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 */ 85*e77fbf99SDavid Sterba __u32 version; /* in */ 86*e77fbf99SDavid 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 390578bda9eSDavid Sterba * - not BTRFS_EXCLOP_NONE - this is intentionally incompatible and the caller 391578bda9eSDavid Sterba * must check the condition first that would allow none -> @type 392578bda9eSDavid Sterba */ 393578bda9eSDavid Sterba bool btrfs_exclop_start_try_lock(struct btrfs_fs_info *fs_info, 394578bda9eSDavid Sterba enum btrfs_exclusive_operation type) 395578bda9eSDavid Sterba { 396578bda9eSDavid Sterba spin_lock(&fs_info->super_lock); 397578bda9eSDavid Sterba if (fs_info->exclusive_operation == type) 398578bda9eSDavid Sterba return true; 399578bda9eSDavid Sterba 400578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 401578bda9eSDavid Sterba return false; 402578bda9eSDavid Sterba } 403578bda9eSDavid Sterba 404578bda9eSDavid Sterba void btrfs_exclop_start_unlock(struct btrfs_fs_info *fs_info) 405578bda9eSDavid Sterba { 406578bda9eSDavid Sterba spin_unlock(&fs_info->super_lock); 407578bda9eSDavid Sterba } 408578bda9eSDavid Sterba 409c3e1f96cSGoldwyn Rodrigues void btrfs_exclop_finish(struct btrfs_fs_info *fs_info) 410c3e1f96cSGoldwyn Rodrigues { 4110d7ed32cSDavid Sterba spin_lock(&fs_info->super_lock); 412c3e1f96cSGoldwyn Rodrigues WRITE_ONCE(fs_info->exclusive_operation, BTRFS_EXCLOP_NONE); 4130d7ed32cSDavid Sterba spin_unlock(&fs_info->super_lock); 41466a2823cSGoldwyn Rodrigues sysfs_notify(&fs_info->fs_devices->fsid_kobj, NULL, "exclusive_operation"); 415c3e1f96cSGoldwyn Rodrigues } 416c3e1f96cSGoldwyn Rodrigues 4176cbff00fSChristoph Hellwig static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 4186cbff00fSChristoph Hellwig { 419496ad9aaSAl Viro struct inode *inode = file_inode(file); 4206cbff00fSChristoph Hellwig 4216cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 4226cbff00fSChristoph Hellwig } 423f46b5a66SChristoph Hellwig 424b929c1d8SMarcos Paulo de Souza static noinline int btrfs_ioctl_fitrim(struct btrfs_fs_info *fs_info, 425b929c1d8SMarcos Paulo de Souza void __user *arg) 426f7039b1dSLi Dongyang { 427f7039b1dSLi Dongyang struct btrfs_device *device; 428f7039b1dSLi Dongyang struct request_queue *q; 429f7039b1dSLi Dongyang struct fstrim_range range; 430f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 431f7039b1dSLi Dongyang u64 num_devices = 0; 432f7039b1dSLi Dongyang int ret; 433f7039b1dSLi Dongyang 434f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 435f7039b1dSLi Dongyang return -EPERM; 436f7039b1dSLi Dongyang 437f35f06c3SFilipe Manana /* 4381cb3dc3fSNaohiro Aota * btrfs_trim_block_group() depends on space cache, which is not 4391cb3dc3fSNaohiro Aota * available in zoned filesystem. So, disallow fitrim on a zoned 4401cb3dc3fSNaohiro Aota * filesystem for now. 4411cb3dc3fSNaohiro Aota */ 4421cb3dc3fSNaohiro Aota if (btrfs_is_zoned(fs_info)) 4431cb3dc3fSNaohiro Aota return -EOPNOTSUPP; 4441cb3dc3fSNaohiro Aota 4451cb3dc3fSNaohiro Aota /* 446f35f06c3SFilipe Manana * If the fs is mounted with nologreplay, which requires it to be 447f35f06c3SFilipe Manana * mounted in RO mode as well, we can not allow discard on free space 448f35f06c3SFilipe Manana * inside block groups, because log trees refer to extents that are not 449f35f06c3SFilipe Manana * pinned in a block group's free space cache (pinning the extents is 450f35f06c3SFilipe Manana * precisely the first phase of replaying a log tree). 451f35f06c3SFilipe Manana */ 452f35f06c3SFilipe Manana if (btrfs_test_opt(fs_info, NOLOGREPLAY)) 453f35f06c3SFilipe Manana return -EROFS; 454f35f06c3SFilipe Manana 4551f78160cSXiao Guangrong rcu_read_lock(); 4561f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 4571f78160cSXiao Guangrong dev_list) { 458f7039b1dSLi Dongyang if (!device->bdev) 459f7039b1dSLi Dongyang continue; 460f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 461f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 462f7039b1dSLi Dongyang num_devices++; 46350d0446eSSeraphime Kirkovski minlen = min_t(u64, q->limits.discard_granularity, 464f7039b1dSLi Dongyang minlen); 465f7039b1dSLi Dongyang } 466f7039b1dSLi Dongyang } 4671f78160cSXiao Guangrong rcu_read_unlock(); 468f4c697e6SLukas Czerner 469f7039b1dSLi Dongyang if (!num_devices) 470f7039b1dSLi Dongyang return -EOPNOTSUPP; 471f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 472f7039b1dSLi Dongyang return -EFAULT; 4736ba9fc8eSQu Wenruo 4746ba9fc8eSQu Wenruo /* 4756ba9fc8eSQu Wenruo * NOTE: Don't truncate the range using super->total_bytes. Bytenr of 4766ba9fc8eSQu Wenruo * block group is in the logical address space, which can be any 4776ba9fc8eSQu Wenruo * sectorsize aligned bytenr in the range [0, U64_MAX]. 4786ba9fc8eSQu Wenruo */ 4796ba9fc8eSQu Wenruo if (range.len < fs_info->sb->s_blocksize) 480f4c697e6SLukas Czerner return -EINVAL; 481f7039b1dSLi Dongyang 482f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 4832ff7e61eSJeff Mahoney ret = btrfs_trim_fs(fs_info, &range); 484f7039b1dSLi Dongyang if (ret < 0) 485f7039b1dSLi Dongyang return ret; 486f7039b1dSLi Dongyang 487f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 488f7039b1dSLi Dongyang return -EFAULT; 489f7039b1dSLi Dongyang 490f7039b1dSLi Dongyang return 0; 491f7039b1dSLi Dongyang } 492f7039b1dSLi Dongyang 493e1f60a65SDavid Sterba int __pure btrfs_is_empty_uuid(u8 *uuid) 494dd5f9615SStefan Behrens { 49546e0f66aSChris Mason int i; 49646e0f66aSChris Mason 49746e0f66aSChris Mason for (i = 0; i < BTRFS_UUID_SIZE; i++) { 49846e0f66aSChris Mason if (uuid[i]) 49946e0f66aSChris Mason return 0; 50046e0f66aSChris Mason } 50146e0f66aSChris Mason return 1; 502dd5f9615SStefan Behrens } 503dd5f9615SStefan Behrens 5044d4340c9SChristian Brauner static noinline int create_subvol(struct user_namespace *mnt_userns, 5054d4340c9SChristian Brauner struct inode *dir, struct dentry *dentry, 50652f75f4fSDavid Sterba const char *name, int namelen, 5078696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 508f46b5a66SChristoph Hellwig { 5090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 510f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 511f46b5a66SChristoph Hellwig struct btrfs_key key; 51249a3c4d9SDavid Sterba struct btrfs_root_item *root_item; 513f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 514f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 515d5c12070SMiao Xie struct btrfs_root *root = BTRFS_I(dir)->root; 51676dda93cSYan, Zheng struct btrfs_root *new_root; 517d5c12070SMiao Xie struct btrfs_block_rsv block_rsv; 51895582b00SDeepa Dinamani struct timespec64 cur_time = current_time(dir); 5195662344bSTsutomu Itoh struct inode *inode; 520f46b5a66SChristoph Hellwig int ret; 521f46b5a66SChristoph Hellwig int err; 5222dfb1e43SQu Wenruo dev_t anon_dev = 0; 523f46b5a66SChristoph Hellwig u64 objectid; 5243de4586cSChris Mason u64 index = 0; 525f46b5a66SChristoph Hellwig 52649a3c4d9SDavid Sterba root_item = kzalloc(sizeof(*root_item), GFP_KERNEL); 52749a3c4d9SDavid Sterba if (!root_item) 52849a3c4d9SDavid Sterba return -ENOMEM; 52949a3c4d9SDavid Sterba 530543068a2SNikolay Borisov ret = btrfs_get_free_objectid(fs_info->tree_root, &objectid); 5312fbe8c8aSAl Viro if (ret) 53249a3c4d9SDavid Sterba goto fail_free; 5336a912213SJosef Bacik 5342dfb1e43SQu Wenruo ret = get_anon_bdev(&anon_dev); 5352dfb1e43SQu Wenruo if (ret < 0) 5362dfb1e43SQu Wenruo goto fail_free; 5372dfb1e43SQu Wenruo 538e09fe2d2SQu Wenruo /* 539e09fe2d2SQu Wenruo * Don't create subvolume whose level is not zero. Or qgroup will be 54001327610SNicholas D Steeves * screwed up since it assumes subvolume qgroup's level to be 0. 541e09fe2d2SQu Wenruo */ 54249a3c4d9SDavid Sterba if (btrfs_qgroup_level(objectid)) { 54349a3c4d9SDavid Sterba ret = -ENOSPC; 54449a3c4d9SDavid Sterba goto fail_free; 54549a3c4d9SDavid Sterba } 546e09fe2d2SQu Wenruo 547d5c12070SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 5489ed74f2dSJosef Bacik /* 549d5c12070SMiao Xie * The same as the snapshot creation, please see the comment 550d5c12070SMiao Xie * of create_snapshot(). 5519ed74f2dSJosef Bacik */ 552c4c129dbSGu JinXiang ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 8, false); 553d5c12070SMiao Xie if (ret) 55449a3c4d9SDavid Sterba goto fail_free; 555f46b5a66SChristoph Hellwig 556d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 557d5c12070SMiao Xie if (IS_ERR(trans)) { 558d5c12070SMiao Xie ret = PTR_ERR(trans); 559e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 56049a3c4d9SDavid Sterba goto fail_free; 561d5c12070SMiao Xie } 562d5c12070SMiao Xie trans->block_rsv = &block_rsv; 563d5c12070SMiao Xie trans->bytes_reserved = block_rsv.size; 564f46b5a66SChristoph Hellwig 565a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, 0, objectid, inherit); 5666f72c7e2SArne Jansen if (ret) 5676f72c7e2SArne Jansen goto fail; 5686f72c7e2SArne Jansen 5699631e4ccSJosef Bacik leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0, 5709631e4ccSJosef Bacik BTRFS_NESTING_NORMAL); 5718e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 5728e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 5738e8a1e31SJosef Bacik goto fail; 5748e8a1e31SJosef Bacik } 575f46b5a66SChristoph Hellwig 576f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 577f46b5a66SChristoph Hellwig 57849a3c4d9SDavid Sterba inode_item = &root_item->inode; 5793cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 5803cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 5813cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 582da17066cSJeff Mahoney btrfs_set_stack_inode_nbytes(inode_item, 5830b246afaSJeff Mahoney fs_info->nodesize); 5843cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 585f46b5a66SChristoph Hellwig 58649a3c4d9SDavid Sterba btrfs_set_root_flags(root_item, 0); 58749a3c4d9SDavid Sterba btrfs_set_root_limit(root_item, 0); 5883cae210fSQu Wenruo btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 58908fe4db1SLi Zefan 59049a3c4d9SDavid Sterba btrfs_set_root_bytenr(root_item, leaf->start); 59149a3c4d9SDavid Sterba btrfs_set_root_generation(root_item, trans->transid); 59249a3c4d9SDavid Sterba btrfs_set_root_level(root_item, 0); 59349a3c4d9SDavid Sterba btrfs_set_root_refs(root_item, 1); 59449a3c4d9SDavid Sterba btrfs_set_root_used(root_item, leaf->len); 59549a3c4d9SDavid Sterba btrfs_set_root_last_snapshot(root_item, 0); 596f46b5a66SChristoph Hellwig 59749a3c4d9SDavid Sterba btrfs_set_root_generation_v2(root_item, 59849a3c4d9SDavid Sterba btrfs_root_generation(root_item)); 599807fc790SAndy Shevchenko generate_random_guid(root_item->uuid); 60049a3c4d9SDavid Sterba btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec); 60149a3c4d9SDavid Sterba btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec); 60249a3c4d9SDavid Sterba root_item->ctime = root_item->otime; 60349a3c4d9SDavid Sterba btrfs_set_root_ctransid(root_item, trans->transid); 60449a3c4d9SDavid Sterba btrfs_set_root_otransid(root_item, trans->transid); 605f46b5a66SChristoph Hellwig 606925baeddSChris Mason btrfs_tree_unlock(leaf); 607f46b5a66SChristoph Hellwig 60869948022SNikolay Borisov btrfs_set_root_dirid(root_item, BTRFS_FIRST_FREE_OBJECTID); 609f46b5a66SChristoph Hellwig 610f46b5a66SChristoph Hellwig key.objectid = objectid; 6115d4f98a2SYan Zheng key.offset = 0; 612962a298fSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 6130b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, &key, 61449a3c4d9SDavid Sterba root_item); 61567addf29SFilipe Manana if (ret) { 61667addf29SFilipe Manana /* 61767addf29SFilipe Manana * Since we don't abort the transaction in this case, free the 61867addf29SFilipe Manana * tree block so that we don't leak space and leave the 61967addf29SFilipe Manana * filesystem in an inconsistent state (an extent item in the 62067addf29SFilipe Manana * extent tree without backreferences). Also no need to have 62167addf29SFilipe Manana * the tree block locked since it is not in any tree at this 62267addf29SFilipe Manana * point, so no other task can find it and use it. 62367addf29SFilipe Manana */ 62467addf29SFilipe Manana btrfs_free_tree_block(trans, root, leaf, 0, 1); 62567addf29SFilipe Manana free_extent_buffer(leaf); 626f46b5a66SChristoph Hellwig goto fail; 62767addf29SFilipe Manana } 62867addf29SFilipe Manana 62967addf29SFilipe Manana free_extent_buffer(leaf); 63067addf29SFilipe Manana leaf = NULL; 631f46b5a66SChristoph Hellwig 63276dda93cSYan, Zheng key.offset = (u64)-1; 6332dfb1e43SQu Wenruo new_root = btrfs_get_new_fs_root(fs_info, objectid, anon_dev); 63479787eaaSJeff Mahoney if (IS_ERR(new_root)) { 6352dfb1e43SQu Wenruo free_anon_bdev(anon_dev); 63679787eaaSJeff Mahoney ret = PTR_ERR(new_root); 63766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 63879787eaaSJeff Mahoney goto fail; 63979787eaaSJeff Mahoney } 6402dfb1e43SQu Wenruo /* Freeing will be done in btrfs_put_root() of new_root */ 6412dfb1e43SQu Wenruo anon_dev = 0; 64276dda93cSYan, Zheng 643221581e4SJosef Bacik ret = btrfs_record_root_in_trans(trans, new_root); 644221581e4SJosef Bacik if (ret) { 645221581e4SJosef Bacik btrfs_put_root(new_root); 646221581e4SJosef Bacik btrfs_abort_transaction(trans, ret); 647221581e4SJosef Bacik goto fail; 648221581e4SJosef Bacik } 64976dda93cSYan, Zheng 6504d4340c9SChristian Brauner ret = btrfs_create_subvol_root(trans, new_root, root, mnt_userns); 65100246528SJosef Bacik btrfs_put_root(new_root); 652ce598979SMark Fasheh if (ret) { 653ce598979SMark Fasheh /* We potentially lose an unused inode item here */ 65466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 655ce598979SMark Fasheh goto fail; 656ce598979SMark Fasheh } 657ce598979SMark Fasheh 658f46b5a66SChristoph Hellwig /* 659f46b5a66SChristoph Hellwig * insert the directory item 660f46b5a66SChristoph Hellwig */ 661877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(dir), &index); 66279787eaaSJeff Mahoney if (ret) { 66366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 66479787eaaSJeff Mahoney goto fail; 66579787eaaSJeff Mahoney } 6663de4586cSChris Mason 667684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, name, namelen, BTRFS_I(dir), &key, 6683de4586cSChris Mason BTRFS_FT_DIR, index); 66979787eaaSJeff Mahoney if (ret) { 67066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 671f46b5a66SChristoph Hellwig goto fail; 67279787eaaSJeff Mahoney } 6730660b5afSChris Mason 6746ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(dir), dir->i_size + namelen * 2); 6759a56fcd1SNikolay Borisov ret = btrfs_update_inode(trans, root, BTRFS_I(dir)); 676c7e54b51SJosef Bacik if (ret) { 677c7e54b51SJosef Bacik btrfs_abort_transaction(trans, ret); 678c7e54b51SJosef Bacik goto fail; 679c7e54b51SJosef Bacik } 68052c26179SYan Zheng 6816025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, root->root_key.objectid, 6824a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(dir)), index, name, namelen); 683c7e54b51SJosef Bacik if (ret) { 684c7e54b51SJosef Bacik btrfs_abort_transaction(trans, ret); 685c7e54b51SJosef Bacik goto fail; 686c7e54b51SJosef Bacik } 6870660b5afSChris Mason 688cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, root_item->uuid, 6896bccf3abSJeff Mahoney BTRFS_UUID_KEY_SUBVOL, objectid); 690dd5f9615SStefan Behrens if (ret) 69166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 692dd5f9615SStefan Behrens 693f46b5a66SChristoph Hellwig fail: 69449a3c4d9SDavid Sterba kfree(root_item); 695d5c12070SMiao Xie trans->block_rsv = NULL; 696d5c12070SMiao Xie trans->bytes_reserved = 0; 697e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &block_rsv); 698de6e8200SLiu Bo 6993a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 700f46b5a66SChristoph Hellwig if (err && !ret) 701f46b5a66SChristoph Hellwig ret = err; 7021a65e24bSChris Mason 7035662344bSTsutomu Itoh if (!ret) { 7045662344bSTsutomu Itoh inode = btrfs_lookup_dentry(dir, dentry); 705de6e8200SLiu Bo if (IS_ERR(inode)) 706de6e8200SLiu Bo return PTR_ERR(inode); 7075662344bSTsutomu Itoh d_instantiate(dentry, inode); 7085662344bSTsutomu Itoh } 709f46b5a66SChristoph Hellwig return ret; 71049a3c4d9SDavid Sterba 71149a3c4d9SDavid Sterba fail_free: 7122dfb1e43SQu Wenruo if (anon_dev) 7132dfb1e43SQu Wenruo free_anon_bdev(anon_dev); 71449a3c4d9SDavid Sterba kfree(root_item); 71549a3c4d9SDavid Sterba return ret; 716f46b5a66SChristoph Hellwig } 717f46b5a66SChristoph Hellwig 718e9662f70SMiao Xie static int create_snapshot(struct btrfs_root *root, struct inode *dir, 7199babda9fSNikolay Borisov struct dentry *dentry, bool readonly, 720e9662f70SMiao Xie struct btrfs_qgroup_inherit *inherit) 721f46b5a66SChristoph Hellwig { 7220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 7232e4bfab9SYan, Zheng struct inode *inode; 724f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 725f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 7262e4bfab9SYan, Zheng int ret; 727f46b5a66SChristoph Hellwig 72892a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 729f46b5a66SChristoph Hellwig return -EINVAL; 730f46b5a66SChristoph Hellwig 731eede2bf3SOmar Sandoval if (atomic_read(&root->nr_swapfiles)) { 732eede2bf3SOmar Sandoval btrfs_warn(fs_info, 733eede2bf3SOmar Sandoval "cannot snapshot subvolume with active swapfile"); 734eede2bf3SOmar Sandoval return -ETXTBSY; 735eede2bf3SOmar Sandoval } 736eede2bf3SOmar Sandoval 73723269bf5SDavid Sterba pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_KERNEL); 738a1ee7362SDavid Sterba if (!pending_snapshot) 739a1ee7362SDavid Sterba return -ENOMEM; 740a1ee7362SDavid Sterba 7412dfb1e43SQu Wenruo ret = get_anon_bdev(&pending_snapshot->anon_dev); 7422dfb1e43SQu Wenruo if (ret < 0) 7432dfb1e43SQu Wenruo goto free_pending; 744b0c0ea63SDavid Sterba pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item), 74523269bf5SDavid Sterba GFP_KERNEL); 7468546b570SDavid Sterba pending_snapshot->path = btrfs_alloc_path(); 7478546b570SDavid Sterba if (!pending_snapshot->root_item || !pending_snapshot->path) { 748b0c0ea63SDavid Sterba ret = -ENOMEM; 749b0c0ea63SDavid Sterba goto free_pending; 750b0c0ea63SDavid Sterba } 751b0c0ea63SDavid Sterba 75266d8f3ddSMiao Xie btrfs_init_block_rsv(&pending_snapshot->block_rsv, 75366d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 754d5c12070SMiao Xie /* 755d5c12070SMiao Xie * 1 - parent dir inode 756d5c12070SMiao Xie * 2 - dir entries 757d5c12070SMiao Xie * 1 - root item 758d5c12070SMiao Xie * 2 - root ref/backref 759d5c12070SMiao Xie * 1 - root of snapshot 760dd5f9615SStefan Behrens * 1 - UUID item 761d5c12070SMiao Xie */ 762d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 763dd5f9615SStefan Behrens &pending_snapshot->block_rsv, 8, 764ee3441b4SJeff Mahoney false); 765d5c12070SMiao Xie if (ret) 766c11fbb6eSRobbie Ko goto free_pending; 767d5c12070SMiao Xie 768a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 769a22285a6SYan, Zheng pending_snapshot->root = root; 770b83cc969SLi Zefan pending_snapshot->readonly = readonly; 771e9662f70SMiao Xie pending_snapshot->dir = dir; 7728696c533SMiao Xie pending_snapshot->inherit = inherit; 773a22285a6SYan, Zheng 774d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 775a22285a6SYan, Zheng if (IS_ERR(trans)) { 776a22285a6SYan, Zheng ret = PTR_ERR(trans); 777a22285a6SYan, Zheng goto fail; 778a22285a6SYan, Zheng } 779a22285a6SYan, Zheng 7800b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 781a22285a6SYan, Zheng list_add(&pending_snapshot->list, 782a22285a6SYan, Zheng &trans->transaction->pending_snapshots); 7830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 7849babda9fSNikolay Borisov 7853a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 786aec8030aSMiao Xie if (ret) 787c37b2b62SJosef Bacik goto fail; 788a22285a6SYan, Zheng 789a22285a6SYan, Zheng ret = pending_snapshot->error; 790f46b5a66SChristoph Hellwig if (ret) 7912e4bfab9SYan, Zheng goto fail; 792f46b5a66SChristoph Hellwig 793d3797308SChris Mason ret = btrfs_orphan_cleanup(pending_snapshot->snap); 794d3797308SChris Mason if (ret) 795d3797308SChris Mason goto fail; 796d3797308SChris Mason 7972b0143b5SDavid Howells inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry); 7982e4bfab9SYan, Zheng if (IS_ERR(inode)) { 7992e4bfab9SYan, Zheng ret = PTR_ERR(inode); 8002e4bfab9SYan, Zheng goto fail; 8012e4bfab9SYan, Zheng } 8025662344bSTsutomu Itoh 8032e4bfab9SYan, Zheng d_instantiate(dentry, inode); 8042e4bfab9SYan, Zheng ret = 0; 8052dfb1e43SQu Wenruo pending_snapshot->anon_dev = 0; 8062e4bfab9SYan, Zheng fail: 8072dfb1e43SQu Wenruo /* Prevent double freeing of anon_dev */ 8082dfb1e43SQu Wenruo if (ret && pending_snapshot->snap) 8092dfb1e43SQu Wenruo pending_snapshot->snap->anon_dev = 0; 81000246528SJosef Bacik btrfs_put_root(pending_snapshot->snap); 811e85fde51SQu Wenruo btrfs_subvolume_release_metadata(root, &pending_snapshot->block_rsv); 812b0c0ea63SDavid Sterba free_pending: 8132dfb1e43SQu Wenruo if (pending_snapshot->anon_dev) 8142dfb1e43SQu Wenruo free_anon_bdev(pending_snapshot->anon_dev); 815b0c0ea63SDavid Sterba kfree(pending_snapshot->root_item); 8168546b570SDavid Sterba btrfs_free_path(pending_snapshot->path); 817a1ee7362SDavid Sterba kfree(pending_snapshot); 818a1ee7362SDavid Sterba 819f46b5a66SChristoph Hellwig return ret; 820f46b5a66SChristoph Hellwig } 821f46b5a66SChristoph Hellwig 8224260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 8234260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 8244260f7c7SSage Weil * whether the type of victim is right. 8254260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 8264260f7c7SSage Weil * 2. We should have write and exec permissions on dir 8274260f7c7SSage Weil * 3. We can't remove anything from append-only dir 8284260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 8294260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 8304260f7c7SSage Weil * a. be owner of dir, or 8314260f7c7SSage Weil * b. be owner of victim, or 8324260f7c7SSage Weil * c. have CAP_FOWNER capability 83301327610SNicholas D Steeves * 6. If the victim is append-only or immutable we can't do anything with 8344260f7c7SSage Weil * links pointing to it. 8354260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 8364260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 8374260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 8384260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 8394260f7c7SSage Weil * nfs_async_unlink(). 