmar Sandoval trans_num_items, false); 820d5c12070SMiao Xie if (ret) 821c11fbb6eSRobbie Ko goto free_pending; 822d5c12070SMiao Xie 823a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 824a22285a6SYan, Zheng pending_snapshot->root = root; 825b83cc969SLi Zefan pending_snapshot->readonly = readonly; 826e9662f70SMiao Xie pending_snapshot->dir = dir; 8278696c533SMiao Xie pending_snapshot->inherit = inherit; 828a22285a6SYan, Zheng 829d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 830a22285a6SYan, Zheng if (IS_ERR(trans)) { 831a22285a6SYan, Zheng ret = PTR_ERR(trans); 832a22285a6SYan, Zheng goto fail; 833a22285a6SYan, Zheng } 834a22285a6SYan, Zheng 83528b21c55SFilipe Manana trans->pending_snapshot = pending_snapshot; 8369babda9fSNikolay Borisov 8373a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 838aec8030aSMiao Xie if (ret) 839c37b2b62SJosef Bacik goto fail; 840a22285a6SYan, Zheng 841a22285a6SYan, Zheng ret = pending_snapshot->error; 842f46b5a66SChristoph Hellwig if (ret) 8432e4bfab9SYan, Zheng goto fail; 844f46b5a66SChristoph Hellwig 845d3797308SChris Mason ret = btrfs_orphan_cleanup(pending_snapshot->snap); 846d3797308SChris Mason if (ret) 847d3797308SChris Mason goto fail; 848d3797308SChris Mason 8492b0143b5SDavid Howells inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry); 8502e4bfab9SYan, Zheng if (IS_ERR(inode)) { 8512e4bfab9SYan, Zheng ret = PTR_ERR(inode); 8522e4bfab9SYan, Zheng goto fail; 8532e4bfab9SYan, Zheng } 8545662344bSTsutomu Itoh 8552e4bfab9SYan, Zheng d_instantiate(dentry, inode); 8562e4bfab9SYan, Zheng ret = 0; 8572dfb1e43SQu Wenruo pending_snapshot->anon_dev = 0; 8582e4bfab9SYan, Zheng fail: 8592dfb1e43SQu Wenruo /* Prevent double freeing of anon_dev */ 8602dfb1e43SQu Wenruo if (ret && pending_snapshot->snap) 8612dfb1e43SQu Wenruo pending_snapshot->snap->anon_dev = 0; 86200246528SJosef Bacik btrfs_put_root(pending_snapshot->snap); 863e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv); 864b0c0ea63SDavid Sterba free_pending: 8652dfb1e43SQu Wenruo if (pending_snapshot->anon_dev) 8662dfb1e43SQu Wenruo free_anon_bdev(pending_snapshot->anon_dev); 867b0c0ea63SDavid Sterba kfree(pending_snapshot->root_item); 8688546b570SDavid Sterba btrfs_free_path(pending_snapshot->path); 869a1ee7362SDavid Sterba kfree(pending_snapshot); 870a1ee7362SDavid Sterba 871f46b5a66SChristoph Hellwig return ret; 872f46b5a66SChristoph Hellwig } 873f46b5a66SChristoph Hellwig 8744260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 8754260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 8764260f7c7SSage Weil * whether the type of victim is right. 8774260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 8784260f7c7SSage Weil * 2. We should have write and exec permissions on dir 8794260f7c7SSage Weil * 3. We can't remove anything from append-only dir 8804260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 8814260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 8824260f7c7SSage Weil * a. be owner of dir, or 8834260f7c7SSage Weil * b. be owner of victim, or 8844260f7c7SSage Weil * c. have CAP_FOWNER capability 88501327610SNicholas D Steeves * 6. If the victim is append-only or immutable we can't do anything with 8864260f7c7SSage Weil * links pointing to it. 8874260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 8884260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 8894260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 8904260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 8914260f7c7SSage Weil * nfs_async_unlink(). 8924260f7c7SSage Weil */ 8934260f7c7SSage Weil 894c4ed533bSChristian Brauner static int btrfs_may_delete(struct user_namespace *mnt_userns, 895c4ed533bSChristian Brauner struct inode *dir, struct dentry *victim, int isdir) 8964260f7c7SSage Weil { 8974260f7c7SSage Weil int error; 8984260f7c7SSage Weil 8992b0143b5SDavid Howells if (d_really_is_negative(victim)) 9004260f7c7SSage Weil return -ENOENT; 9014260f7c7SSage Weil 9022b0143b5SDavid Howells BUG_ON(d_inode(victim->d_parent) != dir); 9034fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 9044260f7c7SSage Weil 905c4ed533bSChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 9064260f7c7SSage Weil if (error) 9074260f7c7SSage Weil return error; 9084260f7c7SSage Weil if (IS_APPEND(dir)) 9094260f7c7SSage Weil return -EPERM; 910c4ed533bSChristian Brauner if (check_sticky(mnt_userns, dir, d_inode(victim)) || 911ba73d987SChristian Brauner IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) || 912ba73d987SChristian Brauner IS_SWAPFILE(d_inode(victim))) 9134260f7c7SSage Weil return -EPERM; 9144260f7c7SSage Weil if (isdir) { 915e36cb0b8SDavid Howells if (!d_is_dir(victim)) 9164260f7c7SSage Weil return -ENOTDIR; 9174260f7c7SSage Weil if (IS_ROOT(victim)) 9184260f7c7SSage Weil return -EBUSY; 919e36cb0b8SDavid Howells } else if (d_is_dir(victim)) 9204260f7c7SSage Weil return -EISDIR; 9214260f7c7SSage Weil if (IS_DEADDIR(dir)) 9224260f7c7SSage Weil return -ENOENT; 9234260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 9244260f7c7SSage Weil return -EBUSY; 9254260f7c7SSage Weil return 0; 9264260f7c7SSage Weil } 9274260f7c7SSage Weil 928cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 9294d4340c9SChristian Brauner static inline int btrfs_may_create(struct user_namespace *mnt_userns, 9304d4340c9SChristian Brauner struct inode *dir, struct dentry *child) 931cb8e7090SChristoph Hellwig { 9322b0143b5SDavid Howells if (d_really_is_positive(child)) 933cb8e7090SChristoph Hellwig return -EEXIST; 934cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 935cb8e7090SChristoph Hellwig return -ENOENT; 9364d4340c9SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 9375474bf40SChristian Brauner return -EOVERFLOW; 9384d4340c9SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 939cb8e7090SChristoph Hellwig } 940cb8e7090SChristoph Hellwig 941cb8e7090SChristoph Hellwig /* 942cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 943cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 944cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 945cb8e7090SChristoph Hellwig */ 94692872094SAl Viro static noinline int btrfs_mksubvol(const struct path *parent, 9474d4340c9SChristian Brauner struct user_namespace *mnt_userns, 94852f75f4fSDavid Sterba const char *name, int namelen, 94972fd032eSSage Weil struct btrfs_root *snap_src, 9509babda9fSNikolay Borisov bool readonly, 9518696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 952cb8e7090SChristoph Hellwig { 9532b0143b5SDavid Howells struct inode *dir = d_inode(parent->dentry); 9540b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 955cb8e7090SChristoph Hellwig struct dentry *dentry; 956cb8e7090SChristoph Hellwig int error; 957cb8e7090SChristoph Hellwig 95800235411SAl Viro error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 95900235411SAl Viro if (error == -EINTR) 96000235411SAl Viro return error; 961cb8e7090SChristoph Hellwig 9624d4340c9SChristian Brauner dentry = lookup_one(mnt_userns, name, parent->dentry, namelen); 963cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 964cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 965cb8e7090SChristoph Hellwig goto out_unlock; 966cb8e7090SChristoph Hellwig 9674d4340c9SChristian Brauner error = btrfs_may_create(mnt_userns, dir, dentry); 968cb8e7090SChristoph Hellwig if (error) 969a874a63eSLiu Bo goto out_dput; 970cb8e7090SChristoph Hellwig 9719c52057cSChris Mason /* 9729c52057cSChris Mason * even if this name doesn't exist, we may get hash collisions. 9739c52057cSChris Mason * check for them now when we can safely fail 9749c52057cSChris Mason */ 9759c52057cSChris Mason error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 9769c52057cSChris Mason dir->i_ino, name, 9779c52057cSChris Mason namelen); 9789c52057cSChris Mason if (error) 9799c52057cSChris Mason goto out_dput; 9809c52057cSChris Mason 9810b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 98276dda93cSYan, Zheng 98376dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 98476dda93cSYan, Zheng goto out_up_read; 98576dda93cSYan, Zheng 9869babda9fSNikolay Borisov if (snap_src) 9879babda9fSNikolay Borisov error = create_snapshot(snap_src, dir, dentry, readonly, inherit); 9889babda9fSNikolay Borisov else 98970dc55f4SOmar Sandoval error = create_subvol(mnt_userns, dir, dentry, inherit); 9909babda9fSNikolay Borisov 99176dda93cSYan, Zheng if (!error) 99276dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 99376dda93cSYan, Zheng out_up_read: 9940b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 995cb8e7090SChristoph Hellwig out_dput: 996cb8e7090SChristoph Hellwig dput(dentry); 997cb8e7090SChristoph Hellwig out_unlock: 99864708539SJosef Bacik btrfs_inode_unlock(dir, 0); 999cb8e7090SChristoph Hellwig return error; 1000cb8e7090SChristoph Hellwig } 1001cb8e7090SChristoph Hellwig 1002c11fbb6eSRobbie Ko static noinline int btrfs_mksnapshot(const struct path *parent, 10034d4340c9SChristian Brauner struct user_namespace *mnt_userns, 1004c11fbb6eSRobbie Ko const char *name, int namelen, 1005c11fbb6eSRobbie Ko struct btrfs_root *root, 1006c11fbb6eSRobbie Ko bool readonly, 1007c11fbb6eSRobbie Ko struct btrfs_qgroup_inherit *inherit) 1008c11fbb6eSRobbie Ko { 1009c11fbb6eSRobbie Ko int ret; 1010c11fbb6eSRobbie Ko bool snapshot_force_cow = false; 1011c11fbb6eSRobbie Ko 1012c11fbb6eSRobbie Ko /* 1013c11fbb6eSRobbie Ko * Force new buffered writes to reserve space even when NOCOW is 1014c11fbb6eSRobbie Ko * possible. This is to avoid later writeback (running dealloc) to 1015c11fbb6eSRobbie Ko * fallback to COW mode and unexpectedly fail with ENOSPC. 1016c11fbb6eSRobbie Ko */ 1017c11fbb6eSRobbie Ko btrfs_drew_read_lock(&root->snapshot_lock); 1018c11fbb6eSRobbie Ko 1019f9baa501SFilipe Manana ret = btrfs_start_delalloc_snapshot(root, false); 1020c11fbb6eSRobbie Ko if (ret) 1021c11fbb6eSRobbie Ko goto out; 1022c11fbb6eSRobbie Ko 1023c11fbb6eSRobbie Ko /* 1024c11fbb6eSRobbie Ko * All previous writes have started writeback in NOCOW mode, so now 1025c11fbb6eSRobbie Ko * we force future writes to fallback to COW mode during snapshot 1026c11fbb6eSRobbie Ko * creation. 1027c11fbb6eSRobbie Ko */ 1028c11fbb6eSRobbie Ko atomic_inc(&root->snapshot_force_cow); 1029c11fbb6eSRobbie Ko snapshot_force_cow = true; 1030c11fbb6eSRobbie Ko 1031c11fbb6eSRobbie Ko btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1); 1032c11fbb6eSRobbie Ko 10334d4340c9SChristian Brauner ret = btrfs_mksubvol(parent, mnt_userns, name, namelen, 1034c11fbb6eSRobbie Ko root, readonly, inherit); 1035c11fbb6eSRobbie Ko out: 1036c11fbb6eSRobbie Ko if (snapshot_force_cow) 1037c11fbb6eSRobbie Ko atomic_dec(&root->snapshot_force_cow); 1038c11fbb6eSRobbie Ko btrfs_drew_read_unlock(&root->snapshot_lock); 1039c11fbb6eSRobbie Ko return ret; 1040c11fbb6eSRobbie Ko } 1041c11fbb6eSRobbie Ko 1042d5633b0dSQu Wenruo /* 1043d5633b0dSQu Wenruo * Defrag specific helper to get an extent map. 1044d5633b0dSQu Wenruo * 1045d5633b0dSQu Wenruo * Differences between this and btrfs_get_extent() are: 1046d5633b0dSQu Wenruo * 1047d5633b0dSQu Wenruo * - No extent_map will be added to inode->extent_tree 1048d5633b0dSQu Wenruo * To reduce memory usage in the long run. 1049d5633b0dSQu Wenruo * 1050d5633b0dSQu Wenruo * - Extra optimization to skip file extents older than @newer_than 1051d5633b0dSQu Wenruo * By using btrfs_search_forward() we can skip entire file ranges that 1052d5633b0dSQu Wenruo * have extents created in past transactions, because btrfs_search_forward() 1053d5633b0dSQu Wenruo * will not visit leaves and nodes with a generation smaller than given 1054d5633b0dSQu Wenruo * minimal generation threshold (@newer_than). 1055d5633b0dSQu Wenruo * 1056d5633b0dSQu Wenruo * Return valid em if we find a file extent matching the requirement. 1057d5633b0dSQu Wenruo * Return NULL if we can not find a file extent matching the requirement. 1058d5633b0dSQu Wenruo * 1059d5633b0dSQu Wenruo * Return ERR_PTR() for error. 1060d5633b0dSQu Wenruo */ 1061d5633b0dSQu Wenruo static struct extent_map *defrag_get_extent(struct btrfs_inode *inode, 1062d5633b0dSQu Wenruo u64 start, u64 newer_than) 1063d5633b0dSQu Wenruo { 1064d5633b0dSQu Wenruo struct btrfs_root *root = inode->root; 1065d5633b0dSQu Wenruo struct btrfs_file_extent_item *fi; 1066d5633b0dSQu Wenruo struct btrfs_path path = { 0 }; 1067d5633b0dSQu Wenruo struct extent_map *em; 1068d5633b0dSQu Wenruo struct btrfs_key key; 1069d5633b0dSQu Wenruo u64 ino = btrfs_ino(inode); 1070d5633b0dSQu Wenruo int ret; 1071d5633b0dSQu Wenruo 1072d5633b0dSQu Wenruo em = alloc_extent_map(); 1073d5633b0dSQu Wenruo if (!em) { 1074d5633b0dSQu Wenruo ret = -ENOMEM; 1075d5633b0dSQu Wenruo goto err; 1076d5633b0dSQu Wenruo } 1077d5633b0dSQu Wenruo 1078d5633b0dSQu Wenruo key.objectid = ino; 1079d5633b0dSQu Wenruo key.type = BTRFS_EXTENT_DATA_KEY; 1080d5633b0dSQu Wenruo key.offset = start; 1081d5633b0dSQu Wenruo 1082d5633b0dSQu Wenruo if (newer_than) { 1083d5633b0dSQu Wenruo ret = btrfs_search_forward(root, &key, &path, newer_than); 1084d5633b0dSQu Wenruo if (ret < 0) 1085d5633b0dSQu Wenruo goto err; 1086d5633b0dSQu Wenruo /* Can't find anything newer */ 1087d5633b0dSQu Wenruo if (ret > 0) 1088d5633b0dSQu Wenruo goto not_found; 1089d5633b0dSQu Wenruo } else { 1090d5633b0dSQu Wenruo ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0); 1091d5633b0dSQu Wenruo if (ret < 0) 1092d5633b0dSQu Wenruo goto err; 1093d5633b0dSQu Wenruo } 1094d5633b0dSQu Wenruo if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) { 1095d5633b0dSQu Wenruo /* 1096d5633b0dSQu Wenruo * If btrfs_search_slot() makes path to point beyond nritems, 1097d5633b0dSQu Wenruo * we should not have an empty leaf, as this inode must at 1098d5633b0dSQu Wenruo * least have its INODE_ITEM. 1099d5633b0dSQu Wenruo */ 1100d5633b0dSQu Wenruo ASSERT(btrfs_header_nritems(path.nodes[0])); 1101d5633b0dSQu Wenruo path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1; 1102d5633b0dSQu Wenruo } 1103d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1104d5633b0dSQu Wenruo /* Perfect match, no need to go one slot back */ 1105d5633b0dSQu Wenruo if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY && 1106d5633b0dSQu Wenruo key.offset == start) 1107d5633b0dSQu Wenruo goto iterate; 1108d5633b0dSQu Wenruo 1109d5633b0dSQu Wenruo /* We didn't find a perfect match, needs to go one slot back */ 1110d5633b0dSQu Wenruo if (path.slots[0] > 0) { 1111d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1112d5633b0dSQu Wenruo if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) 1113d5633b0dSQu Wenruo path.slots[0]--; 1114d5633b0dSQu Wenruo } 1115d5633b0dSQu Wenruo 1116d5633b0dSQu Wenruo iterate: 1117d5633b0dSQu Wenruo /* Iterate through the path to find a file extent covering @start */ 1118d5633b0dSQu Wenruo while (true) { 1119d5633b0dSQu Wenruo u64 extent_end; 1120d5633b0dSQu Wenruo 1121d5633b0dSQu Wenruo if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) 1122d5633b0dSQu Wenruo goto next; 1123d5633b0dSQu Wenruo 1124d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1125d5633b0dSQu Wenruo 1126d5633b0dSQu Wenruo /* 1127d5633b0dSQu Wenruo * We may go one slot back to INODE_REF/XATTR item, then 1128d5633b0dSQu Wenruo * need to go forward until we reach an EXTENT_DATA. 1129d5633b0dSQu Wenruo * But we should still has the correct ino as key.objectid. 1130d5633b0dSQu Wenruo */ 1131d5633b0dSQu Wenruo if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY) 1132d5633b0dSQu Wenruo goto next; 1133d5633b0dSQu Wenruo 1134d5633b0dSQu Wenruo /* It's beyond our target range, definitely not extent found */ 1135d5633b0dSQu Wenruo if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY) 1136d5633b0dSQu Wenruo goto not_found; 1137d5633b0dSQu Wenruo 1138d5633b0dSQu Wenruo /* 1139d5633b0dSQu Wenruo * | |<- File extent ->| 1140d5633b0dSQu Wenruo * \- start 1141d5633b0dSQu Wenruo * 1142d5633b0dSQu Wenruo * This means there is a hole between start and key.offset. 1143d5633b0dSQu Wenruo */ 1144d5633b0dSQu Wenruo if (key.offset > start) { 1145d5633b0dSQu Wenruo em->start = start; 1146d5633b0dSQu Wenruo em->orig_start = start; 1147d5633b0dSQu Wenruo em->block_start = EXTENT_MAP_HOLE; 1148d5633b0dSQu Wenruo em->len = key.offset - start; 1149d5633b0dSQu Wenruo break; 1150d5633b0dSQu Wenruo } 1151d5633b0dSQu Wenruo 1152d5633b0dSQu Wenruo fi = btrfs_item_ptr(path.nodes[0], path.slots[0], 1153d5633b0dSQu Wenruo struct btrfs_file_extent_item); 1154d5633b0dSQu Wenruo extent_end = btrfs_file_extent_end(&path); 1155d5633b0dSQu Wenruo 1156d5633b0dSQu Wenruo /* 1157d5633b0dSQu Wenruo * |<- file extent ->| | 1158d5633b0dSQu Wenruo * \- start 1159d5633b0dSQu Wenruo * 1160d5633b0dSQu Wenruo * We haven't reached start, search next slot. 1161d5633b0dSQu Wenruo */ 1162d5633b0dSQu Wenruo if (extent_end <= start) 1163d5633b0dSQu Wenruo goto next; 1164d5633b0dSQu Wenruo 1165d5633b0dSQu Wenruo /* Now this extent covers @start, convert it to em */ 1166d5633b0dSQu Wenruo btrfs_extent_item_to_extent_map(inode, &path, fi, false, em); 1167d5633b0dSQu Wenruo break; 1168d5633b0dSQu Wenruo next: 1169d5633b0dSQu Wenruo ret = btrfs_next_item(root, &path); 1170d5633b0dSQu Wenruo if (ret < 0) 1171d5633b0dSQu Wenruo goto err; 1172d5633b0dSQu Wenruo if (ret > 0) 1173d5633b0dSQu Wenruo goto not_found; 1174d5633b0dSQu Wenruo } 1175d5633b0dSQu Wenruo btrfs_release_path(&path); 1176d5633b0dSQu Wenruo return em; 1177d5633b0dSQu Wenruo 1178d5633b0dSQu Wenruo not_found: 1179d5633b0dSQu Wenruo btrfs_release_path(&path); 1180d5633b0dSQu Wenruo free_extent_map(em); 1181d5633b0dSQu Wenruo return NULL; 1182d5633b0dSQu Wenruo 1183d5633b0dSQu Wenruo err: 1184d5633b0dSQu Wenruo btrfs_release_path(&path); 1185d5633b0dSQu Wenruo free_extent_map(em); 1186d5633b0dSQu Wenruo return ERR_PTR(ret); 1187d5633b0dSQu Wenruo } 1188d5633b0dSQu Wenruo 1189e9eec721SQu Wenruo static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start, 1190d5633b0dSQu Wenruo u64 newer_than, bool locked) 119117ce6ef8SLiu Bo { 119217ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1193940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 11946c282eb4SLi Zefan struct extent_map *em; 119576068caeSQu Wenruo const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize; 1196940100a4SChris Mason 1197940100a4SChris Mason /* 1198940100a4SChris Mason * hopefully we have this extent in the tree already, try without 1199940100a4SChris Mason * the full extent lock 1200940100a4SChris Mason */ 1201940100a4SChris Mason read_lock(&em_tree->lock); 120276068caeSQu Wenruo em = lookup_extent_mapping(em_tree, start, sectorsize); 1203940100a4SChris Mason read_unlock(&em_tree->lock); 1204940100a4SChris Mason 1205199257a7SQu Wenruo /* 1206199257a7SQu Wenruo * We can get a merged extent, in that case, we need to re-search 1207199257a7SQu Wenruo * tree to get the original em for defrag. 1208199257a7SQu Wenruo * 1209199257a7SQu Wenruo * If @newer_than is 0 or em::generation < newer_than, we can trust 1210199257a7SQu Wenruo * this em, as either we don't care about the generation, or the 1211199257a7SQu Wenruo * merged extent map will be rejected anyway. 1212199257a7SQu Wenruo */ 1213199257a7SQu Wenruo if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) && 1214199257a7SQu Wenruo newer_than && em->generation >= newer_than) { 1215199257a7SQu Wenruo free_extent_map(em); 1216199257a7SQu Wenruo em = NULL; 1217199257a7SQu Wenruo } 1218199257a7SQu Wenruo 1219940100a4SChris Mason if (!em) { 1220308d9800SFilipe Manana struct extent_state *cached = NULL; 122176068caeSQu Wenruo u64 end = start + sectorsize - 1; 1222308d9800SFilipe Manana 1223940100a4SChris Mason /* get the big lock and read metadata off disk */ 1224e9eec721SQu Wenruo if (!locked) 1225ff13db41SDavid Sterba lock_extent_bits(io_tree, start, end, &cached); 1226d5633b0dSQu Wenruo em = defrag_get_extent(BTRFS_I(inode), start, newer_than); 1227e9eec721SQu Wenruo if (!locked) 1228e43bbe5eSDavid Sterba unlock_extent_cached(io_tree, start, end, &cached); 1229940100a4SChris Mason 12306cf8bfbfSDan Carpenter if (IS_ERR(em)) 12316c282eb4SLi Zefan return NULL; 1232940100a4SChris Mason } 1233940100a4SChris Mason 12346c282eb4SLi Zefan return em; 12356c282eb4SLi Zefan } 12366c282eb4SLi Zefan 1237979b25c3SQu Wenruo static u32 get_extent_max_capacity(const struct extent_map *em) 1238979b25c3SQu Wenruo { 1239979b25c3SQu Wenruo if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) 1240979b25c3SQu Wenruo return BTRFS_MAX_COMPRESSED; 1241979b25c3SQu Wenruo return BTRFS_MAX_EXTENT_SIZE; 1242979b25c3SQu Wenruo } 1243979b25c3SQu Wenruo 1244e9eec721SQu Wenruo static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em, 124575a36a7dSQu Wenruo u32 extent_thresh, u64 newer_than, bool locked) 12466c282eb4SLi Zefan { 12476c282eb4SLi Zefan struct extent_map *next; 12487093f152SQu Wenruo bool ret = false; 12496c282eb4SLi Zefan 12506c282eb4SLi Zefan /* this is the last extent */ 12516c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 12526c282eb4SLi Zefan return false; 12536c282eb4SLi Zefan 1254d5633b0dSQu Wenruo /* 125575a36a7dSQu Wenruo * Here we need to pass @newer_then when checking the next extent, or 125675a36a7dSQu Wenruo * we will hit a case we mark current extent for defrag, but the next 125775a36a7dSQu Wenruo * one will not be a target. 125875a36a7dSQu Wenruo * This will just cause extra IO without really reducing the fragments. 1259d5633b0dSQu Wenruo */ 126075a36a7dSQu Wenruo next = defrag_lookup_extent(inode, em->start + em->len, newer_than, locked); 12617093f152SQu Wenruo /* No more em or hole */ 1262e9512d72SChris Mason if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 12637093f152SQu Wenruo goto out; 12647093f152SQu Wenruo if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags)) 12657093f152SQu Wenruo goto out; 1266979b25c3SQu Wenruo /* 1267979b25c3SQu Wenruo * If the next extent is at its max capacity, defragging current extent 1268979b25c3SQu Wenruo * makes no sense, as the total number of extents won't change. 1269979b25c3SQu Wenruo */ 1270979b25c3SQu Wenruo if (next->len >= get_extent_max_capacity(em)) 1271979b25c3SQu Wenruo goto out; 127275a36a7dSQu Wenruo /* Skip older extent */ 127375a36a7dSQu Wenruo if (next->generation < newer_than) 127475a36a7dSQu Wenruo goto out; 127575a36a7dSQu Wenruo /* Also check extent size */ 127675a36a7dSQu Wenruo if (next->len >= extent_thresh) 127775a36a7dSQu Wenruo goto out; 127875a36a7dSQu Wenruo 12797093f152SQu Wenruo ret = true; 12807093f152SQu Wenruo out: 12816c282eb4SLi Zefan free_extent_map(next); 12826c282eb4SLi Zefan return ret; 12836c282eb4SLi Zefan } 12846c282eb4SLi Zefan 12854cb5300bSChris Mason /* 12865767b50cSQu Wenruo * Prepare one page to be defragged. 12875767b50cSQu Wenruo * 12885767b50cSQu Wenruo * This will ensure: 12895767b50cSQu Wenruo * 12905767b50cSQu Wenruo * - Returned page is locked and has been set up properly. 12915767b50cSQu Wenruo * - No ordered extent exists in the page. 12925767b50cSQu Wenruo * - The page is uptodate. 12935767b50cSQu Wenruo * 12945767b50cSQu Wenruo * NOTE: Caller should also wait for page writeback after the cluster is 12955767b50cSQu Wenruo * prepared, here we don't do writeback wait for each page. 12965767b50cSQu Wenruo */ 12975767b50cSQu Wenruo static struct page *defrag_prepare_one_page(struct btrfs_inode *inode, 12985767b50cSQu Wenruo pgoff_t index) 1299940100a4SChris Mason { 13005767b50cSQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 13015767b50cSQu Wenruo gfp_t mask = btrfs_alloc_write_mask(mapping); 13025767b50cSQu Wenruo u64 page_start = (u64)index << PAGE_SHIFT; 13035767b50cSQu Wenruo u64 page_end = page_start + PAGE_SIZE - 1; 13044cb5300bSChris Mason struct extent_state *cached_state = NULL; 13054cb5300bSChris Mason struct page *page; 13065767b50cSQu Wenruo int ret; 13075767b50cSQu Wenruo 1308600a45e1SMiao Xie again: 13095767b50cSQu Wenruo page = find_or_create_page(mapping, index, mask); 13104cb5300bSChris Mason if (!page) 13115767b50cSQu Wenruo return ERR_PTR(-ENOMEM); 13125767b50cSQu Wenruo 131324bcb454SOmar Sandoval /* 131424bcb454SOmar Sandoval * Since we can defragment files opened read-only, we can encounter 131524bcb454SOmar Sandoval * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We 131624bcb454SOmar Sandoval * can't do I/O using huge pages yet, so return an error for now. 131724bcb454SOmar Sandoval * Filesystem transparent huge pages are typically only used for 131824bcb454SOmar Sandoval * executables that explicitly enable them, so this isn't very 131924bcb454SOmar Sandoval * restrictive. 132024bcb454SOmar Sandoval */ 132124bcb454SOmar Sandoval if (PageCompound(page)) { 132224bcb454SOmar Sandoval unlock_page(page); 132324bcb454SOmar Sandoval put_page(page); 132424bcb454SOmar Sandoval return ERR_PTR(-ETXTBSY); 132524bcb454SOmar Sandoval } 13264cb5300bSChris Mason 132732443de3SQu Wenruo ret = set_page_extent_mapped(page); 132832443de3SQu Wenruo if (ret < 0) { 132932443de3SQu Wenruo unlock_page(page); 133032443de3SQu Wenruo put_page(page); 13315767b50cSQu Wenruo return ERR_PTR(ret); 133232443de3SQu Wenruo } 133332443de3SQu Wenruo 13345767b50cSQu Wenruo /* Wait for any existing ordered extent in the range */ 1335600a45e1SMiao Xie while (1) { 13365767b50cSQu Wenruo struct btrfs_ordered_extent *ordered; 13375767b50cSQu Wenruo 13385767b50cSQu Wenruo lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); 13395767b50cSQu Wenruo ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); 13405767b50cSQu Wenruo unlock_extent_cached(&inode->io_tree, page_start, page_end, 1341e43bbe5eSDavid Sterba &cached_state); 1342600a45e1SMiao Xie if (!ordered) 1343600a45e1SMiao Xie break; 1344600a45e1SMiao Xie 1345600a45e1SMiao Xie unlock_page(page); 1346c0a43603SNikolay Borisov btrfs_start_ordered_extent(ordered, 1); 1347600a45e1SMiao Xie btrfs_put_ordered_extent(ordered); 1348600a45e1SMiao Xie lock_page(page); 13491f12bd06SLiu Bo /* 13505767b50cSQu Wenruo * We unlocked the page above, so we need check if it was 13515767b50cSQu Wenruo * released or not. 13521f12bd06SLiu Bo */ 13535767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 13541f12bd06SLiu Bo unlock_page(page); 135509cbfeafSKirill A. Shutemov put_page(page); 13561f12bd06SLiu Bo goto again; 13571f12bd06SLiu Bo } 1358600a45e1SMiao Xie } 1359600a45e1SMiao Xie 13605767b50cSQu Wenruo /* 13615767b50cSQu Wenruo * Now the page range has no ordered extent any more. Read the page to 13625767b50cSQu Wenruo * make it uptodate. 13635767b50cSQu Wenruo */ 13644cb5300bSChris Mason if (!PageUptodate(page)) { 13654cb5300bSChris Mason btrfs_readpage(NULL, page); 13664cb5300bSChris Mason lock_page(page); 13675767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 1368600a45e1SMiao Xie unlock_page(page); 136909cbfeafSKirill A. Shutemov put_page(page); 1370600a45e1SMiao Xie goto again; 1371600a45e1SMiao Xie } 13725767b50cSQu Wenruo if (!PageUptodate(page)) { 13735767b50cSQu Wenruo unlock_page(page); 13745767b50cSQu Wenruo put_page(page); 13755767b50cSQu Wenruo return ERR_PTR(-EIO); 13764cb5300bSChris Mason } 13775767b50cSQu Wenruo } 13785767b50cSQu Wenruo return page; 13795767b50cSQu Wenruo } 13804cb5300bSChris Mason 1381eb793cf8SQu Wenruo struct defrag_target_range { 1382eb793cf8SQu Wenruo struct list_head list; 1383eb793cf8SQu Wenruo u64 start; 1384eb793cf8SQu Wenruo u64 len; 1385eb793cf8SQu Wenruo }; 13864cb5300bSChris Mason 13874cb5300bSChris Mason /* 1388eb793cf8SQu Wenruo * Collect all valid target extents. 1389eb793cf8SQu Wenruo * 1390eb793cf8SQu Wenruo * @start: file offset to lookup 1391eb793cf8SQu Wenruo * @len: length to lookup 1392eb793cf8SQu Wenruo * @extent_thresh: file extent size threshold, any extent size >= this value 1393eb793cf8SQu Wenruo * will be ignored 1394eb793cf8SQu Wenruo * @newer_than: only defrag extents newer than this value 1395eb793cf8SQu Wenruo * @do_compress: whether the defrag is doing compression 1396eb793cf8SQu Wenruo * if true, @extent_thresh will be ignored and all regular 1397eb793cf8SQu Wenruo * file extents meeting @newer_than will be targets. 