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> 31f46b5a66SChristoph Hellwig #include "ctree.h" 32f46b5a66SChristoph Hellwig #include "disk-io.h" 33949964c9SMarcos Paulo de Souza #include "export.h" 34f46b5a66SChristoph Hellwig #include "transaction.h" 35f46b5a66SChristoph Hellwig #include "btrfs_inode.h" 36f46b5a66SChristoph Hellwig #include "print-tree.h" 37f46b5a66SChristoph Hellwig #include "volumes.h" 38925baeddSChris Mason #include "locking.h" 39d7728c96SJan Schmidt #include "backref.h" 40606686eeSJosef Bacik #include "rcu-string.h" 4131db9f7cSAlexander Block #include "send.h" 423f6bcfbdSStefan Behrens #include "dev-replace.h" 4363541927SFilipe David Borba Manana #include "props.h" 443b02a68aSJeff Mahoney #include "sysfs.h" 45fcebe456SJosef Bacik #include "qgroup.h" 461ec9a1aeSFilipe Manana #include "tree-log.h" 47ebb8765bSAnand Jain #include "compression.h" 488719aaaeSJosef Bacik #include "space-info.h" 4986736342SJosef Bacik #include "delalloc-space.h" 50aac0023cSJosef Bacik #include "block-group.h" 5122b398eeSQu Wenruo #include "subpage.h" 52f46b5a66SChristoph Hellwig 53abccd00fSHugo Mills #ifdef CONFIG_64BIT 54abccd00fSHugo Mills /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI 55abccd00fSHugo Mills * structures are incorrect, as the timespec structure from userspace 56abccd00fSHugo Mills * is 4 bytes too small. We define these alternatives here to teach 57abccd00fSHugo Mills * the kernel about the 32-bit struct packing. 58abccd00fSHugo Mills */ 59abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 { 60abccd00fSHugo Mills __u64 sec; 61abccd00fSHugo Mills __u32 nsec; 62abccd00fSHugo Mills } __attribute__ ((__packed__)); 63abccd00fSHugo Mills 64abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 { 65abccd00fSHugo Mills char uuid[BTRFS_UUID_SIZE]; /* in */ 66abccd00fSHugo Mills __u64 stransid; /* in */ 67abccd00fSHugo Mills __u64 rtransid; /* out */ 68abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 stime; /* in */ 69abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 rtime; /* out */ 70abccd00fSHugo Mills __u64 flags; /* in */ 71abccd00fSHugo Mills __u64 reserved[16]; /* in */ 72abccd00fSHugo Mills } __attribute__ ((__packed__)); 73abccd00fSHugo Mills 74abccd00fSHugo Mills #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \ 75abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32) 76abccd00fSHugo Mills #endif 77abccd00fSHugo Mills 782351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 792351f431SJosef Bacik struct btrfs_ioctl_send_args_32 { 802351f431SJosef Bacik __s64 send_fd; /* in */ 812351f431SJosef Bacik __u64 clone_sources_count; /* in */ 822351f431SJosef Bacik compat_uptr_t clone_sources; /* in */ 832351f431SJosef Bacik __u64 parent_root; /* in */ 842351f431SJosef Bacik __u64 flags; /* in */ 85e77fbf99SDavid Sterba __u32 version; /* in */ 86e77fbf99SDavid Sterba __u8 reserved[28]; /* in */ 872351f431SJosef Bacik } __attribute__ ((__packed__)); 882351f431SJosef Bacik 892351f431SJosef Bacik #define BTRFS_IOC_SEND_32 _IOW(BTRFS_IOCTL_MAGIC, 38, \ 902351f431SJosef Bacik struct btrfs_ioctl_send_args_32) 912351f431SJosef Bacik #endif 92abccd00fSHugo Mills 936cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 941905a0f7SDavid Sterba static unsigned int btrfs_mask_fsflags_for_type(struct inode *inode, 951905a0f7SDavid Sterba unsigned int flags) 966cbff00fSChristoph Hellwig { 971905a0f7SDavid Sterba if (S_ISDIR(inode->i_mode)) 986cbff00fSChristoph Hellwig return flags; 991905a0f7SDavid Sterba else if (S_ISREG(inode->i_mode)) 1006cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 1016cbff00fSChristoph Hellwig else 1026cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 1036cbff00fSChristoph Hellwig } 104f46b5a66SChristoph Hellwig 1056cbff00fSChristoph Hellwig /* 106a157d4fdSDavid Sterba * Export internal inode flags to the format expected by the FS_IOC_GETFLAGS 107a157d4fdSDavid Sterba * ioctl. 1086cbff00fSChristoph Hellwig */ 10977eea05eSBoris Burkov static unsigned int btrfs_inode_flags_to_fsflags(struct btrfs_inode *binode) 1106cbff00fSChristoph Hellwig { 1116cbff00fSChristoph Hellwig unsigned int iflags = 0; 11277eea05eSBoris Burkov u32 flags = binode->flags; 11314605409SBoris Burkov u32 ro_flags = binode->ro_flags; 1146cbff00fSChristoph Hellwig 1156cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 1166cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 1176cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 1186cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 1196cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 1206cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 1216cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 1226cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 1236cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 1246cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 1256cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 1266cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 127d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 128d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 12914605409SBoris Burkov if (ro_flags & BTRFS_INODE_RO_VERITY) 13014605409SBoris Burkov iflags |= FS_VERITY_FL; 131d0092bddSLi Zefan 13213f48dc9SSatoru Takeuchi if (flags & BTRFS_INODE_NOCOMPRESS) 133d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 13413f48dc9SSatoru Takeuchi else if (flags & BTRFS_INODE_COMPRESS) 13513f48dc9SSatoru Takeuchi iflags |= FS_COMPR_FL; 1366cbff00fSChristoph Hellwig 1376cbff00fSChristoph Hellwig return iflags; 1386cbff00fSChristoph Hellwig } 1396cbff00fSChristoph Hellwig 1406cbff00fSChristoph Hellwig /* 1416cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 1426cbff00fSChristoph Hellwig */ 1437b6a221eSDavid Sterba void btrfs_sync_inode_flags_to_i_flags(struct inode *inode) 1446cbff00fSChristoph Hellwig { 1455c57b8b6SDavid Sterba struct btrfs_inode *binode = BTRFS_I(inode); 1463cc79392SFilipe Manana unsigned int new_fl = 0; 1476cbff00fSChristoph Hellwig 1485c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_SYNC) 1493cc79392SFilipe Manana new_fl |= S_SYNC; 1505c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_IMMUTABLE) 1513cc79392SFilipe Manana new_fl |= S_IMMUTABLE; 1525c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_APPEND) 1533cc79392SFilipe Manana new_fl |= S_APPEND; 1545c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_NOATIME) 1553cc79392SFilipe Manana new_fl |= S_NOATIME; 1565c57b8b6SDavid Sterba if (binode->flags & BTRFS_INODE_DIRSYNC) 1573cc79392SFilipe Manana new_fl |= S_DIRSYNC; 15814605409SBoris Burkov if (binode->ro_flags & BTRFS_INODE_RO_VERITY) 15914605409SBoris Burkov new_fl |= S_VERITY; 1603cc79392SFilipe Manana 1613cc79392SFilipe Manana set_mask_bits(&inode->i_flags, 16214605409SBoris Burkov S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC | 16314605409SBoris Burkov S_VERITY, new_fl); 1646cbff00fSChristoph Hellwig } 1656cbff00fSChristoph Hellwig 166f37c563bSDavid Sterba /* 167f37c563bSDavid Sterba * Check if @flags are a supported and valid set of FS_*_FL flags and that 168f37c563bSDavid Sterba * the old and new flags are not conflicting 169f37c563bSDavid Sterba */ 170f37c563bSDavid Sterba static int check_fsflags(unsigned int old_flags, unsigned int flags) 17175e7cb7fSLiu Bo { 17275e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 17375e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 17475e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 175e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 176e1e8fb6aSLi Zefan FS_NOCOW_FL)) 17775e7cb7fSLiu Bo return -EOPNOTSUPP; 17875e7cb7fSLiu Bo 179f37c563bSDavid Sterba /* COMPR and NOCOMP on new/old are valid */ 18075e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 18175e7cb7fSLiu Bo return -EINVAL; 18275e7cb7fSLiu Bo 183f37c563bSDavid Sterba if ((flags & FS_COMPR_FL) && (flags & FS_NOCOW_FL)) 184f37c563bSDavid Sterba return -EINVAL; 185f37c563bSDavid Sterba 186f37c563bSDavid Sterba /* NOCOW and compression options are mutually exclusive */ 187f37c563bSDavid Sterba if ((old_flags & FS_NOCOW_FL) && (flags & (FS_COMPR_FL | FS_NOCOMP_FL))) 188f37c563bSDavid Sterba return -EINVAL; 189f37c563bSDavid Sterba if ((flags & FS_NOCOW_FL) && (old_flags & (FS_COMPR_FL | FS_NOCOMP_FL))) 190f37c563bSDavid Sterba return -EINVAL; 191f37c563bSDavid Sterba 19275e7cb7fSLiu Bo return 0; 19375e7cb7fSLiu Bo } 19475e7cb7fSLiu Bo 195d206e9c9SNaohiro Aota static int check_fsflags_compatible(struct btrfs_fs_info *fs_info, 196d206e9c9SNaohiro Aota unsigned int flags) 197d206e9c9SNaohiro Aota { 198d206e9c9SNaohiro Aota if (btrfs_is_zoned(fs_info) && (flags & FS_NOCOW_FL)) 199d206e9c9SNaohiro Aota return -EPERM; 200d206e9c9SNaohiro Aota 201d206e9c9SNaohiro Aota return 0; 202d206e9c9SNaohiro Aota } 203d206e9c9SNaohiro Aota 20497fc2977SMiklos Szeredi /* 20597fc2977SMiklos Szeredi * Set flags/xflags from the internal inode flags. The remaining items of 20697fc2977SMiklos Szeredi * fsxattr are zeroed. 20797fc2977SMiklos Szeredi */ 20897fc2977SMiklos Szeredi int btrfs_fileattr_get(struct dentry *dentry, struct fileattr *fa) 2096cbff00fSChristoph Hellwig { 21097fc2977SMiklos Szeredi struct btrfs_inode *binode = BTRFS_I(d_inode(dentry)); 21197fc2977SMiklos Szeredi 21277eea05eSBoris Burkov fileattr_fill_flags(fa, btrfs_inode_flags_to_fsflags(binode)); 21397fc2977SMiklos Szeredi return 0; 21497fc2977SMiklos Szeredi } 21597fc2977SMiklos Szeredi 21697fc2977SMiklos Szeredi int btrfs_fileattr_set(struct user_namespace *mnt_userns, 21797fc2977SMiklos Szeredi struct dentry *dentry, struct fileattr *fa) 21897fc2977SMiklos Szeredi { 21997fc2977SMiklos Szeredi struct inode *inode = d_inode(dentry); 2200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 2215c57b8b6SDavid Sterba struct btrfs_inode *binode = BTRFS_I(inode); 2225c57b8b6SDavid Sterba struct btrfs_root *root = binode->root; 2236cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 2245aca2842SDarrick J. Wong unsigned int fsflags, old_fsflags; 2256cbff00fSChristoph Hellwig int ret; 226ff9fef55SAnand Jain const char *comp = NULL; 227f37c563bSDavid Sterba u32 binode_flags; 2286cbff00fSChristoph Hellwig 229b83cc969SLi Zefan if (btrfs_root_readonly(root)) 230b83cc969SLi Zefan return -EROFS; 231b83cc969SLi Zefan 23297fc2977SMiklos Szeredi if (fileattr_has_fsx(fa)) 23397fc2977SMiklos Szeredi return -EOPNOTSUPP; 2346cbff00fSChristoph Hellwig 23597fc2977SMiklos Szeredi fsflags = btrfs_mask_fsflags_for_type(inode, fa->flags); 23677eea05eSBoris Burkov old_fsflags = btrfs_inode_flags_to_fsflags(binode); 23797fc2977SMiklos Szeredi ret = check_fsflags(old_fsflags, fsflags); 238e7848683SJan Kara if (ret) 239e7848683SJan Kara return ret; 240e7848683SJan Kara 241d206e9c9SNaohiro Aota ret = check_fsflags_compatible(fs_info, fsflags); 242d206e9c9SNaohiro Aota if (ret) 24397fc2977SMiklos Szeredi return ret; 244d206e9c9SNaohiro Aota 245f37c563bSDavid Sterba binode_flags = binode->flags; 2465c57b8b6SDavid Sterba if (fsflags & FS_SYNC_FL) 247d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_SYNC; 2486cbff00fSChristoph Hellwig else 249d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_SYNC; 2505c57b8b6SDavid Sterba if (fsflags & FS_IMMUTABLE_FL) 251d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_IMMUTABLE; 2526cbff00fSChristoph Hellwig else 253d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_IMMUTABLE; 2545c57b8b6SDavid Sterba if (fsflags & FS_APPEND_FL) 255d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_APPEND; 2566cbff00fSChristoph Hellwig else 257d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_APPEND; 2585c57b8b6SDavid Sterba if (fsflags & FS_NODUMP_FL) 259d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODUMP; 2606cbff00fSChristoph Hellwig else 261d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NODUMP; 2625c57b8b6SDavid Sterba if (fsflags & FS_NOATIME_FL) 263d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NOATIME; 2646cbff00fSChristoph Hellwig else 265d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NOATIME; 26697fc2977SMiklos Szeredi 26797fc2977SMiklos Szeredi /* If coming from FS_IOC_FSSETXATTR then skip unconverted flags */ 26897fc2977SMiklos Szeredi if (!fa->flags_valid) { 26997fc2977SMiklos Szeredi /* 1 item for the inode */ 27097fc2977SMiklos Szeredi trans = btrfs_start_transaction(root, 1); 2719b8a233bSRitesh Harjani if (IS_ERR(trans)) 2729b8a233bSRitesh Harjani return PTR_ERR(trans); 27397fc2977SMiklos Szeredi goto update_flags; 27497fc2977SMiklos Szeredi } 27597fc2977SMiklos Szeredi 2765c57b8b6SDavid Sterba if (fsflags & FS_DIRSYNC_FL) 277d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_DIRSYNC; 2786cbff00fSChristoph Hellwig else 279d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_DIRSYNC; 2805c57b8b6SDavid Sterba if (fsflags & FS_NOCOW_FL) { 28144e5194bSAnand Jain if (S_ISREG(inode->i_mode)) { 2827e97b8daSDavid Sterba /* 2837e97b8daSDavid Sterba * It's safe to turn csums off here, no extents exist. 2847e97b8daSDavid Sterba * Otherwise we want the flag to reflect the real COW 2857e97b8daSDavid Sterba * status of the file and will not set it. 2867e97b8daSDavid Sterba */ 2877e97b8daSDavid Sterba if (inode->i_size == 0) 288d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODATACOW | 289d2b8fcfeSAnand Jain BTRFS_INODE_NODATASUM; 2907e97b8daSDavid Sterba } else { 291d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NODATACOW; 2927e97b8daSDavid Sterba } 2937e97b8daSDavid Sterba } else { 2947e97b8daSDavid Sterba /* 29501327610SNicholas D Steeves * Revert back under same assumptions as above 2967e97b8daSDavid Sterba */ 29744e5194bSAnand Jain if (S_ISREG(inode->i_mode)) { 2987e97b8daSDavid Sterba if (inode->i_size == 0) 299d2b8fcfeSAnand Jain binode_flags &= ~(BTRFS_INODE_NODATACOW | 300d2b8fcfeSAnand Jain BTRFS_INODE_NODATASUM); 3017e97b8daSDavid Sterba } else { 302d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NODATACOW; 3037e97b8daSDavid Sterba } 3047e97b8daSDavid Sterba } 3056cbff00fSChristoph Hellwig 30675e7cb7fSLiu Bo /* 30775e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 30875e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 30975e7cb7fSLiu Bo * things smaller. 31075e7cb7fSLiu Bo */ 3115c57b8b6SDavid Sterba if (fsflags & FS_NOCOMP_FL) { 312d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_COMPRESS; 313d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_NOCOMPRESS; 3145c57b8b6SDavid Sterba } else if (fsflags & FS_COMPR_FL) { 31563541927SFilipe David Borba Manana 31697fc2977SMiklos Szeredi if (IS_SWAPFILE(inode)) 31797fc2977SMiklos Szeredi return -ETXTBSY; 318eede2bf3SOmar Sandoval 319d2b8fcfeSAnand Jain binode_flags |= BTRFS_INODE_COMPRESS; 320d2b8fcfeSAnand Jain binode_flags &= ~BTRFS_INODE_NOCOMPRESS; 32163541927SFilipe David Borba Manana 32293370509SDavid Sterba comp = btrfs_compress_type2str(fs_info->compress_type); 32393370509SDavid Sterba if (!comp || comp[0] == 0) 32493370509SDavid Sterba comp = btrfs_compress_type2str(BTRFS_COMPRESS_ZLIB); 325ebcb904dSLi Zefan } else { 326d2b8fcfeSAnand Jain binode_flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 32775e7cb7fSLiu Bo } 3286cbff00fSChristoph Hellwig 329ff9fef55SAnand Jain /* 330ff9fef55SAnand Jain * 1 for inode item 331ff9fef55SAnand Jain * 2 for properties 332ff9fef55SAnand Jain */ 333ff9fef55SAnand Jain trans = btrfs_start_transaction(root, 3); 33497fc2977SMiklos Szeredi if (IS_ERR(trans)) 33597fc2977SMiklos Szeredi return PTR_ERR(trans); 3366cbff00fSChristoph Hellwig 337ff9fef55SAnand Jain if (comp) { 338ff9fef55SAnand Jain ret = btrfs_set_prop(trans, inode, "btrfs.compression", comp, 339ff9fef55SAnand Jain strlen(comp), 0); 340ff9fef55SAnand Jain if (ret) { 341ff9fef55SAnand Jain btrfs_abort_transaction(trans, ret); 342ff9fef55SAnand Jain goto out_end_trans; 343ff9fef55SAnand Jain } 344ff9fef55SAnand Jain } else { 345ff9fef55SAnand Jain ret = btrfs_set_prop(trans, inode, "btrfs.compression", NULL, 346ff9fef55SAnand Jain 0, 0); 347ff9fef55SAnand Jain if (ret && ret != -ENODATA) { 348ff9fef55SAnand Jain btrfs_abort_transaction(trans, ret); 349ff9fef55SAnand Jain goto out_end_trans; 350ff9fef55SAnand Jain } 351ff9fef55SAnand Jain } 352ff9fef55SAnand Jain 35397fc2977SMiklos Szeredi update_flags: 354d2b8fcfeSAnand Jain binode->flags = binode_flags; 3557b6a221eSDavid Sterba btrfs_sync_inode_flags_to_i_flags(inode); 3560c4d2d95SJosef Bacik inode_inc_iversion(inode); 357c2050a45SDeepa Dinamani inode->i_ctime = current_time(inode); 3589a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(inode)); 3596cbff00fSChristoph Hellwig 360ff9fef55SAnand Jain out_end_trans: 3613a45bb20SJeff Mahoney btrfs_end_transaction(trans); 3622d4e6f6aSliubo return ret; 3636cbff00fSChristoph Hellwig } 3646cbff00fSChristoph Hellwig 3650d7ed32cSDavid Sterba /* 3660d7ed32cSDavid Sterba * Start exclusive operation @type, return true on success 3670d7ed32cSDavid Sterba */ 368c3e1f96cSGoldwyn Rodrigues bool btrfs_exclop_start(struct btrfs_fs_info *fs_info, 369c3e1f96cSGoldwyn Rodrigues enum btrfs_exclusive_operation type) 370c3e1f96cSGoldwyn Rodrigues { 3710d7ed32cSDavid Sterba bool ret = false; 3720d7ed32cSDavid Sterba 3730d7ed32cSDavid Sterba spin_lock(&fs_info->super_lock); 3740d7ed32cSDavid Sterba if (fs_info->exclusive_operation == BTRFS_EXCLOP_NONE) { 3750d7ed32cSDavid Sterba fs_info->exclusive_operation = type; 3760d7ed32cSDavid Sterba ret = true; 3770d7ed32cSDavid Sterba } 3780d7ed32cSDavid Sterba spin_unlock(&fs_info->super_lock); 3790d7ed32cSDavid Sterba 3800d7ed32cSDavid Sterba return ret; 381c3e1f96cSGoldwyn Rodrigues } 382c3e1f96cSGoldwyn Rodrigues 383578bda9eSDavid Sterba /* 384578bda9eSDavid Sterba * Conditionally allow to enter the exclusive operation in case it's compatible 385578bda9eSDavid Sterba * with the running one. This must be paired with btrfs_exclop_start_unlock and 386578bda9eSDavid Sterba * btrfs_exclop_finish. 387578bda9eSDavid Sterba * 388578bda9eSDavid Sterba * Compatibility: 389578bda9eSDavid Sterba * - the same type is already running 390621a1ee1SNikolay Borisov * - when trying to add a device and balance has been paused 391578bda9eSDavid Sterba * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller 392578bda9eSDavid Sterba * must check the condition first that would allow none -> @type 393578bda9eSDavid Sterba */ 394578bda9eSDavid Sterba bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info, 395578bda9eSDavid Sterba enum btrfs_exclusive_operation type) 396578bda9eSDavid Sterba { 397578bda9eSDavid Sterba spin_lock(&fs_info->super_lock); 398621a1ee1SNikolay Borisov if (fs_info->exclusive_operation == type || 399621a1ee1SNikolay Borisov (fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED && 400621a1ee1SNikolay Borisov type == BTRFS_EXCLOP_DEV_ADD)) 401578bda9eSDavid Sterba return true; 402578bda9eSDavid Sterba 403578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 404578bda9eSDavid Sterba return false; 405578bda9eSDavid Sterba } 406578bda9eSDavid Sterba 407578bda9eSDavid Sterba void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info) 408578bda9eSDavid Sterba { 409578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 410578bda9eSDavid Sterba } 411578bda9eSDavid Sterba 412c3e1f96cSGoldwyn Rodrigues void btrfs_exclop_finish(struct btrfs_fs_info *fs_info) 413c3e1f96cSGoldwyn Rodrigues { 4140d7ed32cSDavid Sterba spin_lock(&fs_info->super_lock); 415c3e1f96cSGoldwyn Rodrigues WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE); 4160d7ed32cSDavid Sterba spin_unlock(&fs_info->super_lock); 41766a2823cSGoldwyn Rodrigues sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation"); 418c3e1f96cSGoldwyn Rodrigues } 419c3e1f96cSGoldwyn Rodrigues 420efc0e69cSNikolay Borisov void btrfs_exclop_balance(struct btrfs_fs_info *fs_info, 421efc0e69cSNikolay Borisov enum btrfs_exclusive_operation op) 422efc0e69cSNikolay Borisov { 423efc0e69cSNikolay Borisov switch (op) { 424efc0e69cSNikolay Borisov case BTRFS_EXCLOP_BALANCE_PAUSED: 425efc0e69cSNikolay Borisov spin_lock(&fs_info->super_lock); 426efc0e69cSNikolay Borisov ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE || 427efc0e69cSNikolay Borisov fs_info->exclusive_operation == BTRFS_EXCLOP_DEV_ADD); 428efc0e69cSNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE_PAUSED; 429efc0e69cSNikolay Borisov spin_unlock(&fs_info->super_lock); 430efc0e69cSNikolay Borisov break; 431efc0e69cSNikolay Borisov case BTRFS_EXCLOP_BALANCE: 432efc0e69cSNikolay Borisov spin_lock(&fs_info->super_lock); 433efc0e69cSNikolay Borisov ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); 434efc0e69cSNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; 435efc0e69cSNikolay Borisov spin_unlock(&fs_info->super_lock); 436efc0e69cSNikolay Borisov break; 437efc0e69cSNikolay Borisov default: 438efc0e69cSNikolay Borisov btrfs_warn(fs_info, 439efc0e69cSNikolay Borisov "invalid exclop balance operation %d requested", op); 440efc0e69cSNikolay Borisov } 441efc0e69cSNikolay Borisov } 442efc0e69cSNikolay Borisov 443dc408ccdSSahil Kang static int btrfs_ioctl_getversion(struct inode *inode, int __user *arg) 4446cbff00fSChristoph Hellwig { 4456cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 4466cbff00fSChristoph Hellwig } 447f46b5a66SChristoph Hellwig 448b929c1d8SMarcos Paulo de Souza static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info, 449b929c1d8SMarcos Paulo de Souza void __user *arg) 450f7039b1dSLi Dongyang { 451f7039b1dSLi Dongyang struct btrfs_device *device; 452f7039b1dSLi Dongyang struct request_queue *q; 453f7039b1dSLi Dongyang struct fstrim_range range; 454f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 455f7039b1dSLi Dongyang u64 num_devices = 0; 456f7039b1dSLi Dongyang int ret; 457f7039b1dSLi Dongyang 458f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 459f7039b1dSLi Dongyang return -EPERM; 460f7039b1dSLi Dongyang 461f35f06c3SFilipe Manana /* 4621cb3dc3fSNaohiro Aota * btrfs_trim_block_group() depends on space cache, which is not 4631cb3dc3fSNaohiro Aota * available in zoned filesystem. So, disallow fitrim on a zoned 4641cb3dc3fSNaohiro Aota * filesystem for now. 4651cb3dc3fSNaohiro Aota */ 4661cb3dc3fSNaohiro Aota if (btrfs_is_zoned(fs_info)) 4671cb3dc3fSNaohiro Aota return -EOPNOTSUPP; 4681cb3dc3fSNaohiro Aota 4691cb3dc3fSNaohiro Aota /* 470f35f06c3SFilipe Manana * If the fs is mounted with nologreplay, which requires it to be 471f35f06c3SFilipe Manana * mounted in RO mode as well, we can not allow discard on free space 472f35f06c3SFilipe Manana * inside block groups, because log trees refer to extents that are not 473f35f06c3SFilipe Manana * pinned in a block group's free space cache (pinning the extents is 474f35f06c3SFilipe Manana * precisely the first phase of replaying a log tree). 475f35f06c3SFilipe Manana */ 476f35f06c3SFilipe Manana if (btrfs_test_opt(fs_info, NOLOGREPLAY)) 477f35f06c3SFilipe Manana return -EROFS; 478f35f06c3SFilipe Manana 4791f78160cSXiao Guangrong rcu_read_lock(); 4801f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 4811f78160cSXiao Guangrong dev_list) { 482f7039b1dSLi Dongyang if (!device->bdev) 483f7039b1dSLi Dongyang continue; 484f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 485f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 486f7039b1dSLi Dongyang num_devices++; 48750d0446eSSeraphime Kirkovski minlen = min_t(u64, q->limits.discard_granularity, 488f7039b1dSLi Dongyang minlen); 489f7039b1dSLi Dongyang } 490f7039b1dSLi Dongyang } 4911f78160cSXiao Guangrong rcu_read_unlock(); 492f4c697e6SLukas Czerner 493f7039b1dSLi Dongyang if (!num_devices) 494f7039b1dSLi Dongyang return -EOPNOTSUPP; 495f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 496f7039b1dSLi Dongyang return -EFAULT; 4976ba9fc8eSQu Wenruo 4986ba9fc8eSQu Wenruo /* 4996ba9fc8eSQu Wenruo * NOTE: Don't truncate the range using super->total_bytes. Bytenr of 5006ba9fc8eSQu Wenruo * block group is in the logical address space, which can be any 5016ba9fc8eSQu Wenruo * sectorsize aligned bytenr in the range [0, U64_MAX]. 5026ba9fc8eSQu Wenruo */ 5036ba9fc8eSQu Wenruo if (range.len < fs_info->sb->s_blocksize) 504f4c697e6SLukas Czerner return -EINVAL; 505f7039b1dSLi Dongyang 506f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 5072ff7e61eSJeff Mahoney ret = btrfs_trim_fs(fs_info, &range); 508f7039b1dSLi Dongyang if (ret < 0) 509f7039b1dSLi Dongyang return ret; 510f7039b1dSLi Dongyang 511f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 512f7039b1dSLi Dongyang return -EFAULT; 513f7039b1dSLi Dongyang 514f7039b1dSLi Dongyang return 0; 515f7039b1dSLi Dongyang } 516f7039b1dSLi Dongyang 517e1f60a65SDavid Sterba int __pure btrfs_is_empty_uuid(u8 *uuid) 518dd5f9615SStefan Behrens { 51946e0f66aSChris Mason int i; 52046e0f66aSChris Mason 52146e0f66aSChris Mason for (i = 0; i < BTRFS_UUID_SIZE; i++) { 52246e0f66aSChris Mason if (uuid[i]) 52346e0f66aSChris Mason return 0; 52446e0f66aSChris Mason } 52546e0f66aSChris Mason return 1; 526dd5f9615SStefan Behrens } 527dd5f9615SStefan Behrens 5284d4340c9SChristian Brauner static noinline int create_subvol(struct user_namespace *mnt_userns, 5294d4340c9SChristian Brauner struct inode *dir, struct dentry *dentry, 53052f75f4fSDavid Sterba const char *name, int namelen, 5318696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 532f46b5a66SChristoph Hellwig { 5330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 534f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 535f46b5a66SChristoph Hellwig struct btrfs_key key; 53649a3c4d9SDavid Sterba struct btrfs_root_item *root_item; 537f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 538f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 539d5c12070SMiao Xie struct btrfs_root *root = BTRFS_I(dir)->root; 54076dda93cSYan, Zheng struct btrfs_root *new_root; 541d5c12070SMiao Xie struct btrfs_block_rsv block_rsv; 54295582b00SDeepa Dinamani struct timespec64 cur_time = current_time(dir); 5435662344bSTsutomu Itoh struct inode *inode; 544f46b5a66SChristoph Hellwig int ret; 5452dfb1e43SQu Wenruo dev_t anon_dev = 0; 546f46b5a66SChristoph Hellwig u64 objectid; 5473de4586cSChris Mason u64 index = 0; 548f46b5a66SChristoph Hellwig 54949a3c4d9SDavid Sterba root_item = kzalloc(sizeof(*root_item), GFP_KERNEL); 55049a3c4d9SDavid Sterba if (!root_item) 55149a3c4d9SDavid Sterba return -ENOMEM; 55249a3c4d9SDavid Sterba 553543068a2SNikolay Borisov ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid); 5542fbe8c8aSAl Viro if (ret) 55549a3c4d9SDavid Sterba goto fail_free; 5566a912213SJosef Bacik 5572dfb1e43SQu Wenruo ret = get_anon_bdev(&anon_dev); 5582dfb1e43SQu Wenruo if (ret < 0) 5592dfb1e43SQu Wenruo goto fail_free; 5602dfb1e43SQu Wenruo 561e09fe2d2SQu Wenruo /* 562e09fe2d2SQu Wenruo * Don't create subvolume whose level is not zero. Or qgroup will be 56301327610SNicholas D Steeves * screwed up since it assumes subvolume qgroup's level to be 0. 564e09fe2d2SQu Wenruo */ 56549a3c4d9SDavid Sterba if (btrfs_qgroup_level(objectid)) { 56649a3c4d9SDavid Sterba ret = -ENOSPC; 56749a3c4d9SDavid Sterba goto fail_free; 56849a3c4d9SDavid Sterba } 569e09fe2d2SQu Wenruo 570d5c12070SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 5719ed74f2dSJosef Bacik /* 572d5c12070SMiao Xie * The same as the snapshot creation, please see the comment 573d5c12070SMiao Xie * of create_snapshot(). 5749ed74f2dSJosef Bacik */ 575c4c129dbSGu JinXiang ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false); 576d5c12070SMiao Xie if (ret) 57749a3c4d9SDavid Sterba goto fail_free; 578f46b5a66SChristoph Hellwig 579d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 580d5c12070SMiao Xie if (IS_ERR(trans)) { 581d5c12070SMiao Xie ret = PTR_ERR(trans); 582e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 58349a3c4d9SDavid Sterba goto fail_free; 584d5c12070SMiao Xie } 585d5c12070SMiao Xie trans->block_rsv = &block_rsv; 586d5c12070SMiao Xie trans->bytes_reserved = block_rsv.size; 587f46b5a66SChristoph Hellwig 588a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit); 5896f72c7e2SArne Jansen if (ret) 5906f72c7e2SArne Jansen goto fail; 5916f72c7e2SArne Jansen 5929631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 5939631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 5948e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 5958e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 5968e8a1e31SJosef Bacik goto fail; 5978e8a1e31SJosef Bacik } 598f46b5a66SChristoph Hellwig 599f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 600f46b5a66SChristoph Hellwig 60149a3c4d9SDavid Sterba inode_item = &root_item->inode; 6023cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 6033cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 6043cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 605da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 6060b246afaSJeff Mahoney fs_info->nodesize); 6073cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 608f46b5a66SChristoph Hellwig 60949a3c4d9SDavid Sterba btrfs_set_root_flags(root_item, 0); 61049a3c4d9SDavid Sterba btrfs_set_root_limit(root_item, 0); 6113cae210fSQu Wenruo btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 61208fe4db1SLi Zefan 61349a3c4d9SDavid Sterba btrfs_set_root_bytenr(root_item, leaf->start); 61449a3c4d9SDavid Sterba btrfs_set_root_generation(root_item, trans->transid); 61549a3c4d9SDavid Sterba btrfs_set_root_level(root_item, 0); 61649a3c4d9SDavid Sterba btrfs_set_root_refs(root_item, 1); 61749a3c4d9SDavid Sterba btrfs_set_root_used(root_item, leaf->len); 61849a3c4d9SDavid Sterba btrfs_set_root_last_snapshot(root_item, 0); 619f46b5a66SChristoph Hellwig 62049a3c4d9SDavid Sterba btrfs_set_root_generation_v2(root_item, 62149a3c4d9SDavid Sterba btrfs_root_generation(root_item)); 622807fc790SAndy Shevchenko generate_random_guid(root_item->uuid); 62349a3c4d9SDavid Sterba btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec); 62449a3c4d9SDavid Sterba btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec); 62549a3c4d9SDavid Sterba root_item->ctime = root_item->otime; 62649a3c4d9SDavid Sterba btrfs_set_root_ctransid(root_item, trans->transid); 62749a3c4d9SDavid Sterba btrfs_set_root_otransid(root_item, trans->transid); 628f46b5a66SChristoph Hellwig 629925baeddSChris Mason btrfs_tree_unlock(leaf); 630f46b5a66SChristoph Hellwig 63169948022SNikolay Borisov btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID); 632f46b5a66SChristoph Hellwig 633f46b5a66SChristoph Hellwig key.objectid = objectid; 6345d4f98a2SYan Zheng key.offset = 0; 635962a298fSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 6360b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, &key, 63749a3c4d9SDavid Sterba root_item); 63867addf29SFilipe Manana if (ret) { 63967addf29SFilipe Manana /* 64067addf29SFilipe Manana * Since we don't abort the transaction in this case, free the 64167addf29SFilipe Manana * tree block so that we don't leak space and leave the 64267addf29SFilipe Manana * filesystem in an inconsistent state (an extent item in the 6437a163608SFilipe Manana * extent tree with a backreference for a root that does not 644212a58fdSFilipe Manana * exists). 