8404260f7c7SSage Weil */ 8414260f7c7SSage Weil 842c4ed533bSChristian Brauner static int btrfs_may_delete(struct user_namespace *mnt_userns, 843c4ed533bSChristian Brauner struct inode *dir, struct dentry *victim, int isdir) 8444260f7c7SSage Weil { 8454260f7c7SSage Weil int error; 8464260f7c7SSage Weil 8472b0143b5SDavid Howells if (d_really_is_negative(victim)) 8484260f7c7SSage Weil return -ENOENT; 8494260f7c7SSage Weil 8502b0143b5SDavid Howells BUG_ON(d_inode(victim->d_parent) != dir); 8514fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 8524260f7c7SSage Weil 853c4ed533bSChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 8544260f7c7SSage Weil if (error) 8554260f7c7SSage Weil return error; 8564260f7c7SSage Weil if (IS_APPEND(dir)) 8574260f7c7SSage Weil return -EPERM; 858c4ed533bSChristian Brauner if (check_sticky(mnt_userns, dir, d_inode(victim)) || 859ba73d987SChristian Brauner IS_APPEND(d_inode(victim)) || IS_IMMUTABLE(d_inode(victim)) || 860ba73d987SChristian Brauner IS_SWAPFILE(d_inode(victim))) 8614260f7c7SSage Weil return -EPERM; 8624260f7c7SSage Weil if (isdir) { 863e36cb0b8SDavid Howells if (!d_is_dir(victim)) 8644260f7c7SSage Weil return -ENOTDIR; 8654260f7c7SSage Weil if (IS_ROOT(victim)) 8664260f7c7SSage Weil return -EBUSY; 867e36cb0b8SDavid Howells } else if (d_is_dir(victim)) 8684260f7c7SSage Weil return -EISDIR; 8694260f7c7SSage Weil if (IS_DEADDIR(dir)) 8704260f7c7SSage Weil return -ENOENT; 8714260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 8724260f7c7SSage Weil return -EBUSY; 8734260f7c7SSage Weil return 0; 8744260f7c7SSage Weil } 8754260f7c7SSage Weil 876cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 8774d4340c9SChristian Brauner static inline int btrfs_may_create(struct user_namespace *mnt_userns, 8784d4340c9SChristian Brauner struct inode *dir, struct dentry *child) 879cb8e7090SChristoph Hellwig { 8802b0143b5SDavid Howells if (d_really_is_positive(child)) 881cb8e7090SChristoph Hellwig return -EEXIST; 882cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 883cb8e7090SChristoph Hellwig return -ENOENT; 8844d4340c9SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 8855474bf40SChristian Brauner return -EOVERFLOW; 8864d4340c9SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 887cb8e7090SChristoph Hellwig } 888cb8e7090SChristoph Hellwig 889cb8e7090SChristoph Hellwig /* 890cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 891cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 892cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 893cb8e7090SChristoph Hellwig */ 89492872094SAl Viro static noinline int btrfs_mksubvol(const struct path *parent, 8954d4340c9SChristian Brauner struct user_namespace *mnt_userns, 89652f75f4fSDavid Sterba const char *name, int namelen, 89772fd032eSSage Weil struct btrfs_root *snap_src, 8989babda9fSNikolay Borisov bool readonly, 8998696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 900cb8e7090SChristoph Hellwig { 9012b0143b5SDavid Howells struct inode *dir = d_inode(parent->dentry); 9020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb); 903cb8e7090SChristoph Hellwig struct dentry *dentry; 904cb8e7090SChristoph Hellwig int error; 905cb8e7090SChristoph Hellwig 90600235411SAl Viro error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 90700235411SAl Viro if (error == -EINTR) 90800235411SAl Viro return error; 909cb8e7090SChristoph Hellwig 9104d4340c9SChristian Brauner dentry = lookup_one(mnt_userns, name, parent->dentry, namelen); 911cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 912cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 913cb8e7090SChristoph Hellwig goto out_unlock; 914cb8e7090SChristoph Hellwig 9154d4340c9SChristian Brauner error = btrfs_may_create(mnt_userns, dir, dentry); 916cb8e7090SChristoph Hellwig if (error) 917a874a63eSLiu Bo goto out_dput; 918cb8e7090SChristoph Hellwig 9199c52057cSChris Mason /* 9209c52057cSChris Mason * even if this name doesn't exist, we may get hash collisions. 9219c52057cSChris Mason * check for them now when we can safely fail 9229c52057cSChris Mason */ 9239c52057cSChris Mason error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 9249c52057cSChris Mason dir->i_ino, name, 9259c52057cSChris Mason namelen); 9269c52057cSChris Mason if (error) 9279c52057cSChris Mason goto out_dput; 9289c52057cSChris Mason 9290b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 93076dda93cSYan, Zheng 93176dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 93276dda93cSYan, Zheng goto out_up_read; 93376dda93cSYan, Zheng 9349babda9fSNikolay Borisov if (snap_src) 9359babda9fSNikolay Borisov error = create_snapshot(snap_src, dir, dentry, readonly, inherit); 9369babda9fSNikolay Borisov else 9374d4340c9SChristian Brauner error = create_subvol(mnt_userns, dir, dentry, name, namelen, inherit); 9389babda9fSNikolay Borisov 93976dda93cSYan, Zheng if (!error) 94076dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 94176dda93cSYan, Zheng out_up_read: 9420b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 943cb8e7090SChristoph Hellwig out_dput: 944cb8e7090SChristoph Hellwig dput(dentry); 945cb8e7090SChristoph Hellwig out_unlock: 94664708539SJosef Bacik btrfs_inode_unlock(dir, 0); 947cb8e7090SChristoph Hellwig return error; 948cb8e7090SChristoph Hellwig } 949cb8e7090SChristoph Hellwig 950c11fbb6eSRobbie Ko static noinline int btrfs_mksnapshot(const struct path *parent, 9514d4340c9SChristian Brauner struct user_namespace *mnt_userns, 952c11fbb6eSRobbie Ko const char *name, int namelen, 953c11fbb6eSRobbie Ko struct btrfs_root *root, 954c11fbb6eSRobbie Ko bool readonly, 955c11fbb6eSRobbie Ko struct btrfs_qgroup_inherit *inherit) 956c11fbb6eSRobbie Ko { 957c11fbb6eSRobbie Ko int ret; 958c11fbb6eSRobbie Ko bool snapshot_force_cow = false; 959c11fbb6eSRobbie Ko 960c11fbb6eSRobbie Ko /* 961c11fbb6eSRobbie Ko * Force new buffered writes to reserve space even when NOCOW is 962c11fbb6eSRobbie Ko * possible. This is to avoid later writeback (running dealloc) to 963c11fbb6eSRobbie Ko * fallback to COW mode and unexpectedly fail with ENOSPC. 964c11fbb6eSRobbie Ko */ 965c11fbb6eSRobbie Ko btrfs_drew_read_lock(&root->snapshot_lock); 966c11fbb6eSRobbie Ko 967f9baa501SFilipe Manana ret = btrfs_start_delalloc_snapshot(root, false); 968c11fbb6eSRobbie Ko if (ret) 969c11fbb6eSRobbie Ko goto out; 970c11fbb6eSRobbie Ko 971c11fbb6eSRobbie Ko /* 972c11fbb6eSRobbie Ko * All previous writes have started writeback in NOCOW mode, so now 973c11fbb6eSRobbie Ko * we force future writes to fallback to COW mode during snapshot 974c11fbb6eSRobbie Ko * creation. 975c11fbb6eSRobbie Ko */ 976c11fbb6eSRobbie Ko atomic_inc(&root->snapshot_force_cow); 977c11fbb6eSRobbie Ko snapshot_force_cow = true; 978c11fbb6eSRobbie Ko 979c11fbb6eSRobbie Ko btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1); 980c11fbb6eSRobbie Ko 9814d4340c9SChristian Brauner ret = btrfs_mksubvol(parent, mnt_userns, name, namelen, 982c11fbb6eSRobbie Ko root, readonly, inherit); 983c11fbb6eSRobbie Ko out: 984c11fbb6eSRobbie Ko if (snapshot_force_cow) 985c11fbb6eSRobbie Ko atomic_dec(&root->snapshot_force_cow); 986c11fbb6eSRobbie Ko btrfs_drew_read_unlock(&root->snapshot_lock); 987c11fbb6eSRobbie Ko return ret; 988c11fbb6eSRobbie Ko } 989c11fbb6eSRobbie Ko 990e9eec721SQu Wenruo static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start, 991e9eec721SQu Wenruo bool locked) 99217ce6ef8SLiu Bo { 99317ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 994940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 9956c282eb4SLi Zefan struct extent_map *em; 99676068caeSQu Wenruo const u32 sectorsize = BTRFS_I(inode)->root->fs_info->sectorsize; 997940100a4SChris Mason 998940100a4SChris Mason /* 999940100a4SChris Mason * hopefully we have this extent in the tree already, try without 1000940100a4SChris Mason * the full extent lock 1001940100a4SChris Mason */ 1002940100a4SChris Mason read_lock(&em_tree->lock); 100376068caeSQu Wenruo em = lookup_extent_mapping(em_tree, start, sectorsize); 1004940100a4SChris Mason read_unlock(&em_tree->lock); 1005940100a4SChris Mason 1006940100a4SChris Mason if (!em) { 1007308d9800SFilipe Manana struct extent_state *cached = NULL; 100876068caeSQu Wenruo u64 end = start + sectorsize - 1; 1009308d9800SFilipe Manana 1010940100a4SChris Mason /* get the big lock and read metadata off disk */ 1011e9eec721SQu Wenruo if (!locked) 1012ff13db41SDavid Sterba lock_extent_bits(io_tree, start, end, &cached); 101376068caeSQu Wenruo em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, sectorsize); 1014e9eec721SQu Wenruo if (!locked) 1015e43bbe5eSDavid Sterba unlock_extent_cached(io_tree, start, end, &cached); 1016940100a4SChris Mason 10176cf8bfbfSDan Carpenter if (IS_ERR(em)) 10186c282eb4SLi Zefan return NULL; 1019940100a4SChris Mason } 1020940100a4SChris Mason 10216c282eb4SLi Zefan return em; 10226c282eb4SLi Zefan } 10236c282eb4SLi Zefan 1024e9eec721SQu Wenruo static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em, 1025e9eec721SQu Wenruo bool locked) 10266c282eb4SLi Zefan { 10276c282eb4SLi Zefan struct extent_map *next; 10286c282eb4SLi Zefan bool ret = true; 10296c282eb4SLi Zefan 10306c282eb4SLi Zefan /* this is the last extent */ 10316c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 10326c282eb4SLi Zefan return false; 10336c282eb4SLi Zefan 1034e9eec721SQu Wenruo next = defrag_lookup_extent(inode, em->start + em->len, locked); 1035e9512d72SChris Mason if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 1036e9512d72SChris Mason ret = false; 1037e9512d72SChris Mason else if ((em->block_start + em->block_len == next->block_start) && 1038ee22184bSByongho Lee (em->block_len > SZ_128K && next->block_len > SZ_128K)) 10396c282eb4SLi Zefan ret = false; 10406c282eb4SLi Zefan 10416c282eb4SLi Zefan free_extent_map(next); 10426c282eb4SLi Zefan return ret; 10436c282eb4SLi Zefan } 10446c282eb4SLi Zefan 10454cb5300bSChris Mason /* 10465767b50cSQu Wenruo * Prepare one page to be defragged. 10475767b50cSQu Wenruo * 10485767b50cSQu Wenruo * This will ensure: 10495767b50cSQu Wenruo * 10505767b50cSQu Wenruo * - Returned page is locked and has been set up properly. 10515767b50cSQu Wenruo * - No ordered extent exists in the page. 10525767b50cSQu Wenruo * - The page is uptodate. 10535767b50cSQu Wenruo * 10545767b50cSQu Wenruo * NOTE: Caller should also wait for page writeback after the cluster is 10555767b50cSQu Wenruo * prepared, here we don't do writeback wait for each page. 10565767b50cSQu Wenruo */ 10575767b50cSQu Wenruo static struct page *defrag_prepare_one_page(struct btrfs_inode *inode, 10585767b50cSQu Wenruo pgoff_t index) 10595767b50cSQu Wenruo { 10605767b50cSQu Wenruo struct address_space *mapping = inode->vfs_inode.i_mapping; 10615767b50cSQu Wenruo gfp_t mask = btrfs_alloc_write_mask(mapping); 10625767b50cSQu Wenruo u64 page_start = (u64)index << PAGE_SHIFT; 10635767b50cSQu Wenruo u64 page_end = page_start + PAGE_SIZE - 1; 10645767b50cSQu Wenruo struct extent_state *cached_state = NULL; 10655767b50cSQu Wenruo struct page *page; 10665767b50cSQu Wenruo int ret; 10675767b50cSQu Wenruo 10685767b50cSQu Wenruo again: 10695767b50cSQu Wenruo page = find_or_create_page(mapping, index, mask); 10705767b50cSQu Wenruo if (!page) 10715767b50cSQu Wenruo return ERR_PTR(-ENOMEM); 10725767b50cSQu Wenruo 107324bcb454SOmar Sandoval /* 107424bcb454SOmar Sandoval * Since we can defragment files opened read-only, we can encounter 107524bcb454SOmar Sandoval * transparent huge pages here (see CONFIG_READ_ONLY_THP_FOR_FS). We 107624bcb454SOmar Sandoval * can't do I/O using huge pages yet, so return an error for now. 107724bcb454SOmar Sandoval * Filesystem transparent huge pages are typically only used for 107824bcb454SOmar Sandoval * executables that explicitly enable them, so this isn't very 107924bcb454SOmar Sandoval * restrictive. 108024bcb454SOmar Sandoval */ 108124bcb454SOmar Sandoval if (PageCompound(page)) { 108224bcb454SOmar Sandoval unlock_page(page); 108324bcb454SOmar Sandoval put_page(page); 108424bcb454SOmar Sandoval return ERR_PTR(-ETXTBSY); 108524bcb454SOmar Sandoval } 108624bcb454SOmar Sandoval 10875767b50cSQu Wenruo ret = set_page_extent_mapped(page); 10885767b50cSQu Wenruo if (ret < 0) { 10895767b50cSQu Wenruo unlock_page(page); 10905767b50cSQu Wenruo put_page(page); 10915767b50cSQu Wenruo return ERR_PTR(ret); 10925767b50cSQu Wenruo } 10935767b50cSQu Wenruo 10945767b50cSQu Wenruo /* Wait for any existing ordered extent in the range */ 10955767b50cSQu Wenruo while (1) { 10965767b50cSQu Wenruo struct btrfs_ordered_extent *ordered; 10975767b50cSQu Wenruo 10985767b50cSQu Wenruo lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state); 10995767b50cSQu Wenruo ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE); 11005767b50cSQu Wenruo unlock_extent_cached(&inode->io_tree, page_start, page_end, 11015767b50cSQu Wenruo &cached_state); 11025767b50cSQu Wenruo if (!ordered) 11035767b50cSQu Wenruo break; 11045767b50cSQu Wenruo 11055767b50cSQu Wenruo unlock_page(page); 11065767b50cSQu Wenruo btrfs_start_ordered_extent(ordered, 1); 11075767b50cSQu Wenruo btrfs_put_ordered_extent(ordered); 11085767b50cSQu Wenruo lock_page(page); 11095767b50cSQu Wenruo /* 11105767b50cSQu Wenruo * We unlocked the page above, so we need check if it was 11115767b50cSQu Wenruo * released or not. 11125767b50cSQu Wenruo */ 11135767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 11145767b50cSQu Wenruo unlock_page(page); 11155767b50cSQu Wenruo put_page(page); 11165767b50cSQu Wenruo goto again; 11175767b50cSQu Wenruo } 11185767b50cSQu Wenruo } 11195767b50cSQu Wenruo 11205767b50cSQu Wenruo /* 11215767b50cSQu Wenruo * Now the page range has no ordered extent any more. Read the page to 11225767b50cSQu Wenruo * make it uptodate. 11235767b50cSQu Wenruo */ 11245767b50cSQu Wenruo if (!PageUptodate(page)) { 11255767b50cSQu Wenruo btrfs_readpage(NULL, page); 11265767b50cSQu Wenruo lock_page(page); 11275767b50cSQu Wenruo if (page->mapping != mapping || !PagePrivate(page)) { 11285767b50cSQu Wenruo unlock_page(page); 11295767b50cSQu Wenruo put_page(page); 11305767b50cSQu Wenruo goto again; 11315767b50cSQu Wenruo } 11325767b50cSQu Wenruo if (!PageUptodate(page)) { 11335767b50cSQu Wenruo unlock_page(page); 11345767b50cSQu Wenruo put_page(page); 11355767b50cSQu Wenruo return ERR_PTR(-EIO); 11365767b50cSQu Wenruo } 11375767b50cSQu Wenruo } 11385767b50cSQu Wenruo return page; 11395767b50cSQu Wenruo } 11405767b50cSQu Wenruo 1141eb793cf8SQu Wenruo struct defrag_target_range { 1142eb793cf8SQu Wenruo struct list_head list; 1143eb793cf8SQu Wenruo u64 start; 1144eb793cf8SQu Wenruo u64 len; 1145eb793cf8SQu Wenruo }; 1146eb793cf8SQu Wenruo 1147eb793cf8SQu Wenruo /* 1148eb793cf8SQu Wenruo * Collect all valid target extents. 1149eb793cf8SQu Wenruo * 1150eb793cf8SQu Wenruo * @start: file offset to lookup 1151eb793cf8SQu Wenruo * @len: length to lookup 1152eb793cf8SQu Wenruo * @extent_thresh: file extent size threshold, any extent size >= this value 1153eb793cf8SQu Wenruo * will be ignored 1154eb793cf8SQu Wenruo * @newer_than: only defrag extents newer than this value 1155eb793cf8SQu Wenruo * @do_compress: whether the defrag is doing compression 1156eb793cf8SQu Wenruo * if true, @extent_thresh will be ignored and all regular 1157eb793cf8SQu Wenruo * file extents meeting @newer_than will be targets. 1158e9eec721SQu Wenruo * @locked: if the range has already held extent lock 1159eb793cf8SQu Wenruo * @target_list: list of targets file extents 1160eb793cf8SQu Wenruo */ 1161eb793cf8SQu Wenruo static int defrag_collect_targets(struct btrfs_inode *inode, 1162eb793cf8SQu Wenruo u64 start, u64 len, u32 extent_thresh, 1163eb793cf8SQu Wenruo u64 newer_than, bool do_compress, 1164e9eec721SQu Wenruo bool locked, struct list_head *target_list) 1165eb793cf8SQu Wenruo { 1166eb793cf8SQu Wenruo u64 cur = start; 1167eb793cf8SQu Wenruo int ret = 0; 1168eb793cf8SQu Wenruo 1169eb793cf8SQu Wenruo while (cur < start + len) { 1170eb793cf8SQu Wenruo struct extent_map *em; 1171eb793cf8SQu Wenruo struct defrag_target_range *new; 1172eb793cf8SQu Wenruo bool next_mergeable = true; 1173eb793cf8SQu Wenruo u64 range_len; 1174eb793cf8SQu Wenruo 1175e9eec721SQu Wenruo em = defrag_lookup_extent(&inode->vfs_inode, cur, locked); 1176eb793cf8SQu Wenruo if (!em) 1177eb793cf8SQu Wenruo break; 1178eb793cf8SQu Wenruo 1179eb793cf8SQu Wenruo /* Skip hole/inline/preallocated extents */ 1180eb793cf8SQu Wenruo if (em->block_start >= EXTENT_MAP_LAST_BYTE || 1181eb793cf8SQu Wenruo test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) 1182eb793cf8SQu Wenruo goto next; 1183eb793cf8SQu Wenruo 1184eb793cf8SQu Wenruo /* Skip older extent */ 1185eb793cf8SQu Wenruo if (em->generation < newer_than) 1186eb793cf8SQu Wenruo goto next; 1187eb793cf8SQu Wenruo 1188eb793cf8SQu Wenruo /* 1189eb793cf8SQu Wenruo * For do_compress case, we want to compress all valid file 1190eb793cf8SQu Wenruo * extents, thus no @extent_thresh or mergeable check. 1191eb793cf8SQu Wenruo */ 1192eb793cf8SQu Wenruo if (do_compress) 1193eb793cf8SQu Wenruo goto add; 1194eb793cf8SQu Wenruo 1195eb793cf8SQu Wenruo /* Skip too large extent */ 1196eb793cf8SQu Wenruo if (em->len >= extent_thresh) 1197eb793cf8SQu Wenruo goto next; 1198eb793cf8SQu Wenruo 1199e9eec721SQu Wenruo next_mergeable = defrag_check_next_extent(&inode->vfs_inode, em, 1200e9eec721SQu Wenruo locked); 1201eb793cf8SQu Wenruo if (!next_mergeable) { 1202eb793cf8SQu Wenruo struct defrag_target_range *last; 1203eb793cf8SQu Wenruo 1204eb793cf8SQu Wenruo /* Empty target list, no way to merge with last entry */ 1205eb793cf8SQu Wenruo if (list_empty(target_list)) 1206eb793cf8SQu Wenruo goto next; 1207eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1208eb793cf8SQu Wenruo struct defrag_target_range, list); 1209eb793cf8SQu Wenruo /* Not mergeable with last entry */ 1210eb793cf8SQu Wenruo if (last->start + last->len != cur) 1211eb793cf8SQu Wenruo goto next; 1212eb793cf8SQu Wenruo 1213eb793cf8SQu Wenruo /* Mergeable, fall through to add it to @target_list. */ 1214eb793cf8SQu Wenruo } 1215eb793cf8SQu Wenruo 1216eb793cf8SQu Wenruo add: 1217eb793cf8SQu Wenruo range_len = min(extent_map_end(em), start + len) - cur; 1218eb793cf8SQu Wenruo /* 1219eb793cf8SQu Wenruo * This one is a good target, check if it can be merged into 1220eb793cf8SQu Wenruo * last range of the target list. 1221eb793cf8SQu Wenruo */ 1222eb793cf8SQu Wenruo if (!list_empty(target_list)) { 1223eb793cf8SQu Wenruo struct defrag_target_range *last; 1224eb793cf8SQu Wenruo 1225eb793cf8SQu Wenruo last = list_entry(target_list->prev, 1226eb793cf8SQu Wenruo struct defrag_target_range, list); 1227eb793cf8SQu Wenruo ASSERT(last->start + last->len <= cur); 1228eb793cf8SQu Wenruo if (last->start + last->len == cur) { 1229eb793cf8SQu Wenruo /* Mergeable, enlarge the last entry */ 1230eb793cf8SQu Wenruo last->len += range_len; 1231eb793cf8SQu Wenruo goto next; 1232eb793cf8SQu Wenruo } 1233eb793cf8SQu Wenruo /* Fall through to allocate a new entry */ 1234eb793cf8SQu Wenruo } 1235eb793cf8SQu Wenruo 1236eb793cf8SQu Wenruo /* Allocate new defrag_target_range */ 1237eb793cf8SQu Wenruo new = kmalloc(sizeof(*new), GFP_NOFS); 1238eb793cf8SQu Wenruo if (!new) { 1239eb793cf8SQu Wenruo free_extent_map(em); 1240eb793cf8SQu Wenruo ret = -ENOMEM; 1241eb793cf8SQu Wenruo break; 1242eb793cf8SQu Wenruo } 1243eb793cf8SQu Wenruo new->start = cur; 1244eb793cf8SQu Wenruo new->len = range_len; 1245eb793cf8SQu Wenruo list_add_tail(&new->list, target_list); 1246eb793cf8SQu Wenruo 1247eb793cf8SQu Wenruo next: 1248eb793cf8SQu Wenruo cur = extent_map_end(em); 1249eb793cf8SQu Wenruo free_extent_map(em); 1250eb793cf8SQu Wenruo } 1251eb793cf8SQu Wenruo if (ret < 0) { 1252eb793cf8SQu Wenruo struct defrag_target_range *entry; 1253eb793cf8SQu Wenruo struct defrag_target_range *tmp; 1254eb793cf8SQu Wenruo 1255eb793cf8SQu Wenruo list_for_each_entry_safe(entry, tmp, target_list, list) { 1256eb793cf8SQu Wenruo list_del_init(&entry->list); 1257eb793cf8SQu Wenruo kfree(entry); 1258eb793cf8SQu Wenruo } 1259eb793cf8SQu Wenruo } 1260eb793cf8SQu Wenruo return ret; 1261eb793cf8SQu Wenruo } 1262eb793cf8SQu Wenruo 126322b398eeSQu Wenruo #define CLUSTER_SIZE (SZ_256K) 126422b398eeSQu Wenruo 126522b398eeSQu Wenruo /* 126622b398eeSQu Wenruo * Defrag one contiguous target range. 126722b398eeSQu Wenruo * 126822b398eeSQu Wenruo * @inode: target inode 126922b398eeSQu Wenruo * @target: target range to defrag 127022b398eeSQu Wenruo * @pages: locked pages covering the defrag range 127122b398eeSQu Wenruo * @nr_pages: number of locked pages 127222b398eeSQu Wenruo * 127322b398eeSQu Wenruo * Caller should ensure: 127422b398eeSQu Wenruo * 127522b398eeSQu Wenruo * - Pages are prepared 127622b398eeSQu Wenruo * Pages should be locked, no ordered extent in the pages range, 127722b398eeSQu Wenruo * no writeback. 127822b398eeSQu Wenruo * 127922b398eeSQu Wenruo * - Extent bits are locked 128022b398eeSQu Wenruo */ 128122b398eeSQu Wenruo static int defrag_one_locked_target(struct btrfs_inode *inode, 128222b398eeSQu Wenruo struct defrag_target_range *target, 128322b398eeSQu Wenruo struct page **pages, int nr_pages, 128422b398eeSQu Wenruo struct extent_state **cached_state) 128522b398eeSQu Wenruo { 128622b398eeSQu Wenruo struct btrfs_fs_info *fs_info = inode->root->fs_info; 128722b398eeSQu Wenruo struct extent_changeset *data_reserved = NULL; 128822b398eeSQu Wenruo const u64 start = target->start; 128922b398eeSQu Wenruo const u64 len = target->len; 129022b398eeSQu Wenruo unsigned long last_index = (start + len - 1) >> PAGE_SHIFT; 129122b398eeSQu Wenruo unsigned long start_index = start >> PAGE_SHIFT; 129222b398eeSQu Wenruo unsigned long first_index = page_index(pages[0]); 129322b398eeSQu Wenruo int ret = 0; 129422b398eeSQu Wenruo int i; 129522b398eeSQu Wenruo 129622b398eeSQu Wenruo ASSERT(last_index - first_index + 1 <= nr_pages); 129722b398eeSQu Wenruo 129822b398eeSQu Wenruo ret = btrfs_delalloc_reserve_space(inode, &data_reserved, start, len); 129922b398eeSQu Wenruo if (ret < 0) 130022b398eeSQu Wenruo return ret; 130122b398eeSQu Wenruo clear_extent_bit(&inode->io_tree, start, start + len - 1, 130222b398eeSQu Wenruo EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | 130322b398eeSQu Wenruo EXTENT_DEFRAG, 0, 0, cached_state); 130422b398eeSQu Wenruo set_extent_defrag(&inode->io_tree, start, start + len - 1, cached_state); 130522b398eeSQu Wenruo 130622b398eeSQu Wenruo /* Update the page status */ 130722b398eeSQu Wenruo for (i = start_index - first_index; i <= last_index - first_index; i++) { 130822b398eeSQu Wenruo ClearPageChecked(pages[i]); 130922b398eeSQu Wenruo btrfs_page_clamp_set_dirty(fs_info, pages[i], start, len); 131022b398eeSQu Wenruo } 131122b398eeSQu Wenruo btrfs_delalloc_release_extents(inode, len); 131222b398eeSQu Wenruo extent_changeset_free(data_reserved); 131322b398eeSQu Wenruo 131422b398eeSQu Wenruo return ret; 131522b398eeSQu Wenruo } 131622b398eeSQu Wenruo 1317e9eec721SQu Wenruo static int defrag_one_range(struct btrfs_inode *inode, u64 start, u32 len, 1318e9eec721SQu Wenruo u32 extent_thresh, u64 newer_than, bool do_compress) 1319e9eec721SQu Wenruo { 1320e9eec721SQu Wenruo struct extent_state *cached_state = NULL; 1321e9eec721SQu Wenruo struct defrag_target_range *entry; 1322e9eec721SQu Wenruo struct defrag_target_range *tmp; 1323e9eec721SQu Wenruo LIST_HEAD(target_list); 1324e9eec721SQu Wenruo struct page **pages; 1325e9eec721SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1326e9eec721SQu Wenruo u64 last_index = (start + len - 1) >> PAGE_SHIFT; 1327e9eec721SQu Wenruo u64 start_index = start >> PAGE_SHIFT; 1328e9eec721SQu Wenruo unsigned int nr_pages = last_index - start_index + 1; 1329e9eec721SQu Wenruo int ret = 0; 1330e9eec721SQu Wenruo int i; 1331e9eec721SQu Wenruo 1332e9eec721SQu Wenruo ASSERT(nr_pages <= CLUSTER_SIZE / PAGE_SIZE); 1333e9eec721SQu Wenruo ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(len, sectorsize)); 1334e9eec721SQu Wenruo 1335e9eec721SQu Wenruo pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS); 1336e9eec721SQu Wenruo if (!pages) 1337e9eec721SQu Wenruo return -ENOMEM; 1338e9eec721SQu Wenruo 1339e9eec721SQu Wenruo /* Prepare all pages */ 1340e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1341e9eec721SQu Wenruo pages[i] = defrag_prepare_one_page(inode, start_index + i); 1342e9eec721SQu Wenruo if (IS_ERR(pages[i])) { 1343e9eec721SQu Wenruo ret = PTR_ERR(pages[i]); 1344e9eec721SQu Wenruo pages[i] = NULL; 1345e9eec721SQu Wenruo goto free_pages; 1346e9eec721SQu Wenruo } 1347e9eec721SQu Wenruo } 1348e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) 1349e9eec721SQu Wenruo wait_on_page_writeback(pages[i]); 1350e9eec721SQu Wenruo 1351e9eec721SQu Wenruo /* Lock the pages range */ 1352e9eec721SQu Wenruo lock_extent_bits(&inode->io_tree, start_index << PAGE_SHIFT, 1353e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1354e9eec721SQu Wenruo &cached_state); 1355e9eec721SQu Wenruo /* 1356e9eec721SQu Wenruo * Now we have a consistent view about the extent map, re-check 1357e9eec721SQu Wenruo * which range really needs to be defragged. 1358e9eec721SQu Wenruo * 1359e9eec721SQu Wenruo * And this time we have extent locked already, pass @locked = true 1360e9eec721SQu Wenruo * so that we won't relock the extent range and cause deadlock. 