1398e9eec721SQu Wenruo * @locked: if the range has already held extent lock 1399eb793cf8SQu Wenruo * @target_list: list of targets file extents 14004cb5300bSChris Mason */ 1401eb793cf8SQu Wenruo static int defrag_collect_targets(struct btrfs_inode *inode, 1402eb793cf8SQu Wenruo u64 start, u64 len, u32 extent_thresh, 1403eb793cf8SQu Wenruo u64 newer_than, bool do_compress, 1404966d879bSQu Wenruo bool locked, struct list_head *target_list, 1405966d879bSQu Wenruo u64 *last_scanned_ret) 1406eb793cf8SQu Wenruo { 1407966d879bSQu Wenruo bool last_is_target = false; 1408eb793cf8SQu Wenruo u64 cur = start; 1409eb793cf8SQu Wenruo int ret = 0; 1410eb793cf8SQu Wenruo 1411eb793cf8SQu Wenruo while (cur < start + len) { 1412eb793cf8SQu Wenruo struct extent_map *em; 1413eb793cf8SQu Wenruo struct defrag_target_range *new; 1414eb793cf8SQu Wenruo bool next_mergeable = true; 1415eb793cf8SQu Wenruo u64 range_len; 1416eb793cf8SQu Wenruo 1417966d879bSQu Wenruo last_is_target = false; 1418d5633b0dSQu Wenruo em = defrag_lookup_extent(&inode->vfs_inode, cur, 1419d5633b0dSQu Wenruo newer_than, locked); 1420eb793cf8SQu Wenruo if (!em) 1421eb793cf8SQu Wenruo break; 1422eb793cf8SQu Wenruo 1423eb793cf8SQu Wenruo /* Skip hole/inline/preallocated extents */ 1424eb793cf8SQu Wenruo if (em->block_start >= EXTENT_MAP_LAST_BYTE || 1425eb793cf8SQu Wenruo test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) 1426eb793cf8SQu Wenruo goto next; 1427eb793cf8SQu Wenruo 1428eb793cf8SQu Wenruo /* Skip older extent */ 1429eb793cf8SQu Wenruo if (em->generation < newer_than) 1430eb793cf8SQu Wenruo goto next; 1431eb793cf8SQu Wenruo 14320d1ffa22SQu Wenruo /* This em is under writeback, no need to defrag */ 14330d1ffa22SQu Wenruo if (em->generation == (u64)-1) 14340d1ffa22SQu Wenruo goto next; 14350d1ffa22SQu Wenruo 1436eb793cf8SQu Wenruo /* 14370cb5950fSFilipe Manana * Our start offset might be in the middle of an existing extent 14380cb5950fSFilipe Manana * map, so take that into account. 14390cb5950fSFilipe Manana */ 14400cb5950fSFilipe Manana range_len = em->len - (cur - em->start); 14410cb5950fSFilipe Manana /* 14420cb5950fSFilipe Manana * If this range of the extent map is already flagged for delalloc, 14430cb5950fSFilipe Manana * skip it, because: 14440cb5950fSFilipe Manana * 14450cb5950fSFilipe Manana * 1) We could deadlock later, when trying to reserve space for 14460cb5950fSFilipe Manana * delalloc, because in case we can't immediately reserve space 14470cb5950fSFilipe Manana * the flusher can start delalloc and wait for the respective 14480cb5950fSFilipe Manana * ordered extents to complete. The deadlock would happen 14490cb5950fSFilipe Manana * because we do the space reservation while holding the range 14500cb5950fSFilipe Manana * locked, and starting writeback, or finishing an ordered 14510cb5950fSFilipe Manana * extent, requires locking the range; 14520cb5950fSFilipe Manana * 14530cb5950fSFilipe Manana * 2) If there's delalloc there, it means there's dirty pages for 14540cb5950fSFilipe Manana * which writeback has not started yet (we clean the delalloc 14550cb5950fSFilipe Manana * flag when starting writeback and after creating an ordered 14560cb5950fSFilipe Manana * extent). If we mark pages in an adjacent range for defrag, 14570cb5950fSFilipe Manana * then we will have a larger contiguous range for delalloc, 14580cb5950fSFilipe Manana * very likely resulting in a larger extent after writeback is 14590cb5950fSFilipe Manana * triggered (except in a case of free space fragmentation). 14600cb5950fSFilipe Manana */ 14610cb5950fSFilipe Manana if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1, 14620cb5950fSFilipe Manana EXTENT_DELALLOC, 0, NULL)) 14630cb5950fSFilipe Manana goto next; 14640cb5950fSFilipe Manana 14650cb5950fSFilipe Manana /* 1466eb793cf8SQu Wenruo * For do_compress case, we want to compress all valid file 1467eb793cf8SQu Wenruo * extents, thus no @extent_thresh or mergeable check. 1468eb793cf8SQu Wenruo */ 1469eb793cf8SQu Wenruo if (do_compress) 1470eb793cf8SQu Wenruo goto add; 1471eb793cf8SQu Wenruo 1472eb793cf8SQu Wenruo /* Skip too large extent */ 14730cb5950fSFilipe Manana if (range_len >= extent_thresh) 1474eb793cf8SQu Wenruo goto next; 1475eb793cf8SQu Wenruo 1476979b25c3SQu Wenruo /* 1477979b25c3SQu Wenruo * Skip extents already at its max capacity, this is mostly for 1478979b25c3SQu Wenruo * compressed extents, which max cap is only 128K. 1479979b25c3SQu Wenruo */ 1480979b25c3SQu Wenruo if (em->len >= get_extent_max_capacity(em)) 1481979b25c3SQu Wenruo goto next; 1482979b25c3SQu Wenruo 1483e9eec721SQu Wenruo next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em, 148475a36a7dSQu Wenruo extent_thresh, newer_than, locked); 1485eb793cf8SQu Wenruo if (!next_mergeable) { 1486eb793cf8SQu Wenruo struct defrag_target_range *last; 1487eb793cf8SQu Wenruo 1488eb793cf8SQu Wenruo /* Empty target list, no way to merge with last entry */ 1489eb793cf8SQu Wenruo if (list_empty(target_list)) 1490eb793cf8SQu Wenruo goto next; 1491eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1492eb793cf8SQu Wenruo struct defrag_target_range, list); 1493eb793cf8SQu Wenruo /* Not mergeable with last entry */ 1494eb793cf8SQu Wenruo if (last->start + last->len != cur) 1495eb793cf8SQu Wenruo goto next; 1496eb793cf8SQu Wenruo 1497eb793cf8SQu Wenruo /* Mergeable, fall through to add it to @target_list. */ 1498eb793cf8SQu Wenruo } 1499eb793cf8SQu Wenruo 1500eb793cf8SQu Wenruo add: 1501966d879bSQu Wenruo last_is_target = true; 1502eb793cf8SQu Wenruo range_len = min(extent_map_end(em), start + len) - cur; 1503eb793cf8SQu Wenruo /* 1504eb793cf8SQu Wenruo * This one is a good target, check if it can be merged into 1505eb793cf8SQu Wenruo * last range of the target list. 1506eb793cf8SQu Wenruo */ 1507eb793cf8SQu Wenruo if (!list_empty(target_list)) { 1508eb793cf8SQu Wenruo struct defrag_target_range *last; 1509eb793cf8SQu Wenruo 1510eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1511eb793cf8SQu Wenruo struct defrag_target_range, list); 1512eb793cf8SQu Wenruo ASSERT(last->start + last->len <= cur); 1513eb793cf8SQu Wenruo if (last->start + last->len == cur) { 1514eb793cf8SQu Wenruo /* Mergeable, enlarge the last entry */ 1515eb793cf8SQu Wenruo last->len += range_len; 1516eb793cf8SQu Wenruo goto next; 1517eb793cf8SQu Wenruo } 1518eb793cf8SQu Wenruo /* Fall through to allocate a new entry */ 1519eb793cf8SQu Wenruo } 1520eb793cf8SQu Wenruo 1521eb793cf8SQu Wenruo /* Allocate new defrag_target_range */ 1522eb793cf8SQu Wenruo new = kmalloc(sizeof(*new), GFP_NOFS); 1523eb793cf8SQu Wenruo if (!new) { 1524eb793cf8SQu Wenruo free_extent_map(em); 1525eb793cf8SQu Wenruo ret = -ENOMEM; 1526eb793cf8SQu Wenruo break; 1527eb793cf8SQu Wenruo } 1528eb793cf8SQu Wenruo new->start = cur; 1529eb793cf8SQu Wenruo new->len = range_len; 1530eb793cf8SQu Wenruo list_add_tail(&new->list, target_list); 1531eb793cf8SQu Wenruo 1532eb793cf8SQu Wenruo next: 1533eb793cf8SQu Wenruo cur = extent_map_end(em); 1534eb793cf8SQu Wenruo free_extent_map(em); 1535eb793cf8SQu Wenruo } 1536eb793cf8SQu Wenruo if (ret < 0) { 1537eb793cf8SQu Wenruo struct defrag_target_range *entry; 1538eb793cf8SQu Wenruo struct defrag_target_range *tmp; 1539eb793cf8SQu Wenruo 1540eb793cf8SQu Wenruo list_for_each_entry_safe(entry, tmp, target_list, list) { 1541eb793cf8SQu Wenruo list_del_init(&entry->list); 1542eb793cf8SQu Wenruo kfree(entry); 1543eb793cf8SQu Wenruo } 1544eb793cf8SQu Wenruo } 1545966d879bSQu Wenruo if (!ret && last_scanned_ret) { 1546966d879bSQu Wenruo /* 1547966d879bSQu Wenruo * If the last extent is not a target, the caller can skip to 1548966d879bSQu Wenruo * the end of that extent. 1549966d879bSQu Wenruo * Otherwise, we can only go the end of the specified range. 1550966d879bSQu Wenruo */ 1551966d879bSQu Wenruo if (!last_is_target) 1552966d879bSQu Wenruo *last_scanned_ret = max(cur, *last_scanned_ret); 1553966d879bSQu Wenruo else 1554966d879bSQu Wenruo *last_scanned_ret = max(start + len, *last_scanned_ret); 1555966d879bSQu Wenruo } 1556eb793cf8SQu Wenruo return ret; 1557eb793cf8SQu Wenruo } 1558eb793cf8SQu Wenruo 155922b398eeSQu Wenruo #define CLUSTER_SIZE (SZ_256K) 1560a55e65b8SDavid Sterba static_assert(IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE)); 156122b398eeSQu Wenruo 156222b398eeSQu Wenruo /* 156322b398eeSQu Wenruo * Defrag one contiguous target range. 156422b398eeSQu Wenruo * 156522b398eeSQu Wenruo * @inode: target inode 156622b398eeSQu Wenruo * @target: target range to defrag 156722b398eeSQu Wenruo * @pages: locked pages covering the defrag range 156822b398eeSQu Wenruo * @nr_pages: number of locked pages 156922b398eeSQu Wenruo * 157022b398eeSQu Wenruo * Caller should ensure: 157122b398eeSQu Wenruo * 157222b398eeSQu Wenruo * - Pages are prepared 157322b398eeSQu Wenruo * Pages should be locked, no ordered extent in the pages range, 157422b398eeSQu Wenruo * no writeback. 157522b398eeSQu Wenruo * 157622b398eeSQu Wenruo * - Extent bits are locked 157722b398eeSQu Wenruo */ 157822b398eeSQu Wenruo static int defrag_one_locked_target(struct btrfs_inode *inode, 157922b398eeSQu Wenruo struct defrag_target_range *target, 158022b398eeSQu Wenruo struct page **pages, int nr_pages, 158122b398eeSQu Wenruo struct extent_state **cached_state) 158222b398eeSQu Wenruo { 158322b398eeSQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 158422b398eeSQu Wenruo struct extent_changeset *data_reserved = NULL; 158522b398eeSQu Wenruo const u64 start = target->start; 158622b398eeSQu Wenruo const u64 len = target->len; 158722b398eeSQu Wenruo unsigned long last_index = (start + len - 1) >> PAGE_SHIFT; 158822b398eeSQu Wenruo unsigned long start_index = start >> PAGE_SHIFT; 158922b398eeSQu Wenruo unsigned long first_index = page_index(pages[0]); 159022b398eeSQu Wenruo int ret = 0; 159122b398eeSQu Wenruo int i; 159222b398eeSQu Wenruo 159322b398eeSQu Wenruo ASSERT(last_index - first_index + 1 <= nr_pages); 159422b398eeSQu Wenruo 159522b398eeSQu Wenruo ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len); 159622b398eeSQu Wenruo if (ret < 0) 159722b398eeSQu Wenruo return ret; 159822b398eeSQu Wenruo clear_extent_bit(&inode->io_tree, start, start + len - 1, 159922b398eeSQu Wenruo EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | 160022b398eeSQu Wenruo EXTENT_DEFRAG, 0, 0, cached_state); 160122b398eeSQu Wenruo set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state); 160222b398eeSQu Wenruo 160322b398eeSQu Wenruo /* Update the page status */ 160422b398eeSQu Wenruo for (i = start_index - first_index; i <= last_index - first_index; i++) { 160522b398eeSQu Wenruo ClearPageChecked(pages[i]); 160622b398eeSQu Wenruo btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len); 160722b398eeSQu Wenruo } 160822b398eeSQu Wenruo btrfs_delalloc_release_extents(inode, len); 160922b398eeSQu Wenruo extent_changeset_free(data_reserved); 161022b398eeSQu Wenruo 161122b398eeSQu Wenruo return ret; 161222b398eeSQu Wenruo } 161322b398eeSQu Wenruo 1614e9eec721SQu Wenruo static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len, 1615966d879bSQu Wenruo u32 extent_thresh, u64 newer_than, bool do_compress, 1616966d879bSQu Wenruo u64 *last_scanned_ret) 1617e9eec721SQu Wenruo { 1618e9eec721SQu Wenruo struct extent_state *cached_state = NULL; 1619e9eec721SQu Wenruo struct defrag_target_range *entry; 1620e9eec721SQu Wenruo struct defrag_target_range *tmp; 1621e9eec721SQu Wenruo LIST_HEAD(target_list); 1622e9eec721SQu Wenruo struct page **pages; 1623e9eec721SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1624e9eec721SQu Wenruo u64 last_index = (start + len - 1) >> PAGE_SHIFT; 1625e9eec721SQu Wenruo u64 start_index = start >> PAGE_SHIFT; 1626e9eec721SQu Wenruo unsigned int nr_pages = last_index - start_index + 1; 1627e9eec721SQu Wenruo int ret = 0; 1628e9eec721SQu Wenruo int i; 1629e9eec721SQu Wenruo 1630e9eec721SQu Wenruo ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE); 1631e9eec721SQu Wenruo ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize)); 1632e9eec721SQu Wenruo 1633e9eec721SQu Wenruo pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); 1634e9eec721SQu Wenruo if (!pages) 1635e9eec721SQu Wenruo return -ENOMEM; 1636e9eec721SQu Wenruo 1637e9eec721SQu Wenruo /* Prepare all pages */ 1638e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1639e9eec721SQu Wenruo pages[i] = defrag_prepare_one_page(inode, start_index + i); 1640e9eec721SQu Wenruo if (IS_ERR(pages[i])) { 1641e9eec721SQu Wenruo ret = PTR_ERR(pages[i]); 1642e9eec721SQu Wenruo pages[i] = NULL; 1643e9eec721SQu Wenruo goto free_pages; 1644e9eec721SQu Wenruo } 1645e9eec721SQu Wenruo } 1646e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) 16474cb5300bSChris Mason wait_on_page_writeback(pages[i]); 16484cb5300bSChris Mason 1649e9eec721SQu Wenruo /* Lock the pages range */ 1650e9eec721SQu Wenruo lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT, 1651e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1652e9eec721SQu Wenruo &cached_state); 1653e9eec721SQu Wenruo /* 1654e9eec721SQu Wenruo * Now we have a consistent view about the extent map, re-check 1655e9eec721SQu Wenruo * which range really needs to be defragged. 1656e9eec721SQu Wenruo * 1657e9eec721SQu Wenruo * And this time we have extent locked already, pass @locked = true 1658e9eec721SQu Wenruo * so that we won't relock the extent range and cause deadlock. 1659e9eec721SQu Wenruo */ 1660e9eec721SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1661e9eec721SQu Wenruo newer_than, do_compress, true, 1662966d879bSQu Wenruo &target_list, last_scanned_ret); 1663e9eec721SQu Wenruo if (ret < 0) 1664e9eec721SQu Wenruo goto unlock_extent; 16654cb5300bSChris Mason 1666e9eec721SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1667e9eec721SQu Wenruo ret = defrag_one_locked_target(inode, entry, pages, nr_pages, 1668e9eec721SQu Wenruo &cached_state); 1669e9eec721SQu Wenruo if (ret < 0) 1670e9eec721SQu Wenruo break; 1671e9eec721SQu Wenruo } 1672e9eec721SQu Wenruo 1673e9eec721SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1674e9eec721SQu Wenruo list_del_init(&entry->list); 1675e9eec721SQu Wenruo kfree(entry); 1676e9eec721SQu Wenruo } 1677e9eec721SQu Wenruo unlock_extent: 1678e9eec721SQu Wenruo unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT, 1679e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1680e9eec721SQu Wenruo &cached_state); 1681e9eec721SQu Wenruo free_pages: 1682e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1683e9eec721SQu Wenruo if (pages[i]) { 1684e9eec721SQu Wenruo unlock_page(pages[i]); 1685e9eec721SQu Wenruo put_page(pages[i]); 1686e9eec721SQu Wenruo } 1687e9eec721SQu Wenruo } 1688e9eec721SQu Wenruo kfree(pages); 1689e9eec721SQu Wenruo return ret; 1690e9eec721SQu Wenruo } 1691e9eec721SQu Wenruo 1692b18c3ab2SQu Wenruo static int defrag_one_cluster(struct btrfs_inode *inode, 1693b18c3ab2SQu Wenruo struct file_ra_state *ra, 1694b18c3ab2SQu Wenruo u64 start, u32 len, u32 extent_thresh, 1695b18c3ab2SQu Wenruo u64 newer_than, bool do_compress, 1696b18c3ab2SQu Wenruo unsigned long *sectors_defragged, 1697966d879bSQu Wenruo unsigned long max_sectors, 1698966d879bSQu Wenruo u64 *last_scanned_ret) 1699b18c3ab2SQu Wenruo { 1700b18c3ab2SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1701b18c3ab2SQu Wenruo struct defrag_target_range *entry; 1702b18c3ab2SQu Wenruo struct defrag_target_range *tmp; 1703b18c3ab2SQu Wenruo LIST_HEAD(target_list); 1704b18c3ab2SQu Wenruo int ret; 1705b18c3ab2SQu Wenruo 1706b18c3ab2SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1707b18c3ab2SQu Wenruo newer_than, do_compress, false, 1708966d879bSQu Wenruo &target_list, NULL); 1709b18c3ab2SQu Wenruo if (ret < 0) 1710b18c3ab2SQu Wenruo goto out; 1711b18c3ab2SQu Wenruo 1712b18c3ab2SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1713b18c3ab2SQu Wenruo u32 range_len = entry->len; 1714b18c3ab2SQu Wenruo 1715484167daSQu Wenruo /* Reached or beyond the limit */ 1716c080b414SQu Wenruo if (max_sectors && *sectors_defragged >= max_sectors) { 1717c080b414SQu Wenruo ret = 1; 1718b18c3ab2SQu Wenruo break; 1719c080b414SQu Wenruo } 1720b18c3ab2SQu Wenruo 1721b18c3ab2SQu Wenruo if (max_sectors) 1722b18c3ab2SQu Wenruo range_len = min_t(u32, range_len, 1723b18c3ab2SQu Wenruo (max_sectors - *sectors_defragged) * sectorsize); 1724b18c3ab2SQu Wenruo 1725966d879bSQu Wenruo /* 1726966d879bSQu Wenruo * If defrag_one_range() has updated last_scanned_ret, 1727966d879bSQu Wenruo * our range may already be invalid (e.g. hole punched). 1728966d879bSQu Wenruo * Skip if our range is before last_scanned_ret, as there is 1729966d879bSQu Wenruo * no need to defrag the range anymore. 1730966d879bSQu Wenruo */ 1731966d879bSQu Wenruo if (entry->start + range_len <= *last_scanned_ret) 1732966d879bSQu Wenruo continue; 1733966d879bSQu Wenruo 1734b18c3ab2SQu Wenruo if (ra) 1735b18c3ab2SQu Wenruo page_cache_sync_readahead(inode->vfs_inode.i_mapping, 1736b18c3ab2SQu Wenruo ra, NULL, entry->start >> PAGE_SHIFT, 1737b18c3ab2SQu Wenruo ((entry->start + range_len - 1) >> PAGE_SHIFT) - 1738b18c3ab2SQu Wenruo (entry->start >> PAGE_SHIFT) + 1); 1739b18c3ab2SQu Wenruo /* 1740b18c3ab2SQu Wenruo * Here we may not defrag any range if holes are punched before 1741b18c3ab2SQu Wenruo * we locked the pages. 1742b18c3ab2SQu Wenruo * But that's fine, it only affects the @sectors_defragged 1743b18c3ab2SQu Wenruo * accounting. 1744b18c3ab2SQu Wenruo */ 1745b18c3ab2SQu Wenruo ret = defrag_one_range(inode, entry->start, range_len, 1746966d879bSQu Wenruo extent_thresh, newer_than, do_compress, 1747966d879bSQu Wenruo last_scanned_ret); 1748b18c3ab2SQu Wenruo if (ret < 0) 1749b18c3ab2SQu Wenruo break; 1750484167daSQu Wenruo *sectors_defragged += range_len >> 1751484167daSQu Wenruo inode->root->fs_info->sectorsize_bits; 1752b18c3ab2SQu Wenruo } 1753b18c3ab2SQu Wenruo out: 1754b18c3ab2SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1755b18c3ab2SQu Wenruo list_del_init(&entry->list); 1756b18c3ab2SQu Wenruo kfree(entry); 1757b18c3ab2SQu Wenruo } 1758966d879bSQu Wenruo if (ret >= 0) 1759966d879bSQu Wenruo *last_scanned_ret = max(*last_scanned_ret, start + len); 1760b18c3ab2SQu Wenruo return ret; 1761b18c3ab2SQu Wenruo } 17627f458a38SFilipe Manana 17637f458a38SFilipe Manana /* 17641ccc2e8aSQu Wenruo * Entry point to file defragmentation. 17651ccc2e8aSQu Wenruo * 17661ccc2e8aSQu Wenruo * @inode: inode to be defragged 17671ccc2e8aSQu Wenruo * @ra: readahead state (can be NUL) 17681ccc2e8aSQu Wenruo * @range: defrag options including range and flags 17691ccc2e8aSQu Wenruo * @newer_than: minimum transid to defrag 17701ccc2e8aSQu Wenruo * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode 17711ccc2e8aSQu Wenruo * will be defragged. 1772484167daSQu Wenruo * 1773484167daSQu Wenruo * Return <0 for error. 1774c080b414SQu Wenruo * Return >=0 for the number of sectors defragged, and range->start will be updated 1775c080b414SQu Wenruo * to indicate the file offset where next defrag should be started at. 1776c080b414SQu Wenruo * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without 1777c080b414SQu Wenruo * defragging all the range). 17787f458a38SFilipe Manana */ 17791ccc2e8aSQu Wenruo int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, 17804cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 17814cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 17824cb5300bSChris Mason { 17830b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 17847b508037SQu Wenruo unsigned long sectors_defragged = 0; 1785151a31b2SLi Zefan u64 isize = i_size_read(inode); 17867b508037SQu Wenruo u64 cur; 17877b508037SQu Wenruo u64 last_byte; 17881e2ef46dSDavid Sterba bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS; 17891ccc2e8aSQu Wenruo bool ra_allocated = false; 17907b508037SQu Wenruo int compress_type = BTRFS_COMPRESS_ZLIB; 17917b508037SQu Wenruo int ret = 0; 17927b508037SQu Wenruo u32 extent_thresh = range->extent_thresh; 179327cdfde1SFilipe Manana pgoff_t start_index; 17944cb5300bSChris Mason 17950abd5b17SLiu Bo if (isize == 0) 17960abd5b17SLiu Bo return 0; 17970abd5b17SLiu Bo 17980abd5b17SLiu Bo if (range->start >= isize) 17990abd5b17SLiu Bo return -EINVAL; 18001a419d85SLi Zefan 18011e2ef46dSDavid Sterba if (do_compress) { 1802ce96b7ffSChengguang Xu if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES) 18031a419d85SLi Zefan return -EINVAL; 18041a419d85SLi Zefan if (range->compress_type) 18051a419d85SLi Zefan compress_type = range->compress_type; 18061a419d85SLi Zefan } 1807f46b5a66SChristoph Hellwig 18080abd5b17SLiu Bo if (extent_thresh == 0) 1809ee22184bSByongho Lee extent_thresh = SZ_256K; 1810f46b5a66SChristoph Hellwig 18117b508037SQu Wenruo if (range->start + range->len > range->start) { 18127b508037SQu Wenruo /* Got a specific range */ 18136b34cd8eSFilipe Manana last_byte = min(isize, range->start + range->len); 18147b508037SQu Wenruo } else { 18157b508037SQu Wenruo /* Defrag until file end */ 18166b34cd8eSFilipe Manana last_byte = isize; 18177b508037SQu Wenruo } 18187b508037SQu Wenruo 18196b34cd8eSFilipe Manana /* Align the range */ 18206b34cd8eSFilipe Manana cur = round_down(range->start, fs_info->sectorsize); 18216b34cd8eSFilipe Manana last_byte = round_up(last_byte, fs_info->sectorsize) - 1; 18226b34cd8eSFilipe Manana 18234cb5300bSChris Mason /* 18241ccc2e8aSQu Wenruo * If we were not given a ra, allocate a readahead context. As 18250a52d108SDavid Sterba * readahead is just an optimization, defrag will work without it so 18260a52d108SDavid Sterba * we don't error out. 18274cb5300bSChris Mason */ 18281ccc2e8aSQu Wenruo if (!ra) { 18291ccc2e8aSQu Wenruo ra_allocated = true; 183063e727ecSDavid Sterba ra = kzalloc(sizeof(*ra), GFP_KERNEL); 18310a52d108SDavid Sterba if (ra) 18324cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 18334cb5300bSChris Mason } 18344cb5300bSChris Mason 183527cdfde1SFilipe Manana /* 183627cdfde1SFilipe Manana * Make writeback start from the beginning of the range, so that the 183727cdfde1SFilipe Manana * defrag range can be written sequentially. 183827cdfde1SFilipe Manana */ 183927cdfde1SFilipe Manana start_index = cur >> PAGE_SHIFT; 184027cdfde1SFilipe Manana if (start_index < inode->i_mapping->writeback_index) 184127cdfde1SFilipe Manana inode->i_mapping->writeback_index = start_index; 184227cdfde1SFilipe Manana 18437b508037SQu Wenruo while (cur < last_byte) { 18443c9d31c7SFilipe Manana const unsigned long prev_sectors_defragged = sectors_defragged; 1845966d879bSQu Wenruo u64 last_scanned = cur; 18467b508037SQu Wenruo u64 cluster_end; 18471e701a32SChris Mason 1848b767c2fcSFilipe Manana if (btrfs_defrag_cancelled(fs_info)) { 1849b767c2fcSFilipe Manana ret = -EAGAIN; 1850b767c2fcSFilipe Manana break; 1851b767c2fcSFilipe Manana } 1852b767c2fcSFilipe Manana 18537b508037SQu Wenruo /* We want the cluster end at page boundary when possible */ 18547b508037SQu Wenruo cluster_end = (((cur >> PAGE_SHIFT) + 18557b508037SQu Wenruo (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1; 18567b508037SQu Wenruo cluster_end = min(cluster_end, last_byte); 18574cb5300bSChris Mason 185864708539SJosef Bacik btrfs_inode_lock(inode, 0); 1859eede2bf3SOmar Sandoval if (IS_SWAPFILE(inode)) { 1860eede2bf3SOmar Sandoval ret = -ETXTBSY; 18617b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 18627b508037SQu Wenruo break; 18637b508037SQu Wenruo } 18647b508037SQu Wenruo if (!(inode->i_sb->s_flags & SB_ACTIVE)) { 18657b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 18667b508037SQu Wenruo break; 18677b508037SQu Wenruo } 18681e2ef46dSDavid Sterba if (do_compress) 1869eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = compress_type; 18707b508037SQu Wenruo ret = defrag_one_cluster(BTRFS_I(inode), ra, cur, 18717b508037SQu Wenruo cluster_end + 1 - cur, extent_thresh, 1872966d879bSQu Wenruo newer_than, do_compress, §ors_defragged, 1873966d879bSQu Wenruo max_to_defrag, &last_scanned); 18743c9d31c7SFilipe Manana 18753c9d31c7SFilipe Manana if (sectors_defragged > prev_sectors_defragged) 18763c9d31c7SFilipe Manana balance_dirty_pages_ratelimited(inode->i_mapping); 18773c9d31c7SFilipe Manana 187864708539SJosef Bacik btrfs_inode_unlock(inode, 0); 18797b508037SQu Wenruo if (ret < 0) 18804cb5300bSChris Mason break; 1881966d879bSQu Wenruo cur = max(cluster_end + 1, last_scanned); 1882c080b414SQu Wenruo if (ret > 0) { 1883c080b414SQu Wenruo ret = 0; 1884c080b414SQu Wenruo break; 1885c080b414SQu Wenruo } 1886ea0eba69SQu Wenruo cond_resched(); 18874cb5300bSChris Mason } 1888f46b5a66SChristoph Hellwig 18897b508037SQu Wenruo if (ra_allocated) 18907b508037SQu Wenruo kfree(ra); 1891c080b414SQu Wenruo /* 1892c080b414SQu Wenruo * Update range.start for autodefrag, this will indicate where to start 1893c080b414SQu Wenruo * in next run. 1894c080b414SQu Wenruo */ 1895c080b414SQu Wenruo range->start = cur; 18967b508037SQu Wenruo if (sectors_defragged) { 18977b508037SQu Wenruo /* 18987b508037SQu Wenruo * We have defragged some sectors, for compression case they 18997b508037SQu Wenruo * need to be written back immediately. 19007b508037SQu Wenruo */ 19017b508037SQu Wenruo if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) { 19021e701a32SChris Mason filemap_flush(inode->i_mapping); 1903dec8ef90SFilipe Manana if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, 1904dec8ef90SFilipe Manana &BTRFS_I(inode)->runtime_flags)) 1905dec8ef90SFilipe Manana filemap_flush(inode->i_mapping); 1906dec8ef90SFilipe Manana } 19077b508037SQu Wenruo if (range->compress_type == BTRFS_COMPRESS_LZO) 19080b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, COMPRESS_LZO); 19097b508037SQu Wenruo else if (range->compress_type == BTRFS_COMPRESS_ZSTD) 19105c1aab1dSNick Terrell btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD); 19117b508037SQu Wenruo ret = sectors_defragged; 19121a419d85SLi Zefan } 19131e2ef46dSDavid Sterba if (do_compress) { 191464708539SJosef Bacik btrfs_inode_lock(inode, 0); 1915eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE; 191664708539SJosef Bacik btrfs_inode_unlock(inode, 0); 1917633085c7SFilipe David Borba Manana } 1918940100a4SChris Mason return ret; 1919f46b5a66SChristoph Hellwig } 1920f46b5a66SChristoph Hellwig 192117aaa434SDavid Sterba /* 192217aaa434SDavid Sterba * Try to start exclusive operation @type or cancel it if it's running. 192317aaa434SDavid Sterba * 192417aaa434SDavid Sterba * Return: 192517aaa434SDavid Sterba * 0 - normal mode, newly claimed op started 192617aaa434SDavid Sterba * >0 - normal mode, something else is running, 192717aaa434SDavid Sterba * return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space 192817aaa434SDavid Sterba * ECANCELED - cancel mode, successful cancel 192917aaa434SDavid Sterba * ENOTCONN - cancel mode, operation not running anymore 193017aaa434SDavid Sterba */ 193117aaa434SDavid Sterba static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info, 193217aaa434SDavid Sterba enum btrfs_exclusive_operation type, bool cancel) 193317aaa434SDavid Sterba { 193417aaa434SDavid Sterba if (!cancel) { 193517aaa434SDavid Sterba /* Start normal op */ 193617aaa434SDavid Sterba if (!btrfs_exclop_start(fs_info, type)) 193717aaa434SDavid Sterba return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 193817aaa434SDavid Sterba /* Exclusive operation is now claimed */ 193917aaa434SDavid Sterba return 0; 194017aaa434SDavid Sterba } 194117aaa434SDavid Sterba 194217aaa434SDavid Sterba /* Cancel running op */ 194317aaa434SDavid Sterba if (btrfs_exclop_start_try_lock(fs_info, type)) { 194417aaa434SDavid Sterba /* 194517aaa434SDavid Sterba * This blocks any exclop finish from setting it to NONE, so we 194617aaa434SDavid Sterba * request cancellation. Either it runs and we will wait for it, 194717aaa434SDavid Sterba * or it has finished and no waiting will happen. 