64567addf29SFilipe Manana */ 646212a58fdSFilipe Manana btrfs_tree_lock(leaf); 647212a58fdSFilipe Manana btrfs_clean_tree_block(leaf); 648212a58fdSFilipe Manana btrfs_tree_unlock(leaf); 6497a163608SFilipe Manana btrfs_free_tree_block(trans, objectid, leaf, 0, 1); 65067addf29SFilipe Manana free_extent_buffer(leaf); 651f46b5a66SChristoph Hellwig goto fail; 65267addf29SFilipe Manana } 65367addf29SFilipe Manana 65467addf29SFilipe Manana free_extent_buffer(leaf); 65567addf29SFilipe Manana leaf = NULL; 656f46b5a66SChristoph Hellwig 65776dda93cSYan, Zheng key.offset = (u64)-1; 6582dfb1e43SQu Wenruo new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev); 65979787eaaSJeff Mahoney if (IS_ERR(new_root)) { 6602dfb1e43SQu Wenruo free_anon_bdev(anon_dev); 66179787eaaSJeff Mahoney ret = PTR_ERR(new_root); 66266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 66379787eaaSJeff Mahoney goto fail; 66479787eaaSJeff Mahoney } 6652dfb1e43SQu Wenruo /* Freeing will be done in btrfs_put_root() of new_root */ 6662dfb1e43SQu Wenruo anon_dev = 0; 66776dda93cSYan, Zheng 668221581e4SJosef Bacik ret = btrfs_record_root_in_trans(trans, new_root); 669221581e4SJosef Bacik if (ret) { 670221581e4SJosef Bacik btrfs_put_root(new_root); 671221581e4SJosef Bacik btrfs_abort_transaction(trans, ret); 672221581e4SJosef Bacik goto fail; 673221581e4SJosef Bacik } 67476dda93cSYan, Zheng 6754d4340c9SChristian Brauner ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns); 67600246528SJosef Bacik btrfs_put_root(new_root); 677ce598979SMark Fasheh if (ret) { 678ce598979SMark Fasheh /* We potentially lose an unused inode item here */ 67966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 680ce598979SMark Fasheh goto fail; 681ce598979SMark Fasheh } 682ce598979SMark Fasheh 683f46b5a66SChristoph Hellwig /* 684f46b5a66SChristoph Hellwig * insert the directory item 685f46b5a66SChristoph Hellwig */ 686877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(dir), &index); 68779787eaaSJeff Mahoney if (ret) { 68866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 68979787eaaSJeff Mahoney goto fail; 69079787eaaSJeff Mahoney } 6913de4586cSChris Mason 692684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key, 6933de4586cSChris Mason BTRFS_FT_DIR, index); 69479787eaaSJeff Mahoney if (ret) { 69566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 696f46b5a66SChristoph Hellwig goto fail; 69779787eaaSJeff Mahoney } 6980660b5afSChris Mason 6996ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2); 7009a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); 701c7e54b51SJosef Bacik if (ret) { 702c7e54b51SJosef Bacik btrfs_abort_transaction(trans, ret); 703c7e54b51SJosef Bacik goto fail; 704c7e54b51SJosef Bacik } 70552c26179SYan Zheng 7066025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid, 7074a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(dir)), index, name, namelen); 708c7e54b51SJosef Bacik if (ret) { 709c7e54b51SJosef Bacik btrfs_abort_transaction(trans, ret); 710c7e54b51SJosef Bacik goto fail; 711c7e54b51SJosef Bacik } 7120660b5afSChris Mason 713cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, root_item->uuid, 7146bccf3abSJeff Mahoney BTRFS_UUID_KEY_SUBVOL, objectid); 715dd5f9615SStefan Behrens if (ret) 71666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 717dd5f9615SStefan Behrens 718f46b5a66SChristoph Hellwig fail: 71949a3c4d9SDavid Sterba kfree(root_item); 720d5c12070SMiao Xie trans->block_rsv = NULL; 721d5c12070SMiao Xie trans->bytes_reserved = 0; 722e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 723de6e8200SLiu Bo 7241b58ae0eSFilipe Manana if (ret) 7251b58ae0eSFilipe Manana btrfs_end_transaction(trans); 7261b58ae0eSFilipe Manana else 7271b58ae0eSFilipe Manana ret = btrfs_commit_transaction(trans); 7281a65e24bSChris Mason 7295662344bSTsutomu Itoh if (!ret) { 7305662344bSTsutomu Itoh inode = btrfs_lookup_dentry(dir, dentry); 731de6e8200SLiu Bo if (IS_ERR(inode)) 732de6e8200SLiu Bo return PTR_ERR(inode); 7335662344bSTsutomu Itoh d_instantiate(dentry, inode); 7345662344bSTsutomu Itoh } 735f46b5a66SChristoph Hellwig return ret; 73649a3c4d9SDavid Sterba 73749a3c4d9SDavid Sterba fail_free: 7382dfb1e43SQu Wenruo if (anon_dev) 7392dfb1e43SQu Wenruo free_anon_bdev(anon_dev); 74049a3c4d9SDavid Sterba kfree(root_item); 74149a3c4d9SDavid Sterba return ret; 742f46b5a66SChristoph Hellwig } 743f46b5a66SChristoph Hellwig 744e9662f70SMiao Xie static int create_snapshot(struct btrfs_root *root, struct inode *dir, 7459babda9fSNikolay Borisov struct dentry *dentry, bool readonly, 746e9662f70SMiao Xie struct btrfs_qgroup_inherit *inherit) 747f46b5a66SChristoph Hellwig { 7480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 7492e4bfab9SYan, Zheng struct inode *inode; 750f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 751f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 7522e4bfab9SYan, Zheng int ret; 753f46b5a66SChristoph Hellwig 75492a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 755f46b5a66SChristoph Hellwig return -EINVAL; 756f46b5a66SChristoph Hellwig 757eede2bf3SOmar Sandoval if (atomic_read(&root->nr_swapfiles)) { 758eede2bf3SOmar Sandoval btrfs_warn(fs_info, 759eede2bf3SOmar Sandoval "cannot snapshot subvolume with active swapfile"); 760eede2bf3SOmar Sandoval return -ETXTBSY; 761eede2bf3SOmar Sandoval } 762eede2bf3SOmar Sandoval 76323269bf5SDavid Sterba pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL); 764a1ee7362SDavid Sterba if (!pending_snapshot) 765a1ee7362SDavid Sterba return -ENOMEM; 766a1ee7362SDavid Sterba 7672dfb1e43SQu Wenruo ret = get_anon_bdev(&pending_snapshot->anon_dev); 7682dfb1e43SQu Wenruo if (ret < 0) 7692dfb1e43SQu Wenruo goto free_pending; 770b0c0ea63SDavid Sterba pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item), 77123269bf5SDavid Sterba GFP_KERNEL); 7728546b570SDavid Sterba pending_snapshot->path = btrfs_alloc_path(); 7738546b570SDavid Sterba if (!pending_snapshot->root_item || !pending_snapshot->path) { 774b0c0ea63SDavid Sterba ret = -ENOMEM; 775b0c0ea63SDavid Sterba goto free_pending; 776b0c0ea63SDavid Sterba } 777b0c0ea63SDavid Sterba 77866d8f3ddSMiao Xie btrfs_init_block_rsv(&pending_snapshot->block_rsv, 77966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 780d5c12070SMiao Xie /* 781d5c12070SMiao Xie * 1 - parent dir inode 782d5c12070SMiao Xie * 2 - dir entries 783d5c12070SMiao Xie * 1 - root item 784d5c12070SMiao Xie * 2 - root ref/backref 785d5c12070SMiao Xie * 1 - root of snapshot 786dd5f9615SStefan Behrens * 1 - UUID item 787d5c12070SMiao Xie */ 788d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 789dd5f9615SStefan Behrens &pending_snapshot->block_rsv, 8, 790ee3441b4SJeff Mahoney false); 791d5c12070SMiao Xie if (ret) 792c11fbb6eSRobbie Ko goto free_pending; 793d5c12070SMiao Xie 794a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 795a22285a6SYan, Zheng pending_snapshot->root = root; 796b83cc969SLi Zefan pending_snapshot->readonly = readonly; 797e9662f70SMiao Xie pending_snapshot->dir = dir; 7988696c533SMiao Xie pending_snapshot->inherit = inherit; 799a22285a6SYan, Zheng 800d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 801a22285a6SYan, Zheng if (IS_ERR(trans)) { 802a22285a6SYan, Zheng ret = PTR_ERR(trans); 803a22285a6SYan, Zheng goto fail; 804a22285a6SYan, Zheng } 805a22285a6SYan, Zheng 80628b21c55SFilipe Manana trans->pending_snapshot = pending_snapshot; 8079babda9fSNikolay Borisov 8083a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 809aec8030aSMiao Xie if (ret) 810c37b2b62SJosef Bacik goto fail; 811a22285a6SYan, Zheng 812a22285a6SYan, Zheng ret = pending_snapshot->error; 813f46b5a66SChristoph Hellwig if (ret) 8142e4bfab9SYan, Zheng goto fail; 815f46b5a66SChristoph Hellwig 816d3797308SChris Mason ret = btrfs_orphan_cleanup(pending_snapshot->snap); 817d3797308SChris Mason if (ret) 818d3797308SChris Mason goto fail; 819d3797308SChris Mason 8202b0143b5SDavid Howells inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry); 8212e4bfab9SYan, Zheng if (IS_ERR(inode)) { 8222e4bfab9SYan, Zheng ret = PTR_ERR(inode); 8232e4bfab9SYan, Zheng goto fail; 8242e4bfab9SYan, Zheng } 8255662344bSTsutomu Itoh 8262e4bfab9SYan, Zheng d_instantiate(dentry, inode); 8272e4bfab9SYan, Zheng ret = 0; 8282dfb1e43SQu Wenruo pending_snapshot->anon_dev = 0; 8292e4bfab9SYan, Zheng fail: 8302dfb1e43SQu Wenruo /* Prevent double freeing of anon_dev */ 8312dfb1e43SQu Wenruo if (ret && pending_snapshot->snap) 8322dfb1e43SQu Wenruo pending_snapshot->snap->anon_dev = 0; 83300246528SJosef Bacik btrfs_put_root(pending_snapshot->snap); 834e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv); 835b0c0ea63SDavid Sterba free_pending: 8362dfb1e43SQu Wenruo if (pending_snapshot->anon_dev) 8372dfb1e43SQu Wenruo free_anon_bdev(pending_snapshot->anon_dev); 838b0c0ea63SDavid Sterba kfree(pending_snapshot->root_item); 8398546b570SDavid Sterba btrfs_free_path(pending_snapshot->path); 840a1ee7362SDavid Sterba kfree(pending_snapshot); 841a1ee7362SDavid Sterba 842f46b5a66SChristoph Hellwig return ret; 843f46b5a66SChristoph Hellwig } 844f46b5a66SChristoph Hellwig 8454260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 8464260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 8474260f7c7SSage Weil * whether the type of victim is right. 8484260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 8494260f7c7SSage Weil * 2. We should have write and exec permissions on dir 8504260f7c7SSage Weil * 3. We can't remove anything from append-only dir 8514260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 8524260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 8534260f7c7SSage Weil * a. be owner of dir, or 8544260f7c7SSage Weil * b. be owner of victim, or 8554260f7c7SSage Weil * c. have CAP_FOWNER capability 85601327610SNicholas D Steeves * 6. If the victim is append-only or immutable we can't do anything with 8574260f7c7SSage Weil * links pointing to it. 8584260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 8594260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 8604260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 8614260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 8624260f7c7SSage Weil * nfs_async_unlink(). 8634260f7c7SSage Weil */ 8644260f7c7SSage Weil 865c4ed533bSChristian Brauner static int btrfs_may_delete(struct user_namespace *mnt_userns, 866c4ed533bSChristian Brauner struct inode *dir, struct dentry *victim, int isdir) 8674260f7c7SSage Weil { 8684260f7c7SSage Weil int error; 8694260f7c7SSage Weil 8702b0143b5SDavid Howells if (d_really_is_negative(victim)) 8714260f7c7SSage Weil return -ENOENT; 8724260f7c7SSage Weil 8732b0143b5SDavid Howells BUG_ON(d_inode(victim->d_parent) != dir); 8744fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 8754260f7c7SSage Weil 876c4ed533bSChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 8774260f7c7SSage Weil if (error) 8784260f7c7SSage Weil return error; 8794260f7c7SSage Weil if (IS_APPEND(dir)) 8804260f7c7SSage Weil return -EPERM; 881c4ed533bSChristian Brauner if (check_sticky(mnt_userns, dir, d_inode(victim)) || 882ba73d987SChristian Brauner IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) || 883ba73d987SChristian Brauner IS_SWAPFILE(d_inode(victim))) 8844260f7c7SSage Weil return -EPERM; 8854260f7c7SSage Weil if (isdir) { 886e36cb0b8SDavid Howells if (!d_is_dir(victim)) 8874260f7c7SSage Weil return -ENOTDIR; 8884260f7c7SSage Weil if (IS_ROOT(victim)) 8894260f7c7SSage Weil return -EBUSY; 890e36cb0b8SDavid Howells } else if (d_is_dir(victim)) 8914260f7c7SSage Weil return -EISDIR; 8924260f7c7SSage Weil if (IS_DEADDIR(dir)) 8934260f7c7SSage Weil return -ENOENT; 8944260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 8954260f7c7SSage Weil return -EBUSY; 8964260f7c7SSage Weil return 0; 8974260f7c7SSage Weil } 8984260f7c7SSage Weil 899cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 9004d4340c9SChristian Brauner static inline int btrfs_may_create(struct user_namespace *mnt_userns, 9014d4340c9SChristian Brauner struct inode *dir, struct dentry *child) 902cb8e7090SChristoph Hellwig { 9032b0143b5SDavid Howells if (d_really_is_positive(child)) 904cb8e7090SChristoph Hellwig return -EEXIST; 905cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 906cb8e7090SChristoph Hellwig return -ENOENT; 9074d4340c9SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 9085474bf40SChristian Brauner return -EOVERFLOW; 9094d4340c9SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 910cb8e7090SChristoph Hellwig } 911cb8e7090SChristoph Hellwig 912cb8e7090SChristoph Hellwig /* 913cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 914cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 915cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 916cb8e7090SChristoph Hellwig */ 91792872094SAl Viro static noinline int btrfs_mksubvol(const struct path *parent, 9184d4340c9SChristian Brauner struct user_namespace *mnt_userns, 91952f75f4fSDavid Sterba const char *name, int namelen, 92072fd032eSSage Weil struct btrfs_root *snap_src, 9219babda9fSNikolay Borisov bool readonly, 9228696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 923cb8e7090SChristoph Hellwig { 9242b0143b5SDavid Howells struct inode *dir = d_inode(parent->dentry); 9250b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 926cb8e7090SChristoph Hellwig struct dentry *dentry; 927cb8e7090SChristoph Hellwig int error; 928cb8e7090SChristoph Hellwig 92900235411SAl Viro error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 93000235411SAl Viro if (error == -EINTR) 93100235411SAl Viro return error; 932cb8e7090SChristoph Hellwig 9334d4340c9SChristian Brauner dentry = lookup_one(mnt_userns, name, parent->dentry, namelen); 934cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 935cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 936cb8e7090SChristoph Hellwig goto out_unlock; 937cb8e7090SChristoph Hellwig 9384d4340c9SChristian Brauner error = btrfs_may_create(mnt_userns, dir, dentry); 939cb8e7090SChristoph Hellwig if (error) 940a874a63eSLiu Bo goto out_dput; 941cb8e7090SChristoph Hellwig 9429c52057cSChris Mason /* 9439c52057cSChris Mason * even if this name doesn't exist, we may get hash collisions. 9449c52057cSChris Mason * check for them now when we can safely fail 9459c52057cSChris Mason */ 9469c52057cSChris Mason error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 9479c52057cSChris Mason dir->i_ino, name, 9489c52057cSChris Mason namelen); 9499c52057cSChris Mason if (error) 9509c52057cSChris Mason goto out_dput; 9519c52057cSChris Mason 9520b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 95376dda93cSYan, Zheng 95476dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 95576dda93cSYan, Zheng goto out_up_read; 95676dda93cSYan, Zheng 9579babda9fSNikolay Borisov if (snap_src) 9589babda9fSNikolay Borisov error = create_snapshot(snap_src, dir, dentry, readonly, inherit); 9599babda9fSNikolay Borisov else 9604d4340c9SChristian Brauner error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit); 9619babda9fSNikolay Borisov 96276dda93cSYan, Zheng if (!error) 96376dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 96476dda93cSYan, Zheng out_up_read: 9650b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 966cb8e7090SChristoph Hellwig out_dput: 967cb8e7090SChristoph Hellwig dput(dentry); 968cb8e7090SChristoph Hellwig out_unlock: 96964708539SJosef Bacik btrfs_inode_unlock(dir, 0); 970cb8e7090SChristoph Hellwig return error; 971cb8e7090SChristoph Hellwig } 972cb8e7090SChristoph Hellwig 973c11fbb6eSRobbie Ko static noinline int btrfs_mksnapshot(const struct path *parent, 9744d4340c9SChristian Brauner struct user_namespace *mnt_userns, 975c11fbb6eSRobbie Ko const char *name, int namelen, 976c11fbb6eSRobbie Ko struct btrfs_root *root, 977c11fbb6eSRobbie Ko bool readonly, 978c11fbb6eSRobbie Ko struct btrfs_qgroup_inherit *inherit) 979c11fbb6eSRobbie Ko { 980c11fbb6eSRobbie Ko int ret; 981c11fbb6eSRobbie Ko bool snapshot_force_cow = false; 982c11fbb6eSRobbie Ko 983c11fbb6eSRobbie Ko /* 984c11fbb6eSRobbie Ko * Force new buffered writes to reserve space even when NOCOW is 985c11fbb6eSRobbie Ko * possible. This is to avoid later writeback (running dealloc) to 986c11fbb6eSRobbie Ko * fallback to COW mode and unexpectedly fail with ENOSPC. 987c11fbb6eSRobbie Ko */ 988c11fbb6eSRobbie Ko btrfs_drew_read_lock(&root->snapshot_lock); 989c11fbb6eSRobbie Ko 990f9baa501SFilipe Manana ret = btrfs_start_delalloc_snapshot(root, false); 991c11fbb6eSRobbie Ko if (ret) 992c11fbb6eSRobbie Ko goto out; 993c11fbb6eSRobbie Ko 994c11fbb6eSRobbie Ko /* 995c11fbb6eSRobbie Ko * All previous writes have started writeback in NOCOW mode, so now 996c11fbb6eSRobbie Ko * we force future writes to fallback to COW mode during snapshot 997c11fbb6eSRobbie Ko * creation. 998c11fbb6eSRobbie Ko */ 999c11fbb6eSRobbie Ko atomic_inc(&root->snapshot_force_cow); 1000c11fbb6eSRobbie Ko snapshot_force_cow = true; 1001c11fbb6eSRobbie Ko 1002c11fbb6eSRobbie Ko btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1); 1003c11fbb6eSRobbie Ko 10044d4340c9SChristian Brauner ret = btrfs_mksubvol(parent, mnt_userns, name, namelen, 1005c11fbb6eSRobbie Ko root, readonly, inherit); 1006c11fbb6eSRobbie Ko out: 1007c11fbb6eSRobbie Ko if (snapshot_force_cow) 1008c11fbb6eSRobbie Ko atomic_dec(&root->snapshot_force_cow); 1009c11fbb6eSRobbie Ko btrfs_drew_read_unlock(&root->snapshot_lock); 1010c11fbb6eSRobbie Ko return ret; 1011c11fbb6eSRobbie Ko } 1012c11fbb6eSRobbie Ko 1013d5633b0dSQu Wenruo /* 1014d5633b0dSQu Wenruo * Defrag specific helper to get an extent map. 1015d5633b0dSQu Wenruo * 1016d5633b0dSQu Wenruo * Differences between this and btrfs_get_extent() are: 1017d5633b0dSQu Wenruo * 1018d5633b0dSQu Wenruo * - No extent_map will be added to inode->extent_tree 1019d5633b0dSQu Wenruo * To reduce memory usage in the long run. 1020d5633b0dSQu Wenruo * 1021d5633b0dSQu Wenruo * - Extra optimization to skip file extents older than @newer_than 1022d5633b0dSQu Wenruo * By using btrfs_search_forward() we can skip entire file ranges that 1023d5633b0dSQu Wenruo * have extents created in past transactions, because btrfs_search_forward() 1024d5633b0dSQu Wenruo * will not visit leaves and nodes with a generation smaller than given 1025d5633b0dSQu Wenruo * minimal generation threshold (@newer_than). 1026d5633b0dSQu Wenruo * 1027d5633b0dSQu Wenruo * Return valid em if we find a file extent matching the requirement. 1028d5633b0dSQu Wenruo * Return NULL if we can not find a file extent matching the requirement. 1029d5633b0dSQu Wenruo * 1030d5633b0dSQu Wenruo * Return ERR_PTR() for error. 1031d5633b0dSQu Wenruo */ 1032d5633b0dSQu Wenruo static struct extent_map *defrag_get_extent(struct btrfs_inode *inode, 1033d5633b0dSQu Wenruo u64 start, u64 newer_than) 1034d5633b0dSQu Wenruo { 1035d5633b0dSQu Wenruo struct btrfs_root *root = inode->root; 1036d5633b0dSQu Wenruo struct btrfs_file_extent_item *fi; 1037d5633b0dSQu Wenruo struct btrfs_path path = { 0 }; 1038d5633b0dSQu Wenruo struct extent_map *em; 1039d5633b0dSQu Wenruo struct btrfs_key key; 1040d5633b0dSQu Wenruo u64 ino = btrfs_ino(inode); 1041d5633b0dSQu Wenruo int ret; 1042d5633b0dSQu Wenruo 1043d5633b0dSQu Wenruo em = alloc_extent_map(); 1044d5633b0dSQu Wenruo if (!em) { 1045d5633b0dSQu Wenruo ret = -ENOMEM; 1046d5633b0dSQu Wenruo goto err; 1047d5633b0dSQu Wenruo } 1048d5633b0dSQu Wenruo 1049d5633b0dSQu Wenruo key.objectid = ino; 1050d5633b0dSQu Wenruo key.type = BTRFS_EXTENT_DATA_KEY; 1051d5633b0dSQu Wenruo key.offset = start; 1052d5633b0dSQu Wenruo 1053d5633b0dSQu Wenruo if (newer_than) { 1054d5633b0dSQu Wenruo ret = btrfs_search_forward(root, &key, &path, newer_than); 1055d5633b0dSQu Wenruo if (ret < 0) 1056d5633b0dSQu Wenruo goto err; 1057d5633b0dSQu Wenruo /* Can't find anything newer */ 1058d5633b0dSQu Wenruo if (ret > 0) 1059d5633b0dSQu Wenruo goto not_found; 1060d5633b0dSQu Wenruo } else { 1061d5633b0dSQu Wenruo ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0); 1062d5633b0dSQu Wenruo if (ret < 0) 1063d5633b0dSQu Wenruo goto err; 1064d5633b0dSQu Wenruo } 1065d5633b0dSQu Wenruo if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) { 1066d5633b0dSQu Wenruo /* 1067d5633b0dSQu Wenruo * If btrfs_search_slot() makes path to point beyond nritems, 1068d5633b0dSQu Wenruo * we should not have an empty leaf, as this inode must at 1069d5633b0dSQu Wenruo * least have its INODE_ITEM. 1070d5633b0dSQu Wenruo */ 1071d5633b0dSQu Wenruo ASSERT(btrfs_header_nritems(path.nodes[0])); 1072d5633b0dSQu Wenruo path.slots[0] = btrfs_header_nritems(path.nodes[0]) - 1; 1073d5633b0dSQu Wenruo } 1074d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1075d5633b0dSQu Wenruo /* Perfect match, no need to go one slot back */ 1076d5633b0dSQu Wenruo if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY && 1077d5633b0dSQu Wenruo key.offset == start) 1078d5633b0dSQu Wenruo goto iterate; 1079d5633b0dSQu Wenruo 1080d5633b0dSQu Wenruo /* We didn't find a perfect match, needs to go one slot back */ 1081d5633b0dSQu Wenruo if (path.slots[0] > 0) { 1082d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1083d5633b0dSQu Wenruo if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) 1084d5633b0dSQu Wenruo path.slots[0]--; 1085d5633b0dSQu Wenruo } 1086d5633b0dSQu Wenruo 1087d5633b0dSQu Wenruo iterate: 1088d5633b0dSQu Wenruo /* Iterate through the path to find a file extent covering @start */ 1089d5633b0dSQu Wenruo while (true) { 1090d5633b0dSQu Wenruo u64 extent_end; 1091d5633b0dSQu Wenruo 1092d5633b0dSQu Wenruo if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) 1093d5633b0dSQu Wenruo goto next; 1094d5633b0dSQu Wenruo 1095d5633b0dSQu Wenruo btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]); 1096d5633b0dSQu Wenruo 1097d5633b0dSQu Wenruo /* 1098d5633b0dSQu Wenruo * We may go one slot back to INODE_REF/XATTR item, then 1099d5633b0dSQu Wenruo * need to go forward until we reach an EXTENT_DATA. 1100d5633b0dSQu Wenruo * But we should still has the correct ino as key.objectid. 1101d5633b0dSQu Wenruo */ 1102d5633b0dSQu Wenruo if (WARN_ON(key.objectid < ino) || key.type < BTRFS_EXTENT_DATA_KEY) 1103d5633b0dSQu Wenruo goto next; 1104d5633b0dSQu Wenruo 1105d5633b0dSQu Wenruo /* It's beyond our target range, definitely not extent found */ 1106d5633b0dSQu Wenruo if (key.objectid > ino || key.type > BTRFS_EXTENT_DATA_KEY) 1107d5633b0dSQu Wenruo goto not_found; 1108d5633b0dSQu Wenruo 1109d5633b0dSQu Wenruo /* 1110d5633b0dSQu Wenruo * | |<- File extent ->| 1111d5633b0dSQu Wenruo * \- start 1112d5633b0dSQu Wenruo * 1113d5633b0dSQu Wenruo * This means there is a hole between start and key.offset. 1114d5633b0dSQu Wenruo */ 1115d5633b0dSQu Wenruo if (key.offset > start) { 1116d5633b0dSQu Wenruo em->start = start; 1117d5633b0dSQu Wenruo em->orig_start = start; 1118d5633b0dSQu Wenruo em->block_start = EXTENT_MAP_HOLE; 1119d5633b0dSQu Wenruo em->len = key.offset - start; 1120d5633b0dSQu Wenruo break; 1121d5633b0dSQu Wenruo } 1122d5633b0dSQu Wenruo 1123d5633b0dSQu Wenruo fi = btrfs_item_ptr(path.nodes[0], path.slots[0], 1124d5633b0dSQu Wenruo struct btrfs_file_extent_item); 1125d5633b0dSQu Wenruo extent_end = btrfs_file_extent_end(&path); 1126d5633b0dSQu Wenruo 1127d5633b0dSQu Wenruo /* 1128d5633b0dSQu Wenruo * |<- file extent ->| | 1129d5633b0dSQu Wenruo * \- start 1130d5633b0dSQu Wenruo * 1131d5633b0dSQu Wenruo * We haven't reached start, search next slot. 1132d5633b0dSQu Wenruo */ 1133d5633b0dSQu Wenruo if (extent_end <= start) 1134d5633b0dSQu Wenruo goto next; 1135d5633b0dSQu Wenruo 1136d5633b0dSQu Wenruo /* Now this extent covers @start, convert it to em */ 1137d5633b0dSQu Wenruo btrfs_extent_item_to_extent_map(inode, &path, fi, false, em); 1138d5633b0dSQu Wenruo break; 1139d5633b0dSQu Wenruo next: 1140d5633b0dSQu Wenruo ret = btrfs_next_item(root, &path); 1141d5633b0dSQu Wenruo if (ret < 0) 1142d5633b0dSQu Wenruo goto err; 1143d5633b0dSQu Wenruo if (ret > 0) 1144d5633b0dSQu Wenruo goto not_found; 1145d5633b0dSQu Wenruo } 1146d5633b0dSQu Wenruo btrfs_release_path(&path); 1147d5633b0dSQu Wenruo return em; 1148d5633b0dSQu Wenruo 1149d5633b0dSQu Wenruo not_found: 1150d5633b0dSQu Wenruo btrfs_release_path(&path); 1151d5633b0dSQu Wenruo free_extent_map(em); 1152d5633b0dSQu Wenruo return NULL; 1153d5633b0dSQu Wenruo 1154d5633b0dSQu Wenruo err: 1155d5633b0dSQu Wenruo btrfs_release_path(&path); 1156d5633b0dSQu Wenruo free_extent_map(em); 1157d5633b0dSQu Wenruo return ERR_PTR(ret); 1158d5633b0dSQu Wenruo } 1159d5633b0dSQu Wenruo 1160e9eec721SQu Wenruo static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start, 1161d5633b0dSQu Wenruo u64 newer_than, bool locked) 116217ce6ef8SLiu Bo { 116317ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 1164940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 11656c282eb4SLi Zefan struct extent_map *em; 116676068caeSQu Wenruo const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize; 1167940100a4SChris Mason 1168940100a4SChris Mason /* 1169940100a4SChris Mason * hopefully we have this extent in the tree already, try without 1170940100a4SChris Mason * the full extent lock 1171940100a4SChris Mason */ 1172940100a4SChris Mason read_lock(&em_tree->lock); 117376068caeSQu Wenruo em = lookup_extent_mapping(em_tree, start, sectorsize); 1174940100a4SChris Mason read_unlock(&em_tree->lock); 1175940100a4SChris Mason 1176199257a7SQu Wenruo /* 1177199257a7SQu Wenruo * We can get a merged extent, in that case, we need to re-search 1178199257a7SQu Wenruo * tree to get the original em for defrag. 1179199257a7SQu Wenruo * 1180199257a7SQu Wenruo * If @newer_than is 0 or em::generation < newer_than, we can trust 1181199257a7SQu Wenruo * this em, as either we don't care about the generation, or the 1182199257a7SQu Wenruo * merged extent map will be rejected anyway. 1183199257a7SQu Wenruo */ 1184199257a7SQu Wenruo if (em && test_bit(EXTENT_FLAG_MERGED, &em->flags) && 1185199257a7SQu Wenruo newer_than && em->generation >= newer_than) { 1186199257a7SQu Wenruo free_extent_map(em); 1187199257a7SQu Wenruo em = NULL; 1188199257a7SQu Wenruo } 1189199257a7SQu Wenruo 1190940100a4SChris Mason if (!em) { 1191308d9800SFilipe Manana struct extent_state *cached = NULL; 119276068caeSQu Wenruo u64 end = start + sectorsize - 1; 1193308d9800SFilipe Manana 1194940100a4SChris Mason /* get the big lock and read metadata off disk */ 1195e9eec721SQu Wenruo if (!locked) 1196ff13db41SDavid Sterba lock_extent_bits(io_tree, start, end, &cached); 1197d5633b0dSQu Wenruo em = defrag_get_extent(BTRFS_I(inode), start, newer_than); 1198e9eec721SQu Wenruo if (!locked) 1199e43bbe5eSDavid Sterba unlock_extent_cached(io_tree, start, end, &cached); 1200940100a4SChris Mason 12016cf8bfbfSDan Carpenter if (IS_ERR(em)) 12026c282eb4SLi Zefan return NULL; 1203940100a4SChris Mason } 1204940100a4SChris Mason 12056c282eb4SLi Zefan return em; 12066c282eb4SLi Zefan } 12076c282eb4SLi Zefan 1208979b25c3SQu Wenruo static u32 get_extent_max_capacity(const struct extent_map *em) 1209979b25c3SQu Wenruo { 1210979b25c3SQu Wenruo if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) 1211979b25c3SQu Wenruo return BTRFS_MAX_COMPRESSED; 1212979b25c3SQu Wenruo return BTRFS_MAX_EXTENT_SIZE; 1213979b25c3SQu Wenruo } 1214979b25c3SQu Wenruo 1215e9eec721SQu Wenruo static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em, 1216e9eec721SQu Wenruo bool locked) 12176c282eb4SLi Zefan { 12186c282eb4SLi Zefan struct extent_map *next; 12197093f152SQu Wenruo bool ret = false; 12206c282eb4SLi Zefan 12216c282eb4SLi Zefan /* this is the last extent */ 12226c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 12236c282eb4SLi Zefan return false; 12246c282eb4SLi Zefan 1225d5633b0dSQu Wenruo /* 1226d5633b0dSQu Wenruo * We want to check if the next extent can be merged with the current 1227d5633b0dSQu Wenruo * one, which can be an extent created in a past generation, so we pass 1228d5633b0dSQu Wenruo * a minimum generation of 0 to defrag_lookup_extent(). 1229d5633b0dSQu Wenruo */ 1230d5633b0dSQu Wenruo next = defrag_lookup_extent(inode, em->start + em->len, 0, locked); 12317093f152SQu Wenruo /* No more em or hole */ 1232e9512d72SChris Mason if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 12337093f152SQu Wenruo goto out; 12347093f152SQu Wenruo if (test_bit(EXTENT_FLAG_PREALLOC, &next->flags)) 12357093f152SQu Wenruo goto out; 1236979b25c3SQu Wenruo /* 1237979b25c3SQu Wenruo * If the next extent is at its max capacity, defragging current extent 1238979b25c3SQu Wenruo * makes no sense, as the total number of extents won't change. 1239979b25c3SQu Wenruo */ 1240979b25c3SQu Wenruo if (next->len >= get_extent_max_capacity(em)) 1241979b25c3SQu Wenruo goto out; 12427093f152SQu Wenruo ret = true; 12437093f152SQu Wenruo out: 12446c282eb4SLi Zefan free_extent_map(next); 12456c282eb4SLi Zefan return ret; 12466c282eb4SLi Zefan } 12476c282eb4SLi Zefan 12484cb5300bSChris Mason /* 12495767b50cSQu Wenruo * Prepare one page to be defragged. 12505767b50cSQu Wenruo * 12515767b50cSQu Wenruo * This will ensure: 12525767b50cSQu Wenruo * 12535767b50cSQu Wenruo * - Returned page is locked and has been set up properly. 12545767b50cSQu Wenruo * - No ordered extent exists in the page. 12555767b50cSQu Wenruo * - The page is uptodate. 12565767b50cSQu Wenruo * 12575767b50cSQu Wenruo * NOTE: Caller should also wait for page writeback after the cluster is 12585767b50cSQu Wenruo * prepared, here we don't do writeback wait for each page. 12595767b50cSQu Wenruo */ 12605767b50cSQu Wenruo static struct page *defrag_prepare_one_page(struct btrfs_inode *inode, 12615767b50cSQu Wenruo pgoff_t index) 1262940100a4SChris Mason { 12635767b50cSQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 12645767b50cSQu Wenruo gfp_t mask = btrfs_alloc_write_mask(mapping); 12655767b50cSQu Wenruo u64 page_start = (u64)index << PAGE_SHIFT; 12665767b50cSQu Wenruo u64 page_end = page_start + PAGE_SIZE - 1; 12674cb5300bSChris Mason struct extent_state *cached_state = NULL; 12684cb5300bSChris Mason struct page *page; 12695767b50cSQu Wenruo int ret; 12705767b50cSQu Wenruo 1271600a45e1SMiao Xie again: 12725767b50cSQu Wenruo page = find_or_create_page(mapping, index, mask); 12734cb5300bSChris Mason if (!page) 12745767b50cSQu Wenruo return ERR_PTR(-ENOMEM); 12755767b50cSQu Wenruo 127624bcb454SOmar Sandoval /* 127724bcb454SOmar Sandoval * Since we can defragment files opened read-only, we can encounter 127824bcb454SOmar Sandoval * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We 127924bcb454SOmar Sandoval * can't do I/O using huge pages yet, so return an error for now. 128024bcb454SOmar Sandoval * Filesystem transparent huge pages are typically only used for 128124bcb454SOmar Sandoval * executables that explicitly enable them, so this isn't very 128224bcb454SOmar Sandoval * restrictive. 128324bcb454SOmar Sandoval */ 128424bcb454SOmar Sandoval if (PageCompound(page)) { 128524bcb454SOmar Sandoval unlock_page(page); 128624bcb454SOmar Sandoval put_page(page); 128724bcb454SOmar Sandoval return ERR_PTR(-ETXTBSY); 128824bcb454SOmar Sandoval } 12894cb5300bSChris Mason 129032443de3SQu Wenruo ret = set_page_extent_mapped(page); 129132443de3SQu Wenruo if (ret < 0) { 129232443de3SQu Wenruo unlock_page(page); 129332443de3SQu Wenruo put_page(page); 12945767b50cSQu Wenruo return ERR_PTR(ret); 129532443de3SQu Wenruo } 129632443de3SQu Wenruo 12975767b50cSQu Wenruo /* Wait for any existing ordered extent in the range */ 1298600a45e1SMiao Xie while (1) { 12995767b50cSQu Wenruo struct btrfs_ordered_extent *ordered; 13005767b50cSQu Wenruo 13015767b50cSQu Wenruo lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); 13025767b50cSQu Wenruo ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); 13035767b50cSQu Wenruo unlock_extent_cached(&inode->io_tree, page_start, page_end, 1304e43bbe5eSDavid Sterba &cached_state); 1305600a45e1SMiao Xie if (!ordered) 1306600a45e1SMiao Xie break; 1307600a45e1SMiao Xie 1308600a45e1SMiao Xie unlock_page(page); 1309c0a43603SNikolay Borisov btrfs_start_ordered_extent(ordered, 1); 1310600a45e1SMiao Xie btrfs_put_ordered_extent(ordered); 1311600a45e1SMiao Xie lock_page(page); 13121f12bd06SLiu Bo /* 13135767b50cSQu Wenruo * We unlocked the page above, so we need check if it was 13145767b50cSQu Wenruo * released or not. 13151f12bd06SLiu Bo */ 13165767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 13171f12bd06SLiu Bo unlock_page(page); 131809cbfeafSKirill A. Shutemov put_page(page); 13191f12bd06SLiu Bo goto again; 13201f12bd06SLiu Bo } 1321600a45e1SMiao Xie } 1322600a45e1SMiao Xie 13235767b50cSQu Wenruo /* 13245767b50cSQu Wenruo * Now the page range has no ordered extent any more. Read the page to 13255767b50cSQu Wenruo * make it uptodate. 13265767b50cSQu Wenruo */ 13274cb5300bSChris Mason if (!PageUptodate(page)) { 13284cb5300bSChris Mason btrfs_readpage(NULL, page); 13294cb5300bSChris Mason lock_page(page); 13305767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 1331600a45e1SMiao Xie unlock_page(page); 133209cbfeafSKirill A. Shutemov put_page(page); 1333600a45e1SMiao Xie goto again; 1334600a45e1SMiao Xie } 13355767b50cSQu Wenruo if (!PageUptodate(page)) { 13365767b50cSQu Wenruo unlock_page(page); 13375767b50cSQu Wenruo put_page(page); 13385767b50cSQu Wenruo return ERR_PTR(-EIO); 13394cb5300bSChris Mason } 13405767b50cSQu Wenruo } 13415767b50cSQu Wenruo return page; 13425767b50cSQu Wenruo } 13434cb5300bSChris Mason 1344eb793cf8SQu Wenruo struct defrag_target_range { 1345eb793cf8SQu Wenruo struct list_head list; 1346eb793cf8SQu Wenruo u64 start; 1347eb793cf8SQu Wenruo u64 len; 1348eb793cf8SQu Wenruo }; 13494cb5300bSChris Mason 13504cb5300bSChris Mason /* 1351eb793cf8SQu Wenruo * Collect all valid target extents. 