1361e9eec721SQu Wenruo */ 1362e9eec721SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1363e9eec721SQu Wenruo newer_than, do_compress, true, 1364e9eec721SQu Wenruo &target_list); 1365e9eec721SQu Wenruo if (ret < 0) 1366e9eec721SQu Wenruo goto unlock_extent; 1367e9eec721SQu Wenruo 1368e9eec721SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1369e9eec721SQu Wenruo ret = defrag_one_locked_target(inode, entry, pages, nr_pages, 1370e9eec721SQu Wenruo &cached_state); 1371e9eec721SQu Wenruo if (ret < 0) 1372e9eec721SQu Wenruo break; 1373e9eec721SQu Wenruo } 1374e9eec721SQu Wenruo 1375e9eec721SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1376e9eec721SQu Wenruo list_del_init(&entry->list); 1377e9eec721SQu Wenruo kfree(entry); 1378e9eec721SQu Wenruo } 1379e9eec721SQu Wenruo unlock_extent: 1380e9eec721SQu Wenruo unlock_extent_cached(&inode->io_tree, start_index << PAGE_SHIFT, 1381e9eec721SQu Wenruo (last_index << PAGE_SHIFT) + PAGE_SIZE - 1, 1382e9eec721SQu Wenruo &cached_state); 1383e9eec721SQu Wenruo free_pages: 1384e9eec721SQu Wenruo for (i = 0; i < nr_pages; i++) { 1385e9eec721SQu Wenruo if (pages[i]) { 1386e9eec721SQu Wenruo unlock_page(pages[i]); 1387e9eec721SQu Wenruo put_page(pages[i]); 1388e9eec721SQu Wenruo } 1389e9eec721SQu Wenruo } 1390e9eec721SQu Wenruo kfree(pages); 1391e9eec721SQu Wenruo return ret; 1392e9eec721SQu Wenruo } 1393e9eec721SQu Wenruo 1394b18c3ab2SQu Wenruo static int defrag_one_cluster(struct btrfs_inode *inode, 1395b18c3ab2SQu Wenruo struct file_ra_state *ra, 1396b18c3ab2SQu Wenruo u64 start, u32 len, u32 extent_thresh, 1397b18c3ab2SQu Wenruo u64 newer_than, bool do_compress, 1398b18c3ab2SQu Wenruo unsigned long *sectors_defragged, 1399b18c3ab2SQu Wenruo unsigned long max_sectors) 1400b18c3ab2SQu Wenruo { 1401b18c3ab2SQu Wenruo const u32 sectorsize = inode->root->fs_info->sectorsize; 1402b18c3ab2SQu Wenruo struct defrag_target_range *entry; 1403b18c3ab2SQu Wenruo struct defrag_target_range *tmp; 1404b18c3ab2SQu Wenruo LIST_HEAD(target_list); 1405b18c3ab2SQu Wenruo int ret; 1406b18c3ab2SQu Wenruo 1407b18c3ab2SQu Wenruo BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE)); 1408b18c3ab2SQu Wenruo ret = defrag_collect_targets(inode, start, len, extent_thresh, 1409b18c3ab2SQu Wenruo newer_than, do_compress, false, 1410b18c3ab2SQu Wenruo &target_list); 1411b18c3ab2SQu Wenruo if (ret < 0) 1412b18c3ab2SQu Wenruo goto out; 1413b18c3ab2SQu Wenruo 1414b18c3ab2SQu Wenruo list_for_each_entry(entry, &target_list, list) { 1415b18c3ab2SQu Wenruo u32 range_len = entry->len; 1416b18c3ab2SQu Wenruo 1417b18c3ab2SQu Wenruo /* Reached the limit */ 1418b18c3ab2SQu Wenruo if (max_sectors && max_sectors == *sectors_defragged) 1419b18c3ab2SQu Wenruo break; 1420b18c3ab2SQu Wenruo 1421b18c3ab2SQu Wenruo if (max_sectors) 1422b18c3ab2SQu Wenruo range_len = min_t(u32, range_len, 1423b18c3ab2SQu Wenruo (max_sectors - *sectors_defragged) * sectorsize); 1424b18c3ab2SQu Wenruo 1425b18c3ab2SQu Wenruo if (ra) 1426b18c3ab2SQu Wenruo page_cache_sync_readahead(inode->vfs_inode.i_mapping, 1427b18c3ab2SQu Wenruo ra, NULL, entry->start >> PAGE_SHIFT, 1428b18c3ab2SQu Wenruo ((entry->start + range_len - 1) >> PAGE_SHIFT) - 1429b18c3ab2SQu Wenruo (entry->start >> PAGE_SHIFT) + 1); 1430b18c3ab2SQu Wenruo /* 1431b18c3ab2SQu Wenruo * Here we may not defrag any range if holes are punched before 1432b18c3ab2SQu Wenruo * we locked the pages. 1433b18c3ab2SQu Wenruo * But that's fine, it only affects the @sectors_defragged 1434b18c3ab2SQu Wenruo * accounting. 1435b18c3ab2SQu Wenruo */ 1436b18c3ab2SQu Wenruo ret = defrag_one_range(inode, entry->start, range_len, 1437b18c3ab2SQu Wenruo extent_thresh, newer_than, do_compress); 1438b18c3ab2SQu Wenruo if (ret < 0) 1439b18c3ab2SQu Wenruo break; 1440b18c3ab2SQu Wenruo *sectors_defragged += range_len; 1441b18c3ab2SQu Wenruo } 1442b18c3ab2SQu Wenruo out: 1443b18c3ab2SQu Wenruo list_for_each_entry_safe(entry, tmp, &target_list, list) { 1444b18c3ab2SQu Wenruo list_del_init(&entry->list); 1445b18c3ab2SQu Wenruo kfree(entry); 1446b18c3ab2SQu Wenruo } 1447b18c3ab2SQu Wenruo return ret; 1448b18c3ab2SQu Wenruo } 1449b18c3ab2SQu Wenruo 14501ccc2e8aSQu Wenruo /* 14511ccc2e8aSQu Wenruo * Entry point to file defragmentation. 14521ccc2e8aSQu Wenruo * 14531ccc2e8aSQu Wenruo * @inode: inode to be defragged 14541ccc2e8aSQu Wenruo * @ra: readahead state (can be NUL) 14551ccc2e8aSQu Wenruo * @range: defrag options including range and flags 14561ccc2e8aSQu Wenruo * @newer_than: minimum transid to defrag 14571ccc2e8aSQu Wenruo * @max_to_defrag: max number of sectors to be defragged, if 0, the whole inode 14581ccc2e8aSQu Wenruo * will be defragged. 14591ccc2e8aSQu Wenruo */ 14601ccc2e8aSQu Wenruo int btrfs_defrag_file(struct inode *inode, struct file_ra_state *ra, 14614cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 14624cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 14634cb5300bSChris Mason { 14640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 14657b508037SQu Wenruo unsigned long sectors_defragged = 0; 1466151a31b2SLi Zefan u64 isize = i_size_read(inode); 14677b508037SQu Wenruo u64 cur; 14687b508037SQu Wenruo u64 last_byte; 14691e2ef46dSDavid Sterba bool do_compress = range->flags & BTRFS_DEFRAG_RANGE_COMPRESS; 14701ccc2e8aSQu Wenruo bool ra_allocated = false; 14717b508037SQu Wenruo int compress_type = BTRFS_COMPRESS_ZLIB; 14727b508037SQu Wenruo int ret = 0; 14737b508037SQu Wenruo u32 extent_thresh = range->extent_thresh; 14744cb5300bSChris Mason 14750abd5b17SLiu Bo if (isize == 0) 14760abd5b17SLiu Bo return 0; 14770abd5b17SLiu Bo 14780abd5b17SLiu Bo if (range->start >= isize) 14790abd5b17SLiu Bo return -EINVAL; 14801a419d85SLi Zefan 14811e2ef46dSDavid Sterba if (do_compress) { 1482ce96b7ffSChengguang Xu if (range->compress_type >= BTRFS_NR_COMPRESS_TYPES) 14831a419d85SLi Zefan return -EINVAL; 14841a419d85SLi Zefan if (range->compress_type) 14851a419d85SLi Zefan compress_type = range->compress_type; 14861a419d85SLi Zefan } 1487f46b5a66SChristoph Hellwig 14880abd5b17SLiu Bo if (extent_thresh == 0) 1489ee22184bSByongho Lee extent_thresh = SZ_256K; 1490f46b5a66SChristoph Hellwig 14917b508037SQu Wenruo if (range->start + range->len > range->start) { 14927b508037SQu Wenruo /* Got a specific range */ 14937b508037SQu Wenruo last_byte = min(isize, range->start + range->len) - 1; 14947b508037SQu Wenruo } else { 14957b508037SQu Wenruo /* Defrag until file end */ 14967b508037SQu Wenruo last_byte = isize - 1; 14977b508037SQu Wenruo } 14987b508037SQu Wenruo 14994cb5300bSChris Mason /* 15001ccc2e8aSQu Wenruo * If we were not given a ra, allocate a readahead context. As 15010a52d108SDavid Sterba * readahead is just an optimization, defrag will work without it so 15020a52d108SDavid Sterba * we don't error out. 15034cb5300bSChris Mason */ 15041ccc2e8aSQu Wenruo if (!ra) { 15051ccc2e8aSQu Wenruo ra_allocated = true; 150663e727ecSDavid Sterba ra = kzalloc(sizeof(*ra), GFP_KERNEL); 15070a52d108SDavid Sterba if (ra) 15084cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 15094cb5300bSChris Mason } 15104cb5300bSChris Mason 15117b508037SQu Wenruo /* Align the range */ 15127b508037SQu Wenruo cur = round_down(range->start, fs_info->sectorsize); 15137b508037SQu Wenruo last_byte = round_up(last_byte, fs_info->sectorsize) - 1; 15144cb5300bSChris Mason 15157b508037SQu Wenruo while (cur < last_byte) { 15167b508037SQu Wenruo u64 cluster_end; 15171e701a32SChris Mason 15187b508037SQu Wenruo /* The cluster size 256K should always be page aligned */ 15197b508037SQu Wenruo BUILD_BUG_ON(!IS_ALIGNED(CLUSTER_SIZE, PAGE_SIZE)); 15204cb5300bSChris Mason 15217b508037SQu Wenruo /* We want the cluster end at page boundary when possible */ 15227b508037SQu Wenruo cluster_end = (((cur >> PAGE_SHIFT) + 15237b508037SQu Wenruo (SZ_256K >> PAGE_SHIFT)) << PAGE_SHIFT) - 1; 15247b508037SQu Wenruo cluster_end = min(cluster_end, last_byte); 15254cb5300bSChris Mason 152664708539SJosef Bacik btrfs_inode_lock(inode, 0); 1527eede2bf3SOmar Sandoval if (IS_SWAPFILE(inode)) { 1528eede2bf3SOmar Sandoval ret = -ETXTBSY; 15297b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 15307b508037SQu Wenruo break; 15317b508037SQu Wenruo } 15327b508037SQu Wenruo if (!(inode->i_sb->s_flags & SB_ACTIVE)) { 15337b508037SQu Wenruo btrfs_inode_unlock(inode, 0); 15347b508037SQu Wenruo break; 15357b508037SQu Wenruo } 15361e2ef46dSDavid Sterba if (do_compress) 1537eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = compress_type; 15387b508037SQu Wenruo ret = defrag_one_cluster(BTRFS_I(inode), ra, cur, 15397b508037SQu Wenruo cluster_end + 1 - cur, extent_thresh, 15407b508037SQu Wenruo newer_than, do_compress, 15417b508037SQu Wenruo §ors_defragged, max_to_defrag); 154264708539SJosef Bacik btrfs_inode_unlock(inode, 0); 15437b508037SQu Wenruo if (ret < 0) 15444cb5300bSChris Mason break; 15457b508037SQu Wenruo cur = cluster_end + 1; 15464cb5300bSChris Mason } 1547f46b5a66SChristoph Hellwig 15487b508037SQu Wenruo if (ra_allocated) 15497b508037SQu Wenruo kfree(ra); 15507b508037SQu Wenruo if (sectors_defragged) { 15517b508037SQu Wenruo /* 15527b508037SQu Wenruo * We have defragged some sectors, for compression case they 15537b508037SQu Wenruo * need to be written back immediately. 15547b508037SQu Wenruo */ 15557b508037SQu Wenruo if (range->flags & BTRFS_DEFRAG_RANGE_START_IO) { 15561e701a32SChris Mason filemap_flush(inode->i_mapping); 1557dec8ef90SFilipe Manana if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, 1558dec8ef90SFilipe Manana &BTRFS_I(inode)->runtime_flags)) 1559dec8ef90SFilipe Manana filemap_flush(inode->i_mapping); 1560dec8ef90SFilipe Manana } 15617b508037SQu Wenruo if (range->compress_type == BTRFS_COMPRESS_LZO) 15620b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, COMPRESS_LZO); 15637b508037SQu Wenruo else if (range->compress_type == BTRFS_COMPRESS_ZSTD) 15645c1aab1dSNick Terrell btrfs_set_fs_incompat(fs_info, COMPRESS_ZSTD); 15657b508037SQu Wenruo ret = sectors_defragged; 15661a419d85SLi Zefan } 15671e2ef46dSDavid Sterba if (do_compress) { 156864708539SJosef Bacik btrfs_inode_lock(inode, 0); 1569eec63c65SDavid Sterba BTRFS_I(inode)->defrag_compress = BTRFS_COMPRESS_NONE; 157064708539SJosef Bacik btrfs_inode_unlock(inode, 0); 1571633085c7SFilipe David Borba Manana } 1572940100a4SChris Mason return ret; 1573f46b5a66SChristoph Hellwig } 1574f46b5a66SChristoph Hellwig 157517aaa434SDavid Sterba /* 157617aaa434SDavid Sterba * Try to start exclusive operation @type or cancel it if it's running. 157717aaa434SDavid Sterba * 157817aaa434SDavid Sterba * Return: 157917aaa434SDavid Sterba * 0 - normal mode, newly claimed op started 158017aaa434SDavid Sterba * >0 - normal mode, something else is running, 158117aaa434SDavid Sterba * return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS to user space 158217aaa434SDavid Sterba * ECANCELED - cancel mode, successful cancel 158317aaa434SDavid Sterba * ENOTCONN - cancel mode, operation not running anymore 158417aaa434SDavid Sterba */ 158517aaa434SDavid Sterba static int exclop_start_or_cancel_reloc(struct btrfs_fs_info *fs_info, 158617aaa434SDavid Sterba enum btrfs_exclusive_operation type, bool cancel) 158717aaa434SDavid Sterba { 158817aaa434SDavid Sterba if (!cancel) { 158917aaa434SDavid Sterba /* Start normal op */ 159017aaa434SDavid Sterba if (!btrfs_exclop_start(fs_info, type)) 159117aaa434SDavid Sterba return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 159217aaa434SDavid Sterba /* Exclusive operation is now claimed */ 159317aaa434SDavid Sterba return 0; 159417aaa434SDavid Sterba } 159517aaa434SDavid Sterba 159617aaa434SDavid Sterba /* Cancel running op */ 159717aaa434SDavid Sterba if (btrfs_exclop_start_try_lock(fs_info, type)) { 159817aaa434SDavid Sterba /* 159917aaa434SDavid Sterba * This blocks any exclop finish from setting it to NONE, so we 160017aaa434SDavid Sterba * request cancellation. Either it runs and we will wait for it, 160117aaa434SDavid Sterba * or it has finished and no waiting will happen. 160217aaa434SDavid Sterba */ 160317aaa434SDavid Sterba atomic_inc(&fs_info->reloc_cancel_req); 160417aaa434SDavid Sterba btrfs_exclop_start_unlock(fs_info); 160517aaa434SDavid Sterba 160617aaa434SDavid Sterba if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 160717aaa434SDavid Sterba wait_on_bit(&fs_info->flags, BTRFS_FS_RELOC_RUNNING, 160817aaa434SDavid Sterba TASK_INTERRUPTIBLE); 160917aaa434SDavid Sterba 161017aaa434SDavid Sterba return -ECANCELED; 161117aaa434SDavid Sterba } 161217aaa434SDavid Sterba 161317aaa434SDavid Sterba /* Something else is running or none */ 161417aaa434SDavid Sterba return -ENOTCONN; 161517aaa434SDavid Sterba } 161617aaa434SDavid Sterba 1617198605a8SMiao Xie static noinline int btrfs_ioctl_resize(struct file *file, 161876dda93cSYan, Zheng void __user *arg) 1619f46b5a66SChristoph Hellwig { 1620562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 16210b246afaSJeff Mahoney struct inode *inode = file_inode(file); 16220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 1623f46b5a66SChristoph Hellwig u64 new_size; 1624f46b5a66SChristoph Hellwig u64 old_size; 1625f46b5a66SChristoph Hellwig u64 devid = 1; 16260b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 1627f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1628f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1629f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1630f46b5a66SChristoph Hellwig char *sizestr; 16319a40f122SGui Hecheng char *retptr; 1632f46b5a66SChristoph Hellwig char *devstr = NULL; 1633f46b5a66SChristoph Hellwig int ret = 0; 1634f46b5a66SChristoph Hellwig int mod = 0; 1635bb059a37SDavid Sterba bool cancel; 1636f46b5a66SChristoph Hellwig 1637e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1638e441d54dSChris Mason return -EPERM; 1639e441d54dSChris Mason 1640198605a8SMiao Xie ret = mnt_want_write_file(file); 1641198605a8SMiao Xie if (ret) 1642198605a8SMiao Xie return ret; 1643198605a8SMiao Xie 1644bb059a37SDavid Sterba /* 1645bb059a37SDavid Sterba * Read the arguments before checking exclusivity to be able to 1646bb059a37SDavid Sterba * distinguish regular resize and cancel 1647bb059a37SDavid Sterba */ 1648dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1649c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1650c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1651bb059a37SDavid Sterba goto out_drop; 1652c9e9f97bSIlya Dryomov } 16535516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1654f46b5a66SChristoph Hellwig sizestr = vol_args->name; 1655bb059a37SDavid Sterba cancel = (strcmp("cancel", sizestr) == 0); 1656bb059a37SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_RESIZE, cancel); 1657bb059a37SDavid Sterba if (ret) 1658bb059a37SDavid Sterba goto out_free; 1659bb059a37SDavid Sterba /* Exclusive operation is now claimed */ 1660bb059a37SDavid Sterba 1661f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 1662f46b5a66SChristoph Hellwig if (devstr) { 1663f46b5a66SChristoph Hellwig sizestr = devstr + 1; 1664f46b5a66SChristoph Hellwig *devstr = '\0'; 1665f46b5a66SChristoph Hellwig devstr = vol_args->name; 166658dfae63SZhangZhen ret = kstrtoull(devstr, 10, &devid); 166758dfae63SZhangZhen if (ret) 1668bb059a37SDavid Sterba goto out_finish; 1669dfd79829SMiao Xie if (!devid) { 1670dfd79829SMiao Xie ret = -EINVAL; 1671bb059a37SDavid Sterba goto out_finish; 1672dfd79829SMiao Xie } 16730b246afaSJeff Mahoney btrfs_info(fs_info, "resizing devid %llu", devid); 1674f46b5a66SChristoph Hellwig } 1675dba60f3fSMiao Xie 1676562d7b15SJosef Bacik args.devid = devid; 1677562d7b15SJosef Bacik device = btrfs_find_device(fs_info->fs_devices, &args); 1678f46b5a66SChristoph Hellwig if (!device) { 16790b246afaSJeff Mahoney btrfs_info(fs_info, "resizer unable to find device %llu", 1680c1c9ff7cSGeert Uytterhoeven devid); 1681dfd79829SMiao Xie ret = -ENODEV; 1682bb059a37SDavid Sterba goto out_finish; 1683f46b5a66SChristoph Hellwig } 1684dba60f3fSMiao Xie 1685ebbede42SAnand Jain if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { 16860b246afaSJeff Mahoney btrfs_info(fs_info, 1687efe120a0SFrank Holton "resizer unable to apply on readonly device %llu", 1688c1c9ff7cSGeert Uytterhoeven devid); 1689dfd79829SMiao Xie ret = -EPERM; 1690bb059a37SDavid Sterba goto out_finish; 16914e42ae1bSLiu Bo } 16924e42ae1bSLiu Bo 1693f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 1694f46b5a66SChristoph Hellwig new_size = device->bdev->bd_inode->i_size; 1695f46b5a66SChristoph Hellwig else { 1696f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 1697f46b5a66SChristoph Hellwig mod = -1; 1698f46b5a66SChristoph Hellwig sizestr++; 1699f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 1700f46b5a66SChristoph Hellwig mod = 1; 1701f46b5a66SChristoph Hellwig sizestr++; 1702f46b5a66SChristoph Hellwig } 17039a40f122SGui Hecheng new_size = memparse(sizestr, &retptr); 17049a40f122SGui Hecheng if (*retptr != '\0' || new_size == 0) { 1705f46b5a66SChristoph Hellwig ret = -EINVAL; 1706bb059a37SDavid Sterba goto out_finish; 1707f46b5a66SChristoph Hellwig } 1708f46b5a66SChristoph Hellwig } 1709f46b5a66SChristoph Hellwig 1710401e29c1SAnand Jain if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { 1711dfd79829SMiao Xie ret = -EPERM; 1712bb059a37SDavid Sterba goto out_finish; 171363a212abSStefan Behrens } 171463a212abSStefan Behrens 17157cc8e58dSMiao Xie old_size = btrfs_device_get_total_bytes(device); 1716f46b5a66SChristoph Hellwig 1717f46b5a66SChristoph Hellwig if (mod < 0) { 1718f46b5a66SChristoph Hellwig if (new_size > old_size) { 1719f46b5a66SChristoph Hellwig ret = -EINVAL; 1720bb059a37SDavid Sterba goto out_finish; 1721f46b5a66SChristoph Hellwig } 1722f46b5a66SChristoph Hellwig new_size = old_size - new_size; 1723f46b5a66SChristoph Hellwig } else if (mod > 0) { 1724eb8052e0SWenliang Fan if (new_size > ULLONG_MAX - old_size) { 1725902c68a4SGui Hecheng ret = -ERANGE; 1726bb059a37SDavid Sterba goto out_finish; 1727eb8052e0SWenliang Fan } 1728f46b5a66SChristoph Hellwig new_size = old_size + new_size; 1729f46b5a66SChristoph Hellwig } 1730f46b5a66SChristoph Hellwig 1731ee22184bSByongho Lee if (new_size < SZ_256M) { 1732f46b5a66SChristoph Hellwig ret = -EINVAL; 1733bb059a37SDavid Sterba goto out_finish; 1734f46b5a66SChristoph Hellwig } 1735f46b5a66SChristoph Hellwig if (new_size > device->bdev->bd_inode->i_size) { 1736f46b5a66SChristoph Hellwig ret = -EFBIG; 1737bb059a37SDavid Sterba goto out_finish; 1738f46b5a66SChristoph Hellwig } 1739f46b5a66SChristoph Hellwig 174047f08b96SNikolay Borisov new_size = round_down(new_size, fs_info->sectorsize); 1741f46b5a66SChristoph Hellwig 1742f46b5a66SChristoph Hellwig if (new_size > old_size) { 1743a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 174498d5dc13STsutomu Itoh if (IS_ERR(trans)) { 174598d5dc13STsutomu Itoh ret = PTR_ERR(trans); 1746bb059a37SDavid Sterba goto out_finish; 174798d5dc13STsutomu Itoh } 1748f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 17493a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 1750ece7d20eSMike Fleetwood } else if (new_size < old_size) { 1751f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 17520253f40eSjeff.liu } /* equal, nothing need to do */ 1753f46b5a66SChristoph Hellwig 1754faf8f7b9SMarcos Paulo de Souza if (ret == 0 && new_size != old_size) 1755faf8f7b9SMarcos Paulo de Souza btrfs_info_in_rcu(fs_info, 1756faf8f7b9SMarcos Paulo de Souza "resize device %s (devid %llu) from %llu to %llu", 1757faf8f7b9SMarcos Paulo de Souza rcu_str_deref(device->name), device->devid, 1758faf8f7b9SMarcos Paulo de Souza old_size, new_size); 1759bb059a37SDavid Sterba out_finish: 1760bb059a37SDavid Sterba btrfs_exclop_finish(fs_info); 1761c9e9f97bSIlya Dryomov out_free: 1762f46b5a66SChristoph Hellwig kfree(vol_args); 1763bb059a37SDavid Sterba out_drop: 176418f39c41SIlya Dryomov mnt_drop_write_file(file); 1765f46b5a66SChristoph Hellwig return ret; 1766f46b5a66SChristoph Hellwig } 1767f46b5a66SChristoph Hellwig 17685d54c67eSNikolay Borisov static noinline int __btrfs_ioctl_snap_create(struct file *file, 17694d4340c9SChristian Brauner struct user_namespace *mnt_userns, 177052f75f4fSDavid Sterba const char *name, unsigned long fd, int subvol, 17715d54c67eSNikolay Borisov bool readonly, 17728696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 1773f46b5a66SChristoph Hellwig { 1774f46b5a66SChristoph Hellwig int namelen; 17753de4586cSChris Mason int ret = 0; 1776f46b5a66SChristoph Hellwig 1777325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 1778325c50e3SJeff Mahoney return -ENOTDIR; 1779325c50e3SJeff Mahoney 1780a874a63eSLiu Bo ret = mnt_want_write_file(file); 1781a874a63eSLiu Bo if (ret) 1782a874a63eSLiu Bo goto out; 1783a874a63eSLiu Bo 178472fd032eSSage Weil namelen = strlen(name); 178572fd032eSSage Weil if (strchr(name, '/')) { 1786f46b5a66SChristoph Hellwig ret = -EINVAL; 1787a874a63eSLiu Bo goto out_drop_write; 1788f46b5a66SChristoph Hellwig } 1789f46b5a66SChristoph Hellwig 179016780cabSChris Mason if (name[0] == '.' && 179116780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 179216780cabSChris Mason ret = -EEXIST; 1793a874a63eSLiu Bo goto out_drop_write; 179416780cabSChris Mason } 179516780cabSChris Mason 17963de4586cSChris Mason if (subvol) { 17974d4340c9SChristian Brauner ret = btrfs_mksubvol(&file->f_path, mnt_userns, name, 17984d4340c9SChristian Brauner namelen, NULL, readonly, inherit); 1799cb8e7090SChristoph Hellwig } else { 18002903ff01SAl Viro struct fd src = fdget(fd); 18013de4586cSChris Mason struct inode *src_inode; 18022903ff01SAl Viro if (!src.file) { 18033de4586cSChris Mason ret = -EINVAL; 1804a874a63eSLiu Bo goto out_drop_write; 18053de4586cSChris Mason } 18063de4586cSChris Mason 1807496ad9aaSAl Viro src_inode = file_inode(src.file); 1808496ad9aaSAl Viro if (src_inode->i_sb != file_inode(file)->i_sb) { 1809c79b4713SJosef Bacik btrfs_info(BTRFS_I(file_inode(file))->root->fs_info, 1810efe120a0SFrank Holton "Snapshot src from another FS"); 181123ad5b17SKusanagi Kouichi ret = -EXDEV; 18124d4340c9SChristian Brauner } else if (!inode_owner_or_capable(mnt_userns, src_inode)) { 1813d0242061SDavid Sterba /* 1814d0242061SDavid Sterba * Subvolume creation is not restricted, but snapshots 1815d0242061SDavid Sterba * are limited to own subvolumes only 1816d0242061SDavid Sterba */ 1817d0242061SDavid Sterba ret = -EPERM; 1818ecd18815SAl Viro } else { 18194d4340c9SChristian Brauner ret = btrfs_mksnapshot(&file->f_path, mnt_userns, 18204d4340c9SChristian Brauner name, namelen, 182172fd032eSSage Weil BTRFS_I(src_inode)->root, 18225d54c67eSNikolay Borisov readonly, inherit); 1823ecd18815SAl Viro } 18242903ff01SAl Viro fdput(src); 1825cb8e7090SChristoph Hellwig } 1826a874a63eSLiu Bo out_drop_write: 1827a874a63eSLiu Bo mnt_drop_write_file(file); 1828f46b5a66SChristoph Hellwig out: 182972fd032eSSage Weil return ret; 183072fd032eSSage Weil } 183172fd032eSSage Weil 183272fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 1833fa0d2b9bSLi Zefan void __user *arg, int subvol) 183472fd032eSSage Weil { 1835fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 183672fd032eSSage Weil int ret; 183772fd032eSSage Weil 1838325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 1839325c50e3SJeff Mahoney return -ENOTDIR; 1840325c50e3SJeff Mahoney 1841fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1842fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1843fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1844fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1845fa0d2b9bSLi Zefan 18464d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 18474d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 18484d4340c9SChristian Brauner false, NULL); 1849fa0d2b9bSLi Zefan 1850fa0d2b9bSLi Zefan kfree(vol_args); 1851fa0d2b9bSLi Zefan return ret; 1852fa0d2b9bSLi Zefan } 1853fa0d2b9bSLi Zefan 1854fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1855fa0d2b9bSLi Zefan void __user *arg, int subvol) 1856fa0d2b9bSLi Zefan { 1857fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 1858fa0d2b9bSLi Zefan int ret; 1859b83cc969SLi Zefan bool readonly = false; 18606f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 186172fd032eSSage Weil 1862325c50e3SJeff Mahoney if (!S_ISDIR(file_inode(file)->i_mode)) 1863325c50e3SJeff Mahoney return -ENOTDIR; 1864325c50e3SJeff Mahoney 1865fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1866fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1867fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1868fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1869fdfb1e4fSLi Zefan 1870673990dbSDavid Sterba if (vol_args->flags & ~BTRFS_SUBVOL_CREATE_ARGS_MASK) { 1871b83cc969SLi Zefan ret = -EOPNOTSUPP; 1872c47ca32dSDan Carpenter goto free_args; 1873fdfb1e4fSLi Zefan } 1874fdfb1e4fSLi Zefan 1875b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1876b83cc969SLi Zefan readonly = true; 18776f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 18785011c5a6SDan Carpenter u64 nums; 18795011c5a6SDan Carpenter 18805011c5a6SDan Carpenter if (vol_args->size < sizeof(*inherit) || 18815011c5a6SDan Carpenter vol_args->size > PAGE_SIZE) { 18826f72c7e2SArne Jansen ret = -EINVAL; 1883c47ca32dSDan Carpenter goto free_args; 18846f72c7e2SArne