194817aaa434SDavid Sterba */ 194917aaa434SDavid Sterba atomic_inc(&fs_info->reloc_cancel_req); 195017aaa434SDavid Sterba btrfs_exclop_start_unlock(fs_info); 195117aaa434SDavid Sterba 195217aaa434SDavid Sterba if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 195317aaa434SDavid Sterba wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING, 195417aaa434SDavid Sterba TASK_INTERRUPTIBLE); 195517aaa434SDavid Sterba 195617aaa434SDavid Sterba return -ECANCELED; 195717aaa434SDavid Sterba } 195817aaa434SDavid Sterba 195917aaa434SDavid Sterba /* Something else is running or none */ 196017aaa434SDavid Sterba return -ENOTCONN; 196117aaa434SDavid Sterba } 196217aaa434SDavid Sterba 1963198605a8SMiao Xie static noinline int btrfs_ioctl_resize(struct file *file, 196476dda93cSYan, Zheng void __user *arg) 1965f46b5a66SChristoph Hellwig { 1966562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 19670b246afaSJeff Mahoney struct inode *inode = file_inode(file); 19680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1969f46b5a66SChristoph Hellwig u64 new_size; 1970f46b5a66SChristoph Hellwig u64 old_size; 1971f46b5a66SChristoph Hellwig u64 devid = 1; 19720b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 1973f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1974f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1975f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1976f46b5a66SChristoph Hellwig char *sizestr; 19779a40f122SGui Hecheng char *retptr; 1978f46b5a66SChristoph Hellwig char *devstr = NULL; 1979f46b5a66SChristoph Hellwig int ret = 0; 1980f46b5a66SChristoph Hellwig int mod = 0; 1981bb059a37SDavid Sterba bool cancel; 1982f46b5a66SChristoph Hellwig 1983e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1984e441d54dSChris Mason return -EPERM; 1985e441d54dSChris Mason 1986198605a8SMiao Xie ret = mnt_want_write_file(file); 1987198605a8SMiao Xie if (ret) 1988198605a8SMiao Xie return ret; 1989198605a8SMiao Xie 1990bb059a37SDavid Sterba /* 1991bb059a37SDavid Sterba * Read the arguments before checking exclusivity to be able to 1992bb059a37SDavid Sterba * distinguish regular resize and cancel 1993bb059a37SDavid Sterba */ 1994dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1995c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1996c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1997bb059a37SDavid Sterba goto out_drop; 1998c9e9f97bSIlya Dryomov } 19995516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2000f46b5a66SChristoph Hellwig sizestr = vol_args->name; 2001bb059a37SDavid Sterba cancel = (strcmp("cancel", sizestr) == 0); 2002bb059a37SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel); 2003bb059a37SDavid Sterba if (ret) 2004bb059a37SDavid Sterba goto out_free; 2005bb059a37SDavid Sterba /* Exclusive operation is now claimed */ 2006bb059a37SDavid Sterba 2007f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 2008f46b5a66SChristoph Hellwig if (devstr) { 2009f46b5a66SChristoph Hellwig sizestr = devstr + 1; 2010f46b5a66SChristoph Hellwig *devstr = '\0'; 2011f46b5a66SChristoph Hellwig devstr = vol_args->name; 201258dfae63SZhangZhen ret = kstrtoull(devstr, 10, &devid); 201358dfae63SZhangZhen if (ret) 2014bb059a37SDavid Sterba goto out_finish; 2015dfd79829SMiao Xie if (!devid) { 2016dfd79829SMiao Xie ret = -EINVAL; 2017bb059a37SDavid Sterba goto out_finish; 2018dfd79829SMiao Xie } 20190b246afaSJeff Mahoney btrfs_info(fs_info, "resizing devid %llu", devid); 2020f46b5a66SChristoph Hellwig } 2021dba60f3fSMiao Xie 2022562d7b15SJosef Bacik args.devid = devid; 2023562d7b15SJosef Bacik device = btrfs_find_device(fs_info->fs_devices, &args); 2024f46b5a66SChristoph Hellwig if (!device) { 20250b246afaSJeff Mahoney btrfs_info(fs_info, "resizer unable to find device %llu", 2026c1c9ff7cSGeert Uytterhoeven devid); 2027dfd79829SMiao Xie ret = -ENODEV; 2028bb059a37SDavid Sterba goto out_finish; 2029f46b5a66SChristoph Hellwig } 2030dba60f3fSMiao Xie 2031ebbede42SAnand Jain if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 20320b246afaSJeff Mahoney btrfs_info(fs_info, 2033efe120a0SFrank Holton "resizer unable to apply on readonly device %llu", 2034c1c9ff7cSGeert Uytterhoeven devid); 2035dfd79829SMiao Xie ret = -EPERM; 2036bb059a37SDavid Sterba goto out_finish; 20374e42ae1bSLiu Bo } 20384e42ae1bSLiu Bo 2039f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 2040cda00ebaSChristoph Hellwig new_size = bdev_nr_bytes(device->bdev); 2041f46b5a66SChristoph Hellwig else { 2042f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 2043f46b5a66SChristoph Hellwig mod = -1; 2044f46b5a66SChristoph Hellwig sizestr++; 2045f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 2046f46b5a66SChristoph Hellwig mod = 1; 2047f46b5a66SChristoph Hellwig sizestr++; 2048f46b5a66SChristoph Hellwig } 20499a40f122SGui Hecheng new_size = memparse(sizestr, &retptr); 20509a40f122SGui Hecheng if (*retptr != '\0' || new_size == 0) { 2051f46b5a66SChristoph Hellwig ret = -EINVAL; 2052bb059a37SDavid Sterba goto out_finish; 2053f46b5a66SChristoph Hellwig } 2054f46b5a66SChristoph Hellwig } 2055f46b5a66SChristoph Hellwig 2056401e29c1SAnand Jain if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2057dfd79829SMiao Xie ret = -EPERM; 2058bb059a37SDavid Sterba goto out_finish; 205963a212abSStefan Behrens } 206063a212abSStefan Behrens 20617cc8e58dSMiao Xie old_size = btrfs_device_get_total_bytes(device); 2062f46b5a66SChristoph Hellwig 2063f46b5a66SChristoph Hellwig if (mod < 0) { 2064f46b5a66SChristoph Hellwig if (new_size > old_size) { 2065f46b5a66SChristoph Hellwig ret = -EINVAL; 2066bb059a37SDavid Sterba goto out_finish; 2067f46b5a66SChristoph Hellwig } 2068f46b5a66SChristoph Hellwig new_size = old_size - new_size; 2069f46b5a66SChristoph Hellwig } else if (mod > 0) { 2070eb8052e0SWenliang Fan if (new_size > ULLONG_MAX - old_size) { 2071902c68a4SGui Hecheng ret = -ERANGE; 2072bb059a37SDavid Sterba goto out_finish; 2073eb8052e0SWenliang Fan } 2074f46b5a66SChristoph Hellwig new_size = old_size + new_size; 2075f46b5a66SChristoph Hellwig } 2076f46b5a66SChristoph Hellwig 2077ee22184bSByongho Lee if (new_size < SZ_256M) { 2078f46b5a66SChristoph Hellwig ret = -EINVAL; 2079bb059a37SDavid Sterba goto out_finish; 2080f46b5a66SChristoph Hellwig } 2081cda00ebaSChristoph Hellwig if (new_size > bdev_nr_bytes(device->bdev)) { 2082f46b5a66SChristoph Hellwig ret = -EFBIG; 2083bb059a37SDavid Sterba goto out_finish; 2084f46b5a66SChristoph Hellwig } 2085f46b5a66SChristoph Hellwig 208647f08b96SNikolay Borisov new_size = round_down(new_size, fs_info->sectorsize); 2087f46b5a66SChristoph Hellwig 2088f46b5a66SChristoph Hellwig if (new_size > old_size) { 2089a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 209098d5dc13STsutomu Itoh if (IS_ERR(trans)) { 209198d5dc13STsutomu Itoh ret = PTR_ERR(trans); 2092bb059a37SDavid Sterba goto out_finish; 209398d5dc13STsutomu Itoh } 2094f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 20953a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 2096ece7d20eSMike Fleetwood } else if (new_size < old_size) { 2097f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 20980253f40eSjeff.liu } /* equal, nothing need to do */ 2099f46b5a66SChristoph Hellwig 2100faf8f7b9SMarcos Paulo de Souza if (ret == 0 && new_size != old_size) 2101faf8f7b9SMarcos Paulo de Souza btrfs_info_in_rcu(fs_info, 2102faf8f7b9SMarcos Paulo de Souza "resize device %s (devid %llu) from %llu to %llu", 2103faf8f7b9SMarcos Paulo de Souza rcu_str_deref(device->name), device->devid, 2104faf8f7b9SMarcos Paulo de Souza old_size, new_size); 2105bb059a37SDavid Sterba out_finish: 2106bb059a37SDavid Sterba btrfs_exclop_finish(fs_info); 2107c9e9f97bSIlya Dryomov out_free: 2108f46b5a66SChristoph Hellwig kfree(vol_args); 2109bb059a37SDavid Sterba out_drop: 211018f39c41SIlya Dryomov mnt_drop_write_file(file); 2111f46b5a66SChristoph Hellwig return ret; 2112f46b5a66SChristoph Hellwig } 2113f46b5a66SChristoph Hellwig 21145d54c67eSNikolay Borisov static noinline int __btrfs_ioctl_snap_create(struct file *file, 21154d4340c9SChristian Brauner struct user_namespace *mnt_userns, 211652f75f4fSDavid Sterba const char *name, unsigned long fd, int subvol, 21175d54c67eSNikolay Borisov bool readonly, 21188696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 2119f46b5a66SChristoph Hellwig { 2120f46b5a66SChristoph Hellwig int namelen; 21213de4586cSChris Mason int ret = 0; 2122f46b5a66SChristoph Hellwig 2123325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2124325c50e3SJeff Mahoney return -ENOTDIR; 2125325c50e3SJeff Mahoney 2126a874a63eSLiu Bo ret = mnt_want_write_file(file); 2127a874a63eSLiu Bo if (ret) 2128a874a63eSLiu Bo goto out; 2129a874a63eSLiu Bo 213072fd032eSSage Weil namelen = strlen(name); 213172fd032eSSage Weil if (strchr(name, '/')) { 2132f46b5a66SChristoph Hellwig ret = -EINVAL; 2133a874a63eSLiu Bo goto out_drop_write; 2134f46b5a66SChristoph Hellwig } 2135f46b5a66SChristoph Hellwig 213616780cabSChris Mason if (name[0] == '.' && 213716780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 213816780cabSChris Mason ret = -EEXIST; 2139a874a63eSLiu Bo goto out_drop_write; 214016780cabSChris Mason } 214116780cabSChris Mason 21423de4586cSChris Mason if (subvol) { 21434d4340c9SChristian Brauner ret = btrfs_mksubvol(&file->f_path, mnt_userns, name, 21444d4340c9SChristian Brauner namelen, NULL, readonly, inherit); 2145cb8e7090SChristoph Hellwig } else { 21462903ff01SAl Viro struct fd src = fdget(fd); 21473de4586cSChris Mason struct inode *src_inode; 21482903ff01SAl Viro if (!src.file) { 21493de4586cSChris Mason ret = -EINVAL; 2150a874a63eSLiu Bo goto out_drop_write; 21513de4586cSChris Mason } 21523de4586cSChris Mason 2153496ad9aaSAl Viro src_inode = file_inode(src.file); 2154496ad9aaSAl Viro if (src_inode->i_sb != file_inode(file)->i_sb) { 2155c79b4713SJosef Bacik btrfs_info(BTRFS_I(file_inode(file))->root->fs_info, 2156efe120a0SFrank Holton "Snapshot src from another FS"); 215723ad5b17SKusanagi Kouichi ret = -EXDEV; 21584d4340c9SChristian Brauner } else if (!inode_owner_or_capable(mnt_userns, src_inode)) { 2159d0242061SDavid Sterba /* 2160d0242061SDavid Sterba * Subvolume creation is not restricted, but snapshots 2161d0242061SDavid Sterba * are limited to own subvolumes only 2162d0242061SDavid Sterba */ 2163d0242061SDavid Sterba ret = -EPERM; 2164ecd18815SAl Viro } else { 21654d4340c9SChristian Brauner ret = btrfs_mksnapshot(&file->f_path, mnt_userns, 21664d4340c9SChristian Brauner name, namelen, 216772fd032eSSage Weil BTRFS_I(src_inode)->root, 21685d54c67eSNikolay Borisov readonly, inherit); 2169ecd18815SAl Viro } 21702903ff01SAl Viro fdput(src); 2171cb8e7090SChristoph Hellwig } 2172a874a63eSLiu Bo out_drop_write: 2173a874a63eSLiu Bo mnt_drop_write_file(file); 2174f46b5a66SChristoph Hellwig out: 217572fd032eSSage Weil return ret; 217672fd032eSSage Weil } 217772fd032eSSage Weil 217872fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 2179fa0d2b9bSLi Zefan void __user *arg, int subvol) 218072fd032eSSage Weil { 2181fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 218272fd032eSSage Weil int ret; 218372fd032eSSage Weil 2184325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2185325c50e3SJeff Mahoney return -ENOTDIR; 2186325c50e3SJeff Mahoney 2187fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2188fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 2189fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 2190fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2191fa0d2b9bSLi Zefan 21924d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 21934d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 21944d4340c9SChristian Brauner false, NULL); 2195fa0d2b9bSLi Zefan 2196fa0d2b9bSLi Zefan kfree(vol_args); 2197fa0d2b9bSLi Zefan return ret; 2198fa0d2b9bSLi Zefan } 2199fa0d2b9bSLi Zefan 2200fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 2201fa0d2b9bSLi Zefan void __user *arg, int subvol) 2202fa0d2b9bSLi Zefan { 2203fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 2204fa0d2b9bSLi Zefan int ret; 2205b83cc969SLi Zefan bool readonly = false; 22066f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 220772fd032eSSage Weil 2208325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2209325c50e3SJeff Mahoney return -ENOTDIR; 2210325c50e3SJeff Mahoney 2211fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2212fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 2213fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 2214fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 2215fdfb1e4fSLi Zefan 2216673990dbSDavid Sterba if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) { 2217b83cc969SLi Zefan ret = -EOPNOTSUPP; 2218c47ca32dSDan Carpenter goto free_args; 2219fdfb1e4fSLi Zefan } 2220fdfb1e4fSLi Zefan 2221b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 2222b83cc969SLi Zefan readonly = true; 22236f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 22245011c5a6SDan Carpenter u64 nums; 22255011c5a6SDan Carpenter 22265011c5a6SDan Carpenter if (vol_args->size < sizeof(*inherit) || 22275011c5a6SDan Carpenter vol_args->size > PAGE_SIZE) { 22286f72c7e2SArne Jansen ret = -EINVAL; 2229c47ca32dSDan Carpenter goto free_args; 22306f72c7e2SArne Jansen } 22316f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 22326f72c7e2SArne Jansen if (IS_ERR(inherit)) { 22336f72c7e2SArne Jansen ret = PTR_ERR(inherit); 2234c47ca32dSDan Carpenter goto free_args; 22356f72c7e2SArne Jansen } 22365011c5a6SDan Carpenter 22375011c5a6SDan Carpenter if (inherit->num_qgroups > PAGE_SIZE || 22385011c5a6SDan Carpenter inherit->num_ref_copies > PAGE_SIZE || 22395011c5a6SDan Carpenter inherit->num_excl_copies > PAGE_SIZE) { 22405011c5a6SDan Carpenter ret = -EINVAL; 22415011c5a6SDan Carpenter goto free_inherit; 22425011c5a6SDan Carpenter } 22435011c5a6SDan Carpenter 22445011c5a6SDan Carpenter nums = inherit->num_qgroups + 2 * inherit->num_ref_copies + 22455011c5a6SDan Carpenter 2 * inherit->num_excl_copies; 22465011c5a6SDan Carpenter if (vol_args->size != struct_size(inherit, qgroups, nums)) { 22475011c5a6SDan Carpenter ret = -EINVAL; 22485011c5a6SDan Carpenter goto free_inherit; 22495011c5a6SDan Carpenter } 22506f72c7e2SArne Jansen } 225175eaa0e2SSage Weil 22524d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 22534d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 22544d4340c9SChristian Brauner readonly, inherit); 2255c47ca32dSDan Carpenter if (ret) 2256c47ca32dSDan Carpenter goto free_inherit; 2257c47ca32dSDan Carpenter free_inherit: 22586f72c7e2SArne Jansen kfree(inherit); 2259c47ca32dSDan Carpenter free_args: 2260c47ca32dSDan Carpenter kfree(vol_args); 2261f46b5a66SChristoph Hellwig return ret; 2262f46b5a66SChristoph Hellwig } 2263f46b5a66SChristoph Hellwig 22649ad12305SSahil Kang static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode, 22650caa102dSLi Zefan void __user *arg) 22660caa102dSLi Zefan { 22670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 22680caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 22690caa102dSLi Zefan int ret = 0; 22700caa102dSLi Zefan u64 flags = 0; 22710caa102dSLi Zefan 22724a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) 22730caa102dSLi Zefan return -EINVAL; 22740caa102dSLi Zefan 22750b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 22760caa102dSLi Zefan if (btrfs_root_readonly(root)) 22770caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 22780b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 22790caa102dSLi Zefan 22800caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 22810caa102dSLi Zefan ret = -EFAULT; 22820caa102dSLi Zefan 22830caa102dSLi Zefan return ret; 22840caa102dSLi Zefan } 22850caa102dSLi Zefan 22860caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 22870caa102dSLi Zefan void __user *arg) 22880caa102dSLi Zefan { 2289496ad9aaSAl Viro struct inode *inode = file_inode(file); 22900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 22910caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 22920caa102dSLi Zefan struct btrfs_trans_handle *trans; 22930caa102dSLi Zefan u64 root_flags; 22940caa102dSLi Zefan u64 flags; 22950caa102dSLi Zefan int ret = 0; 22960caa102dSLi Zefan 229739e1674fSChristian Brauner if (!inode_owner_or_capable(file_mnt_user_ns(file), inode)) 2298bd60ea0fSDavid Sterba return -EPERM; 2299bd60ea0fSDavid Sterba 2300b9ca0664SLiu Bo ret = mnt_want_write_file(file); 2301b9ca0664SLiu Bo if (ret) 2302b9ca0664SLiu Bo goto out; 23030caa102dSLi Zefan 23044a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 2305b9ca0664SLiu Bo ret = -EINVAL; 2306b9ca0664SLiu Bo goto out_drop_write; 2307b9ca0664SLiu Bo } 23080caa102dSLi Zefan 2309b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 2310b9ca0664SLiu Bo ret = -EFAULT; 2311b9ca0664SLiu Bo goto out_drop_write; 2312b9ca0664SLiu Bo } 23130caa102dSLi Zefan 2314b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 2315b9ca0664SLiu Bo ret = -EOPNOTSUPP; 2316b9ca0664SLiu Bo goto out_drop_write; 2317b9ca0664SLiu Bo } 23180caa102dSLi Zefan 23190b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 23200caa102dSLi Zefan 23210caa102dSLi Zefan /* nothing to do */ 23220caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 2323b9ca0664SLiu Bo goto out_drop_sem; 23240caa102dSLi Zefan 23250caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 23262c686537SDavid Sterba if (flags & BTRFS_SUBVOL_RDONLY) { 23270caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 23280caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 23292c686537SDavid Sterba } else { 23302c686537SDavid Sterba /* 23312c686537SDavid Sterba * Block RO -> RW transition if this subvolume is involved in 23322c686537SDavid Sterba * send 23332c686537SDavid Sterba */ 23342c686537SDavid Sterba spin_lock(&root->root_item_lock); 23352c686537SDavid Sterba if (root->send_in_progress == 0) { 23360caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 23370caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 23382c686537SDavid Sterba spin_unlock(&root->root_item_lock); 23392c686537SDavid Sterba } else { 23402c686537SDavid Sterba spin_unlock(&root->root_item_lock); 23410b246afaSJeff Mahoney btrfs_warn(fs_info, 23422c686537SDavid Sterba "Attempt to set subvolume %llu read-write during send", 23432c686537SDavid Sterba root->root_key.objectid); 23442c686537SDavid Sterba ret = -EPERM; 23452c686537SDavid Sterba goto out_drop_sem; 23462c686537SDavid Sterba } 23472c686537SDavid Sterba } 23480caa102dSLi Zefan 23490caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 23500caa102dSLi Zefan if (IS_ERR(trans)) { 23510caa102dSLi Zefan ret = PTR_ERR(trans); 23520caa102dSLi Zefan goto out_reset; 23530caa102dSLi Zefan } 23540caa102dSLi Zefan 23550b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 23560caa102dSLi Zefan &root->root_key, &root->root_item); 23579417ebc8SNikolay Borisov if (ret < 0) { 23589417ebc8SNikolay Borisov btrfs_end_transaction(trans); 23599417ebc8SNikolay Borisov goto out_reset; 23609417ebc8SNikolay Borisov } 23610caa102dSLi Zefan 23629417ebc8SNikolay Borisov ret = btrfs_commit_transaction(trans); 23639417ebc8SNikolay Borisov 23640caa102dSLi Zefan out_reset: 23650caa102dSLi Zefan if (ret) 23660caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 2367b9ca0664SLiu Bo out_drop_sem: 23680b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 2369b9ca0664SLiu Bo out_drop_write: 2370b9ca0664SLiu Bo mnt_drop_write_file(file); 2371b9ca0664SLiu Bo out: 23720caa102dSLi Zefan return ret; 23730caa102dSLi Zefan } 23740caa102dSLi Zefan 2375ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 2376ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 2377ac8e9819SChris Mason { 2378abc6e134SChris Mason struct btrfs_key test; 2379abc6e134SChris Mason int ret; 2380abc6e134SChris Mason 2381abc6e134SChris Mason test.objectid = sk->min_objectid; 2382abc6e134SChris Mason test.type = sk->min_type; 2383abc6e134SChris Mason test.offset = sk->min_offset; 2384abc6e134SChris Mason 2385abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2386abc6e134SChris Mason if (ret < 0) 2387ac8e9819SChris Mason return 0; 2388abc6e134SChris Mason 2389abc6e134SChris Mason test.objectid = sk->max_objectid; 2390abc6e134SChris Mason test.type = sk->max_type; 2391abc6e134SChris Mason test.offset = sk->max_offset; 2392abc6e134SChris Mason 2393abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2394abc6e134SChris Mason if (ret > 0) 2395ac8e9819SChris Mason return 0; 2396ac8e9819SChris Mason return 1; 2397ac8e9819SChris Mason } 2398ac8e9819SChris Mason 2399df397565SJeff Mahoney static noinline int copy_to_sk(struct btrfs_path *path, 2400ac8e9819SChris Mason struct btrfs_key *key, 2401ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 24029b6e817dSGerhard Heift size_t *buf_size, 2403ba346b35SGerhard Heift char __user *ubuf, 2404ac8e9819SChris Mason unsigned long *sk_offset, 2405ac8e9819SChris Mason int *num_found) 2406ac8e9819SChris Mason { 2407ac8e9819SChris Mason u64 found_transid; 2408ac8e9819SChris Mason struct extent_buffer *leaf; 2409ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 2410dd81d459SNaohiro Aota struct btrfs_key test; 2411ac8e9819SChris Mason unsigned long item_off; 2412ac8e9819SChris Mason unsigned long item_len; 2413ac8e9819SChris Mason int nritems; 2414ac8e9819SChris Mason int i; 2415ac8e9819SChris Mason int slot; 2416ac8e9819SChris Mason int ret = 0; 2417ac8e9819SChris Mason 2418ac8e9819SChris Mason leaf = path->nodes[0]; 2419ac8e9819SChris Mason slot = path->slots[0]; 2420ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 2421ac8e9819SChris Mason 2422ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 2423ac8e9819SChris Mason i = nritems; 2424ac8e9819SChris Mason goto advance_key; 2425ac8e9819SChris Mason } 2426ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 2427ac8e9819SChris Mason 2428ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 2429ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 24303212fa14SJosef Bacik item_len = btrfs_item_size(leaf, i); 2431ac8e9819SChris Mason 243203b71c6cSGabriel de Perthuis btrfs_item_key_to_cpu(leaf, key, i); 243303b71c6cSGabriel de Perthuis if (!key_in_sk(key, sk)) 243403b71c6cSGabriel de Perthuis continue; 243503b71c6cSGabriel de Perthuis 24369b6e817dSGerhard Heift if (sizeof(sh) + item_len > *buf_size) { 24378f5f6178SGerhard Heift if (*num_found) { 2438ac8e9819SChris Mason ret = 1; 24398f5f6178SGerhard Heift goto out; 24408f5f6178SGerhard Heift } 24418f5f6178SGerhard Heift 24428f5f6178SGerhard Heift /* 24438f5f6178SGerhard Heift * return one empty item back for v1, which does not 24448f5f6178SGerhard Heift * handle -EOVERFLOW 24458f5f6178SGerhard Heift */ 24468f5f6178SGerhard Heift 24479b6e817dSGerhard Heift *buf_size = sizeof(sh) + item_len; 2448ac8e9819SChris Mason item_len = 0; 24498f5f6178SGerhard Heift ret = -EOVERFLOW; 24508f5f6178SGerhard Heift } 2451ac8e9819SChris Mason 24529b6e817dSGerhard Heift if (sizeof(sh) + item_len + *sk_offset > *buf_size) { 2453ac8e9819SChris Mason ret = 1; 245425c9bc2eSGerhard Heift goto out; 2455ac8e9819SChris Mason } 2456ac8e9819SChris Mason 2457ac8e9819SChris Mason sh.objectid = key->objectid; 2458ac8e9819SChris Mason sh.offset = key->offset; 2459ac8e9819SChris Mason sh.type = key->type; 2460ac8e9819SChris Mason sh.len = item_len; 2461ac8e9819SChris Mason sh.transid = found_transid; 2462ac8e9819SChris Mason 2463a48b73ecSJosef Bacik /* 2464a48b73ecSJosef Bacik * Copy search result header. If we fault then loop again so we 2465a48b73ecSJosef Bacik * can fault in the pages and -EFAULT there if there's a 2466a48b73ecSJosef Bacik * problem. Otherwise we'll fault and then copy the buffer in 2467a48b73ecSJosef Bacik * properly this next time through 2468a48b73ecSJosef Bacik */ 2469a48b73ecSJosef Bacik if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) { 2470a48b73ecSJosef Bacik ret = 0; 2471ba346b35SGerhard Heift goto out; 2472ba346b35SGerhard Heift } 2473ba346b35SGerhard Heift 2474ac8e9819SChris Mason *sk_offset += sizeof(sh); 2475ac8e9819SChris Mason 2476ac8e9819SChris Mason if (item_len) { 2477ba346b35SGerhard Heift char __user *up = ubuf + *sk_offset; 2478a48b73ecSJosef Bacik /* 2479a48b73ecSJosef Bacik * Copy the item, same behavior as above, but reset the 2480a48b73ecSJosef Bacik * * sk_offset so we copy the full thing again. 2481a48b73ecSJosef Bacik */ 2482a48b73ecSJosef Bacik if (read_extent_buffer_to_user_nofault(leaf, up, 2483ba346b35SGerhard Heift item_off, item_len)) { 2484a48b73ecSJosef Bacik ret = 0; 2485a48b73ecSJosef Bacik *sk_offset -= sizeof(sh); 2486ba346b35SGerhard Heift goto out; 2487ba346b35SGerhard Heift } 2488ba346b35SGerhard Heift 2489ac8e9819SChris Mason *sk_offset += item_len; 2490ac8e9819SChris Mason } 2491e2156867SHugo Mills (*num_found)++; 2492ac8e9819SChris Mason 24938f5f6178SGerhard Heift if (ret) /* -EOVERFLOW from above */ 24948f5f6178SGerhard Heift goto out; 24958f5f6178SGerhard Heift 249625c9bc2eSGerhard Heift if (*num_found >= sk->nr_items) { 249725c9bc2eSGerhard Heift ret = 1; 249825c9bc2eSGerhard Heift goto out; 249925c9bc2eSGerhard Heift } 2500ac8e9819SChris Mason } 2501ac8e9819SChris Mason advance_key: 2502ac8e9819SChris Mason ret = 0; 2503dd81d459SNaohiro Aota test.objectid = sk->max_objectid; 2504dd81d459SNaohiro Aota test.type = sk->max_type; 2505dd81d459SNaohiro Aota test.offset = sk->max_offset; 2506dd81d459SNaohiro Aota if (btrfs_comp_cpu_keys(key, &test) >= 0) 2507dd81d459SNaohiro Aota ret = 1; 2508dd81d459SNaohiro Aota else if (key->offset < (u64)-1) 2509abc6e134SChris Mason key->offset++; 2510dd81d459SNaohiro Aota else if (key->type < (u8)-1) { 2511abc6e134SChris Mason key->offset = 0; 2512abc6e134SChris Mason key->type++; 2513dd81d459SNaohiro Aota } else if (key->objectid < (u64)-1) { 2514abc6e134SChris Mason key->offset = 0; 2515abc6e134SChris Mason key->type = 0; 2516abc6e134SChris Mason key->objectid++; 2517abc6e134SChris Mason } else 2518abc6e134SChris Mason ret = 1; 251925c9bc2eSGerhard Heift out: 2520ba346b35SGerhard Heift /* 2521ba346b35SGerhard Heift * 0: all items from this leaf copied, continue with next 2522ba346b35SGerhard Heift * 1: * more items can be copied, but unused buffer is too small 2523ba346b35SGerhard Heift * * all items were found 2524ba346b35SGerhard Heift * Either way, it will stops the loop which iterates to the next 2525ba346b35SGerhard Heift * leaf 2526ba346b35SGerhard Heift * -EOVERFLOW: item was to large for buffer 2527ba346b35SGerhard Heift * -EFAULT: could not copy extent buffer back to userspace 2528ba346b35SGerhard Heift */ 2529ac8e9819SChris Mason return ret; 2530ac8e9819SChris Mason } 2531ac8e9819SChris Mason 2532ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 253312544442SGerhard Heift struct btrfs_ioctl_search_key *sk, 25349b6e817dSGerhard Heift size_t *buf_size, 2535ba346b35SGerhard Heift char __user *ubuf) 2536ac8e9819SChris Mason { 25370b246afaSJeff Mahoney struct btrfs_fs_info *info = btrfs_sb(inode->i_sb); 2538ac8e9819SChris Mason struct btrfs_root *root; 2539ac8e9819SChris Mason struct btrfs_key key; 2540ac8e9819SChris Mason struct btrfs_path *path; 2541ac8e9819SChris Mason int ret; 2542ac8e9819SChris Mason int num_found = 0; 2543ac8e9819SChris Mason unsigned long sk_offset = 0; 2544ac8e9819SChris Mason 25459b6e817dSGerhard Heift if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) { 25469b6e817dSGerhard Heift *buf_size = sizeof(struct btrfs_ioctl_search_header); 254712544442SGerhard Heift return -EOVERFLOW; 25489b6e817dSGerhard Heift } 254912544442SGerhard Heift 2550ac8e9819SChris Mason path = btrfs_alloc_path(); 2551ac8e9819SChris Mason if (!path) 2552ac8e9819SChris Mason return -ENOMEM; 2553ac8e9819SChris Mason 2554ac8e9819SChris Mason if (sk->tree_id == 0) { 2555ac8e9819SChris Mason /* search the root of the inode that was passed */ 255600246528SJosef Bacik root = btrfs_grab_root(BTRFS_I(inode)->root); 2557ac8e9819SChris Mason } else { 255856e9357aSDavid Sterba root = btrfs_get_fs_root(info, sk->tree_id, true); 2559ac8e9819SChris Mason if (IS_ERR(root)) { 2560ac8e9819SChris Mason btrfs_free_path(path); 2561ad1e3d56SMisono Tomohiro return PTR_ERR(root); 2562ac8e9819SChris Mason } 2563ac8e9819SChris Mason } 2564ac8e9819SChris Mason 2565ac8e9819SChris Mason key.objectid = sk->min_objectid; 2566ac8e9819SChris Mason key.type = sk->min_type; 2567ac8e9819SChris Mason key.offset = sk->min_offset; 2568ac8e9819SChris Mason 2569ac8e9819SChris Mason while (1) { 2570bb523b40SAndreas Gruenbacher ret = -EFAULT; 2571bb523b40SAndreas Gruenbacher if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset)) 2572a48b73ecSJosef Bacik break; 2573a48b73ecSJosef Bacik 25746174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &key, path, sk->min_transid); 2575ac8e9819SChris Mason if (ret != 0) { 2576ac8e9819SChris Mason if (ret > 0) 2577ac8e9819SChris Mason ret = 0; 2578ac8e9819SChris Mason goto err; 2579ac8e9819SChris Mason } 2580df397565SJeff Mahoney ret = copy_to_sk(path, &key, sk, buf_size, ubuf, 2581ac8e9819SChris Mason &sk_offset, &num_found); 2582b3b4aa74SDavid Sterba btrfs_release_path(path); 258325c9bc2eSGerhard Heift if (ret) 2584ac8e9819SChris Mason break; 2585ac8e9819SChris Mason 2586ac8e9819SChris Mason } 25878f5f6178SGerhard Heift if (ret > 0) 2588ac8e9819SChris Mason ret = 0; 2589ac8e9819SChris Mason err: 2590ac8e9819SChris Mason sk->nr_items = num_found; 259100246528SJosef Bacik btrfs_put_root(root); 2592ac8e9819SChris Mason btrfs_free_path(path); 2593ac8e9819SChris Mason return ret; 2594ac8e9819SChris Mason } 2595ac8e9819SChris Mason 25969ad12305SSahil Kang static noinline int btrfs_ioctl_tree_search(struct inode *inode, 2597ac8e9819SChris Mason void __user *argp) 2598ac8e9819SChris Mason { 2599ba346b35SGerhard Heift struct btrfs_ioctl_search_args __user *uargs; 2600ba346b35SGerhard Heift struct btrfs_ioctl_search_key sk; 2601ac8e9819SChris Mason int ret; 26029b6e817dSGerhard Heift size_t buf_size; 2603ac8e9819SChris Mason 2604ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2605ac8e9819SChris Mason return -EPERM; 2606ac8e9819SChris Mason 2607ba346b35SGerhard Heift uargs = (struct btrfs_ioctl_search_args __user *)argp; 2608ac8e9819SChris Mason 2609ba346b35SGerhard Heift if (copy_from_user(&sk, &uargs->key, sizeof(sk))) 2610ba346b35SGerhard Heift return -EFAULT; 