1352eb793cf8SQu Wenruo * 1353eb793cf8SQu Wenruo * @start: file offset to lookup 1354eb793cf8SQu Wenruo * @len: length to lookup 1355eb793cf8SQu Wenruo * @extent_thresh: file extent size threshold, any extent size >= this value 1356eb793cf8SQu Wenruo * will be ignored 1357eb793cf8SQu Wenruo * @newer_than: only defrag extents newer than this value 1358eb793cf8SQu Wenruo * @do_compress: whether the defrag is doing compression 1359eb793cf8SQu Wenruo * if true, @extent_thresh will be ignored and all regular 1360eb793cf8SQu Wenruo * file extents meeting @newer_than will be targets. 1361e9eec721SQu Wenruo * @locked: if the range has already held extent lock 1362eb793cf8SQu Wenruo * @target_list: list of targets file extents 13634cb5300bSChris Mason */ 1364eb793cf8SQu Wenruo static int defrag_collect_targets(struct btrfs_inode *inode, 1365eb793cf8SQu Wenruo u64 start, u64 len, u32 extent_thresh, 1366eb793cf8SQu Wenruo u64 newer_than, bool do_compress, 1367966d879bSQu Wenruo bool locked, struct list_head *target_list, 1368966d879bSQu Wenruo u64 *last_scanned_ret) 1369eb793cf8SQu Wenruo { 1370966d879bSQu Wenruo bool last_is_target = false; 1371eb793cf8SQu Wenruo u64 cur = start; 1372eb793cf8SQu Wenruo int ret = 0; 1373eb793cf8SQu Wenruo 1374eb793cf8SQu Wenruo while (cur < start + len) { 1375eb793cf8SQu Wenruo struct extent_map *em; 1376eb793cf8SQu Wenruo struct defrag_target_range *new; 1377eb793cf8SQu Wenruo bool next_mergeable = true; 1378eb793cf8SQu Wenruo u64 range_len; 1379eb793cf8SQu Wenruo 1380966d879bSQu Wenruo last_is_target = false; 1381d5633b0dSQu Wenruo em = defrag_lookup_extent(&inode->vfs_inode, cur, 1382d5633b0dSQu Wenruo newer_than, locked); 1383eb793cf8SQu Wenruo if (!em) 1384eb793cf8SQu Wenruo break; 1385eb793cf8SQu Wenruo 1386eb793cf8SQu Wenruo /* Skip hole/inline/preallocated extents */ 1387eb793cf8SQu Wenruo if (em->block_start >= EXTENT_MAP_LAST_BYTE || 1388eb793cf8SQu Wenruo test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) 1389eb793cf8SQu Wenruo goto next; 1390eb793cf8SQu Wenruo 1391eb793cf8SQu Wenruo /* Skip older extent */ 1392eb793cf8SQu Wenruo if (em->generation < newer_than) 1393eb793cf8SQu Wenruo goto next; 1394eb793cf8SQu Wenruo 13950d1ffa22SQu Wenruo /* This em is under writeback, no need to defrag */ 13960d1ffa22SQu Wenruo if (em->generation == (u64)-1) 13970d1ffa22SQu Wenruo goto next; 13980d1ffa22SQu Wenruo 1399eb793cf8SQu Wenruo /* 14000cb5950fSFilipe Manana * Our start offset might be in the middle of an existing extent 14010cb5950fSFilipe Manana * map, so take that into account. 14020cb5950fSFilipe Manana */ 14030cb5950fSFilipe Manana range_len = em->len - (cur - em->start); 14040cb5950fSFilipe Manana /* 14050cb5950fSFilipe Manana * If this range of the extent map is already flagged for delalloc, 14060cb5950fSFilipe Manana * skip it, because: 14070cb5950fSFilipe Manana * 14080cb5950fSFilipe Manana * 1) We could deadlock later, when trying to reserve space for 14090cb5950fSFilipe Manana * delalloc, because in case we can't immediately reserve space 14100cb5950fSFilipe Manana * the flusher can start delalloc and wait for the respective 14110cb5950fSFilipe Manana * ordered extents to complete. The deadlock would happen 14120cb5950fSFilipe Manana * because we do the space reservation while holding the range 14130cb5950fSFilipe Manana * locked, and starting writeback, or finishing an ordered 14140cb5950fSFilipe Manana * extent, requires locking the range; 14150cb5950fSFilipe Manana * 14160cb5950fSFilipe Manana * 2) If there's delalloc there, it means there's dirty pages for 14170cb5950fSFilipe Manana * which writeback has not started yet (we clean the delalloc 14180cb5950fSFilipe Manana * flag when starting writeback and after creating an ordered 14190cb5950fSFilipe Manana * extent). If we mark pages in an adjacent range for defrag, 14200cb5950fSFilipe Manana * then we will have a larger contiguous range for delalloc, 14210cb5950fSFilipe Manana * very likely resulting in a larger extent after writeback is 14220cb5950fSFilipe Manana * triggered (except in a case of free space fragmentation). 14230cb5950fSFilipe Manana */ 14240cb5950fSFilipe Manana if (test_range_bit(&inode->io_tree, cur, cur + range_len - 1, 14250cb5950fSFilipe Manana EXTENT_DELALLOC, 0, NULL)) 14260cb5950fSFilipe Manana goto next; 14270cb5950fSFilipe Manana 14280cb5950fSFilipe Manana /* 1429eb793cf8SQu Wenruo * For do_compress case, we want to compress all valid file 1430eb793cf8SQu Wenruo * extents, thus no @extent_thresh or mergeable check. 1431eb793cf8SQu Wenruo */ 1432eb793cf8SQu Wenruo if (do_compress) 1433eb793cf8SQu Wenruo goto add; 1434eb793cf8SQu Wenruo 1435eb793cf8SQu Wenruo /* Skip too large extent */ 14360cb5950fSFilipe Manana if (range_len >= extent_thresh) 1437eb793cf8SQu Wenruo goto next; 1438eb793cf8SQu Wenruo 1439979b25c3SQu Wenruo /* 1440979b25c3SQu Wenruo * Skip extents already at its max capacity, this is mostly for 1441979b25c3SQu Wenruo * compressed extents, which max cap is only 128K. 1442979b25c3SQu Wenruo */ 1443979b25c3SQu Wenruo if (em->len >= get_extent_max_capacity(em)) 1444979b25c3SQu Wenruo goto next; 1445979b25c3SQu Wenruo 1446e9eec721SQu Wenruo next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em, 1447e9eec721SQu Wenruo locked); 1448eb793cf8SQu Wenruo if (!next_mergeable) { 1449eb793cf8SQu Wenruo struct defrag_target_range *last; 1450eb793cf8SQu Wenruo 1451eb793cf8SQu Wenruo /* Empty target list, no way to merge with last entry */ 1452eb793cf8SQu Wenruo if (list_empty(target_list)) 1453eb793cf8SQu Wenruo goto next; 1454eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1455eb793cf8SQu Wenruo struct defrag_target_range, list); 1456eb793cf8SQu Wenruo /* Not mergeable with last entry */ 1457eb793cf8SQu Wenruo if (last->start + last->len != cur) 1458eb793cf8SQu Wenruo goto next; 1459eb793cf8SQu Wenruo 1460eb793cf8SQu Wenruo /* Mergeable, fall through to add it to @target_list. */ 1461eb793cf8SQu Wenruo } 1462eb793cf8SQu Wenruo 1463eb793cf8SQu Wenruo add: 1464966d879bSQu Wenruo last_is_target = true; 1465eb793cf8SQu Wenruo range_len = min(extent_map_end(em), start + len) - cur; 1466eb793cf8SQu Wenruo /* 1467eb793cf8SQu Wenruo * This one is a good target, check if it can be merged into 1468eb793cf8SQu Wenruo * last range of the target list. 1469eb793cf8SQu Wenruo */ 1470eb793cf8SQu Wenruo if (!list_empty(target_list)) { 1471eb793cf8SQu Wenruo struct defrag_target_range *last; 1472eb793cf8SQu Wenruo 1473eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1474eb793cf8SQu Wenruo struct defrag_target_range, list); 1475eb793cf8SQu Wenruo ASSERT(last->start + last->len <= cur); 1476eb793cf8SQu Wenruo if (last->start + last->len == cur) { 1477eb793cf8SQu Wenruo /* Mergeable, enlarge the last entry */ 1478eb793cf8SQu Wenruo last->len += range_len; 1479eb793cf8SQu Wenruo goto next; 1480eb793cf8SQu Wenruo } 1481eb793cf8SQu Wenruo /* Fall through to allocate a new entry */ 1482eb793cf8SQu Wenruo } 1483eb793cf8SQu Wenruo 1484eb793cf8SQu Wenruo /* Allocate new defrag_target_range */ 1485eb793cf8SQu Wenruo new = kmalloc(sizeof(*new), GFP_NOFS); 1486eb793cf8SQu Wenruo if (!new) { 1487eb793cf8SQu Wenruo free_extent_map(em); 1488eb793cf8SQu Wenruo ret = -ENOMEM; 1489eb793cf8SQu Wenruo break; 1490eb793cf8SQu Wenruo } 1491eb793cf8SQu Wenruo new->start = cur; 1492eb793cf8SQu Wenruo new->len = range_len; 1493eb793cf8SQu Wenruo list_add_tail(&new->list, target_list); 1494eb793cf8SQu Wenruo 1495eb793cf8SQu Wenruo next: 1496eb793cf8SQu Wenruo cur = extent_map_end(em); 1497eb793cf8SQu Wenruo free_extent_map(em); 1498eb793cf8SQu Wenruo } 1499eb793cf8SQu Wenruo if (ret < 0) { 1500eb793cf8SQu Wenruo struct defrag_target_range *entry; 1501eb793cf8SQu Wenruo struct defrag_target_range *tmp; 1502eb793cf8SQu Wenruo 1503eb793cf8SQu Wenruo list_for_each_entry_safe(entry, tmp, target_list, list) { 1504eb793cf8SQu Wenruo list_del_init(&entry->list); 1505eb793cf8SQu Wenruo kfree(entry); 1506eb793cf8SQu Wenruo } 1507eb793cf8SQu Wenruo } 1508966d879bSQu Wenruo if (!ret && last_scanned_ret) { 1509966d879bSQu Wenruo /* 1510966d879bSQu Wenruo * If the last extent is not a target, the caller can skip to 1511966d879bSQu Wenruo * the end of that extent. 1512966d879bSQu Wenruo * Otherwise, we can only go the end of the specified range. 1513966d879bSQu Wenruo */ 1514966d879bSQu Wenruo if (!last_is_target) 1515966d879bSQu Wenruo *last_scanned_ret = max(cur, *last_scanned_ret); 1516966d879bSQu Wenruo else 1517966d879bSQu Wenruo *last_scanned_ret = max(start + len, *last_scanned_ret); 1518966d879bSQu Wenruo } 1519eb793cf8SQu Wenruo return ret; 1520eb793cf8SQu Wenruo } 1521eb793cf8SQu Wenruo 152222b398eeSQu Wenruo #define CLUSTER_SIZE (SZ_256K) 152322b398eeSQu Wenruo 152422b398eeSQu Wenruo /* 152522b398eeSQu Wenruo * Defrag one contiguous target range. 152622b398eeSQu Wenruo * 152722b398eeSQu Wenruo * @inode: target inode 152822b398eeSQu Wenruo * @target: target range to defrag 152922b398eeSQu Wenruo * @pages: locked pages covering the defrag range 153022b398eeSQu Wenruo * @nr_pages: number of locked pages 153122b398eeSQu Wenruo * 153222b398eeSQu Wenruo * Caller should ensure: 153322b398eeSQu Wenruo * 153422b398eeSQu Wenruo * - Pages are prepared 153522b398eeSQu Wenruo * Pages should be locked, no ordered extent in the pages range, 153622b398eeSQu Wenruo * no writeback. 153722b398eeSQu Wenruo * 153822b398eeSQu Wenruo * - Extent bits are locked 153922b398eeSQu Wenruo */ 154022b398eeSQu Wenruo static int defrag_one_locked_target(struct btrfs_inode *inode, 154122b398eeSQu Wenruo struct defrag_target_range *target, 154222b398eeSQu Wenruo struct page **pages, int nr_pages, 154322b398eeSQu Wenruo struct extent_state **cached_state) 154422b398eeSQu Wenruo { 154522b398eeSQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 154622b398eeSQu Wenruo struct extent_changeset *data_reserved = NULL; 154722b398eeSQu Wenruo const u64 start = target->start; 154822b398eeSQu Wenruo const u64 len = target->len; 154922b398eeSQu Wenruo unsigned long last_index = (start + len - 1) >> PAGE_SHIFT; 155022b398eeSQu Wenruo unsigned long start_index = start >> PAGE_SHIFT; 155122b398eeSQu Wenruo unsigned long first_index = page_index(pages[0]); 155222b398eeSQu Wenruo int ret = 0; 155322b398eeSQu Wenruo int i; 155422b398eeSQu Wenruo 155522b398eeSQu Wenruo ASSERT(last_index - first_index + 1 <= nr_pages); 155622b398eeSQu Wenruo 155722b398eeSQu Wenruo ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len); 155822b398eeSQu Wenruo if (ret < 0) 155922b398eeSQu Wenruo return ret; 156022b398eeSQu Wenruo clear_extent_bit(&inode->io_tree, start, start + len - 1, 156122b398eeSQu Wenruo EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | 156222b398eeSQu Wenruo EXTENT_DEFRAG, 0, 0, cached_state); 156322b398eeSQu Wenruo set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state); 156422b398eeSQu Wenruo 156522b398eeSQu Wenruo /* Update the page status */ 156622b398eeSQu Wenruo for (i = start_index - first_index; i <= last_index - first_index; i++) { 156722b398eeSQu Wenruo ClearPageChecked(pages[i]); 156822b398eeSQu Wenruo btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len); 156922b398eeSQu Wenruo } 157022b398eeSQu Wenruo btrfs_delalloc_release_extents(inode, len); 157122b398eeSQu Wenruo extent_changeset_free(data_reserved); 157222b398eeSQu Wenruo 157322b398eeSQu Wenruo return ret; 157422b398eeSQu Wenruo } 157522b398eeSQu Wenruo 1576e9eec721SQu Wenruo static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len, 1577966d879bSQu Wenruo u32 extent_thresh, u64 newer_than, bool do_compress, 1578966d879bSQu Wenruo u64 *last_scanned_ret) 1579e9eec721SQu Wenruo { 1580e9eec721SQu Wenruo struct extent_state *cached_state = NULL; 1581e9eec721SQu Wenruo struct defrag_target_range *entry; 1582e9eec721SQu Wenruo struct defrag_target_range *tmp; 1583e9eec721SQu Wenruo LIST_HEAD(target_list); 1584e9eec721SQu Wenruo struct page **pages; 1585e9eec721SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1586e9eec721SQu Wenruo u64 last_index = (start + len - 1) >> PAGE_SHIFT; 1587e9eec721SQu Wenruo u64 start_index = start >> PAGE_SHIFT; 1588e9eec721SQu Wenruo unsigned int nr_pages = last_index - start_index + 1; 1589e9eec721SQu Wenruo int ret = 0; 1590e9eec721SQu Wenruo int i; 1591e9eec721SQu Wenruo 1592e9eec721SQu Wenruo ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE); 1593e9eec721SQu Wenruo ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize)); 1594e9eec721SQu Wenruo 1595e9eec721SQu Wenruo pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); 1596e9eec721SQu Wenruo if (!pages) 1597e9eec721SQu Wenruo return -ENOMEM; 1598e9eec721SQu Wenruo 1599e9eec721SQu Wenruo /* Prepare all pages */ 1600e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1601e9eec721SQu Wenruo pages[i] = defrag_prepare_one_page(inode, start_index + i); 1602e9eec721SQu Wenruo if (IS_ERR(pages[i])) { 1603e9eec721SQu Wenruo ret = PTR_ERR(pages[i]); 1604e9eec721SQu Wenruo pages[i] = NULL; 1605e9eec721SQu Wenruo goto free_pages; 1606e9eec721SQu Wenruo } 1607e9eec721SQu Wenruo } 1608e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) 16094cb5300bSChris Mason wait_on_page_writeback(pages[i]); 16104cb5300bSChris Mason 1611e9eec721SQu Wenruo /* Lock the pages range */ 1612e9eec721SQu Wenruo lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT, 1613e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1614e9eec721SQu Wenruo &cached_state); 1615e9eec721SQu Wenruo /* 1616e9eec721SQu Wenruo * Now we have a consistent view about the extent map, re-check 1617e9eec721SQu Wenruo * which range really needs to be defragged. 1618e9eec721SQu Wenruo * 1619e9eec721SQu Wenruo * And this time we have extent locked already, pass @locked = true 1620e9eec721SQu Wenruo * so that we won't relock the extent range and cause deadlock. 1621e9eec721SQu Wenruo */ 1622e9eec721SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1623e9eec721SQu Wenruo newer_than, do_compress, true, 1624966d879bSQu Wenruo &target_list, last_scanned_ret); 1625e9eec721SQu Wenruo if (ret < 0) 1626e9eec721SQu Wenruo goto unlock_extent; 16274cb5300bSChris Mason 1628e9eec721SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1629e9eec721SQu Wenruo ret = defrag_one_locked_target(inode, entry, pages, nr_pages, 1630e9eec721SQu Wenruo &cached_state); 1631e9eec721SQu Wenruo if (ret < 0) 1632e9eec721SQu Wenruo break; 1633e9eec721SQu Wenruo } 1634e9eec721SQu Wenruo 1635e9eec721SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1636e9eec721SQu Wenruo list_del_init(&entry->list); 1637e9eec721SQu Wenruo kfree(entry); 1638e9eec721SQu Wenruo } 1639e9eec721SQu Wenruo unlock_extent: 1640e9eec721SQu Wenruo unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT, 1641e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1642e9eec721SQu Wenruo &cached_state); 1643e9eec721SQu Wenruo free_pages: 1644e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1645e9eec721SQu Wenruo if (pages[i]) { 1646e9eec721SQu Wenruo unlock_page(pages[i]); 1647e9eec721SQu Wenruo put_page(pages[i]); 1648e9eec721SQu Wenruo } 1649e9eec721SQu Wenruo } 1650e9eec721SQu Wenruo kfree(pages); 1651e9eec721SQu Wenruo return ret; 1652e9eec721SQu Wenruo } 1653e9eec721SQu Wenruo 1654b18c3ab2SQu Wenruo static int defrag_one_cluster(struct btrfs_inode *inode, 1655b18c3ab2SQu Wenruo struct file_ra_state *ra, 1656b18c3ab2SQu Wenruo u64 start, u32 len, u32 extent_thresh, 1657b18c3ab2SQu Wenruo u64 newer_than, bool do_compress, 1658b18c3ab2SQu Wenruo unsigned long *sectors_defragged, 1659966d879bSQu Wenruo unsigned long max_sectors, 1660966d879bSQu Wenruo u64 *last_scanned_ret) 1661b18c3ab2SQu Wenruo { 1662b18c3ab2SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1663b18c3ab2SQu Wenruo struct defrag_target_range *entry; 1664b18c3ab2SQu Wenruo struct defrag_target_range *tmp; 1665b18c3ab2SQu Wenruo LIST_HEAD(target_list); 1666b18c3ab2SQu Wenruo int ret; 1667b18c3ab2SQu Wenruo 1668b18c3ab2SQu Wenruo BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE)); 1669b18c3ab2SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1670b18c3ab2SQu Wenruo newer_than, do_compress, false, 1671966d879bSQu Wenruo &target_list, NULL); 1672b18c3ab2SQu Wenruo if (ret < 0) 1673b18c3ab2SQu Wenruo goto out; 1674b18c3ab2SQu Wenruo 1675b18c3ab2SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1676b18c3ab2SQu Wenruo u32 range_len = entry->len; 1677b18c3ab2SQu Wenruo 1678484167daSQu Wenruo /* Reached or beyond the limit */ 1679c080b414SQu Wenruo if (max_sectors && *sectors_defragged >= max_sectors) { 1680c080b414SQu Wenruo ret = 1; 1681b18c3ab2SQu Wenruo break; 1682c080b414SQu Wenruo } 1683b18c3ab2SQu Wenruo 1684b18c3ab2SQu Wenruo if (max_sectors) 1685b18c3ab2SQu Wenruo range_len = min_t(u32, range_len, 1686b18c3ab2SQu Wenruo (max_sectors - *sectors_defragged) * sectorsize); 1687b18c3ab2SQu Wenruo 1688966d879bSQu Wenruo /* 1689966d879bSQu Wenruo * If defrag_one_range() has updated last_scanned_ret, 1690966d879bSQu Wenruo * our range may already be invalid (e.g. hole punched). 1691966d879bSQu Wenruo * Skip if our range is before last_scanned_ret, as there is 1692966d879bSQu Wenruo * no need to defrag the range anymore. 1693966d879bSQu Wenruo */ 1694966d879bSQu Wenruo if (entry->start + range_len <= *last_scanned_ret) 1695966d879bSQu Wenruo continue; 1696966d879bSQu Wenruo 1697b18c3ab2SQu Wenruo if (ra) 1698b18c3ab2SQu Wenruo page_cache_sync_readahead(inode->vfs_inode.i_mapping, 1699b18c3ab2SQu Wenruo ra, NULL, entry->start >> PAGE_SHIFT, 1700b18c3ab2SQu Wenruo ((entry->start + range_len - 1) >> PAGE_SHIFT) - 1701b18c3ab2SQu Wenruo (entry->start >> PAGE_SHIFT) + 1); 1702b18c3ab2SQu Wenruo /* 1703b18c3ab2SQu Wenruo * Here we may not defrag any range if holes are punched before 1704b18c3ab2SQu Wenruo * we locked the pages. 1705b18c3ab2SQu Wenruo * But that's fine, it only affects the @sectors_defragged 1706b18c3ab2SQu Wenruo * accounting. 1707b18c3ab2SQu Wenruo */ 1708b18c3ab2SQu Wenruo ret = defrag_one_range(inode, entry->start, range_len, 1709966d879bSQu Wenruo extent_thresh, newer_than, do_compress, 1710966d879bSQu Wenruo last_scanned_ret); 1711b18c3ab2SQu Wenruo if (ret < 0) 1712b18c3ab2SQu Wenruo break; 1713484167daSQu Wenruo *sectors_defragged += range_len >> 1714484167daSQu Wenruo inode->root->fs_info->sectorsize_bits; 1715b18c3ab2SQu Wenruo } 1716b18c3ab2SQu Wenruo out: 1717b18c3ab2SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1718b18c3ab2SQu Wenruo list_del_init(&entry->list); 1719b18c3ab2SQu Wenruo kfree(entry); 1720b18c3ab2SQu Wenruo } 1721966d879bSQu Wenruo if (ret >= 0) 1722966d879bSQu Wenruo *last_scanned_ret = max(*last_scanned_ret, start + len); 1723b18c3ab2SQu Wenruo return ret; 1724b18c3ab2SQu Wenruo } 17257f458a38SFilipe Manana 17267f458a38SFilipe Manana /* 17271ccc2e8aSQu Wenruo * Entry point to file defragmentation. 17281ccc2e8aSQu Wenruo * 17291ccc2e8aSQu Wenruo * @inode: inode to be defragged 17301ccc2e8aSQu Wenruo * @ra: readahead state (can be NUL) 17311ccc2e8aSQu Wenruo * @range: defrag options including range and flags 17321ccc2e8aSQu Wenruo * @newer_than: minimum transid to defrag 17331ccc2e8aSQu Wenruo * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode 17341ccc2e8aSQu Wenruo * will be defragged. 1735484167daSQu Wenruo * 1736484167daSQu Wenruo * Return <0 for error. 1737c080b414SQu Wenruo * Return >=0 for the number of sectors defragged, and range->start will be updated 1738c080b414SQu Wenruo * to indicate the file offset where next defrag should be started at. 1739c080b414SQu Wenruo * (Mostly for autodefrag, which sets @max_to_defrag thus we may exit early without 1740c080b414SQu Wenruo * defragging all the range). 17417f458a38SFilipe Manana */ 17421ccc2e8aSQu Wenruo int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, 17434cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 17444cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 17454cb5300bSChris Mason { 17460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 17477b508037SQu Wenruo unsigned long sectors_defragged = 0; 1748151a31b2SLi Zefan u64 isize = i_size_read(inode); 17497b508037SQu Wenruo u64 cur; 17507b508037SQu Wenruo u64 last_byte; 17511e2ef46dSDavid Sterba bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS; 17521ccc2e8aSQu Wenruo bool ra_allocated = false; 17537b508037SQu Wenruo int compress_type = BTRFS_COMPRESS_ZLIB; 17547b508037SQu Wenruo int ret = 0; 17557b508037SQu Wenruo u32 extent_thresh = range->extent_thresh; 175627cdfde1SFilipe Manana pgoff_t start_index; 17574cb5300bSChris Mason 17580abd5b17SLiu Bo if (isize == 0) 17590abd5b17SLiu Bo return 0; 17600abd5b17SLiu Bo 17610abd5b17SLiu Bo if (range->start >= isize) 17620abd5b17SLiu Bo return -EINVAL; 17631a419d85SLi Zefan 17641e2ef46dSDavid Sterba if (do_compress) { 1765ce96b7ffSChengguang Xu if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES) 17661a419d85SLi Zefan return -EINVAL; 17671a419d85SLi Zefan if (range->compress_type) 17681a419d85SLi Zefan compress_type = range->compress_type; 17691a419d85SLi Zefan } 1770f46b5a66SChristoph Hellwig 17710abd5b17SLiu Bo if (extent_thresh == 0) 1772ee22184bSByongho Lee extent_thresh = SZ_256K; 1773f46b5a66SChristoph Hellwig 17747b508037SQu Wenruo if (range->start + range->len > range->start) { 17757b508037SQu Wenruo /* Got a specific range */ 17766b34cd8eSFilipe Manana last_byte = min(isize, range->start + range->len); 17777b508037SQu Wenruo } else { 17787b508037SQu Wenruo /* Defrag until file end */ 17796b34cd8eSFilipe Manana last_byte = isize; 17807b508037SQu Wenruo } 17817b508037SQu Wenruo 17826b34cd8eSFilipe Manana /* Align the range */ 17836b34cd8eSFilipe Manana cur = round_down(range->start, fs_info->sectorsize); 17846b34cd8eSFilipe Manana last_byte = round_up(last_byte, fs_info->sectorsize) - 1; 17856b34cd8eSFilipe Manana 17864cb5300bSChris Mason /* 17871ccc2e8aSQu Wenruo * If we were not given a ra, allocate a readahead context. As 17880a52d108SDavid Sterba * readahead is just an optimization, defrag will work without it so 17890a52d108SDavid Sterba * we don't error out. 17904cb5300bSChris Mason */ 17911ccc2e8aSQu Wenruo if (!ra) { 17921ccc2e8aSQu Wenruo ra_allocated = true; 179363e727ecSDavid Sterba ra = kzalloc(sizeof(*ra), GFP_KERNEL); 17940a52d108SDavid Sterba if (ra) 17954cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 17964cb5300bSChris Mason } 17974cb5300bSChris Mason 179827cdfde1SFilipe Manana /* 179927cdfde1SFilipe Manana * Make writeback start from the beginning of the range, so that the 180027cdfde1SFilipe Manana * defrag range can be written sequentially. 180127cdfde1SFilipe Manana */ 180227cdfde1SFilipe Manana start_index = cur >> PAGE_SHIFT; 180327cdfde1SFilipe Manana if (start_index < inode->i_mapping->writeback_index) 180427cdfde1SFilipe Manana inode->i_mapping->writeback_index = start_index; 180527cdfde1SFilipe Manana 18067b508037SQu Wenruo while (cur < last_byte) { 18073c9d31c7SFilipe Manana const unsigned long prev_sectors_defragged = sectors_defragged; 1808966d879bSQu Wenruo u64 last_scanned = cur; 18097b508037SQu Wenruo u64 cluster_end; 18101e701a32SChris Mason 18117b508037SQu Wenruo /* The cluster size 256K should always be page aligned */ 18127b508037SQu Wenruo BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE)); 18134cb5300bSChris Mason 1814b767c2fcSFilipe Manana if (btrfs_defrag_cancelled(fs_info)) { 1815b767c2fcSFilipe Manana ret = -EAGAIN; 1816b767c2fcSFilipe Manana break; 1817b767c2fcSFilipe Manana } 1818b767c2fcSFilipe Manana 18197b508037SQu Wenruo /* We want the cluster end at page boundary when possible */ 18207b508037SQu Wenruo cluster_end = (((cur >> PAGE_SHIFT) + 18217b508037SQu Wenruo (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1; 18227b508037SQu Wenruo cluster_end = min(cluster_end, last_byte); 18234cb5300bSChris Mason 182464708539SJosef Bacik btrfs_inode_lock(inode, 0); 1825eede2bf3SOmar Sandoval if (IS_SWAPFILE(inode)) { 1826eede2bf3SOmar Sandoval ret = -ETXTBSY; 18277b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 18287b508037SQu Wenruo break; 18297b508037SQu Wenruo } 18307b508037SQu Wenruo if (!(inode->i_sb->s_flags & SB_ACTIVE)) { 18317b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 18327b508037SQu Wenruo break; 18337b508037SQu Wenruo } 18341e2ef46dSDavid Sterba if (do_compress) 1835eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = compress_type; 18367b508037SQu Wenruo ret = defrag_one_cluster(BTRFS_I(inode), ra, cur, 18377b508037SQu Wenruo cluster_end + 1 - cur, extent_thresh, 1838966d879bSQu Wenruo newer_than, do_compress, §ors_defragged, 1839966d879bSQu Wenruo max_to_defrag, &last_scanned); 18403c9d31c7SFilipe Manana 18413c9d31c7SFilipe Manana if (sectors_defragged > prev_sectors_defragged) 18423c9d31c7SFilipe Manana balance_dirty_pages_ratelimited(inode->i_mapping); 18433c9d31c7SFilipe Manana 184464708539SJosef Bacik btrfs_inode_unlock(inode, 0); 18457b508037SQu Wenruo if (ret < 0) 18464cb5300bSChris Mason break; 1847966d879bSQu Wenruo cur = max(cluster_end + 1, last_scanned); 1848c080b414SQu Wenruo if (ret > 0) { 1849c080b414SQu Wenruo ret = 0; 1850c080b414SQu Wenruo break; 1851c080b414SQu Wenruo } 1852ea0eba69SQu Wenruo cond_resched(); 18534cb5300bSChris Mason } 1854f46b5a66SChristoph Hellwig 18557b508037SQu Wenruo if (ra_allocated) 18567b508037SQu Wenruo kfree(ra); 1857c080b414SQu Wenruo /* 1858c080b414SQu Wenruo * Update range.start for autodefrag, this will indicate where to start 1859c080b414SQu Wenruo * in next run. 1860c080b414SQu Wenruo */ 1861c080b414SQu Wenruo range->start = cur; 18627b508037SQu Wenruo if (sectors_defragged) { 18637b508037SQu Wenruo /* 18647b508037SQu Wenruo * We have defragged some sectors, for compression case they 18657b508037SQu Wenruo * need to be written back immediately. 18667b508037SQu Wenruo */ 18677b508037SQu Wenruo if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) { 18681e701a32SChris Mason filemap_flush(inode->i_mapping); 1869dec8ef90SFilipe Manana if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, 1870dec8ef90SFilipe Manana &BTRFS_I(inode)->runtime_flags)) 1871dec8ef90SFilipe Manana filemap_flush(inode->i_mapping); 1872dec8ef90SFilipe Manana } 18737b508037SQu Wenruo if (range->compress_type == BTRFS_COMPRESS_LZO) 18740b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, COMPRESS_LZO); 18757b508037SQu Wenruo else if (range->compress_type == BTRFS_COMPRESS_ZSTD) 18765c1aab1dSNick Terrell btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD); 18777b508037SQu Wenruo ret = sectors_defragged; 18781a419d85SLi Zefan } 18791e2ef46dSDavid Sterba if (do_compress) { 188064708539SJosef Bacik btrfs_inode_lock(inode, 0); 1881eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE; 188264708539SJosef Bacik btrfs_inode_unlock(inode, 0); 1883633085c7SFilipe David Borba Manana } 1884940100a4SChris Mason return ret; 1885f46b5a66SChristoph Hellwig } 1886f46b5a66SChristoph Hellwig 188717aaa434SDavid Sterba /* 188817aaa434SDavid Sterba * Try to start exclusive operation @type or cancel it if it's running. 188917aaa434SDavid Sterba * 189017aaa434SDavid Sterba * Return: 189117aaa434SDavid Sterba * 0 - normal mode, newly claimed op started 189217aaa434SDavid Sterba * >0 - normal mode, something else is running, 189317aaa434SDavid Sterba * return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space 189417aaa434SDavid Sterba * ECANCELED - cancel mode, successful cancel 189517aaa434SDavid Sterba * ENOTCONN - cancel mode, operation not running anymore 189617aaa434SDavid Sterba */ 189717aaa434SDavid Sterba static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info, 189817aaa434SDavid Sterba enum btrfs_exclusive_operation type, bool cancel) 189917aaa434SDavid Sterba { 190017aaa434SDavid Sterba if (!cancel) { 190117aaa434SDavid Sterba /* Start normal op */ 190217aaa434SDavid Sterba if (!btrfs_exclop_start(fs_info, type)) 190317aaa434SDavid Sterba return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 190417aaa434SDavid Sterba /* Exclusive operation is now claimed */ 190517aaa434SDavid Sterba return 0; 190617aaa434SDavid Sterba } 190717aaa434SDavid Sterba 190817aaa434SDavid Sterba /* Cancel running op */ 190917aaa434SDavid Sterba if (btrfs_exclop_start_try_lock(fs_info, type)) { 191017aaa434SDavid Sterba /* 191117aaa434SDavid Sterba * This blocks any exclop finish from setting it to NONE, so we 191217aaa434SDavid Sterba * request cancellation. Either it runs and we will wait for it, 191317aaa434SDavid Sterba * or it has finished and no waiting will happen. 