Jansen } 18856f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 18866f72c7e2SArne Jansen if (IS_ERR(inherit)) { 18876f72c7e2SArne Jansen ret = PTR_ERR(inherit); 1888c47ca32dSDan Carpenter goto free_args; 18896f72c7e2SArne Jansen } 18905011c5a6SDan Carpenter 18915011c5a6SDan Carpenter if (inherit->num_qgroups > PAGE_SIZE || 18925011c5a6SDan Carpenter inherit->num_ref_copies > PAGE_SIZE || 18935011c5a6SDan Carpenter inherit->num_excl_copies > PAGE_SIZE) { 18945011c5a6SDan Carpenter ret = -EINVAL; 18955011c5a6SDan Carpenter goto free_inherit; 18965011c5a6SDan Carpenter } 18975011c5a6SDan Carpenter 18985011c5a6SDan Carpenter nums = inherit->num_qgroups + 2 * inherit->num_ref_copies + 18995011c5a6SDan Carpenter 2 * inherit->num_excl_copies; 19005011c5a6SDan Carpenter if (vol_args->size != struct_size(inherit, qgroups, nums)) { 19015011c5a6SDan Carpenter ret = -EINVAL; 19025011c5a6SDan Carpenter goto free_inherit; 19035011c5a6SDan Carpenter } 19046f72c7e2SArne Jansen } 190575eaa0e2SSage Weil 19064d4340c9SChristian Brauner ret = __btrfs_ioctl_snap_create(file, file_mnt_user_ns(file), 19074d4340c9SChristian Brauner vol_args->name, vol_args->fd, subvol, 19084d4340c9SChristian Brauner readonly, inherit); 1909c47ca32dSDan Carpenter if (ret) 1910c47ca32dSDan Carpenter goto free_inherit; 1911c47ca32dSDan Carpenter free_inherit: 19126f72c7e2SArne Jansen kfree(inherit); 1913c47ca32dSDan Carpenter free_args: 1914c47ca32dSDan Carpenter kfree(vol_args); 1915f46b5a66SChristoph Hellwig return ret; 1916f46b5a66SChristoph Hellwig } 1917f46b5a66SChristoph Hellwig 19180caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 19190caa102dSLi Zefan void __user *arg) 19200caa102dSLi Zefan { 1921496ad9aaSAl Viro struct inode *inode = file_inode(file); 19220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 19230caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 19240caa102dSLi Zefan int ret = 0; 19250caa102dSLi Zefan u64 flags = 0; 19260caa102dSLi Zefan 19274a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) 19280caa102dSLi Zefan return -EINVAL; 19290caa102dSLi Zefan 19300b246afaSJeff Mahoney down_read(&fs_info->subvol_sem); 19310caa102dSLi Zefan if (btrfs_root_readonly(root)) 19320caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 19330b246afaSJeff Mahoney up_read(&fs_info->subvol_sem); 19340caa102dSLi Zefan 19350caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 19360caa102dSLi Zefan ret = -EFAULT; 19370caa102dSLi Zefan 19380caa102dSLi Zefan return ret; 19390caa102dSLi Zefan } 19400caa102dSLi Zefan 19410caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 19420caa102dSLi Zefan void __user *arg) 19430caa102dSLi Zefan { 1944496ad9aaSAl Viro struct inode *inode = file_inode(file); 19450b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 19460caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 19470caa102dSLi Zefan struct btrfs_trans_handle *trans; 19480caa102dSLi Zefan u64 root_flags; 19490caa102dSLi Zefan u64 flags; 19500caa102dSLi Zefan int ret = 0; 19510caa102dSLi Zefan 195239e1674fSChristian Brauner if (!inode_owner_or_capable(file_mnt_user_ns(file), inode)) 1953bd60ea0fSDavid Sterba return -EPERM; 1954bd60ea0fSDavid Sterba 1955b9ca0664SLiu Bo ret = mnt_want_write_file(file); 1956b9ca0664SLiu Bo if (ret) 1957b9ca0664SLiu Bo goto out; 19580caa102dSLi Zefan 19594a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 1960b9ca0664SLiu Bo ret = -EINVAL; 1961b9ca0664SLiu Bo goto out_drop_write; 1962b9ca0664SLiu Bo } 19630caa102dSLi Zefan 1964b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 1965b9ca0664SLiu Bo ret = -EFAULT; 1966b9ca0664SLiu Bo goto out_drop_write; 1967b9ca0664SLiu Bo } 19680caa102dSLi Zefan 1969b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 1970b9ca0664SLiu Bo ret = -EOPNOTSUPP; 1971b9ca0664SLiu Bo goto out_drop_write; 1972b9ca0664SLiu Bo } 19730caa102dSLi Zefan 19740b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 19750caa102dSLi Zefan 19760caa102dSLi Zefan /* nothing to do */ 19770caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 1978b9ca0664SLiu Bo goto out_drop_sem; 19790caa102dSLi Zefan 19800caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 19812c686537SDavid Sterba if (flags & BTRFS_SUBVOL_RDONLY) { 19820caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 19830caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 19842c686537SDavid Sterba } else { 19852c686537SDavid Sterba /* 19862c686537SDavid Sterba * Block RO -> RW transition if this subvolume is involved in 19872c686537SDavid Sterba * send 19882c686537SDavid Sterba */ 19892c686537SDavid Sterba spin_lock(&root->root_item_lock); 19902c686537SDavid Sterba if (root->send_in_progress == 0) { 19910caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 19920caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 19932c686537SDavid Sterba spin_unlock(&root->root_item_lock); 19942c686537SDavid Sterba } else { 19952c686537SDavid Sterba spin_unlock(&root->root_item_lock); 19960b246afaSJeff Mahoney btrfs_warn(fs_info, 19972c686537SDavid Sterba "Attempt to set subvolume %llu read-write during send", 19982c686537SDavid Sterba root->root_key.objectid); 19992c686537SDavid Sterba ret = -EPERM; 20002c686537SDavid Sterba goto out_drop_sem; 20012c686537SDavid Sterba } 20022c686537SDavid Sterba } 20030caa102dSLi Zefan 20040caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 20050caa102dSLi Zefan if (IS_ERR(trans)) { 20060caa102dSLi Zefan ret = PTR_ERR(trans); 20070caa102dSLi Zefan goto out_reset; 20080caa102dSLi Zefan } 20090caa102dSLi Zefan 20100b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 20110caa102dSLi Zefan &root->root_key, &root->root_item); 20129417ebc8SNikolay Borisov if (ret < 0) { 20139417ebc8SNikolay Borisov btrfs_end_transaction(trans); 20149417ebc8SNikolay Borisov goto out_reset; 20159417ebc8SNikolay Borisov } 20160caa102dSLi Zefan 20179417ebc8SNikolay Borisov ret = btrfs_commit_transaction(trans); 20189417ebc8SNikolay Borisov 20190caa102dSLi Zefan out_reset: 20200caa102dSLi Zefan if (ret) 20210caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 2022b9ca0664SLiu Bo out_drop_sem: 20230b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 2024b9ca0664SLiu Bo out_drop_write: 2025b9ca0664SLiu Bo mnt_drop_write_file(file); 2026b9ca0664SLiu Bo out: 20270caa102dSLi Zefan return ret; 20280caa102dSLi Zefan } 20290caa102dSLi Zefan 2030ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 2031ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 2032ac8e9819SChris Mason { 2033abc6e134SChris Mason struct btrfs_key test; 2034abc6e134SChris Mason int ret; 2035abc6e134SChris Mason 2036abc6e134SChris Mason test.objectid = sk->min_objectid; 2037abc6e134SChris Mason test.type = sk->min_type; 2038abc6e134SChris Mason test.offset = sk->min_offset; 2039abc6e134SChris Mason 2040abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2041abc6e134SChris Mason if (ret < 0) 2042ac8e9819SChris Mason return 0; 2043abc6e134SChris Mason 2044abc6e134SChris Mason test.objectid = sk->max_objectid; 2045abc6e134SChris Mason test.type = sk->max_type; 2046abc6e134SChris Mason test.offset = sk->max_offset; 2047abc6e134SChris Mason 2048abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 2049abc6e134SChris Mason if (ret > 0) 2050ac8e9819SChris Mason return 0; 2051ac8e9819SChris Mason return 1; 2052ac8e9819SChris Mason } 2053ac8e9819SChris Mason 2054df397565SJeff Mahoney static noinline int copy_to_sk(struct btrfs_path *path, 2055ac8e9819SChris Mason struct btrfs_key *key, 2056ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 20579b6e817dSGerhard Heift size_t *buf_size, 2058ba346b35SGerhard Heift char __user *ubuf, 2059ac8e9819SChris Mason unsigned long *sk_offset, 2060ac8e9819SChris Mason int *num_found) 2061ac8e9819SChris Mason { 2062ac8e9819SChris Mason u64 found_transid; 2063ac8e9819SChris Mason struct extent_buffer *leaf; 2064ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 2065dd81d459SNaohiro Aota struct btrfs_key test; 2066ac8e9819SChris Mason unsigned long item_off; 2067ac8e9819SChris Mason unsigned long item_len; 2068ac8e9819SChris Mason int nritems; 2069ac8e9819SChris Mason int i; 2070ac8e9819SChris Mason int slot; 2071ac8e9819SChris Mason int ret = 0; 2072ac8e9819SChris Mason 2073ac8e9819SChris Mason leaf = path->nodes[0]; 2074ac8e9819SChris Mason slot = path->slots[0]; 2075ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 2076ac8e9819SChris Mason 2077ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 2078ac8e9819SChris Mason i = nritems; 2079ac8e9819SChris Mason goto advance_key; 2080ac8e9819SChris Mason } 2081ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 2082ac8e9819SChris Mason 2083ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 2084ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 2085ac8e9819SChris Mason item_len = btrfs_item_size_nr(leaf, i); 2086ac8e9819SChris Mason 208703b71c6cSGabriel de Perthuis btrfs_item_key_to_cpu(leaf, key, i); 208803b71c6cSGabriel de Perthuis if (!key_in_sk(key, sk)) 208903b71c6cSGabriel de Perthuis continue; 209003b71c6cSGabriel de Perthuis 20919b6e817dSGerhard Heift if (sizeof(sh) + item_len > *buf_size) { 20928f5f6178SGerhard Heift if (*num_found) { 2093ac8e9819SChris Mason ret = 1; 20948f5f6178SGerhard Heift goto out; 20958f5f6178SGerhard Heift } 20968f5f6178SGerhard Heift 20978f5f6178SGerhard Heift /* 20988f5f6178SGerhard Heift * return one empty item back for v1, which does not 20998f5f6178SGerhard Heift * handle -EOVERFLOW 21008f5f6178SGerhard Heift */ 21018f5f6178SGerhard Heift 21029b6e817dSGerhard Heift *buf_size = sizeof(sh) + item_len; 2103ac8e9819SChris Mason item_len = 0; 21048f5f6178SGerhard Heift ret = -EOVERFLOW; 21058f5f6178SGerhard Heift } 2106ac8e9819SChris Mason 21079b6e817dSGerhard Heift if (sizeof(sh) + item_len + *sk_offset > *buf_size) { 2108ac8e9819SChris Mason ret = 1; 210925c9bc2eSGerhard Heift goto out; 2110ac8e9819SChris Mason } 2111ac8e9819SChris Mason 2112ac8e9819SChris Mason sh.objectid = key->objectid; 2113ac8e9819SChris Mason sh.offset = key->offset; 2114ac8e9819SChris Mason sh.type = key->type; 2115ac8e9819SChris Mason sh.len = item_len; 2116ac8e9819SChris Mason sh.transid = found_transid; 2117ac8e9819SChris Mason 2118a48b73ecSJosef Bacik /* 2119a48b73ecSJosef Bacik * Copy search result header. If we fault then loop again so we 2120a48b73ecSJosef Bacik * can fault in the pages and -EFAULT there if there's a 2121a48b73ecSJosef Bacik * problem. Otherwise we'll fault and then copy the buffer in 2122a48b73ecSJosef Bacik * properly this next time through 2123a48b73ecSJosef Bacik */ 2124a48b73ecSJosef Bacik if (copy_to_user_nofault(ubuf + *sk_offset, &sh, sizeof(sh))) { 2125a48b73ecSJosef Bacik ret = 0; 2126ba346b35SGerhard Heift goto out; 2127ba346b35SGerhard Heift } 2128ba346b35SGerhard Heift 2129ac8e9819SChris Mason *sk_offset += sizeof(sh); 2130ac8e9819SChris Mason 2131ac8e9819SChris Mason if (item_len) { 2132ba346b35SGerhard Heift char __user *up = ubuf + *sk_offset; 2133a48b73ecSJosef Bacik /* 2134a48b73ecSJosef Bacik * Copy the item, same behavior as above, but reset the 2135a48b73ecSJosef Bacik * * sk_offset so we copy the full thing again. 2136a48b73ecSJosef Bacik */ 2137a48b73ecSJosef Bacik if (read_extent_buffer_to_user_nofault(leaf, up, 2138ba346b35SGerhard Heift item_off, item_len)) { 2139a48b73ecSJosef Bacik ret = 0; 2140a48b73ecSJosef Bacik *sk_offset -= sizeof(sh); 2141ba346b35SGerhard Heift goto out; 2142ba346b35SGerhard Heift } 2143ba346b35SGerhard Heift 2144ac8e9819SChris Mason *sk_offset += item_len; 2145ac8e9819SChris Mason } 2146e2156867SHugo Mills (*num_found)++; 2147ac8e9819SChris Mason 21488f5f6178SGerhard Heift if (ret) /* -EOVERFLOW from above */ 21498f5f6178SGerhard Heift goto out; 21508f5f6178SGerhard Heift 215125c9bc2eSGerhard Heift if (*num_found >= sk->nr_items) { 215225c9bc2eSGerhard Heift ret = 1; 215325c9bc2eSGerhard Heift goto out; 215425c9bc2eSGerhard Heift } 2155ac8e9819SChris Mason } 2156ac8e9819SChris Mason advance_key: 2157ac8e9819SChris Mason ret = 0; 2158dd81d459SNaohiro Aota test.objectid = sk->max_objectid; 2159dd81d459SNaohiro Aota test.type = sk->max_type; 2160dd81d459SNaohiro Aota test.offset = sk->max_offset; 2161dd81d459SNaohiro Aota if (btrfs_comp_cpu_keys(key, &test) >= 0) 2162dd81d459SNaohiro Aota ret = 1; 2163dd81d459SNaohiro Aota else if (key->offset < (u64)-1) 2164abc6e134SChris Mason key->offset++; 2165dd81d459SNaohiro Aota else if (key->type < (u8)-1) { 2166abc6e134SChris Mason key->offset = 0; 2167abc6e134SChris Mason key->type++; 2168dd81d459SNaohiro Aota } else if (key->objectid < (u64)-1) { 2169abc6e134SChris Mason key->offset = 0; 2170abc6e134SChris Mason key->type = 0; 2171abc6e134SChris Mason key->objectid++; 2172abc6e134SChris Mason } else 2173abc6e134SChris Mason ret = 1; 217425c9bc2eSGerhard Heift out: 2175ba346b35SGerhard Heift /* 2176ba346b35SGerhard Heift * 0: all items from this leaf copied, continue with next 2177ba346b35SGerhard Heift * 1: * more items can be copied, but unused buffer is too small 2178ba346b35SGerhard Heift * * all items were found 2179ba346b35SGerhard Heift * Either way, it will stops the loop which iterates to the next 2180ba346b35SGerhard Heift * leaf 2181ba346b35SGerhard Heift * -EOVERFLOW: item was to large for buffer 2182ba346b35SGerhard Heift * -EFAULT: could not copy extent buffer back to userspace 2183ba346b35SGerhard Heift */ 2184ac8e9819SChris Mason return ret; 2185ac8e9819SChris Mason } 2186ac8e9819SChris Mason 2187ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 218812544442SGerhard Heift struct btrfs_ioctl_search_key *sk, 21899b6e817dSGerhard Heift size_t *buf_size, 2190ba346b35SGerhard Heift char __user *ubuf) 2191ac8e9819SChris Mason { 21920b246afaSJeff Mahoney struct btrfs_fs_info *info = btrfs_sb(inode->i_sb); 2193ac8e9819SChris Mason struct btrfs_root *root; 2194ac8e9819SChris Mason struct btrfs_key key; 2195ac8e9819SChris Mason struct btrfs_path *path; 2196ac8e9819SChris Mason int ret; 2197ac8e9819SChris Mason int num_found = 0; 2198ac8e9819SChris Mason unsigned long sk_offset = 0; 2199ac8e9819SChris Mason 22009b6e817dSGerhard Heift if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) { 22019b6e817dSGerhard Heift *buf_size = sizeof(struct btrfs_ioctl_search_header); 220212544442SGerhard Heift return -EOVERFLOW; 22039b6e817dSGerhard Heift } 220412544442SGerhard Heift 2205ac8e9819SChris Mason path = btrfs_alloc_path(); 2206ac8e9819SChris Mason if (!path) 2207ac8e9819SChris Mason return -ENOMEM; 2208ac8e9819SChris Mason 2209ac8e9819SChris Mason if (sk->tree_id == 0) { 2210ac8e9819SChris Mason /* search the root of the inode that was passed */ 221100246528SJosef Bacik root = btrfs_grab_root(BTRFS_I(inode)->root); 2212ac8e9819SChris Mason } else { 221356e9357aSDavid Sterba root = btrfs_get_fs_root(info, sk->tree_id, true); 2214ac8e9819SChris Mason if (IS_ERR(root)) { 2215ac8e9819SChris Mason btrfs_free_path(path); 2216ad1e3d56SMisono Tomohiro return PTR_ERR(root); 2217ac8e9819SChris Mason } 2218ac8e9819SChris Mason } 2219ac8e9819SChris Mason 2220ac8e9819SChris Mason key.objectid = sk->min_objectid; 2221ac8e9819SChris Mason key.type = sk->min_type; 2222ac8e9819SChris Mason key.offset = sk->min_offset; 2223ac8e9819SChris Mason 2224ac8e9819SChris Mason while (1) { 22251c78544eSFilipe Manana ret = fault_in_pages_writeable(ubuf + sk_offset, 22261c78544eSFilipe Manana *buf_size - sk_offset); 2227a48b73ecSJosef Bacik if (ret) 2228a48b73ecSJosef Bacik break; 2229a48b73ecSJosef Bacik 22306174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &key, path, sk->min_transid); 2231ac8e9819SChris Mason if (ret != 0) { 2232ac8e9819SChris Mason if (ret > 0) 2233ac8e9819SChris Mason ret = 0; 2234ac8e9819SChris Mason goto err; 2235ac8e9819SChris Mason } 2236df397565SJeff Mahoney ret = copy_to_sk(path, &key, sk, buf_size, ubuf, 2237ac8e9819SChris Mason &sk_offset, &num_found); 2238b3b4aa74SDavid Sterba btrfs_release_path(path); 223925c9bc2eSGerhard Heift if (ret) 2240ac8e9819SChris Mason break; 2241ac8e9819SChris Mason 2242ac8e9819SChris Mason } 22438f5f6178SGerhard Heift if (ret > 0) 2244ac8e9819SChris Mason ret = 0; 2245ac8e9819SChris Mason err: 2246ac8e9819SChris Mason sk->nr_items = num_found; 224700246528SJosef Bacik btrfs_put_root(root); 2248ac8e9819SChris Mason btrfs_free_path(path); 2249ac8e9819SChris Mason return ret; 2250ac8e9819SChris Mason } 2251ac8e9819SChris Mason 2252ac8e9819SChris Mason static noinline int btrfs_ioctl_tree_search(struct file *file, 2253ac8e9819SChris Mason void __user *argp) 2254ac8e9819SChris Mason { 2255ba346b35SGerhard Heift struct btrfs_ioctl_search_args __user *uargs; 2256ba346b35SGerhard Heift struct btrfs_ioctl_search_key sk; 2257ac8e9819SChris Mason struct inode *inode; 2258ac8e9819SChris Mason int ret; 22599b6e817dSGerhard Heift size_t buf_size; 2260ac8e9819SChris Mason 2261ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2262ac8e9819SChris Mason return -EPERM; 2263ac8e9819SChris Mason 2264ba346b35SGerhard Heift uargs = (struct btrfs_ioctl_search_args __user *)argp; 2265ac8e9819SChris Mason 2266ba346b35SGerhard Heift if (copy_from_user(&sk, &uargs->key, sizeof(sk))) 2267ba346b35SGerhard Heift return -EFAULT; 2268ba346b35SGerhard Heift 2269ba346b35SGerhard Heift buf_size = sizeof(uargs->buf); 2270ac8e9819SChris Mason 2271496ad9aaSAl Viro inode = file_inode(file); 2272ba346b35SGerhard Heift ret = search_ioctl(inode, &sk, &buf_size, uargs->buf); 22738f5f6178SGerhard Heift 22748f5f6178SGerhard Heift /* 22758f5f6178SGerhard Heift * In the origin implementation an overflow is handled by returning a 22768f5f6178SGerhard Heift * search header with a len of zero, so reset ret. 22778f5f6178SGerhard Heift */ 22788f5f6178SGerhard Heift if (ret == -EOVERFLOW) 22798f5f6178SGerhard Heift ret = 0; 22808f5f6178SGerhard Heift 2281ba346b35SGerhard Heift if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk))) 2282ac8e9819SChris Mason ret = -EFAULT; 2283ac8e9819SChris Mason return ret; 2284ac8e9819SChris Mason } 2285ac8e9819SChris Mason 2286cc68a8a5SGerhard Heift static noinline int btrfs_ioctl_tree_search_v2(struct file *file, 2287cc68a8a5SGerhard Heift void __user *argp) 2288cc68a8a5SGerhard Heift { 2289cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 __user *uarg; 2290cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 args; 2291cc68a8a5SGerhard Heift struct inode *inode; 2292cc68a8a5SGerhard Heift int ret; 2293cc68a8a5SGerhard Heift size_t buf_size; 2294ee22184bSByongho Lee const size_t buf_limit = SZ_16M; 2295cc68a8a5SGerhard Heift 2296cc68a8a5SGerhard Heift if (!capable(CAP_SYS_ADMIN)) 2297cc68a8a5SGerhard Heift return -EPERM; 2298cc68a8a5SGerhard Heift 2299cc68a8a5SGerhard Heift /* copy search header and buffer size */ 2300cc68a8a5SGerhard Heift uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp; 2301cc68a8a5SGerhard Heift if (copy_from_user(&args, uarg, sizeof(args))) 2302cc68a8a5SGerhard Heift return -EFAULT; 2303cc68a8a5SGerhard Heift 2304cc68a8a5SGerhard Heift buf_size = args.buf_size; 2305cc68a8a5SGerhard Heift 2306cc68a8a5SGerhard Heift /* limit result size to 16MB */ 2307cc68a8a5SGerhard Heift if (buf_size > buf_limit) 2308cc68a8a5SGerhard Heift buf_size = buf_limit; 2309cc68a8a5SGerhard Heift 2310cc68a8a5SGerhard Heift inode = file_inode(file); 2311cc68a8a5SGerhard Heift ret = search_ioctl(inode, &args.key, &buf_size, 2312718dc5faSOmar Sandoval (char __user *)(&uarg->buf[0])); 2313cc68a8a5SGerhard Heift if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key))) 2314cc68a8a5SGerhard Heift ret = -EFAULT; 2315cc68a8a5SGerhard Heift else if (ret == -EOVERFLOW && 2316cc68a8a5SGerhard Heift copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size))) 2317cc68a8a5SGerhard Heift ret = -EFAULT; 2318cc68a8a5SGerhard Heift 231976dda93cSYan, Zheng return ret; 232076dda93cSYan, Zheng } 232176dda93cSYan, Zheng 232298d377a0STARUISI Hiroaki /* 2323ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 2324ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 232598d377a0STARUISI Hiroaki */ 232698d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 232798d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 232898d377a0STARUISI Hiroaki { 232998d377a0STARUISI Hiroaki struct btrfs_root *root; 233098d377a0STARUISI Hiroaki struct btrfs_key key; 2331ac8e9819SChris Mason char *ptr; 233298d377a0STARUISI Hiroaki int ret = -1; 233398d377a0STARUISI Hiroaki int slot; 233498d377a0STARUISI Hiroaki int len; 233598d377a0STARUISI Hiroaki int total_len = 0; 233698d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 233798d377a0STARUISI Hiroaki struct extent_buffer *l; 233898d377a0STARUISI Hiroaki struct btrfs_path *path; 233998d377a0STARUISI Hiroaki 234098d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 234198d377a0STARUISI Hiroaki name[0]='\0'; 234298d377a0STARUISI Hiroaki return 0; 234398d377a0STARUISI Hiroaki } 234498d377a0STARUISI Hiroaki 234598d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 234698d377a0STARUISI Hiroaki if (!path) 234798d377a0STARUISI Hiroaki return -ENOMEM; 234898d377a0STARUISI Hiroaki 2349c8bcbfbdSNikolay Borisov ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX - 1]; 235098d377a0STARUISI Hiroaki 235156e9357aSDavid Sterba root = btrfs_get_fs_root(info, tree_id, true); 235298d377a0STARUISI Hiroaki if (IS_ERR(root)) { 2353ad1e3d56SMisono Tomohiro ret = PTR_ERR(root); 235488234012SJosef Bacik root = NULL; 235588234012SJosef Bacik goto out; 235688234012SJosef Bacik } 235798d377a0STARUISI Hiroaki 235898d377a0STARUISI Hiroaki key.objectid = dirid; 235998d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 23608ad6fcabSChris Mason key.offset = (u64)-1; 236198d377a0STARUISI Hiroaki 236298d377a0STARUISI Hiroaki while (1) { 23630ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 236418674c6cSFilipe David Borba Manana if (ret < 0) 236518674c6cSFilipe David Borba Manana goto out; 236618674c6cSFilipe David Borba Manana else if (ret > 0) { 2367ac8e9819SChris Mason ret = -ENOENT; 236898d377a0STARUISI Hiroaki goto out; 2369ac8e9819SChris Mason } 237018674c6cSFilipe David Borba Manana 237118674c6cSFilipe David Borba Manana l = path->nodes[0]; 237218674c6cSFilipe David Borba Manana slot = path->slots[0]; 237398d377a0STARUISI Hiroaki 237498d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 237598d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 237698d377a0STARUISI Hiroaki ptr -= len + 1; 237798d377a0STARUISI Hiroaki total_len += len + 1; 2378a696cf35SFilipe David Borba Manana if (ptr < name) { 2379a696cf35SFilipe David Borba Manana ret = -ENAMETOOLONG; 238098d377a0STARUISI Hiroaki goto out; 2381a696cf35SFilipe David Borba Manana } 238298d377a0STARUISI Hiroaki 238398d377a0STARUISI Hiroaki *(ptr + len) = '/'; 238498d377a0STARUISI Hiroaki read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 238598d377a0STARUISI Hiroaki 238698d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 238798d377a0STARUISI Hiroaki break; 238898d377a0STARUISI Hiroaki 2389b3b4aa74SDavid Sterba btrfs_release_path(path); 239098d377a0STARUISI Hiroaki key.objectid = key.offset; 23918ad6fcabSChris Mason key.offset = (u64)-1; 239298d377a0STARUISI Hiroaki dirid = key.objectid; 239398d377a0STARUISI Hiroaki } 239477906a50SLi Zefan memmove(name, ptr, total_len); 239598d377a0STARUISI Hiroaki name[total_len] = '\0'; 239698d377a0STARUISI Hiroaki ret = 0; 239798d377a0STARUISI Hiroaki out: 239800246528SJosef Bacik btrfs_put_root(root); 239998d377a0STARUISI Hiroaki btrfs_free_path(path); 2400ac8e9819SChris Mason return ret; 2401ac8e9819SChris Mason } 2402ac8e9819SChris Mason 24036623d9a0SChristian Brauner static int btrfs_search_path_in_tree_user(struct user_namespace *mnt_userns, 24046623d9a0SChristian Brauner struct inode *inode, 240523d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args) 240623d0b79dSTomohiro Misono { 240723d0b79dSTomohiro Misono struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info; 240823d0b79dSTomohiro Misono struct super_block *sb = inode->i_sb; 240923d0b79dSTomohiro Misono struct btrfs_key upper_limit = BTRFS_I(inode)->location; 241023d0b79dSTomohiro Misono u64 treeid = BTRFS_I(inode)->root->root_key.objectid; 241123d0b79dSTomohiro Misono u64 dirid = args->dirid; 241223d0b79dSTomohiro Misono unsigned long item_off; 