2611ba346b35SGerhard Heift 2612ba346b35SGerhard Heift buf_size = sizeof(uargs->buf); 2613ac8e9819SChris Mason 2614ba346b35SGerhard Heift ret = search_ioctl(inode, &sk, &buf_size, uargs->buf); 26158f5f6178SGerhard Heift 26168f5f6178SGerhard Heift /* 26178f5f6178SGerhard Heift * In the origin implementation an overflow is handled by returning a 26188f5f6178SGerhard Heift * search header with a len of zero, so reset ret. 26198f5f6178SGerhard Heift */ 26208f5f6178SGerhard Heift if (ret == -EOVERFLOW) 26218f5f6178SGerhard Heift ret = 0; 26228f5f6178SGerhard Heift 2623ba346b35SGerhard Heift if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk))) 2624ac8e9819SChris Mason ret = -EFAULT; 2625ac8e9819SChris Mason return ret; 2626ac8e9819SChris Mason } 2627ac8e9819SChris Mason 26289ad12305SSahil Kang static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode, 2629cc68a8a5SGerhard Heift void __user *argp) 2630cc68a8a5SGerhard Heift { 2631cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 __user *uarg; 2632cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 args; 2633cc68a8a5SGerhard Heift int ret; 2634cc68a8a5SGerhard Heift size_t buf_size; 2635ee22184bSByongho Lee const size_t buf_limit = SZ_16M; 2636cc68a8a5SGerhard Heift 2637cc68a8a5SGerhard Heift if (!capable(CAP_SYS_ADMIN)) 2638cc68a8a5SGerhard Heift return -EPERM; 2639cc68a8a5SGerhard Heift 2640cc68a8a5SGerhard Heift /* copy search header and buffer size */ 2641cc68a8a5SGerhard Heift uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp; 2642cc68a8a5SGerhard Heift if (copy_from_user(&args, uarg, sizeof(args))) 2643cc68a8a5SGerhard Heift return -EFAULT; 2644cc68a8a5SGerhard Heift 2645cc68a8a5SGerhard Heift buf_size = args.buf_size; 2646cc68a8a5SGerhard Heift 2647cc68a8a5SGerhard Heift /* limit result size to 16MB */ 2648cc68a8a5SGerhard Heift if (buf_size > buf_limit) 2649cc68a8a5SGerhard Heift buf_size = buf_limit; 2650cc68a8a5SGerhard Heift 2651cc68a8a5SGerhard Heift ret = search_ioctl(inode, &args.key, &buf_size, 2652718dc5faSOmar Sandoval (char __user *)(&uarg->buf[0])); 2653cc68a8a5SGerhard Heift if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key))) 2654cc68a8a5SGerhard Heift ret = -EFAULT; 2655cc68a8a5SGerhard Heift else if (ret == -EOVERFLOW && 2656cc68a8a5SGerhard Heift copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size))) 2657cc68a8a5SGerhard Heift ret = -EFAULT; 2658cc68a8a5SGerhard Heift 265976dda93cSYan, Zheng return ret; 266076dda93cSYan, Zheng } 266176dda93cSYan, Zheng 266298d377a0STARUISI Hiroaki /* 2663ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 2664ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 266598d377a0STARUISI Hiroaki */ 266698d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 266798d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 266898d377a0STARUISI Hiroaki { 266998d377a0STARUISI Hiroaki struct btrfs_root *root; 267098d377a0STARUISI Hiroaki struct btrfs_key key; 2671ac8e9819SChris Mason char *ptr; 267298d377a0STARUISI Hiroaki int ret = -1; 267398d377a0STARUISI Hiroaki int slot; 267498d377a0STARUISI Hiroaki int len; 267598d377a0STARUISI Hiroaki int total_len = 0; 267698d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 267798d377a0STARUISI Hiroaki struct extent_buffer *l; 267898d377a0STARUISI Hiroaki struct btrfs_path *path; 267998d377a0STARUISI Hiroaki 268098d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 268198d377a0STARUISI Hiroaki name[0]='\0'; 268298d377a0STARUISI Hiroaki return 0; 268398d377a0STARUISI Hiroaki } 268498d377a0STARUISI Hiroaki 268598d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 268698d377a0STARUISI Hiroaki if (!path) 268798d377a0STARUISI Hiroaki return -ENOMEM; 268898d377a0STARUISI Hiroaki 2689c8bcbfbdSNikolay Borisov ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1]; 269098d377a0STARUISI Hiroaki 269156e9357aSDavid Sterba root = btrfs_get_fs_root(info, tree_id, true); 269298d377a0STARUISI Hiroaki if (IS_ERR(root)) { 2693ad1e3d56SMisono Tomohiro ret = PTR_ERR(root); 269488234012SJosef Bacik root = NULL; 269588234012SJosef Bacik goto out; 269688234012SJosef Bacik } 269798d377a0STARUISI Hiroaki 269898d377a0STARUISI Hiroaki key.objectid = dirid; 269998d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 27008ad6fcabSChris Mason key.offset = (u64)-1; 270198d377a0STARUISI Hiroaki 270298d377a0STARUISI Hiroaki while (1) { 27030ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 270418674c6cSFilipe David Borba Manana if (ret < 0) 270518674c6cSFilipe David Borba Manana goto out; 270618674c6cSFilipe David Borba Manana else if (ret > 0) { 2707ac8e9819SChris Mason ret = -ENOENT; 270898d377a0STARUISI Hiroaki goto out; 2709ac8e9819SChris Mason } 271018674c6cSFilipe David Borba Manana 271118674c6cSFilipe David Borba Manana l = path->nodes[0]; 271218674c6cSFilipe David Borba Manana slot = path->slots[0]; 271398d377a0STARUISI Hiroaki 271498d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 271598d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 271698d377a0STARUISI Hiroaki ptr -= len + 1; 271798d377a0STARUISI Hiroaki total_len += len + 1; 2718a696cf35SFilipe David Borba Manana if (ptr < name) { 2719a696cf35SFilipe David Borba Manana ret = -ENAMETOOLONG; 272098d377a0STARUISI Hiroaki goto out; 2721a696cf35SFilipe David Borba Manana } 272298d377a0STARUISI Hiroaki 272398d377a0STARUISI Hiroaki *(ptr + len) = '/'; 272498d377a0STARUISI Hiroaki read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 272598d377a0STARUISI Hiroaki 272698d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 272798d377a0STARUISI Hiroaki break; 272898d377a0STARUISI Hiroaki 2729b3b4aa74SDavid Sterba btrfs_release_path(path); 273098d377a0STARUISI Hiroaki key.objectid = key.offset; 27318ad6fcabSChris Mason key.offset = (u64)-1; 273298d377a0STARUISI Hiroaki dirid = key.objectid; 273398d377a0STARUISI Hiroaki } 273477906a50SLi Zefan memmove(name, ptr, total_len); 273598d377a0STARUISI Hiroaki name[total_len] = '\0'; 273698d377a0STARUISI Hiroaki ret = 0; 273798d377a0STARUISI Hiroaki out: 273800246528SJosef Bacik btrfs_put_root(root); 273998d377a0STARUISI Hiroaki btrfs_free_path(path); 2740ac8e9819SChris Mason return ret; 2741ac8e9819SChris Mason } 2742ac8e9819SChris Mason 27436623d9a0SChristian Brauner static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns, 27446623d9a0SChristian Brauner struct inode *inode, 274523d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args) 274623d0b79dSTomohiro Misono { 274723d0b79dSTomohiro Misono struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 274823d0b79dSTomohiro Misono struct super_block *sb = inode->i_sb; 274923d0b79dSTomohiro Misono struct btrfs_key upper_limit = BTRFS_I(inode)->location; 275023d0b79dSTomohiro Misono u64 treeid = BTRFS_I(inode)->root->root_key.objectid; 275123d0b79dSTomohiro Misono u64 dirid = args->dirid; 275223d0b79dSTomohiro Misono unsigned long item_off; 275323d0b79dSTomohiro Misono unsigned long item_len; 275423d0b79dSTomohiro Misono struct btrfs_inode_ref *iref; 275523d0b79dSTomohiro Misono struct btrfs_root_ref *rref; 2756b8a49ae1SJosef Bacik struct btrfs_root *root = NULL; 275723d0b79dSTomohiro Misono struct btrfs_path *path; 275823d0b79dSTomohiro Misono struct btrfs_key key, key2; 275923d0b79dSTomohiro Misono struct extent_buffer *leaf; 276023d0b79dSTomohiro Misono struct inode *temp_inode; 276123d0b79dSTomohiro Misono char *ptr; 276223d0b79dSTomohiro Misono int slot; 276323d0b79dSTomohiro Misono int len; 276423d0b79dSTomohiro Misono int total_len = 0; 276523d0b79dSTomohiro Misono int ret; 276623d0b79dSTomohiro Misono 276723d0b79dSTomohiro Misono path = btrfs_alloc_path(); 276823d0b79dSTomohiro Misono if (!path) 276923d0b79dSTomohiro Misono return -ENOMEM; 277023d0b79dSTomohiro Misono 277123d0b79dSTomohiro Misono /* 277223d0b79dSTomohiro Misono * If the bottom subvolume does not exist directly under upper_limit, 277323d0b79dSTomohiro Misono * construct the path in from the bottom up. 277423d0b79dSTomohiro Misono */ 277523d0b79dSTomohiro Misono if (dirid != upper_limit.objectid) { 277623d0b79dSTomohiro Misono ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1]; 277723d0b79dSTomohiro Misono 277856e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, treeid, true); 277923d0b79dSTomohiro Misono if (IS_ERR(root)) { 278023d0b79dSTomohiro Misono ret = PTR_ERR(root); 278123d0b79dSTomohiro Misono goto out; 278223d0b79dSTomohiro Misono } 278323d0b79dSTomohiro Misono 278423d0b79dSTomohiro Misono key.objectid = dirid; 278523d0b79dSTomohiro Misono key.type = BTRFS_INODE_REF_KEY; 278623d0b79dSTomohiro Misono key.offset = (u64)-1; 278723d0b79dSTomohiro Misono while (1) { 27880ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 27890ff40a91SMarcos Paulo de Souza if (ret < 0) 2790b8a49ae1SJosef Bacik goto out_put; 27910ff40a91SMarcos Paulo de Souza else if (ret > 0) { 279223d0b79dSTomohiro Misono ret = -ENOENT; 2793b8a49ae1SJosef Bacik goto out_put; 279423d0b79dSTomohiro Misono } 279523d0b79dSTomohiro Misono 279623d0b79dSTomohiro Misono leaf = path->nodes[0]; 279723d0b79dSTomohiro Misono slot = path->slots[0]; 279823d0b79dSTomohiro Misono 279923d0b79dSTomohiro Misono iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref); 280023d0b79dSTomohiro Misono len = btrfs_inode_ref_name_len(leaf, iref); 280123d0b79dSTomohiro Misono ptr -= len + 1; 280223d0b79dSTomohiro Misono total_len += len + 1; 280323d0b79dSTomohiro Misono if (ptr < args->path) { 280423d0b79dSTomohiro Misono ret = -ENAMETOOLONG; 2805b8a49ae1SJosef Bacik goto out_put; 280623d0b79dSTomohiro Misono } 280723d0b79dSTomohiro Misono 280823d0b79dSTomohiro Misono *(ptr + len) = '/'; 280923d0b79dSTomohiro Misono read_extent_buffer(leaf, ptr, 281023d0b79dSTomohiro Misono (unsigned long)(iref + 1), len); 281123d0b79dSTomohiro Misono 281223d0b79dSTomohiro Misono /* Check the read+exec permission of this directory */ 281323d0b79dSTomohiro Misono ret = btrfs_previous_item(root, path, dirid, 281423d0b79dSTomohiro Misono BTRFS_INODE_ITEM_KEY); 281523d0b79dSTomohiro Misono if (ret < 0) { 2816b8a49ae1SJosef Bacik goto out_put; 281723d0b79dSTomohiro Misono } else if (ret > 0) { 281823d0b79dSTomohiro Misono ret = -ENOENT; 2819b8a49ae1SJosef Bacik goto out_put; 282023d0b79dSTomohiro Misono } 282123d0b79dSTomohiro Misono 282223d0b79dSTomohiro Misono leaf = path->nodes[0]; 282323d0b79dSTomohiro Misono slot = path->slots[0]; 282423d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key2, slot); 282523d0b79dSTomohiro Misono if (key2.objectid != dirid) { 282623d0b79dSTomohiro Misono ret = -ENOENT; 2827b8a49ae1SJosef Bacik goto out_put; 282823d0b79dSTomohiro Misono } 282923d0b79dSTomohiro Misono 28300202e83fSDavid Sterba temp_inode = btrfs_iget(sb, key2.objectid, root); 28313ca57bd6SMisono Tomohiro if (IS_ERR(temp_inode)) { 28323ca57bd6SMisono Tomohiro ret = PTR_ERR(temp_inode); 2833b8a49ae1SJosef Bacik goto out_put; 28343ca57bd6SMisono Tomohiro } 28356623d9a0SChristian Brauner ret = inode_permission(mnt_userns, temp_inode, 283647291baaSChristian Brauner MAY_READ | MAY_EXEC); 283723d0b79dSTomohiro Misono iput(temp_inode); 283823d0b79dSTomohiro Misono if (ret) { 283923d0b79dSTomohiro Misono ret = -EACCES; 2840b8a49ae1SJosef Bacik goto out_put; 284123d0b79dSTomohiro Misono } 284223d0b79dSTomohiro Misono 284323d0b79dSTomohiro Misono if (key.offset == upper_limit.objectid) 284423d0b79dSTomohiro Misono break; 284523d0b79dSTomohiro Misono if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) { 284623d0b79dSTomohiro Misono ret = -EACCES; 2847b8a49ae1SJosef Bacik goto out_put; 284823d0b79dSTomohiro Misono } 284923d0b79dSTomohiro Misono 285023d0b79dSTomohiro Misono btrfs_release_path(path); 285123d0b79dSTomohiro Misono key.objectid = key.offset; 285223d0b79dSTomohiro Misono key.offset = (u64)-1; 285323d0b79dSTomohiro Misono dirid = key.objectid; 285423d0b79dSTomohiro Misono } 285523d0b79dSTomohiro Misono 285623d0b79dSTomohiro Misono memmove(args->path, ptr, total_len); 285723d0b79dSTomohiro Misono args->path[total_len] = '\0'; 285800246528SJosef Bacik btrfs_put_root(root); 2859b8a49ae1SJosef Bacik root = NULL; 286023d0b79dSTomohiro Misono btrfs_release_path(path); 286123d0b79dSTomohiro Misono } 286223d0b79dSTomohiro Misono 286323d0b79dSTomohiro Misono /* Get the bottom subvolume's name from ROOT_REF */ 286423d0b79dSTomohiro Misono key.objectid = treeid; 286523d0b79dSTomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 286623d0b79dSTomohiro Misono key.offset = args->treeid; 2867b8a49ae1SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 286823d0b79dSTomohiro Misono if (ret < 0) { 286923d0b79dSTomohiro Misono goto out; 287023d0b79dSTomohiro Misono } else if (ret > 0) { 287123d0b79dSTomohiro Misono ret = -ENOENT; 287223d0b79dSTomohiro Misono goto out; 287323d0b79dSTomohiro Misono } 287423d0b79dSTomohiro Misono 287523d0b79dSTomohiro Misono leaf = path->nodes[0]; 287623d0b79dSTomohiro Misono slot = path->slots[0]; 287723d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 287823d0b79dSTomohiro Misono 287923d0b79dSTomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot); 28803212fa14SJosef Bacik item_len = btrfs_item_size(leaf, slot); 288123d0b79dSTomohiro Misono /* Check if dirid in ROOT_REF corresponds to passed dirid */ 288223d0b79dSTomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 288323d0b79dSTomohiro Misono if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) { 288423d0b79dSTomohiro Misono ret = -EINVAL; 288523d0b79dSTomohiro Misono goto out; 288623d0b79dSTomohiro Misono } 288723d0b79dSTomohiro Misono 288823d0b79dSTomohiro Misono /* Copy subvolume's name */ 288923d0b79dSTomohiro Misono item_off += sizeof(struct btrfs_root_ref); 289023d0b79dSTomohiro Misono item_len -= sizeof(struct btrfs_root_ref); 289123d0b79dSTomohiro Misono read_extent_buffer(leaf, args->name, item_off, item_len); 289223d0b79dSTomohiro Misono args->name[item_len] = 0; 289323d0b79dSTomohiro Misono 2894b8a49ae1SJosef Bacik out_put: 289500246528SJosef Bacik btrfs_put_root(root); 289623d0b79dSTomohiro Misono out: 289723d0b79dSTomohiro Misono btrfs_free_path(path); 289823d0b79dSTomohiro Misono return ret; 289923d0b79dSTomohiro Misono } 290023d0b79dSTomohiro Misono 2901dc408ccdSSahil Kang static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root, 2902ac8e9819SChris Mason void __user *argp) 2903ac8e9819SChris Mason { 2904ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 290501b810b8SDavid Sterba int ret = 0; 2906ac8e9819SChris Mason 29072354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 29082354d08fSJulia Lawall if (IS_ERR(args)) 29092354d08fSJulia Lawall return PTR_ERR(args); 2910c2b96929SDan Carpenter 291101b810b8SDavid Sterba /* 291201b810b8SDavid Sterba * Unprivileged query to obtain the containing subvolume root id. The 291301b810b8SDavid Sterba * path is reset so it's consistent with btrfs_search_path_in_tree. 291401b810b8SDavid Sterba */ 29151b53ac4dSChris Mason if (args->treeid == 0) 2916dc408ccdSSahil Kang args->treeid = root->root_key.objectid; 29171b53ac4dSChris Mason 291801b810b8SDavid Sterba if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) { 291901b810b8SDavid Sterba args->name[0] = 0; 292001b810b8SDavid Sterba goto out; 292101b810b8SDavid Sterba } 292201b810b8SDavid Sterba 292301b810b8SDavid Sterba if (!capable(CAP_SYS_ADMIN)) { 292401b810b8SDavid Sterba ret = -EPERM; 292501b810b8SDavid Sterba goto out; 292601b810b8SDavid Sterba } 292701b810b8SDavid Sterba 2928dc408ccdSSahil Kang ret = btrfs_search_path_in_tree(root->fs_info, 2929ac8e9819SChris Mason args->treeid, args->objectid, 2930ac8e9819SChris Mason args->name); 2931ac8e9819SChris Mason 293201b810b8SDavid Sterba out: 2933ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2934ac8e9819SChris Mason ret = -EFAULT; 2935ac8e9819SChris Mason 2936ac8e9819SChris Mason kfree(args); 293798d377a0STARUISI Hiroaki return ret; 293898d377a0STARUISI Hiroaki } 293998d377a0STARUISI Hiroaki 294023d0b79dSTomohiro Misono /* 294123d0b79dSTomohiro Misono * Version of ino_lookup ioctl (unprivileged) 294223d0b79dSTomohiro Misono * 294323d0b79dSTomohiro Misono * The main differences from ino_lookup ioctl are: 294423d0b79dSTomohiro Misono * 294523d0b79dSTomohiro Misono * 1. Read + Exec permission will be checked using inode_permission() during 294623d0b79dSTomohiro Misono * path construction. -EACCES will be returned in case of failure. 294723d0b79dSTomohiro Misono * 2. Path construction will be stopped at the inode number which corresponds 294823d0b79dSTomohiro Misono * to the fd with which this ioctl is called. If constructed path does not 294923d0b79dSTomohiro Misono * exist under fd's inode, -EACCES will be returned. 295023d0b79dSTomohiro Misono * 3. The name of bottom subvolume is also searched and filled. 295123d0b79dSTomohiro Misono */ 295223d0b79dSTomohiro Misono static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp) 295323d0b79dSTomohiro Misono { 295423d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args; 295523d0b79dSTomohiro Misono struct inode *inode; 295623d0b79dSTomohiro Misono int ret; 295723d0b79dSTomohiro Misono 295823d0b79dSTomohiro Misono args = memdup_user(argp, sizeof(*args)); 295923d0b79dSTomohiro Misono if (IS_ERR(args)) 296023d0b79dSTomohiro Misono return PTR_ERR(args); 296123d0b79dSTomohiro Misono 296223d0b79dSTomohiro Misono inode = file_inode(file); 296323d0b79dSTomohiro Misono 296423d0b79dSTomohiro Misono if (args->dirid == BTRFS_FIRST_FREE_OBJECTID && 296523d0b79dSTomohiro Misono BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) { 296623d0b79dSTomohiro Misono /* 296723d0b79dSTomohiro Misono * The subvolume does not exist under fd with which this is 296823d0b79dSTomohiro Misono * called 296923d0b79dSTomohiro Misono */ 297023d0b79dSTomohiro Misono kfree(args); 297123d0b79dSTomohiro Misono return -EACCES; 297223d0b79dSTomohiro Misono } 297323d0b79dSTomohiro Misono 29746623d9a0SChristian Brauner ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args); 297523d0b79dSTomohiro Misono 297623d0b79dSTomohiro Misono if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 297723d0b79dSTomohiro Misono ret = -EFAULT; 297823d0b79dSTomohiro Misono 297923d0b79dSTomohiro Misono kfree(args); 298023d0b79dSTomohiro Misono return ret; 298123d0b79dSTomohiro Misono } 298223d0b79dSTomohiro Misono 2983b64ec075STomohiro Misono /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */ 29849ad12305SSahil Kang static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp) 2985b64ec075STomohiro Misono { 2986b64ec075STomohiro Misono struct btrfs_ioctl_get_subvol_info_args *subvol_info; 2987b64ec075STomohiro Misono struct btrfs_fs_info *fs_info; 2988b64ec075STomohiro Misono struct btrfs_root *root; 2989b64ec075STomohiro Misono struct btrfs_path *path; 2990b64ec075STomohiro Misono struct btrfs_key key; 2991b64ec075STomohiro Misono struct btrfs_root_item *root_item; 2992b64ec075STomohiro Misono struct btrfs_root_ref *rref; 2993b64ec075STomohiro Misono struct extent_buffer *leaf; 2994b64ec075STomohiro Misono unsigned long item_off; 2995b64ec075STomohiro Misono unsigned long item_len; 2996b64ec075STomohiro Misono int slot; 2997b64ec075STomohiro Misono int ret = 0; 2998b64ec075STomohiro Misono 2999b64ec075STomohiro Misono path = btrfs_alloc_path(); 3000b64ec075STomohiro Misono if (!path) 3001b64ec075STomohiro Misono return -ENOMEM; 3002b64ec075STomohiro Misono 3003b64ec075STomohiro Misono subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL); 3004b64ec075STomohiro Misono if (!subvol_info) { 3005b64ec075STomohiro Misono btrfs_free_path(path); 3006b64ec075STomohiro Misono return -ENOMEM; 3007b64ec075STomohiro Misono } 3008b64ec075STomohiro Misono 3009b64ec075STomohiro Misono fs_info = BTRFS_I(inode)->root->fs_info; 3010b64ec075STomohiro Misono 3011b64ec075STomohiro Misono /* Get root_item of inode's subvolume */ 3012b64ec075STomohiro Misono key.objectid = BTRFS_I(inode)->root->root_key.objectid; 301356e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, key.objectid, true); 3014b64ec075STomohiro Misono if (IS_ERR(root)) { 3015b64ec075STomohiro Misono ret = PTR_ERR(root); 301604734e84SJosef Bacik goto out_free; 301704734e84SJosef Bacik } 3018b64ec075STomohiro Misono root_item = &root->root_item; 3019b64ec075STomohiro Misono 3020b64ec075STomohiro Misono subvol_info->treeid = key.objectid; 3021b64ec075STomohiro Misono 3022b64ec075STomohiro Misono subvol_info->generation = btrfs_root_generation(root_item); 3023b64ec075STomohiro Misono subvol_info->flags = btrfs_root_flags(root_item); 3024b64ec075STomohiro Misono 3025b64ec075STomohiro Misono memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE); 3026b64ec075STomohiro Misono memcpy(subvol_info->parent_uuid, root_item->parent_uuid, 3027b64ec075STomohiro Misono BTRFS_UUID_SIZE); 3028b64ec075STomohiro Misono memcpy(subvol_info->received_uuid, root_item->received_uuid, 3029b64ec075STomohiro Misono BTRFS_UUID_SIZE); 3030b64ec075STomohiro Misono 3031b64ec075STomohiro Misono subvol_info->ctransid = btrfs_root_ctransid(root_item); 3032b64ec075STomohiro Misono subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime); 3033b64ec075STomohiro Misono subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime); 3034b64ec075STomohiro Misono 3035b64ec075STomohiro Misono subvol_info->otransid = btrfs_root_otransid(root_item); 3036b64ec075STomohiro Misono subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime); 3037b64ec075STomohiro Misono subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime); 3038b64ec075STomohiro Misono 3039b64ec075STomohiro Misono subvol_info->stransid = btrfs_root_stransid(root_item); 3040b64ec075STomohiro Misono subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime); 3041b64ec075STomohiro Misono subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime); 3042b64ec075STomohiro Misono 3043b64ec075STomohiro Misono subvol_info->rtransid = btrfs_root_rtransid(root_item); 3044b64ec075STomohiro Misono subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime); 3045b64ec075STomohiro Misono subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime); 3046b64ec075STomohiro Misono 3047b64ec075STomohiro Misono if (key.objectid != BTRFS_FS_TREE_OBJECTID) { 3048b64ec075STomohiro Misono /* Search root tree for ROOT_BACKREF of this subvolume */ 3049b64ec075STomohiro Misono key.type = BTRFS_ROOT_BACKREF_KEY; 3050b64ec075STomohiro Misono key.offset = 0; 305104734e84SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 3052b64ec075STomohiro Misono if (ret < 0) { 3053b64ec075STomohiro Misono goto out; 3054b64ec075STomohiro Misono } else if (path->slots[0] >= 3055b64ec075STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 305604734e84SJosef Bacik ret = btrfs_next_leaf(fs_info->tree_root, path); 3057b64ec075STomohiro Misono if (ret < 0) { 3058b64ec075STomohiro Misono goto out; 3059b64ec075STomohiro Misono } else if (ret > 0) { 3060b64ec075STomohiro Misono ret = -EUCLEAN; 3061b64ec075STomohiro Misono goto out; 3062b64ec075STomohiro Misono } 3063b64ec075STomohiro Misono } 3064b64ec075STomohiro Misono 3065b64ec075STomohiro Misono leaf = path->nodes[0]; 3066b64ec075STomohiro Misono slot = path->slots[0]; 3067b64ec075STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 3068b64ec075STomohiro Misono if (key.objectid == subvol_info->treeid && 3069b64ec075STomohiro Misono key.type == BTRFS_ROOT_BACKREF_KEY) { 3070b64ec075STomohiro Misono subvol_info->parent_id = key.offset; 3071b64ec075STomohiro Misono 3072b64ec075STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 3073b64ec075STomohiro Misono subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref); 3074b64ec075STomohiro Misono 3075b64ec075STomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot) 3076b64ec075STomohiro Misono + sizeof(struct btrfs_root_ref); 30773212fa14SJosef Bacik item_len = btrfs_item_size(leaf, slot) 3078b64ec075STomohiro Misono - sizeof(struct btrfs_root_ref); 3079b64ec075STomohiro Misono read_extent_buffer(leaf, subvol_info->name, 3080b64ec075STomohiro Misono item_off, item_len); 3081b64ec075STomohiro Misono } else { 3082b64ec075STomohiro Misono ret = -ENOENT; 3083b64ec075STomohiro Misono goto out; 3084b64ec075STomohiro Misono } 3085b64ec075STomohiro Misono } 3086b64ec075STomohiro Misono 3087b64ec075STomohiro Misono if (copy_to_user(argp, subvol_info, sizeof(*subvol_info))) 3088b64ec075STomohiro Misono ret = -EFAULT; 3089b64ec075STomohiro Misono 3090b64ec075STomohiro Misono out: 309100246528SJosef Bacik btrfs_put_root(root); 309204734e84SJosef Bacik out_free: 3093b64ec075STomohiro Misono btrfs_free_path(path); 3094b091f7feSWaiman Long kfree(subvol_info); 3095b64ec075STomohiro Misono return ret; 3096b64ec075STomohiro Misono } 3097b64ec075STomohiro Misono 309842e4b520STomohiro Misono /* 309942e4b520STomohiro Misono * Return ROOT_REF information of the subvolume containing this inode 310042e4b520STomohiro Misono * except the subvolume name. 310142e4b520STomohiro Misono */ 31029ad12305SSahil Kang static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root, 3103dc408ccdSSahil Kang void __user *argp) 310442e4b520STomohiro Misono { 310542e4b520STomohiro Misono struct btrfs_ioctl_get_subvol_rootref_args *rootrefs; 310642e4b520STomohiro Misono struct btrfs_root_ref *rref; 310742e4b520STomohiro Misono struct btrfs_path *path; 310842e4b520STomohiro Misono struct btrfs_key key; 310942e4b520STomohiro Misono struct extent_buffer *leaf; 311042e4b520STomohiro Misono u64 objectid; 311142e4b520STomohiro Misono int slot; 311242e4b520STomohiro Misono int ret; 311342e4b520STomohiro Misono u8 found; 311442e4b520STomohiro Misono 311542e4b520STomohiro Misono path = btrfs_alloc_path(); 311642e4b520STomohiro Misono if (!path) 311742e4b520STomohiro Misono return -ENOMEM; 311842e4b520STomohiro Misono 311942e4b520STomohiro Misono rootrefs = memdup_user(argp, sizeof(*rootrefs)); 312042e4b520STomohiro Misono if (IS_ERR(rootrefs)) { 312142e4b520STomohiro Misono btrfs_free_path(path); 312242e4b520STomohiro Misono return PTR_ERR(rootrefs); 312342e4b520STomohiro Misono } 312442e4b520STomohiro Misono 31259ad12305SSahil Kang objectid = root->root_key.objectid; 312642e4b520STomohiro Misono key.objectid = objectid; 312742e4b520STomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 312842e4b520STomohiro Misono key.offset = rootrefs->min_treeid; 312942e4b520STomohiro Misono found = 0; 313042e4b520STomohiro Misono 31319ad12305SSahil Kang root = root->fs_info->tree_root; 313242e4b520STomohiro Misono ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 313342e4b520STomohiro Misono if (ret < 0) { 313442e4b520STomohiro Misono goto out; 313542e4b520STomohiro Misono } else if (path->slots[0] >= 313642e4b520STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 313742e4b520STomohiro Misono ret = btrfs_next_leaf(root, path); 313842e4b520STomohiro Misono if (ret < 0) { 313942e4b520STomohiro Misono goto out; 314042e4b520STomohiro Misono } else if (ret > 0) { 314142e4b520STomohiro Misono ret = -EUCLEAN; 314242e4b520STomohiro Misono goto out; 314342e4b520STomohiro Misono } 314442e4b520STomohiro Misono } 314542e4b520STomohiro Misono while (1) { 314642e4b520STomohiro Misono leaf = path->nodes[0]; 314742e4b520STomohiro Misono slot = path->slots[0]; 314842e4b520STomohiro Misono 314942e4b520STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 315042e4b520STomohiro Misono if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) { 315142e4b520STomohiro Misono ret = 0; 315242e4b520STomohiro Misono goto out; 315342e4b520STomohiro Misono } 315442e4b520STomohiro Misono 315542e4b520STomohiro Misono if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) { 315642e4b520STomohiro Misono ret = -EOVERFLOW; 315742e4b520STomohiro Misono goto out; 315842e4b520STomohiro Misono } 315942e4b520STomohiro Misono 316042e4b520STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 316142e4b520STomohiro Misono rootrefs->rootref[found].treeid = key.offset; 316242e4b520STomohiro Misono rootrefs->rootref[found].dirid = 316342e4b520STomohiro Misono btrfs_root_ref_dirid(leaf, rref); 316442e4b520STomohiro Misono found++; 316542e4b520STomohiro Misono 316642e4b520STomohiro Misono ret = btrfs_next_item(root, path); 316742e4b520STomohiro Misono if (ret < 0) { 316842e4b520STomohiro Misono goto out; 316942e4b520STomohiro Misono } else if (ret > 0) { 317042e4b520STomohiro Misono ret = -EUCLEAN; 317142e4b520STomohiro Misono goto out; 317242e4b520STomohiro Misono } 317342e4b520STomohiro Misono } 317442e4b520STomohiro Misono 317542e4b520STomohiro Misono out: 317642e4b520STomohiro Misono if (!ret || ret == -EOVERFLOW) { 317742e4b520STomohiro Misono rootrefs->num_items = found; 317842e4b520STomohiro Misono /* update