191417aaa434SDavid Sterba */ 191517aaa434SDavid Sterba atomic_inc(&fs_info->reloc_cancel_req); 191617aaa434SDavid Sterba btrfs_exclop_start_unlock(fs_info); 191717aaa434SDavid Sterba 191817aaa434SDavid Sterba if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 191917aaa434SDavid Sterba wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING, 192017aaa434SDavid Sterba TASK_INTERRUPTIBLE); 192117aaa434SDavid Sterba 192217aaa434SDavid Sterba return -ECANCELED; 192317aaa434SDavid Sterba } 192417aaa434SDavid Sterba 192517aaa434SDavid Sterba /* Something else is running or none */ 192617aaa434SDavid Sterba return -ENOTCONN; 192717aaa434SDavid Sterba } 192817aaa434SDavid Sterba 1929198605a8SMiao Xie static noinline int btrfs_ioctl_resize(struct file *file, 193076dda93cSYan, Zheng void __user *arg) 1931f46b5a66SChristoph Hellwig { 1932562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 19330b246afaSJeff Mahoney struct inode *inode = file_inode(file); 19340b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1935f46b5a66SChristoph Hellwig u64 new_size; 1936f46b5a66SChristoph Hellwig u64 old_size; 1937f46b5a66SChristoph Hellwig u64 devid = 1; 19380b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 1939f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1940f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1941f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1942f46b5a66SChristoph Hellwig char *sizestr; 19439a40f122SGui Hecheng char *retptr; 1944f46b5a66SChristoph Hellwig char *devstr = NULL; 1945f46b5a66SChristoph Hellwig int ret = 0; 1946f46b5a66SChristoph Hellwig int mod = 0; 1947bb059a37SDavid Sterba bool cancel; 1948f46b5a66SChristoph Hellwig 1949e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1950e441d54dSChris Mason return -EPERM; 1951e441d54dSChris Mason 1952198605a8SMiao Xie ret = mnt_want_write_file(file); 1953198605a8SMiao Xie if (ret) 1954198605a8SMiao Xie return ret; 1955198605a8SMiao Xie 1956bb059a37SDavid Sterba /* 1957bb059a37SDavid Sterba * Read the arguments before checking exclusivity to be able to 1958bb059a37SDavid Sterba * distinguish regular resize and cancel 1959bb059a37SDavid Sterba */ 1960dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1961c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1962c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1963bb059a37SDavid Sterba goto out_drop; 1964c9e9f97bSIlya Dryomov } 19655516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1966f46b5a66SChristoph Hellwig sizestr = vol_args->name; 1967bb059a37SDavid Sterba cancel = (strcmp("cancel", sizestr) == 0); 1968bb059a37SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel); 1969bb059a37SDavid Sterba if (ret) 1970bb059a37SDavid Sterba goto out_free; 1971bb059a37SDavid Sterba /* Exclusive operation is now claimed */ 1972bb059a37SDavid Sterba 1973f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 1974f46b5a66SChristoph Hellwig if (devstr) { 1975f46b5a66SChristoph Hellwig sizestr = devstr + 1; 1976f46b5a66SChristoph Hellwig *devstr = '\0'; 1977f46b5a66SChristoph Hellwig devstr = vol_args->name; 197858dfae63SZhangZhen ret = kstrtoull(devstr, 10, &devid); 197958dfae63SZhangZhen if (ret) 1980bb059a37SDavid Sterba goto out_finish; 1981dfd79829SMiao Xie if (!devid) { 1982dfd79829SMiao Xie ret = -EINVAL; 1983bb059a37SDavid Sterba goto out_finish; 1984dfd79829SMiao Xie } 19850b246afaSJeff Mahoney btrfs_info(fs_info, "resizing devid %llu", devid); 1986f46b5a66SChristoph Hellwig } 1987dba60f3fSMiao Xie 1988562d7b15SJosef Bacik args.devid = devid; 1989562d7b15SJosef Bacik device = btrfs_find_device(fs_info->fs_devices, &args); 1990f46b5a66SChristoph Hellwig if (!device) { 19910b246afaSJeff Mahoney btrfs_info(fs_info, "resizer unable to find device %llu", 1992c1c9ff7cSGeert Uytterhoeven devid); 1993dfd79829SMiao Xie ret = -ENODEV; 1994bb059a37SDavid Sterba goto out_finish; 1995f46b5a66SChristoph Hellwig } 1996dba60f3fSMiao Xie 1997ebbede42SAnand Jain if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 19980b246afaSJeff Mahoney btrfs_info(fs_info, 1999efe120a0SFrank Holton "resizer unable to apply on readonly device %llu", 2000c1c9ff7cSGeert Uytterhoeven devid); 2001dfd79829SMiao Xie ret = -EPERM; 2002bb059a37SDavid Sterba goto out_finish; 20034e42ae1bSLiu Bo } 20044e42ae1bSLiu Bo 2005f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 2006cda00ebaSChristoph Hellwig new_size = bdev_nr_bytes(device->bdev); 2007f46b5a66SChristoph Hellwig else { 2008f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 2009f46b5a66SChristoph Hellwig mod = -1; 2010f46b5a66SChristoph Hellwig sizestr++; 2011f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 2012f46b5a66SChristoph Hellwig mod = 1; 2013f46b5a66SChristoph Hellwig sizestr++; 2014f46b5a66SChristoph Hellwig } 20159a40f122SGui Hecheng new_size = memparse(sizestr, &retptr); 20169a40f122SGui Hecheng if (*retptr != '\0' || new_size == 0) { 2017f46b5a66SChristoph Hellwig ret = -EINVAL; 2018bb059a37SDavid Sterba goto out_finish; 2019f46b5a66SChristoph Hellwig } 2020f46b5a66SChristoph Hellwig } 2021f46b5a66SChristoph Hellwig 2022401e29c1SAnand Jain if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 2023dfd79829SMiao Xie ret = -EPERM; 2024bb059a37SDavid Sterba goto out_finish; 202563a212abSStefan Behrens } 202663a212abSStefan Behrens 20277cc8e58dSMiao Xie old_size = btrfs_device_get_total_bytes(device); 2028f46b5a66SChristoph Hellwig 2029f46b5a66SChristoph Hellwig if (mod < 0) { 2030f46b5a66SChristoph Hellwig if (new_size > old_size) { 2031f46b5a66SChristoph Hellwig ret = -EINVAL; 2032bb059a37SDavid Sterba goto out_finish; 2033f46b5a66SChristoph Hellwig } 2034f46b5a66SChristoph Hellwig new_size = old_size - new_size; 2035f46b5a66SChristoph Hellwig } else if (mod > 0) { 2036eb8052e0SWenliang Fan if (new_size > ULLONG_MAX - old_size) { 2037902c68a4SGui Hecheng ret = -ERANGE; 2038bb059a37SDavid Sterba goto out_finish; 2039eb8052e0SWenliang Fan } 2040f46b5a66SChristoph Hellwig new_size = old_size + new_size; 2041f46b5a66SChristoph Hellwig } 2042f46b5a66SChristoph Hellwig 2043ee22184bSByongho Lee if (new_size < SZ_256M) { 2044f46b5a66SChristoph Hellwig ret = -EINVAL; 2045bb059a37SDavid Sterba goto out_finish; 2046f46b5a66SChristoph Hellwig } 2047cda00ebaSChristoph Hellwig if (new_size > bdev_nr_bytes(device->bdev)) { 2048f46b5a66SChristoph Hellwig ret = -EFBIG; 2049bb059a37SDavid Sterba goto out_finish; 2050f46b5a66SChristoph Hellwig } 2051f46b5a66SChristoph Hellwig 205247f08b96SNikolay Borisov new_size = round_down(new_size, fs_info->sectorsize); 2053f46b5a66SChristoph Hellwig 2054f46b5a66SChristoph Hellwig if (new_size > old_size) { 2055a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 205698d5dc13STsutomu Itoh if (IS_ERR(trans)) { 205798d5dc13STsutomu Itoh ret = PTR_ERR(trans); 2058bb059a37SDavid Sterba goto out_finish; 205998d5dc13STsutomu Itoh } 2060f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 20613a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 2062ece7d20eSMike Fleetwood } else if (new_size < old_size) { 2063f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 20640253f40eSjeff.liu } /* equal, nothing need to do */ 2065f46b5a66SChristoph Hellwig 2066faf8f7b9SMarcos Paulo de Souza if (ret == 0 && new_size != old_size) 2067faf8f7b9SMarcos Paulo de Souza btrfs_info_in_rcu(fs_info, 2068faf8f7b9SMarcos Paulo de Souza "resize device %s (devid %llu) from %llu to %llu", 2069faf8f7b9SMarcos Paulo de Souza rcu_str_deref(device->name), device->devid, 2070faf8f7b9SMarcos Paulo de Souza old_size, new_size); 2071bb059a37SDavid Sterba out_finish: 2072bb059a37SDavid Sterba btrfs_exclop_finish(fs_info); 2073c9e9f97bSIlya Dryomov out_free: 2074f46b5a66SChristoph Hellwig kfree(vol_args); 2075bb059a37SDavid Sterba out_drop: 207618f39c41SIlya Dryomov mnt_drop_write_file(file); 2077f46b5a66SChristoph Hellwig return ret; 2078f46b5a66SChristoph Hellwig } 2079f46b5a66SChristoph Hellwig 20805d54c67eSNikolay Borisov static noinline int __btrfs_ioctl_snap_create(struct file *file, 20814d4340c9SChristian Brauner struct user_namespace *mnt_userns, 208252f75f4fSDavid Sterba const char *name, unsigned long fd, int subvol, 20835d54c67eSNikolay Borisov bool readonly, 20848696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 2085f46b5a66SChristoph Hellwig { 2086f46b5a66SChristoph Hellwig int namelen; 20873de4586cSChris Mason int ret = 0; 2088f46b5a66SChristoph Hellwig 2089325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2090325c50e3SJeff Mahoney return -ENOTDIR; 2091325c50e3SJeff Mahoney 2092a874a63eSLiu Bo ret = mnt_want_write_file(file); 2093a874a63eSLiu Bo if (ret) 2094a874a63eSLiu Bo goto out; 2095a874a63eSLiu Bo 209672fd032eSSage Weil namelen = strlen(name); 209772fd032eSSage Weil if (strchr(name, '/')) { 2098f46b5a66SChristoph Hellwig ret = -EINVAL; 2099a874a63eSLiu Bo goto out_drop_write; 2100f46b5a66SChristoph Hellwig } 2101f46b5a66SChristoph Hellwig 210216780cabSChris Mason if (name[0] == '.' && 210316780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 210416780cabSChris Mason ret = -EEXIST; 2105a874a63eSLiu Bo goto out_drop_write; 210616780cabSChris Mason } 210716780cabSChris Mason 21083de4586cSChris Mason if (subvol) { 21094d4340c9SChristian Brauner ret = btrfs_mksubvol(&file->f_path, mnt_userns, name, 21104d4340c9SChristian Brauner namelen, NULL, readonly, inherit); 2111cb8e7090SChristoph Hellwig } else { 21122903ff01SAl Viro struct fd src = fdget(fd); 21133de4586cSChris Mason struct inode *src_inode; 21142903ff01SAl Viro if (!src.file) { 21153de4586cSChris Mason ret = -EINVAL; 2116a874a63eSLiu Bo goto out_drop_write; 21173de4586cSChris Mason } 21183de4586cSChris Mason 2119496ad9aaSAl Viro src_inode = file_inode(src.file); 2120496ad9aaSAl Viro if (src_inode->i_sb != file_inode(file)->i_sb) { 2121c79b4713SJosef Bacik btrfs_info(BTRFS_I(file_inode(file))->root->fs_info, 2122efe120a0SFrank Holton "Snapshot src from another FS"); 212323ad5b17SKusanagi Kouichi ret = -EXDEV; 21244d4340c9SChristian Brauner } else if (!inode_owner_or_capable(mnt_userns, src_inode)) { 2125d0242061SDavid Sterba /* 2126d0242061SDavid Sterba * Subvolume creation is not restricted, but snapshots 2127d0242061SDavid Sterba * are limited to own subvolumes only 2128d0242061SDavid Sterba */ 2129d0242061SDavid Sterba ret = -EPERM; 2130ecd18815SAl Viro } else { 21314d4340c9SChristian Brauner ret = btrfs_mksnapshot(&file->f_path, mnt_userns, 21324d4340c9SChristian Brauner name, namelen, 213372fd032eSSage Weil BTRFS_I(src_inode)->root, 21345d54c67eSNikolay Borisov readonly, inherit); 2135ecd18815SAl Viro } 21362903ff01SAl Viro fdput(src); 2137cb8e7090SChristoph Hellwig } 2138a874a63eSLiu Bo out_drop_write: 2139a874a63eSLiu Bo mnt_drop_write_file(file); 2140f46b5a66SChristoph Hellwig out: 214172fd032eSSage Weil return ret; 214272fd032eSSage Weil } 214372fd032eSSage Weil 214472fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 2145fa0d2b9bSLi Zefan void __user *arg, int subvol) 214672fd032eSSage Weil { 2147fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 214872fd032eSSage Weil int ret; 214972fd032eSSage Weil 2150325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2151325c50e3SJeff Mahoney return -ENOTDIR; 2152325c50e3SJeff Mahoney 2153fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2154fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 2155fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 2156fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2157fa0d2b9bSLi Zefan 21584d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 21594d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 21604d4340c9SChristian Brauner false, NULL); 2161fa0d2b9bSLi Zefan 2162fa0d2b9bSLi Zefan kfree(vol_args); 2163fa0d2b9bSLi Zefan return ret; 2164fa0d2b9bSLi Zefan } 2165fa0d2b9bSLi Zefan 2166fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 2167fa0d2b9bSLi Zefan void __user *arg, int subvol) 2168fa0d2b9bSLi Zefan { 2169fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 2170fa0d2b9bSLi Zefan int ret; 2171b83cc969SLi Zefan bool readonly = false; 21726f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 217372fd032eSSage Weil 2174325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 2175325c50e3SJeff Mahoney return -ENOTDIR; 2176325c50e3SJeff Mahoney 2177fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2178fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 2179fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 2180fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 2181fdfb1e4fSLi Zefan 2182673990dbSDavid Sterba if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) { 2183b83cc969SLi Zefan ret = -EOPNOTSUPP; 2184c47ca32dSDan Carpenter goto free_args; 2185fdfb1e4fSLi Zefan } 2186fdfb1e4fSLi Zefan 2187b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 2188b83cc969SLi Zefan readonly = true; 21896f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 21905011c5a6SDan Carpenter u64 nums; 21915011c5a6SDan Carpenter 21925011c5a6SDan Carpenter if (vol_args->size < sizeof(*inherit) || 21935011c5a6SDan Carpenter vol_args->size > PAGE_SIZE) { 21946f72c7e2SArne Jansen ret = -EINVAL; 2195c47ca32dSDan Carpenter goto free_args; 21966f72c7e2SArne Jansen } 21976f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 21986f72c7e2SArne Jansen if (IS_ERR(inherit)) { 21996f72c7e2SArne Jansen ret = PTR_ERR(inherit); 2200c47ca32dSDan Carpenter goto free_args; 22016f72c7e2SArne Jansen } 22025011c5a6SDan Carpenter 22035011c5a6SDan Carpenter if (inherit->num_qgroups > PAGE_SIZE || 22045011c5a6SDan Carpenter inherit->num_ref_copies > PAGE_SIZE || 22055011c5a6SDan Carpenter inherit->num_excl_copies > PAGE_SIZE) { 22065011c5a6SDan Carpenter ret = -EINVAL; 22075011c5a6SDan Carpenter goto free_inherit; 22085011c5a6SDan Carpenter } 22095011c5a6SDan Carpenter 22105011c5a6SDan Carpenter nums = inherit->num_qgroups + 2 * inherit->num_ref_copies + 22115011c5a6SDan Carpenter 2 * inherit->num_excl_copies; 22125011c5a6SDan Carpenter if (vol_args->size != struct_size(inherit, qgroups, nums)) { 22135011c5a6SDan Carpenter ret = -EINVAL; 22145011c5a6SDan Carpenter goto free_inherit; 22155011c5a6SDan Carpenter } 22166f72c7e2SArne Jansen } 221775eaa0e2SSage Weil 22184d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 22194d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 22204d4340c9SChristian Brauner readonly, inherit); 2221c47ca32dSDan Carpenter if (ret) 2222c47ca32dSDan Carpenter goto free_inherit; 2223c47ca32dSDan Carpenter free_inherit: 22246f72c7e2SArne Jansen kfree(inherit); 2225c47ca32dSDan Carpenter free_args: 2226c47ca32dSDan Carpenter kfree(vol_args); 2227f46b5a66SChristoph Hellwig return ret; 2228f46b5a66SChristoph Hellwig } 2229f46b5a66SChristoph Hellwig 22309ad12305SSahil Kang static noinline int btrfs_ioctl_subvol_getflags(struct inode *inode, 22310caa102dSLi Zefan void __user *arg) 22320caa102dSLi Zefan { 22330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 22340caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 22350caa102dSLi Zefan int ret = 0; 22360caa102dSLi Zefan u64 flags = 0; 22370caa102dSLi Zefan 22384a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) 22390caa102dSLi Zefan return -EINVAL; 22400caa102dSLi Zefan 22410b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 22420caa102dSLi Zefan if (btrfs_root_readonly(root)) 22430caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 22440b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 22450caa102dSLi Zefan 22460caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 22470caa102dSLi Zefan ret = -EFAULT; 22480caa102dSLi Zefan 22490caa102dSLi Zefan return ret; 22500caa102dSLi Zefan } 22510caa102dSLi Zefan 22520caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 22530caa102dSLi Zefan void __user *arg) 22540caa102dSLi Zefan { 2255496ad9aaSAl Viro struct inode *inode = file_inode(file); 22560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 22570caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 22580caa102dSLi Zefan struct btrfs_trans_handle *trans; 22590caa102dSLi Zefan u64 root_flags; 22600caa102dSLi Zefan u64 flags; 22610caa102dSLi Zefan int ret = 0; 22620caa102dSLi Zefan 226339e1674fSChristian Brauner if (!inode_owner_or_capable(file_mnt_user_ns(file), inode)) 2264bd60ea0fSDavid Sterba return -EPERM; 2265bd60ea0fSDavid Sterba 2266b9ca0664SLiu Bo ret = mnt_want_write_file(file); 2267b9ca0664SLiu Bo if (ret) 2268b9ca0664SLiu Bo goto out; 22690caa102dSLi Zefan 22704a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 2271b9ca0664SLiu Bo ret = -EINVAL; 2272b9ca0664SLiu Bo goto out_drop_write; 2273b9ca0664SLiu Bo } 22740caa102dSLi Zefan 2275b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 2276b9ca0664SLiu Bo ret = -EFAULT; 2277b9ca0664SLiu Bo goto out_drop_write; 2278b9ca0664SLiu Bo } 22790caa102dSLi Zefan 2280b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 2281b9ca0664SLiu Bo ret = -EOPNOTSUPP; 2282b9ca0664SLiu Bo goto out_drop_write; 2283b9ca0664SLiu Bo } 22840caa102dSLi Zefan 22850b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 22860caa102dSLi Zefan 22870caa102dSLi Zefan /* nothing to do */ 22880caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 2289b9ca0664SLiu Bo goto out_drop_sem; 22900caa102dSLi Zefan 22910caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 22922c686537SDavid Sterba if (flags & BTRFS_SUBVOL_RDONLY) { 22930caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 22940caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 22952c686537SDavid Sterba } else { 22962c686537SDavid Sterba /* 22972c686537SDavid Sterba * Block RO -> RW transition if this subvolume is involved in 22982c686537SDavid Sterba * send 22992c686537SDavid Sterba */ 23002c686537SDavid Sterba spin_lock(&root->root_item_lock); 23012c686537SDavid Sterba if (root->send_in_progress == 0) { 23020caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 23030caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 23042c686537SDavid Sterba spin_unlock(&root->root_item_lock); 23052c686537SDavid Sterba } else { 23062c686537SDavid Sterba spin_unlock(&root->root_item_lock); 23070b246afaSJeff Mahoney btrfs_warn(fs_info, 23082c686537SDavid Sterba "Attempt to set subvolume %llu read-write during send", 23092c686537SDavid Sterba root->root_key.objectid); 23102c686537SDavid Sterba ret = -EPERM; 23112c686537SDavid Sterba goto out_drop_sem; 23122c686537SDavid Sterba } 23132c686537SDavid Sterba } 23140caa102dSLi Zefan 23150caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 23160caa102dSLi Zefan if (IS_ERR(trans)) { 23170caa102dSLi Zefan ret = PTR_ERR(trans); 23180caa102dSLi Zefan goto out_reset; 23190caa102dSLi Zefan } 23200caa102dSLi Zefan 23210b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 23220caa102dSLi Zefan &root->root_key, &root->root_item); 23239417ebc8SNikolay Borisov if (ret < 0) { 23249417ebc8SNikolay Borisov btrfs_end_transaction(trans); 23259417ebc8SNikolay Borisov goto out_reset; 23269417ebc8SNikolay Borisov } 23270caa102dSLi Zefan 23289417ebc8SNikolay Borisov ret = btrfs_commit_transaction(trans); 23299417ebc8SNikolay Borisov 23300caa102dSLi Zefan out_reset: 23310caa102dSLi Zefan if (ret) 23320caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 2333b9ca0664SLiu Bo out_drop_sem: 23340b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 2335b9ca0664SLiu Bo out_drop_write: 2336b9ca0664SLiu Bo mnt_drop_write_file(file); 2337b9ca0664SLiu Bo out: 23380caa102dSLi Zefan return ret; 23390caa102dSLi Zefan } 23400caa102dSLi Zefan 2341ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 2342ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 2343ac8e9819SChris Mason { 2344abc6e134SChris Mason struct btrfs_key test; 2345abc6e134SChris Mason int ret; 2346abc6e134SChris Mason 2347abc6e134SChris Mason test.objectid = sk->min_objectid; 2348abc6e134SChris Mason test.type = sk->min_type; 2349abc6e134SChris Mason test.offset = sk->min_offset; 2350abc6e134SChris Mason 2351abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2352abc6e134SChris Mason if (ret < 0) 2353ac8e9819SChris Mason return 0; 2354abc6e134SChris Mason 2355abc6e134SChris Mason test.objectid = sk->max_objectid; 2356abc6e134SChris Mason test.type = sk->max_type; 2357abc6e134SChris Mason test.offset = sk->max_offset; 2358abc6e134SChris Mason 2359abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2360abc6e134SChris Mason if (ret > 0) 2361ac8e9819SChris Mason return 0; 2362ac8e9819SChris Mason return 1; 2363ac8e9819SChris Mason } 2364ac8e9819SChris Mason 2365df397565SJeff Mahoney static noinline int copy_to_sk(struct btrfs_path *path, 2366ac8e9819SChris Mason struct btrfs_key *key, 2367ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 23689b6e817dSGerhard Heift size_t *buf_size, 2369ba346b35SGerhard Heift char __user *ubuf, 2370ac8e9819SChris Mason unsigned long *sk_offset, 2371ac8e9819SChris Mason int *num_found) 2372ac8e9819SChris Mason { 2373ac8e9819SChris Mason u64 found_transid; 2374ac8e9819SChris Mason struct extent_buffer *leaf; 2375ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 2376dd81d459SNaohiro Aota struct btrfs_key test; 2377ac8e9819SChris Mason unsigned long item_off; 2378ac8e9819SChris Mason unsigned long item_len; 2379ac8e9819SChris Mason int nritems; 2380ac8e9819SChris Mason int i; 2381ac8e9819SChris Mason int slot; 2382ac8e9819SChris Mason int ret = 0; 2383ac8e9819SChris Mason 2384ac8e9819SChris Mason leaf = path->nodes[0]; 2385ac8e9819SChris Mason slot = path->slots[0]; 2386ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 2387ac8e9819SChris Mason 2388ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 2389ac8e9819SChris Mason i = nritems; 2390ac8e9819SChris Mason goto advance_key; 2391ac8e9819SChris Mason } 2392ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 2393ac8e9819SChris Mason 2394ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 2395ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 23963212fa14SJosef Bacik item_len = btrfs_item_size(leaf, i); 2397ac8e9819SChris Mason 239803b71c6cSGabriel de Perthuis btrfs_item_key_to_cpu(leaf, key, i); 239903b71c6cSGabriel de Perthuis if (!key_in_sk(key, sk)) 240003b71c6cSGabriel de Perthuis continue; 240103b71c6cSGabriel de Perthuis 24029b6e817dSGerhard Heift if (sizeof(sh) + item_len > *buf_size) { 24038f5f6178SGerhard Heift if (*num_found) { 2404ac8e9819SChris Mason ret = 1; 24058f5f6178SGerhard Heift goto out; 24068f5f6178SGerhard Heift } 24078f5f6178SGerhard Heift 24088f5f6178SGerhard Heift /* 24098f5f6178SGerhard Heift * return one empty item back for v1, which does not 24108f5f6178SGerhard Heift * handle -EOVERFLOW 24118f5f6178SGerhard Heift */ 24128f5f6178SGerhard Heift 24139b6e817dSGerhard Heift *buf_size = sizeof(sh) + item_len; 2414ac8e9819SChris Mason item_len = 0; 24158f5f6178SGerhard Heift ret = -EOVERFLOW; 24168f5f6178SGerhard Heift } 2417ac8e9819SChris Mason 24189b6e817dSGerhard Heift if (sizeof(sh) + item_len + *sk_offset > *buf_size) { 2419ac8e9819SChris Mason ret = 1; 242025c9bc2eSGerhard Heift goto out; 2421ac8e9819SChris Mason } 2422ac8e9819SChris Mason 2423ac8e9819SChris Mason sh.objectid = key->objectid; 2424ac8e9819SChris Mason sh.offset = key->offset; 2425ac8e9819SChris Mason sh.type = key->type; 2426ac8e9819SChris Mason sh.len = item_len; 2427ac8e9819SChris Mason sh.transid = found_transid; 2428ac8e9819SChris Mason 2429a48b73ecSJosef Bacik /* 2430a48b73ecSJosef Bacik * Copy search result header. If we fault then loop again so we 2431a48b73ecSJosef Bacik * can fault in the pages and -EFAULT there if there's a 2432a48b73ecSJosef Bacik * problem. Otherwise we'll fault and then copy the buffer in 2433a48b73ecSJosef Bacik * properly this next time through 2434a48b73ecSJosef Bacik */ 2435a48b73ecSJosef Bacik if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) { 2436a48b73ecSJosef Bacik ret = 0; 2437ba346b35SGerhard Heift goto out; 2438ba346b35SGerhard Heift } 2439ba346b35SGerhard Heift 2440ac8e9819SChris Mason *sk_offset += sizeof(sh); 2441ac8e9819SChris Mason 2442ac8e9819SChris Mason if (item_len) { 2443ba346b35SGerhard Heift char __user *up = ubuf + *sk_offset; 2444a48b73ecSJosef Bacik /* 2445a48b73ecSJosef Bacik * Copy the item, same behavior as above, but reset the 2446a48b73ecSJosef Bacik * * sk_offset so we copy the full thing again. 2447a48b73ecSJosef Bacik */ 2448a48b73ecSJosef Bacik if (read_extent_buffer_to_user_nofault(leaf, up, 2449ba346b35SGerhard Heift item_off, item_len)) { 2450a48b73ecSJosef Bacik ret = 0; 2451a48b73ecSJosef Bacik *sk_offset -= sizeof(sh); 2452ba346b35SGerhard Heift goto out; 2453ba346b35SGerhard Heift } 2454ba346b35SGerhard Heift 2455ac8e9819SChris Mason *sk_offset += item_len; 2456ac8e9819SChris Mason } 2457e2156867SHugo Mills (*num_found)++; 2458ac8e9819SChris Mason 24598f5f6178SGerhard Heift if (ret) /* -EOVERFLOW from above */ 24608f5f6178SGerhard Heift goto out; 24618f5f6178SGerhard Heift 246225c9bc2eSGerhard Heift if (*num_found >= sk->nr_items) { 246325c9bc2eSGerhard Heift ret = 1; 246425c9bc2eSGerhard Heift goto out; 246525c9bc2eSGerhard Heift } 2466ac8e9819SChris Mason } 2467ac8e9819SChris Mason advance_key: 2468ac8e9819SChris Mason ret = 0; 2469dd81d459SNaohiro Aota test.objectid = sk->max_objectid; 2470dd81d459SNaohiro Aota test.type = sk->max_type; 2471dd81d459SNaohiro Aota test.offset = sk->max_offset; 2472dd81d459SNaohiro Aota if (btrfs_comp_cpu_keys(key, &test) >= 0) 2473dd81d459SNaohiro Aota ret = 1; 2474dd81d459SNaohiro Aota else if (key->offset < (u64)-1) 2475abc6e134SChris Mason key->offset++; 2476dd81d459SNaohiro Aota else if (key->type < (u8)-1) { 2477abc6e134SChris Mason key->offset = 0; 2478abc6e134SChris Mason key->type++; 2479dd81d459SNaohiro Aota } else if (key->objectid < (u64)-1) { 2480abc6e134SChris Mason key->offset = 0; 2481abc6e134SChris Mason key->type = 0; 2482abc6e134SChris Mason key->objectid++; 2483abc6e134SChris Mason } else 2484abc6e134SChris Mason ret = 1; 248525c9bc2eSGerhard Heift out: 2486ba346b35SGerhard Heift /* 2487ba346b35SGerhard Heift * 0: all items from this leaf copied, continue with next 2488ba346b35SGerhard Heift * 1: * more items can be copied, but unused buffer is too small 2489ba346b35SGerhard Heift * * all items were found 2490ba346b35SGerhard Heift * Either way, it will stops the loop which iterates to the next 2491ba346b35SGerhard Heift * leaf 2492ba346b35SGerhard Heift * -EOVERFLOW: item was to large for buffer 2493ba346b35SGerhard Heift * -EFAULT: could not copy extent buffer back to userspace 2494ba346b35SGerhard Heift */ 2495ac8e9819SChris Mason return ret; 2496ac8e9819SChris Mason } 2497ac8e9819SChris Mason 2498ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 249912544442SGerhard Heift struct btrfs_ioctl_search_key *sk, 25009b6e817dSGerhard Heift size_t *buf_size, 2501ba346b35SGerhard Heift char __user *ubuf) 2502ac8e9819SChris Mason { 25030b246afaSJeff Mahoney struct btrfs_fs_info *info = btrfs_sb(inode->i_sb); 2504ac8e9819SChris Mason struct btrfs_root *root; 2505ac8e9819SChris Mason struct btrfs_key key; 2506ac8e9819SChris Mason struct btrfs_path *path; 2507ac8e9819SChris Mason int ret; 2508ac8e9819SChris Mason int num_found = 0; 2509ac8e9819SChris Mason unsigned long sk_offset = 0; 2510ac8e9819SChris Mason 25119b6e817dSGerhard Heift if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) { 25129b6e817dSGerhard Heift *buf_size = sizeof(struct btrfs_ioctl_search_header); 251312544442SGerhard Heift return -EOVERFLOW; 25149b6e817dSGerhard Heift } 251512544442SGerhard Heift 2516ac8e9819SChris Mason path = btrfs_alloc_path(); 2517ac8e9819SChris Mason if (!path) 2518ac8e9819SChris Mason return -ENOMEM; 2519ac8e9819SChris Mason 2520ac8e9819SChris Mason if (sk->tree_id == 0) { 2521ac8e9819SChris Mason /* search the root of the inode that was passed */ 252200246528SJosef Bacik root = btrfs_grab_root(BTRFS_I(inode)->root); 2523ac8e9819SChris Mason } else { 252456e9357aSDavid Sterba root = btrfs_get_fs_root(info, sk->tree_id, true); 2525ac8e9819SChris Mason if (IS_ERR(root)) { 2526ac8e9819SChris Mason btrfs_free_path(path); 2527ad1e3d56SMisono Tomohiro return PTR_ERR(root); 2528ac8e9819SChris Mason } 2529ac8e9819SChris Mason } 2530ac8e9819SChris Mason 2531ac8e9819SChris Mason key.objectid = sk->min_objectid; 2532ac8e9819SChris Mason key.type = sk->min_type; 2533ac8e9819SChris Mason key.offset = sk->min_offset; 2534ac8e9819SChris Mason 2535ac8e9819SChris Mason while (1) { 2536bb523b40SAndreas Gruenbacher ret = -EFAULT; 2537bb523b40SAndreas Gruenbacher if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset)) 2538a48b73ecSJosef Bacik break; 2539a48b73ecSJosef Bacik 25406174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &key, path, sk->min_transid); 2541ac8e9819SChris Mason if (ret != 0) { 2542ac8e9819SChris Mason if (ret > 0) 2543ac8e9819SChris Mason ret = 0; 2544ac8e9819SChris Mason goto err; 2545ac8e9819SChris Mason } 2546df397565SJeff Mahoney ret = copy_to_sk(path, &key, sk, buf_size, ubuf, 2547ac8e9819SChris Mason &sk_offset, &num_found); 2548b3b4aa74SDavid Sterba btrfs_release_path(path); 254925c9bc2eSGerhard Heift if (ret) 2550ac8e9819SChris Mason break; 2551ac8e9819SChris Mason 2552ac8e9819SChris Mason } 25538f5f6178SGerhard Heift if (ret > 0) 2554ac8e9819SChris Mason ret = 0; 2555ac8e9819SChris Mason err: 2556ac8e9819SChris Mason sk->nr_items = num_found; 255700246528SJosef Bacik btrfs_put_root(root); 2558ac8e9819SChris Mason btrfs_free_path(path); 2559ac8e9819SChris Mason return ret; 2560ac8e9819SChris Mason } 2561ac8e9819SChris Mason 25629ad12305SSahil Kang static noinline int btrfs_ioctl_tree_search(struct inode *inode, 2563ac8e9819SChris Mason void __user *argp) 2564ac8e9819SChris Mason { 2565ba346b35SGerhard Heift struct btrfs_ioctl_search_args __user *uargs; 2566ba346b35SGerhard Heift struct btrfs_ioctl_search_key sk; 2567ac8e9819SChris Mason int ret; 25689b6e817dSGerhard Heift size_t buf_size; 2569ac8e9819SChris Mason 2570ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2571ac8e9819SChris Mason return -EPERM; 2572ac8e9819SChris Mason 2573ba346b35SGerhard Heift uargs = (struct btrfs_ioctl_search_args __user *)argp; 2574ac8e9819SChris Mason 2575ba346b35SGerhard Heift if (copy_from_user(&sk, &uargs->key, sizeof(sk))) 2576ba346b35SGerhard Heift return -EFAULT; 2577ba346b35SGerhard Heift 2578ba346b35SGerhard Heift buf_size = sizeof(uargs->buf); 2579ac8e9819SChris Mason 2580ba346b35SGerhard Heift ret = search_ioctl(inode, &sk, &buf_size, uargs->buf); 25818f5f6178SGerhard Heift 25828f5f6178SGerhard Heift /* 25838f5f6178SGerhard Heift * In the origin implementation an overflow is handled by returning a 25848f5f6178SGerhard Heift * search header with a len of zero, so reset ret. 25858f5f6178SGerhard Heift */ 25868f5f6178SGerhard Heift if (ret == -EOVERFLOW) 25878f5f6178SGerhard Heift ret = 0; 25888f5f6178SGerhard Heift 2589ba346b35SGerhard Heift if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk))) 2590ac8e9819SChris Mason ret = -EFAULT; 2591ac8e9819SChris Mason return ret; 2592ac8e9819SChris Mason } 2593ac8e9819SChris Mason 25949ad12305SSahil Kang static noinline int btrfs_ioctl_tree_search_v2(struct inode *inode, 2595cc68a8a5SGerhard Heift void __user *argp) 2596cc68a8a5SGerhard Heift { 2597cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 __user *uarg; 2598cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 args; 2599cc68a8a5SGerhard Heift int ret; 2600cc68a8a5SGerhard Heift size_t buf_size; 2601ee22184bSByongho Lee const size_t buf_limit = SZ_16M; 2602cc68a8a5SGerhard Heift 2603cc68a8a5SGerhard Heift if (!capable(CAP_SYS_ADMIN)) 2604cc68a8a5SGerhard Heift return -EPERM; 2605cc68a8a5SGerhard Heift 2606cc68a8a5SGerhard Heift /* copy search header and buffer size */ 2607cc68a8a5SGerhard Heift uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp; 2608cc68a8a5SGerhard Heift if (copy_from_user(&args, uarg, sizeof(args))) 2609cc68a8a5SGerhard Heift return -EFAULT; 2610cc68a8a5SGerhard Heift 2611cc68a8a5SGerhard Heift buf_size = args.buf_size; 2612cc68a8a5SGerhard Heift 2613cc68a8a5SGerhard Heift /* limit result size to 16MB */ 2614cc68a8a5SGerhard Heift if (buf_size > buf_limit) 2615cc68a8a5SGerhard Heift buf_size = buf_limit; 2616cc68a8a5SGerhard Heift 2617cc68a8a5SGerhard Heift ret = search_ioctl(inode, &args.key, &buf_size, 2618718dc5faSOmar Sandoval (char __user *)(&uarg->buf[0])); 2619cc68a8a5SGerhard Heift if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key))) 2620cc68a8a5SGerhard Heift ret = -EFAULT; 2621cc68a8a5SGerhard Heift else if (ret == -EOVERFLOW && 2622cc68a8a5SGerhard Heift copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size))) 2623cc68a8a5SGerhard Heift ret = -EFAULT; 2624cc68a8a5SGerhard Heift 262576dda93cSYan, Zheng return ret; 262676dda93cSYan, Zheng } 262776dda93cSYan, Zheng 262898d377a0STARUISI Hiroaki /* 2629ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 2630ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 263198d377a0STARUISI Hiroaki */ 263298d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 263398d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 263498d377a0STARUISI Hiroaki { 263598d377a0STARUISI Hiroaki struct btrfs_root *root; 263698d377a0STARUISI Hiroaki struct btrfs_key key; 2637ac8e9819SChris Mason char *ptr; 263898d377a0STARUISI Hiroaki int ret = -1; 263998d377a0STARUISI Hiroaki int slot; 264098d377a0STARUISI Hiroaki int len; 264198d377a0STARUISI Hiroaki int total_len = 0; 264298d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 264398d377a0STARUISI Hiroaki struct extent_buffer *l; 264498d377a0STARUISI Hiroaki struct btrfs_path *path; 264598d377a0STARUISI Hiroaki 264698d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 264798d377a0STARUISI Hiroaki name[0]='\0'; 264898d377a0STARUISI Hiroaki return 0; 264998d377a0STARUISI Hiroaki } 265098d377a0STARUISI Hiroaki 265198d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 265298d377a0STARUISI Hiroaki if (!path) 265398d377a0STARUISI Hiroaki return -ENOMEM; 265498d377a0STARUISI Hiroaki 2655c8bcbfbdSNikolay Borisov ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1]; 265698d377a0STARUISI Hiroaki 265756e9357aSDavid Sterba root = btrfs_get_fs_root(info, tree_id, true); 265898d377a0STARUISI Hiroaki if (IS_ERR(root)) { 2659ad1e3d56SMisono Tomohiro ret = PTR_ERR(root); 266088234012SJosef Bacik root = NULL; 266188234012SJosef Bacik goto out; 266288234012SJosef Bacik } 266398d377a0STARUISI Hiroaki 266498d377a0STARUISI Hiroaki key.objectid = dirid; 266598d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 26668ad6fcabSChris Mason key.offset = (u64)-1; 266798d377a0STARUISI Hiroaki 266898d377a0STARUISI Hiroaki while (1) { 26690ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 267018674c6cSFilipe David Borba Manana if (ret < 0) 267118674c6cSFilipe David Borba Manana goto out; 267218674c6cSFilipe David Borba Manana else if (ret > 0) { 2673ac8e9819SChris Mason ret = -ENOENT; 267498d377a0STARUISI Hiroaki goto out; 2675ac8e9819SChris Mason } 267618674c6cSFilipe David Borba Manana 267718674c6cSFilipe David Borba Manana l = path->nodes[0]; 267818674c6cSFilipe David Borba Manana slot = path->slots[0]; 267998d377a0STARUISI Hiroaki 268098d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 268198d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 268298d377a0STARUISI Hiroaki ptr -= len + 1; 268398d377a0STARUISI Hiroaki total_len += len + 1; 2684a696cf35SFilipe David Borba Manana if (ptr < name) { 2685a696cf35SFilipe David Borba Manana ret = -ENAMETOOLONG; 268698d377a0STARUISI Hiroaki goto out; 2687a696cf35SFilipe David Borba Manana } 268898d377a0STARUISI Hiroaki 268998d377a0STARUISI Hiroaki *(ptr + len) = '/'; 269098d377a0STARUISI