241323d0b79dSTomohiro Misono unsigned long item_len; 241423d0b79dSTomohiro Misono struct btrfs_inode_ref *iref; 241523d0b79dSTomohiro Misono struct btrfs_root_ref *rref; 2416b8a49ae1SJosef Bacik struct btrfs_root *root = NULL; 241723d0b79dSTomohiro Misono struct btrfs_path *path; 241823d0b79dSTomohiro Misono struct btrfs_key key, key2; 241923d0b79dSTomohiro Misono struct extent_buffer *leaf; 242023d0b79dSTomohiro Misono struct inode *temp_inode; 242123d0b79dSTomohiro Misono char *ptr; 242223d0b79dSTomohiro Misono int slot; 242323d0b79dSTomohiro Misono int len; 242423d0b79dSTomohiro Misono int total_len = 0; 242523d0b79dSTomohiro Misono int ret; 242623d0b79dSTomohiro Misono 242723d0b79dSTomohiro Misono path = btrfs_alloc_path(); 242823d0b79dSTomohiro Misono if (!path) 242923d0b79dSTomohiro Misono return -ENOMEM; 243023d0b79dSTomohiro Misono 243123d0b79dSTomohiro Misono /* 243223d0b79dSTomohiro Misono * If the bottom subvolume does not exist directly under upper_limit, 243323d0b79dSTomohiro Misono * construct the path in from the bottom up. 243423d0b79dSTomohiro Misono */ 243523d0b79dSTomohiro Misono if (dirid != upper_limit.objectid) { 243623d0b79dSTomohiro Misono ptr = &args->path[BTRFS_INO_LOOKUP_USER_PATH_MAX - 1]; 243723d0b79dSTomohiro Misono 243856e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, treeid, true); 243923d0b79dSTomohiro Misono if (IS_ERR(root)) { 244023d0b79dSTomohiro Misono ret = PTR_ERR(root); 244123d0b79dSTomohiro Misono goto out; 244223d0b79dSTomohiro Misono } 244323d0b79dSTomohiro Misono 244423d0b79dSTomohiro Misono key.objectid = dirid; 244523d0b79dSTomohiro Misono key.type = BTRFS_INODE_REF_KEY; 244623d0b79dSTomohiro Misono key.offset = (u64)-1; 244723d0b79dSTomohiro Misono while (1) { 24480ff40a91SMarcos Paulo de Souza ret = btrfs_search_backwards(root, &key, path); 24490ff40a91SMarcos Paulo de Souza if (ret < 0) 2450b8a49ae1SJosef Bacik goto out_put; 24510ff40a91SMarcos Paulo de Souza else if (ret > 0) { 245223d0b79dSTomohiro Misono ret = -ENOENT; 2453b8a49ae1SJosef Bacik goto out_put; 245423d0b79dSTomohiro Misono } 245523d0b79dSTomohiro Misono 245623d0b79dSTomohiro Misono leaf = path->nodes[0]; 245723d0b79dSTomohiro Misono slot = path->slots[0]; 245823d0b79dSTomohiro Misono 245923d0b79dSTomohiro Misono iref = btrfs_item_ptr(leaf, slot, struct btrfs_inode_ref); 246023d0b79dSTomohiro Misono len = btrfs_inode_ref_name_len(leaf, iref); 246123d0b79dSTomohiro Misono ptr -= len + 1; 246223d0b79dSTomohiro Misono total_len += len + 1; 246323d0b79dSTomohiro Misono if (ptr < args->path) { 246423d0b79dSTomohiro Misono ret = -ENAMETOOLONG; 2465b8a49ae1SJosef Bacik goto out_put; 246623d0b79dSTomohiro Misono } 246723d0b79dSTomohiro Misono 246823d0b79dSTomohiro Misono *(ptr + len) = '/'; 246923d0b79dSTomohiro Misono read_extent_buffer(leaf, ptr, 247023d0b79dSTomohiro Misono (unsigned long)(iref + 1), len); 247123d0b79dSTomohiro Misono 247223d0b79dSTomohiro Misono /* Check the read+exec permission of this directory */ 247323d0b79dSTomohiro Misono ret = btrfs_previous_item(root, path, dirid, 247423d0b79dSTomohiro Misono BTRFS_INODE_ITEM_KEY); 247523d0b79dSTomohiro Misono if (ret < 0) { 2476b8a49ae1SJosef Bacik goto out_put; 247723d0b79dSTomohiro Misono } else if (ret > 0) { 247823d0b79dSTomohiro Misono ret = -ENOENT; 2479b8a49ae1SJosef Bacik goto out_put; 248023d0b79dSTomohiro Misono } 248123d0b79dSTomohiro Misono 248223d0b79dSTomohiro Misono leaf = path->nodes[0]; 248323d0b79dSTomohiro Misono slot = path->slots[0]; 248423d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key2, slot); 248523d0b79dSTomohiro Misono if (key2.objectid != dirid) { 248623d0b79dSTomohiro Misono ret = -ENOENT; 2487b8a49ae1SJosef Bacik goto out_put; 248823d0b79dSTomohiro Misono } 248923d0b79dSTomohiro Misono 24900202e83fSDavid Sterba temp_inode = btrfs_iget(sb, key2.objectid, root); 24913ca57bd6SMisono Tomohiro if (IS_ERR(temp_inode)) { 24923ca57bd6SMisono Tomohiro ret = PTR_ERR(temp_inode); 2493b8a49ae1SJosef Bacik goto out_put; 24943ca57bd6SMisono Tomohiro } 24956623d9a0SChristian Brauner ret = inode_permission(mnt_userns, temp_inode, 249647291baaSChristian Brauner MAY_READ | MAY_EXEC); 249723d0b79dSTomohiro Misono iput(temp_inode); 249823d0b79dSTomohiro Misono if (ret) { 249923d0b79dSTomohiro Misono ret = -EACCES; 2500b8a49ae1SJosef Bacik goto out_put; 250123d0b79dSTomohiro Misono } 250223d0b79dSTomohiro Misono 250323d0b79dSTomohiro Misono if (key.offset == upper_limit.objectid) 250423d0b79dSTomohiro Misono break; 250523d0b79dSTomohiro Misono if (key.objectid == BTRFS_FIRST_FREE_OBJECTID) { 250623d0b79dSTomohiro Misono ret = -EACCES; 2507b8a49ae1SJosef Bacik goto out_put; 250823d0b79dSTomohiro Misono } 250923d0b79dSTomohiro Misono 251023d0b79dSTomohiro Misono btrfs_release_path(path); 251123d0b79dSTomohiro Misono key.objectid = key.offset; 251223d0b79dSTomohiro Misono key.offset = (u64)-1; 251323d0b79dSTomohiro Misono dirid = key.objectid; 251423d0b79dSTomohiro Misono } 251523d0b79dSTomohiro Misono 251623d0b79dSTomohiro Misono memmove(args->path, ptr, total_len); 251723d0b79dSTomohiro Misono args->path[total_len] = '\0'; 251800246528SJosef Bacik btrfs_put_root(root); 2519b8a49ae1SJosef Bacik root = NULL; 252023d0b79dSTomohiro Misono btrfs_release_path(path); 252123d0b79dSTomohiro Misono } 252223d0b79dSTomohiro Misono 252323d0b79dSTomohiro Misono /* Get the bottom subvolume's name from ROOT_REF */ 252423d0b79dSTomohiro Misono key.objectid = treeid; 252523d0b79dSTomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 252623d0b79dSTomohiro Misono key.offset = args->treeid; 2527b8a49ae1SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 252823d0b79dSTomohiro Misono if (ret < 0) { 252923d0b79dSTomohiro Misono goto out; 253023d0b79dSTomohiro Misono } else if (ret > 0) { 253123d0b79dSTomohiro Misono ret = -ENOENT; 253223d0b79dSTomohiro Misono goto out; 253323d0b79dSTomohiro Misono } 253423d0b79dSTomohiro Misono 253523d0b79dSTomohiro Misono leaf = path->nodes[0]; 253623d0b79dSTomohiro Misono slot = path->slots[0]; 253723d0b79dSTomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 253823d0b79dSTomohiro Misono 253923d0b79dSTomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot); 254023d0b79dSTomohiro Misono item_len = btrfs_item_size_nr(leaf, slot); 254123d0b79dSTomohiro Misono /* Check if dirid in ROOT_REF corresponds to passed dirid */ 254223d0b79dSTomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 254323d0b79dSTomohiro Misono if (args->dirid != btrfs_root_ref_dirid(leaf, rref)) { 254423d0b79dSTomohiro Misono ret = -EINVAL; 254523d0b79dSTomohiro Misono goto out; 254623d0b79dSTomohiro Misono } 254723d0b79dSTomohiro Misono 254823d0b79dSTomohiro Misono /* Copy subvolume's name */ 254923d0b79dSTomohiro Misono item_off += sizeof(struct btrfs_root_ref); 255023d0b79dSTomohiro Misono item_len -= sizeof(struct btrfs_root_ref); 255123d0b79dSTomohiro Misono read_extent_buffer(leaf, args->name, item_off, item_len); 255223d0b79dSTomohiro Misono args->name[item_len] = 0; 255323d0b79dSTomohiro Misono 2554b8a49ae1SJosef Bacik out_put: 255500246528SJosef Bacik btrfs_put_root(root); 255623d0b79dSTomohiro Misono out: 255723d0b79dSTomohiro Misono btrfs_free_path(path); 255823d0b79dSTomohiro Misono return ret; 255923d0b79dSTomohiro Misono } 256023d0b79dSTomohiro Misono 2561ac8e9819SChris Mason static noinline int btrfs_ioctl_ino_lookup(struct file *file, 2562ac8e9819SChris Mason void __user *argp) 2563ac8e9819SChris Mason { 2564ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 2565ac8e9819SChris Mason struct inode *inode; 256601b810b8SDavid Sterba int ret = 0; 2567ac8e9819SChris Mason 25682354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 25692354d08fSJulia Lawall if (IS_ERR(args)) 25702354d08fSJulia Lawall return PTR_ERR(args); 2571c2b96929SDan Carpenter 2572496ad9aaSAl Viro inode = file_inode(file); 2573ac8e9819SChris Mason 257401b810b8SDavid Sterba /* 257501b810b8SDavid Sterba * Unprivileged query to obtain the containing subvolume root id. The 257601b810b8SDavid Sterba * path is reset so it's consistent with btrfs_search_path_in_tree. 257701b810b8SDavid Sterba */ 25781b53ac4dSChris Mason if (args->treeid == 0) 25791b53ac4dSChris Mason args->treeid = BTRFS_I(inode)->root->root_key.objectid; 25801b53ac4dSChris Mason 258101b810b8SDavid Sterba if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) { 258201b810b8SDavid Sterba args->name[0] = 0; 258301b810b8SDavid Sterba goto out; 258401b810b8SDavid Sterba } 258501b810b8SDavid Sterba 258601b810b8SDavid Sterba if (!capable(CAP_SYS_ADMIN)) { 258701b810b8SDavid Sterba ret = -EPERM; 258801b810b8SDavid Sterba goto out; 258901b810b8SDavid Sterba } 259001b810b8SDavid Sterba 2591ac8e9819SChris Mason ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 2592ac8e9819SChris Mason args->treeid, args->objectid, 2593ac8e9819SChris Mason args->name); 2594ac8e9819SChris Mason 259501b810b8SDavid Sterba out: 2596ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2597ac8e9819SChris Mason ret = -EFAULT; 2598ac8e9819SChris Mason 2599ac8e9819SChris Mason kfree(args); 260098d377a0STARUISI Hiroaki return ret; 260198d377a0STARUISI Hiroaki } 260298d377a0STARUISI Hiroaki 260323d0b79dSTomohiro Misono /* 260423d0b79dSTomohiro Misono * Version of ino_lookup ioctl (unprivileged) 260523d0b79dSTomohiro Misono * 260623d0b79dSTomohiro Misono * The main differences from ino_lookup ioctl are: 260723d0b79dSTomohiro Misono * 260823d0b79dSTomohiro Misono * 1. Read + Exec permission will be checked using inode_permission() during 260923d0b79dSTomohiro Misono * path construction. -EACCES will be returned in case of failure. 261023d0b79dSTomohiro Misono * 2. Path construction will be stopped at the inode number which corresponds 261123d0b79dSTomohiro Misono * to the fd with which this ioctl is called. If constructed path does not 261223d0b79dSTomohiro Misono * exist under fd's inode, -EACCES will be returned. 261323d0b79dSTomohiro Misono * 3. The name of bottom subvolume is also searched and filled. 261423d0b79dSTomohiro Misono */ 261523d0b79dSTomohiro Misono static int btrfs_ioctl_ino_lookup_user(struct file *file, void __user *argp) 261623d0b79dSTomohiro Misono { 261723d0b79dSTomohiro Misono struct btrfs_ioctl_ino_lookup_user_args *args; 261823d0b79dSTomohiro Misono struct inode *inode; 261923d0b79dSTomohiro Misono int ret; 262023d0b79dSTomohiro Misono 262123d0b79dSTomohiro Misono args = memdup_user(argp, sizeof(*args)); 262223d0b79dSTomohiro Misono if (IS_ERR(args)) 262323d0b79dSTomohiro Misono return PTR_ERR(args); 262423d0b79dSTomohiro Misono 262523d0b79dSTomohiro Misono inode = file_inode(file); 262623d0b79dSTomohiro Misono 262723d0b79dSTomohiro Misono if (args->dirid == BTRFS_FIRST_FREE_OBJECTID && 262823d0b79dSTomohiro Misono BTRFS_I(inode)->location.objectid != BTRFS_FIRST_FREE_OBJECTID) { 262923d0b79dSTomohiro Misono /* 263023d0b79dSTomohiro Misono * The subvolume does not exist under fd with which this is 263123d0b79dSTomohiro Misono * called 263223d0b79dSTomohiro Misono */ 263323d0b79dSTomohiro Misono kfree(args); 263423d0b79dSTomohiro Misono return -EACCES; 263523d0b79dSTomohiro Misono } 263623d0b79dSTomohiro Misono 26376623d9a0SChristian Brauner ret = btrfs_search_path_in_tree_user(file_mnt_user_ns(file), inode, args); 263823d0b79dSTomohiro Misono 263923d0b79dSTomohiro Misono if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 264023d0b79dSTomohiro Misono ret = -EFAULT; 264123d0b79dSTomohiro Misono 264223d0b79dSTomohiro Misono kfree(args); 264323d0b79dSTomohiro Misono return ret; 264423d0b79dSTomohiro Misono } 264523d0b79dSTomohiro Misono 2646b64ec075STomohiro Misono /* Get the subvolume information in BTRFS_ROOT_ITEM and BTRFS_ROOT_BACKREF */ 2647b64ec075STomohiro Misono static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp) 2648b64ec075STomohiro Misono { 2649b64ec075STomohiro Misono struct btrfs_ioctl_get_subvol_info_args *subvol_info; 2650b64ec075STomohiro Misono struct btrfs_fs_info *fs_info; 2651b64ec075STomohiro Misono struct btrfs_root *root; 2652b64ec075STomohiro Misono struct btrfs_path *path; 2653b64ec075STomohiro Misono struct btrfs_key key; 2654b64ec075STomohiro Misono struct btrfs_root_item *root_item; 2655b64ec075STomohiro Misono struct btrfs_root_ref *rref; 2656b64ec075STomohiro Misono struct extent_buffer *leaf; 2657b64ec075STomohiro Misono unsigned long item_off; 2658b64ec075STomohiro Misono unsigned long item_len; 2659b64ec075STomohiro Misono struct inode *inode; 2660b64ec075STomohiro Misono int slot; 2661b64ec075STomohiro Misono int ret = 0; 2662b64ec075STomohiro Misono 2663b64ec075STomohiro Misono path = btrfs_alloc_path(); 2664b64ec075STomohiro Misono if (!path) 2665b64ec075STomohiro Misono return -ENOMEM; 2666b64ec075STomohiro Misono 2667b64ec075STomohiro Misono subvol_info = kzalloc(sizeof(*subvol_info), GFP_KERNEL); 2668b64ec075STomohiro Misono if (!subvol_info) { 2669b64ec075STomohiro Misono btrfs_free_path(path); 2670b64ec075STomohiro Misono return -ENOMEM; 2671b64ec075STomohiro Misono } 2672b64ec075STomohiro Misono 2673b64ec075STomohiro Misono inode = file_inode(file); 2674b64ec075STomohiro Misono fs_info = BTRFS_I(inode)->root->fs_info; 2675b64ec075STomohiro Misono 2676b64ec075STomohiro Misono /* Get root_item of inode's subvolume */ 2677b64ec075STomohiro Misono key.objectid = BTRFS_I(inode)->root->root_key.objectid; 267856e9357aSDavid Sterba root = btrfs_get_fs_root(fs_info, key.objectid, true); 2679b64ec075STomohiro Misono if (IS_ERR(root)) { 2680b64ec075STomohiro Misono ret = PTR_ERR(root); 268104734e84SJosef Bacik goto out_free; 268204734e84SJosef Bacik } 2683b64ec075STomohiro Misono root_item = &root->root_item; 2684b64ec075STomohiro Misono 2685b64ec075STomohiro Misono subvol_info->treeid = key.objectid; 2686b64ec075STomohiro Misono 2687b64ec075STomohiro Misono subvol_info->generation = btrfs_root_generation(root_item); 2688b64ec075STomohiro Misono subvol_info->flags = btrfs_root_flags(root_item); 2689b64ec075STomohiro Misono 2690b64ec075STomohiro Misono memcpy(subvol_info->uuid, root_item->uuid, BTRFS_UUID_SIZE); 2691b64ec075STomohiro Misono memcpy(subvol_info->parent_uuid, root_item->parent_uuid, 2692b64ec075STomohiro Misono BTRFS_UUID_SIZE); 2693b64ec075STomohiro Misono memcpy(subvol_info->received_uuid, root_item->received_uuid, 2694b64ec075STomohiro Misono BTRFS_UUID_SIZE); 2695b64ec075STomohiro Misono 2696b64ec075STomohiro Misono subvol_info->ctransid = btrfs_root_ctransid(root_item); 2697b64ec075STomohiro Misono subvol_info->ctime.sec = btrfs_stack_timespec_sec(&root_item->ctime); 2698b64ec075STomohiro Misono subvol_info->ctime.nsec = btrfs_stack_timespec_nsec(&root_item->ctime); 2699b64ec075STomohiro Misono 2700b64ec075STomohiro Misono subvol_info->otransid = btrfs_root_otransid(root_item); 2701b64ec075STomohiro Misono subvol_info->otime.sec = btrfs_stack_timespec_sec(&root_item->otime); 2702b64ec075STomohiro Misono subvol_info->otime.nsec = btrfs_stack_timespec_nsec(&root_item->otime); 2703b64ec075STomohiro Misono 2704b64ec075STomohiro Misono subvol_info->stransid = btrfs_root_stransid(root_item); 2705b64ec075STomohiro Misono subvol_info->stime.sec = btrfs_stack_timespec_sec(&root_item->stime); 2706b64ec075STomohiro Misono subvol_info->stime.nsec = btrfs_stack_timespec_nsec(&root_item->stime); 2707b64ec075STomohiro Misono 2708b64ec075STomohiro Misono subvol_info->rtransid = btrfs_root_rtransid(root_item); 2709b64ec075STomohiro Misono subvol_info->rtime.sec = btrfs_stack_timespec_sec(&root_item->rtime); 2710b64ec075STomohiro Misono subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item->rtime); 2711b64ec075STomohiro Misono 2712b64ec075STomohiro Misono if (key.objectid != BTRFS_FS_TREE_OBJECTID) { 2713b64ec075STomohiro Misono /* Search root tree for ROOT_BACKREF of this subvolume */ 2714b64ec075STomohiro Misono key.type = BTRFS_ROOT_BACKREF_KEY; 2715b64ec075STomohiro Misono key.offset = 0; 271604734e84SJosef Bacik ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); 2717b64ec075STomohiro Misono if (ret < 0) { 2718b64ec075STomohiro Misono goto out; 2719b64ec075STomohiro Misono } else if (path->slots[0] >= 2720b64ec075STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 272104734e84SJosef Bacik ret = btrfs_next_leaf(fs_info->tree_root, path); 2722b64ec075STomohiro Misono if (ret < 0) { 2723b64ec075STomohiro Misono goto out; 2724b64ec075STomohiro Misono } else if (ret > 0) { 2725b64ec075STomohiro Misono ret = -EUCLEAN; 2726b64ec075STomohiro Misono goto out; 2727b64ec075STomohiro Misono } 2728b64ec075STomohiro Misono } 2729b64ec075STomohiro Misono 2730b64ec075STomohiro Misono leaf = path->nodes[0]; 2731b64ec075STomohiro Misono slot = path->slots[0]; 2732b64ec075STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 2733b64ec075STomohiro Misono if (key.objectid == subvol_info->treeid && 2734b64ec075STomohiro Misono key.type == BTRFS_ROOT_BACKREF_KEY) { 2735b64ec075STomohiro Misono subvol_info->parent_id = key.offset; 2736b64ec075STomohiro Misono 2737b64ec075STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 2738b64ec075STomohiro Misono subvol_info->dirid = btrfs_root_ref_dirid(leaf, rref); 2739b64ec075STomohiro Misono 2740b64ec075STomohiro Misono item_off = btrfs_item_ptr_offset(leaf, slot) 2741b64ec075STomohiro Misono + sizeof(struct btrfs_root_ref); 2742b64ec075STomohiro Misono item_len = btrfs_item_size_nr(leaf, slot) 2743b64ec075STomohiro Misono - sizeof(struct btrfs_root_ref); 2744b64ec075STomohiro Misono read_extent_buffer(leaf, subvol_info->name, 2745b64ec075STomohiro Misono item_off, item_len); 2746b64ec075STomohiro Misono } else { 2747b64ec075STomohiro Misono ret = -ENOENT; 2748b64ec075STomohiro Misono goto out; 2749b64ec075STomohiro Misono } 2750b64ec075STomohiro Misono } 2751b64ec075STomohiro Misono 2752b64ec075STomohiro Misono if (copy_to_user(argp, subvol_info, sizeof(*subvol_info))) 2753b64ec075STomohiro Misono ret = -EFAULT; 2754b64ec075STomohiro Misono 2755b64ec075STomohiro Misono out: 275600246528SJosef Bacik btrfs_put_root(root); 275704734e84SJosef Bacik out_free: 2758b64ec075STomohiro Misono btrfs_free_path(path); 2759b091f7feSWaiman Long kfree(subvol_info); 2760b64ec075STomohiro Misono return ret; 2761b64ec075STomohiro Misono } 2762b64ec075STomohiro Misono 276342e4b520STomohiro Misono /* 276442e4b520STomohiro Misono * Return ROOT_REF information of the subvolume containing this inode 276542e4b520STomohiro Misono * except the subvolume name. 276642e4b520STomohiro Misono */ 276742e4b520STomohiro Misono static int btrfs_ioctl_get_subvol_rootref(struct file *file, void __user *argp) 276842e4b520STomohiro Misono { 276942e4b520STomohiro Misono struct btrfs_ioctl_get_subvol_rootref_args *rootrefs; 277042e4b520STomohiro Misono struct btrfs_root_ref *rref; 277142e4b520STomohiro Misono struct btrfs_root *root; 277242e4b520STomohiro Misono struct btrfs_path *path; 277342e4b520STomohiro Misono struct btrfs_key key; 277442e4b520STomohiro Misono struct extent_buffer *leaf; 277542e4b520STomohiro Misono struct inode *inode; 277642e4b520STomohiro Misono u64 objectid; 277742e4b520STomohiro Misono int slot; 277842e4b520STomohiro Misono int ret; 277942e4b520STomohiro Misono u8 found; 278042e4b520STomohiro Misono 278142e4b520STomohiro Misono path = btrfs_alloc_path(); 278242e4b520STomohiro Misono if (!path) 278342e4b520STomohiro Misono return -ENOMEM; 278442e4b520STomohiro Misono 278542e4b520STomohiro Misono rootrefs = memdup_user(argp, sizeof(*rootrefs)); 278642e4b520STomohiro Misono if (IS_ERR(rootrefs)) { 278742e4b520STomohiro Misono btrfs_free_path(path); 278842e4b520STomohiro Misono return PTR_ERR(rootrefs); 278942e4b520STomohiro Misono } 279042e4b520STomohiro Misono 279142e4b520STomohiro Misono inode = file_inode(file); 279242e4b520STomohiro Misono root = BTRFS_I(inode)->root->fs_info->tree_root; 279342e4b520STomohiro Misono objectid = BTRFS_I(inode)->root->root_key.objectid; 279442e4b520STomohiro Misono 279542e4b520STomohiro Misono key.objectid = objectid; 279642e4b520STomohiro Misono key.type = BTRFS_ROOT_REF_KEY; 279742e4b520STomohiro Misono key.offset = rootrefs->min_treeid; 279842e4b520STomohiro Misono found = 0; 279942e4b520STomohiro Misono 280042e4b520STomohiro Misono ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 280142e4b520STomohiro Misono if (ret < 0) { 280242e4b520STomohiro Misono goto out; 280342e4b520STomohiro Misono } else if (path->slots[0] >= 280442e4b520STomohiro Misono btrfs_header_nritems(path->nodes[0])) { 280542e4b520STomohiro Misono ret = btrfs_next_leaf(root, path); 280642e4b520STomohiro Misono if (ret < 0) { 280742e4b520STomohiro Misono goto out; 280842e4b520STomohiro Misono } else if (ret > 0) { 280942e4b520STomohiro Misono ret = -EUCLEAN; 281042e4b520STomohiro Misono goto out; 281142e4b520STomohiro Misono } 281242e4b520STomohiro Misono } 281342e4b520STomohiro Misono while (1) { 281442e4b520STomohiro Misono leaf = path->nodes[0]; 281542e4b520STomohiro Misono slot = path->slots[0]; 281642e4b520STomohiro Misono 281742e4b520STomohiro Misono btrfs_item_key_to_cpu(leaf, &key, slot); 281842e4b520STomohiro Misono if (key.objectid != objectid || key.type != BTRFS_ROOT_REF_KEY) { 281942e4b520STomohiro Misono ret = 0; 282042e4b520STomohiro Misono goto out; 282142e4b520STomohiro Misono } 282242e4b520STomohiro Misono 282342e4b520STomohiro Misono if (found == BTRFS_MAX_ROOTREF_BUFFER_NUM) { 282442e4b520STomohiro Misono ret = -EOVERFLOW; 282542e4b520STomohiro Misono goto out; 282642e4b520STomohiro Misono } 282742e4b520STomohiro Misono 282842e4b520STomohiro Misono rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 282942e4b520STomohiro Misono rootrefs->rootref[found].treeid = key.offset; 283042e4b520STomohiro Misono rootrefs->rootref[found].dirid = 283142e4b520STomohiro Misono btrfs_root_ref_dirid(leaf, rref); 283242e4b520STomohiro Misono found++; 283342e4b520STomohiro Misono 283442e4b520STomohiro Misono ret = btrfs_next_item(root, path); 283542e4b520STomohiro Misono if (ret < 0) { 283642e4b520STomohiro Misono goto out; 283742e4b520STomohiro Misono } else if (ret > 0) { 283842e4b520STomohiro Misono ret = -EUCLEAN; 283942e4b520STomohiro Misono goto out; 284042e4b520STomohiro Misono } 284142e4b520STomohiro Misono } 284242e4b520STomohiro Misono 284342e4b520STomohiro Misono out: 284442e4b520STomohiro Misono if (!ret || ret == -EOVERFLOW) { 284542e4b520STomohiro Misono rootrefs->num_items = found; 284642e4b520STomohiro Misono /* update min_treeid for next search */ 284742e4b520STomohiro Misono if (found) 284842e4b520STomohiro Misono rootrefs->min_treeid = 284942e4b520STomohiro Misono rootrefs->rootref[found - 1].treeid + 1; 285042e4b520STomohiro Misono if (copy_to_user(argp, rootrefs, sizeof(*rootrefs))) 285142e4b520STomohiro Misono ret = -EFAULT; 285242e4b520STomohiro Misono } 285342e4b520STomohiro Misono 285442e4b520STomohiro Misono kfree(rootrefs); 285542e4b520STomohiro Misono btrfs_free_path(path); 285642e4b520STomohiro Misono 285742e4b520STomohiro Misono return ret; 285842e4b520STomohiro Misono } 285942e4b520STomohiro Misono 286076dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 2861949964c9SMarcos Paulo de Souza void __user *arg, 2862949964c9SMarcos Paulo de Souza bool destroy_v2) 286376dda93cSYan, Zheng { 286454563d41SAl Viro struct dentry *parent = file->f_path.dentry; 28650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(parent->d_sb); 286676dda93cSYan, Zheng struct dentry *dentry; 28672b0143b5SDavid Howells struct inode *dir = d_inode(parent); 286876dda93cSYan, Zheng struct inode *inode; 286976dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 287076dda93cSYan, Zheng struct btrfs_root *dest = NULL; 2871949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args *vol_args = NULL; 2872949964c9SMarcos Paulo de Souza struct btrfs_ioctl_vol_args_v2 *vol_args2 = NULL; 2873c4ed533bSChristian Brauner struct user_namespace *mnt_userns = file_mnt_user_ns(file); 2874949964c9SMarcos Paulo de Souza char *subvol_name, *subvol_name_ptr = NULL; 2875949964c9SMarcos Paulo de Souza int subvol_namelen; 287676dda93cSYan, Zheng int err = 0; 2877949964c9SMarcos Paulo de Souza bool destroy_parent = false; 287876dda93cSYan, Zheng 2879949964c9SMarcos Paulo de Souza if (destroy_v2) { 2880949964c9SMarcos Paulo de Souza vol_args2 = memdup_user(arg, sizeof(*vol_args2)); 2881949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args2)) 2882949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args2); 2883325c50e3SJeff Mahoney 2884949964c9SMarcos Paulo de Souza if (vol_args2->flags & ~BTRFS_SUBVOL_DELETE_ARGS_MASK) { 2885949964c9SMarcos Paulo de Souza err = -EOPNOTSUPP; 2886949964c9SMarcos Paulo de Souza goto out; 2887949964c9SMarcos Paulo de Souza } 288876dda93cSYan, Zheng 2889949964c9SMarcos Paulo de Souza /* 2890949964c9SMarcos Paulo de Souza * If SPEC_BY_ID is not set, we are looking for the subvolume by 2891949964c9SMarcos Paulo de Souza * name, same as v1 currently does. 2892949964c9SMarcos Paulo de Souza */ 2893949964c9SMarcos Paulo de Souza if (!