min_treeid for next search */ 317942e4b520STomohiro Misono if (found) 318042e4b520STomohiro Misono rootrefs->min_treeid = 318142e4b520STomohiro Misono rootrefs->rootref[found - 1].treeid + 1; 318242e4b520STomohiro Misono if (copy_to_user(argp, rootrefs, sizeof(*rootrefs))) 318342e4b520STomohiro Misono ret = -EFAULT; 318442e4b520STomohiro Misono } 318542e4b520STomohiro Misono 318642e4b520STomohiro Misono kfree(rootrefs); 318742e4b520STomohiro Misono btrfs_free_path(path); 318842e4b520STomohiro Misono 318942e4b520STomohiro Misono return ret; 319042e4b520STomohiro Misono } 319142e4b520STomohiro Misono 319276dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 3193949964c9SMarcos Paulo de Souza void __user *arg, 3194949964c9SMarcos Paulo de Souza bool destroy_v2) 319576dda93cSYan, Zheng { 319654563d41SAl Viro struct dentry *parent = file->f_path.dentry; 31970b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb); 319876dda93cSYan, Zheng struct dentry *dentry; 31992b0143b5SDavid Howells struct inode *dir = d_inode(parent); 320076dda93cSYan, Zheng struct inode *inode; 320176dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 320276dda93cSYan, Zheng struct btrfs_root *dest = NULL; 3203949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args *vol_args = NULL; 3204949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL; 3205c4ed533bSChristian Brauner struct user_namespace *mnt_userns = file_mnt_user_ns(file); 3206949964c9SMarcos Paulo de Souza char *subvol_name, *subvol_name_ptr = NULL; 3207949964c9SMarcos Paulo de Souza int subvol_namelen; 320876dda93cSYan, Zheng int err = 0; 3209949964c9SMarcos Paulo de Souza bool destroy_parent = false; 321076dda93cSYan, Zheng 3211813febdbSJosef Bacik /* We don't support snapshots with extent tree v2 yet. */ 3212813febdbSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3213813febdbSJosef Bacik btrfs_err(fs_info, 3214813febdbSJosef Bacik "extent tree v2 doesn't support snapshot deletion yet"); 3215813febdbSJosef Bacik return -EOPNOTSUPP; 3216813febdbSJosef Bacik } 3217813febdbSJosef Bacik 3218949964c9SMarcos Paulo de Souza if (destroy_v2) { 3219949964c9SMarcos Paulo de Souza vol_args2 = memdup_user(arg, sizeof(*vol_args2)); 3220949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args2)) 3221949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args2); 3222325c50e3SJeff Mahoney 3223949964c9SMarcos Paulo de Souza if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) { 3224949964c9SMarcos Paulo de Souza err = -EOPNOTSUPP; 3225949964c9SMarcos Paulo de Souza goto out; 3226949964c9SMarcos Paulo de Souza } 322776dda93cSYan, Zheng 3228949964c9SMarcos Paulo de Souza /* 3229949964c9SMarcos Paulo de Souza * If SPEC_BY_ID is not set, we are looking for the subvolume by 3230949964c9SMarcos Paulo de Souza * name, same as v1 currently does. 3231949964c9SMarcos Paulo de Souza */ 3232949964c9SMarcos Paulo de Souza if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) { 3233949964c9SMarcos Paulo de Souza vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0; 3234949964c9SMarcos Paulo de Souza subvol_name = vol_args2->name; 3235949964c9SMarcos Paulo de Souza 3236949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 3237949964c9SMarcos Paulo de Souza if (err) 3238949964c9SMarcos Paulo de Souza goto out; 3239949964c9SMarcos Paulo de Souza } else { 3240aabb34e7SChristian Brauner struct inode *old_dir; 3241c4ed533bSChristian Brauner 3242949964c9SMarcos Paulo de Souza if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) { 324376dda93cSYan, Zheng err = -EINVAL; 324476dda93cSYan, Zheng goto out; 324576dda93cSYan, Zheng } 324676dda93cSYan, Zheng 3247a561be71SAl Viro err = mnt_want_write_file(file); 324876dda93cSYan, Zheng if (err) 324976dda93cSYan, Zheng goto out; 325076dda93cSYan, Zheng 3251949964c9SMarcos Paulo de Souza dentry = btrfs_get_dentry(fs_info->sb, 3252949964c9SMarcos Paulo de Souza BTRFS_FIRST_FREE_OBJECTID, 3253949964c9SMarcos Paulo de Souza vol_args2->subvolid, 0, 0); 3254949964c9SMarcos Paulo de Souza if (IS_ERR(dentry)) { 3255949964c9SMarcos Paulo de Souza err = PTR_ERR(dentry); 3256949964c9SMarcos Paulo de Souza goto out_drop_write; 3257949964c9SMarcos Paulo de Souza } 3258949964c9SMarcos Paulo de Souza 3259949964c9SMarcos Paulo de Souza /* 3260949964c9SMarcos Paulo de Souza * Change the default parent since the subvolume being 3261949964c9SMarcos Paulo de Souza * deleted can be outside of the current mount point. 3262949964c9SMarcos Paulo de Souza */ 3263949964c9SMarcos Paulo de Souza parent = btrfs_get_parent(dentry); 3264949964c9SMarcos Paulo de Souza 3265949964c9SMarcos Paulo de Souza /* 3266949964c9SMarcos Paulo de Souza * At this point dentry->d_name can point to '/' if the 3267949964c9SMarcos Paulo de Souza * subvolume we want to destroy is outsite of the 3268949964c9SMarcos Paulo de Souza * current mount point, so we need to release the 3269949964c9SMarcos Paulo de Souza * current dentry and execute the lookup to return a new 3270949964c9SMarcos Paulo de Souza * one with ->d_name pointing to the 3271949964c9SMarcos Paulo de Souza * <mount point>/subvol_name. 3272949964c9SMarcos Paulo de Souza */ 3273949964c9SMarcos Paulo de Souza dput(dentry); 3274949964c9SMarcos Paulo de Souza if (IS_ERR(parent)) { 3275949964c9SMarcos Paulo de Souza err = PTR_ERR(parent); 3276949964c9SMarcos Paulo de Souza goto out_drop_write; 3277949964c9SMarcos Paulo de Souza } 3278aabb34e7SChristian Brauner old_dir = dir; 3279949964c9SMarcos Paulo de Souza dir = d_inode(parent); 3280949964c9SMarcos Paulo de Souza 3281949964c9SMarcos Paulo de Souza /* 3282949964c9SMarcos Paulo de Souza * If v2 was used with SPEC_BY_ID, a new parent was 3283949964c9SMarcos Paulo de Souza * allocated since the subvolume can be outside of the 3284949964c9SMarcos Paulo de Souza * current mount point. Later on we need to release this 3285949964c9SMarcos Paulo de Souza * new parent dentry. 3286949964c9SMarcos Paulo de Souza */ 3287949964c9SMarcos Paulo de Souza destroy_parent = true; 3288949964c9SMarcos Paulo de Souza 3289aabb34e7SChristian Brauner /* 3290aabb34e7SChristian Brauner * On idmapped mounts, deletion via subvolid is 3291aabb34e7SChristian Brauner * restricted to subvolumes that are immediate 3292aabb34e7SChristian Brauner * ancestors of the inode referenced by the file 3293aabb34e7SChristian Brauner * descriptor in the ioctl. Otherwise the idmapping 3294aabb34e7SChristian Brauner * could potentially be abused to delete subvolumes 3295aabb34e7SChristian Brauner * anywhere in the filesystem the user wouldn't be able 3296aabb34e7SChristian Brauner * to delete without an idmapped mount. 3297aabb34e7SChristian Brauner */ 3298aabb34e7SChristian Brauner if (old_dir != dir && mnt_userns != &init_user_ns) { 3299aabb34e7SChristian Brauner err = -EOPNOTSUPP; 3300aabb34e7SChristian Brauner goto free_parent; 3301aabb34e7SChristian Brauner } 3302aabb34e7SChristian Brauner 3303949964c9SMarcos Paulo de Souza subvol_name_ptr = btrfs_get_subvol_name_from_objectid( 3304949964c9SMarcos Paulo de Souza fs_info, vol_args2->subvolid); 3305949964c9SMarcos Paulo de Souza if (IS_ERR(subvol_name_ptr)) { 3306949964c9SMarcos Paulo de Souza err = PTR_ERR(subvol_name_ptr); 3307949964c9SMarcos Paulo de Souza goto free_parent; 3308949964c9SMarcos Paulo de Souza } 33091a9fd417SDavid Sterba /* subvol_name_ptr is already nul terminated */ 3310949964c9SMarcos Paulo de Souza subvol_name = (char *)kbasename(subvol_name_ptr); 3311949964c9SMarcos Paulo de Souza } 3312949964c9SMarcos Paulo de Souza } else { 3313949964c9SMarcos Paulo de Souza vol_args = memdup_user(arg, sizeof(*vol_args)); 3314949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args)) 3315949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args); 3316949964c9SMarcos Paulo de Souza 3317949964c9SMarcos Paulo de Souza vol_args->name[BTRFS_PATH_NAME_MAX] = 0; 3318949964c9SMarcos Paulo de Souza subvol_name = vol_args->name; 3319949964c9SMarcos Paulo de Souza 3320949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 3321949964c9SMarcos Paulo de Souza if (err) 3322949964c9SMarcos Paulo de Souza goto out; 3323949964c9SMarcos Paulo de Souza } 3324949964c9SMarcos Paulo de Souza 3325949964c9SMarcos Paulo de Souza subvol_namelen = strlen(subvol_name); 3326949964c9SMarcos Paulo de Souza 3327949964c9SMarcos Paulo de Souza if (strchr(subvol_name, '/') || 3328949964c9SMarcos Paulo de Souza strncmp(subvol_name, "..", subvol_namelen) == 0) { 3329949964c9SMarcos Paulo de Souza err = -EINVAL; 3330949964c9SMarcos Paulo de Souza goto free_subvol_name; 3331949964c9SMarcos Paulo de Souza } 3332949964c9SMarcos Paulo de Souza 3333949964c9SMarcos Paulo de Souza if (!S_ISDIR(dir->i_mode)) { 3334949964c9SMarcos Paulo de Souza err = -ENOTDIR; 3335949964c9SMarcos Paulo de Souza goto free_subvol_name; 3336949964c9SMarcos Paulo de Souza } 3337521e0546SDavid Sterba 333800235411SAl Viro err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 333900235411SAl Viro if (err == -EINTR) 3340949964c9SMarcos Paulo de Souza goto free_subvol_name; 3341c4ed533bSChristian Brauner dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen); 334276dda93cSYan, Zheng if (IS_ERR(dentry)) { 334376dda93cSYan, Zheng err = PTR_ERR(dentry); 334476dda93cSYan, Zheng goto out_unlock_dir; 334576dda93cSYan, Zheng } 334676dda93cSYan, Zheng 33472b0143b5SDavid Howells if (d_really_is_negative(dentry)) { 334876dda93cSYan, Zheng err = -ENOENT; 334976dda93cSYan, Zheng goto out_dput; 335076dda93cSYan, Zheng } 335176dda93cSYan, Zheng 33522b0143b5SDavid Howells inode = d_inode(dentry); 33534260f7c7SSage Weil dest = BTRFS_I(inode)->root; 33544260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)) { 33554260f7c7SSage Weil /* 33564260f7c7SSage Weil * Regular user. Only allow this with a special mount 33574260f7c7SSage Weil * option, when the user has write+exec access to the 33584260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 33594260f7c7SSage Weil * allowed. 33604260f7c7SSage Weil * 33614260f7c7SSage Weil * Note that this is _not_ check that the subvol is 33624260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 33634260f7c7SSage Weil * otherwise be able to delete. 33644260f7c7SSage Weil * 33654260f7c7SSage Weil * Users who want to delete empty subvols should try 33664260f7c7SSage Weil * rmdir(2). 33674260f7c7SSage Weil */ 33684260f7c7SSage Weil err = -EPERM; 33690b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED)) 33704260f7c7SSage Weil goto out_dput; 33714260f7c7SSage Weil 33724260f7c7SSage Weil /* 33734260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 33744260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 33754260f7c7SSage Weil * must be called on the dentry referencing the root 33764260f7c7SSage Weil * of the subvol, not a random directory contained 33774260f7c7SSage Weil * within it. 33784260f7c7SSage Weil */ 33794260f7c7SSage Weil err = -EINVAL; 33804260f7c7SSage Weil if (root == dest) 33814260f7c7SSage Weil goto out_dput; 33824260f7c7SSage Weil 3383c4ed533bSChristian Brauner err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC); 33844260f7c7SSage Weil if (err) 33854260f7c7SSage Weil goto out_dput; 33865c39da5bSMiao Xie } 33874260f7c7SSage Weil 33885c39da5bSMiao Xie /* check if subvolume may be deleted by a user */ 3389c4ed533bSChristian Brauner err = btrfs_may_delete(mnt_userns, dir, dentry, 1); 33904260f7c7SSage Weil if (err) 33914260f7c7SSage Weil goto out_dput; 33924260f7c7SSage Weil 33934a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 339476dda93cSYan, Zheng err = -EINVAL; 339576dda93cSYan, Zheng goto out_dput; 339676dda93cSYan, Zheng } 339776dda93cSYan, Zheng 339864708539SJosef Bacik btrfs_inode_lock(inode, 0); 3399f60a2364SMisono Tomohiro err = btrfs_delete_subvolume(dir, dentry); 340064708539SJosef Bacik btrfs_inode_unlock(inode, 0); 3401a37d9a17SAmir Goldstein if (!err) 3402a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 3403fa6ac876SLiu Bo 340476dda93cSYan, Zheng out_dput: 340576dda93cSYan, Zheng dput(dentry); 340676dda93cSYan, Zheng out_unlock_dir: 340764708539SJosef Bacik btrfs_inode_unlock(dir, 0); 3408949964c9SMarcos Paulo de Souza free_subvol_name: 3409949964c9SMarcos Paulo de Souza kfree(subvol_name_ptr); 3410949964c9SMarcos Paulo de Souza free_parent: 3411949964c9SMarcos Paulo de Souza if (destroy_parent) 3412949964c9SMarcos Paulo de Souza dput(parent); 341300235411SAl Viro out_drop_write: 34142a79f17eSAl Viro mnt_drop_write_file(file); 341576dda93cSYan, Zheng out: 3416949964c9SMarcos Paulo de Souza kfree(vol_args2); 341776dda93cSYan, Zheng kfree(vol_args); 341876dda93cSYan, Zheng return err; 341976dda93cSYan, Zheng } 342076dda93cSYan, Zheng 34211e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 3422f46b5a66SChristoph Hellwig { 3423496ad9aaSAl Viro struct inode *inode = file_inode(file); 3424f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3425c853a578SGoldwyn Rodrigues struct btrfs_ioctl_defrag_range_args range = {0}; 3426c146afadSYan Zheng int ret; 3427c146afadSYan Zheng 342825122d15SIlya Dryomov ret = mnt_want_write_file(file); 342925122d15SIlya Dryomov if (ret) 343025122d15SIlya Dryomov return ret; 3431b83cc969SLi Zefan 343225122d15SIlya Dryomov if (btrfs_root_readonly(root)) { 343325122d15SIlya Dryomov ret = -EROFS; 343425122d15SIlya Dryomov goto out; 34355ac00addSStefan Behrens } 3436f46b5a66SChristoph Hellwig 3437f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3438f46b5a66SChristoph Hellwig case S_IFDIR: 3439e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 3440e441d54dSChris Mason ret = -EPERM; 3441e441d54dSChris Mason goto out; 3442e441d54dSChris Mason } 3443de78b51aSEric Sandeen ret = btrfs_defrag_root(root); 3444f46b5a66SChristoph Hellwig break; 3445f46b5a66SChristoph Hellwig case S_IFREG: 3446616d374eSAdam Borowski /* 3447616d374eSAdam Borowski * Note that this does not check the file descriptor for write 3448616d374eSAdam Borowski * access. This prevents defragmenting executables that are 3449616d374eSAdam Borowski * running and allows defrag on files open in read-only mode. 3450616d374eSAdam Borowski */ 3451616d374eSAdam Borowski if (!capable(CAP_SYS_ADMIN) && 345247291baaSChristian Brauner inode_permission(&init_user_ns, inode, MAY_WRITE)) { 3453616d374eSAdam Borowski ret = -EPERM; 3454e441d54dSChris Mason goto out; 3455e441d54dSChris Mason } 34561e701a32SChris Mason 34571e701a32SChris Mason if (argp) { 3458c853a578SGoldwyn Rodrigues if (copy_from_user(&range, argp, sizeof(range))) { 34591e701a32SChris Mason ret = -EFAULT; 3460683be16eSDan Carpenter goto out; 34611e701a32SChris Mason } 34621e701a32SChris Mason /* compression requires us to start the IO */ 3463c853a578SGoldwyn Rodrigues if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 3464c853a578SGoldwyn Rodrigues range.flags |= BTRFS_DEFRAG_RANGE_START_IO; 3465c853a578SGoldwyn Rodrigues range.extent_thresh = (u32)-1; 34661e701a32SChris Mason } 34671e701a32SChris Mason } else { 34681e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 3469c853a578SGoldwyn Rodrigues range.len = (u64)-1; 34701e701a32SChris Mason } 34711ccc2e8aSQu Wenruo ret = btrfs_defrag_file(file_inode(file), &file->f_ra, 3472c853a578SGoldwyn Rodrigues &range, BTRFS_OLDEST_GENERATION, 0); 34734cb5300bSChris Mason if (ret > 0) 34744cb5300bSChris Mason ret = 0; 3475f46b5a66SChristoph Hellwig break; 34768929ecfaSYan, Zheng default: 34778929ecfaSYan, Zheng ret = -EINVAL; 3478f46b5a66SChristoph Hellwig } 3479e441d54dSChris Mason out: 348025122d15SIlya Dryomov mnt_drop_write_file(file); 3481e441d54dSChris Mason return ret; 3482f46b5a66SChristoph Hellwig } 3483f46b5a66SChristoph Hellwig 34842ff7e61eSJeff Mahoney static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg) 3485f46b5a66SChristoph Hellwig { 3486f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 3487a174c0a2SNikolay Borisov bool restore_op = false; 3488f46b5a66SChristoph Hellwig int ret; 3489f46b5a66SChristoph Hellwig 3490e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3491e441d54dSChris Mason return -EPERM; 3492e441d54dSChris Mason 3493914a519bSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3494914a519bSJosef Bacik btrfs_err(fs_info, "device add not supported on extent tree v2 yet"); 3495914a519bSJosef Bacik return -EINVAL; 3496914a519bSJosef Bacik } 3497914a519bSJosef Bacik 3498a174c0a2SNikolay Borisov if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) { 3499a174c0a2SNikolay Borisov if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD)) 3500e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3501c9e9f97bSIlya Dryomov 3502a174c0a2SNikolay Borisov /* 3503a174c0a2SNikolay Borisov * We can do the device add because we have a paused balanced, 3504a174c0a2SNikolay Borisov * change the exclusive op type and remember we should bring 3505a174c0a2SNikolay Borisov * back the paused balance 3506a174c0a2SNikolay Borisov */ 3507a174c0a2SNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD; 3508a174c0a2SNikolay Borisov btrfs_exclop_start_unlock(fs_info); 3509a174c0a2SNikolay Borisov restore_op = true; 3510a174c0a2SNikolay Borisov } 3511a174c0a2SNikolay Borisov 3512dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3513c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 3514c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 3515c9e9f97bSIlya Dryomov goto out; 3516c9e9f97bSIlya Dryomov } 3517f46b5a66SChristoph Hellwig 35185516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 35192ff7e61eSJeff Mahoney ret = btrfs_init_new_device(fs_info, vol_args->name); 3520f46b5a66SChristoph Hellwig 352143d20761SAnand Jain if (!ret) 35220b246afaSJeff Mahoney btrfs_info(fs_info, "disk added %s", vol_args->name); 352343d20761SAnand Jain 3524f46b5a66SChristoph Hellwig kfree(vol_args); 3525c9e9f97bSIlya Dryomov out: 3526a174c0a2SNikolay Borisov if (restore_op) 3527a174c0a2SNikolay Borisov btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 3528a174c0a2SNikolay Borisov else 3529c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3530f46b5a66SChristoph Hellwig return ret; 3531f46b5a66SChristoph Hellwig } 3532f46b5a66SChristoph Hellwig 35336b526ed7SAnand Jain static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg) 3534f46b5a66SChristoph Hellwig { 35351a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 35360b246afaSJeff Mahoney struct inode *inode = file_inode(file); 35370b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 35386b526ed7SAnand Jain struct btrfs_ioctl_vol_args_v2 *vol_args; 35393fa421deSJosef Bacik struct block_device *bdev = NULL; 35403fa421deSJosef Bacik fmode_t mode; 3541f46b5a66SChristoph Hellwig int ret; 354267ae34b6SDavid Sterba bool cancel = false; 3543f46b5a66SChristoph Hellwig 3544e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3545e441d54dSChris Mason return -EPERM; 3546e441d54dSChris Mason 3547dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3548d815b3f2SDan Carpenter if (IS_ERR(vol_args)) 3549d815b3f2SDan Carpenter return PTR_ERR(vol_args); 3550f46b5a66SChristoph Hellwig 3551748449cdSDavid Sterba if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) { 3552fd4e994bSOmar Sandoval ret = -EOPNOTSUPP; 3553fd4e994bSOmar Sandoval goto out; 3554fd4e994bSOmar Sandoval } 35551a15eb72SJosef Bacik 35566b526ed7SAnand Jain vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 35571a15eb72SJosef Bacik if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) { 35581a15eb72SJosef Bacik args.devid = vol_args->devid; 35591a15eb72SJosef Bacik } else if (!strcmp("cancel", vol_args->name)) { 356067ae34b6SDavid Sterba cancel = true; 35611a15eb72SJosef Bacik } else { 35621a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 35631a15eb72SJosef Bacik if (ret) 35641a15eb72SJosef Bacik goto out; 35651a15eb72SJosef Bacik } 35661a15eb72SJosef Bacik 35671a15eb72SJosef Bacik ret = mnt_want_write_file(file); 35681a15eb72SJosef Bacik if (ret) 35691a15eb72SJosef Bacik goto out; 357067ae34b6SDavid Sterba 357167ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 357267ae34b6SDavid Sterba cancel); 357367ae34b6SDavid Sterba if (ret) 35741a15eb72SJosef Bacik goto err_drop; 357567ae34b6SDavid Sterba 35761a15eb72SJosef Bacik /* Exclusive operation is now claimed */ 35771a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 357867ae34b6SDavid Sterba 3579c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3580183860f6SAnand Jain 35816b526ed7SAnand Jain if (!ret) { 3582735654eaSDavid Sterba if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) 35830b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: id %llu", 35846b526ed7SAnand Jain vol_args->devid); 35856b526ed7SAnand Jain else 35860b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: %s", 35876b526ed7SAnand Jain vol_args->name); 35886b526ed7SAnand Jain } 3589c47ca32dSDan Carpenter err_drop: 35904ac20c70SIlya Dryomov mnt_drop_write_file(file); 35913fa421deSJosef Bacik if (bdev) 35923fa421deSJosef Bacik blkdev_put(bdev, mode); 35931a15eb72SJosef Bacik out: 35941a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 35951a15eb72SJosef Bacik kfree(vol_args); 3596f46b5a66SChristoph Hellwig return ret; 3597f46b5a66SChristoph Hellwig } 3598f46b5a66SChristoph Hellwig 3599f46b5a66SChristoph Hellwig static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 3600f46b5a66SChristoph Hellwig { 36011a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 36020b246afaSJeff Mahoney struct inode *inode = file_inode(file); 36030b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3604f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 36053fa421deSJosef Bacik struct block_device *bdev = NULL; 36063fa421deSJosef Bacik fmode_t mode; 3607f46b5a66SChristoph Hellwig int ret; 360837b45995STom Rix bool cancel = false; 3609f46b5a66SChristoph Hellwig 3610f46b5a66SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 3611f46b5a66SChristoph Hellwig return -EPERM; 3612f46b5a66SChristoph Hellwig 36131a15eb72SJosef Bacik vol_args = memdup_user(arg, sizeof(*vol_args)); 36141a15eb72SJosef Bacik if (IS_ERR(vol_args)) 36151a15eb72SJosef Bacik return PTR_ERR(vol_args); 36161a15eb72SJosef Bacik 36171a15eb72SJosef Bacik vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 36181a15eb72SJosef Bacik if (!strcmp("cancel", vol_args->name)) { 36191a15eb72SJosef Bacik cancel = true; 36201a15eb72SJosef Bacik } else { 36211a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 36221a15eb72SJosef Bacik if (ret) 36231a15eb72SJosef Bacik goto out; 36241a15eb72SJosef Bacik } 36251a15eb72SJosef Bacik 3626f46b5a66SChristoph Hellwig ret = mnt_want_write_file(file); 3627f46b5a66SChristoph Hellwig if (ret) 36281a15eb72SJosef Bacik goto out; 3629f46b5a66SChristoph Hellwig 363067ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 363167ae34b6SDavid Sterba cancel); 363267ae34b6SDavid Sterba if (ret == 0) { 36331a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 3634f46b5a66SChristoph Hellwig if (!ret) 36350b246afaSJeff Mahoney btrfs_info(fs_info, "disk deleted %s", vol_args->name); 3636c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 363767ae34b6SDavid Sterba } 363867ae34b6SDavid Sterba 3639f46b5a66SChristoph Hellwig mnt_drop_write_file(file); 36403fa421deSJosef Bacik if (bdev) 36413fa421deSJosef Bacik blkdev_put(bdev, mode); 36421a15eb72SJosef Bacik out: 36431a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 36441a15eb72SJosef Bacik kfree(vol_args); 3645f46b5a66SChristoph Hellwig return ret; 3646f46b5a66SChristoph Hellwig } 3647f46b5a66SChristoph Hellwig 36482ff7e61eSJeff Mahoney static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info, 36492ff7e61eSJeff Mahoney void __user *arg) 3650475f6387SJan Schmidt { 3651027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 3652475f6387SJan Schmidt struct btrfs_device *device; 36530b246afaSJeff Mahoney struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3654137c5418SJohannes Thumshirn u64 flags_in; 3655027ed2f0SLi Zefan int ret = 0; 3656475f6387SJan Schmidt 3657137c5418SJohannes Thumshirn fi_args = memdup_user(arg, sizeof(*fi_args)); 3658137c5418SJohannes Thumshirn if (IS_ERR(fi_args)) 3659137c5418SJohannes Thumshirn return PTR_ERR(fi_args); 3660137c5418SJohannes Thumshirn 3661137c5418SJohannes Thumshirn flags_in = fi_args->flags; 3662137c5418SJohannes Thumshirn memset(fi_args, 0, sizeof(*fi_args)); 3663027ed2f0SLi Zefan 3664d03262c7SDavid Sterba rcu_read_lock(); 3665027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 3666475f6387SJan Schmidt 3667d03262c7SDavid Sterba list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 3668027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 3669027ed2f0SLi Zefan fi_args->max_id = device->devid; 3670475f6387SJan Schmidt } 3671d03262c7SDavid Sterba rcu_read_unlock(); 3672475f6387SJan Schmidt 3673de37aa51SNikolay Borisov memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid)); 3674bea7eafdSOmar Sandoval fi_args->nodesize = fs_info->nodesize; 3675bea7eafdSOmar Sandoval fi_args->sectorsize = fs_info->sectorsize; 3676bea7eafdSOmar Sandoval fi_args->clone_alignment = fs_info->sectorsize; 367780a773fbSDavid Sterba 3678137c5418SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) { 3679137c5418SJohannes Thumshirn fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy); 3680137c5418SJohannes Thumshirn fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy); 3681137c5418SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO; 3682137c5418SJohannes Thumshirn } 3683137c5418SJohannes Thumshirn 36840fb408a5SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) { 36850fb408a5SJohannes Thumshirn fi_args->generation = fs_info->generation; 36860fb408a5SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION; 36870fb408a5SJohannes Thumshirn } 36880fb408a5SJohannes Thumshirn 368949bac897SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) { 369049bac897SJohannes Thumshirn memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid, 369149bac897SJohannes Thumshirn sizeof(fi_args->metadata_uuid)); 369249bac897SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID; 369349bac897SJohannes Thumshirn } 369449bac897SJohannes Thumshirn 3695027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 3696027ed2f0SLi Zefan ret = -EFAULT; 3697475f6387SJan Schmidt 3698027ed2f0SLi Zefan kfree(fi_args); 3699027ed2f0SLi Zefan return ret; 3700475f6387SJan Schmidt } 3701475f6387SJan Schmidt 37022ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info, 37032ff7e61eSJeff Mahoney void __user *arg) 3704475f6387SJan Schmidt { 3705562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 3706475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 3707475f6387SJan Schmidt struct btrfs_device *dev; 3708475f6387SJan Schmidt int ret = 0; 3709475f6387SJan Schmidt 3710475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 3711475f6387SJan Schmidt if (IS_ERR(di_args)) 3712475f6387SJan Schmidt return PTR_ERR(di_args); 3713475f6387SJan Schmidt 3714562d7b15SJosef Bacik args.devid = di_args->devid; 3715dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(di_args->uuid)) 3716562d7b15SJosef Bacik args.uuid = di_args->uuid; 3717475f6387SJan Schmidt 3718c5593ca3SDavid Sterba rcu_read_lock(); 3719562d7b15SJosef Bacik dev = btrfs_find_device(fs_info->fs_devices, &args); 3720475f6387SJan Schmidt if (!dev) { 3721475f6387SJan Schmidt ret = -ENODEV; 3722475f6387SJan Schmidt goto out; 3723475f6387SJan Schmidt } 3724475f6387SJan Schmidt 3725475f6387SJan Schmidt di_args->devid = dev->devid; 37267cc8e58dSMiao Xie di_args->bytes_used = btrfs_device_get_bytes_used(dev); 37277cc8e58dSMiao Xie di_args->total_bytes = btrfs_device_get_total_bytes(dev); 3728475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 3729a27202fbSJim Meyering if (dev->name) { 3730672d5990SMisono Tomohiro strncpy(di_args->path, rcu_str_deref(dev->name), 3731672d5990SMisono Tomohiro sizeof(di_args->path) - 1); 3732a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 3733a27202fbSJim Meyering } else { 373499ba55adSStefan Behrens di_args->path[0] = '\0'; 3735a27202fbSJim Meyering } 3736475f6387SJan Schmidt 3737475f6387SJan Schmidt out: 3738c5593ca3SDavid Sterba rcu_read_unlock(); 3739475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 3740475f6387SJan Schmidt ret = -EFAULT; 3741475f6387SJan Schmidt 3742475f6387SJan Schmidt kfree(di_args); 3743475f6387SJan Schmidt return ret; 3744475f6387SJan Schmidt } 3745475f6387SJan Schmidt 37466ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 37476ef5ed0dSJosef Bacik { 3748496ad9aaSAl Viro struct inode *inode = file_inode(file); 37490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 37506ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 37516ef5ed0dSJosef Bacik struct btrfs_root *new_root; 37526ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 37536ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 37542a2b5d62SJosef Bacik struct btrfs_path *path = NULL; 37556ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 37566ef5ed0dSJosef Bacik u64 objectid = 0; 37576ef5ed0dSJosef Bacik u64 dir_id; 37583c04ce01SMiao Xie int ret; 37596ef5ed0dSJosef Bacik 37606ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 37616ef5ed0dSJosef Bacik return -EPERM; 37626ef5ed0dSJosef Bacik 37633c04ce01SMiao Xie ret = mnt_want_write_file(file); 37643c04ce01SMiao Xie if (ret) 37653c04ce01SMiao Xie return ret; 37663c04ce01SMiao Xie 37673c04ce01SMiao Xie if (copy_from_user(&objectid, argp, sizeof(objectid))) { 37683c04ce01SMiao Xie ret = -EFAULT; 37693c04ce01SMiao Xie goto out; 37703c04ce01SMiao Xie } 37716ef5ed0dSJosef Bacik 37726ef5ed0dSJosef Bacik if (!objectid) 37731cecf579Schandan objectid = BTRFS_FS_TREE_OBJECTID; 37746ef5ed0dSJosef Bacik 377556e9357aSDavid Sterba new_root = btrfs_get_fs_root(fs_info, objectid, true); 37763c04ce01SMiao Xie if (IS_ERR(new_root)) { 37773c04ce01SMiao Xie ret = PTR_ERR(new_root); 37783c04ce01SMiao Xie goto out; 37793c04ce01SMiao Xie } 37802a2b5d62SJosef Bacik if (!is_fstree(new_root->root_key.objectid)) { 37812a2b5d62SJosef Bacik ret = -ENOENT; 37822a2b5d62SJosef Bacik goto out_free; 37832a2b5d62SJosef Bacik } 37846ef5ed0dSJosef Bacik 37856ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 37863c04ce01SMiao Xie if (!path) { 37873c04ce01SMiao Xie ret = -ENOMEM; 37882a2b5d62SJosef Bacik goto out_free; 37893c04ce01SMiao Xie } 37906ef5ed0dSJosef Bacik 37916ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 379298d5dc13STsutomu Itoh if (IS_ERR(trans)) { 37933c04ce01SMiao Xie ret = PTR_ERR(trans); 37942a2b5d62SJosef Bacik goto out_free; 37956ef5ed0dSJosef Bacik } 37966ef5ed0dSJosef Bacik 37970b246afaSJeff Mahoney dir_id = btrfs_super_root_dir(fs_info->super_copy); 37980b246afaSJeff Mahoney di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path, 37996ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 3800cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 38012a2b5d62SJosef Bacik btrfs_release_path(path); 38023a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38030b246afaSJeff Mahoney btrfs_err(fs_info, 38045d163e0eSJeff Mahoney "Umm, you don't have the default diritem, this isn't going to work"); 38053c04ce01SMiao Xie ret = -ENOENT; 38062a2b5d62SJosef Bacik goto out_free; 38076ef5ed0dSJosef Bacik } 38086ef5ed0dSJosef Bacik 38096ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 38106ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 38116ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 38122a2b5d62SJosef Bacik btrfs_release_path(path); 38136ef5ed0dSJosef Bacik 38140b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL); 38153a45bb20SJeff Mahoney btrfs_end_transaction(trans); 38162a2b5d62SJosef Bacik out_free: 381700246528SJosef Bacik btrfs_put_root(new_root); 38182a2b5d62SJosef Bacik btrfs_free_path(path); 38193c04ce01SMiao Xie out: 38203c04ce01SMiao Xie mnt_drop_write_file(file); 38213c04ce01SMiao Xie return ret; 38226ef5ed0dSJosef Bacik } 38236ef5ed0dSJosef Bacik 3824c065f5b1SSu Yue static void get_block_group_info(struct list_head *groups_list, 3825bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 3826bf5fc093SJosef Bacik { 382732da5386SDavid Sterba struct btrfs_block_group *block_group; 3828bf5fc093SJosef Bacik 3829bf5fc093SJosef Bacik space->total_bytes = 0; 3830bf5fc093SJosef Bacik space->used_bytes = 0; 3831bf5fc093SJosef Bacik space->flags = 0; 3832bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 3833bf5fc093SJosef Bacik space->flags = block_group->flags; 3834b3470b5dSDavid Sterba space->total_bytes += block_group->length; 3835bf38be65SDavid Sterba space->used_bytes += block_group->used; 3836bf5fc093SJosef Bacik } 3837bf5fc093SJosef Bacik } 3838bf5fc093SJosef Bacik 38392ff7e61eSJeff Mahoney static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info, 38402ff7e61eSJeff Mahoney void __user *arg) 38411406e432SJosef Bacik { 38421406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 38431406e432SJosef Bacik struct btrfs_ioctl_space_info space; 38441406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 38457fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 384613f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 38471406e432SJosef Bacik struct btrfs_space_info *info; 3848315d8e98SColin Ian King static const u64 types[] = { 3849315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA, 3850bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 3851bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 3852315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA 3853315d8e98SColin Ian King }; 3854bf5fc093SJosef Bacik int num_types = 4; 38557fde62bfSChris Mason int alloc_size; 38561406e432SJosef Bacik int ret = 0; 385751788b1bSDan Rosenberg u64 slot_count = 0; 3858bf5fc093SJosef Bacik int i, c; 38591406e432SJosef Bacik 38601406e432SJosef Bacik if (copy_from_user(&space_args, 38611406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 38621406e432SJosef Bacik sizeof(space_args))) 38631406e432SJosef Bacik return -EFAULT; 38641406e432SJosef Bacik 3865bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3866bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3867bf5fc093SJosef Bacik 3868bf5fc093SJosef Bacik info = NULL; 386972804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3870bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3871bf5fc093SJosef Bacik info = tmp; 3872bf5fc093SJosef Bacik break; 3873bf5fc093SJosef Bacik } 3874bf5fc093SJosef Bacik } 38751406e432SJosef Bacik 3876bf5fc093SJosef Bacik if (!info) 3877bf5fc093SJosef Bacik continue; 3878bf5fc093SJosef Bacik 3879bf5fc093SJosef Bacik down_read(&info->groups_sem); 3880bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3881bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 3882bf5fc093SJosef Bacik slot_count++; 3883bf5fc093SJosef Bacik } 3884bf5fc093SJosef Bacik up_read(&info->groups_sem); 3885bf5fc093SJosef Bacik } 3886bf5fc093SJosef Bacik 388736523e95SDavid Sterba /* 388836523e95SDavid Sterba * Global block reserve, exported as a space_info 388936523e95SDavid Sterba */ 389036523e95SDavid Sterba slot_count++; 389136523e95SDavid Sterba 38927fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 38937fde62bfSChris Mason if (space_args.space_slots == 0) { 38947fde62bfSChris Mason space_args.total_spaces = slot_count; 38957fde62bfSChris Mason goto out; 38967fde62bfSChris Mason } 3897bf5fc093SJosef Bacik 389851788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 3899bf5fc093SJosef Bacik 39007fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 3901bf5fc093SJosef Bacik 39027fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 39037fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 39047fde62bfSChris Mason */ 390509cbfeafSKirill A. Shutemov if (alloc_size > PAGE_SIZE) 39067fde62bfSChris Mason return -ENOMEM; 39077fde62bfSChris Mason 39087fde62bfSChris Mason space_args.total_spaces = 0; 39098d2db785SDavid Sterba dest = kmalloc(alloc_size, GFP_KERNEL); 39107fde62bfSChris Mason if (!dest) 39117fde62bfSChris Mason return -ENOMEM; 39127fde62bfSChris Mason dest_orig = dest; 39137fde62bfSChris Mason 39147fde62bfSChris Mason /* now we have a buffer to copy into */ 3915bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3916bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3917bf5fc093SJosef Bacik 391851788b1bSDan Rosenberg if (!slot_count) 391951788b1bSDan Rosenberg break; 392051788b1bSDan Rosenberg 3921bf5fc093SJosef Bacik info = NULL; 392272804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3923bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3924bf5fc093SJosef Bacik info = tmp; 39257fde62bfSChris Mason break; 3926bf5fc093SJosef Bacik } 3927bf5fc093SJosef Bacik } 39287fde62bfSChris Mason 3929bf5fc093SJosef Bacik if (!info) 3930bf5fc093SJosef Bacik continue; 3931bf5fc093SJosef Bacik down_read(&info->groups_sem); 3932bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3933bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 3934c065f5b1SSu Yue get_block_group_info(&info->block_groups[c], 3935c065f5b1SSu Yue &space); 39367fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 39371406e432SJosef Bacik dest++; 39381406e432SJosef Bacik space_args.total_spaces++; 393951788b1bSDan Rosenberg slot_count--; 39401406e432SJosef Bacik } 394151788b1bSDan Rosenberg if (!slot_count) 394251788b1bSDan Rosenberg break; 3943bf5fc093SJosef Bacik } 3944bf5fc093SJosef Bacik up_read(&info->groups_sem); 3945bf5fc093SJosef Bacik } 39461406e432SJosef Bacik 394736523e95SDavid Sterba /* 394836523e95SDavid Sterba * Add global block reserve 394936523e95SDavid Sterba */ 395036523e95SDavid Sterba if (slot_count) { 39510b246afaSJeff Mahoney struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; 395236523e95SDavid Sterba 395336523e95SDavid Sterba spin_lock(&block_rsv->lock); 395436523e95SDavid Sterba space.total_bytes = block_rsv->size; 395536523e95SDavid Sterba space.used_bytes = block_rsv->size - block_rsv->reserved; 395636523e95SDavid Sterba spin_unlock(&block_rsv->lock); 395736523e95SDavid Sterba space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV; 395836523e95SDavid Sterba memcpy(dest, &space, sizeof(space)); 395936523e95SDavid Sterba space_args.total_spaces++; 396036523e95SDavid Sterba } 396136523e95SDavid Sterba 39622eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 39637fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 39647fde62bfSChris Mason 39657fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 39667fde62bfSChris Mason ret = -EFAULT; 39677fde62bfSChris Mason 39687fde62bfSChris Mason kfree(dest_orig); 39697fde62bfSChris Mason out: 39707fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 39711406e432SJosef Bacik ret = -EFAULT; 39721406e432SJosef Bacik 39731406e432SJosef Bacik return ret; 39741406e432SJosef Bacik } 39751406e432SJosef Bacik 39769a8c28beSMiao Xie static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 39779a8c28beSMiao Xie void __user *argp) 397846204592SSage Weil { 397946204592SSage Weil struct btrfs_trans_handle *trans; 398046204592SSage Weil u64 transid; 398146204592SSage Weil 3982d4edf39bSMiao Xie trans = btrfs_attach_transaction_barrier(root); 3983ff7c1d33SMiao Xie if (IS_ERR(trans)) { 3984ff7c1d33SMiao Xie if (PTR_ERR(trans) != -ENOENT) 398598d5dc13STsutomu Itoh return PTR_ERR(trans); 3986ff7c1d33SMiao Xie 3987ff7c1d33SMiao Xie /* No running transaction, don't bother */ 3988ff7c1d33SMiao Xie transid = root->fs_info->last_trans_committed; 3989ff7c1d33SMiao Xie goto out; 3990ff7c1d33SMiao Xie } 399146204592SSage Weil transid = trans->transid; 3992fdfbf020SJosef Bacik btrfs_commit_transaction_async(trans); 3993ff7c1d33SMiao Xie out: 399446204592SSage Weil if (argp) 399546204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 399646204592SSage Weil return -EFAULT; 399746204592SSage Weil return 0; 399846204592SSage Weil } 399946204592SSage Weil 40002ff7e61eSJeff Mahoney static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info, 40019a8c28beSMiao Xie void __user *argp) 400246204592SSage Weil { 400346204592SSage Weil u64 transid; 400446204592SSage Weil 400546204592SSage Weil if (argp) { 400646204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 400746204592SSage Weil return -EFAULT; 400846204592SSage Weil } else { 400946204592SSage Weil transid = 0; /* current trans */ 401046204592SSage Weil } 40112ff7e61eSJeff Mahoney return btrfs_wait_for_commit(fs_info, transid); 401246204592SSage Weil } 401346204592SSage Weil 4014b8e95489SMiao Xie static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 4015475f6387SJan Schmidt { 40160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 4017475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 4018b8e95489SMiao Xie int ret; 4019475f6387SJan Schmidt 4020475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4021475f6387SJan Schmidt return -EPERM; 4022475f6387SJan Schmidt 4023da32c6d5SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 4024da32c6d5SJosef Bacik btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet"); 4025da32c6d5SJosef Bacik return -EINVAL; 4026da32c6d5SJosef Bacik } 4027da32c6d5SJosef Bacik 4028475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 4029475f6387SJan Schmidt if (IS_ERR(sa)) 4030475f6387SJan Schmidt return PTR_ERR(sa); 4031475f6387SJan Schmidt 4032b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 4033b8e95489SMiao Xie ret = mnt_want_write_file(file); 4034b8e95489SMiao Xie if (ret) 4035b8e95489SMiao Xie goto out; 4036b8e95489SMiao Xie } 4037b8e95489SMiao Xie 40380b246afaSJeff Mahoney ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end, 403963a212abSStefan Behrens &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 404063a212abSStefan Behrens 0); 4041475f6387SJan Schmidt 40425afe6ce7SFilipe Manana /* 40435afe6ce7SFilipe Manana * Copy scrub args to user space even if btrfs_scrub_dev() returned an 40445afe6ce7SFilipe Manana * error. This is important as it allows user space to know how much 40455afe6ce7SFilipe Manana * progress scrub has done. For example, if scrub is canceled we get 40465afe6ce7SFilipe Manana * -ECANCELED from btrfs_scrub_dev() and return that error back to user 40475afe6ce7SFilipe Manana * space. Later user space can inspect the progress from the structure 40485afe6ce7SFilipe Manana * btrfs_ioctl_scrub_args and resume scrub from where it left off 40495afe6ce7SFilipe Manana * previously (btrfs-progs does this). 40505afe6ce7SFilipe Manana * If we fail to copy the btrfs_ioctl_scrub_args structure to user space 40515afe6ce7SFilipe Manana * then return -EFAULT to signal the structure was not copied or it may 40525afe6ce7SFilipe Manana * be corrupt and unreliable due to a partial copy. 40535afe6ce7SFilipe Manana */ 40545afe6ce7SFilipe Manana if (copy_to_user(arg, sa, sizeof(*sa))) 4055475f6387SJan Schmidt ret = -EFAULT; 4056475f6387SJan Schmidt 4057b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) 4058b8e95489SMiao Xie mnt_drop_write_file(file); 4059b8e95489SMiao Xie out: 4060475f6387SJan Schmidt kfree(sa); 4061475f6387SJan Schmidt return ret; 4062475f6387SJan Schmidt } 4063475f6387SJan Schmidt 40642ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info) 4065475f6387SJan Schmidt { 4066475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4067475f6387SJan Schmidt return -EPERM; 4068475f6387SJan Schmidt 40692ff7e61eSJeff Mahoney return btrfs_scrub_cancel(fs_info); 4070475f6387SJan Schmidt } 4071475f6387SJan Schmidt 40722ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info, 4073475f6387SJan Schmidt void __user *arg) 4074475f6387SJan Schmidt { 4075475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 4076475f6387SJan Schmidt int ret; 4077475f6387SJan Schmidt 4078475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4079475f6387SJan Schmidt return -EPERM; 4080475f6387SJan Schmidt 4081475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 4082475f6387SJan Schmidt if (IS_ERR(sa)) 4083475f6387SJan Schmidt return PTR_ERR(sa); 4084475f6387SJan Schmidt 40852ff7e61eSJeff Mahoney ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress); 4086475f6387SJan Schmidt 40874fa99b00SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 4088475f6387SJan Schmidt ret = -EFAULT; 4089475f6387SJan Schmidt 4090475f6387SJan Schmidt kfree(sa); 4091475f6387SJan Schmidt return ret; 4092475f6387SJan Schmidt } 4093475f6387SJan Schmidt 40942ff7e61eSJeff Mahoney static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info, 4095b27f7c0cSDavid Sterba void __user *arg) 4096c11d2c23SStefan Behrens { 4097c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 4098c11d2c23SStefan Behrens int ret; 4099c11d2c23SStefan Behrens 4100c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 4101c11d2c23SStefan Behrens if (IS_ERR(sa)) 4102c11d2c23SStefan Behrens return PTR_ERR(sa); 4103c11d2c23SStefan Behrens 4104b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 4105b27f7c0cSDavid Sterba kfree(sa); 4106b27f7c0cSDavid Sterba return -EPERM; 4107b27f7c0cSDavid Sterba } 4108b27f7c0cSDavid Sterba 41092ff7e61eSJeff Mahoney ret = btrfs_get_dev_stats(fs_info, sa); 4110c11d2c23SStefan Behrens 4111eee99577SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 4112c11d2c23SStefan Behrens ret = -EFAULT; 4113c11d2c23SStefan Behrens 4114c11d2c23SStefan Behrens kfree(sa); 4115c11d2c23SStefan Behrens return ret; 4116c11d2c23SStefan Behrens } 4117c11d2c23SStefan Behrens 41182ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info, 41192ff7e61eSJeff Mahoney void __user *arg) 41203f6bcfbdSStefan Behrens { 41213f6bcfbdSStefan Behrens struct btrfs_ioctl_dev_replace_args *p; 41223f6bcfbdSStefan Behrens int ret; 41233f6bcfbdSStefan Behrens 41243f6bcfbdSStefan Behrens if (!capable(CAP_SYS_ADMIN)) 41253f6bcfbdSStefan Behrens return -EPERM; 41263f6bcfbdSStefan Behrens 4127914a519bSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 4128914a519bSJosef Bacik btrfs_err(fs_info, "device replace not supported on extent tree v2 yet"); 4129914a519bSJosef Bacik return -EINVAL; 4130914a519bSJosef Bacik } 4131914a519bSJosef Bacik 41323f6bcfbdSStefan Behrens p = memdup_user(arg, sizeof(*p)); 41333f6bcfbdSStefan Behrens if (IS_ERR(p)) 41343f6bcfbdSStefan Behrens return PTR_ERR(p); 41353f6bcfbdSStefan Behrens 41363f6bcfbdSStefan Behrens switch (p->cmd) { 41373f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 4138bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 4139adfa97cbSIlya Dryomov ret = -EROFS; 4140adfa97cbSIlya Dryomov goto out; 4141adfa97cbSIlya Dryomov } 4142c3e1f96cSGoldwyn Rodrigues if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { 4143e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 41443f6bcfbdSStefan Behrens } else { 41452ff7e61eSJeff Mahoney ret = btrfs_dev_replace_by_ioctl(fs_info, p); 4146c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 41473f6bcfbdSStefan Behrens } 41483f6bcfbdSStefan Behrens break; 41493f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 41500b246afaSJeff Mahoney btrfs_dev_replace_status(fs_info, p); 41513f6bcfbdSStefan Behrens ret = 0; 41523f6bcfbdSStefan Behrens break; 41533f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 415417d202b9SAnand Jain p->result = btrfs_dev_replace_cancel(fs_info); 415597282031SAnand Jain ret = 0; 41563f6bcfbdSStefan Behrens break; 41573f6bcfbdSStefan Behrens default: 41583f6bcfbdSStefan Behrens ret = -EINVAL; 41593f6bcfbdSStefan Behrens break; 41603f6bcfbdSStefan Behrens } 41613f6bcfbdSStefan Behrens 4162d3a53286SFilipe Manana if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p))) 41633f6bcfbdSStefan Behrens ret = -EFAULT; 4164adfa97cbSIlya Dryomov out: 41653f6bcfbdSStefan Behrens kfree(p); 41663f6bcfbdSStefan Behrens return ret; 41673f6bcfbdSStefan Behrens } 41683f6bcfbdSStefan Behrens 4169d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 4170d7728c96SJan Schmidt { 4171d7728c96SJan Schmidt int ret = 0; 4172d7728c96SJan Schmidt int i; 4173740c3d22SChris Mason u64 rel_ptr; 4174d7728c96SJan Schmidt int size; 4175806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 4176d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 4177d7728c96SJan Schmidt struct btrfs_path *path; 4178d7728c96SJan Schmidt 417982b22ac8SKusanagi Kouichi if (!capable(CAP_DAC_READ_SEARCH)) 4180d7728c96SJan Schmidt return -EPERM; 4181d7728c96SJan Schmidt 4182d7728c96SJan Schmidt path = btrfs_alloc_path(); 4183d7728c96SJan Schmidt if (!path) { 4184d7728c96SJan Schmidt ret = -ENOMEM; 4185d7728c96SJan Schmidt goto out; 4186d7728c96SJan Schmidt } 4187d7728c96SJan Schmidt 4188d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 4189d7728c96SJan Schmidt if (IS_ERR(ipa)) { 4190d7728c96SJan Schmidt ret = PTR_ERR(ipa); 4191d7728c96SJan Schmidt ipa = NULL; 4192d7728c96SJan Schmidt goto out; 4193d7728c96SJan Schmidt } 4194d7728c96SJan Schmidt 4195d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 4196d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 4197d7728c96SJan Schmidt if (IS_ERR(ipath)) { 4198d7728c96SJan Schmidt ret = PTR_ERR(ipath); 4199d7728c96SJan Schmidt ipath = NULL; 4200d7728c96SJan Schmidt goto out; 4201d7728c96SJan Schmidt } 4202d7728c96SJan Schmidt 4203d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 4204d7728c96SJan Schmidt if (ret < 0) 4205d7728c96SJan Schmidt goto out; 4206d7728c96SJan Schmidt 4207d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 4208745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 4209745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 4210740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 4211d7728c96SJan Schmidt } 4212d7728c96SJan Schmidt 4213718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)ipa->fspath, 4214718dc5faSOmar Sandoval ipath->fspath, size); 4215d7728c96SJan Schmidt if (ret) { 4216d7728c96SJan Schmidt ret = -EFAULT; 4217d7728c96SJan Schmidt goto out; 4218d7728c96SJan Schmidt } 4219d7728c96SJan Schmidt 4220d7728c96SJan Schmidt out: 4221d7728c96SJan Schmidt btrfs_free_path(path); 4222d7728c96SJan Schmidt free_ipath(ipath); 4223d7728c96SJan Schmidt kfree(ipa); 4224d7728c96SJan Schmidt 4225d7728c96SJan Schmidt return ret; 4226d7728c96SJan Schmidt } 4227d7728c96SJan Schmidt 4228d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 4229d7728c96SJan Schmidt { 4230d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 4231d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 4232d7728c96SJan Schmidt 4233d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 4234d7728c96SJan Schmidt inodes->bytes_left -= c; 4235d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 4236d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 4237d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 4238d7728c96SJan Schmidt inodes->elem_cnt += 3; 4239d7728c96SJan Schmidt } else { 4240d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 4241d7728c96SJan Schmidt inodes->bytes_left = 0; 4242d7728c96SJan Schmidt inodes->elem_missed += 3; 4243d7728c96SJan Schmidt } 4244d7728c96SJan Schmidt 4245d7728c96SJan Schmidt return 0; 4246d7728c96SJan Schmidt } 4247d7728c96SJan Schmidt 42482ff7e61eSJeff Mahoney static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info, 4249d24a67b2SZygo Blaxell void __user *arg, int version) 4250d7728c96SJan Schmidt { 4251d7728c96SJan Schmidt int ret = 0; 4252d7728c96SJan Schmidt int size; 4253d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 4254d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 4255d7728c96SJan Schmidt struct btrfs_path *path = NULL; 4256d24a67b2SZygo Blaxell bool ignore_offset; 4257d7728c96SJan Schmidt 4258d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4259d7728c96SJan Schmidt return -EPERM; 4260d7728c96SJan Schmidt 4261d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 42627b9ea627SShailendra Verma if (IS_ERR(loi)) 42637b9ea627SShailendra Verma return PTR_ERR(loi); 4264d7728c96SJan Schmidt 4265d24a67b2SZygo Blaxell if (version == 1) { 4266d24a67b2SZygo Blaxell ignore_offset = false; 4267b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_64K); 4268d24a67b2SZygo Blaxell } else { 4269d24a67b2SZygo Blaxell /* All reserved bits must be 0 for now */ 4270d24a67b2SZygo Blaxell if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) { 4271d24a67b2SZygo Blaxell ret = -EINVAL; 4272d24a67b2SZygo Blaxell goto out_loi; 4273d24a67b2SZygo Blaxell } 4274d24a67b2SZygo Blaxell /* Only accept flags we have defined so far */ 4275d24a67b2SZygo Blaxell if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) { 4276d24a67b2SZygo Blaxell ret = -EINVAL; 4277d24a67b2SZygo Blaxell goto out_loi; 4278d24a67b2SZygo Blaxell } 4279d24a67b2SZygo Blaxell ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET; 4280b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_16M); 4281d24a67b2SZygo Blaxell } 4282d24a67b2SZygo Blaxell 4283d7728c96SJan Schmidt path = btrfs_alloc_path(); 4284d7728c96SJan Schmidt if (!path) { 4285d7728c96SJan Schmidt ret = -ENOMEM; 4286d7728c96SJan Schmidt goto out; 4287d7728c96SJan Schmidt } 4288d7728c96SJan Schmidt 4289d7728c96SJan Schmidt inodes = init_data_container(size); 4290d7728c96SJan Schmidt if (IS_ERR(inodes)) { 4291d7728c96SJan Schmidt ret = PTR_ERR(inodes); 4292d7728c96SJan Schmidt inodes = NULL; 4293d7728c96SJan Schmidt goto out; 4294d7728c96SJan Schmidt } 4295d7728c96SJan Schmidt 42962ff7e61eSJeff Mahoney ret = iterate_inodes_from_logical(loi->logical, fs_info, path, 4297d24a67b2SZygo Blaxell build_ino_list, inodes, ignore_offset); 4298df031f07SLiu Bo if (ret == -EINVAL) 4299d7728c96SJan Schmidt ret = -ENOENT; 4300d7728c96SJan Schmidt if (ret < 0) 4301d7728c96SJan Schmidt goto out; 4302d7728c96SJan Schmidt 4303718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes, 4304718dc5faSOmar Sandoval size); 4305d7728c96SJan Schmidt if (ret) 4306d7728c96SJan Schmidt ret = -EFAULT; 4307d7728c96SJan Schmidt 4308d7728c96SJan Schmidt out: 4309d7728c96SJan Schmidt btrfs_free_path(path); 4310f54de068SDavid Sterba kvfree(inodes); 4311d24a67b2SZygo Blaxell out_loi: 4312d7728c96SJan Schmidt kfree(loi); 4313d7728c96SJan Schmidt 4314d7728c96SJan Schmidt return ret; 4315d7728c96SJan Schmidt } 4316d7728c96SJan Schmidt 4317008ef096SDavid Sterba void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info, 4318c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 4319c9e9f97bSIlya Dryomov { 4320c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4321c9e9f97bSIlya Dryomov 4322c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 4323c9e9f97bSIlya Dryomov 43243009a62fSDavid Sterba if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) 4325837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 4326837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 4327837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 4328a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 4329a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 4330837d5b6eSIlya Dryomov 4331c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 4332c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 4333c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 433419a39dceSIlya Dryomov 433519a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 433619a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 433719a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 4338c9e9f97bSIlya Dryomov } 4339c9e9f97bSIlya Dryomov 43409ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 4341c9e9f97bSIlya Dryomov { 4342496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4343c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 4344c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 4345c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 4346ed0fb78fSIlya Dryomov bool need_unlock; /* for mut. excl. ops lock */ 4347c9e9f97bSIlya Dryomov int ret; 4348c9e9f97bSIlya Dryomov 4349c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4350c9e9f97bSIlya Dryomov return -EPERM; 4351c9e9f97bSIlya Dryomov 4352e54bfa31SLiu Bo ret = mnt_want_write_file(file); 43539ba1f6e4SLiu Bo if (ret) 43549ba1f6e4SLiu Bo return ret; 43559ba1f6e4SLiu Bo 4356*d8645462SNikolay Borisov bargs = memdup_user(arg, sizeof(*bargs)); 4357*d8645462SNikolay Borisov if (IS_ERR(bargs)) { 4358*d8645462SNikolay Borisov ret = PTR_ERR(bargs); 4359*d8645462SNikolay Borisov bargs = NULL; 4360*d8645462SNikolay Borisov goto out; 4361*d8645462SNikolay Borisov } 4362*d8645462SNikolay Borisov 4363ed0fb78fSIlya Dryomov again: 4364c3e1f96cSGoldwyn Rodrigues if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 4365c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4366ed0fb78fSIlya Dryomov need_unlock = true; 4367ed0fb78fSIlya Dryomov goto locked; 4368ed0fb78fSIlya Dryomov } 4369ed0fb78fSIlya Dryomov 4370ed0fb78fSIlya Dryomov /* 437101327610SNicholas D Steeves * mut. excl. ops lock is locked. Three possibilities: 4372ed0fb78fSIlya Dryomov * (1) some other op is running 4373ed0fb78fSIlya Dryomov * (2) balance is running 4374ed0fb78fSIlya Dryomov * (3) balance is paused -- special case (think resume) 4375ed0fb78fSIlya Dryomov */ 4376ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4377ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4378ed0fb78fSIlya Dryomov /* this is either (2) or (3) */ 43793009a62fSDavid Sterba if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4380ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4381dccdb07bSDavid Sterba /* 4382dccdb07bSDavid Sterba * Lock released to allow other waiters to continue, 4383dccdb07bSDavid Sterba * we'll reexamine the status again. 