Hiroaki read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 269198d377a0STARUISI Hiroaki 269298d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 269398d377a0STARUISI Hiroaki break; 269498d377a0STARUISI Hiroaki 2695b3b4aa74SDavid Sterba btrfs_release_path(path); 269698d377a0STARUISI Hiroaki key.objectid = key.offset; 26978ad6fcabSChris Mason key.offset = (u64)-1; 269898d377a0STARUISI Hiroaki dirid = key.objectid; 269998d377a0STARUISI Hiroaki } 270077906a50SLi Zefan memmove(name, ptr, total_len); 270198d377a0STARUISI Hiroaki name[total_len] = '\0'; 270298d377a0STARUISI Hiroaki ret = 0; 270398d377a0STARUISI Hiroaki out: 270400246528SJosef Bacik btrfs_put_root(root); 270598d377a0STARUISI Hiroaki btrfs_free_path(path); 2706ac8e9819SChris Mason return ret; 2707ac8e9819SChris Mason } 2708ac8e9819SChris Mason 27096623d9a0SChristian Brauner static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns, 27106623d9a0SChristian Brauner struct inode *inode, 271123d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args) 271223d0b79dSTomohiro Misono { 271323d0b79dSTomohiro Misono struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 271423d0b79dSTomohiro Misono struct super_block *sb = inode->i_sb; 271523d0b79dSTomohiro Misono struct btrfs_key upper_limit = BTRFS_I(inode)->location; 271623d0b79dSTomohiro Misono u64 treeid = BTRFS_I(inode)->root->root_key.objectid; 271723d0b79dSTomohiro Misono u64 dirid = args->dirid; 271823d0b79dSTomohiro Misono unsigned long item_off; 271923d0b79dSTomohiro Misono unsigned long item_len; 272023d0b79dSTomohiro Misono struct btrfs_inode_ref *iref; 272123d0b79dSTomohiro Misono struct btrfs_root_ref *rref; 2722b8a49ae1SJosef Bacik struct btrfs_root *root = NULL; 272323d0b79dSTomohiro Misono struct btrfs_path *path; 272423d0b79dSTomohiro Misono struct btrfs_key key, key2; 272523d0b79dSTomohiro Misono struct extent_buffer *leaf; 272623d0b79dSTomohiro Misono struct inode *temp_inode; 272723d0b79dSTomohiro Misono char *ptr; 272823d0b79dSTomohiro Misono int slot; 272923d0b79dSTomohiro Misono int len; 273023d0b79dSTomohiro Misono int total_len = 0; 273123d0b79dSTomohiro Misono int ret; 273223d0b79dSTomohiro Misono 273323d0b79dSTomohiro Misono path = btrfs_alloc_path(); 273423d0b79dSTomohiro Misono if (!path) 273523d0b79dSTomohiro Misono return -ENOMEM; 273623d0b79dSTomohiro Misono 273723d0b79dSTomohiro Misono /* 273823d0b79dSTomohiro Misono * If the bottom subvolume does not exist directly under upper_limit, 273923d0b79dSTomohiro Misono * construct the path in from the bottom up. 274023d0b79dSTomohiro Misono */ 274123d0b79dSTomohiro Misono if (dirid != upper_limit.objectid) { 274223d0b79dSTomohiro Misono ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1]; 274323d0b79dSTomohiro Misono 274456e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, treeid, true); 274523d0b79dSTomohiro Misono if (IS_ERR(root)) { 274623d0b79dSTomohiro Misono ret = PTR_ERR(root); 274723d0b79dSTomohiro Misono goto out; 274823d0b79dSTomohiro Misono } 274923d0b79dSTomohiro Misono 275023d0b79dSTomohiro Misono key.objectid = dirid; 275123d0b79dSTomohiro Misono key.type = BTRFS_INODE_REF_KEY; 275223d0b79dSTomohiro Misono key.offset = (u64)-1; 275323d0b79dSTomohiro Misono while (1) { 27540ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 27550ff40a91SMarcos Paulo de Souza if (ret < 0) 2756b8a49ae1SJosef Bacik goto out_put; 27570ff40a91SMarcos Paulo de Souza else if (ret > 0) { 275823d0b79dSTomohiro Misono ret = -ENOENT; 2759b8a49ae1SJosef Bacik goto out_put; 276023d0b79dSTomohiro Misono } 276123d0b79dSTomohiro Misono 276223d0b79dSTomohiro Misono leaf = path->nodes[0]; 276323d0b79dSTomohiro Misono slot = path->slots[0]; 276423d0b79dSTomohiro Misono 276523d0b79dSTomohiro Misono iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref); 276623d0b79dSTomohiro Misono len = btrfs_inode_ref_name_len(leaf, iref); 276723d0b79dSTomohiro Misono ptr -= len + 1; 276823d0b79dSTomohiro Misono total_len += len + 1; 276923d0b79dSTomohiro Misono if (ptr < args->path) { 277023d0b79dSTomohiro Misono ret = -ENAMETOOLONG; 2771b8a49ae1SJosef Bacik goto out_put; 277223d0b79dSTomohiro Misono } 277323d0b79dSTomohiro Misono 277423d0b79dSTomohiro Misono *(ptr + len) = '/'; 277523d0b79dSTomohiro Misono read_extent_buffer(leaf, ptr, 277623d0b79dSTomohiro Misono (unsigned long)(iref + 1), len); 277723d0b79dSTomohiro Misono 277823d0b79dSTomohiro Misono /* Check the read+exec permission of this directory */ 277923d0b79dSTomohiro Misono ret = btrfs_previous_item(root, path, dirid, 278023d0b79dSTomohiro Misono BTRFS_INODE_ITEM_KEY); 278123d0b79dSTomohiro Misono if (ret < 0) { 2782b8a49ae1SJosef Bacik goto out_put; 278323d0b79dSTomohiro Misono } else if (ret > 0) { 278423d0b79dSTomohiro Misono ret = -ENOENT; 2785b8a49ae1SJosef Bacik goto out_put; 278623d0b79dSTomohiro Misono } 278723d0b79dSTomohiro Misono 278823d0b79dSTomohiro Misono leaf = path->nodes[0]; 278923d0b79dSTomohiro Misono slot = path->slots[0]; 279023d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key2, slot); 279123d0b79dSTomohiro Misono if (key2.objectid != dirid) { 279223d0b79dSTomohiro Misono ret = -ENOENT; 2793b8a49ae1SJosef Bacik goto out_put; 279423d0b79dSTomohiro Misono } 279523d0b79dSTomohiro Misono 27960202e83fSDavid Sterba temp_inode = btrfs_iget(sb, key2.objectid, root); 27973ca57bd6SMisono Tomohiro if (IS_ERR(temp_inode)) { 27983ca57bd6SMisono Tomohiro ret = PTR_ERR(temp_inode); 2799b8a49ae1SJosef Bacik goto out_put; 28003ca57bd6SMisono Tomohiro } 28016623d9a0SChristian Brauner ret = inode_permission(mnt_userns, temp_inode, 280247291baaSChristian Brauner MAY_READ | MAY_EXEC); 280323d0b79dSTomohiro Misono iput(temp_inode); 280423d0b79dSTomohiro Misono if (ret) { 280523d0b79dSTomohiro Misono ret = -EACCES; 2806b8a49ae1SJosef Bacik goto out_put; 280723d0b79dSTomohiro Misono } 280823d0b79dSTomohiro Misono 280923d0b79dSTomohiro Misono if (key.offset == upper_limit.objectid) 281023d0b79dSTomohiro Misono break; 281123d0b79dSTomohiro Misono if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) { 281223d0b79dSTomohiro Misono ret = -EACCES; 2813b8a49ae1SJosef Bacik goto out_put; 281423d0b79dSTomohiro Misono } 281523d0b79dSTomohiro Misono 281623d0b79dSTomohiro Misono btrfs_release_path(path); 281723d0b79dSTomohiro Misono key.objectid = key.offset; 281823d0b79dSTomohiro Misono key.offset = (u64)-1; 281923d0b79dSTomohiro Misono dirid = key.objectid; 282023d0b79dSTomohiro Misono } 282123d0b79dSTomohiro Misono 282223d0b79dSTomohiro Misono memmove(args->path, ptr, total_len); 282323d0b79dSTomohiro Misono args->path[total_len] = '\0'; 282400246528SJosef Bacik btrfs_put_root(root); 2825b8a49ae1SJosef Bacik root = NULL; 282623d0b79dSTomohiro Misono btrfs_release_path(path); 282723d0b79dSTomohiro Misono } 282823d0b79dSTomohiro Misono 282923d0b79dSTomohiro Misono /* Get the bottom subvolume's name from ROOT_REF */ 283023d0b79dSTomohiro Misono key.objectid = treeid; 283123d0b79dSTomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 283223d0b79dSTomohiro Misono key.offset = args->treeid; 2833b8a49ae1SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 283423d0b79dSTomohiro Misono if (ret < 0) { 283523d0b79dSTomohiro Misono goto out; 283623d0b79dSTomohiro Misono } else if (ret > 0) { 283723d0b79dSTomohiro Misono ret = -ENOENT; 283823d0b79dSTomohiro Misono goto out; 283923d0b79dSTomohiro Misono } 284023d0b79dSTomohiro Misono 284123d0b79dSTomohiro Misono leaf = path->nodes[0]; 284223d0b79dSTomohiro Misono slot = path->slots[0]; 284323d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 284423d0b79dSTomohiro Misono 284523d0b79dSTomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot); 28463212fa14SJosef Bacik item_len = btrfs_item_size(leaf, slot); 284723d0b79dSTomohiro Misono /* Check if dirid in ROOT_REF corresponds to passed dirid */ 284823d0b79dSTomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 284923d0b79dSTomohiro Misono if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) { 285023d0b79dSTomohiro Misono ret = -EINVAL; 285123d0b79dSTomohiro Misono goto out; 285223d0b79dSTomohiro Misono } 285323d0b79dSTomohiro Misono 285423d0b79dSTomohiro Misono /* Copy subvolume's name */ 285523d0b79dSTomohiro Misono item_off += sizeof(struct btrfs_root_ref); 285623d0b79dSTomohiro Misono item_len -= sizeof(struct btrfs_root_ref); 285723d0b79dSTomohiro Misono read_extent_buffer(leaf, args->name, item_off, item_len); 285823d0b79dSTomohiro Misono args->name[item_len] = 0; 285923d0b79dSTomohiro Misono 2860b8a49ae1SJosef Bacik out_put: 286100246528SJosef Bacik btrfs_put_root(root); 286223d0b79dSTomohiro Misono out: 286323d0b79dSTomohiro Misono btrfs_free_path(path); 286423d0b79dSTomohiro Misono return ret; 286523d0b79dSTomohiro Misono } 286623d0b79dSTomohiro Misono 2867dc408ccdSSahil Kang static noinline int btrfs_ioctl_ino_lookup(struct btrfs_root *root, 2868ac8e9819SChris Mason void __user *argp) 2869ac8e9819SChris Mason { 2870ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 287101b810b8SDavid Sterba int ret = 0; 2872ac8e9819SChris Mason 28732354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 28742354d08fSJulia Lawall if (IS_ERR(args)) 28752354d08fSJulia Lawall return PTR_ERR(args); 2876c2b96929SDan Carpenter 287701b810b8SDavid Sterba /* 287801b810b8SDavid Sterba * Unprivileged query to obtain the containing subvolume root id. The 287901b810b8SDavid Sterba * path is reset so it's consistent with btrfs_search_path_in_tree. 288001b810b8SDavid Sterba */ 28811b53ac4dSChris Mason if (args->treeid == 0) 2882dc408ccdSSahil Kang args->treeid = root->root_key.objectid; 28831b53ac4dSChris Mason 288401b810b8SDavid Sterba if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) { 288501b810b8SDavid Sterba args->name[0] = 0; 288601b810b8SDavid Sterba goto out; 288701b810b8SDavid Sterba } 288801b810b8SDavid Sterba 288901b810b8SDavid Sterba if (!capable(CAP_SYS_ADMIN)) { 289001b810b8SDavid Sterba ret = -EPERM; 289101b810b8SDavid Sterba goto out; 289201b810b8SDavid Sterba } 289301b810b8SDavid Sterba 2894dc408ccdSSahil Kang ret = btrfs_search_path_in_tree(root->fs_info, 2895ac8e9819SChris Mason args->treeid, args->objectid, 2896ac8e9819SChris Mason args->name); 2897ac8e9819SChris Mason 289801b810b8SDavid Sterba out: 2899ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2900ac8e9819SChris Mason ret = -EFAULT; 2901ac8e9819SChris Mason 2902ac8e9819SChris Mason kfree(args); 290398d377a0STARUISI Hiroaki return ret; 290498d377a0STARUISI Hiroaki } 290598d377a0STARUISI Hiroaki 290623d0b79dSTomohiro Misono /* 290723d0b79dSTomohiro Misono * Version of ino_lookup ioctl (unprivileged) 290823d0b79dSTomohiro Misono * 290923d0b79dSTomohiro Misono * The main differences from ino_lookup ioctl are: 291023d0b79dSTomohiro Misono * 291123d0b79dSTomohiro Misono * 1. Read + Exec permission will be checked using inode_permission() during 291223d0b79dSTomohiro Misono * path construction. -EACCES will be returned in case of failure. 291323d0b79dSTomohiro Misono * 2. Path construction will be stopped at the inode number which corresponds 291423d0b79dSTomohiro Misono * to the fd with which this ioctl is called. If constructed path does not 291523d0b79dSTomohiro Misono * exist under fd's inode, -EACCES will be returned. 291623d0b79dSTomohiro Misono * 3. The name of bottom subvolume is also searched and filled. 291723d0b79dSTomohiro Misono */ 291823d0b79dSTomohiro Misono static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp) 291923d0b79dSTomohiro Misono { 292023d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args; 292123d0b79dSTomohiro Misono struct inode *inode; 292223d0b79dSTomohiro Misono int ret; 292323d0b79dSTomohiro Misono 292423d0b79dSTomohiro Misono args = memdup_user(argp, sizeof(*args)); 292523d0b79dSTomohiro Misono if (IS_ERR(args)) 292623d0b79dSTomohiro Misono return PTR_ERR(args); 292723d0b79dSTomohiro Misono 292823d0b79dSTomohiro Misono inode = file_inode(file); 292923d0b79dSTomohiro Misono 293023d0b79dSTomohiro Misono if (args->dirid == BTRFS_FIRST_FREE_OBJECTID && 293123d0b79dSTomohiro Misono BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) { 293223d0b79dSTomohiro Misono /* 293323d0b79dSTomohiro Misono * The subvolume does not exist under fd with which this is 293423d0b79dSTomohiro Misono * called 293523d0b79dSTomohiro Misono */ 293623d0b79dSTomohiro Misono kfree(args); 293723d0b79dSTomohiro Misono return -EACCES; 293823d0b79dSTomohiro Misono } 293923d0b79dSTomohiro Misono 29406623d9a0SChristian Brauner ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args); 294123d0b79dSTomohiro Misono 294223d0b79dSTomohiro Misono if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 294323d0b79dSTomohiro Misono ret = -EFAULT; 294423d0b79dSTomohiro Misono 294523d0b79dSTomohiro Misono kfree(args); 294623d0b79dSTomohiro Misono return ret; 294723d0b79dSTomohiro Misono } 294823d0b79dSTomohiro Misono 2949b64ec075STomohiro Misono /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */ 29509ad12305SSahil Kang static int btrfs_ioctl_get_subvol_info(struct inode *inode, void __user *argp) 2951b64ec075STomohiro Misono { 2952b64ec075STomohiro Misono struct btrfs_ioctl_get_subvol_info_args *subvol_info; 2953b64ec075STomohiro Misono struct btrfs_fs_info *fs_info; 2954b64ec075STomohiro Misono struct btrfs_root *root; 2955b64ec075STomohiro Misono struct btrfs_path *path; 2956b64ec075STomohiro Misono struct btrfs_key key; 2957b64ec075STomohiro Misono struct btrfs_root_item *root_item; 2958b64ec075STomohiro Misono struct btrfs_root_ref *rref; 2959b64ec075STomohiro Misono struct extent_buffer *leaf; 2960b64ec075STomohiro Misono unsigned long item_off; 2961b64ec075STomohiro Misono unsigned long item_len; 2962b64ec075STomohiro Misono int slot; 2963b64ec075STomohiro Misono int ret = 0; 2964b64ec075STomohiro Misono 2965b64ec075STomohiro Misono path = btrfs_alloc_path(); 2966b64ec075STomohiro Misono if (!path) 2967b64ec075STomohiro Misono return -ENOMEM; 2968b64ec075STomohiro Misono 2969b64ec075STomohiro Misono subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL); 2970b64ec075STomohiro Misono if (!subvol_info) { 2971b64ec075STomohiro Misono btrfs_free_path(path); 2972b64ec075STomohiro Misono return -ENOMEM; 2973b64ec075STomohiro Misono } 2974b64ec075STomohiro Misono 2975b64ec075STomohiro Misono fs_info = BTRFS_I(inode)->root->fs_info; 2976b64ec075STomohiro Misono 2977b64ec075STomohiro Misono /* Get root_item of inode's subvolume */ 2978b64ec075STomohiro Misono key.objectid = BTRFS_I(inode)->root->root_key.objectid; 297956e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, key.objectid, true); 2980b64ec075STomohiro Misono if (IS_ERR(root)) { 2981b64ec075STomohiro Misono ret = PTR_ERR(root); 298204734e84SJosef Bacik goto out_free; 298304734e84SJosef Bacik } 2984b64ec075STomohiro Misono root_item = &root->root_item; 2985b64ec075STomohiro Misono 2986b64ec075STomohiro Misono subvol_info->treeid = key.objectid; 2987b64ec075STomohiro Misono 2988b64ec075STomohiro Misono subvol_info->generation = btrfs_root_generation(root_item); 2989b64ec075STomohiro Misono subvol_info->flags = btrfs_root_flags(root_item); 2990b64ec075STomohiro Misono 2991b64ec075STomohiro Misono memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE); 2992b64ec075STomohiro Misono memcpy(subvol_info->parent_uuid, root_item->parent_uuid, 2993b64ec075STomohiro Misono BTRFS_UUID_SIZE); 2994b64ec075STomohiro Misono memcpy(subvol_info->received_uuid, root_item->received_uuid, 2995b64ec075STomohiro Misono BTRFS_UUID_SIZE); 2996b64ec075STomohiro Misono 2997b64ec075STomohiro Misono subvol_info->ctransid = btrfs_root_ctransid(root_item); 2998b64ec075STomohiro Misono subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime); 2999b64ec075STomohiro Misono subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime); 3000b64ec075STomohiro Misono 3001b64ec075STomohiro Misono subvol_info->otransid = btrfs_root_otransid(root_item); 3002b64ec075STomohiro Misono subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime); 3003b64ec075STomohiro Misono subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime); 3004b64ec075STomohiro Misono 3005b64ec075STomohiro Misono subvol_info->stransid = btrfs_root_stransid(root_item); 3006b64ec075STomohiro Misono subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime); 3007b64ec075STomohiro Misono subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime); 3008b64ec075STomohiro Misono 3009b64ec075STomohiro Misono subvol_info->rtransid = btrfs_root_rtransid(root_item); 3010b64ec075STomohiro Misono subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime); 3011b64ec075STomohiro Misono subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime); 3012b64ec075STomohiro Misono 3013b64ec075STomohiro Misono if (key.objectid != BTRFS_FS_TREE_OBJECTID) { 3014b64ec075STomohiro Misono /* Search root tree for ROOT_BACKREF of this subvolume */ 3015b64ec075STomohiro Misono key.type = BTRFS_ROOT_BACKREF_KEY; 3016b64ec075STomohiro Misono key.offset = 0; 301704734e84SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 3018b64ec075STomohiro Misono if (ret < 0) { 3019b64ec075STomohiro Misono goto out; 3020b64ec075STomohiro Misono } else if (path->slots[0] >= 3021b64ec075STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 302204734e84SJosef Bacik ret = btrfs_next_leaf(fs_info->tree_root, path); 3023b64ec075STomohiro Misono if (ret < 0) { 3024b64ec075STomohiro Misono goto out; 3025b64ec075STomohiro Misono } else if (ret > 0) { 3026b64ec075STomohiro Misono ret = -EUCLEAN; 3027b64ec075STomohiro Misono goto out; 3028b64ec075STomohiro Misono } 3029b64ec075STomohiro Misono } 3030b64ec075STomohiro Misono 3031b64ec075STomohiro Misono leaf = path->nodes[0]; 3032b64ec075STomohiro Misono slot = path->slots[0]; 3033b64ec075STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 3034b64ec075STomohiro Misono if (key.objectid == subvol_info->treeid && 3035b64ec075STomohiro Misono key.type == BTRFS_ROOT_BACKREF_KEY) { 3036b64ec075STomohiro Misono subvol_info->parent_id = key.offset; 3037b64ec075STomohiro Misono 3038b64ec075STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 3039b64ec075STomohiro Misono subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref); 3040b64ec075STomohiro Misono 3041b64ec075STomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot) 3042b64ec075STomohiro Misono + sizeof(struct btrfs_root_ref); 30433212fa14SJosef Bacik item_len = btrfs_item_size(leaf, slot) 3044b64ec075STomohiro Misono - sizeof(struct btrfs_root_ref); 3045b64ec075STomohiro Misono read_extent_buffer(leaf, subvol_info->name, 3046b64ec075STomohiro Misono item_off, item_len); 3047b64ec075STomohiro Misono } else { 3048b64ec075STomohiro Misono ret = -ENOENT; 3049b64ec075STomohiro Misono goto out; 3050b64ec075STomohiro Misono } 3051b64ec075STomohiro Misono } 3052b64ec075STomohiro Misono 3053b64ec075STomohiro Misono if (copy_to_user(argp, subvol_info, sizeof(*subvol_info))) 3054b64ec075STomohiro Misono ret = -EFAULT; 3055b64ec075STomohiro Misono 3056b64ec075STomohiro Misono out: 305700246528SJosef Bacik btrfs_put_root(root); 305804734e84SJosef Bacik out_free: 3059b64ec075STomohiro Misono btrfs_free_path(path); 3060b091f7feSWaiman Long kfree(subvol_info); 3061b64ec075STomohiro Misono return ret; 3062b64ec075STomohiro Misono } 3063b64ec075STomohiro Misono 306442e4b520STomohiro Misono /* 306542e4b520STomohiro Misono * Return ROOT_REF information of the subvolume containing this inode 306642e4b520STomohiro Misono * except the subvolume name. 306742e4b520STomohiro Misono */ 30689ad12305SSahil Kang static int btrfs_ioctl_get_subvol_rootref(struct btrfs_root *root, 3069dc408ccdSSahil Kang void __user *argp) 307042e4b520STomohiro Misono { 307142e4b520STomohiro Misono struct btrfs_ioctl_get_subvol_rootref_args *rootrefs; 307242e4b520STomohiro Misono struct btrfs_root_ref *rref; 307342e4b520STomohiro Misono struct btrfs_path *path; 307442e4b520STomohiro Misono struct btrfs_key key; 307542e4b520STomohiro Misono struct extent_buffer *leaf; 307642e4b520STomohiro Misono u64 objectid; 307742e4b520STomohiro Misono int slot; 307842e4b520STomohiro Misono int ret; 307942e4b520STomohiro Misono u8 found; 308042e4b520STomohiro Misono 308142e4b520STomohiro Misono path = btrfs_alloc_path(); 308242e4b520STomohiro Misono if (!path) 308342e4b520STomohiro Misono return -ENOMEM; 308442e4b520STomohiro Misono 308542e4b520STomohiro Misono rootrefs = memdup_user(argp, sizeof(*rootrefs)); 308642e4b520STomohiro Misono if (IS_ERR(rootrefs)) { 308742e4b520STomohiro Misono btrfs_free_path(path); 308842e4b520STomohiro Misono return PTR_ERR(rootrefs); 308942e4b520STomohiro Misono } 309042e4b520STomohiro Misono 30919ad12305SSahil Kang objectid = root->root_key.objectid; 309242e4b520STomohiro Misono key.objectid = objectid; 309342e4b520STomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 309442e4b520STomohiro Misono key.offset = rootrefs->min_treeid; 309542e4b520STomohiro Misono found = 0; 309642e4b520STomohiro Misono 30979ad12305SSahil Kang root = root->fs_info->tree_root; 309842e4b520STomohiro Misono ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 309942e4b520STomohiro Misono if (ret < 0) { 310042e4b520STomohiro Misono goto out; 310142e4b520STomohiro Misono } else if (path->slots[0] >= 310242e4b520STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 310342e4b520STomohiro Misono ret = btrfs_next_leaf(root, path); 310442e4b520STomohiro Misono if (ret < 0) { 310542e4b520STomohiro Misono goto out; 310642e4b520STomohiro Misono } else if (ret > 0) { 310742e4b520STomohiro Misono ret = -EUCLEAN; 310842e4b520STomohiro Misono goto out; 310942e4b520STomohiro Misono } 311042e4b520STomohiro Misono } 311142e4b520STomohiro Misono while (1) { 311242e4b520STomohiro Misono leaf = path->nodes[0]; 311342e4b520STomohiro Misono slot = path->slots[0]; 311442e4b520STomohiro Misono 311542e4b520STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 311642e4b520STomohiro Misono if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) { 311742e4b520STomohiro Misono ret = 0; 311842e4b520STomohiro Misono goto out; 311942e4b520STomohiro Misono } 312042e4b520STomohiro Misono 312142e4b520STomohiro Misono if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) { 312242e4b520STomohiro Misono ret = -EOVERFLOW; 312342e4b520STomohiro Misono goto out; 312442e4b520STomohiro Misono } 312542e4b520STomohiro Misono 312642e4b520STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 312742e4b520STomohiro Misono rootrefs->rootref[found].treeid = key.offset; 312842e4b520STomohiro Misono rootrefs->rootref[found].dirid = 312942e4b520STomohiro Misono btrfs_root_ref_dirid(leaf, rref); 313042e4b520STomohiro Misono found++; 313142e4b520STomohiro Misono 313242e4b520STomohiro Misono ret = btrfs_next_item(root, path); 313342e4b520STomohiro Misono if (ret < 0) { 313442e4b520STomohiro Misono goto out; 313542e4b520STomohiro Misono } else if (ret > 0) { 313642e4b520STomohiro Misono ret = -EUCLEAN; 313742e4b520STomohiro Misono goto out; 313842e4b520STomohiro Misono } 313942e4b520STomohiro Misono } 314042e4b520STomohiro Misono 314142e4b520STomohiro Misono out: 314242e4b520STomohiro Misono if (!ret || ret == -EOVERFLOW) { 314342e4b520STomohiro Misono rootrefs->num_items = found; 314442e4b520STomohiro Misono /* update min_treeid for next search */ 314542e4b520STomohiro Misono if (found) 314642e4b520STomohiro Misono rootrefs->min_treeid = 314742e4b520STomohiro Misono rootrefs->rootref[found - 1].treeid + 1; 314842e4b520STomohiro Misono if (copy_to_user(argp, rootrefs, sizeof(*rootrefs))) 314942e4b520STomohiro Misono ret = -EFAULT; 315042e4b520STomohiro Misono } 315142e4b520STomohiro Misono 315242e4b520STomohiro Misono kfree(rootrefs); 315342e4b520STomohiro Misono btrfs_free_path(path); 315442e4b520STomohiro Misono 315542e4b520STomohiro Misono return ret; 315642e4b520STomohiro Misono } 315742e4b520STomohiro Misono 315876dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 3159949964c9SMarcos Paulo de Souza void __user *arg, 3160949964c9SMarcos Paulo de Souza bool destroy_v2) 316176dda93cSYan, Zheng { 316254563d41SAl Viro struct dentry *parent = file->f_path.dentry; 31630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb); 316476dda93cSYan, Zheng struct dentry *dentry; 31652b0143b5SDavid Howells struct inode *dir = d_inode(parent); 316676dda93cSYan, Zheng struct inode *inode; 316776dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 316876dda93cSYan, Zheng struct btrfs_root *dest = NULL; 3169949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args *vol_args = NULL; 3170949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL; 3171c4ed533bSChristian Brauner struct user_namespace *mnt_userns = file_mnt_user_ns(file); 3172949964c9SMarcos Paulo de Souza char *subvol_name, *subvol_name_ptr = NULL; 3173949964c9SMarcos Paulo de Souza int subvol_namelen; 317476dda93cSYan, Zheng int err = 0; 3175949964c9SMarcos Paulo de Souza bool destroy_parent = false; 317676dda93cSYan, Zheng 3177949964c9SMarcos Paulo de Souza if (destroy_v2) { 3178949964c9SMarcos Paulo de Souza vol_args2 = memdup_user(arg, sizeof(*vol_args2)); 3179949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args2)) 3180949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args2); 3181325c50e3SJeff Mahoney 3182949964c9SMarcos Paulo de Souza if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) { 3183949964c9SMarcos Paulo de Souza err = -EOPNOTSUPP; 3184949964c9SMarcos Paulo de Souza goto out; 3185949964c9SMarcos Paulo de Souza } 318676dda93cSYan, Zheng 3187949964c9SMarcos Paulo de Souza /* 3188949964c9SMarcos Paulo de Souza * If SPEC_BY_ID is not set, we are looking for the subvolume by 3189949964c9SMarcos Paulo de Souza * name, same as v1 currently does. 3190949964c9SMarcos Paulo de Souza */ 3191949964c9SMarcos Paulo de Souza if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) { 3192949964c9SMarcos Paulo de Souza vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0; 3193949964c9SMarcos Paulo de Souza subvol_name = vol_args2->name; 3194949964c9SMarcos Paulo de Souza 3195949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 3196949964c9SMarcos Paulo de Souza if (err) 3197949964c9SMarcos Paulo de Souza goto out; 3198949964c9SMarcos Paulo de Souza } else { 3199aabb34e7SChristian Brauner struct inode *old_dir; 3200c4ed533bSChristian Brauner 3201949964c9SMarcos Paulo de Souza if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) { 320276dda93cSYan, Zheng err = -EINVAL; 320376dda93cSYan, Zheng goto out; 320476dda93cSYan, Zheng } 320576dda93cSYan, Zheng 3206a561be71SAl Viro err = mnt_want_write_file(file); 320776dda93cSYan, Zheng if (err) 320876dda93cSYan, Zheng goto out; 320976dda93cSYan, Zheng 3210949964c9SMarcos Paulo de Souza dentry = btrfs_get_dentry(fs_info->sb, 3211949964c9SMarcos Paulo de Souza BTRFS_FIRST_FREE_OBJECTID, 3212949964c9SMarcos Paulo de Souza vol_args2->subvolid, 0, 0); 3213949964c9SMarcos Paulo de Souza if (IS_ERR(dentry)) { 3214949964c9SMarcos Paulo de Souza err = PTR_ERR(dentry); 3215949964c9SMarcos Paulo de Souza goto out_drop_write; 3216949964c9SMarcos Paulo de Souza } 3217949964c9SMarcos Paulo de Souza 3218949964c9SMarcos Paulo de Souza /* 3219949964c9SMarcos Paulo de Souza * Change the default parent since the subvolume being 3220949964c9SMarcos Paulo de Souza * deleted can be outside of the current mount point. 3221949964c9SMarcos Paulo de Souza */ 3222949964c9SMarcos Paulo de Souza parent = btrfs_get_parent(dentry); 3223949964c9SMarcos Paulo de Souza 3224949964c9SMarcos Paulo de Souza /* 3225949964c9SMarcos Paulo de Souza * At this point dentry->d_name can point to '/' if the 3226949964c9SMarcos Paulo de Souza * subvolume we want to destroy is outsite of the 3227949964c9SMarcos Paulo de Souza * current mount point, so we need to release the 3228949964c9SMarcos Paulo de Souza * current dentry and execute the lookup to return a new 3229949964c9SMarcos Paulo de Souza * one with ->d_name pointing to the 3230949964c9SMarcos Paulo de Souza * <mount point>/subvol_name. 3231949964c9SMarcos Paulo de Souza */ 3232949964c9SMarcos Paulo de Souza dput(dentry); 3233949964c9SMarcos Paulo de Souza if (IS_ERR(parent)) { 3234949964c9SMarcos Paulo de Souza err = PTR_ERR(parent); 3235949964c9SMarcos Paulo de Souza goto out_drop_write; 3236949964c9SMarcos Paulo de Souza } 3237aabb34e7SChristian Brauner old_dir = dir; 3238949964c9SMarcos Paulo de Souza dir = d_inode(parent); 3239949964c9SMarcos Paulo de Souza 3240949964c9SMarcos Paulo de Souza /* 3241949964c9SMarcos Paulo de Souza * If v2 was used with SPEC_BY_ID, a new parent was 3242949964c9SMarcos Paulo de Souza * allocated since the subvolume can be outside of the 3243949964c9SMarcos Paulo de Souza * current mount point. Later on we need to release this 3244949964c9SMarcos Paulo de Souza * new parent dentry. 3245949964c9SMarcos Paulo de Souza */ 3246949964c9SMarcos Paulo de Souza destroy_parent = true; 3247949964c9SMarcos Paulo de Souza 3248aabb34e7SChristian Brauner /* 3249aabb34e7SChristian Brauner * On idmapped mounts, deletion via subvolid is 3250aabb34e7SChristian Brauner * restricted to subvolumes that are immediate 3251aabb34e7SChristian Brauner * ancestors of the inode referenced by the file 3252aabb34e7SChristian Brauner * descriptor in the ioctl. Otherwise the idmapping 3253aabb34e7SChristian Brauner * could potentially be abused to delete subvolumes 3254aabb34e7SChristian Brauner * anywhere in the filesystem the user wouldn't be able 3255aabb34e7SChristian Brauner * to delete without an idmapped mount. 3256aabb34e7SChristian Brauner */ 3257aabb34e7SChristian Brauner if (old_dir != dir && mnt_userns != &init_user_ns) { 3258aabb34e7SChristian Brauner err = -EOPNOTSUPP; 3259aabb34e7SChristian Brauner goto free_parent; 3260aabb34e7SChristian Brauner } 3261aabb34e7SChristian Brauner 3262949964c9SMarcos Paulo de Souza subvol_name_ptr = btrfs_get_subvol_name_from_objectid( 3263949964c9SMarcos Paulo de Souza fs_info, vol_args2->subvolid); 3264949964c9SMarcos Paulo de Souza if (IS_ERR(subvol_name_ptr)) { 3265949964c9SMarcos Paulo de Souza err = PTR_ERR(subvol_name_ptr); 3266949964c9SMarcos Paulo de Souza goto free_parent; 3267949964c9SMarcos Paulo de Souza } 32681a9fd417SDavid Sterba /* subvol_name_ptr is already nul terminated */ 3269949964c9SMarcos Paulo de Souza subvol_name = (char *)kbasename(subvol_name_ptr); 3270949964c9SMarcos Paulo de Souza } 3271949964c9SMarcos Paulo de Souza } else { 3272949964c9SMarcos Paulo de Souza vol_args = memdup_user(arg, sizeof(*vol_args)); 3273949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args)) 3274949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args); 3275949964c9SMarcos Paulo de Souza 3276949964c9SMarcos Paulo de Souza vol_args->name[BTRFS_PATH_NAME_MAX] = 0; 3277949964c9SMarcos Paulo de Souza subvol_name = vol_args->name; 3278949964c9SMarcos Paulo de Souza 3279949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 3280949964c9SMarcos Paulo de Souza if (err) 3281949964c9SMarcos Paulo de Souza goto out; 3282949964c9SMarcos Paulo de Souza } 3283949964c9SMarcos Paulo de Souza 3284949964c9SMarcos Paulo de Souza subvol_namelen = strlen(subvol_name); 3285949964c9SMarcos Paulo de Souza 3286949964c9SMarcos Paulo de Souza if (strchr(subvol_name, '/') || 3287949964c9SMarcos Paulo de Souza strncmp(subvol_name, "..", subvol_namelen) == 0) { 3288949964c9SMarcos Paulo de Souza err = -EINVAL; 3289949964c9SMarcos Paulo de Souza goto free_subvol_name; 3290949964c9SMarcos Paulo de Souza } 3291949964c9SMarcos Paulo de Souza 3292949964c9SMarcos Paulo de Souza if (!S_ISDIR(dir->i_mode)) { 3293949964c9SMarcos Paulo de Souza err = -ENOTDIR; 3294949964c9SMarcos Paulo de Souza goto free_subvol_name; 3295949964c9SMarcos Paulo de Souza } 3296521e0546SDavid Sterba 329700235411SAl Viro err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 329800235411SAl Viro if (err == -EINTR) 3299949964c9SMarcos Paulo de Souza goto free_subvol_name; 3300c4ed533bSChristian Brauner dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen); 330176dda93cSYan, Zheng if (IS_ERR(dentry)) { 330276dda93cSYan, Zheng err = PTR_ERR(dentry); 330376dda93cSYan, Zheng goto out_unlock_dir; 330476dda93cSYan, Zheng } 330576dda93cSYan, Zheng 33062b0143b5SDavid Howells if (d_really_is_negative(dentry)) { 330776dda93cSYan, Zheng err = -ENOENT; 330876dda93cSYan, Zheng goto out_dput; 330976dda93cSYan, Zheng } 331076dda93cSYan, Zheng 33112b0143b5SDavid Howells inode = d_inode(dentry); 33124260f7c7SSage Weil dest = BTRFS_I(inode)->root; 33134260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)) { 33144260f7c7SSage Weil /* 33154260f7c7SSage Weil * Regular user. Only allow this with a special mount 33164260f7c7SSage Weil * option, when the user has write+exec access to the 33174260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 33184260f7c7SSage Weil * allowed. 