(vol_args2->flags & BTRFS_SUBVOL_SPEC_BY_ID)) { 2894949964c9SMarcos Paulo de Souza vol_args2->name[BTRFS_SUBVOL_NAME_MAX] = 0; 2895949964c9SMarcos Paulo de Souza subvol_name = vol_args2->name; 2896949964c9SMarcos Paulo de Souza 2897949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 2898949964c9SMarcos Paulo de Souza if (err) 2899949964c9SMarcos Paulo de Souza goto out; 2900949964c9SMarcos Paulo de Souza } else { 2901aabb34e7SChristian Brauner struct inode *old_dir; 2902c4ed533bSChristian Brauner 2903949964c9SMarcos Paulo de Souza if (vol_args2->subvolid < BTRFS_FIRST_FREE_OBJECTID) { 290476dda93cSYan, Zheng err = -EINVAL; 290576dda93cSYan, Zheng goto out; 290676dda93cSYan, Zheng } 290776dda93cSYan, Zheng 2908a561be71SAl Viro err = mnt_want_write_file(file); 290976dda93cSYan, Zheng if (err) 291076dda93cSYan, Zheng goto out; 291176dda93cSYan, Zheng 2912949964c9SMarcos Paulo de Souza dentry = btrfs_get_dentry(fs_info->sb, 2913949964c9SMarcos Paulo de Souza BTRFS_FIRST_FREE_OBJECTID, 2914949964c9SMarcos Paulo de Souza vol_args2->subvolid, 0, 0); 2915949964c9SMarcos Paulo de Souza if (IS_ERR(dentry)) { 2916949964c9SMarcos Paulo de Souza err = PTR_ERR(dentry); 2917949964c9SMarcos Paulo de Souza goto out_drop_write; 2918949964c9SMarcos Paulo de Souza } 2919949964c9SMarcos Paulo de Souza 2920949964c9SMarcos Paulo de Souza /* 2921949964c9SMarcos Paulo de Souza * Change the default parent since the subvolume being 2922949964c9SMarcos Paulo de Souza * deleted can be outside of the current mount point. 2923949964c9SMarcos Paulo de Souza */ 2924949964c9SMarcos Paulo de Souza parent = btrfs_get_parent(dentry); 2925949964c9SMarcos Paulo de Souza 2926949964c9SMarcos Paulo de Souza /* 2927949964c9SMarcos Paulo de Souza * At this point dentry->d_name can point to '/' if the 2928949964c9SMarcos Paulo de Souza * subvolume we want to destroy is outsite of the 2929949964c9SMarcos Paulo de Souza * current mount point, so we need to release the 2930949964c9SMarcos Paulo de Souza * current dentry and execute the lookup to return a new 2931949964c9SMarcos Paulo de Souza * one with ->d_name pointing to the 2932949964c9SMarcos Paulo de Souza * <mount point>/subvol_name. 2933949964c9SMarcos Paulo de Souza */ 2934949964c9SMarcos Paulo de Souza dput(dentry); 2935949964c9SMarcos Paulo de Souza if (IS_ERR(parent)) { 2936949964c9SMarcos Paulo de Souza err = PTR_ERR(parent); 2937949964c9SMarcos Paulo de Souza goto out_drop_write; 2938949964c9SMarcos Paulo de Souza } 2939aabb34e7SChristian Brauner old_dir = dir; 2940949964c9SMarcos Paulo de Souza dir = d_inode(parent); 2941949964c9SMarcos Paulo de Souza 2942949964c9SMarcos Paulo de Souza /* 2943949964c9SMarcos Paulo de Souza * If v2 was used with SPEC_BY_ID, a new parent was 2944949964c9SMarcos Paulo de Souza * allocated since the subvolume can be outside of the 2945949964c9SMarcos Paulo de Souza * current mount point. Later on we need to release this 2946949964c9SMarcos Paulo de Souza * new parent dentry. 2947949964c9SMarcos Paulo de Souza */ 2948949964c9SMarcos Paulo de Souza destroy_parent = true; 2949949964c9SMarcos Paulo de Souza 2950aabb34e7SChristian Brauner /* 2951aabb34e7SChristian Brauner * On idmapped mounts, deletion via subvolid is 2952aabb34e7SChristian Brauner * restricted to subvolumes that are immediate 2953aabb34e7SChristian Brauner * ancestors of the inode referenced by the file 2954aabb34e7SChristian Brauner * descriptor in the ioctl. Otherwise the idmapping 2955aabb34e7SChristian Brauner * could potentially be abused to delete subvolumes 2956aabb34e7SChristian Brauner * anywhere in the filesystem the user wouldn't be able 2957aabb34e7SChristian Brauner * to delete without an idmapped mount. 2958aabb34e7SChristian Brauner */ 2959aabb34e7SChristian Brauner if (old_dir != dir && mnt_userns != &init_user_ns) { 2960aabb34e7SChristian Brauner err = -EOPNOTSUPP; 2961aabb34e7SChristian Brauner goto free_parent; 2962aabb34e7SChristian Brauner } 2963aabb34e7SChristian Brauner 2964949964c9SMarcos Paulo de Souza subvol_name_ptr = btrfs_get_subvol_name_from_objectid( 2965949964c9SMarcos Paulo de Souza fs_info, vol_args2->subvolid); 2966949964c9SMarcos Paulo de Souza if (IS_ERR(subvol_name_ptr)) { 2967949964c9SMarcos Paulo de Souza err = PTR_ERR(subvol_name_ptr); 2968949964c9SMarcos Paulo de Souza goto free_parent; 2969949964c9SMarcos Paulo de Souza } 29701a9fd417SDavid Sterba /* subvol_name_ptr is already nul terminated */ 2971949964c9SMarcos Paulo de Souza subvol_name = (char *)kbasename(subvol_name_ptr); 2972949964c9SMarcos Paulo de Souza } 2973949964c9SMarcos Paulo de Souza } else { 2974949964c9SMarcos Paulo de Souza vol_args = memdup_user(arg, sizeof(*vol_args)); 2975949964c9SMarcos Paulo de Souza if (IS_ERR(vol_args)) 2976949964c9SMarcos Paulo de Souza return PTR_ERR(vol_args); 2977949964c9SMarcos Paulo de Souza 2978949964c9SMarcos Paulo de Souza vol_args->name[BTRFS_PATH_NAME_MAX] = 0; 2979949964c9SMarcos Paulo de Souza subvol_name = vol_args->name; 2980949964c9SMarcos Paulo de Souza 2981949964c9SMarcos Paulo de Souza err = mnt_want_write_file(file); 2982949964c9SMarcos Paulo de Souza if (err) 2983949964c9SMarcos Paulo de Souza goto out; 2984949964c9SMarcos Paulo de Souza } 2985949964c9SMarcos Paulo de Souza 2986949964c9SMarcos Paulo de Souza subvol_namelen = strlen(subvol_name); 2987949964c9SMarcos Paulo de Souza 2988949964c9SMarcos Paulo de Souza if (strchr(subvol_name, '/') || 2989949964c9SMarcos Paulo de Souza strncmp(subvol_name, "..", subvol_namelen) == 0) { 2990949964c9SMarcos Paulo de Souza err = -EINVAL; 2991949964c9SMarcos Paulo de Souza goto free_subvol_name; 2992949964c9SMarcos Paulo de Souza } 2993949964c9SMarcos Paulo de Souza 2994949964c9SMarcos Paulo de Souza if (!S_ISDIR(dir->i_mode)) { 2995949964c9SMarcos Paulo de Souza err = -ENOTDIR; 2996949964c9SMarcos Paulo de Souza goto free_subvol_name; 2997949964c9SMarcos Paulo de Souza } 2998521e0546SDavid Sterba 299900235411SAl Viro err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT); 300000235411SAl Viro if (err == -EINTR) 3001949964c9SMarcos Paulo de Souza goto free_subvol_name; 3002c4ed533bSChristian Brauner dentry = lookup_one(mnt_userns, subvol_name, parent, subvol_namelen); 300376dda93cSYan, Zheng if (IS_ERR(dentry)) { 300476dda93cSYan, Zheng err = PTR_ERR(dentry); 300576dda93cSYan, Zheng goto out_unlock_dir; 300676dda93cSYan, Zheng } 300776dda93cSYan, Zheng 30082b0143b5SDavid Howells if (d_really_is_negative(dentry)) { 300976dda93cSYan, Zheng err = -ENOENT; 301076dda93cSYan, Zheng goto out_dput; 301176dda93cSYan, Zheng } 301276dda93cSYan, Zheng 30132b0143b5SDavid Howells inode = d_inode(dentry); 30144260f7c7SSage Weil dest = BTRFS_I(inode)->root; 30154260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)) { 30164260f7c7SSage Weil /* 30174260f7c7SSage Weil * Regular user. Only allow this with a special mount 30184260f7c7SSage Weil * option, when the user has write+exec access to the 30194260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 30204260f7c7SSage Weil * allowed. 30214260f7c7SSage Weil * 30224260f7c7SSage Weil * Note that this is _not_ check that the subvol is 30234260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 30244260f7c7SSage Weil * otherwise be able to delete. 30254260f7c7SSage Weil * 30264260f7c7SSage Weil * Users who want to delete empty subvols should try 30274260f7c7SSage Weil * rmdir(2). 30284260f7c7SSage Weil */ 30294260f7c7SSage Weil err = -EPERM; 30300b246afaSJeff Mahoney if (!btrfs_test_opt(fs_info, USER_SUBVOL_RM_ALLOWED)) 30314260f7c7SSage Weil goto out_dput; 30324260f7c7SSage Weil 30334260f7c7SSage Weil /* 30344260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 30354260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 30364260f7c7SSage Weil * must be called on the dentry referencing the root 30374260f7c7SSage Weil * of the subvol, not a random directory contained 30384260f7c7SSage Weil * within it. 30394260f7c7SSage Weil */ 30404260f7c7SSage Weil err = -EINVAL; 30414260f7c7SSage Weil if (root == dest) 30424260f7c7SSage Weil goto out_dput; 30434260f7c7SSage Weil 3044c4ed533bSChristian Brauner err = inode_permission(mnt_userns, inode, MAY_WRITE | MAY_EXEC); 30454260f7c7SSage Weil if (err) 30464260f7c7SSage Weil goto out_dput; 30475c39da5bSMiao Xie } 30484260f7c7SSage Weil 30495c39da5bSMiao Xie /* check if subvolume may be deleted by a user */ 3050c4ed533bSChristian Brauner err = btrfs_may_delete(mnt_userns, dir, dentry, 1); 30514260f7c7SSage Weil if (err) 30524260f7c7SSage Weil goto out_dput; 30534260f7c7SSage Weil 30544a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 305576dda93cSYan, Zheng err = -EINVAL; 305676dda93cSYan, Zheng goto out_dput; 305776dda93cSYan, Zheng } 305876dda93cSYan, Zheng 305964708539SJosef Bacik btrfs_inode_lock(inode, 0); 3060f60a2364SMisono Tomohiro err = btrfs_delete_subvolume(dir, dentry); 306164708539SJosef Bacik btrfs_inode_unlock(inode, 0); 306246008d9dSAmir Goldstein if (!err) { 306346008d9dSAmir Goldstein fsnotify_rmdir(dir, dentry); 306476dda93cSYan, Zheng d_delete(dentry); 306546008d9dSAmir Goldstein } 3066fa6ac876SLiu Bo 306776dda93cSYan, Zheng out_dput: 306876dda93cSYan, Zheng dput(dentry); 306976dda93cSYan, Zheng out_unlock_dir: 307064708539SJosef Bacik btrfs_inode_unlock(dir, 0); 3071949964c9SMarcos Paulo de Souza free_subvol_name: 3072949964c9SMarcos Paulo de Souza kfree(subvol_name_ptr); 3073949964c9SMarcos Paulo de Souza free_parent: 3074949964c9SMarcos Paulo de Souza if (destroy_parent) 3075949964c9SMarcos Paulo de Souza dput(parent); 307600235411SAl Viro out_drop_write: 30772a79f17eSAl Viro mnt_drop_write_file(file); 307876dda93cSYan, Zheng out: 3079949964c9SMarcos Paulo de Souza kfree(vol_args2); 308076dda93cSYan, Zheng kfree(vol_args); 308176dda93cSYan, Zheng return err; 308276dda93cSYan, Zheng } 308376dda93cSYan, Zheng 30841e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 3085f46b5a66SChristoph Hellwig { 3086496ad9aaSAl Viro struct inode *inode = file_inode(file); 3087f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3088c853a578SGoldwyn Rodrigues struct btrfs_ioctl_defrag_range_args range = {0}; 3089c146afadSYan Zheng int ret; 3090c146afadSYan Zheng 309125122d15SIlya Dryomov ret = mnt_want_write_file(file); 309225122d15SIlya Dryomov if (ret) 309325122d15SIlya Dryomov return ret; 3094b83cc969SLi Zefan 309525122d15SIlya Dryomov if (btrfs_root_readonly(root)) { 309625122d15SIlya Dryomov ret = -EROFS; 309725122d15SIlya Dryomov goto out; 30985ac00addSStefan Behrens } 3099f46b5a66SChristoph Hellwig 3100f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3101f46b5a66SChristoph Hellwig case S_IFDIR: 3102e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 3103e441d54dSChris Mason ret = -EPERM; 3104e441d54dSChris Mason goto out; 3105e441d54dSChris Mason } 3106de78b51aSEric Sandeen ret = btrfs_defrag_root(root); 3107f46b5a66SChristoph Hellwig break; 3108f46b5a66SChristoph Hellwig case S_IFREG: 3109616d374eSAdam Borowski /* 3110616d374eSAdam Borowski * Note that this does not check the file descriptor for write 3111616d374eSAdam Borowski * access. This prevents defragmenting executables that are 3112616d374eSAdam Borowski * running and allows defrag on files open in read-only mode. 3113616d374eSAdam Borowski */ 3114616d374eSAdam Borowski if (!capable(CAP_SYS_ADMIN) && 311547291baaSChristian Brauner inode_permission(&init_user_ns, inode, MAY_WRITE)) { 3116616d374eSAdam Borowski ret = -EPERM; 3117e441d54dSChris Mason goto out; 3118e441d54dSChris Mason } 31191e701a32SChris Mason 31201e701a32SChris Mason if (argp) { 3121c853a578SGoldwyn Rodrigues if (copy_from_user(&range, argp, sizeof(range))) { 31221e701a32SChris Mason ret = -EFAULT; 3123683be16eSDan Carpenter goto out; 31241e701a32SChris Mason } 31251e701a32SChris Mason /* compression requires us to start the IO */ 3126c853a578SGoldwyn Rodrigues if ((range.flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 3127c853a578SGoldwyn Rodrigues range.flags |= BTRFS_DEFRAG_RANGE_START_IO; 3128c853a578SGoldwyn Rodrigues range.extent_thresh = (u32)-1; 31291e701a32SChris Mason } 31301e701a32SChris Mason } else { 31311e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 3132c853a578SGoldwyn Rodrigues range.len = (u64)-1; 31331e701a32SChris Mason } 31341ccc2e8aSQu Wenruo ret = btrfs_defrag_file(file_inode(file), &file->f_ra, 3135c853a578SGoldwyn Rodrigues &range, BTRFS_OLDEST_GENERATION, 0); 31364cb5300bSChris Mason if (ret > 0) 31374cb5300bSChris Mason ret = 0; 3138f46b5a66SChristoph Hellwig break; 31398929ecfaSYan, Zheng default: 31408929ecfaSYan, Zheng ret = -EINVAL; 3141f46b5a66SChristoph Hellwig } 3142e441d54dSChris Mason out: 314325122d15SIlya Dryomov mnt_drop_write_file(file); 3144e441d54dSChris Mason return ret; 3145f46b5a66SChristoph Hellwig } 3146f46b5a66SChristoph Hellwig 31472ff7e61eSJeff Mahoney static long btrfs_ioctl_add_dev(struct btrfs_fs_info *fs_info, void __user *arg) 3148f46b5a66SChristoph Hellwig { 3149f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 3150f46b5a66SChristoph Hellwig int ret; 3151f46b5a66SChristoph Hellwig 3152e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3153e441d54dSChris Mason return -EPERM; 3154e441d54dSChris Mason 3155c3e1f96cSGoldwyn Rodrigues if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_ADD)) 3156e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3157c9e9f97bSIlya Dryomov 3158dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3159c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 3160c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 3161c9e9f97bSIlya Dryomov goto out; 3162c9e9f97bSIlya Dryomov } 3163f46b5a66SChristoph Hellwig 31645516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 31652ff7e61eSJeff Mahoney ret = btrfs_init_new_device(fs_info, vol_args->name); 3166f46b5a66SChristoph Hellwig 316743d20761SAnand Jain if (!ret) 31680b246afaSJeff Mahoney btrfs_info(fs_info, "disk added %s", vol_args->name); 316943d20761SAnand Jain 3170f46b5a66SChristoph Hellwig kfree(vol_args); 3171c9e9f97bSIlya Dryomov out: 3172c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3173f46b5a66SChristoph Hellwig return ret; 3174f46b5a66SChristoph Hellwig } 3175f46b5a66SChristoph Hellwig 31766b526ed7SAnand Jain static long btrfs_ioctl_rm_dev_v2(struct file *file, void __user *arg) 3177f46b5a66SChristoph Hellwig { 31781a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 31790b246afaSJeff Mahoney struct inode *inode = file_inode(file); 31800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31816b526ed7SAnand Jain struct btrfs_ioctl_vol_args_v2 *vol_args; 31823fa421deSJosef Bacik struct block_device *bdev = NULL; 31833fa421deSJosef Bacik fmode_t mode; 3184f46b5a66SChristoph Hellwig int ret; 318567ae34b6SDavid Sterba bool cancel = false; 3186f46b5a66SChristoph Hellwig 3187e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 3188e441d54dSChris Mason return -EPERM; 3189e441d54dSChris Mason 3190dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 3191c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 3192c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 31931a15eb72SJosef Bacik goto out; 3194c9e9f97bSIlya Dryomov } 3195f46b5a66SChristoph Hellwig 3196748449cdSDavid Sterba if (vol_args->flags & ~BTRFS_DEVICE_REMOVE_ARGS_MASK) { 3197fd4e994bSOmar Sandoval ret = -EOPNOTSUPP; 3198fd4e994bSOmar Sandoval goto out; 3199fd4e994bSOmar Sandoval } 32001a15eb72SJosef Bacik 32016b526ed7SAnand Jain vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 32021a15eb72SJosef Bacik if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) { 32031a15eb72SJosef Bacik args.devid = vol_args->devid; 32041a15eb72SJosef Bacik } else if (!strcmp("cancel", vol_args->name)) { 320567ae34b6SDavid Sterba cancel = true; 32061a15eb72SJosef Bacik } else { 32071a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 32081a15eb72SJosef Bacik if (ret) 32091a15eb72SJosef Bacik goto out; 32101a15eb72SJosef Bacik } 32111a15eb72SJosef Bacik 32121a15eb72SJosef Bacik ret = mnt_want_write_file(file); 32131a15eb72SJosef Bacik if (ret) 32141a15eb72SJosef Bacik goto out; 321567ae34b6SDavid Sterba 321667ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 321767ae34b6SDavid Sterba cancel); 321867ae34b6SDavid Sterba if (ret) 32191a15eb72SJosef Bacik goto err_drop; 322067ae34b6SDavid Sterba 32211a15eb72SJosef Bacik /* Exclusive operation is now claimed */ 32221a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 322367ae34b6SDavid Sterba 3224c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 3225183860f6SAnand Jain 32266b526ed7SAnand Jain if (!ret) { 3227735654eaSDavid Sterba if (vol_args->flags & BTRFS_DEVICE_SPEC_BY_ID) 32280b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: id %llu", 32296b526ed7SAnand Jain vol_args->devid); 32306b526ed7SAnand Jain else 32310b246afaSJeff Mahoney btrfs_info(fs_info, "device deleted: %s", 32326b526ed7SAnand Jain vol_args->name); 32336b526ed7SAnand Jain } 3234c47ca32dSDan Carpenter err_drop: 32354ac20c70SIlya Dryomov mnt_drop_write_file(file); 32363fa421deSJosef Bacik if (bdev) 32373fa421deSJosef Bacik blkdev_put(bdev, mode); 32381a15eb72SJosef Bacik out: 32391a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 32401a15eb72SJosef Bacik kfree(vol_args); 3241f46b5a66SChristoph Hellwig return ret; 3242f46b5a66SChristoph Hellwig } 3243f46b5a66SChristoph Hellwig 3244f46b5a66SChristoph Hellwig static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 3245f46b5a66SChristoph Hellwig { 32461a15eb72SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 32470b246afaSJeff Mahoney struct inode *inode = file_inode(file); 32480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 3249f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 32503fa421deSJosef Bacik struct block_device *bdev = NULL; 32513fa421deSJosef Bacik fmode_t mode; 3252f46b5a66SChristoph Hellwig int ret; 325367ae34b6SDavid Sterba bool cancel; 3254f46b5a66SChristoph Hellwig 3255f46b5a66SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 3256f46b5a66SChristoph Hellwig return -EPERM; 3257f46b5a66SChristoph Hellwig 32581a15eb72SJosef Bacik vol_args = memdup_user(arg, sizeof(*vol_args)); 32591a15eb72SJosef Bacik if (IS_ERR(vol_args)) 32601a15eb72SJosef Bacik return PTR_ERR(vol_args); 32611a15eb72SJosef Bacik 32621a15eb72SJosef Bacik vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 32631a15eb72SJosef Bacik if (!strcmp("cancel", vol_args->name)) { 32641a15eb72SJosef Bacik cancel = true; 32651a15eb72SJosef Bacik } else { 32661a15eb72SJosef Bacik ret = btrfs_get_dev_args_from_path(fs_info, &args, vol_args->name); 32671a15eb72SJosef Bacik if (ret) 32681a15eb72SJosef Bacik goto out; 32691a15eb72SJosef Bacik } 32701a15eb72SJosef Bacik 3271f46b5a66SChristoph Hellwig ret = mnt_want_write_file(file); 3272f46b5a66SChristoph Hellwig if (ret) 32731a15eb72SJosef Bacik goto out; 3274f46b5a66SChristoph Hellwig 327567ae34b6SDavid Sterba ret = exclop_start_or_cancel_reloc(fs_info, BTRFS_EXCLOP_DEV_REMOVE, 327667ae34b6SDavid Sterba cancel); 327767ae34b6SDavid Sterba if (ret == 0) { 32781a15eb72SJosef Bacik ret = btrfs_rm_device(fs_info, &args, &bdev, &mode); 3279f46b5a66SChristoph Hellwig if (!ret) 32800b246afaSJeff Mahoney btrfs_info(fs_info, "disk deleted %s", vol_args->name); 3281c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 328267ae34b6SDavid Sterba } 328367ae34b6SDavid Sterba 3284f46b5a66SChristoph Hellwig mnt_drop_write_file(file); 32853fa421deSJosef Bacik if (bdev) 32863fa421deSJosef Bacik blkdev_put(bdev, mode); 32871a15eb72SJosef Bacik out: 32881a15eb72SJosef Bacik btrfs_put_dev_args_from_path(&args); 32891a15eb72SJosef Bacik kfree(vol_args); 3290f46b5a66SChristoph Hellwig return ret; 3291f46b5a66SChristoph Hellwig } 3292f46b5a66SChristoph Hellwig 32932ff7e61eSJeff Mahoney static long btrfs_ioctl_fs_info(struct btrfs_fs_info *fs_info, 32942ff7e61eSJeff Mahoney void __user *arg) 3295475f6387SJan Schmidt { 3296027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 3297475f6387SJan Schmidt struct btrfs_device *device; 32980b246afaSJeff Mahoney struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3299137c5418SJohannes Thumshirn u64 flags_in; 3300027ed2f0SLi Zefan int ret = 0; 3301475f6387SJan Schmidt 3302137c5418SJohannes Thumshirn fi_args = memdup_user(arg, sizeof(*fi_args)); 3303137c5418SJohannes Thumshirn if (IS_ERR(fi_args)) 3304137c5418SJohannes Thumshirn return PTR_ERR(fi_args); 3305137c5418SJohannes Thumshirn 3306137c5418SJohannes Thumshirn flags_in = fi_args->flags; 3307137c5418SJohannes Thumshirn memset(fi_args, 0, sizeof(*fi_args)); 3308027ed2f0SLi Zefan 3309d03262c7SDavid Sterba rcu_read_lock(); 3310027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 3311475f6387SJan Schmidt 3312d03262c7SDavid Sterba list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { 3313027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 3314027ed2f0SLi Zefan fi_args->max_id = device->devid; 3315475f6387SJan Schmidt } 3316d03262c7SDavid Sterba rcu_read_unlock(); 3317475f6387SJan Schmidt 3318de37aa51SNikolay Borisov memcpy(&fi_args->fsid, fs_devices->fsid, sizeof(fi_args->fsid)); 3319bea7eafdSOmar Sandoval fi_args->nodesize = fs_info->nodesize; 3320bea7eafdSOmar Sandoval fi_args->sectorsize = fs_info->sectorsize; 3321bea7eafdSOmar Sandoval fi_args->clone_alignment = fs_info->sectorsize; 332280a773fbSDavid Sterba 3323137c5418SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_CSUM_INFO) { 3324137c5418SJohannes Thumshirn fi_args->csum_type = btrfs_super_csum_type(fs_info->super_copy); 3325137c5418SJohannes Thumshirn fi_args->csum_size = btrfs_super_csum_size(fs_info->super_copy); 3326137c5418SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_CSUM_INFO; 3327137c5418SJohannes Thumshirn } 3328137c5418SJohannes Thumshirn 33290fb408a5SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_GENERATION) { 33300fb408a5SJohannes Thumshirn fi_args->generation = fs_info->generation; 33310fb408a5SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_GENERATION; 33320fb408a5SJohannes Thumshirn } 33330fb408a5SJohannes Thumshirn 333449bac897SJohannes Thumshirn if (flags_in & BTRFS_FS_INFO_FLAG_METADATA_UUID) { 333549bac897SJohannes Thumshirn memcpy(&fi_args->metadata_uuid, fs_devices->metadata_uuid, 333649bac897SJohannes Thumshirn sizeof(fi_args->metadata_uuid)); 333749bac897SJohannes Thumshirn fi_args->flags |= BTRFS_FS_INFO_FLAG_METADATA_UUID; 333849bac897SJohannes Thumshirn } 333949bac897SJohannes Thumshirn 3340027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 3341027ed2f0SLi Zefan ret = -EFAULT; 3342475f6387SJan Schmidt 3343027ed2f0SLi Zefan kfree(fi_args); 3344027ed2f0SLi Zefan return ret; 3345475f6387SJan Schmidt } 3346475f6387SJan Schmidt 33472ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_info(struct btrfs_fs_info *fs_info, 33482ff7e61eSJeff Mahoney void __user *arg) 3349475f6387SJan Schmidt { 3350562d7b15SJosef Bacik BTRFS_DEV_LOOKUP_ARGS(args); 3351475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 3352475f6387SJan Schmidt struct btrfs_device *dev; 3353475f6387SJan Schmidt int ret = 0; 3354475f6387SJan Schmidt 3355475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 3356475f6387SJan Schmidt if (IS_ERR(di_args)) 3357475f6387SJan Schmidt return PTR_ERR(di_args); 3358475f6387SJan Schmidt 3359562d7b15SJosef Bacik args.devid = di_args->devid; 3360dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(di_args->uuid)) 3361562d7b15SJosef Bacik args.uuid = di_args->uuid; 3362475f6387SJan Schmidt 3363c5593ca3SDavid Sterba rcu_read_lock(); 3364562d7b15SJosef Bacik dev = btrfs_find_device(fs_info->fs_devices, &args); 3365475f6387SJan