4384dccdb07bSDavid Sterba */ 4385ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4386ed0fb78fSIlya Dryomov 4387ed0fb78fSIlya Dryomov if (fs_info->balance_ctl && 43883009a62fSDavid Sterba !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4389ed0fb78fSIlya Dryomov /* this is (3) */ 4390ed0fb78fSIlya Dryomov need_unlock = false; 4391ed0fb78fSIlya Dryomov goto locked; 4392ed0fb78fSIlya Dryomov } 4393ed0fb78fSIlya Dryomov 4394ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4395ed0fb78fSIlya Dryomov goto again; 4396ed0fb78fSIlya Dryomov } else { 4397ed0fb78fSIlya Dryomov /* this is (2) */ 4398ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4399ed0fb78fSIlya Dryomov ret = -EINPROGRESS; 4400ed0fb78fSIlya Dryomov goto out; 4401ed0fb78fSIlya Dryomov } 4402ed0fb78fSIlya Dryomov } else { 4403ed0fb78fSIlya Dryomov /* this is (1) */ 4404ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4405e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4406ed0fb78fSIlya Dryomov goto out; 4407ed0fb78fSIlya Dryomov } 4408ed0fb78fSIlya Dryomov 4409ed0fb78fSIlya Dryomov locked: 4410de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 4411de322263SIlya Dryomov if (!fs_info->balance_ctl) { 4412de322263SIlya Dryomov ret = -ENOTCONN; 4413*d8645462SNikolay Borisov goto out_unlock; 4414de322263SIlya Dryomov } 4415de322263SIlya Dryomov 4416de322263SIlya Dryomov bctl = fs_info->balance_ctl; 4417de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 4418de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 4419de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 4420efc0e69cSNikolay Borisov btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE); 4421de322263SIlya Dryomov 4422de322263SIlya Dryomov goto do_balance; 4423de322263SIlya Dryomov } 4424c9e9f97bSIlya Dryomov 4425*d8645462SNikolay Borisov if (bargs->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) { 4426*d8645462SNikolay Borisov ret = -EINVAL; 4427*d8645462SNikolay Borisov goto out_unlock; 4428*d8645462SNikolay Borisov } 4429*d8645462SNikolay Borisov 4430ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4431837d5b6eSIlya Dryomov ret = -EINPROGRESS; 4432*d8645462SNikolay Borisov goto out_unlock; 4433837d5b6eSIlya Dryomov } 4434837d5b6eSIlya Dryomov 44358d2db785SDavid Sterba bctl = kzalloc(sizeof(*bctl), GFP_KERNEL); 4436c9e9f97bSIlya Dryomov if (!bctl) { 4437c9e9f97bSIlya Dryomov ret = -ENOMEM; 4438*d8645462SNikolay Borisov goto out_unlock; 4439c9e9f97bSIlya Dryomov } 4440c9e9f97bSIlya Dryomov 4441c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 4442c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 4443c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 4444c9e9f97bSIlya Dryomov 4445c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 4446de322263SIlya Dryomov do_balance: 4447c9e9f97bSIlya Dryomov /* 4448c3e1f96cSGoldwyn Rodrigues * Ownership of bctl and exclusive operation goes to btrfs_balance. 4449c3e1f96cSGoldwyn Rodrigues * bctl is freed in reset_balance_state, or, if restriper was paused 4450c3e1f96cSGoldwyn Rodrigues * all the way until unmount, in free_fs_info. The flag should be 4451c3e1f96cSGoldwyn Rodrigues * cleared after reset_balance_state. 4452c9e9f97bSIlya Dryomov */ 4453ed0fb78fSIlya Dryomov need_unlock = false; 4454ed0fb78fSIlya Dryomov 44556fcf6e2bSDavid Sterba ret = btrfs_balance(fs_info, bctl, bargs); 44560f89abf5SChristian Engelmayer bctl = NULL; 4457ed0fb78fSIlya Dryomov 445839864601SNikolay Borisov if (ret == 0 || ret == -ECANCELED) { 4459c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 4460c9e9f97bSIlya Dryomov ret = -EFAULT; 4461c9e9f97bSIlya Dryomov } 4462c9e9f97bSIlya Dryomov 44630f89abf5SChristian Engelmayer kfree(bctl); 4464ed0fb78fSIlya Dryomov out_unlock: 4465c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4466ed0fb78fSIlya Dryomov if (need_unlock) 4467c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 4468ed0fb78fSIlya Dryomov out: 4469e54bfa31SLiu Bo mnt_drop_write_file(file); 4470*d8645462SNikolay Borisov kfree(bargs); 4471c9e9f97bSIlya Dryomov return ret; 4472c9e9f97bSIlya Dryomov } 4473c9e9f97bSIlya Dryomov 44742ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd) 4475837d5b6eSIlya Dryomov { 4476837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4477837d5b6eSIlya Dryomov return -EPERM; 4478837d5b6eSIlya Dryomov 4479837d5b6eSIlya Dryomov switch (cmd) { 4480837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 44810b246afaSJeff Mahoney return btrfs_pause_balance(fs_info); 4482a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 44830b246afaSJeff Mahoney return btrfs_cancel_balance(fs_info); 4484837d5b6eSIlya Dryomov } 4485837d5b6eSIlya Dryomov 4486837d5b6eSIlya Dryomov return -EINVAL; 4487837d5b6eSIlya Dryomov } 4488837d5b6eSIlya Dryomov 44892ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info, 449019a39dceSIlya Dryomov void __user *arg) 449119a39dceSIlya Dryomov { 449219a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 449319a39dceSIlya Dryomov int ret = 0; 449419a39dceSIlya Dryomov 449519a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 449619a39dceSIlya Dryomov return -EPERM; 449719a39dceSIlya Dryomov 449819a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 449919a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 450019a39dceSIlya Dryomov ret = -ENOTCONN; 450119a39dceSIlya Dryomov goto out; 450219a39dceSIlya Dryomov } 450319a39dceSIlya Dryomov 45048d2db785SDavid Sterba bargs = kzalloc(sizeof(*bargs), GFP_KERNEL); 450519a39dceSIlya Dryomov if (!bargs) { 450619a39dceSIlya Dryomov ret = -ENOMEM; 450719a39dceSIlya Dryomov goto out; 450819a39dceSIlya Dryomov } 450919a39dceSIlya Dryomov 4510008ef096SDavid Sterba btrfs_update_ioctl_balance_args(fs_info, bargs); 451119a39dceSIlya Dryomov 451219a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 451319a39dceSIlya Dryomov ret = -EFAULT; 451419a39dceSIlya Dryomov 451519a39dceSIlya Dryomov kfree(bargs); 451619a39dceSIlya Dryomov out: 451719a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 451819a39dceSIlya Dryomov return ret; 451919a39dceSIlya Dryomov } 452019a39dceSIlya Dryomov 4521905b0ddaSMiao Xie static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 45225d13a37bSArne Jansen { 45230b246afaSJeff Mahoney struct inode *inode = file_inode(file); 45240b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45255d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 45265d13a37bSArne Jansen int ret; 45275d13a37bSArne Jansen 45285d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 45295d13a37bSArne Jansen return -EPERM; 45305d13a37bSArne Jansen 4531905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4532905b0ddaSMiao Xie if (ret) 4533905b0ddaSMiao Xie return ret; 45345d13a37bSArne Jansen 45355d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4536905b0ddaSMiao Xie if (IS_ERR(sa)) { 4537905b0ddaSMiao Xie ret = PTR_ERR(sa); 4538905b0ddaSMiao Xie goto drop_write; 4539905b0ddaSMiao Xie } 45405d13a37bSArne Jansen 45410b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 45425d13a37bSArne Jansen 45435d13a37bSArne Jansen switch (sa->cmd) { 45445d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 4545340f1aa2SNikolay Borisov ret = btrfs_quota_enable(fs_info); 45465d13a37bSArne Jansen break; 45475d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 4548340f1aa2SNikolay Borisov ret = btrfs_quota_disable(fs_info); 45495d13a37bSArne Jansen break; 45505d13a37bSArne Jansen default: 45515d13a37bSArne Jansen ret = -EINVAL; 45525d13a37bSArne Jansen break; 45535d13a37bSArne Jansen } 45545d13a37bSArne Jansen 45555d13a37bSArne Jansen kfree(sa); 45560b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4557905b0ddaSMiao Xie drop_write: 4558905b0ddaSMiao Xie mnt_drop_write_file(file); 45595d13a37bSArne Jansen return ret; 45605d13a37bSArne Jansen } 45615d13a37bSArne Jansen 4562905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 45635d13a37bSArne Jansen { 45640b246afaSJeff Mahoney struct inode *inode = file_inode(file); 45650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45660b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 45675d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 45685d13a37bSArne Jansen struct btrfs_trans_handle *trans; 45695d13a37bSArne Jansen int ret; 45705d13a37bSArne Jansen int err; 45715d13a37bSArne Jansen 45725d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 45735d13a37bSArne Jansen return -EPERM; 45745d13a37bSArne Jansen 4575905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4576905b0ddaSMiao Xie if (ret) 4577905b0ddaSMiao Xie return ret; 45785d13a37bSArne Jansen 45795d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4580905b0ddaSMiao Xie if (IS_ERR(sa)) { 4581905b0ddaSMiao Xie ret = PTR_ERR(sa); 4582905b0ddaSMiao Xie goto drop_write; 4583905b0ddaSMiao Xie } 45845d13a37bSArne Jansen 45855d13a37bSArne Jansen trans = btrfs_join_transaction(root); 45865d13a37bSArne Jansen if (IS_ERR(trans)) { 45875d13a37bSArne Jansen ret = PTR_ERR(trans); 45885d13a37bSArne Jansen goto out; 45895d13a37bSArne Jansen } 45905d13a37bSArne Jansen 45915d13a37bSArne Jansen if (sa->assign) { 45929f8a6ce6SLu Fengqi ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst); 45935d13a37bSArne Jansen } else { 459439616c27SLu Fengqi ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst); 45955d13a37bSArne Jansen } 45965d13a37bSArne Jansen 4597e082f563SQu Wenruo /* update qgroup status and info */ 4598280f8bd2SLu Fengqi err = btrfs_run_qgroups(trans); 4599e082f563SQu Wenruo if (err < 0) 46000b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, err, 4601ad8403dfSAnand Jain "failed to update qgroup status and info"); 46023a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 46035d13a37bSArne Jansen if (err && !ret) 46045d13a37bSArne Jansen ret = err; 46055d13a37bSArne Jansen 46065d13a37bSArne Jansen out: 46075d13a37bSArne Jansen kfree(sa); 4608905b0ddaSMiao Xie drop_write: 4609905b0ddaSMiao Xie mnt_drop_write_file(file); 46105d13a37bSArne Jansen return ret; 46115d13a37bSArne Jansen } 46125d13a37bSArne Jansen 4613905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 46145d13a37bSArne Jansen { 46150b246afaSJeff Mahoney struct inode *inode = file_inode(file); 46160b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 46175d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 46185d13a37bSArne Jansen struct btrfs_trans_handle *trans; 46195d13a37bSArne Jansen int ret; 46205d13a37bSArne Jansen int err; 46215d13a37bSArne Jansen 46225d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 46235d13a37bSArne Jansen return -EPERM; 46245d13a37bSArne Jansen 4625905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4626905b0ddaSMiao Xie if (ret) 4627905b0ddaSMiao Xie return ret; 46285d13a37bSArne Jansen 46295d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4630905b0ddaSMiao Xie if (IS_ERR(sa)) { 4631905b0ddaSMiao Xie ret = PTR_ERR(sa); 4632905b0ddaSMiao Xie goto drop_write; 4633905b0ddaSMiao Xie } 46345d13a37bSArne Jansen 4635d86e56cfSMiao Xie if (!sa->qgroupid) { 4636d86e56cfSMiao Xie ret = -EINVAL; 4637d86e56cfSMiao Xie goto out; 4638d86e56cfSMiao Xie } 4639d86e56cfSMiao Xie 46405d13a37bSArne Jansen trans = btrfs_join_transaction(root); 46415d13a37bSArne Jansen if (IS_ERR(trans)) { 46425d13a37bSArne Jansen ret = PTR_ERR(trans); 46435d13a37bSArne Jansen goto out; 46445d13a37bSArne Jansen } 46455d13a37bSArne Jansen 46465d13a37bSArne Jansen if (sa->create) { 464749a05ecdSLu Fengqi ret = btrfs_create_qgroup(trans, sa->qgroupid); 46485d13a37bSArne Jansen } else { 46493efbee1dSLu Fengqi ret = btrfs_remove_qgroup(trans, sa->qgroupid); 46505d13a37bSArne Jansen } 46515d13a37bSArne Jansen 46523a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 46535d13a37bSArne Jansen if (err && !ret) 46545d13a37bSArne Jansen ret = err; 46555d13a37bSArne Jansen 46565d13a37bSArne Jansen out: 46575d13a37bSArne Jansen kfree(sa); 4658905b0ddaSMiao Xie drop_write: 4659905b0ddaSMiao Xie mnt_drop_write_file(file); 46605d13a37bSArne Jansen return ret; 46615d13a37bSArne Jansen } 46625d13a37bSArne Jansen 4663905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 46645d13a37bSArne Jansen { 46650b246afaSJeff Mahoney struct inode *inode = file_inode(file); 46660b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 46675d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 46685d13a37bSArne Jansen struct btrfs_trans_handle *trans; 46695d13a37bSArne Jansen int ret; 46705d13a37bSArne Jansen int err; 46715d13a37bSArne Jansen u64 qgroupid; 46725d13a37bSArne Jansen 46735d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 46745d13a37bSArne Jansen return -EPERM; 46755d13a37bSArne Jansen 4676905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4677905b0ddaSMiao Xie if (ret) 4678905b0ddaSMiao Xie return ret; 46795d13a37bSArne Jansen 46805d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4681905b0ddaSMiao Xie if (IS_ERR(sa)) { 4682905b0ddaSMiao Xie ret = PTR_ERR(sa); 4683905b0ddaSMiao Xie goto drop_write; 4684905b0ddaSMiao Xie } 46855d13a37bSArne Jansen 46865d13a37bSArne Jansen trans = btrfs_join_transaction(root); 46875d13a37bSArne Jansen if (IS_ERR(trans)) { 46885d13a37bSArne Jansen ret = PTR_ERR(trans); 46895d13a37bSArne Jansen goto out; 46905d13a37bSArne Jansen } 46915d13a37bSArne Jansen 46925d13a37bSArne Jansen qgroupid = sa->qgroupid; 46935d13a37bSArne Jansen if (!qgroupid) { 46945d13a37bSArne Jansen /* take the current subvol as qgroup */ 46955d13a37bSArne Jansen qgroupid = root->root_key.objectid; 46965d13a37bSArne Jansen } 46975d13a37bSArne Jansen 4698f0042d5eSLu Fengqi ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim); 46995d13a37bSArne Jansen 47003a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 47015d13a37bSArne Jansen if (err && !ret) 47025d13a37bSArne Jansen ret = err; 47035d13a37bSArne Jansen 47045d13a37bSArne Jansen out: 47055d13a37bSArne Jansen kfree(sa); 4706905b0ddaSMiao Xie drop_write: 4707905b0ddaSMiao Xie mnt_drop_write_file(file); 47085d13a37bSArne Jansen return ret; 47095d13a37bSArne Jansen } 47105d13a37bSArne Jansen 47112f232036SJan Schmidt static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 47122f232036SJan Schmidt { 47130b246afaSJeff Mahoney struct inode *inode = file_inode(file); 47140b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47152f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 47162f232036SJan Schmidt int ret; 47172f232036SJan Schmidt 47182f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 47192f232036SJan Schmidt return -EPERM; 47202f232036SJan Schmidt 47212f232036SJan Schmidt ret = mnt_want_write_file(file); 47222f232036SJan Schmidt if (ret) 47232f232036SJan Schmidt return ret; 47242f232036SJan Schmidt 47252f232036SJan Schmidt qsa = memdup_user(arg, sizeof(*qsa)); 47262f232036SJan Schmidt if (IS_ERR(qsa)) { 47272f232036SJan Schmidt ret = PTR_ERR(qsa); 47282f232036SJan Schmidt goto drop_write; 47292f232036SJan Schmidt } 47302f232036SJan Schmidt 47312f232036SJan Schmidt if (qsa->flags) { 47322f232036SJan Schmidt ret = -EINVAL; 47332f232036SJan Schmidt goto out; 47342f232036SJan Schmidt } 47352f232036SJan Schmidt 47360b246afaSJeff Mahoney ret = btrfs_qgroup_rescan(fs_info); 47372f232036SJan Schmidt 47382f232036SJan Schmidt out: 47392f232036SJan Schmidt kfree(qsa); 47402f232036SJan Schmidt drop_write: 47412f232036SJan Schmidt mnt_drop_write_file(file); 47422f232036SJan Schmidt return ret; 47432f232036SJan Schmidt } 47442f232036SJan Schmidt 4745b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info, 4746b929c1d8SMarcos Paulo de Souza void __user *arg) 47472f232036SJan Schmidt { 47480afb603aSGoldwyn Rodrigues struct btrfs_ioctl_quota_rescan_args qsa = {0}; 47492f232036SJan Schmidt 47502f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 47512f232036SJan Schmidt return -EPERM; 47522f232036SJan Schmidt 47530b246afaSJeff Mahoney if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 47540afb603aSGoldwyn Rodrigues qsa.flags = 1; 47550afb603aSGoldwyn Rodrigues qsa.progress = fs_info->qgroup_rescan_progress.objectid; 47562f232036SJan Schmidt } 47572f232036SJan Schmidt 47580afb603aSGoldwyn Rodrigues if (copy_to_user(arg, &qsa, sizeof(qsa))) 4759991a3daeSAnand Jain return -EFAULT; 47602f232036SJan Schmidt 4761991a3daeSAnand Jain return 0; 47622f232036SJan Schmidt } 47632f232036SJan Schmidt 4764b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info, 4765b929c1d8SMarcos Paulo de Souza void __user *arg) 476657254b6eSJan Schmidt { 476757254b6eSJan Schmidt if (!capable(CAP_SYS_ADMIN)) 476857254b6eSJan Schmidt return -EPERM; 476957254b6eSJan Schmidt 47700b246afaSJeff Mahoney return btrfs_qgroup_wait_for_completion(fs_info, true); 477157254b6eSJan Schmidt } 477257254b6eSJan Schmidt 4773abccd00fSHugo Mills static long _btrfs_ioctl_set_received_subvol(struct file *file, 4774e4fed17aSChristian Brauner struct user_namespace *mnt_userns, 4775abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa) 47768ea05e3aSAlexander Block { 4777496ad9aaSAl Viro struct inode *inode = file_inode(file); 47780b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47798ea05e3aSAlexander Block struct btrfs_root *root = BTRFS_I(inode)->root; 47808ea05e3aSAlexander Block struct btrfs_root_item *root_item = &root->root_item; 47818ea05e3aSAlexander Block struct btrfs_trans_handle *trans; 478295582b00SDeepa Dinamani struct timespec64 ct = current_time(inode); 47838ea05e3aSAlexander Block int ret = 0; 4784dd5f9615SStefan Behrens int received_uuid_changed; 47858ea05e3aSAlexander Block 4786e4fed17aSChristian Brauner if (!inode_owner_or_capable(mnt_userns, inode)) 4787bd60ea0fSDavid Sterba return -EPERM; 4788bd60ea0fSDavid Sterba 47898ea05e3aSAlexander Block ret = mnt_want_write_file(file); 47908ea05e3aSAlexander Block if (ret < 0) 47918ea05e3aSAlexander Block return ret; 47928ea05e3aSAlexander Block 47930b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 47948ea05e3aSAlexander Block 47954a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 47968ea05e3aSAlexander Block ret = -EINVAL; 47978ea05e3aSAlexander Block goto out; 47988ea05e3aSAlexander Block } 47998ea05e3aSAlexander Block 48008ea05e3aSAlexander Block if (btrfs_root_readonly(root)) { 48018ea05e3aSAlexander Block ret = -EROFS; 48028ea05e3aSAlexander Block goto out; 48038ea05e3aSAlexander Block } 48048ea05e3aSAlexander Block 4805dd5f9615SStefan Behrens /* 4806dd5f9615SStefan Behrens * 1 - root item 4807dd5f9615SStefan Behrens * 2 - uuid items (received uuid + subvol uuid) 4808dd5f9615SStefan Behrens */ 4809dd5f9615SStefan Behrens trans = btrfs_start_transaction(root, 3); 48108ea05e3aSAlexander Block if (IS_ERR(trans)) { 48118ea05e3aSAlexander Block ret = PTR_ERR(trans); 48128ea05e3aSAlexander Block trans = NULL; 48138ea05e3aSAlexander Block goto out; 48148ea05e3aSAlexander Block } 48158ea05e3aSAlexander Block 48168ea05e3aSAlexander Block sa->rtransid = trans->transid; 48178ea05e3aSAlexander Block sa->rtime.sec = ct.tv_sec; 48188ea05e3aSAlexander Block sa->rtime.nsec = ct.tv_nsec; 48198ea05e3aSAlexander Block 4820dd5f9615SStefan Behrens received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 4821dd5f9615SStefan Behrens BTRFS_UUID_SIZE); 4822dd5f9615SStefan Behrens if (received_uuid_changed && 4823d87ff758SNikolay Borisov !btrfs_is_empty_uuid(root_item->received_uuid)) { 4824d1957791SLu Fengqi ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid, 4825dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4826dd5f9615SStefan Behrens root->root_key.objectid); 4827d87ff758SNikolay Borisov if (ret && ret != -ENOENT) { 4828d87ff758SNikolay Borisov btrfs_abort_transaction(trans, ret); 4829d87ff758SNikolay Borisov btrfs_end_transaction(trans); 4830d87ff758SNikolay Borisov goto out; 4831d87ff758SNikolay Borisov } 4832d87ff758SNikolay Borisov } 48338ea05e3aSAlexander Block memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 48348ea05e3aSAlexander Block btrfs_set_root_stransid(root_item, sa->stransid); 48358ea05e3aSAlexander Block btrfs_set_root_rtransid(root_item, sa->rtransid); 48363cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 48373cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 48383cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 48393cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 48408ea05e3aSAlexander Block 48410b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 48428ea05e3aSAlexander Block &root->root_key, &root->root_item); 48438ea05e3aSAlexander Block if (ret < 0) { 48443a45bb20SJeff Mahoney btrfs_end_transaction(trans); 48458ea05e3aSAlexander Block goto out; 4846dd5f9615SStefan Behrens } 4847dd5f9615SStefan Behrens if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 4848cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, sa->uuid, 4849dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4850dd5f9615SStefan Behrens root->root_key.objectid); 4851dd5f9615SStefan Behrens if (ret < 0 && ret != -EEXIST) { 485266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 4853efd38150SNikolay Borisov btrfs_end_transaction(trans); 4854dd5f9615SStefan Behrens goto out; 4855dd5f9615SStefan Behrens } 4856dd5f9615SStefan Behrens } 48573a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 4858abccd00fSHugo Mills out: 48590b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4860abccd00fSHugo Mills mnt_drop_write_file(file); 4861abccd00fSHugo Mills return ret; 4862abccd00fSHugo Mills } 4863abccd00fSHugo Mills 4864abccd00fSHugo Mills #ifdef CONFIG_64BIT 4865abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol_32(struct file *file, 4866abccd00fSHugo Mills void __user *arg) 4867abccd00fSHugo Mills { 4868abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL; 4869abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *args64 = NULL; 4870abccd00fSHugo Mills int ret = 0; 4871abccd00fSHugo Mills 4872abccd00fSHugo Mills args32 = memdup_user(arg, sizeof(*args32)); 48737b9ea627SShailendra Verma if (IS_ERR(args32)) 48747b9ea627SShailendra Verma return PTR_ERR(args32); 4875abccd00fSHugo Mills 48768d2db785SDavid Sterba args64 = kmalloc(sizeof(*args64), GFP_KERNEL); 487784dbeb87SDan Carpenter if (!args64) { 487884dbeb87SDan Carpenter ret = -ENOMEM; 4879abccd00fSHugo Mills goto out; 4880abccd00fSHugo Mills } 4881abccd00fSHugo Mills 4882abccd00fSHugo Mills memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE); 4883abccd00fSHugo Mills args64->stransid = args32->stransid; 4884abccd00fSHugo Mills args64->rtransid = args32->rtransid; 4885abccd00fSHugo Mills args64->stime.sec = args32->stime.sec; 4886abccd00fSHugo Mills args64->stime.nsec = args32->stime.nsec; 4887abccd00fSHugo Mills args64->rtime.sec = args32->rtime.sec; 4888abccd00fSHugo Mills args64->rtime.nsec = args32->rtime.nsec; 4889abccd00fSHugo Mills args64->flags = args32->flags; 4890abccd00fSHugo Mills 4891e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64); 4892abccd00fSHugo Mills if (ret) 4893abccd00fSHugo Mills goto out; 4894abccd00fSHugo Mills 4895abccd00fSHugo Mills memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE); 4896abccd00fSHugo Mills args32->stransid = args64->stransid; 4897abccd00fSHugo Mills args32->rtransid = args64->rtransid; 4898abccd00fSHugo Mills args32->stime.sec = args64->stime.sec; 4899abccd00fSHugo Mills args32->stime.nsec = args64->stime.nsec; 4900abccd00fSHugo Mills args32->rtime.sec = args64->rtime.sec; 4901abccd00fSHugo Mills args32->rtime.nsec = args64->rtime.nsec; 4902abccd00fSHugo Mills args32->flags = args64->flags; 4903abccd00fSHugo Mills 4904abccd00fSHugo Mills ret = copy_to_user(arg, args32, sizeof(*args32)); 4905abccd00fSHugo Mills if (ret) 4906abccd00fSHugo Mills ret = -EFAULT; 4907abccd00fSHugo Mills 4908abccd00fSHugo Mills out: 4909abccd00fSHugo Mills kfree(args32); 4910abccd00fSHugo Mills kfree(args64); 4911abccd00fSHugo Mills return ret; 4912abccd00fSHugo Mills } 4913abccd00fSHugo Mills #endif 4914abccd00fSHugo Mills 4915abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol(struct file *file, 4916abccd00fSHugo Mills void __user *arg) 4917abccd00fSHugo Mills { 4918abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa = NULL; 4919abccd00fSHugo Mills int ret = 0; 4920abccd00fSHugo Mills 4921abccd00fSHugo Mills sa = memdup_user(arg, sizeof(*sa)); 49227b9ea627SShailendra Verma if (IS_ERR(sa)) 49237b9ea627SShailendra Verma return PTR_ERR(sa); 4924abccd00fSHugo Mills 4925e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa); 4926abccd00fSHugo Mills 4927abccd00fSHugo Mills if (ret) 4928abccd00fSHugo Mills goto out; 4929abccd00fSHugo Mills 49308ea05e3aSAlexander Block ret = copy_to_user(arg, sa, sizeof(*sa)); 49318ea05e3aSAlexander Block if (ret) 49328ea05e3aSAlexander Block ret = -EFAULT; 49338ea05e3aSAlexander Block 49348ea05e3aSAlexander Block out: 49358ea05e3aSAlexander Block kfree(sa); 49368ea05e3aSAlexander Block return ret; 49378ea05e3aSAlexander Block } 49388ea05e3aSAlexander Block 4939b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info, 4940b929c1d8SMarcos Paulo de Souza void __user *arg) 4941867ab667Sjeff.liu { 4942a1b83ac5SAnand Jain size_t len; 4943867ab667Sjeff.liu int ret; 4944a1b83ac5SAnand Jain char label[BTRFS_LABEL_SIZE]; 4945a1b83ac5SAnand Jain 49460b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 49470b246afaSJeff Mahoney memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE); 49480b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 4949a1b83ac5SAnand Jain 4950a1b83ac5SAnand Jain len = strnlen(label, BTRFS_LABEL_SIZE); 4951867ab667Sjeff.liu 4952867ab667Sjeff.liu if (len == BTRFS_LABEL_SIZE) { 49530b246afaSJeff Mahoney btrfs_warn(fs_info, 49540b246afaSJeff Mahoney "label is too long, return the first %zu bytes", 49550b246afaSJeff Mahoney --len); 4956867ab667Sjeff.liu } 4957867ab667Sjeff.liu 4958867ab667Sjeff.liu ret = copy_to_user(arg, label, len); 4959867ab667Sjeff.liu 4960867ab667Sjeff.liu return ret ? -EFAULT : 0; 4961867ab667Sjeff.liu } 4962867ab667Sjeff.liu 4963a8bfd4abSjeff.liu static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 4964a8bfd4abSjeff.liu { 49650b246afaSJeff Mahoney struct inode *inode = file_inode(file); 49660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 49670b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 49680b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 4969a8bfd4abSjeff.liu struct btrfs_trans_handle *trans; 4970a8bfd4abSjeff.liu char label[BTRFS_LABEL_SIZE]; 4971a8bfd4abSjeff.liu int ret; 4972a8bfd4abSjeff.liu 4973a8bfd4abSjeff.liu if (!capable(CAP_SYS_ADMIN)) 4974a8bfd4abSjeff.liu return -EPERM; 4975a8bfd4abSjeff.liu 4976a8bfd4abSjeff.liu if (copy_from_user(label, arg, sizeof(label))) 4977a8bfd4abSjeff.liu return -EFAULT; 4978a8bfd4abSjeff.liu 4979a8bfd4abSjeff.liu if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 49800b246afaSJeff Mahoney btrfs_err(fs_info, 49815d163e0eSJeff Mahoney "unable to set label with more than %d bytes", 4982a8bfd4abSjeff.liu BTRFS_LABEL_SIZE - 1); 4983a8bfd4abSjeff.liu return -EINVAL; 4984a8bfd4abSjeff.liu } 4985a8bfd4abSjeff.liu 4986a8bfd4abSjeff.liu ret = mnt_want_write_file(file); 4987a8bfd4abSjeff.liu if (ret) 4988a8bfd4abSjeff.liu return ret; 4989a8bfd4abSjeff.liu 4990a8bfd4abSjeff.liu trans = btrfs_start_transaction(root, 0); 4991a8bfd4abSjeff.liu if (IS_ERR(trans)) { 4992a8bfd4abSjeff.liu ret = PTR_ERR(trans); 4993a8bfd4abSjeff.liu goto out_unlock; 4994a8bfd4abSjeff.liu } 4995a8bfd4abSjeff.liu 49960b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 4997a8bfd4abSjeff.liu strcpy(super_block->label, label); 49980b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 49993a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 5000a8bfd4abSjeff.liu 5001a8bfd4abSjeff.liu out_unlock: 5002a8bfd4abSjeff.liu mnt_drop_write_file(file); 5003a8bfd4abSjeff.liu return ret; 5004a8bfd4abSjeff.liu } 5005a8bfd4abSjeff.liu 50062eaa055fSJeff Mahoney #define INIT_FEATURE_FLAGS(suffix) \ 50072eaa055fSJeff Mahoney { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 50082eaa055fSJeff Mahoney .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 50092eaa055fSJeff Mahoney .