33194260f7c7SSage Weil * 33204260f7c7SSage Weil * Note that this is _not_ check that the subvol is 33214260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 33224260f7c7SSage Weil * otherwise be able to delete. 33234260f7c7SSage Weil * 33244260f7c7SSage Weil * Users who want to delete empty subvols should try 33254260f7c7SSage Weil * rmdir(2). 33264260f7c7SSage Weil */ 33274260f7c7SSage Weil err = -EPERM; 33280b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED)) 33294260f7c7SSage Weil goto out_dput; 33304260f7c7SSage Weil 33314260f7c7SSage Weil /* 33324260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 33334260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 33344260f7c7SSage Weil * must be called on the dentry referencing the root 33354260f7c7SSage Weil * of the subvol, not a random directory contained 33364260f7c7SSage Weil * within it. 33374260f7c7SSage Weil */ 33384260f7c7SSage Weil err = -EINVAL; 33394260f7c7SSage Weil if (root == dest) 33404260f7c7SSage Weil goto out_dput; 33414260f7c7SSage Weil 3342c4ed533bSChristian Brauner err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC); 33434260f7c7SSage Weil if (err) 33444260f7c7SSage Weil goto out_dput; 33455c39da5bSMiao Xie } 33464260f7c7SSage Weil 33475c39da5bSMiao Xie /* check if subvolume may be deleted by a user */ 3348c4ed533bSChristian Brauner err = btrfs_may_delete(mnt_userns, dir, dentry, 1); 33494260f7c7SSage Weil if (err) 33504260f7c7SSage Weil goto out_dput; 33514260f7c7SSage Weil 33524a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 335376dda93cSYan, Zheng err = -EINVAL; 335476dda93cSYan, Zheng goto out_dput; 335576dda93cSYan, Zheng } 335676dda93cSYan, Zheng 335764708539SJosef Bacik btrfs_inode_lock(inode, 0); 3358f60a2364SMisono Tomohiro err = btrfs_delete_subvolume(dir, dentry); 335964708539SJosef Bacik btrfs_inode_unlock(inode, 0); 3360a37d9a17SAmir Goldstein if (!err) 3361a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 3362fa6ac876SLiu Bo 336376dda93cSYan, Zheng out_dput: 336476dda93cSYan, Zheng dput(dentry); 336576dda93cSYan, Zheng out_unlock_dir: 336664708539SJosef Bacik btrfs_inode_unlock(dir, 0); 3367949964c9SMarcos Paulo de Souza free_subvol_name: 3368949964c9SMarcos Paulo de Souza kfree(subvol_name_ptr); 3369949964c9SMarcos Paulo de Souza free_parent: 3370949964c9SMarcos Paulo de Souza if (destroy_parent) 3371949964c9SMarcos Paulo de Souza dput(parent); 337200235411SAl Viro out_drop_write: 33732a79f17eSAl Viro mnt_drop_write_file(file); 337476dda93cSYan, Zheng out: 3375949964c9SMarcos Paulo de Souza kfree(vol_args2); 337676dda93cSYan, Zheng kfree(vol_args); 337776dda93cSYan, Zheng return err; 337876dda93cSYan, Zheng } 337976dda93cSYan, Zheng 33801e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 3381f46b5a66SChristoph Hellwig { 3382496ad9aaSAl Viro struct inode *inode = file_inode(file); 3383f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3384c853a578SGoldwyn Rodrigues struct btrfs_ioctl_defrag_range_args range = {0}; 3385c146afadSYan Zheng int ret; 3386c146afadSYan Zheng 338725122d15SIlya Dryomov ret = mnt_want_write_file(file); 338825122d15SIlya Dryomov if (ret) 338925122d15SIlya Dryomov return ret; 3390b83cc969SLi Zefan 339125122d15SIlya Dryomov if (btrfs_root_readonly(root)) { 339225122d15SIlya Dryomov ret = -EROFS; 339325122d15SIlya Dryomov goto out; 33945ac00addSStefan Behrens } 3395f46b5a66SChristoph Hellwig 3396f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3397f46b5a66SChristoph Hellwig case S_IFDIR: 3398e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 3399e441d54dSChris Mason ret = -EPERM; 3400e441d54dSChris Mason goto out; 3401e441d54dSChris Mason } 3402de78b51aSEric Sandeen ret = btrfs_defrag_root(root); 3403f46b5a66SChristoph Hellwig break; 3404f46b5a66SChristoph Hellwig case S_IFREG: 3405616d374eSAdam Borowski /* 3406616d374eSAdam Borowski * Note that this does not check the file descriptor for write 3407616d374eSAdam Borowski * access. This prevents defragmenting executables that are 3408616d374eSAdam Borowski * running and allows defrag on files open in read-only mode. 3409616d374eSAdam Borowski */ 3410616d374eSAdam Borowski if (!capable(CAP_SYS_ADMIN) && 341147291baaSChristian Brauner inode_permission(&init_user_ns, inode, MAY_WRITE)) { 3412616d374eSAdam Borowski ret = -EPERM; 3413e441d54dSChris Mason goto out; 3414e441d54dSChris Mason } 34151e701a32SChris Mason 34161e701a32SChris Mason if (argp) { 3417c853a578SGoldwyn Rodrigues if (copy_from_user(&range, argp, sizeof(range))) { 34181e701a32SChris Mason ret = -EFAULT; 3419683be16eSDan Carpenter goto out; 34201e701a32SChris Mason } 34211e701a32SChris Mason /* compression requires us to start the IO */ 3422c853a578SGoldwyn Rodrigues if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 3423c853a578SGoldwyn Rodrigues range.flags |= BTRFS_DEFRAG_RANGE_START_IO; 3424c853a578SGoldwyn Rodrigues range.extent_thresh = (u32)-1; 34251e701a32SChris Mason } 34261e701a32SChris Mason } else { 34271e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 3428c853a578SGoldwyn Rodrigues range.len = (u64)-1; 34291e701a32SChris Mason } 34301ccc2e8aSQu Wenruo ret = btrfs_defrag_file(file_inode(file), &file->f_ra, 3431c853a578SGoldwyn Rodrigues &range, BTRFS_OLDEST_GENERATION, 0); 34324cb5300bSChris Mason if (ret > 0) 34334cb5300bSChris Mason ret = 0; 3434f46b5a66SChristoph Hellwig break; 34358929ecfaSYan, Zheng default: 34368929ecfaSYan, Zheng ret = -EINVAL; 3437f46b5a66SChristoph Hellwig } 3438e441d54dSChris Mason out: 343925122d15SIlya Dryomov mnt_drop_write_file(file); 3440e441d54dSChris Mason return ret; 3441f46b5a66SChristoph Hellwig } 3442f46b5a66SChristoph Hellwig 34432ff7e61eSJeff Mahoney static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg) 3444f46b5a66SChristoph Hellwig { 3445f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 3446a174c0a2SNikolay Borisov bool restore_op = false; 3447f46b5a66SChristoph Hellwig int ret; 3448f46b5a66SChristoph Hellwig 3449e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3450e441d54dSChris Mason return -EPERM; 3451e441d54dSChris Mason 3452914a519bSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3453914a519bSJosef Bacik btrfs_err(fs_info, "device add not supported on extent tree v2 yet"); 3454914a519bSJosef Bacik return -EINVAL; 3455914a519bSJosef Bacik } 3456914a519bSJosef Bacik 3457a174c0a2SNikolay Borisov if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) { 3458a174c0a2SNikolay Borisov if (!btrfs_exclop_start_try_lock(fs_info, BTRFS_EXCLOP_DEV_ADD)) 3459e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3460c9e9f97bSIlya Dryomov 3461a174c0a2SNikolay Borisov /* 3462a174c0a2SNikolay Borisov * We can do the device add because we have a paused balanced, 3463a174c0a2SNikolay Borisov * change the exclusive op type and remember we should bring 3464a174c0a2SNikolay Borisov * back the paused balance 3465a174c0a2SNikolay Borisov */ 3466a174c0a2SNikolay Borisov fs_info->exclusive_operation = BTRFS_EXCLOP_DEV_ADD; 3467a174c0a2SNikolay Borisov btrfs_exclop_start_unlock(fs_info); 3468a174c0a2SNikolay Borisov restore_op = true; 3469a174c0a2SNikolay Borisov } 3470a174c0a2SNikolay Borisov 3471dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3472c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 3473c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 3474c9e9f97bSIlya Dryomov goto out; 3475c9e9f97bSIlya Dryomov } 3476f46b5a66SChristoph Hellwig 34775516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 34782ff7e61eSJeff Mahoney ret = btrfs_init_new_device(fs_info, vol_args->name); 3479f46b5a66SChristoph Hellwig 348043d20761SAnand Jain if (!ret) 34810b246afaSJeff Mahoney btrfs_info(fs_info, "disk added %s", vol_args->name); 348243d20761SAnand Jain 3483f46b5a66SChristoph Hellwig kfree(vol_args); 3484c9e9f97bSIlya Dryomov out: 3485a174c0a2SNikolay Borisov if (restore_op) 3486a174c0a2SNikolay Borisov btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); 3487a174c0a2SNikolay Borisov else 3488c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3489f46b5a66SChristoph Hellwig return ret; 3490f46b5a66SChristoph Hellwig } 3491f46b5a66SChristoph Hellwig 34926b526ed7SAnand Jain static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg) 3493f46b5a66SChristoph Hellwig { 34941a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 34950b246afaSJeff Mahoney struct inode *inode = file_inode(file); 34960b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 34976b526ed7SAnand Jain struct btrfs_ioctl_vol_args_v2 *vol_args; 34983fa421deSJosef Bacik struct block_device *bdev = NULL; 34993fa421deSJosef Bacik fmode_t mode; 3500f46b5a66SChristoph Hellwig int ret; 350167ae34b6SDavid Sterba bool cancel = false; 3502f46b5a66SChristoph Hellwig 3503e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3504e441d54dSChris Mason return -EPERM; 3505e441d54dSChris Mason 3506dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3507d815b3f2SDan Carpenter if (IS_ERR(vol_args)) 3508d815b3f2SDan Carpenter return PTR_ERR(vol_args); 3509f46b5a66SChristoph Hellwig 3510748449cdSDavid Sterba if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) { 3511fd4e994bSOmar Sandoval ret = -EOPNOTSUPP; 3512fd4e994bSOmar Sandoval goto out; 3513fd4e994bSOmar Sandoval } 35141a15eb72SJosef Bacik 35156b526ed7SAnand Jain vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 35161a15eb72SJosef Bacik if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) { 35171a15eb72SJosef Bacik args.devid = vol_args->devid; 35181a15eb72SJosef Bacik } else if (!strcmp("cancel", vol_args->name)) { 351967ae34b6SDavid Sterba cancel = true; 35201a15eb72SJosef Bacik } else { 35211a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 35221a15eb72SJosef Bacik if (ret) 35231a15eb72SJosef Bacik goto out; 35241a15eb72SJosef Bacik } 35251a15eb72SJosef Bacik 35261a15eb72SJosef Bacik ret = mnt_want_write_file(file); 35271a15eb72SJosef Bacik if (ret) 35281a15eb72SJosef Bacik goto out; 352967ae34b6SDavid Sterba 353067ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 353167ae34b6SDavid Sterba cancel); 353267ae34b6SDavid Sterba if (ret) 35331a15eb72SJosef Bacik goto err_drop; 353467ae34b6SDavid Sterba 35351a15eb72SJosef Bacik /* Exclusive operation is now claimed */ 35361a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 353767ae34b6SDavid Sterba 3538c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3539183860f6SAnand Jain 35406b526ed7SAnand Jain if (!ret) { 3541735654eaSDavid Sterba if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) 35420b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: id %llu", 35436b526ed7SAnand Jain vol_args->devid); 35446b526ed7SAnand Jain else 35450b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: %s", 35466b526ed7SAnand Jain vol_args->name); 35476b526ed7SAnand Jain } 3548c47ca32dSDan Carpenter err_drop: 35494ac20c70SIlya Dryomov mnt_drop_write_file(file); 35503fa421deSJosef Bacik if (bdev) 35513fa421deSJosef Bacik blkdev_put(bdev, mode); 35521a15eb72SJosef Bacik out: 35531a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 35541a15eb72SJosef Bacik kfree(vol_args); 3555f46b5a66SChristoph Hellwig return ret; 3556f46b5a66SChristoph Hellwig } 3557f46b5a66SChristoph Hellwig 3558f46b5a66SChristoph Hellwig static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 3559f46b5a66SChristoph Hellwig { 35601a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 35610b246afaSJeff Mahoney struct inode *inode = file_inode(file); 35620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3563f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 35643fa421deSJosef Bacik struct block_device *bdev = NULL; 35653fa421deSJosef Bacik fmode_t mode; 3566f46b5a66SChristoph Hellwig int ret; 356737b45995STom Rix bool cancel = false; 3568f46b5a66SChristoph Hellwig 3569f46b5a66SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 3570f46b5a66SChristoph Hellwig return -EPERM; 3571f46b5a66SChristoph Hellwig 35721a15eb72SJosef Bacik vol_args = memdup_user(arg, sizeof(*vol_args)); 35731a15eb72SJosef Bacik if (IS_ERR(vol_args)) 35741a15eb72SJosef Bacik return PTR_ERR(vol_args); 35751a15eb72SJosef Bacik 35761a15eb72SJosef Bacik vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 35771a15eb72SJosef Bacik if (!strcmp("cancel", vol_args->name)) { 35781a15eb72SJosef Bacik cancel = true; 35791a15eb72SJosef Bacik } else { 35801a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 35811a15eb72SJosef Bacik if (ret) 35821a15eb72SJosef Bacik goto out; 35831a15eb72SJosef Bacik } 35841a15eb72SJosef Bacik 3585f46b5a66SChristoph Hellwig ret = mnt_want_write_file(file); 3586f46b5a66SChristoph Hellwig if (ret) 35871a15eb72SJosef Bacik goto out; 3588f46b5a66SChristoph Hellwig 358967ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 359067ae34b6SDavid Sterba cancel); 359167ae34b6SDavid Sterba if (ret == 0) { 35921a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 3593f46b5a66SChristoph Hellwig if (!ret) 35940b246afaSJeff Mahoney btrfs_info(fs_info, "disk deleted %s", vol_args->name); 3595c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 359667ae34b6SDavid Sterba } 359767ae34b6SDavid Sterba 3598f46b5a66SChristoph Hellwig mnt_drop_write_file(file); 35993fa421deSJosef Bacik if (bdev) 36003fa421deSJosef Bacik blkdev_put(bdev, mode); 36011a15eb72SJosef Bacik out: 36021a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 36031a15eb72SJosef Bacik kfree(vol_args); 3604f46b5a66SChristoph Hellwig return ret; 3605f46b5a66SChristoph Hellwig } 3606f46b5a66SChristoph Hellwig 36072ff7e61eSJeff Mahoney static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info, 36082ff7e61eSJeff Mahoney void __user *arg) 3609475f6387SJan Schmidt { 3610027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 3611475f6387SJan Schmidt struct btrfs_device *device; 36120b246afaSJeff Mahoney struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3613137c5418SJohannes Thumshirn u64 flags_in; 3614027ed2f0SLi Zefan int ret = 0; 3615475f6387SJan Schmidt 3616137c5418SJohannes Thumshirn fi_args = memdup_user(arg, sizeof(*fi_args)); 3617137c5418SJohannes Thumshirn if (IS_ERR(fi_args)) 3618137c5418SJohannes Thumshirn return PTR_ERR(fi_args); 3619137c5418SJohannes Thumshirn 3620137c5418SJohannes Thumshirn flags_in = fi_args->flags; 3621137c5418SJohannes Thumshirn memset(fi_args, 0, sizeof(*fi_args)); 3622027ed2f0SLi Zefan 3623d03262c7SDavid Sterba rcu_read_lock(); 3624027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 3625475f6387SJan Schmidt 3626d03262c7SDavid Sterba list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 3627027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 3628027ed2f0SLi Zefan fi_args->max_id = device->devid; 3629475f6387SJan Schmidt } 3630d03262c7SDavid Sterba rcu_read_unlock(); 3631475f6387SJan Schmidt 3632de37aa51SNikolay Borisov memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid)); 3633bea7eafdSOmar Sandoval fi_args->nodesize = fs_info->nodesize; 3634bea7eafdSOmar Sandoval fi_args->sectorsize = fs_info->sectorsize; 3635bea7eafdSOmar Sandoval fi_args->clone_alignment = fs_info->sectorsize; 363680a773fbSDavid Sterba 3637137c5418SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) { 3638137c5418SJohannes Thumshirn fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy); 3639137c5418SJohannes Thumshirn fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy); 3640137c5418SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO; 3641137c5418SJohannes Thumshirn } 3642137c5418SJohannes Thumshirn 36430fb408a5SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) { 36440fb408a5SJohannes Thumshirn fi_args->generation = fs_info->generation; 36450fb408a5SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION; 36460fb408a5SJohannes Thumshirn } 36470fb408a5SJohannes Thumshirn 364849bac897SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) { 364949bac897SJohannes Thumshirn memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid, 365049bac897SJohannes Thumshirn sizeof(fi_args->metadata_uuid)); 365149bac897SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID; 365249bac897SJohannes Thumshirn } 365349bac897SJohannes Thumshirn 3654027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 3655027ed2f0SLi Zefan ret = -EFAULT; 3656475f6387SJan Schmidt 3657027ed2f0SLi Zefan kfree(fi_args); 3658027ed2f0SLi Zefan return ret; 3659475f6387SJan Schmidt } 3660475f6387SJan Schmidt 36612ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info, 36622ff7e61eSJeff Mahoney void __user *arg) 3663475f6387SJan Schmidt { 3664562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 3665475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 3666475f6387SJan Schmidt struct btrfs_device *dev; 3667475f6387SJan Schmidt int ret = 0; 3668475f6387SJan Schmidt 3669475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 3670475f6387SJan Schmidt if (IS_ERR(di_args)) 3671475f6387SJan Schmidt return PTR_ERR(di_args); 3672475f6387SJan Schmidt 3673562d7b15SJosef Bacik args.devid = di_args->devid; 3674dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(di_args->uuid)) 3675562d7b15SJosef Bacik args.uuid = di_args->uuid; 3676475f6387SJan Schmidt 3677c5593ca3SDavid Sterba rcu_read_lock(); 3678562d7b15SJosef Bacik dev = btrfs_find_device(fs_info->fs_devices, &args); 3679475f6387SJan Schmidt if (!dev) { 3680475f6387SJan Schmidt ret = -ENODEV; 3681475f6387SJan Schmidt goto out; 3682475f6387SJan Schmidt } 3683475f6387SJan Schmidt 3684475f6387SJan Schmidt di_args->devid = dev->devid; 36857cc8e58dSMiao Xie di_args->bytes_used = btrfs_device_get_bytes_used(dev); 36867cc8e58dSMiao Xie di_args->total_bytes = btrfs_device_get_total_bytes(dev); 3687475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 3688a27202fbSJim Meyering if (dev->name) { 3689672d5990SMisono Tomohiro strncpy(di_args->path, rcu_str_deref(dev->name), 3690672d5990SMisono Tomohiro sizeof(di_args->path) - 1); 3691a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 3692a27202fbSJim Meyering } else { 369399ba55adSStefan Behrens di_args->path[0] = '\0'; 3694a27202fbSJim Meyering } 3695475f6387SJan Schmidt 3696475f6387SJan Schmidt out: 3697c5593ca3SDavid Sterba rcu_read_unlock(); 3698475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 3699475f6387SJan Schmidt ret = -EFAULT; 3700475f6387SJan Schmidt 3701475f6387SJan Schmidt kfree(di_args); 3702475f6387SJan Schmidt return ret; 3703475f6387SJan Schmidt } 3704475f6387SJan Schmidt 37056ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 37066ef5ed0dSJosef Bacik { 3707496ad9aaSAl Viro struct inode *inode = file_inode(file); 37080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 37096ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 37106ef5ed0dSJosef Bacik struct btrfs_root *new_root; 37116ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 37126ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 37132a2b5d62SJosef Bacik struct btrfs_path *path = NULL; 37146ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 37156ef5ed0dSJosef Bacik u64 objectid = 0; 37166ef5ed0dSJosef Bacik u64 dir_id; 37173c04ce01SMiao Xie int ret; 37186ef5ed0dSJosef Bacik 37196ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 37206ef5ed0dSJosef Bacik return -EPERM; 37216ef5ed0dSJosef Bacik 37223c04ce01SMiao Xie ret = mnt_want_write_file(file); 37233c04ce01SMiao Xie if (ret) 37243c04ce01SMiao Xie return ret; 37253c04ce01SMiao Xie 37263c04ce01SMiao Xie if (copy_from_user(&objectid, argp, sizeof(objectid))) { 37273c04ce01SMiao Xie ret = -EFAULT; 37283c04ce01SMiao Xie goto out; 37293c04ce01SMiao Xie } 37306ef5ed0dSJosef Bacik 37316ef5ed0dSJosef Bacik if (!objectid) 37321cecf579Schandan objectid = BTRFS_FS_TREE_OBJECTID; 37336ef5ed0dSJosef Bacik 373456e9357aSDavid Sterba new_root = btrfs_get_fs_root(fs_info, objectid, true); 37353c04ce01SMiao Xie if (IS_ERR(new_root)) { 37363c04ce01SMiao Xie ret = PTR_ERR(new_root); 37373c04ce01SMiao Xie goto out; 37383c04ce01SMiao Xie } 37392a2b5d62SJosef Bacik if (!is_fstree(new_root->root_key.objectid)) { 37402a2b5d62SJosef Bacik ret = -ENOENT; 37412a2b5d62SJosef Bacik goto out_free; 37422a2b5d62SJosef Bacik } 37436ef5ed0dSJosef Bacik 37446ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 37453c04ce01SMiao Xie if (!path) { 37463c04ce01SMiao Xie ret = -ENOMEM; 37472a2b5d62SJosef Bacik goto out_free; 37483c04ce01SMiao Xie } 37496ef5ed0dSJosef Bacik 37506ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 375198d5dc13STsutomu Itoh if (IS_ERR(trans)) { 37523c04ce01SMiao Xie ret = PTR_ERR(trans); 37532a2b5d62SJosef Bacik goto out_free; 37546ef5ed0dSJosef Bacik } 37556ef5ed0dSJosef Bacik 37560b246afaSJeff Mahoney dir_id = btrfs_super_root_dir(fs_info->super_copy); 37570b246afaSJeff Mahoney di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path, 37586ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 3759cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 37602a2b5d62SJosef Bacik btrfs_release_path(path); 37613a45bb20SJeff Mahoney btrfs_end_transaction(trans); 37620b246afaSJeff Mahoney btrfs_err(fs_info, 37635d163e0eSJeff Mahoney "Umm, you don't have the default diritem, this isn't going to work"); 37643c04ce01SMiao Xie ret = -ENOENT; 37652a2b5d62SJosef Bacik goto out_free; 37666ef5ed0dSJosef Bacik } 37676ef5ed0dSJosef Bacik 37686ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 37696ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 37706ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 37712a2b5d62SJosef Bacik btrfs_release_path(path); 37726ef5ed0dSJosef Bacik 37730b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL); 37743a45bb20SJeff Mahoney btrfs_end_transaction(trans); 37752a2b5d62SJosef Bacik out_free: 377600246528SJosef Bacik btrfs_put_root(new_root); 37772a2b5d62SJosef Bacik btrfs_free_path(path); 37783c04ce01SMiao Xie out: 37793c04ce01SMiao Xie mnt_drop_write_file(file); 37803c04ce01SMiao Xie return ret; 37816ef5ed0dSJosef Bacik } 37826ef5ed0dSJosef Bacik 3783c065f5b1SSu Yue static void get_block_group_info(struct list_head *groups_list, 3784bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 3785bf5fc093SJosef Bacik { 378632da5386SDavid Sterba struct btrfs_block_group *block_group; 3787bf5fc093SJosef Bacik 3788bf5fc093SJosef Bacik space->total_bytes = 0; 3789bf5fc093SJosef Bacik space->used_bytes = 0; 3790bf5fc093SJosef Bacik space->flags = 0; 3791bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 3792bf5fc093SJosef Bacik space->flags = block_group->flags; 3793b3470b5dSDavid Sterba space->total_bytes += block_group->length; 3794bf38be65SDavid Sterba space->used_bytes += block_group->used; 3795bf5fc093SJosef Bacik } 3796bf5fc093SJosef Bacik } 3797bf5fc093SJosef Bacik 37982ff7e61eSJeff Mahoney static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info, 37992ff7e61eSJeff Mahoney void __user *arg) 38001406e432SJosef Bacik { 38011406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 38021406e432SJosef Bacik struct btrfs_ioctl_space_info space; 38031406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 38047fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 380513f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 38061406e432SJosef Bacik struct btrfs_space_info *info; 3807315d8e98SColin Ian King static const u64 types[] = { 3808315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA, 3809bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 3810bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 3811315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA 3812315d8e98SColin Ian King }; 3813bf5fc093SJosef Bacik int num_types = 4; 38147fde62bfSChris Mason int alloc_size; 38151406e432SJosef Bacik int ret = 0; 381651788b1bSDan Rosenberg u64 slot_count = 0; 3817bf5fc093SJosef Bacik int i, c; 38181406e432SJosef Bacik 38191406e432SJosef Bacik if (copy_from_user(&space_args, 38201406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 38211406e432SJosef Bacik sizeof(space_args))) 38221406e432SJosef Bacik return -EFAULT; 38231406e432SJosef Bacik 3824bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3825bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3826bf5fc093SJosef Bacik 3827bf5fc093SJosef Bacik info = NULL; 382872804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3829bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3830bf5fc093SJosef Bacik info = tmp; 3831bf5fc093SJosef Bacik break; 3832bf5fc093SJosef Bacik } 3833bf5fc093SJosef Bacik } 38341406e432SJosef Bacik 3835bf5fc093SJosef Bacik if (!info) 3836bf5fc093SJosef Bacik continue; 3837bf5fc093SJosef Bacik 3838bf5fc093SJosef Bacik down_read(&info->groups_sem); 3839bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3840bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 3841bf5fc093SJosef Bacik slot_count++; 3842bf5fc093SJosef Bacik } 3843bf5fc093SJosef Bacik up_read(&info->groups_sem); 3844bf5fc093SJosef Bacik } 3845bf5fc093SJosef Bacik 384636523e95SDavid Sterba /* 384736523e95SDavid Sterba * Global block reserve, exported as a space_info 384836523e95SDavid Sterba */ 384936523e95SDavid Sterba slot_count++; 385036523e95SDavid Sterba 38517fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 38527fde62bfSChris Mason if (space_args.space_slots == 0) { 38537fde62bfSChris Mason space_args.total_spaces = slot_count; 38547fde62bfSChris Mason goto out; 38557fde62bfSChris Mason } 3856bf5fc093SJosef Bacik 385751788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 3858bf5fc093SJosef Bacik 38597fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 3860bf5fc093SJosef Bacik 38617fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 38627fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 38637fde62bfSChris Mason */ 386409cbfeafSKirill A. Shutemov if (alloc_size > PAGE_SIZE) 38657fde62bfSChris Mason return -ENOMEM; 38667fde62bfSChris Mason 38677fde62bfSChris Mason space_args.total_spaces = 0; 38688d2db785SDavid Sterba dest = kmalloc(alloc_size, GFP_KERNEL); 38697fde62bfSChris Mason if (!dest) 38707fde62bfSChris Mason return -ENOMEM; 38717fde62bfSChris Mason dest_orig = dest; 38727fde62bfSChris Mason 38737fde62bfSChris Mason /* now we have a buffer to copy into */ 3874bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3875bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3876bf5fc093SJosef Bacik 387751788b1bSDan Rosenberg if (!slot_count) 387851788b1bSDan Rosenberg break; 387951788b1bSDan Rosenberg 3880bf5fc093SJosef Bacik info = NULL; 388172804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3882bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3883bf5fc093SJosef Bacik info = tmp; 38847fde62bfSChris Mason break; 3885bf5fc093SJosef Bacik } 3886bf5fc093SJosef Bacik } 38877fde62bfSChris Mason 3888bf5fc093SJosef Bacik if (!info) 3889bf5fc093SJosef Bacik continue; 3890bf5fc093SJosef Bacik down_read(&info->groups_sem); 3891bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3892bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 3893c065f5b1SSu Yue get_block_group_info(&info->block_groups[c], 3894c065f5b1SSu Yue &space); 38957fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 38961406e432SJosef Bacik dest++; 38971406e432SJosef Bacik space_args.total_spaces++; 389851788b1bSDan Rosenberg slot_count--; 38991406e432SJosef Bacik } 390051788b1bSDan Rosenberg if (!slot_count) 390151788b1bSDan Rosenberg break; 3902bf5fc093SJosef Bacik } 3903bf5fc093SJosef Bacik up_read(&info->groups_sem); 3904bf5fc093SJosef Bacik } 39051406e432SJosef Bacik 390636523e95SDavid Sterba /* 390736523e95SDavid Sterba * Add global block reserve 390836523e95SDavid Sterba */ 390936523e95SDavid Sterba if (slot_count) { 39100b246afaSJeff Mahoney struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; 391136523e95SDavid Sterba 391236523e95SDavid Sterba spin_lock(&block_rsv->lock); 391336523e95SDavid Sterba space.total_bytes = block_rsv->size; 391436523e95SDavid Sterba space.used_bytes = block_rsv->size - block_rsv->reserved; 391536523e95SDavid Sterba spin_unlock(&block_rsv->lock); 391636523e95SDavid Sterba space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV; 391736523e95SDavid Sterba memcpy(dest, &space, sizeof(space)); 391836523e95SDavid Sterba space_args.total_spaces++; 391936523e95SDavid Sterba } 392036523e95SDavid Sterba 39212eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 39227fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 39237fde62bfSChris Mason 39247fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 39257fde62bfSChris Mason ret = -EFAULT; 39267fde62bfSChris Mason 39277fde62bfSChris Mason kfree(dest_orig); 39287fde62bfSChris Mason out: 39297fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 39301406e432SJosef Bacik ret = -EFAULT; 39311406e432SJosef Bacik 39321406e432SJosef Bacik return ret; 39331406e432SJosef Bacik } 39341406e432SJosef Bacik 39359a8c28beSMiao Xie static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 39369a8c28beSMiao Xie void __user *argp) 393746204592SSage Weil { 393846204592SSage Weil struct btrfs_trans_handle *trans; 393946204592SSage Weil u64 transid; 394046204592SSage Weil 3941d4edf39bSMiao Xie trans = btrfs_attach_transaction_barrier(root); 3942ff7c1d33SMiao Xie if (IS_ERR(trans)) { 3943ff7c1d33SMiao Xie if (PTR_ERR(trans) != -ENOENT) 394498d5dc13STsutomu Itoh return PTR_ERR(trans); 3945ff7c1d33SMiao Xie 3946ff7c1d33SMiao Xie /* No running transaction, don't bother */ 3947ff7c1d33SMiao Xie transid = root->fs_info->last_trans_committed; 3948ff7c1d33SMiao Xie goto out; 3949ff7c1d33SMiao Xie } 395046204592SSage Weil transid = trans->transid; 3951fdfbf020SJosef Bacik btrfs_commit_transaction_async(trans); 3952ff7c1d33SMiao Xie out: 395346204592SSage Weil if (argp) 395446204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 395546204592SSage Weil return -EFAULT; 395646204592SSage Weil return 0; 395746204592SSage Weil } 395846204592SSage Weil 39592ff7e61eSJeff Mahoney static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info, 39609a8c28beSMiao Xie void __user *argp) 396146204592SSage Weil { 396246204592SSage Weil u64 transid; 396346204592SSage Weil 396446204592SSage Weil if (argp) { 396546204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 396646204592SSage Weil return -EFAULT; 396746204592SSage Weil } else { 396846204592SSage Weil transid = 0; /* current trans */ 396946204592SSage Weil } 39702ff7e61eSJeff Mahoney return btrfs_wait_for_commit(fs_info, transid); 397146204592SSage Weil } 397246204592SSage Weil 3973b8e95489SMiao Xie static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 3974475f6387SJan Schmidt { 39750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 3976475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3977b8e95489SMiao Xie int ret; 3978475f6387SJan Schmidt 3979475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3980475f6387SJan Schmidt return -EPERM; 3981475f6387SJan Schmidt 3982*da32c6d5SJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 3983*da32c6d5SJosef Bacik btrfs_err(fs_info, "scrub is not supported on extent tree v2 yet"); 3984*da32c6d5SJosef Bacik return -EINVAL; 3985*da32c6d5SJosef Bacik } 3986*da32c6d5SJosef Bacik 3987475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3988475f6387SJan Schmidt if (IS_ERR(sa)) 3989475f6387SJan Schmidt return PTR_ERR(sa); 3990475f6387SJan Schmidt 3991b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 3992b8e95489SMiao Xie ret = mnt_want_write_file(file); 3993b8e95489SMiao Xie if (ret) 3994b8e95489SMiao Xie goto out; 3995b8e95489SMiao Xie } 3996b8e95489SMiao Xie 39970b246afaSJeff Mahoney ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end, 399863a212abSStefan Behrens &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 399963a212abSStefan Behrens 0); 4000475f6387SJan Schmidt 40015afe6ce7SFilipe Manana /* 40025afe6ce7SFilipe Manana * Copy scrub args to user space even if btrfs_scrub_dev() returned an 40035afe6ce7SFilipe Manana * error. This is important as it allows user space to know how much 40045afe6ce7SFilipe Manana * progress scrub has done. For example, if scrub is canceled we get 40055afe6ce7SFilipe Manana * -ECANCELED from btrfs_scrub_dev() and return that error back to user 40065afe6ce7SFilipe Manana * space. Later user space can inspect the progress from the structure 40075afe6ce7SFilipe Manana * btrfs_ioctl_scrub_args and resume scrub from where it left off 40085afe6ce7SFilipe Manana * previously (btrfs-progs does this). 