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 50102eaa055fSJeff Mahoney 5011d5131b65SDavid Sterba int btrfs_ioctl_get_supported_features(void __user *arg) 50122eaa055fSJeff Mahoney { 50134d4ab6d6SDavid Sterba static const struct btrfs_ioctl_feature_flags features[3] = { 50142eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SUPP), 50152eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_SET), 50162eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_CLEAR) 50172eaa055fSJeff Mahoney }; 50182eaa055fSJeff Mahoney 50192eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 50202eaa055fSJeff Mahoney return -EFAULT; 50212eaa055fSJeff Mahoney 50222eaa055fSJeff Mahoney return 0; 50232eaa055fSJeff Mahoney } 50242eaa055fSJeff Mahoney 5025b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info, 5026b929c1d8SMarcos Paulo de Souza void __user *arg) 50272eaa055fSJeff Mahoney { 50280b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 50292eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags features; 50302eaa055fSJeff Mahoney 50312eaa055fSJeff Mahoney features.compat_flags = btrfs_super_compat_flags(super_block); 50322eaa055fSJeff Mahoney features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 50332eaa055fSJeff Mahoney features.incompat_flags = btrfs_super_incompat_flags(super_block); 50342eaa055fSJeff Mahoney 50352eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 50362eaa055fSJeff Mahoney return -EFAULT; 50372eaa055fSJeff Mahoney 50382eaa055fSJeff Mahoney return 0; 50392eaa055fSJeff Mahoney } 50402eaa055fSJeff Mahoney 50412ff7e61eSJeff Mahoney static int check_feature_bits(struct btrfs_fs_info *fs_info, 50423b02a68aSJeff Mahoney enum btrfs_feature_set set, 50432eaa055fSJeff Mahoney u64 change_mask, u64 flags, u64 supported_flags, 50442eaa055fSJeff Mahoney u64 safe_set, u64 safe_clear) 50452eaa055fSJeff Mahoney { 5046f10152bcSDavid Sterba const char *type = btrfs_feature_set_name(set); 50473b02a68aSJeff Mahoney char *names; 50482eaa055fSJeff Mahoney u64 disallowed, unsupported; 50492eaa055fSJeff Mahoney u64 set_mask = flags & change_mask; 50502eaa055fSJeff Mahoney u64 clear_mask = ~flags & change_mask; 50512eaa055fSJeff Mahoney 50522eaa055fSJeff Mahoney unsupported = set_mask & ~supported_flags; 50532eaa055fSJeff Mahoney if (unsupported) { 50543b02a68aSJeff Mahoney names = btrfs_printable_features(set, unsupported); 50553b02a68aSJeff Mahoney if (names) { 50560b246afaSJeff Mahoney btrfs_warn(fs_info, 50573b02a68aSJeff Mahoney "this kernel does not support the %s feature bit%s", 50583b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50593b02a68aSJeff Mahoney kfree(names); 50603b02a68aSJeff Mahoney } else 50610b246afaSJeff Mahoney btrfs_warn(fs_info, 50622eaa055fSJeff Mahoney "this kernel does not support %s bits 0x%llx", 50632eaa055fSJeff Mahoney type, unsupported); 50642eaa055fSJeff Mahoney return -EOPNOTSUPP; 50652eaa055fSJeff Mahoney } 50662eaa055fSJeff Mahoney 50672eaa055fSJeff Mahoney disallowed = set_mask & ~safe_set; 50682eaa055fSJeff Mahoney if (disallowed) { 50693b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 50703b02a68aSJeff Mahoney if (names) { 50710b246afaSJeff Mahoney btrfs_warn(fs_info, 50723b02a68aSJeff Mahoney "can't set the %s feature bit%s while mounted", 50733b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50743b02a68aSJeff Mahoney kfree(names); 50753b02a68aSJeff Mahoney } else 50760b246afaSJeff Mahoney btrfs_warn(fs_info, 50772eaa055fSJeff Mahoney "can't set %s bits 0x%llx while mounted", 50782eaa055fSJeff Mahoney type, disallowed); 50792eaa055fSJeff Mahoney return -EPERM; 50802eaa055fSJeff Mahoney } 50812eaa055fSJeff Mahoney 50822eaa055fSJeff Mahoney disallowed = clear_mask & ~safe_clear; 50832eaa055fSJeff Mahoney if (disallowed) { 50843b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 50853b02a68aSJeff Mahoney if (names) { 50860b246afaSJeff Mahoney btrfs_warn(fs_info, 50873b02a68aSJeff Mahoney "can't clear the %s feature bit%s while mounted", 50883b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50893b02a68aSJeff Mahoney kfree(names); 50903b02a68aSJeff Mahoney } else 50910b246afaSJeff Mahoney btrfs_warn(fs_info, 50922eaa055fSJeff Mahoney "can't clear %s bits 0x%llx while mounted", 50932eaa055fSJeff Mahoney type, disallowed); 50942eaa055fSJeff Mahoney return -EPERM; 50952eaa055fSJeff Mahoney } 50962eaa055fSJeff Mahoney 50972eaa055fSJeff Mahoney return 0; 50982eaa055fSJeff Mahoney } 50992eaa055fSJeff Mahoney 51002ff7e61eSJeff Mahoney #define check_feature(fs_info, change_mask, flags, mask_base) \ 51012ff7e61eSJeff Mahoney check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags, \ 51022eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 51032eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 51042eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 51052eaa055fSJeff Mahoney 51062eaa055fSJeff Mahoney static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 51072eaa055fSJeff Mahoney { 51080b246afaSJeff Mahoney struct inode *inode = file_inode(file); 51090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 51100b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 51110b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 51122eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags flags[2]; 51132eaa055fSJeff Mahoney struct btrfs_trans_handle *trans; 51142eaa055fSJeff Mahoney u64 newflags; 51152eaa055fSJeff Mahoney int ret; 51162eaa055fSJeff Mahoney 51172eaa055fSJeff Mahoney if (!capable(CAP_SYS_ADMIN)) 51182eaa055fSJeff Mahoney return -EPERM; 51192eaa055fSJeff Mahoney 51202eaa055fSJeff Mahoney if (copy_from_user(flags, arg, sizeof(flags))) 51212eaa055fSJeff Mahoney return -EFAULT; 51222eaa055fSJeff Mahoney 51232eaa055fSJeff Mahoney /* Nothing to do */ 51242eaa055fSJeff Mahoney if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 51252eaa055fSJeff Mahoney !flags[0].incompat_flags) 51262eaa055fSJeff Mahoney return 0; 51272eaa055fSJeff Mahoney 51282ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_flags, 51292eaa055fSJeff Mahoney flags[1].compat_flags, COMPAT); 51302eaa055fSJeff Mahoney if (ret) 51312eaa055fSJeff Mahoney return ret; 51322eaa055fSJeff Mahoney 51332ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_ro_flags, 51342eaa055fSJeff Mahoney flags[1].compat_ro_flags, COMPAT_RO); 51352eaa055fSJeff Mahoney if (ret) 51362eaa055fSJeff Mahoney return ret; 51372eaa055fSJeff Mahoney 51382ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].incompat_flags, 51392eaa055fSJeff Mahoney flags[1].incompat_flags, INCOMPAT); 51402eaa055fSJeff Mahoney if (ret) 51412eaa055fSJeff Mahoney return ret; 51422eaa055fSJeff Mahoney 51437ab19625SDavid Sterba ret = mnt_want_write_file(file); 51447ab19625SDavid Sterba if (ret) 51457ab19625SDavid Sterba return ret; 51467ab19625SDavid Sterba 51478051aa1aSDavid Sterba trans = btrfs_start_transaction(root, 0); 51487ab19625SDavid Sterba if (IS_ERR(trans)) { 51497ab19625SDavid Sterba ret = PTR_ERR(trans); 51507ab19625SDavid Sterba goto out_drop_write; 51517ab19625SDavid Sterba } 51522eaa055fSJeff Mahoney 51530b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 51542eaa055fSJeff Mahoney newflags = btrfs_super_compat_flags(super_block); 51552eaa055fSJeff Mahoney newflags |= flags[0].compat_flags & flags[1].compat_flags; 51562eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 51572eaa055fSJeff Mahoney btrfs_set_super_compat_flags(super_block, newflags); 51582eaa055fSJeff Mahoney 51592eaa055fSJeff Mahoney newflags = btrfs_super_compat_ro_flags(super_block); 51602eaa055fSJeff Mahoney newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 51612eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 51622eaa055fSJeff Mahoney btrfs_set_super_compat_ro_flags(super_block, newflags); 51632eaa055fSJeff Mahoney 51642eaa055fSJeff Mahoney newflags = btrfs_super_incompat_flags(super_block); 51652eaa055fSJeff Mahoney newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 51662eaa055fSJeff Mahoney newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 51672eaa055fSJeff Mahoney btrfs_set_super_incompat_flags(super_block, newflags); 51680b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 51692eaa055fSJeff Mahoney 51703a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 51717ab19625SDavid Sterba out_drop_write: 51727ab19625SDavid Sterba mnt_drop_write_file(file); 51737ab19625SDavid Sterba 51747ab19625SDavid Sterba return ret; 51752eaa055fSJeff Mahoney } 51762eaa055fSJeff Mahoney 51779ad12305SSahil Kang static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat) 51782351f431SJosef Bacik { 51792351f431SJosef Bacik struct btrfs_ioctl_send_args *arg; 51802351f431SJosef Bacik int ret; 51812351f431SJosef Bacik 51822351f431SJosef Bacik if (compat) { 51832351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 51842351f431SJosef Bacik struct btrfs_ioctl_send_args_32 args32; 51852351f431SJosef Bacik 51862351f431SJosef Bacik ret = copy_from_user(&args32, argp, sizeof(args32)); 51872351f431SJosef Bacik if (ret) 51882351f431SJosef Bacik return -EFAULT; 51892351f431SJosef Bacik arg = kzalloc(sizeof(*arg), GFP_KERNEL); 51902351f431SJosef Bacik if (!arg) 51912351f431SJosef Bacik return -ENOMEM; 51922351f431SJosef Bacik arg->send_fd = args32.send_fd; 51932351f431SJosef Bacik arg->clone_sources_count = args32.clone_sources_count; 51942351f431SJosef Bacik arg->clone_sources = compat_ptr(args32.clone_sources); 51952351f431SJosef Bacik arg->parent_root = args32.parent_root; 51962351f431SJosef Bacik arg->flags = args32.flags; 51972351f431SJosef Bacik memcpy(arg->reserved, args32.reserved, 51982351f431SJosef Bacik sizeof(args32.reserved)); 51992351f431SJosef Bacik #else 52002351f431SJosef Bacik return -ENOTTY; 52012351f431SJosef Bacik #endif 52022351f431SJosef Bacik } else { 52032351f431SJosef Bacik arg = memdup_user(argp, sizeof(*arg)); 52042351f431SJosef Bacik if (IS_ERR(arg)) 52052351f431SJosef Bacik return PTR_ERR(arg); 52062351f431SJosef Bacik } 52079ad12305SSahil Kang ret = btrfs_ioctl_send(inode, arg); 52082351f431SJosef Bacik kfree(arg); 52092351f431SJosef Bacik return ret; 52102351f431SJosef Bacik } 52112351f431SJosef Bacik 52121881fba8SOmar Sandoval static int btrfs_ioctl_encoded_read(struct file *file, void __user *argp, 52131881fba8SOmar Sandoval bool compat) 52141881fba8SOmar Sandoval { 52151881fba8SOmar Sandoval struct btrfs_ioctl_encoded_io_args args = { 0 }; 52161881fba8SOmar Sandoval size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args, 52171881fba8SOmar Sandoval flags); 52181881fba8SOmar Sandoval size_t copy_end; 52191881fba8SOmar Sandoval struct iovec iovstack[UIO_FASTIOV]; 52201881fba8SOmar Sandoval struct iovec *iov = iovstack; 52211881fba8SOmar Sandoval struct iov_iter iter; 52221881fba8SOmar Sandoval loff_t pos; 52231881fba8SOmar Sandoval struct kiocb kiocb; 52241881fba8SOmar Sandoval ssize_t ret; 52251881fba8SOmar Sandoval 52261881fba8SOmar Sandoval if (!capable(CAP_SYS_ADMIN)) { 52271881fba8SOmar Sandoval ret = -EPERM; 52281881fba8SOmar Sandoval goto out_acct; 52291881fba8SOmar Sandoval } 52301881fba8SOmar Sandoval 52311881fba8SOmar Sandoval if (compat) { 52321881fba8SOmar Sandoval #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 52331881fba8SOmar Sandoval struct btrfs_ioctl_encoded_io_args_32 args32; 52341881fba8SOmar Sandoval 52351881fba8SOmar Sandoval copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32, 52361881fba8SOmar Sandoval flags); 52371881fba8SOmar Sandoval if (copy_from_user(&args32, argp, copy_end)) { 52381881fba8SOmar Sandoval ret = -EFAULT; 52391881fba8SOmar Sandoval goto out_acct; 52401881fba8SOmar Sandoval } 52411881fba8SOmar Sandoval args.iov = compat_ptr(args32.iov); 52421881fba8SOmar Sandoval args.iovcnt = args32.iovcnt; 52431881fba8SOmar Sandoval args.offset = args32.offset; 52441881fba8SOmar Sandoval args.flags = args32.flags; 52451881fba8SOmar Sandoval #else 52461881fba8SOmar Sandoval return -ENOTTY; 52471881fba8SOmar Sandoval #endif 52481881fba8SOmar Sandoval } else { 52491881fba8SOmar Sandoval copy_end = copy_end_kernel; 52501881fba8SOmar Sandoval if (copy_from_user(&args, argp, copy_end)) { 52511881fba8SOmar Sandoval ret = -EFAULT; 52521881fba8SOmar Sandoval goto out_acct; 52531881fba8SOmar Sandoval } 52541881fba8SOmar Sandoval } 52551881fba8SOmar Sandoval if (args.flags != 0) { 52561881fba8SOmar Sandoval ret = -EINVAL; 52571881fba8SOmar Sandoval goto out_acct; 52581881fba8SOmar Sandoval } 52591881fba8SOmar Sandoval 52601881fba8SOmar Sandoval ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack), 52611881fba8SOmar Sandoval &iov, &iter); 52621881fba8SOmar Sandoval if (ret < 0) 52631881fba8SOmar Sandoval goto out_acct; 52641881fba8SOmar Sandoval 52651881fba8SOmar Sandoval if (iov_iter_count(&iter) == 0) { 52661881fba8SOmar Sandoval ret = 0; 52671881fba8SOmar Sandoval goto out_iov; 52681881fba8SOmar Sandoval } 52691881fba8SOmar Sandoval pos = args.offset; 52701881fba8SOmar Sandoval ret = rw_verify_area(READ, file, &pos, args.len); 52711881fba8SOmar Sandoval if (ret < 0) 52721881fba8SOmar Sandoval goto out_iov; 52731881fba8SOmar Sandoval 52741881fba8SOmar Sandoval init_sync_kiocb(&kiocb, file); 52751881fba8SOmar Sandoval kiocb.ki_pos = pos; 52761881fba8SOmar Sandoval 52771881fba8SOmar Sandoval ret = btrfs_encoded_read(&kiocb, &iter, &args); 52781881fba8SOmar Sandoval if (ret >= 0) { 52791881fba8SOmar Sandoval fsnotify_access(file); 52801881fba8SOmar Sandoval if (copy_to_user(argp + copy_end, 52811881fba8SOmar Sandoval (char *)&args + copy_end_kernel, 52821881fba8SOmar Sandoval sizeof(args) - copy_end_kernel)) 52831881fba8SOmar Sandoval ret = -EFAULT; 52841881fba8SOmar Sandoval } 52851881fba8SOmar Sandoval 52861881fba8SOmar Sandoval out_iov: 52871881fba8SOmar Sandoval kfree(iov); 52881881fba8SOmar Sandoval out_acct: 52891881fba8SOmar Sandoval if (ret > 0) 52901881fba8SOmar Sandoval add_rchar(current, ret); 52911881fba8SOmar Sandoval inc_syscr(current); 52921881fba8SOmar Sandoval return ret; 52931881fba8SOmar Sandoval } 52941881fba8SOmar Sandoval 52957c0c7269SOmar Sandoval static int btrfs_ioctl_encoded_write(struct file *file, void __user *argp, bool compat) 52967c0c7269SOmar Sandoval { 52977c0c7269SOmar Sandoval struct btrfs_ioctl_encoded_io_args args; 52987c0c7269SOmar Sandoval struct iovec iovstack[UIO_FASTIOV]; 52997c0c7269SOmar Sandoval struct iovec *iov = iovstack; 53007c0c7269SOmar Sandoval struct iov_iter iter; 53017c0c7269SOmar Sandoval loff_t pos; 53027c0c7269SOmar Sandoval struct kiocb kiocb; 53037c0c7269SOmar Sandoval ssize_t ret; 53047c0c7269SOmar Sandoval 53057c0c7269SOmar Sandoval if (!capable(CAP_SYS_ADMIN)) { 53067c0c7269SOmar Sandoval ret = -EPERM; 53077c0c7269SOmar Sandoval goto out_acct; 53087c0c7269SOmar Sandoval } 53097c0c7269SOmar Sandoval 53107c0c7269SOmar Sandoval if (!(file->f_mode & FMODE_WRITE)) { 53117c0c7269SOmar Sandoval ret = -EBADF; 53127c0c7269SOmar Sandoval goto out_acct; 53137c0c7269SOmar Sandoval } 53147c0c7269SOmar Sandoval 53157c0c7269SOmar Sandoval if (compat) { 53167c0c7269SOmar Sandoval #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 53177c0c7269SOmar Sandoval struct btrfs_ioctl_encoded_io_args_32 args32; 53187c0c7269SOmar Sandoval 53197c0c7269SOmar Sandoval if (copy_from_user(&args32, argp, sizeof(args32))) { 53207c0c7269SOmar Sandoval ret = -EFAULT; 53217c0c7269SOmar Sandoval goto out_acct; 53227c0c7269SOmar Sandoval } 53237c0c7269SOmar Sandoval args.iov = compat_ptr(args32.iov); 53247c0c7269SOmar Sandoval args.iovcnt = args32.iovcnt; 53257c0c7269SOmar Sandoval args.offset = args32.offset; 53267c0c7269SOmar Sandoval args.flags = args32.flags; 53277c0c7269SOmar Sandoval args.len = args32.len; 53287c0c7269SOmar Sandoval args.unencoded_len = args32.unencoded_len; 53297c0c7269SOmar Sandoval args.unencoded_offset = args32.unencoded_offset; 53307c0c7269SOmar Sandoval args.compression = args32.compression; 53317c0c7269SOmar Sandoval args.encryption = args32.encryption; 53327c0c7269SOmar Sandoval memcpy(args.reserved, args32.reserved, sizeof(args.reserved)); 53337c0c7269SOmar Sandoval #else 53347c0c7269SOmar Sandoval return -ENOTTY; 53357c0c7269SOmar Sandoval #endif 53367c0c7269SOmar Sandoval } else { 53377c0c7269SOmar Sandoval if (copy_from_user(&args, argp, sizeof(args))) { 53387c0c7269SOmar Sandoval ret = -EFAULT; 53397c0c7269SOmar Sandoval goto out_acct; 53407c0c7269SOmar Sandoval } 53417c0c7269SOmar Sandoval } 53427c0c7269SOmar Sandoval 53437c0c7269SOmar Sandoval ret = -EINVAL; 53447c0c7269SOmar Sandoval if (args.flags != 0) 53457c0c7269SOmar Sandoval goto out_acct; 53467c0c7269SOmar Sandoval if (memchr_inv(args.reserved, 0, sizeof(args.reserved))) 53477c0c7269SOmar Sandoval goto out_acct; 53487c0c7269SOmar Sandoval if (args.compression == BTRFS_ENCODED_IO_COMPRESSION_NONE && 53497c0c7269SOmar Sandoval args.encryption == BTRFS_ENCODED_IO_ENCRYPTION_NONE) 53507c0c7269SOmar Sandoval goto out_acct; 53517c0c7269SOmar Sandoval if (args.compression >= BTRFS_ENCODED_IO_COMPRESSION_TYPES || 53527c0c7269SOmar Sandoval args.encryption >= BTRFS_ENCODED_IO_ENCRYPTION_TYPES) 53537c0c7269SOmar Sandoval goto out_acct; 53547c0c7269SOmar Sandoval if (args.unencoded_offset > args.unencoded_len) 53557c0c7269SOmar Sandoval goto out_acct; 53567c0c7269SOmar Sandoval if (args.len > args.unencoded_len - args.unencoded_offset) 53577c0c7269SOmar Sandoval goto out_acct; 53587c0c7269SOmar Sandoval 53597c0c7269SOmar Sandoval ret = import_iovec(WRITE, args.iov, args.iovcnt, ARRAY_SIZE(iovstack), 53607c0c7269SOmar Sandoval &iov, &iter); 53617c0c7269SOmar Sandoval if (ret < 0) 53627c0c7269SOmar Sandoval goto out_acct; 53637c0c7269SOmar Sandoval 53647c0c7269SOmar Sandoval file_start_write(file); 53657c0c7269SOmar Sandoval 53667c0c7269SOmar Sandoval if (iov_iter_count(&iter) == 0) { 53677c0c7269SOmar Sandoval ret = 0; 53687c0c7269SOmar Sandoval goto out_end_write; 53697c0c7269SOmar Sandoval } 53707c0c7269SOmar Sandoval pos = args.offset; 53717c0c7269SOmar Sandoval ret = rw_verify_area(WRITE, file, &pos, args.len); 53727c0c7269SOmar Sandoval if (ret < 0) 53737c0c7269SOmar Sandoval goto out_end_write; 53747c0c7269SOmar Sandoval 53757c0c7269SOmar Sandoval init_sync_kiocb(&kiocb, file); 53767c0c7269SOmar Sandoval ret = kiocb_set_rw_flags(&kiocb, 0); 53777c0c7269SOmar Sandoval if (ret) 53787c0c7269SOmar Sandoval goto out_end_write; 53797c0c7269SOmar Sandoval kiocb.ki_pos = pos; 53807c0c7269SOmar Sandoval 53817c0c7269SOmar Sandoval ret = btrfs_do_write_iter(&kiocb, &iter, &args); 53827c0c7269SOmar Sandoval if (ret > 0) 53837c0c7269SOmar Sandoval fsnotify_modify(file); 53847c0c7269SOmar Sandoval 53857c0c7269SOmar Sandoval out_end_write: 53867c0c7269SOmar Sandoval file_end_write(file); 53877c0c7269SOmar Sandoval kfree(iov); 53887c0c7269SOmar Sandoval out_acct: 53897c0c7269SOmar Sandoval if (ret > 0) 53907c0c7269SOmar Sandoval add_wchar(current, ret); 53917c0c7269SOmar Sandoval inc_syscw(current); 53927c0c7269SOmar Sandoval return ret; 53937c0c7269SOmar Sandoval } 53947c0c7269SOmar Sandoval 5395f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 5396f46b5a66SChristoph Hellwig cmd, unsigned long arg) 5397f46b5a66SChristoph Hellwig { 53980b246afaSJeff Mahoney struct inode *inode = file_inode(file); 53990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 54000b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 54014bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 5402f46b5a66SChristoph Hellwig 5403f46b5a66SChristoph Hellwig switch (cmd) { 54046cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 5405dc408ccdSSahil Kang return btrfs_ioctl_getversion(inode, argp); 540640cf931fSEric Sandeen case FS_IOC_GETFSLABEL: 5407b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_fslabel(fs_info, argp); 540840cf931fSEric Sandeen case FS_IOC_SETFSLABEL: 540940cf931fSEric Sandeen return btrfs_ioctl_set_fslabel(file, argp); 5410f7039b1dSLi Dongyang case FITRIM: 5411b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_fitrim(fs_info, argp); 5412f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 5413fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 5414fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 5415fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 54163de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 5417fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 54186f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 54196f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 542076dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 5421949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, false); 5422949964c9SMarcos Paulo de Souza case BTRFS_IOC_SNAP_DESTROY_V2: 5423949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, true); 54240caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 54259ad12305SSahil Kang return btrfs_ioctl_subvol_getflags(inode, argp); 54260caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 54270caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 54286ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 54296ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 5430f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 54311e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 54321e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 54331e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 5434f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 5435198605a8SMiao Xie return btrfs_ioctl_resize(file, argp); 5436f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 54372ff7e61eSJeff Mahoney return btrfs_ioctl_add_dev(fs_info, argp); 5438f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 5439da24927bSMiao Xie return btrfs_ioctl_rm_dev(file, argp); 54406b526ed7SAnand Jain case BTRFS_IOC_RM_DEV_V2: 54416b526ed7SAnand Jain return btrfs_ioctl_rm_dev_v2(file, argp); 5442475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 54432ff7e61eSJeff Mahoney return btrfs_ioctl_fs_info(fs_info, argp); 5444475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 54452ff7e61eSJeff Mahoney return btrfs_ioctl_dev_info(fs_info, argp); 5446ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 54479ad12305SSahil Kang return btrfs_ioctl_tree_search(inode, argp); 5448cc68a8a5SGerhard Heift case BTRFS_IOC_TREE_SEARCH_V2: 54499ad12305SSahil Kang return btrfs_ioctl_tree_search_v2(inode, argp); 5450ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 5451dc408ccdSSahil Kang return btrfs_ioctl_ino_lookup(root, argp); 5452d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 5453d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 5454d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 5455d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 1); 5456d24a67b2SZygo Blaxell case BTRFS_IOC_LOGICAL_INO_V2: 5457d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 2); 54581406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 54592ff7e61eSJeff Mahoney return btrfs_ioctl_space_info(fs_info, argp); 54609b199859SFilipe David Borba Manana case BTRFS_IOC_SYNC: { 54619b199859SFilipe David Borba Manana int ret; 54629b199859SFilipe David Borba Manana 54639db4dc24SNikolay Borisov ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); 54649b199859SFilipe David Borba Manana if (ret) 54659b199859SFilipe David Borba Manana return ret; 54660b246afaSJeff Mahoney ret = btrfs_sync_fs(inode->i_sb, 1); 54672fad4e83SDavid Sterba /* 54682fad4e83SDavid Sterba * The transaction thread may want to do more work, 546901327610SNicholas D Steeves * namely it pokes the cleaner kthread that will start 54702fad4e83SDavid Sterba * processing uncleaned subvols. 54712fad4e83SDavid Sterba */ 54720b246afaSJeff Mahoney wake_up_process(fs_info->transaction_kthread); 54739b199859SFilipe David Borba Manana return ret; 54749b199859SFilipe David Borba Manana } 547546204592SSage Weil case BTRFS_IOC_START_SYNC: 54769a8c28beSMiao Xie return btrfs_ioctl_start_sync(root, argp); 547746204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 54782ff7e61eSJeff Mahoney return btrfs_ioctl_wait_sync(fs_info, argp); 5479475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 5480b8e95489SMiao Xie return btrfs_ioctl_scrub(file, argp); 5481475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 54822ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_cancel(fs_info); 5483475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 54842ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_progress(fs_info, argp); 5485c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 54869ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 5487837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 54882ff7e61eSJeff Mahoney return btrfs_ioctl_balance_ctl(fs_info, arg); 548919a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 54902ff7e61eSJeff Mahoney return btrfs_ioctl_balance_progress(fs_info, argp); 54918ea05e3aSAlexander Block case BTRFS_IOC_SET_RECEIVED_SUBVOL: 54928ea05e3aSAlexander Block return btrfs_ioctl_set_received_subvol(file, argp); 5493abccd00fSHugo Mills #ifdef CONFIG_64BIT 5494abccd00fSHugo Mills case BTRFS_IOC_SET_RECEIVED_SUBVOL_32: 5495abccd00fSHugo Mills return btrfs_ioctl_set_received_subvol_32(file, argp); 5496abccd00fSHugo Mills #endif 549731db9f7cSAlexander Block case BTRFS_IOC_SEND: 54989ad12305SSahil Kang return _btrfs_ioctl_send(inode, argp, false); 54992351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 55002351f431SJosef Bacik case BTRFS_IOC_SEND_32: 55019ad12305SSahil Kang return _btrfs_ioctl_send(inode, argp, true); 55022351f431SJosef Bacik #endif 5503c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 55042ff7e61eSJeff Mahoney return btrfs_ioctl_get_dev_stats(fs_info, argp); 55055d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 5506905b0ddaSMiao Xie return btrfs_ioctl_quota_ctl(file, argp); 55075d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 5508905b0ddaSMiao Xie return btrfs_ioctl_qgroup_assign(file, argp); 55095d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 5510905b0ddaSMiao Xie return btrfs_ioctl_qgroup_create(file, argp); 55115d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 5512905b0ddaSMiao Xie return btrfs_ioctl_qgroup_limit(file, argp); 55132f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN: 55142f232036SJan Schmidt return btrfs_ioctl_quota_rescan(file, argp); 55152f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_STATUS: 5516b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_status(fs_info, argp); 551757254b6eSJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_WAIT: 5518b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_wait(fs_info, argp); 55193f6bcfbdSStefan Behrens case BTRFS_IOC_DEV_REPLACE: 55202ff7e61eSJeff Mahoney return btrfs_ioctl_dev_replace(fs_info, argp); 55212eaa055fSJeff Mahoney case BTRFS_IOC_GET_SUPPORTED_FEATURES: 5522d5131b65SDavid Sterba return btrfs_ioctl_get_supported_features(argp); 55232eaa055fSJeff Mahoney case BTRFS_IOC_GET_FEATURES: 5524b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_features(fs_info, argp); 55252eaa055fSJeff Mahoney case BTRFS_IOC_SET_FEATURES: 55262eaa055fSJeff Mahoney return btrfs_ioctl_set_features(file, argp); 5527b64ec075STomohiro Misono case BTRFS_IOC_GET_SUBVOL_INFO: 55289ad12305SSahil Kang return btrfs_ioctl_get_subvol_info(inode, argp); 552942e4b520STomohiro Misono case BTRFS_IOC_GET_SUBVOL_ROOTREF: 5530dc408ccdSSahil Kang return btrfs_ioctl_get_subvol_rootref(root, argp); 553123d0b79dSTomohiro Misono case BTRFS_IOC_INO_LOOKUP_USER: 553223d0b79dSTomohiro Misono return btrfs_ioctl_ino_lookup_user(file, argp); 553314605409SBoris Burkov case FS_IOC_ENABLE_VERITY: 553414605409SBoris Burkov return fsverity_ioctl_enable(file, (const void __user *)argp); 553514605409SBoris Burkov case FS_IOC_MEASURE_VERITY: 553614605409SBoris Burkov return fsverity_ioctl_measure(file, argp); 55371881fba8SOmar Sandoval case BTRFS_IOC_ENCODED_READ: 55381881fba8SOmar Sandoval return btrfs_ioctl_encoded_read(file, argp, false); 55397c0c7269SOmar Sandoval case BTRFS_IOC_ENCODED_WRITE: 55407c0c7269SOmar Sandoval return btrfs_ioctl_encoded_write(file, argp, false); 55411881fba8SOmar Sandoval #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 55421881fba8SOmar Sandoval case BTRFS_IOC_ENCODED_READ_32: 55431881fba8SOmar Sandoval return btrfs_ioctl_encoded_read(file, argp, true); 55447c0c7269SOmar Sandoval case BTRFS_IOC_ENCODED_WRITE_32: 55457c0c7269SOmar Sandoval return btrfs_ioctl_encoded_write(file, argp, true); 55461881fba8SOmar Sandoval #endif 5547f46b5a66SChristoph Hellwig } 5548f46b5a66SChristoph Hellwig 5549f46b5a66SChristoph Hellwig return -ENOTTY; 5550f46b5a66SChristoph Hellwig } 55514c63c245SLuke Dashjr 55524c63c245SLuke Dashjr #ifdef CONFIG_COMPAT 55534c63c245SLuke Dashjr long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 55544c63c245SLuke Dashjr { 55552a362249SJeff Mahoney /* 55562a362249SJeff Mahoney * These all access 32-bit values anyway so no further 55572a362249SJeff Mahoney * handling is necessary. 55582a362249SJeff Mahoney */ 55594c63c245SLuke Dashjr switch (cmd) { 55604c63c245SLuke Dashjr case FS_IOC32_GETVERSION: 55614c63c245SLuke Dashjr cmd = FS_IOC_GETVERSION; 55624c63c245SLuke Dashjr break; 55634c63c245SLuke Dashjr } 55644c63c245SLuke Dashjr 55654c63c245SLuke Dashjr return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 55664c63c245SLuke Dashjr } 55674c63c245SLuke Dashjr #endif 5568