40095afe6ce7SFilipe Manana * If we fail to copy the btrfs_ioctl_scrub_args structure to user space 40105afe6ce7SFilipe Manana * then return -EFAULT to signal the structure was not copied or it may 40115afe6ce7SFilipe Manana * be corrupt and unreliable due to a partial copy. 40125afe6ce7SFilipe Manana */ 40135afe6ce7SFilipe Manana if (copy_to_user(arg, sa, sizeof(*sa))) 4014475f6387SJan Schmidt ret = -EFAULT; 4015475f6387SJan Schmidt 4016b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) 4017b8e95489SMiao Xie mnt_drop_write_file(file); 4018b8e95489SMiao Xie out: 4019475f6387SJan Schmidt kfree(sa); 4020475f6387SJan Schmidt return ret; 4021475f6387SJan Schmidt } 4022475f6387SJan Schmidt 40232ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info) 4024475f6387SJan Schmidt { 4025475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4026475f6387SJan Schmidt return -EPERM; 4027475f6387SJan Schmidt 40282ff7e61eSJeff Mahoney return btrfs_scrub_cancel(fs_info); 4029475f6387SJan Schmidt } 4030475f6387SJan Schmidt 40312ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info, 4032475f6387SJan Schmidt void __user *arg) 4033475f6387SJan Schmidt { 4034475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 4035475f6387SJan Schmidt int ret; 4036475f6387SJan Schmidt 4037475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4038475f6387SJan Schmidt return -EPERM; 4039475f6387SJan Schmidt 4040475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 4041475f6387SJan Schmidt if (IS_ERR(sa)) 4042475f6387SJan Schmidt return PTR_ERR(sa); 4043475f6387SJan Schmidt 40442ff7e61eSJeff Mahoney ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress); 4045475f6387SJan Schmidt 40464fa99b00SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 4047475f6387SJan Schmidt ret = -EFAULT; 4048475f6387SJan Schmidt 4049475f6387SJan Schmidt kfree(sa); 4050475f6387SJan Schmidt return ret; 4051475f6387SJan Schmidt } 4052475f6387SJan Schmidt 40532ff7e61eSJeff Mahoney static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info, 4054b27f7c0cSDavid Sterba void __user *arg) 4055c11d2c23SStefan Behrens { 4056c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 4057c11d2c23SStefan Behrens int ret; 4058c11d2c23SStefan Behrens 4059c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 4060c11d2c23SStefan Behrens if (IS_ERR(sa)) 4061c11d2c23SStefan Behrens return PTR_ERR(sa); 4062c11d2c23SStefan Behrens 4063b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 4064b27f7c0cSDavid Sterba kfree(sa); 4065b27f7c0cSDavid Sterba return -EPERM; 4066b27f7c0cSDavid Sterba } 4067b27f7c0cSDavid Sterba 40682ff7e61eSJeff Mahoney ret = btrfs_get_dev_stats(fs_info, sa); 4069c11d2c23SStefan Behrens 4070eee99577SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 4071c11d2c23SStefan Behrens ret = -EFAULT; 4072c11d2c23SStefan Behrens 4073c11d2c23SStefan Behrens kfree(sa); 4074c11d2c23SStefan Behrens return ret; 4075c11d2c23SStefan Behrens } 4076c11d2c23SStefan Behrens 40772ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info, 40782ff7e61eSJeff Mahoney void __user *arg) 40793f6bcfbdSStefan Behrens { 40803f6bcfbdSStefan Behrens struct btrfs_ioctl_dev_replace_args *p; 40813f6bcfbdSStefan Behrens int ret; 40823f6bcfbdSStefan Behrens 40833f6bcfbdSStefan Behrens if (!capable(CAP_SYS_ADMIN)) 40843f6bcfbdSStefan Behrens return -EPERM; 40853f6bcfbdSStefan Behrens 4086914a519bSJosef Bacik if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 4087914a519bSJosef Bacik btrfs_err(fs_info, "device replace not supported on extent tree v2 yet"); 4088914a519bSJosef Bacik return -EINVAL; 4089914a519bSJosef Bacik } 4090914a519bSJosef Bacik 40913f6bcfbdSStefan Behrens p = memdup_user(arg, sizeof(*p)); 40923f6bcfbdSStefan Behrens if (IS_ERR(p)) 40933f6bcfbdSStefan Behrens return PTR_ERR(p); 40943f6bcfbdSStefan Behrens 40953f6bcfbdSStefan Behrens switch (p->cmd) { 40963f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 4097bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 4098adfa97cbSIlya Dryomov ret = -EROFS; 4099adfa97cbSIlya Dryomov goto out; 4100adfa97cbSIlya Dryomov } 4101c3e1f96cSGoldwyn Rodrigues if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { 4102e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 41033f6bcfbdSStefan Behrens } else { 41042ff7e61eSJeff Mahoney ret = btrfs_dev_replace_by_ioctl(fs_info, p); 4105c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 41063f6bcfbdSStefan Behrens } 41073f6bcfbdSStefan Behrens break; 41083f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 41090b246afaSJeff Mahoney btrfs_dev_replace_status(fs_info, p); 41103f6bcfbdSStefan Behrens ret = 0; 41113f6bcfbdSStefan Behrens break; 41123f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 411317d202b9SAnand Jain p->result = btrfs_dev_replace_cancel(fs_info); 411497282031SAnand Jain ret = 0; 41153f6bcfbdSStefan Behrens break; 41163f6bcfbdSStefan Behrens default: 41173f6bcfbdSStefan Behrens ret = -EINVAL; 41183f6bcfbdSStefan Behrens break; 41193f6bcfbdSStefan Behrens } 41203f6bcfbdSStefan Behrens 4121d3a53286SFilipe Manana if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p))) 41223f6bcfbdSStefan Behrens ret = -EFAULT; 4123adfa97cbSIlya Dryomov out: 41243f6bcfbdSStefan Behrens kfree(p); 41253f6bcfbdSStefan Behrens return ret; 41263f6bcfbdSStefan Behrens } 41273f6bcfbdSStefan Behrens 4128d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 4129d7728c96SJan Schmidt { 4130d7728c96SJan Schmidt int ret = 0; 4131d7728c96SJan Schmidt int i; 4132740c3d22SChris Mason u64 rel_ptr; 4133d7728c96SJan Schmidt int size; 4134806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 4135d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 4136d7728c96SJan Schmidt struct btrfs_path *path; 4137d7728c96SJan Schmidt 413882b22ac8SKusanagi Kouichi if (!capable(CAP_DAC_READ_SEARCH)) 4139d7728c96SJan Schmidt return -EPERM; 4140d7728c96SJan Schmidt 4141d7728c96SJan Schmidt path = btrfs_alloc_path(); 4142d7728c96SJan Schmidt if (!path) { 4143d7728c96SJan Schmidt ret = -ENOMEM; 4144d7728c96SJan Schmidt goto out; 4145d7728c96SJan Schmidt } 4146d7728c96SJan Schmidt 4147d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 4148d7728c96SJan Schmidt if (IS_ERR(ipa)) { 4149d7728c96SJan Schmidt ret = PTR_ERR(ipa); 4150d7728c96SJan Schmidt ipa = NULL; 4151d7728c96SJan Schmidt goto out; 4152d7728c96SJan Schmidt } 4153d7728c96SJan Schmidt 4154d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 4155d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 4156d7728c96SJan Schmidt if (IS_ERR(ipath)) { 4157d7728c96SJan Schmidt ret = PTR_ERR(ipath); 4158d7728c96SJan Schmidt ipath = NULL; 4159d7728c96SJan Schmidt goto out; 4160d7728c96SJan Schmidt } 4161d7728c96SJan Schmidt 4162d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 4163d7728c96SJan Schmidt if (ret < 0) 4164d7728c96SJan Schmidt goto out; 4165d7728c96SJan Schmidt 4166d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 4167745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 4168745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 4169740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 4170d7728c96SJan Schmidt } 4171d7728c96SJan Schmidt 4172718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)ipa->fspath, 4173718dc5faSOmar Sandoval ipath->fspath, size); 4174d7728c96SJan Schmidt if (ret) { 4175d7728c96SJan Schmidt ret = -EFAULT; 4176d7728c96SJan Schmidt goto out; 4177d7728c96SJan Schmidt } 4178d7728c96SJan Schmidt 4179d7728c96SJan Schmidt out: 4180d7728c96SJan Schmidt btrfs_free_path(path); 4181d7728c96SJan Schmidt free_ipath(ipath); 4182d7728c96SJan Schmidt kfree(ipa); 4183d7728c96SJan Schmidt 4184d7728c96SJan Schmidt return ret; 4185d7728c96SJan Schmidt } 4186d7728c96SJan Schmidt 4187d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 4188d7728c96SJan Schmidt { 4189d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 4190d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 4191d7728c96SJan Schmidt 4192d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 4193d7728c96SJan Schmidt inodes->bytes_left -= c; 4194d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 4195d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 4196d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 4197d7728c96SJan Schmidt inodes->elem_cnt += 3; 4198d7728c96SJan Schmidt } else { 4199d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 4200d7728c96SJan Schmidt inodes->bytes_left = 0; 4201d7728c96SJan Schmidt inodes->elem_missed += 3; 4202d7728c96SJan Schmidt } 4203d7728c96SJan Schmidt 4204d7728c96SJan Schmidt return 0; 4205d7728c96SJan Schmidt } 4206d7728c96SJan Schmidt 42072ff7e61eSJeff Mahoney static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info, 4208d24a67b2SZygo Blaxell void __user *arg, int version) 4209d7728c96SJan Schmidt { 4210d7728c96SJan Schmidt int ret = 0; 4211d7728c96SJan Schmidt int size; 4212d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 4213d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 4214d7728c96SJan Schmidt struct btrfs_path *path = NULL; 4215d24a67b2SZygo Blaxell bool ignore_offset; 4216d7728c96SJan Schmidt 4217d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4218d7728c96SJan Schmidt return -EPERM; 4219d7728c96SJan Schmidt 4220d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 42217b9ea627SShailendra Verma if (IS_ERR(loi)) 42227b9ea627SShailendra Verma return PTR_ERR(loi); 4223d7728c96SJan Schmidt 4224d24a67b2SZygo Blaxell if (version == 1) { 4225d24a67b2SZygo Blaxell ignore_offset = false; 4226b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_64K); 4227d24a67b2SZygo Blaxell } else { 4228d24a67b2SZygo Blaxell /* All reserved bits must be 0 for now */ 4229d24a67b2SZygo Blaxell if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) { 4230d24a67b2SZygo Blaxell ret = -EINVAL; 4231d24a67b2SZygo Blaxell goto out_loi; 4232d24a67b2SZygo Blaxell } 4233d24a67b2SZygo Blaxell /* Only accept flags we have defined so far */ 4234d24a67b2SZygo Blaxell if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) { 4235d24a67b2SZygo Blaxell ret = -EINVAL; 4236d24a67b2SZygo Blaxell goto out_loi; 4237d24a67b2SZygo Blaxell } 4238d24a67b2SZygo Blaxell ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET; 4239b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_16M); 4240d24a67b2SZygo Blaxell } 4241d24a67b2SZygo Blaxell 4242d7728c96SJan Schmidt path = btrfs_alloc_path(); 4243d7728c96SJan Schmidt if (!path) { 4244d7728c96SJan Schmidt ret = -ENOMEM; 4245d7728c96SJan Schmidt goto out; 4246d7728c96SJan Schmidt } 4247d7728c96SJan Schmidt 4248d7728c96SJan Schmidt inodes = init_data_container(size); 4249d7728c96SJan Schmidt if (IS_ERR(inodes)) { 4250d7728c96SJan Schmidt ret = PTR_ERR(inodes); 4251d7728c96SJan Schmidt inodes = NULL; 4252d7728c96SJan Schmidt goto out; 4253d7728c96SJan Schmidt } 4254d7728c96SJan Schmidt 42552ff7e61eSJeff Mahoney ret = iterate_inodes_from_logical(loi->logical, fs_info, path, 4256d24a67b2SZygo Blaxell build_ino_list, inodes, ignore_offset); 4257df031f07SLiu Bo if (ret == -EINVAL) 4258d7728c96SJan Schmidt ret = -ENOENT; 4259d7728c96SJan Schmidt if (ret < 0) 4260d7728c96SJan Schmidt goto out; 4261d7728c96SJan Schmidt 4262718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes, 4263718dc5faSOmar Sandoval size); 4264d7728c96SJan Schmidt if (ret) 4265d7728c96SJan Schmidt ret = -EFAULT; 4266d7728c96SJan Schmidt 4267d7728c96SJan Schmidt out: 4268d7728c96SJan Schmidt btrfs_free_path(path); 4269f54de068SDavid Sterba kvfree(inodes); 4270d24a67b2SZygo Blaxell out_loi: 4271d7728c96SJan Schmidt kfree(loi); 4272d7728c96SJan Schmidt 4273d7728c96SJan Schmidt return ret; 4274d7728c96SJan Schmidt } 4275d7728c96SJan Schmidt 4276008ef096SDavid Sterba void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info, 4277c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 4278c9e9f97bSIlya Dryomov { 4279c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4280c9e9f97bSIlya Dryomov 4281c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 4282c9e9f97bSIlya Dryomov 42833009a62fSDavid Sterba if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) 4284837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 4285837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 4286837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 4287a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 4288a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 4289837d5b6eSIlya Dryomov 4290c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 4291c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 4292c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 429319a39dceSIlya Dryomov 429419a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 429519a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 429619a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 4297c9e9f97bSIlya Dryomov } 4298c9e9f97bSIlya Dryomov 42999ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 4300c9e9f97bSIlya Dryomov { 4301496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4302c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 4303c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 4304c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 4305ed0fb78fSIlya Dryomov bool need_unlock; /* for mut. excl. ops lock */ 4306c9e9f97bSIlya Dryomov int ret; 4307c9e9f97bSIlya Dryomov 43086c405b24SNikolay Borisov if (!arg) 43096c405b24SNikolay Borisov btrfs_warn(fs_info, 43106c405b24SNikolay Borisov "IOC_BALANCE ioctl (v1) is deprecated and will be removed in kernel 5.18"); 43116c405b24SNikolay Borisov 4312c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4313c9e9f97bSIlya Dryomov return -EPERM; 4314c9e9f97bSIlya Dryomov 4315e54bfa31SLiu Bo ret = mnt_want_write_file(file); 43169ba1f6e4SLiu Bo if (ret) 43179ba1f6e4SLiu Bo return ret; 43189ba1f6e4SLiu Bo 4319ed0fb78fSIlya Dryomov again: 4320c3e1f96cSGoldwyn Rodrigues if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 4321c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4322ed0fb78fSIlya Dryomov need_unlock = true; 4323ed0fb78fSIlya Dryomov goto locked; 4324ed0fb78fSIlya Dryomov } 4325ed0fb78fSIlya Dryomov 4326ed0fb78fSIlya Dryomov /* 432701327610SNicholas D Steeves * mut. excl. ops lock is locked. Three possibilities: 4328ed0fb78fSIlya Dryomov * (1) some other op is running 4329ed0fb78fSIlya Dryomov * (2) balance is running 4330ed0fb78fSIlya Dryomov * (3) balance is paused -- special case (think resume) 4331ed0fb78fSIlya Dryomov */ 4332ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4333ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4334ed0fb78fSIlya Dryomov /* this is either (2) or (3) */ 43353009a62fSDavid Sterba if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4336ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4337dccdb07bSDavid Sterba /* 4338dccdb07bSDavid Sterba * Lock released to allow other waiters to continue, 4339dccdb07bSDavid Sterba * we'll reexamine the status again. 4340dccdb07bSDavid Sterba */ 4341ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4342ed0fb78fSIlya Dryomov 4343ed0fb78fSIlya Dryomov if (fs_info->balance_ctl && 43443009a62fSDavid Sterba !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4345ed0fb78fSIlya Dryomov /* this is (3) */ 4346ed0fb78fSIlya Dryomov need_unlock = false; 4347ed0fb78fSIlya Dryomov goto locked; 4348ed0fb78fSIlya Dryomov } 4349ed0fb78fSIlya Dryomov 4350ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4351ed0fb78fSIlya Dryomov goto again; 4352ed0fb78fSIlya Dryomov } else { 4353ed0fb78fSIlya Dryomov /* this is (2) */ 4354ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4355ed0fb78fSIlya Dryomov ret = -EINPROGRESS; 4356ed0fb78fSIlya Dryomov goto out; 4357ed0fb78fSIlya Dryomov } 4358ed0fb78fSIlya Dryomov } else { 4359ed0fb78fSIlya Dryomov /* this is (1) */ 4360ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4361e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4362ed0fb78fSIlya Dryomov goto out; 4363ed0fb78fSIlya Dryomov } 4364ed0fb78fSIlya Dryomov 4365ed0fb78fSIlya Dryomov locked: 4366c9e9f97bSIlya Dryomov 4367c9e9f97bSIlya Dryomov if (arg) { 4368c9e9f97bSIlya Dryomov bargs = memdup_user(arg, sizeof(*bargs)); 4369c9e9f97bSIlya Dryomov if (IS_ERR(bargs)) { 4370c9e9f97bSIlya Dryomov ret = PTR_ERR(bargs); 4371ed0fb78fSIlya Dryomov goto out_unlock; 4372c9e9f97bSIlya Dryomov } 4373de322263SIlya Dryomov 4374de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 4375de322263SIlya Dryomov if (!fs_info->balance_ctl) { 4376de322263SIlya Dryomov ret = -ENOTCONN; 4377de322263SIlya Dryomov goto out_bargs; 4378de322263SIlya Dryomov } 4379de322263SIlya Dryomov 4380de322263SIlya Dryomov bctl = fs_info->balance_ctl; 4381de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 4382de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 4383de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 4384efc0e69cSNikolay Borisov btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE); 4385de322263SIlya Dryomov 4386de322263SIlya Dryomov goto do_balance; 4387de322263SIlya Dryomov } 4388c9e9f97bSIlya Dryomov } else { 4389c9e9f97bSIlya Dryomov bargs = NULL; 4390c9e9f97bSIlya Dryomov } 4391c9e9f97bSIlya Dryomov 4392ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4393837d5b6eSIlya Dryomov ret = -EINPROGRESS; 4394837d5b6eSIlya Dryomov goto out_bargs; 4395837d5b6eSIlya Dryomov } 4396837d5b6eSIlya Dryomov 43978d2db785SDavid Sterba bctl = kzalloc(sizeof(*bctl), GFP_KERNEL); 4398c9e9f97bSIlya Dryomov if (!bctl) { 4399c9e9f97bSIlya Dryomov ret = -ENOMEM; 4400c9e9f97bSIlya Dryomov goto out_bargs; 4401c9e9f97bSIlya Dryomov } 4402c9e9f97bSIlya Dryomov 4403c9e9f97bSIlya Dryomov if (arg) { 4404c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 4405c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 4406c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 4407c9e9f97bSIlya Dryomov 4408c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 4409f43ffb60SIlya Dryomov } else { 4410f43ffb60SIlya Dryomov /* balance everything - no filters */ 4411f43ffb60SIlya Dryomov bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 4412c9e9f97bSIlya Dryomov } 4413c9e9f97bSIlya Dryomov 44148eb93459SDavid Sterba if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) { 44158eb93459SDavid Sterba ret = -EINVAL; 44160f89abf5SChristian Engelmayer goto out_bctl; 44178eb93459SDavid Sterba } 44188eb93459SDavid Sterba 4419de322263SIlya Dryomov do_balance: 4420c9e9f97bSIlya Dryomov /* 4421c3e1f96cSGoldwyn Rodrigues * Ownership of bctl and exclusive operation goes to btrfs_balance. 4422c3e1f96cSGoldwyn Rodrigues * bctl is freed in reset_balance_state, or, if restriper was paused 4423c3e1f96cSGoldwyn Rodrigues * all the way until unmount, in free_fs_info. The flag should be 4424c3e1f96cSGoldwyn Rodrigues * cleared after reset_balance_state. 4425c9e9f97bSIlya Dryomov */ 4426ed0fb78fSIlya Dryomov need_unlock = false; 4427ed0fb78fSIlya Dryomov 44286fcf6e2bSDavid Sterba ret = btrfs_balance(fs_info, bctl, bargs); 44290f89abf5SChristian Engelmayer bctl = NULL; 4430ed0fb78fSIlya Dryomov 4431d00c2d9cSFilipe Manana if ((ret == 0 || ret == -ECANCELED) && arg) { 4432c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 4433c9e9f97bSIlya Dryomov ret = -EFAULT; 4434c9e9f97bSIlya Dryomov } 4435c9e9f97bSIlya Dryomov 44360f89abf5SChristian Engelmayer out_bctl: 44370f89abf5SChristian Engelmayer kfree(bctl); 4438c9e9f97bSIlya Dryomov out_bargs: 4439c9e9f97bSIlya Dryomov kfree(bargs); 4440ed0fb78fSIlya Dryomov out_unlock: 4441c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4442ed0fb78fSIlya Dryomov if (need_unlock) 4443c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 4444ed0fb78fSIlya Dryomov out: 4445e54bfa31SLiu Bo mnt_drop_write_file(file); 4446c9e9f97bSIlya Dryomov return ret; 4447c9e9f97bSIlya Dryomov } 4448c9e9f97bSIlya Dryomov 44492ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd) 4450837d5b6eSIlya Dryomov { 4451837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4452837d5b6eSIlya Dryomov return -EPERM; 4453837d5b6eSIlya Dryomov 4454837d5b6eSIlya Dryomov switch (cmd) { 4455837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 44560b246afaSJeff Mahoney return btrfs_pause_balance(fs_info); 4457a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 44580b246afaSJeff Mahoney return btrfs_cancel_balance(fs_info); 4459837d5b6eSIlya Dryomov } 4460837d5b6eSIlya Dryomov 4461837d5b6eSIlya Dryomov return -EINVAL; 4462837d5b6eSIlya Dryomov } 4463837d5b6eSIlya Dryomov 44642ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info, 446519a39dceSIlya Dryomov void __user *arg) 446619a39dceSIlya Dryomov { 446719a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 446819a39dceSIlya Dryomov int ret = 0; 446919a39dceSIlya Dryomov 447019a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 447119a39dceSIlya Dryomov return -EPERM; 447219a39dceSIlya Dryomov 447319a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 447419a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 447519a39dceSIlya Dryomov ret = -ENOTCONN; 447619a39dceSIlya Dryomov goto out; 447719a39dceSIlya Dryomov } 447819a39dceSIlya Dryomov 44798d2db785SDavid Sterba bargs = kzalloc(sizeof(*bargs), GFP_KERNEL); 448019a39dceSIlya Dryomov if (!bargs) { 448119a39dceSIlya Dryomov ret = -ENOMEM; 448219a39dceSIlya Dryomov goto out; 448319a39dceSIlya Dryomov } 448419a39dceSIlya Dryomov 4485008ef096SDavid Sterba btrfs_update_ioctl_balance_args(fs_info, bargs); 448619a39dceSIlya Dryomov 448719a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 448819a39dceSIlya Dryomov ret = -EFAULT; 448919a39dceSIlya Dryomov 449019a39dceSIlya Dryomov kfree(bargs); 449119a39dceSIlya Dryomov out: 449219a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 449319a39dceSIlya Dryomov return ret; 449419a39dceSIlya Dryomov } 449519a39dceSIlya Dryomov 4496905b0ddaSMiao Xie static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 44975d13a37bSArne Jansen { 44980b246afaSJeff Mahoney struct inode *inode = file_inode(file); 44990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45005d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 45015d13a37bSArne Jansen int ret; 45025d13a37bSArne Jansen 45035d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 45045d13a37bSArne Jansen return -EPERM; 45055d13a37bSArne Jansen 4506905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4507905b0ddaSMiao Xie if (ret) 4508905b0ddaSMiao Xie return ret; 45095d13a37bSArne Jansen 45105d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4511905b0ddaSMiao Xie if (IS_ERR(sa)) { 4512905b0ddaSMiao Xie ret = PTR_ERR(sa); 4513905b0ddaSMiao Xie goto drop_write; 4514905b0ddaSMiao Xie } 45155d13a37bSArne Jansen 45160b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 45175d13a37bSArne Jansen 45185d13a37bSArne Jansen switch (sa->cmd) { 45195d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 4520340f1aa2SNikolay Borisov ret = btrfs_quota_enable(fs_info); 45215d13a37bSArne Jansen break; 45225d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 4523340f1aa2SNikolay Borisov ret = btrfs_quota_disable(fs_info); 45245d13a37bSArne Jansen break; 45255d13a37bSArne Jansen default: 45265d13a37bSArne Jansen ret = -EINVAL; 45275d13a37bSArne Jansen break; 45285d13a37bSArne Jansen } 45295d13a37bSArne Jansen 45305d13a37bSArne Jansen kfree(sa); 45310b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4532905b0ddaSMiao Xie drop_write: 4533905b0ddaSMiao Xie mnt_drop_write_file(file); 45345d13a37bSArne Jansen return ret; 45355d13a37bSArne Jansen } 45365d13a37bSArne Jansen 4537905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 45385d13a37bSArne Jansen { 45390b246afaSJeff Mahoney struct inode *inode = file_inode(file); 45400b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45410b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 45425d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 45435d13a37bSArne Jansen struct btrfs_trans_handle *trans; 45445d13a37bSArne Jansen int ret; 45455d13a37bSArne Jansen int err; 45465d13a37bSArne Jansen 45475d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 45485d13a37bSArne Jansen return -EPERM; 45495d13a37bSArne Jansen 4550905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4551905b0ddaSMiao Xie if (ret) 4552905b0ddaSMiao Xie return ret; 45535d13a37bSArne Jansen 45545d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4555905b0ddaSMiao Xie if (IS_ERR(sa)) { 4556905b0ddaSMiao Xie ret = PTR_ERR(sa); 4557905b0ddaSMiao Xie goto drop_write; 4558905b0ddaSMiao Xie } 45595d13a37bSArne Jansen 45605d13a37bSArne Jansen trans = btrfs_join_transaction(root); 45615d13a37bSArne Jansen if (IS_ERR(trans)) { 45625d13a37bSArne Jansen ret = PTR_ERR(trans); 45635d13a37bSArne Jansen goto out; 45645d13a37bSArne Jansen } 45655d13a37bSArne Jansen 45665d13a37bSArne Jansen if (sa->assign) { 45679f8a6ce6SLu Fengqi ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst); 45685d13a37bSArne Jansen } else { 456939616c27SLu Fengqi ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst); 45705d13a37bSArne Jansen } 45715d13a37bSArne Jansen 4572e082f563SQu Wenruo /* update qgroup status and info */ 4573280f8bd2SLu Fengqi err = btrfs_run_qgroups(trans); 4574e082f563SQu Wenruo if (err < 0) 45750b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, err, 4576ad8403dfSAnand Jain "failed to update qgroup status and info"); 45773a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 45785d13a37bSArne Jansen if (err && !ret) 45795d13a37bSArne Jansen ret = err; 45805d13a37bSArne Jansen 45815d13a37bSArne Jansen out: 45825d13a37bSArne Jansen kfree(sa); 4583905b0ddaSMiao Xie drop_write: 4584905b0ddaSMiao Xie mnt_drop_write_file(file); 45855d13a37bSArne Jansen return ret; 45865d13a37bSArne Jansen } 45875d13a37bSArne Jansen 4588905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 45895d13a37bSArne Jansen { 45900b246afaSJeff Mahoney struct inode *inode = file_inode(file); 45910b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 45925d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 45935d13a37bSArne Jansen struct btrfs_trans_handle *trans; 45945d13a37bSArne Jansen int ret; 45955d13a37bSArne Jansen int err; 45965d13a37bSArne Jansen 45975d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 45985d13a37bSArne Jansen return -EPERM; 45995d13a37bSArne Jansen 4600905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4601905b0ddaSMiao Xie if (ret) 4602905b0ddaSMiao Xie return ret; 46035d13a37bSArne Jansen 46045d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4605905b0ddaSMiao Xie if (IS_ERR(sa)) { 4606905b0ddaSMiao Xie ret = PTR_ERR(sa); 4607905b0ddaSMiao Xie goto drop_write; 4608905b0ddaSMiao Xie } 46095d13a37bSArne Jansen 4610d86e56cfSMiao Xie if (!sa->qgroupid) { 4611d86e56cfSMiao Xie ret = -EINVAL; 4612d86e56cfSMiao Xie goto out; 4613d86e56cfSMiao Xie } 4614d86e56cfSMiao Xie 46155d13a37bSArne Jansen trans = btrfs_join_transaction(root); 46165d13a37bSArne Jansen if (IS_ERR(trans)) { 46175d13a37bSArne Jansen ret = PTR_ERR(trans); 46185d13a37bSArne Jansen goto out; 46195d13a37bSArne Jansen } 46205d13a37bSArne Jansen 46215d13a37bSArne Jansen if (sa->create) { 462249a05ecdSLu Fengqi ret = btrfs_create_qgroup(trans, sa->qgroupid); 46235d13a37bSArne Jansen } else { 46243efbee1dSLu Fengqi ret = btrfs_remove_qgroup(trans, sa->qgroupid); 46255d13a37bSArne Jansen } 46265d13a37bSArne Jansen 46273a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 46285d13a37bSArne Jansen if (err && !ret) 46295d13a37bSArne Jansen ret = err; 46305d13a37bSArne Jansen 46315d13a37bSArne Jansen out: 46325d13a37bSArne Jansen kfree(sa); 4633905b0ddaSMiao Xie drop_write: 4634905b0ddaSMiao Xie mnt_drop_write_file(file); 46355d13a37bSArne Jansen return ret; 46365d13a37bSArne Jansen } 46375d13a37bSArne Jansen 4638905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 46395d13a37bSArne Jansen { 46400b246afaSJeff Mahoney struct inode *inode = file_inode(file); 46410b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 46425d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 46435d13a37bSArne Jansen struct btrfs_trans_handle *trans; 46445d13a37bSArne Jansen int ret; 46455d13a37bSArne Jansen int err; 46465d13a37bSArne Jansen u64 qgroupid; 46475d13a37bSArne Jansen 46485d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 46495d13a37bSArne Jansen return -EPERM; 46505d13a37bSArne Jansen 4651905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4652905b0ddaSMiao Xie if (ret) 4653905b0ddaSMiao Xie return ret; 46545d13a37bSArne Jansen 46555d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4656905b0ddaSMiao Xie if (IS_ERR(sa)) { 4657905b0ddaSMiao Xie ret = PTR_ERR(sa); 4658905b0ddaSMiao Xie goto drop_write; 4659905b0ddaSMiao Xie } 46605d13a37bSArne Jansen 46615d13a37bSArne Jansen trans = btrfs_join_transaction(root); 46625d13a37bSArne Jansen if (IS_ERR(trans)) { 46635d13a37bSArne Jansen ret = PTR_ERR(trans); 46645d13a37bSArne Jansen goto out; 46655d13a37bSArne Jansen } 46665d13a37bSArne Jansen 46675d13a37bSArne Jansen qgroupid = sa->qgroupid; 46685d13a37bSArne Jansen if (!qgroupid) { 46695d13a37bSArne Jansen /* take the current subvol as qgroup */ 46705d13a37bSArne Jansen qgroupid = root->root_key.objectid; 46715d13a37bSArne Jansen } 46725d13a37bSArne Jansen 4673f0042d5eSLu Fengqi ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim); 46745d13a37bSArne Jansen 46753a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 46765d13a37bSArne Jansen if (err && !ret) 46775d13a37bSArne Jansen ret = err; 46785d13a37bSArne Jansen 46795d13a37bSArne Jansen out: 46805d13a37bSArne Jansen kfree(sa); 4681905b0ddaSMiao Xie drop_write: 4682905b0ddaSMiao Xie mnt_drop_write_file(file); 46835d13a37bSArne Jansen return ret; 46845d13a37bSArne Jansen } 46855d13a37bSArne Jansen 46862f232036SJan Schmidt static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 46872f232036SJan Schmidt { 46880b246afaSJeff Mahoney struct inode *inode = file_inode(file); 46890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46902f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 46912f232036SJan Schmidt int ret; 46922f232036SJan Schmidt 46932f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 46942f232036SJan Schmidt return -EPERM; 46952f232036SJan Schmidt 46962f232036SJan Schmidt ret = mnt_want_write_file(file); 46972f232036SJan Schmidt if (ret) 46982f232036SJan Schmidt return ret; 46992f232036SJan Schmidt 47002f232036SJan Schmidt qsa = memdup_user(arg, sizeof(*qsa)); 47012f232036SJan Schmidt if (IS_ERR(qsa)) { 47022f232036SJan Schmidt ret = PTR_ERR(qsa); 47032f232036SJan Schmidt goto drop_write; 47042f232036SJan Schmidt } 47052f232036SJan Schmidt 47062f232036SJan Schmidt if (qsa->flags) { 47072f232036SJan Schmidt ret = -EINVAL; 47082f232036SJan Schmidt goto out; 47092f232036SJan Schmidt } 47102f232036SJan Schmidt 47110b246afaSJeff Mahoney ret = btrfs_qgroup_rescan(fs_info); 47122f232036SJan Schmidt 47132f232036SJan Schmidt out: 47142f232036SJan Schmidt kfree(qsa); 47152f232036SJan Schmidt drop_write: 47162f232036SJan Schmidt mnt_drop_write_file(file); 47172f232036SJan Schmidt return ret; 47182f232036SJan Schmidt } 47192f232036SJan Schmidt 4720b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info, 4721b929c1d8SMarcos Paulo de Souza void __user *arg) 47222f232036SJan Schmidt { 47230afb603aSGoldwyn Rodrigues struct btrfs_ioctl_quota_rescan_args qsa = {0}; 47242f232036SJan Schmidt 47252f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 47262f232036SJan Schmidt return -EPERM; 47272f232036SJan Schmidt 47280b246afaSJeff Mahoney if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 47290afb603aSGoldwyn Rodrigues qsa.flags = 1; 47300afb603aSGoldwyn Rodrigues qsa.progress = fs_info->qgroup_rescan_progress.objectid; 47312f232036SJan Schmidt } 47322f232036SJan Schmidt 47330afb603aSGoldwyn Rodrigues if (copy_to_user(arg, &qsa, sizeof(qsa))) 4734991a3daeSAnand Jain return -EFAULT; 47352f232036SJan Schmidt 4736991a3daeSAnand Jain return 0; 47372f232036SJan Schmidt } 47382f232036SJan Schmidt 4739b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info, 4740b929c1d8SMarcos Paulo de Souza void __user *arg) 474157254b6eSJan Schmidt { 474257254b6eSJan Schmidt if (!capable(CAP_SYS_ADMIN)) 474357254b6eSJan Schmidt return -EPERM; 474457254b6eSJan Schmidt 47450b246afaSJeff Mahoney return btrfs_qgroup_wait_for_completion(fs_info, true); 474657254b6eSJan Schmidt } 474757254b6eSJan Schmidt 4748abccd00fSHugo Mills static long _btrfs_ioctl_set_received_subvol(struct file *file, 4749e4fed17aSChristian Brauner struct user_namespace *mnt_userns, 4750abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa) 47518ea05e3aSAlexander Block { 4752496ad9aaSAl Viro struct inode *inode = file_inode(file); 47530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47548ea05e3aSAlexander Block struct btrfs_root *root = BTRFS_I(inode)->root; 47558ea05e3aSAlexander Block struct btrfs_root_item *root_item = &root->root_item; 47568ea05e3aSAlexander Block struct btrfs_trans_handle *trans; 475795582b00SDeepa Dinamani struct timespec64 ct = current_time(inode); 47588ea05e3aSAlexander Block int ret = 0; 4759dd5f9615SStefan Behrens int received_uuid_changed; 47608ea05e3aSAlexander Block 4761e4fed17aSChristian Brauner if (!inode_owner_or_capable(mnt_userns, inode)) 4762bd60ea0fSDavid Sterba return -EPERM; 4763bd60ea0fSDavid Sterba 47648ea05e3aSAlexander Block ret = mnt_want_write_file(file); 47658ea05e3aSAlexander Block if (ret < 0) 47668ea05e3aSAlexander Block return ret; 47678ea05e3aSAlexander Block 47680b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 47698ea05e3aSAlexander Block 47704a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 47718ea05e3aSAlexander Block ret = -EINVAL; 47728ea05e3aSAlexander Block goto out; 47738ea05e3aSAlexander Block } 47748ea05e3aSAlexander Block 47758ea05e3aSAlexander Block if (btrfs_root_readonly(root)) { 47768ea05e3aSAlexander Block ret = -EROFS; 47778ea05e3aSAlexander Block goto out; 47788ea05e3aSAlexander Block } 47798ea05e3aSAlexander Block 4780dd5f9615SStefan Behrens /* 4781dd5f9615SStefan Behrens * 1 - root item 4782dd5f9615SStefan Behrens * 2 - uuid items (received uuid + subvol uuid) 4783dd5f9615SStefan Behrens */ 4784dd5f9615SStefan Behrens trans = btrfs_start_transaction(root, 3); 47858ea05e3aSAlexander Block if (IS_ERR(trans)) { 47868ea05e3aSAlexander Block ret = PTR_ERR(trans); 47878ea05e3aSAlexander Block trans = NULL; 47888ea05e3aSAlexander Block goto out; 47898ea05e3aSAlexander Block } 47908ea05e3aSAlexander Block 47918ea05e3aSAlexander Block sa->rtransid = trans->transid; 47928ea05e3aSAlexander Block sa->rtime.sec = ct.tv_sec; 47938ea05e3aSAlexander Block sa->rtime.nsec = ct.tv_nsec; 47948ea05e3aSAlexander Block 4795dd5f9615SStefan Behrens received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 4796dd5f9615SStefan Behrens BTRFS_UUID_SIZE); 4797dd5f9615SStefan Behrens if (received_uuid_changed && 4798d87ff758SNikolay Borisov !btrfs_is_empty_uuid(root_item->received_uuid)) { 4799d1957791SLu Fengqi ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid, 4800dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4801dd5f9615SStefan Behrens root->root_key.objectid); 4802d87ff758SNikolay Borisov if (ret && ret != -ENOENT) { 4803d87ff758SNikolay Borisov btrfs_abort_transaction(trans, ret); 4804d87ff758SNikolay Borisov btrfs_end_transaction(trans); 4805d87ff758SNikolay Borisov goto out; 4806d87ff758SNikolay Borisov } 4807d87ff758SNikolay Borisov } 48088ea05e3aSAlexander Block memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 48098ea05e3aSAlexander Block btrfs_set_root_stransid(root_item, sa->stransid); 48108ea05e3aSAlexander Block btrfs_set_root_rtransid(root_item, sa->rtransid); 48113cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 48123cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 48133cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 48143cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 48158ea05e3aSAlexander Block 48160b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 48178ea05e3aSAlexander Block &root->root_key, &root->root_item); 48188ea05e3aSAlexander Block if (ret < 0) { 48193a45bb20SJeff Mahoney btrfs_end_transaction(trans); 48208ea05e3aSAlexander Block goto out; 4821dd5f9615SStefan Behrens } 4822dd5f9615SStefan Behrens if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 4823cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, sa->uuid, 4824dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4825dd5f9615SStefan Behrens root->root_key.objectid); 4826dd5f9615SStefan Behrens if (ret < 0 && ret != -EEXIST) { 482766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 4828efd38150SNikolay Borisov btrfs_end_transaction(trans); 4829dd5f9615SStefan Behrens goto out; 4830dd5f9615SStefan Behrens } 4831dd5f9615SStefan Behrens } 48323a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 4833abccd00fSHugo Mills out: 48340b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4835abccd00fSHugo Mills mnt_drop_write_file(file); 4836abccd00fSHugo Mills return ret; 4837abccd00fSHugo Mills } 4838abccd00fSHugo Mills 4839abccd00fSHugo Mills #ifdef CONFIG_64BIT 4840abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol_32(struct file *file, 4841abccd00fSHugo Mills void __user *arg) 4842abccd00fSHugo Mills { 4843abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL; 4844abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *args64 = NULL; 4845abccd00fSHugo Mills int ret = 0; 4846abccd00fSHugo Mills 4847abccd00fSHugo Mills args32 = memdup_user(arg, sizeof(*args32)); 48487b9ea627SShailendra Verma if (IS_ERR(args32)) 48497b9ea627SShailendra Verma return PTR_ERR(args32); 4850abccd00fSHugo Mills 48518d2db785SDavid Sterba args64 = kmalloc(sizeof(*args64), GFP_KERNEL); 485284dbeb87SDan Carpenter if (!args64) { 485384dbeb87SDan Carpenter ret = -ENOMEM; 4854abccd00fSHugo Mills goto out; 4855abccd00fSHugo Mills } 4856abccd00fSHugo Mills 4857abccd00fSHugo Mills memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE); 4858abccd00fSHugo Mills args64->stransid = args32->stransid; 4859abccd00fSHugo Mills args64->rtransid = args32->rtransid; 4860abccd00fSHugo Mills args64->stime.sec = args32->stime.sec; 4861abccd00fSHugo Mills args64->stime.nsec = args32->stime.nsec; 4862abccd00fSHugo Mills args64->rtime.sec = args32->rtime.sec; 4863abccd00fSHugo Mills args64->rtime.nsec = args32->rtime.nsec; 4864abccd00fSHugo Mills args64->flags = args32->flags; 4865abccd00fSHugo Mills 4866e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64); 4867abccd00fSHugo Mills if (ret) 4868abccd00fSHugo Mills goto out; 4869abccd00fSHugo Mills 4870abccd00fSHugo Mills memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE); 4871abccd00fSHugo Mills args32->stransid = args64->stransid; 4872abccd00fSHugo Mills args32->rtransid = args64->rtransid; 4873abccd00fSHugo Mills args32->stime.sec = args64->stime.sec; 4874abccd00fSHugo Mills args32->stime.nsec = args64->stime.nsec; 4875abccd00fSHugo Mills args32->rtime.sec = args64->rtime.sec; 4876abccd00fSHugo Mills args32->rtime.nsec = args64->rtime.nsec; 4877abccd00fSHugo Mills args32->flags = args64->flags; 4878abccd00fSHugo Mills 4879abccd00fSHugo Mills ret = copy_to_user(arg, args32, sizeof(*args32)); 4880abccd00fSHugo Mills if (ret) 4881abccd00fSHugo Mills ret = -EFAULT; 4882abccd00fSHugo Mills 4883abccd00fSHugo Mills out: 4884abccd00fSHugo Mills kfree(args32); 4885abccd00fSHugo Mills kfree(args64); 4886abccd00fSHugo Mills return ret; 4887abccd00fSHugo Mills } 4888abccd00fSHugo Mills #endif 4889abccd00fSHugo Mills 4890abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol(struct file *file, 4891abccd00fSHugo Mills void __user *arg) 4892abccd00fSHugo Mills { 4893abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa = NULL; 4894abccd00fSHugo Mills int ret = 0; 4895abccd00fSHugo Mills 4896abccd00fSHugo Mills sa = memdup_user(arg, sizeof(*sa)); 48977b9ea627SShailendra Verma if (IS_ERR(sa)) 48987b9ea627SShailendra Verma return PTR_ERR(sa); 4899abccd00fSHugo Mills 4900e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa); 4901abccd00fSHugo Mills 4902abccd00fSHugo Mills if (ret) 4903abccd00fSHugo Mills goto out; 4904abccd00fSHugo Mills 49058ea05e3aSAlexander Block ret = copy_to_user(arg, sa, sizeof(*sa)); 49068ea05e3aSAlexander Block if (ret) 49078ea05e3aSAlexander Block ret = -EFAULT; 49088ea05e3aSAlexander Block 49098ea05e3aSAlexander Block out: 49108ea05e3aSAlexander Block kfree(sa); 49118ea05e3aSAlexander Block return ret; 49128ea05e3aSAlexander Block } 49138ea05e3aSAlexander Block 4914b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info, 4915b929c1d8SMarcos Paulo de Souza void __user *arg) 4916867ab667Sjeff.liu { 4917a1b83ac5SAnand Jain size_t len; 4918867ab667Sjeff.liu int ret; 4919a1b83ac5SAnand Jain char label[BTRFS_LABEL_SIZE]; 4920a1b83ac5SAnand Jain 49210b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 49220b246afaSJeff Mahoney memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE); 49230b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 4924a1b83ac5SAnand Jain 4925a1b83ac5SAnand Jain len = strnlen(label, BTRFS_LABEL_SIZE); 4926867ab667Sjeff.liu 4927867ab667Sjeff.liu if (len == BTRFS_LABEL_SIZE) { 49280b246afaSJeff Mahoney btrfs_warn(fs_info, 49290b246afaSJeff Mahoney "label is too long, return the first %zu bytes", 49300b246afaSJeff Mahoney --len); 4931867ab667Sjeff.liu } 4932867ab667Sjeff.liu 4933867ab667Sjeff.liu ret = copy_to_user(arg, label, len); 4934867ab667Sjeff.liu 4935867ab667Sjeff.liu return ret ? -EFAULT : 0; 4936867ab667Sjeff.liu } 4937867ab667Sjeff.liu 4938a8bfd4abSjeff.liu static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 4939a8bfd4abSjeff.liu { 49400b246afaSJeff Mahoney struct inode *inode = file_inode(file); 49410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 49420b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 49430b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 4944a8bfd4abSjeff.liu struct btrfs_trans_handle *trans; 4945a8bfd4abSjeff.liu char label[BTRFS_LABEL_SIZE]; 4946a8bfd4abSjeff.liu int ret; 4947a8bfd4abSjeff.liu 4948a8bfd4abSjeff.liu if (!capable(CAP_SYS_ADMIN)) 4949a8bfd4abSjeff.liu return -EPERM; 4950a8bfd4abSjeff.liu 4951a8bfd4abSjeff.liu if (copy_from_user(label, arg, sizeof(label))) 4952a8bfd4abSjeff.liu return -EFAULT; 4953a8bfd4abSjeff.liu 4954a8bfd4abSjeff.liu if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 49550b246afaSJeff Mahoney btrfs_err(fs_info, 49565d163e0eSJeff Mahoney "unable to set label with more than %d bytes", 4957a8bfd4abSjeff.liu BTRFS_LABEL_SIZE - 1); 4958a8bfd4abSjeff.liu return -EINVAL; 4959a8bfd4abSjeff.liu } 4960a8bfd4abSjeff.liu 4961a8bfd4abSjeff.liu ret = mnt_want_write_file(file); 4962a8bfd4abSjeff.liu if (ret) 4963a8bfd4abSjeff.liu return ret; 4964a8bfd4abSjeff.liu 4965a8bfd4abSjeff.liu trans = btrfs_start_transaction(root, 0); 4966a8bfd4abSjeff.liu if (IS_ERR(trans)) { 4967a8bfd4abSjeff.liu ret = PTR_ERR(trans); 4968a8bfd4abSjeff.liu goto out_unlock; 4969a8bfd4abSjeff.liu } 4970a8bfd4abSjeff.liu 49710b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 4972a8bfd4abSjeff.liu strcpy(super_block->label, label); 49730b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 49743a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 4975a8bfd4abSjeff.liu 4976a8bfd4abSjeff.liu out_unlock: 4977a8bfd4abSjeff.liu mnt_drop_write_file(file); 4978a8bfd4abSjeff.liu return ret; 4979a8bfd4abSjeff.liu } 4980a8bfd4abSjeff.liu 49812eaa055fSJeff Mahoney #define INIT_FEATURE_FLAGS(suffix) \ 49822eaa055fSJeff Mahoney { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 49832eaa055fSJeff Mahoney .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 49842eaa055fSJeff Mahoney .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 49852eaa055fSJeff Mahoney 4986d5131b65SDavid Sterba int btrfs_ioctl_get_supported_features(void __user *arg) 49872eaa055fSJeff Mahoney { 49884d4ab6d6SDavid Sterba static const struct btrfs_ioctl_feature_flags features[3] = { 49892eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SUPP), 49902eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_SET), 49912eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_CLEAR) 49922eaa055fSJeff Mahoney }; 49932eaa055fSJeff Mahoney 49942eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 49952eaa055fSJeff Mahoney return -EFAULT; 49962eaa055fSJeff Mahoney 49972eaa055fSJeff Mahoney return 0; 49982eaa055fSJeff Mahoney } 49992eaa055fSJeff Mahoney 5000b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info, 5001b929c1d8SMarcos Paulo de Souza void __user *arg) 50022eaa055fSJeff Mahoney { 50030b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 50042eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags features; 50052eaa055fSJeff Mahoney 50062eaa055fSJeff Mahoney features.compat_flags = btrfs_super_compat_flags(super_block); 50072eaa055fSJeff Mahoney features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 50082eaa055fSJeff Mahoney features.incompat_flags = btrfs_super_incompat_flags(super_block); 50092eaa055fSJeff Mahoney 50102eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 50112eaa055fSJeff Mahoney return -EFAULT; 50122eaa055fSJeff Mahoney 50132eaa055fSJeff Mahoney return 0; 50142eaa055fSJeff Mahoney } 50152eaa055fSJeff Mahoney 50162ff7e61eSJeff Mahoney static int check_feature_bits(struct btrfs_fs_info *fs_info, 50173b02a68aSJeff Mahoney enum btrfs_feature_set set, 50182eaa055fSJeff Mahoney u64 change_mask, u64 flags, u64 supported_flags, 50192eaa055fSJeff Mahoney u64 safe_set, u64 safe_clear) 50202eaa055fSJeff Mahoney { 5021f10152bcSDavid Sterba const char *type = btrfs_feature_set_name(set); 50223b02a68aSJeff Mahoney char *names; 50232eaa055fSJeff Mahoney u64 disallowed, unsupported; 50242eaa055fSJeff Mahoney u64 set_mask = flags & change_mask; 50252eaa055fSJeff Mahoney u64 clear_mask = ~flags & change_mask; 50262eaa055fSJeff Mahoney 50272eaa055fSJeff Mahoney unsupported = set_mask & ~supported_flags; 50282eaa055fSJeff Mahoney if (unsupported) { 50293b02a68aSJeff Mahoney names = btrfs_printable_features(set, unsupported); 50303b02a68aSJeff Mahoney if (names) { 50310b246afaSJeff Mahoney btrfs_warn(fs_info, 50323b02a68aSJeff Mahoney "this kernel does not support the %s feature bit%s", 50333b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50343b02a68aSJeff Mahoney kfree(names); 50353b02a68aSJeff Mahoney } else 50360b246afaSJeff Mahoney btrfs_warn(fs_info, 50372eaa055fSJeff Mahoney "this kernel does not support %s bits 0x%llx", 50382eaa055fSJeff Mahoney type, unsupported); 50392eaa055fSJeff Mahoney return -EOPNOTSUPP; 50402eaa055fSJeff Mahoney } 50412eaa055fSJeff Mahoney 50422eaa055fSJeff Mahoney disallowed = set_mask & ~safe_set; 50432eaa055fSJeff Mahoney if (disallowed) { 50443b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 50453b02a68aSJeff Mahoney if (names) { 50460b246afaSJeff Mahoney btrfs_warn(fs_info, 50473b02a68aSJeff Mahoney "can't set the %s feature bit%s while mounted", 50483b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50493b02a68aSJeff Mahoney kfree(names); 50503b02a68aSJeff Mahoney } else 50510b246afaSJeff Mahoney btrfs_warn(fs_info, 50522eaa055fSJeff Mahoney "can't set %s bits 0x%llx while mounted", 50532eaa055fSJeff Mahoney type, disallowed); 50542eaa055fSJeff Mahoney return -EPERM; 50552eaa055fSJeff Mahoney } 50562eaa055fSJeff Mahoney 50572eaa055fSJeff Mahoney disallowed = clear_mask & ~safe_clear; 50582eaa055fSJeff Mahoney if (disallowed) { 50593b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 50603b02a68aSJeff Mahoney if (names) { 50610b246afaSJeff Mahoney btrfs_warn(fs_info, 50623b02a68aSJeff Mahoney "can't clear the %s feature bit%s while mounted", 50633b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 50643b02a68aSJeff Mahoney kfree(names); 50653b02a68aSJeff Mahoney } else 50660b246afaSJeff Mahoney btrfs_warn(fs_info, 50672eaa055fSJeff Mahoney "can't clear %s bits 0x%llx while mounted", 50682eaa055fSJeff Mahoney type, disallowed); 50692eaa055fSJeff Mahoney return -EPERM; 50702eaa055fSJeff Mahoney } 50712eaa055fSJeff Mahoney 50722eaa055fSJeff Mahoney return 0; 50732eaa055fSJeff Mahoney } 50742eaa055fSJeff Mahoney 50752ff7e61eSJeff Mahoney #define check_feature(fs_info, change_mask, flags, mask_base) \ 50762ff7e61eSJeff Mahoney check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags, \ 50772eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 50782eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 50792eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 50802eaa055fSJeff Mahoney 50812eaa055fSJeff Mahoney static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 50822eaa055fSJeff Mahoney { 50830b246afaSJeff Mahoney struct inode *inode = file_inode(file); 50840b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 50850b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 50860b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 50872eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags flags[2]; 50882eaa055fSJeff Mahoney struct btrfs_trans_handle *trans; 50892eaa055fSJeff Mahoney u64 newflags; 50902eaa055fSJeff Mahoney int ret; 50912eaa055fSJeff Mahoney 50922eaa055fSJeff Mahoney if (!capable(CAP_SYS_ADMIN)) 50932eaa055fSJeff Mahoney return -EPERM; 50942eaa055fSJeff Mahoney 50952eaa055fSJeff Mahoney if (copy_from_user(flags, arg, sizeof(flags))) 50962eaa055fSJeff Mahoney return -EFAULT; 50972eaa055fSJeff Mahoney 50982eaa055fSJeff Mahoney /* Nothing to do */ 50992eaa055fSJeff Mahoney if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 51002eaa055fSJeff Mahoney !flags[0].incompat_flags) 51012eaa055fSJeff Mahoney return 0; 51022eaa055fSJeff Mahoney 51032ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_flags, 51042eaa055fSJeff Mahoney flags[1].compat_flags, COMPAT); 51052eaa055fSJeff Mahoney if (ret) 51062eaa055fSJeff Mahoney return ret; 51072eaa055fSJeff Mahoney 51082ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_ro_flags, 51092eaa055fSJeff Mahoney flags[1].compat_ro_flags, COMPAT_RO); 51102eaa055fSJeff Mahoney if (ret) 51112eaa055fSJeff Mahoney return ret; 51122eaa055fSJeff Mahoney 51132ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].incompat_flags, 51142eaa055fSJeff Mahoney flags[1].incompat_flags, INCOMPAT); 51152eaa055fSJeff Mahoney if (ret) 51162eaa055fSJeff Mahoney return ret; 51172eaa055fSJeff Mahoney 51187ab19625SDavid Sterba ret = mnt_want_write_file(file); 51197ab19625SDavid Sterba if (ret) 51207ab19625SDavid Sterba return ret; 51217ab19625SDavid Sterba 51228051aa1aSDavid Sterba trans = btrfs_start_transaction(root, 0); 51237ab19625SDavid Sterba if (IS_ERR(trans)) { 51247ab19625SDavid Sterba ret = PTR_ERR(trans); 51257ab19625SDavid Sterba goto out_drop_write; 51267ab19625SDavid Sterba } 51272eaa055fSJeff Mahoney 51280b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 51292eaa055fSJeff Mahoney newflags = btrfs_super_compat_flags(super_block); 51302eaa055fSJeff Mahoney newflags |= flags[0].compat_flags & flags[1].compat_flags; 51312eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 51322eaa055fSJeff Mahoney btrfs_set_super_compat_flags(super_block, newflags); 51332eaa055fSJeff Mahoney 51342eaa055fSJeff Mahoney newflags = btrfs_super_compat_ro_flags(super_block); 51352eaa055fSJeff Mahoney newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 51362eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 51372eaa055fSJeff Mahoney btrfs_set_super_compat_ro_flags(super_block, newflags); 51382eaa055fSJeff Mahoney 51392eaa055fSJeff Mahoney newflags = btrfs_super_incompat_flags(super_block); 51402eaa055fSJeff Mahoney newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 51412eaa055fSJeff Mahoney newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 51422eaa055fSJeff Mahoney btrfs_set_super_incompat_flags(super_block, newflags); 51430b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 51442eaa055fSJeff Mahoney 51453a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 51467ab19625SDavid Sterba out_drop_write: 51477ab19625SDavid Sterba mnt_drop_write_file(file); 51487ab19625SDavid Sterba 51497ab19625SDavid Sterba return ret; 51502eaa055fSJeff Mahoney } 51512eaa055fSJeff Mahoney 51529ad12305SSahil Kang static int _btrfs_ioctl_send(struct inode *inode, void __user *argp, bool compat) 51532351f431SJosef Bacik { 51542351f431SJosef Bacik struct btrfs_ioctl_send_args *arg; 51552351f431SJosef Bacik int ret; 51562351f431SJosef Bacik 51572351f431SJosef Bacik if (compat) { 51582351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 51592351f431SJosef Bacik struct btrfs_ioctl_send_args_32 args32; 51602351f431SJosef Bacik 51612351f431SJosef Bacik ret = copy_from_user(&args32, argp, sizeof(args32)); 51622351f431SJosef Bacik if (ret) 51632351f431SJosef Bacik return -EFAULT; 51642351f431SJosef Bacik arg = kzalloc(sizeof(*arg), GFP_KERNEL); 51652351f431SJosef Bacik if (!arg) 51662351f431SJosef Bacik return -ENOMEM; 51672351f431SJosef Bacik arg->send_fd = args32.send_fd; 51682351f431SJosef Bacik arg->clone_sources_count = args32.clone_sources_count; 51692351f431SJosef Bacik arg->clone_sources = compat_ptr(args32.clone_sources); 51702351f431SJosef Bacik arg->parent_root = args32.parent_root; 51712351f431SJosef Bacik arg->flags = args32.flags; 51722351f431SJosef Bacik memcpy(arg->reserved, args32.reserved, 51732351f431SJosef Bacik sizeof(args32.reserved)); 51742351f431SJosef Bacik #else 51752351f431SJosef Bacik return -ENOTTY; 51762351f431SJosef Bacik #endif 51772351f431SJosef Bacik } else { 51782351f431SJosef Bacik arg = memdup_user(argp, sizeof(*arg)); 51792351f431SJosef Bacik if (IS_ERR(arg)) 51802351f431SJosef Bacik return PTR_ERR(arg); 51812351f431SJosef Bacik } 51829ad12305SSahil Kang ret = btrfs_ioctl_send(inode, arg); 51832351f431SJosef Bacik kfree(arg); 51842351f431SJosef Bacik return ret; 51852351f431SJosef Bacik } 51862351f431SJosef Bacik 5187f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 5188f46b5a66SChristoph Hellwig cmd, unsigned long arg) 5189f46b5a66SChristoph Hellwig { 51900b246afaSJeff Mahoney struct inode *inode = file_inode(file); 51910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 51920b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 51934bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 5194f46b5a66SChristoph Hellwig 5195f46b5a66SChristoph Hellwig switch (cmd) { 51966cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 5197dc408ccdSSahil Kang return btrfs_ioctl_getversion(inode, argp); 519840cf931fSEric Sandeen case FS_IOC_GETFSLABEL: 5199b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_fslabel(fs_info, argp); 520040cf931fSEric Sandeen case FS_IOC_SETFSLABEL: 520140cf931fSEric Sandeen return btrfs_ioctl_set_fslabel(file, argp); 5202f7039b1dSLi Dongyang case FITRIM: 5203b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_fitrim(fs_info, argp); 5204f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 5205fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 5206fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 5207fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 52083de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 5209fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 52106f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 52116f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 521276dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 5213949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, false); 5214949964c9SMarcos Paulo de Souza case BTRFS_IOC_SNAP_DESTROY_V2: 5215949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, true); 52160caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 52179ad12305SSahil Kang return btrfs_ioctl_subvol_getflags(inode, argp); 52180caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 52190caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 52206ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 52216ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 5222f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 52231e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 52241e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 52251e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 5226f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 5227198605a8SMiao Xie return btrfs_ioctl_resize(file, argp); 5228f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 52292ff7e61eSJeff Mahoney return btrfs_ioctl_add_dev(fs_info, argp); 5230f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 5231da24927bSMiao Xie return btrfs_ioctl_rm_dev(file, argp); 52326b526ed7SAnand Jain case BTRFS_IOC_RM_DEV_V2: 52336b526ed7SAnand Jain return btrfs_ioctl_rm_dev_v2(file, argp); 5234475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 52352ff7e61eSJeff Mahoney return btrfs_ioctl_fs_info(fs_info, argp); 5236475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 52372ff7e61eSJeff Mahoney return btrfs_ioctl_dev_info(fs_info, argp); 5238f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 52399ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, NULL); 5240ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 52419ad12305SSahil Kang return btrfs_ioctl_tree_search(inode, argp); 5242cc68a8a5SGerhard Heift case BTRFS_IOC_TREE_SEARCH_V2: 52439ad12305SSahil Kang return btrfs_ioctl_tree_search_v2(inode, argp); 5244ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 5245dc408ccdSSahil Kang return btrfs_ioctl_ino_lookup(root, argp); 5246d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 5247d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 5248d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 5249d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 1); 5250d24a67b2SZygo Blaxell case BTRFS_IOC_LOGICAL_INO_V2: 5251d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 2); 52521406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 52532ff7e61eSJeff Mahoney return btrfs_ioctl_space_info(fs_info, argp); 52549b199859SFilipe David Borba Manana case BTRFS_IOC_SYNC: { 52559b199859SFilipe David Borba Manana int ret; 52569b199859SFilipe David Borba Manana 52579db4dc24SNikolay Borisov ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); 52589b199859SFilipe David Borba Manana if (ret) 52599b199859SFilipe David Borba Manana return ret; 52600b246afaSJeff Mahoney ret = btrfs_sync_fs(inode->i_sb, 1); 52612fad4e83SDavid Sterba /* 52622fad4e83SDavid Sterba * The transaction thread may want to do more work, 526301327610SNicholas D Steeves * namely it pokes the cleaner kthread that will start 52642fad4e83SDavid Sterba * processing uncleaned subvols. 52652fad4e83SDavid Sterba */ 52660b246afaSJeff Mahoney wake_up_process(fs_info->transaction_kthread); 52679b199859SFilipe David Borba Manana return ret; 52689b199859SFilipe David Borba Manana } 526946204592SSage Weil case BTRFS_IOC_START_SYNC: 52709a8c28beSMiao Xie return btrfs_ioctl_start_sync(root, argp); 527146204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 52722ff7e61eSJeff Mahoney return btrfs_ioctl_wait_sync(fs_info, argp); 5273475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 5274b8e95489SMiao Xie return btrfs_ioctl_scrub(file, argp); 5275475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 52762ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_cancel(fs_info); 5277475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 52782ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_progress(fs_info, argp); 5279c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 52809ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 5281837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 52822ff7e61eSJeff Mahoney return btrfs_ioctl_balance_ctl(fs_info, arg); 528319a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 52842ff7e61eSJeff Mahoney return btrfs_ioctl_balance_progress(fs_info, argp); 52858ea05e3aSAlexander Block case BTRFS_IOC_SET_RECEIVED_SUBVOL: 52868ea05e3aSAlexander Block return btrfs_ioctl_set_received_subvol(file, argp); 5287abccd00fSHugo Mills #ifdef CONFIG_64BIT 5288abccd00fSHugo Mills case BTRFS_IOC_SET_RECEIVED_SUBVOL_32: 5289abccd00fSHugo Mills return btrfs_ioctl_set_received_subvol_32(file, argp); 5290abccd00fSHugo Mills #endif 529131db9f7cSAlexander Block case BTRFS_IOC_SEND: 52929ad12305SSahil Kang return _btrfs_ioctl_send(inode, argp, false); 52932351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 52942351f431SJosef Bacik case BTRFS_IOC_SEND_32: 52959ad12305SSahil Kang return _btrfs_ioctl_send(inode, argp, true); 52962351f431SJosef Bacik #endif 5297c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 52982ff7e61eSJeff Mahoney return btrfs_ioctl_get_dev_stats(fs_info, argp); 52995d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 5300905b0ddaSMiao Xie return btrfs_ioctl_quota_ctl(file, argp); 53015d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 5302905b0ddaSMiao Xie return btrfs_ioctl_qgroup_assign(file, argp); 53035d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 5304905b0ddaSMiao Xie return btrfs_ioctl_qgroup_create(file, argp); 53055d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 5306905b0ddaSMiao Xie return btrfs_ioctl_qgroup_limit(file, argp); 53072f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN: 53082f232036SJan Schmidt return btrfs_ioctl_quota_rescan(file, argp); 53092f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_STATUS: 5310b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_status(fs_info, argp); 531157254b6eSJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_WAIT: 5312b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_wait(fs_info, argp); 53133f6bcfbdSStefan Behrens case BTRFS_IOC_DEV_REPLACE: 53142ff7e61eSJeff Mahoney return btrfs_ioctl_dev_replace(fs_info, argp); 53152eaa055fSJeff Mahoney case BTRFS_IOC_GET_SUPPORTED_FEATURES: 5316d5131b65SDavid Sterba return btrfs_ioctl_get_supported_features(argp); 53172eaa055fSJeff Mahoney case BTRFS_IOC_GET_FEATURES: 5318b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_features(fs_info, argp); 53192eaa055fSJeff Mahoney case BTRFS_IOC_SET_FEATURES: 53202eaa055fSJeff Mahoney return btrfs_ioctl_set_features(file, argp); 5321b64ec075STomohiro Misono case BTRFS_IOC_GET_SUBVOL_INFO: 53229ad12305SSahil Kang return btrfs_ioctl_get_subvol_info(inode, argp); 532342e4b520STomohiro Misono case BTRFS_IOC_GET_SUBVOL_ROOTREF: 5324dc408ccdSSahil Kang return btrfs_ioctl_get_subvol_rootref(root, argp); 532523d0b79dSTomohiro Misono case BTRFS_IOC_INO_LOOKUP_USER: 532623d0b79dSTomohiro Misono return btrfs_ioctl_ino_lookup_user(file, argp); 532714605409SBoris Burkov case FS_IOC_ENABLE_VERITY: 532814605409SBoris Burkov return fsverity_ioctl_enable(file, (const void __user *)argp); 532914605409SBoris Burkov case FS_IOC_MEASURE_VERITY: 533014605409SBoris Burkov return fsverity_ioctl_measure(file, argp); 5331f46b5a66SChristoph Hellwig } 5332f46b5a66SChristoph Hellwig 5333f46b5a66SChristoph Hellwig return -ENOTTY; 5334f46b5a66SChristoph Hellwig } 53354c63c245SLuke Dashjr 53364c63c245SLuke Dashjr #ifdef CONFIG_COMPAT 53374c63c245SLuke Dashjr long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 53384c63c245SLuke Dashjr { 53392a362249SJeff Mahoney /* 53402a362249SJeff Mahoney * These all access 32-bit values anyway so no further 53412a362249SJeff Mahoney * handling is necessary. 53422a362249SJeff Mahoney */ 53434c63c245SLuke Dashjr switch (cmd) { 53444c63c245SLuke Dashjr case FS_IOC32_GETVERSION: 53454c63c245SLuke Dashjr cmd = FS_IOC_GETVERSION; 53464c63c245SLuke Dashjr break; 53474c63c245SLuke Dashjr } 53484c63c245SLuke Dashjr 53494c63c245SLuke Dashjr return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 53504c63c245SLuke Dashjr } 53514c63c245SLuke Dashjr #endif 5352