Schmidt if (!dev) { 3366475f6387SJan Schmidt ret = -ENODEV; 3367475f6387SJan Schmidt goto out; 3368475f6387SJan Schmidt } 3369475f6387SJan Schmidt 3370475f6387SJan Schmidt di_args->devid = dev->devid; 33717cc8e58dSMiao Xie di_args->bytes_used = btrfs_device_get_bytes_used(dev); 33727cc8e58dSMiao Xie di_args->total_bytes = btrfs_device_get_total_bytes(dev); 3373475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 3374a27202fbSJim Meyering if (dev->name) { 3375672d5990SMisono Tomohiro strncpy(di_args->path, rcu_str_deref(dev->name), 3376672d5990SMisono Tomohiro sizeof(di_args->path) - 1); 3377a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 3378a27202fbSJim Meyering } else { 337999ba55adSStefan Behrens di_args->path[0] = '\0'; 3380a27202fbSJim Meyering } 3381475f6387SJan Schmidt 3382475f6387SJan Schmidt out: 3383c5593ca3SDavid Sterba rcu_read_unlock(); 3384475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 3385475f6387SJan Schmidt ret = -EFAULT; 3386475f6387SJan Schmidt 3387475f6387SJan Schmidt kfree(di_args); 3388475f6387SJan Schmidt return ret; 3389475f6387SJan Schmidt } 3390475f6387SJan Schmidt 33916ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 33926ef5ed0dSJosef Bacik { 3393496ad9aaSAl Viro struct inode *inode = file_inode(file); 33940b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 33956ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 33966ef5ed0dSJosef Bacik struct btrfs_root *new_root; 33976ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 33986ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 33992a2b5d62SJosef Bacik struct btrfs_path *path = NULL; 34006ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 34016ef5ed0dSJosef Bacik u64 objectid = 0; 34026ef5ed0dSJosef Bacik u64 dir_id; 34033c04ce01SMiao Xie int ret; 34046ef5ed0dSJosef Bacik 34056ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 34066ef5ed0dSJosef Bacik return -EPERM; 34076ef5ed0dSJosef Bacik 34083c04ce01SMiao Xie ret = mnt_want_write_file(file); 34093c04ce01SMiao Xie if (ret) 34103c04ce01SMiao Xie return ret; 34113c04ce01SMiao Xie 34123c04ce01SMiao Xie if (copy_from_user(&objectid, argp, sizeof(objectid))) { 34133c04ce01SMiao Xie ret = -EFAULT; 34143c04ce01SMiao Xie goto out; 34153c04ce01SMiao Xie } 34166ef5ed0dSJosef Bacik 34176ef5ed0dSJosef Bacik if (!objectid) 34181cecf579Schandan objectid = BTRFS_FS_TREE_OBJECTID; 34196ef5ed0dSJosef Bacik 342056e9357aSDavid Sterba new_root = btrfs_get_fs_root(fs_info, objectid, true); 34213c04ce01SMiao Xie if (IS_ERR(new_root)) { 34223c04ce01SMiao Xie ret = PTR_ERR(new_root); 34233c04ce01SMiao Xie goto out; 34243c04ce01SMiao Xie } 34252a2b5d62SJosef Bacik if (!is_fstree(new_root->root_key.objectid)) { 34262a2b5d62SJosef Bacik ret = -ENOENT; 34272a2b5d62SJosef Bacik goto out_free; 34282a2b5d62SJosef Bacik } 34296ef5ed0dSJosef Bacik 34306ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 34313c04ce01SMiao Xie if (!path) { 34323c04ce01SMiao Xie ret = -ENOMEM; 34332a2b5d62SJosef Bacik goto out_free; 34343c04ce01SMiao Xie } 34356ef5ed0dSJosef Bacik 34366ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 343798d5dc13STsutomu Itoh if (IS_ERR(trans)) { 34383c04ce01SMiao Xie ret = PTR_ERR(trans); 34392a2b5d62SJosef Bacik goto out_free; 34406ef5ed0dSJosef Bacik } 34416ef5ed0dSJosef Bacik 34420b246afaSJeff Mahoney dir_id = btrfs_super_root_dir(fs_info->super_copy); 34430b246afaSJeff Mahoney di = btrfs_lookup_dir_item(trans, fs_info->tree_root, path, 34446ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 3445cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 34462a2b5d62SJosef Bacik btrfs_release_path(path); 34473a45bb20SJeff Mahoney btrfs_end_transaction(trans); 34480b246afaSJeff Mahoney btrfs_err(fs_info, 34495d163e0eSJeff Mahoney "Umm, you don't have the default diritem, this isn't going to work"); 34503c04ce01SMiao Xie ret = -ENOENT; 34512a2b5d62SJosef Bacik goto out_free; 34526ef5ed0dSJosef Bacik } 34536ef5ed0dSJosef Bacik 34546ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 34556ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 34566ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 34572a2b5d62SJosef Bacik btrfs_release_path(path); 34586ef5ed0dSJosef Bacik 34590b246afaSJeff Mahoney btrfs_set_fs_incompat(fs_info, DEFAULT_SUBVOL); 34603a45bb20SJeff Mahoney btrfs_end_transaction(trans); 34612a2b5d62SJosef Bacik out_free: 346200246528SJosef Bacik btrfs_put_root(new_root); 34632a2b5d62SJosef Bacik btrfs_free_path(path); 34643c04ce01SMiao Xie out: 34653c04ce01SMiao Xie mnt_drop_write_file(file); 34663c04ce01SMiao Xie return ret; 34676ef5ed0dSJosef Bacik } 34686ef5ed0dSJosef Bacik 3469c065f5b1SSu Yue static void get_block_group_info(struct list_head *groups_list, 3470bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 3471bf5fc093SJosef Bacik { 347232da5386SDavid Sterba struct btrfs_block_group *block_group; 3473bf5fc093SJosef Bacik 3474bf5fc093SJosef Bacik space->total_bytes = 0; 3475bf5fc093SJosef Bacik space->used_bytes = 0; 3476bf5fc093SJosef Bacik space->flags = 0; 3477bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 3478bf5fc093SJosef Bacik space->flags = block_group->flags; 3479b3470b5dSDavid Sterba space->total_bytes += block_group->length; 3480bf38be65SDavid Sterba space->used_bytes += block_group->used; 3481bf5fc093SJosef Bacik } 3482bf5fc093SJosef Bacik } 3483bf5fc093SJosef Bacik 34842ff7e61eSJeff Mahoney static long btrfs_ioctl_space_info(struct btrfs_fs_info *fs_info, 34852ff7e61eSJeff Mahoney void __user *arg) 34861406e432SJosef Bacik { 34871406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 34881406e432SJosef Bacik struct btrfs_ioctl_space_info space; 34891406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 34907fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 349113f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 34921406e432SJosef Bacik struct btrfs_space_info *info; 3493315d8e98SColin Ian King static const u64 types[] = { 3494315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA, 3495bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 3496bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 3497315d8e98SColin Ian King BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA 3498315d8e98SColin Ian King }; 3499bf5fc093SJosef Bacik int num_types = 4; 35007fde62bfSChris Mason int alloc_size; 35011406e432SJosef Bacik int ret = 0; 350251788b1bSDan Rosenberg u64 slot_count = 0; 3503bf5fc093SJosef Bacik int i, c; 35041406e432SJosef Bacik 35051406e432SJosef Bacik if (copy_from_user(&space_args, 35061406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 35071406e432SJosef Bacik sizeof(space_args))) 35081406e432SJosef Bacik return -EFAULT; 35091406e432SJosef Bacik 3510bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3511bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3512bf5fc093SJosef Bacik 3513bf5fc093SJosef Bacik info = NULL; 351472804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3515bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3516bf5fc093SJosef Bacik info = tmp; 3517bf5fc093SJosef Bacik break; 3518bf5fc093SJosef Bacik } 3519bf5fc093SJosef Bacik } 35201406e432SJosef Bacik 3521bf5fc093SJosef Bacik if (!info) 3522bf5fc093SJosef Bacik continue; 3523bf5fc093SJosef Bacik 3524bf5fc093SJosef Bacik down_read(&info->groups_sem); 3525bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3526bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 3527bf5fc093SJosef Bacik slot_count++; 3528bf5fc093SJosef Bacik } 3529bf5fc093SJosef Bacik up_read(&info->groups_sem); 3530bf5fc093SJosef Bacik } 3531bf5fc093SJosef Bacik 353236523e95SDavid Sterba /* 353336523e95SDavid Sterba * Global block reserve, exported as a space_info 353436523e95SDavid Sterba */ 353536523e95SDavid Sterba slot_count++; 353636523e95SDavid Sterba 35377fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 35387fde62bfSChris Mason if (space_args.space_slots == 0) { 35397fde62bfSChris Mason space_args.total_spaces = slot_count; 35407fde62bfSChris Mason goto out; 35417fde62bfSChris Mason } 3542bf5fc093SJosef Bacik 354351788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 3544bf5fc093SJosef Bacik 35457fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 3546bf5fc093SJosef Bacik 35477fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 35487fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 35497fde62bfSChris Mason */ 355009cbfeafSKirill A. Shutemov if (alloc_size > PAGE_SIZE) 35517fde62bfSChris Mason return -ENOMEM; 35527fde62bfSChris Mason 35537fde62bfSChris Mason space_args.total_spaces = 0; 35548d2db785SDavid Sterba dest = kmalloc(alloc_size, GFP_KERNEL); 35557fde62bfSChris Mason if (!dest) 35567fde62bfSChris Mason return -ENOMEM; 35577fde62bfSChris Mason dest_orig = dest; 35587fde62bfSChris Mason 35597fde62bfSChris Mason /* now we have a buffer to copy into */ 3560bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3561bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3562bf5fc093SJosef Bacik 356351788b1bSDan Rosenberg if (!slot_count) 356451788b1bSDan Rosenberg break; 356551788b1bSDan Rosenberg 3566bf5fc093SJosef Bacik info = NULL; 356772804905SJosef Bacik list_for_each_entry(tmp, &fs_info->space_info, list) { 3568bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3569bf5fc093SJosef Bacik info = tmp; 35707fde62bfSChris Mason break; 3571bf5fc093SJosef Bacik } 3572bf5fc093SJosef Bacik } 35737fde62bfSChris Mason 3574bf5fc093SJosef Bacik if (!info) 3575bf5fc093SJosef Bacik continue; 3576bf5fc093SJosef Bacik down_read(&info->groups_sem); 3577bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3578bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 3579c065f5b1SSu Yue get_block_group_info(&info->block_groups[c], 3580c065f5b1SSu Yue &space); 35817fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 35821406e432SJosef Bacik dest++; 35831406e432SJosef Bacik space_args.total_spaces++; 358451788b1bSDan Rosenberg slot_count--; 35851406e432SJosef Bacik } 358651788b1bSDan Rosenberg if (!slot_count) 358751788b1bSDan Rosenberg break; 3588bf5fc093SJosef Bacik } 3589bf5fc093SJosef Bacik up_read(&info->groups_sem); 3590bf5fc093SJosef Bacik } 35911406e432SJosef Bacik 359236523e95SDavid Sterba /* 359336523e95SDavid Sterba * Add global block reserve 359436523e95SDavid Sterba */ 359536523e95SDavid Sterba if (slot_count) { 35960b246afaSJeff Mahoney struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; 359736523e95SDavid Sterba 359836523e95SDavid Sterba spin_lock(&block_rsv->lock); 359936523e95SDavid Sterba space.total_bytes = block_rsv->size; 360036523e95SDavid Sterba space.used_bytes = block_rsv->size - block_rsv->reserved; 360136523e95SDavid Sterba spin_unlock(&block_rsv->lock); 360236523e95SDavid Sterba space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV; 360336523e95SDavid Sterba memcpy(dest, &space, sizeof(space)); 360436523e95SDavid Sterba space_args.total_spaces++; 360536523e95SDavid Sterba } 360636523e95SDavid Sterba 36072eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 36087fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 36097fde62bfSChris Mason 36107fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 36117fde62bfSChris Mason ret = -EFAULT; 36127fde62bfSChris Mason 36137fde62bfSChris Mason kfree(dest_orig); 36147fde62bfSChris Mason out: 36157fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 36161406e432SJosef Bacik ret = -EFAULT; 36171406e432SJosef Bacik 36181406e432SJosef Bacik return ret; 36191406e432SJosef Bacik } 36201406e432SJosef Bacik 36219a8c28beSMiao Xie static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 36229a8c28beSMiao Xie void __user *argp) 362346204592SSage Weil { 362446204592SSage Weil struct btrfs_trans_handle *trans; 362546204592SSage Weil u64 transid; 3626db5b493aSTsutomu Itoh int ret; 362746204592SSage Weil 3628d4edf39bSMiao Xie trans = btrfs_attach_transaction_barrier(root); 3629ff7c1d33SMiao Xie if (IS_ERR(trans)) { 3630ff7c1d33SMiao Xie if (PTR_ERR(trans) != -ENOENT) 363198d5dc13STsutomu Itoh return PTR_ERR(trans); 3632ff7c1d33SMiao Xie 3633ff7c1d33SMiao Xie /* No running transaction, don't bother */ 3634ff7c1d33SMiao Xie transid = root->fs_info->last_trans_committed; 3635ff7c1d33SMiao Xie goto out; 3636ff7c1d33SMiao Xie } 363746204592SSage Weil transid = trans->transid; 363832cc4f87SDavid Sterba ret = btrfs_commit_transaction_async(trans); 36398b2b2d3cSTsutomu Itoh if (ret) { 36403a45bb20SJeff Mahoney btrfs_end_transaction(trans); 3641db5b493aSTsutomu Itoh return ret; 36428b2b2d3cSTsutomu Itoh } 3643ff7c1d33SMiao Xie out: 364446204592SSage Weil if (argp) 364546204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 364646204592SSage Weil return -EFAULT; 364746204592SSage Weil return 0; 364846204592SSage Weil } 364946204592SSage Weil 36502ff7e61eSJeff Mahoney static noinline long btrfs_ioctl_wait_sync(struct btrfs_fs_info *fs_info, 36519a8c28beSMiao Xie void __user *argp) 365246204592SSage Weil { 365346204592SSage Weil u64 transid; 365446204592SSage Weil 365546204592SSage Weil if (argp) { 365646204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 365746204592SSage Weil return -EFAULT; 365846204592SSage Weil } else { 365946204592SSage Weil transid = 0; /* current trans */ 366046204592SSage Weil } 36612ff7e61eSJeff Mahoney return btrfs_wait_for_commit(fs_info, transid); 366246204592SSage Weil } 366346204592SSage Weil 3664b8e95489SMiao Xie static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 3665475f6387SJan Schmidt { 36660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 3667475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3668b8e95489SMiao Xie int ret; 3669475f6387SJan Schmidt 3670475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3671475f6387SJan Schmidt return -EPERM; 3672475f6387SJan Schmidt 3673475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3674475f6387SJan Schmidt if (IS_ERR(sa)) 3675475f6387SJan Schmidt return PTR_ERR(sa); 3676475f6387SJan Schmidt 3677b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 3678b8e95489SMiao Xie ret = mnt_want_write_file(file); 3679b8e95489SMiao Xie if (ret) 3680b8e95489SMiao Xie goto out; 3681b8e95489SMiao Xie } 3682b8e95489SMiao Xie 36830b246afaSJeff Mahoney ret = btrfs_scrub_dev(fs_info, sa->devid, sa->start, sa->end, 368463a212abSStefan Behrens &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 368563a212abSStefan Behrens 0); 3686475f6387SJan Schmidt 36875afe6ce7SFilipe Manana /* 36885afe6ce7SFilipe Manana * Copy scrub args to user space even if btrfs_scrub_dev() returned an 36895afe6ce7SFilipe Manana * error. This is important as it allows user space to know how much 36905afe6ce7SFilipe Manana * progress scrub has done. For example, if scrub is canceled we get 36915afe6ce7SFilipe Manana * -ECANCELED from btrfs_scrub_dev() and return that error back to user 36925afe6ce7SFilipe Manana * space. Later user space can inspect the progress from the structure 36935afe6ce7SFilipe Manana * btrfs_ioctl_scrub_args and resume scrub from where it left off 36945afe6ce7SFilipe Manana * previously (btrfs-progs does this). 36955afe6ce7SFilipe Manana * If we fail to copy the btrfs_ioctl_scrub_args structure to user space 36965afe6ce7SFilipe Manana * then return -EFAULT to signal the structure was not copied or it may 36975afe6ce7SFilipe Manana * be corrupt and unreliable due to a partial copy. 36985afe6ce7SFilipe Manana */ 36995afe6ce7SFilipe Manana if (copy_to_user(arg, sa, sizeof(*sa))) 3700475f6387SJan Schmidt ret = -EFAULT; 3701475f6387SJan Schmidt 3702b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) 3703b8e95489SMiao Xie mnt_drop_write_file(file); 3704b8e95489SMiao Xie out: 3705475f6387SJan Schmidt kfree(sa); 3706475f6387SJan Schmidt return ret; 3707475f6387SJan Schmidt } 3708475f6387SJan Schmidt 37092ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_cancel(struct btrfs_fs_info *fs_info) 3710475f6387SJan Schmidt { 3711475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3712475f6387SJan Schmidt return -EPERM; 3713475f6387SJan Schmidt 37142ff7e61eSJeff Mahoney return btrfs_scrub_cancel(fs_info); 3715475f6387SJan Schmidt } 3716475f6387SJan Schmidt 37172ff7e61eSJeff Mahoney static long btrfs_ioctl_scrub_progress(struct btrfs_fs_info *fs_info, 3718475f6387SJan Schmidt void __user *arg) 3719475f6387SJan Schmidt { 3720475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3721475f6387SJan Schmidt int ret; 3722475f6387SJan Schmidt 3723475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3724475f6387SJan Schmidt return -EPERM; 3725475f6387SJan Schmidt 3726475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3727475f6387SJan Schmidt if (IS_ERR(sa)) 3728475f6387SJan Schmidt return PTR_ERR(sa); 3729475f6387SJan Schmidt 37302ff7e61eSJeff Mahoney ret = btrfs_scrub_progress(fs_info, sa->devid, &sa->progress); 3731475f6387SJan Schmidt 37324fa99b00SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 3733475f6387SJan Schmidt ret = -EFAULT; 3734475f6387SJan Schmidt 3735475f6387SJan Schmidt kfree(sa); 3736475f6387SJan Schmidt return ret; 3737475f6387SJan Schmidt } 3738475f6387SJan Schmidt 37392ff7e61eSJeff Mahoney static long btrfs_ioctl_get_dev_stats(struct btrfs_fs_info *fs_info, 3740b27f7c0cSDavid Sterba void __user *arg) 3741c11d2c23SStefan Behrens { 3742c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 3743c11d2c23SStefan Behrens int ret; 3744c11d2c23SStefan Behrens 3745c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 3746c11d2c23SStefan Behrens if (IS_ERR(sa)) 3747c11d2c23SStefan Behrens return PTR_ERR(sa); 3748c11d2c23SStefan Behrens 3749b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 3750b27f7c0cSDavid Sterba kfree(sa); 3751b27f7c0cSDavid Sterba return -EPERM; 3752b27f7c0cSDavid Sterba } 3753b27f7c0cSDavid Sterba 37542ff7e61eSJeff Mahoney ret = btrfs_get_dev_stats(fs_info, sa); 3755c11d2c23SStefan Behrens 3756eee99577SFilipe Manana if (ret == 0 && copy_to_user(arg, sa, sizeof(*sa))) 3757c11d2c23SStefan Behrens ret = -EFAULT; 3758c11d2c23SStefan Behrens 3759c11d2c23SStefan Behrens kfree(sa); 3760c11d2c23SStefan Behrens return ret; 3761c11d2c23SStefan Behrens } 3762c11d2c23SStefan Behrens 37632ff7e61eSJeff Mahoney static long btrfs_ioctl_dev_replace(struct btrfs_fs_info *fs_info, 37642ff7e61eSJeff Mahoney void __user *arg) 37653f6bcfbdSStefan Behrens { 37663f6bcfbdSStefan Behrens struct btrfs_ioctl_dev_replace_args *p; 37673f6bcfbdSStefan Behrens int ret; 37683f6bcfbdSStefan Behrens 37693f6bcfbdSStefan Behrens if (!capable(CAP_SYS_ADMIN)) 37703f6bcfbdSStefan Behrens return -EPERM; 37713f6bcfbdSStefan Behrens 37723f6bcfbdSStefan Behrens p = memdup_user(arg, sizeof(*p)); 37733f6bcfbdSStefan Behrens if (IS_ERR(p)) 37743f6bcfbdSStefan Behrens return PTR_ERR(p); 37753f6bcfbdSStefan Behrens 37763f6bcfbdSStefan Behrens switch (p->cmd) { 37773f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 3778bc98a42cSDavid Howells if (sb_rdonly(fs_info->sb)) { 3779adfa97cbSIlya Dryomov ret = -EROFS; 3780adfa97cbSIlya Dryomov goto out; 3781adfa97cbSIlya Dryomov } 3782c3e1f96cSGoldwyn Rodrigues if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) { 3783e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 37843f6bcfbdSStefan Behrens } else { 37852ff7e61eSJeff Mahoney ret = btrfs_dev_replace_by_ioctl(fs_info, p); 3786c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 37873f6bcfbdSStefan Behrens } 37883f6bcfbdSStefan Behrens break; 37893f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 37900b246afaSJeff Mahoney btrfs_dev_replace_status(fs_info, p); 37913f6bcfbdSStefan Behrens ret = 0; 37923f6bcfbdSStefan Behrens break; 37933f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 379417d202b9SAnand Jain p->result = btrfs_dev_replace_cancel(fs_info); 379597282031SAnand Jain ret = 0; 37963f6bcfbdSStefan Behrens break; 37973f6bcfbdSStefan Behrens default: 37983f6bcfbdSStefan Behrens ret = -EINVAL; 37993f6bcfbdSStefan Behrens break; 38003f6bcfbdSStefan Behrens } 38013f6bcfbdSStefan Behrens 3802d3a53286SFilipe Manana if ((ret == 0 || ret == -ECANCELED) && copy_to_user(arg, p, sizeof(*p))) 38033f6bcfbdSStefan Behrens ret = -EFAULT; 3804adfa97cbSIlya Dryomov out: 38053f6bcfbdSStefan Behrens kfree(p); 38063f6bcfbdSStefan Behrens return ret; 38073f6bcfbdSStefan Behrens } 38083f6bcfbdSStefan Behrens 3809d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 3810d7728c96SJan Schmidt { 3811d7728c96SJan Schmidt int ret = 0; 3812d7728c96SJan Schmidt int i; 3813740c3d22SChris Mason u64 rel_ptr; 3814d7728c96SJan Schmidt int size; 3815806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 3816d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 3817d7728c96SJan Schmidt struct btrfs_path *path; 3818d7728c96SJan Schmidt 381982b22ac8SKusanagi Kouichi if (!capable(CAP_DAC_READ_SEARCH)) 3820d7728c96SJan Schmidt return -EPERM; 3821d7728c96SJan Schmidt 3822d7728c96SJan Schmidt path = btrfs_alloc_path(); 3823d7728c96SJan Schmidt if (!path) { 3824d7728c96SJan Schmidt ret = -ENOMEM; 3825d7728c96SJan Schmidt goto out; 3826d7728c96SJan Schmidt } 3827d7728c96SJan Schmidt 3828d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 3829d7728c96SJan Schmidt if (IS_ERR(ipa)) { 3830d7728c96SJan Schmidt ret = PTR_ERR(ipa); 3831d7728c96SJan Schmidt ipa = NULL; 3832d7728c96SJan Schmidt goto out; 3833d7728c96SJan Schmidt } 3834d7728c96SJan Schmidt 3835d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 3836d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 3837d7728c96SJan Schmidt if (IS_ERR(ipath)) { 3838d7728c96SJan Schmidt ret = PTR_ERR(ipath); 3839d7728c96SJan Schmidt ipath = NULL; 3840d7728c96SJan Schmidt goto out; 3841d7728c96SJan Schmidt } 3842d7728c96SJan Schmidt 3843d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 3844d7728c96SJan Schmidt if (ret < 0) 3845d7728c96SJan Schmidt goto out; 3846d7728c96SJan Schmidt 3847d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 3848745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 3849745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 3850740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 3851d7728c96SJan Schmidt } 3852d7728c96SJan Schmidt 3853718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)ipa->fspath, 3854718dc5faSOmar Sandoval ipath->fspath, size); 3855d7728c96SJan Schmidt if (ret) { 3856d7728c96SJan Schmidt ret = -EFAULT; 3857d7728c96SJan Schmidt goto out; 3858d7728c96SJan Schmidt } 3859d7728c96SJan Schmidt 3860d7728c96SJan Schmidt out: 3861d7728c96SJan Schmidt btrfs_free_path(path); 3862d7728c96SJan Schmidt free_ipath(ipath); 3863d7728c96SJan Schmidt kfree(ipa); 3864d7728c96SJan Schmidt 3865d7728c96SJan Schmidt return ret; 3866d7728c96SJan Schmidt } 3867d7728c96SJan Schmidt 3868d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 3869d7728c96SJan Schmidt { 3870d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 3871d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 3872d7728c96SJan Schmidt 3873d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 3874d7728c96SJan Schmidt inodes->bytes_left -= c; 3875d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 3876d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 3877d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 3878d7728c96SJan Schmidt inodes->elem_cnt += 3; 3879d7728c96SJan Schmidt } else { 3880d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 3881d7728c96SJan Schmidt inodes->bytes_left = 0; 3882d7728c96SJan Schmidt inodes->elem_missed += 3; 3883d7728c96SJan Schmidt } 3884d7728c96SJan Schmidt 3885d7728c96SJan Schmidt return 0; 3886d7728c96SJan Schmidt } 3887d7728c96SJan Schmidt 38882ff7e61eSJeff Mahoney static long btrfs_ioctl_logical_to_ino(struct btrfs_fs_info *fs_info, 3889d24a67b2SZygo Blaxell void __user *arg, int version) 3890d7728c96SJan Schmidt { 3891d7728c96SJan Schmidt int ret = 0; 3892d7728c96SJan Schmidt int size; 3893d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 3894d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 3895d7728c96SJan Schmidt struct btrfs_path *path = NULL; 3896d24a67b2SZygo Blaxell bool ignore_offset; 3897d7728c96SJan Schmidt 3898d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3899d7728c96SJan Schmidt return -EPERM; 3900d7728c96SJan Schmidt 3901d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 39027b9ea627SShailendra Verma if (IS_ERR(loi)) 39037b9ea627SShailendra Verma return PTR_ERR(loi); 3904d7728c96SJan Schmidt 3905d24a67b2SZygo Blaxell if (version == 1) { 3906d24a67b2SZygo Blaxell ignore_offset = false; 3907b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_64K); 3908d24a67b2SZygo Blaxell } else { 3909d24a67b2SZygo Blaxell /* All reserved bits must be 0 for now */ 3910d24a67b2SZygo Blaxell if (memchr_inv(loi->reserved, 0, sizeof(loi->reserved))) { 3911d24a67b2SZygo Blaxell ret = -EINVAL; 3912d24a67b2SZygo Blaxell goto out_loi; 3913d24a67b2SZygo Blaxell } 3914d24a67b2SZygo Blaxell /* Only accept flags we have defined so far */ 3915d24a67b2SZygo Blaxell if (loi->flags & ~(BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET)) { 3916d24a67b2SZygo Blaxell ret = -EINVAL; 3917d24a67b2SZygo Blaxell goto out_loi; 3918d24a67b2SZygo Blaxell } 3919d24a67b2SZygo Blaxell ignore_offset = loi->flags & BTRFS_LOGICAL_INO_ARGS_IGNORE_OFFSET; 3920b115e3bcSZygo Blaxell size = min_t(u32, loi->size, SZ_16M); 3921d24a67b2SZygo Blaxell } 3922d24a67b2SZygo Blaxell 3923d7728c96SJan Schmidt path = btrfs_alloc_path(); 3924d7728c96SJan Schmidt if (!path) { 3925d7728c96SJan Schmidt ret = -ENOMEM; 3926d7728c96SJan Schmidt goto out; 3927d7728c96SJan Schmidt } 3928d7728c96SJan Schmidt 3929d7728c96SJan Schmidt inodes = init_data_container(size); 3930d7728c96SJan Schmidt if (IS_ERR(inodes)) { 3931d7728c96SJan Schmidt ret = PTR_ERR(inodes); 3932d7728c96SJan Schmidt inodes = NULL; 3933d7728c96SJan Schmidt goto out; 3934d7728c96SJan Schmidt } 3935d7728c96SJan Schmidt 39362ff7e61eSJeff Mahoney ret = iterate_inodes_from_logical(loi->logical, fs_info, path, 3937d24a67b2SZygo Blaxell build_ino_list, inodes, ignore_offset); 3938df031f07SLiu Bo if (ret == -EINVAL) 3939d7728c96SJan Schmidt ret = -ENOENT; 3940d7728c96SJan Schmidt if (ret < 0) 3941d7728c96SJan Schmidt goto out; 3942d7728c96SJan Schmidt 3943718dc5faSOmar Sandoval ret = copy_to_user((void __user *)(unsigned long)loi->inodes, inodes, 3944718dc5faSOmar Sandoval size); 3945d7728c96SJan Schmidt if (ret) 3946d7728c96SJan Schmidt ret = -EFAULT; 3947d7728c96SJan Schmidt 3948d7728c96SJan Schmidt out: 3949d7728c96SJan Schmidt btrfs_free_path(path); 3950f54de068SDavid Sterba kvfree(inodes); 3951d24a67b2SZygo Blaxell out_loi: 3952d7728c96SJan Schmidt kfree(loi); 3953d7728c96SJan Schmidt 3954d7728c96SJan Schmidt return ret; 3955d7728c96SJan Schmidt } 3956d7728c96SJan Schmidt 3957008ef096SDavid Sterba void btrfs_update_ioctl_balance_args(struct btrfs_fs_info *fs_info, 3958c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 3959c9e9f97bSIlya Dryomov { 3960c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3961c9e9f97bSIlya Dryomov 3962c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 3963c9e9f97bSIlya Dryomov 39643009a62fSDavid Sterba if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) 3965837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 3966837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 3967837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 3968a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 3969a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 3970837d5b6eSIlya Dryomov 3971c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 3972c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 3973c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 397419a39dceSIlya Dryomov 397519a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 397619a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 397719a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 3978c9e9f97bSIlya Dryomov } 3979c9e9f97bSIlya Dryomov 39809ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 3981c9e9f97bSIlya Dryomov { 3982496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 3983c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 3984c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 3985c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 3986ed0fb78fSIlya Dryomov bool need_unlock; /* for mut. excl. ops lock */ 3987c9e9f97bSIlya Dryomov int ret; 3988c9e9f97bSIlya Dryomov 3989c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 3990c9e9f97bSIlya Dryomov return -EPERM; 3991c9e9f97bSIlya Dryomov 3992e54bfa31SLiu Bo ret = mnt_want_write_file(file); 39939ba1f6e4SLiu Bo if (ret) 39949ba1f6e4SLiu Bo return ret; 39959ba1f6e4SLiu Bo 3996ed0fb78fSIlya Dryomov again: 3997c3e1f96cSGoldwyn Rodrigues if (btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { 3998c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 3999ed0fb78fSIlya Dryomov need_unlock = true; 4000ed0fb78fSIlya Dryomov goto locked; 4001ed0fb78fSIlya Dryomov } 4002ed0fb78fSIlya Dryomov 4003ed0fb78fSIlya Dryomov /* 400401327610SNicholas D Steeves * mut. excl. ops lock is locked. Three possibilities: 4005ed0fb78fSIlya Dryomov * (1) some other op is running 4006ed0fb78fSIlya Dryomov * (2) balance is running 4007ed0fb78fSIlya Dryomov * (3) balance is paused -- special case (think resume) 4008ed0fb78fSIlya Dryomov */ 4009ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4010ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4011ed0fb78fSIlya Dryomov /* this is either (2) or (3) */ 40123009a62fSDavid Sterba if (!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4013ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4014dccdb07bSDavid Sterba /* 4015dccdb07bSDavid Sterba * Lock released to allow other waiters to continue, 4016dccdb07bSDavid Sterba * we'll reexamine the status again. 4017dccdb07bSDavid Sterba */ 4018ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4019ed0fb78fSIlya Dryomov 4020ed0fb78fSIlya Dryomov if (fs_info->balance_ctl && 40213009a62fSDavid Sterba !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { 4022ed0fb78fSIlya Dryomov /* this is (3) */ 4023ed0fb78fSIlya Dryomov need_unlock = false; 4024ed0fb78fSIlya Dryomov goto locked; 4025ed0fb78fSIlya Dryomov } 4026ed0fb78fSIlya Dryomov 4027ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4028ed0fb78fSIlya Dryomov goto again; 4029ed0fb78fSIlya Dryomov } else { 4030ed0fb78fSIlya Dryomov /* this is (2) */ 4031ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4032ed0fb78fSIlya Dryomov ret = -EINPROGRESS; 4033ed0fb78fSIlya Dryomov goto out; 4034ed0fb78fSIlya Dryomov } 4035ed0fb78fSIlya Dryomov } else { 4036ed0fb78fSIlya Dryomov /* this is (1) */ 4037ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4038e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4039ed0fb78fSIlya Dryomov goto out; 4040ed0fb78fSIlya Dryomov } 4041ed0fb78fSIlya Dryomov 4042ed0fb78fSIlya Dryomov locked: 4043c9e9f97bSIlya Dryomov 4044c9e9f97bSIlya Dryomov if (arg) { 4045c9e9f97bSIlya Dryomov bargs = memdup_user(arg, sizeof(*bargs)); 4046c9e9f97bSIlya Dryomov if (IS_ERR(bargs)) { 4047c9e9f97bSIlya Dryomov ret = PTR_ERR(bargs); 4048ed0fb78fSIlya Dryomov goto out_unlock; 4049c9e9f97bSIlya Dryomov } 4050de322263SIlya Dryomov 4051de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 4052de322263SIlya Dryomov if (!fs_info->balance_ctl) { 4053de322263SIlya Dryomov ret = -ENOTCONN; 4054de322263SIlya Dryomov goto out_bargs; 4055de322263SIlya Dryomov } 4056de322263SIlya Dryomov 4057de322263SIlya Dryomov bctl = fs_info->balance_ctl; 4058de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 4059de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 4060de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 4061de322263SIlya Dryomov 4062de322263SIlya Dryomov goto do_balance; 4063de322263SIlya Dryomov } 4064c9e9f97bSIlya Dryomov } else { 4065c9e9f97bSIlya Dryomov bargs = NULL; 4066c9e9f97bSIlya Dryomov } 4067c9e9f97bSIlya Dryomov 4068ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4069837d5b6eSIlya Dryomov ret = -EINPROGRESS; 4070837d5b6eSIlya Dryomov goto out_bargs; 4071837d5b6eSIlya Dryomov } 4072837d5b6eSIlya Dryomov 40738d2db785SDavid Sterba bctl = kzalloc(sizeof(*bctl), GFP_KERNEL); 4074c9e9f97bSIlya Dryomov if (!bctl) { 4075c9e9f97bSIlya Dryomov ret = -ENOMEM; 4076c9e9f97bSIlya Dryomov goto out_bargs; 4077c9e9f97bSIlya Dryomov } 4078c9e9f97bSIlya Dryomov 4079c9e9f97bSIlya Dryomov if (arg) { 4080c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 4081c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 4082c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 4083c9e9f97bSIlya Dryomov 4084c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 4085f43ffb60SIlya Dryomov } else { 4086f43ffb60SIlya Dryomov /* balance everything - no filters */ 4087f43ffb60SIlya Dryomov bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 4088c9e9f97bSIlya Dryomov } 4089c9e9f97bSIlya Dryomov 40908eb93459SDavid Sterba if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) { 40918eb93459SDavid Sterba ret = -EINVAL; 40920f89abf5SChristian Engelmayer goto out_bctl; 40938eb93459SDavid Sterba } 40948eb93459SDavid Sterba 4095de322263SIlya Dryomov do_balance: 4096c9e9f97bSIlya Dryomov /* 4097c3e1f96cSGoldwyn Rodrigues * Ownership of bctl and exclusive operation goes to btrfs_balance. 4098c3e1f96cSGoldwyn Rodrigues * bctl is freed in reset_balance_state, or, if restriper was paused 4099c3e1f96cSGoldwyn Rodrigues * all the way until unmount, in free_fs_info. The flag should be 4100c3e1f96cSGoldwyn Rodrigues * cleared after reset_balance_state. 4101c9e9f97bSIlya Dryomov */ 4102ed0fb78fSIlya Dryomov need_unlock = false; 4103ed0fb78fSIlya Dryomov 41046fcf6e2bSDavid Sterba ret = btrfs_balance(fs_info, bctl, bargs); 41050f89abf5SChristian Engelmayer bctl = NULL; 4106ed0fb78fSIlya Dryomov 4107d00c2d9cSFilipe Manana if ((ret == 0 || ret == -ECANCELED) && arg) { 4108c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 4109c9e9f97bSIlya Dryomov ret = -EFAULT; 4110c9e9f97bSIlya Dryomov } 4111c9e9f97bSIlya Dryomov 41120f89abf5SChristian Engelmayer out_bctl: 41130f89abf5SChristian Engelmayer kfree(bctl); 4114c9e9f97bSIlya Dryomov out_bargs: 4115c9e9f97bSIlya Dryomov kfree(bargs); 4116ed0fb78fSIlya Dryomov out_unlock: 4117c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4118ed0fb78fSIlya Dryomov if (need_unlock) 4119c3e1f96cSGoldwyn Rodrigues btrfs_exclop_finish(fs_info); 4120ed0fb78fSIlya Dryomov out: 4121e54bfa31SLiu Bo mnt_drop_write_file(file); 4122c9e9f97bSIlya Dryomov return ret; 4123c9e9f97bSIlya Dryomov } 4124c9e9f97bSIlya Dryomov 41252ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_ctl(struct btrfs_fs_info *fs_info, int cmd) 4126837d5b6eSIlya Dryomov { 4127837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4128837d5b6eSIlya Dryomov return -EPERM; 4129837d5b6eSIlya Dryomov 4130837d5b6eSIlya Dryomov switch (cmd) { 4131837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 41320b246afaSJeff Mahoney return btrfs_pause_balance(fs_info); 4133a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 41340b246afaSJeff Mahoney return btrfs_cancel_balance(fs_info); 4135837d5b6eSIlya Dryomov } 4136837d5b6eSIlya Dryomov 4137837d5b6eSIlya Dryomov return -EINVAL; 4138837d5b6eSIlya Dryomov } 4139837d5b6eSIlya Dryomov 41402ff7e61eSJeff Mahoney static long btrfs_ioctl_balance_progress(struct btrfs_fs_info *fs_info, 414119a39dceSIlya Dryomov void __user *arg) 414219a39dceSIlya Dryomov { 414319a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 414419a39dceSIlya Dryomov int ret = 0; 414519a39dceSIlya Dryomov 414619a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 414719a39dceSIlya Dryomov return -EPERM; 414819a39dceSIlya Dryomov 414919a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 415019a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 415119a39dceSIlya Dryomov ret = -ENOTCONN; 415219a39dceSIlya Dryomov goto out; 415319a39dceSIlya Dryomov } 415419a39dceSIlya Dryomov 41558d2db785SDavid Sterba bargs = kzalloc(sizeof(*bargs), GFP_KERNEL); 415619a39dceSIlya Dryomov if (!bargs) { 415719a39dceSIlya Dryomov ret = -ENOMEM; 415819a39dceSIlya Dryomov goto out; 415919a39dceSIlya Dryomov } 416019a39dceSIlya Dryomov 4161008ef096SDavid Sterba btrfs_update_ioctl_balance_args(fs_info, bargs); 416219a39dceSIlya Dryomov 416319a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 416419a39dceSIlya Dryomov ret = -EFAULT; 416519a39dceSIlya Dryomov 416619a39dceSIlya Dryomov kfree(bargs); 416719a39dceSIlya Dryomov out: 416819a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 416919a39dceSIlya Dryomov return ret; 417019a39dceSIlya Dryomov } 417119a39dceSIlya Dryomov 4172905b0ddaSMiao Xie static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 41735d13a37bSArne Jansen { 41740b246afaSJeff Mahoney struct inode *inode = file_inode(file); 41750b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 41765d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 41775d13a37bSArne Jansen int ret; 41785d13a37bSArne Jansen 41795d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 41805d13a37bSArne Jansen return -EPERM; 41815d13a37bSArne Jansen 4182905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4183905b0ddaSMiao Xie if (ret) 4184905b0ddaSMiao Xie return ret; 41855d13a37bSArne Jansen 41865d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4187905b0ddaSMiao Xie if (IS_ERR(sa)) { 4188905b0ddaSMiao Xie ret = PTR_ERR(sa); 4189905b0ddaSMiao Xie goto drop_write; 4190905b0ddaSMiao Xie } 41915d13a37bSArne Jansen 41920b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 41935d13a37bSArne Jansen 41945d13a37bSArne Jansen switch (sa->cmd) { 41955d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 4196340f1aa2SNikolay Borisov ret = btrfs_quota_enable(fs_info); 41975d13a37bSArne Jansen break; 41985d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 4199340f1aa2SNikolay Borisov ret = btrfs_quota_disable(fs_info); 42005d13a37bSArne Jansen break; 42015d13a37bSArne Jansen default: 42025d13a37bSArne Jansen ret = -EINVAL; 42035d13a37bSArne Jansen break; 42045d13a37bSArne Jansen } 42055d13a37bSArne Jansen 42065d13a37bSArne Jansen kfree(sa); 42070b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4208905b0ddaSMiao Xie drop_write: 4209905b0ddaSMiao Xie mnt_drop_write_file(file); 42105d13a37bSArne Jansen return ret; 42115d13a37bSArne Jansen } 42125d13a37bSArne Jansen 4213905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 42145d13a37bSArne Jansen { 42150b246afaSJeff Mahoney struct inode *inode = file_inode(file); 42160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 42170b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 42185d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 42195d13a37bSArne Jansen struct btrfs_trans_handle *trans; 42205d13a37bSArne Jansen int ret; 42215d13a37bSArne Jansen int err; 42225d13a37bSArne Jansen 42235d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 42245d13a37bSArne Jansen return -EPERM; 42255d13a37bSArne Jansen 4226905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4227905b0ddaSMiao Xie if (ret) 4228905b0ddaSMiao Xie return ret; 42295d13a37bSArne Jansen 42305d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4231905b0ddaSMiao Xie if (IS_ERR(sa)) { 4232905b0ddaSMiao Xie ret = PTR_ERR(sa); 4233905b0ddaSMiao Xie goto drop_write; 4234905b0ddaSMiao Xie } 42355d13a37bSArne Jansen 42365d13a37bSArne Jansen trans = btrfs_join_transaction(root); 42375d13a37bSArne Jansen if (IS_ERR(trans)) { 42385d13a37bSArne Jansen ret = PTR_ERR(trans); 42395d13a37bSArne Jansen goto out; 42405d13a37bSArne Jansen } 42415d13a37bSArne Jansen 42425d13a37bSArne Jansen if (sa->assign) { 42439f8a6ce6SLu Fengqi ret = btrfs_add_qgroup_relation(trans, sa->src, sa->dst); 42445d13a37bSArne Jansen } else { 424539616c27SLu Fengqi ret = btrfs_del_qgroup_relation(trans, sa->src, sa->dst); 42465d13a37bSArne Jansen } 42475d13a37bSArne Jansen 4248e082f563SQu Wenruo /* update qgroup status and info */ 4249280f8bd2SLu Fengqi err = btrfs_run_qgroups(trans); 4250e082f563SQu Wenruo if (err < 0) 42510b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, err, 4252ad8403dfSAnand Jain "failed to update qgroup status and info"); 42533a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 42545d13a37bSArne Jansen if (err && !ret) 42555d13a37bSArne Jansen ret = err; 42565d13a37bSArne Jansen 42575d13a37bSArne Jansen out: 42585d13a37bSArne Jansen kfree(sa); 4259905b0ddaSMiao Xie drop_write: 4260905b0ddaSMiao Xie mnt_drop_write_file(file); 42615d13a37bSArne Jansen return ret; 42625d13a37bSArne Jansen } 42635d13a37bSArne Jansen 4264905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 42655d13a37bSArne Jansen { 42660b246afaSJeff Mahoney struct inode *inode = file_inode(file); 42670b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 42685d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 42695d13a37bSArne Jansen struct btrfs_trans_handle *trans; 42705d13a37bSArne Jansen int ret; 42715d13a37bSArne Jansen int err; 42725d13a37bSArne Jansen 42735d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 42745d13a37bSArne Jansen return -EPERM; 42755d13a37bSArne Jansen 4276905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4277905b0ddaSMiao Xie if (ret) 4278905b0ddaSMiao Xie return ret; 42795d13a37bSArne Jansen 42805d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4281905b0ddaSMiao Xie if (IS_ERR(sa)) { 4282905b0ddaSMiao Xie ret = PTR_ERR(sa); 4283905b0ddaSMiao Xie goto drop_write; 4284905b0ddaSMiao Xie } 42855d13a37bSArne Jansen 4286d86e56cfSMiao Xie if (!sa->qgroupid) { 4287d86e56cfSMiao Xie ret = -EINVAL; 4288d86e56cfSMiao Xie goto out; 4289d86e56cfSMiao Xie } 4290d86e56cfSMiao Xie 42915d13a37bSArne Jansen trans = btrfs_join_transaction(root); 42925d13a37bSArne Jansen if (IS_ERR(trans)) { 42935d13a37bSArne Jansen ret = PTR_ERR(trans); 42945d13a37bSArne Jansen goto out; 42955d13a37bSArne Jansen } 42965d13a37bSArne Jansen 42975d13a37bSArne Jansen if (sa->create) { 429849a05ecdSLu Fengqi ret = btrfs_create_qgroup(trans, sa->qgroupid); 42995d13a37bSArne Jansen } else { 43003efbee1dSLu Fengqi ret = btrfs_remove_qgroup(trans, sa->qgroupid); 43015d13a37bSArne Jansen } 43025d13a37bSArne Jansen 43033a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 43045d13a37bSArne Jansen if (err && !ret) 43055d13a37bSArne Jansen ret = err; 43065d13a37bSArne Jansen 43075d13a37bSArne Jansen out: 43085d13a37bSArne Jansen kfree(sa); 4309905b0ddaSMiao Xie drop_write: 4310905b0ddaSMiao Xie mnt_drop_write_file(file); 43115d13a37bSArne Jansen return ret; 43125d13a37bSArne Jansen } 43135d13a37bSArne Jansen 4314905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 43155d13a37bSArne Jansen { 43160b246afaSJeff Mahoney struct inode *inode = file_inode(file); 43170b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 43185d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 43195d13a37bSArne Jansen struct btrfs_trans_handle *trans; 43205d13a37bSArne Jansen int ret; 43215d13a37bSArne Jansen int err; 43225d13a37bSArne Jansen u64 qgroupid; 43235d13a37bSArne Jansen 43245d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 43255d13a37bSArne Jansen return -EPERM; 43265d13a37bSArne Jansen 4327905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4328905b0ddaSMiao Xie if (ret) 4329905b0ddaSMiao Xie return ret; 43305d13a37bSArne Jansen 43315d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4332905b0ddaSMiao Xie if (IS_ERR(sa)) { 4333905b0ddaSMiao Xie ret = PTR_ERR(sa); 4334905b0ddaSMiao Xie goto drop_write; 4335905b0ddaSMiao Xie } 43365d13a37bSArne Jansen 43375d13a37bSArne Jansen trans = btrfs_join_transaction(root); 43385d13a37bSArne Jansen if (IS_ERR(trans)) { 43395d13a37bSArne Jansen ret = PTR_ERR(trans); 43405d13a37bSArne Jansen goto out; 43415d13a37bSArne Jansen } 43425d13a37bSArne Jansen 43435d13a37bSArne Jansen qgroupid = sa->qgroupid; 43445d13a37bSArne Jansen if (!qgroupid) { 43455d13a37bSArne Jansen /* take the current subvol as qgroup */ 43465d13a37bSArne Jansen qgroupid = root->root_key.objectid; 43475d13a37bSArne Jansen } 43485d13a37bSArne Jansen 4349f0042d5eSLu Fengqi ret = btrfs_limit_qgroup(trans, qgroupid, &sa->lim); 43505d13a37bSArne Jansen 43513a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 43525d13a37bSArne Jansen if (err && !ret) 43535d13a37bSArne Jansen ret = err; 43545d13a37bSArne Jansen 43555d13a37bSArne Jansen out: 43565d13a37bSArne Jansen kfree(sa); 4357905b0ddaSMiao Xie drop_write: 4358905b0ddaSMiao Xie mnt_drop_write_file(file); 43595d13a37bSArne Jansen return ret; 43605d13a37bSArne Jansen } 43615d13a37bSArne Jansen 43622f232036SJan Schmidt static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 43632f232036SJan Schmidt { 43640b246afaSJeff Mahoney struct inode *inode = file_inode(file); 43650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 43662f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 43672f232036SJan Schmidt int ret; 43682f232036SJan Schmidt 43692f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 43702f232036SJan Schmidt return -EPERM; 43712f232036SJan Schmidt 43722f232036SJan Schmidt ret = mnt_want_write_file(file); 43732f232036SJan Schmidt if (ret) 43742f232036SJan Schmidt return ret; 43752f232036SJan Schmidt 43762f232036SJan Schmidt qsa = memdup_user(arg, sizeof(*qsa)); 43772f232036SJan Schmidt if (IS_ERR(qsa)) { 43782f232036SJan Schmidt ret = PTR_ERR(qsa); 43792f232036SJan Schmidt goto drop_write; 43802f232036SJan Schmidt } 43812f232036SJan Schmidt 43822f232036SJan Schmidt if (qsa->flags) { 43832f232036SJan Schmidt ret = -EINVAL; 43842f232036SJan Schmidt goto out; 43852f232036SJan Schmidt } 43862f232036SJan Schmidt 43870b246afaSJeff Mahoney ret = btrfs_qgroup_rescan(fs_info); 43882f232036SJan Schmidt 43892f232036SJan Schmidt out: 43902f232036SJan Schmidt kfree(qsa); 43912f232036SJan Schmidt drop_write: 43922f232036SJan Schmidt mnt_drop_write_file(file); 43932f232036SJan Schmidt return ret; 43942f232036SJan Schmidt } 43952f232036SJan Schmidt 4396b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_status(struct btrfs_fs_info *fs_info, 4397b929c1d8SMarcos Paulo de Souza void __user *arg) 43982f232036SJan Schmidt { 43990afb603aSGoldwyn Rodrigues struct btrfs_ioctl_quota_rescan_args qsa = {0}; 44002f232036SJan Schmidt 44012f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 44022f232036SJan Schmidt return -EPERM; 44032f232036SJan Schmidt 44040b246afaSJeff Mahoney if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 44050afb603aSGoldwyn Rodrigues qsa.flags = 1; 44060afb603aSGoldwyn Rodrigues qsa.progress = fs_info->qgroup_rescan_progress.objectid; 44072f232036SJan Schmidt } 44082f232036SJan Schmidt 44090afb603aSGoldwyn Rodrigues if (copy_to_user(arg, &qsa, sizeof(qsa))) 4410991a3daeSAnand Jain return -EFAULT; 44112f232036SJan Schmidt 4412991a3daeSAnand Jain return 0; 44132f232036SJan Schmidt } 44142f232036SJan Schmidt 4415b929c1d8SMarcos Paulo de Souza static long btrfs_ioctl_quota_rescan_wait(struct btrfs_fs_info *fs_info, 4416b929c1d8SMarcos Paulo de Souza void __user *arg) 441757254b6eSJan Schmidt { 441857254b6eSJan Schmidt if (!capable(CAP_SYS_ADMIN)) 441957254b6eSJan Schmidt return -EPERM; 442057254b6eSJan Schmidt 44210b246afaSJeff Mahoney return btrfs_qgroup_wait_for_completion(fs_info, true); 442257254b6eSJan Schmidt } 442357254b6eSJan Schmidt 4424abccd00fSHugo Mills static long _btrfs_ioctl_set_received_subvol(struct file *file, 4425e4fed17aSChristian Brauner struct user_namespace *mnt_userns, 4426abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa) 44278ea05e3aSAlexander Block { 4428496ad9aaSAl Viro struct inode *inode = file_inode(file); 44290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 44308ea05e3aSAlexander Block struct btrfs_root *root = BTRFS_I(inode)->root; 44318ea05e3aSAlexander Block struct btrfs_root_item *root_item = &root->root_item; 44328ea05e3aSAlexander Block struct btrfs_trans_handle *trans; 443395582b00SDeepa Dinamani struct timespec64 ct = current_time(inode); 44348ea05e3aSAlexander Block int ret = 0; 4435dd5f9615SStefan Behrens int received_uuid_changed; 44368ea05e3aSAlexander Block 4437e4fed17aSChristian Brauner if (!inode_owner_or_capable(mnt_userns, inode)) 4438bd60ea0fSDavid Sterba return -EPERM; 4439bd60ea0fSDavid Sterba 44408ea05e3aSAlexander Block ret = mnt_want_write_file(file); 44418ea05e3aSAlexander Block if (ret < 0) 44428ea05e3aSAlexander Block return ret; 44438ea05e3aSAlexander Block 44440b246afaSJeff Mahoney down_write(&fs_info->subvol_sem); 44458ea05e3aSAlexander Block 44464a0cc7caSNikolay Borisov if (btrfs_ino(BTRFS_I(inode)) != BTRFS_FIRST_FREE_OBJECTID) { 44478ea05e3aSAlexander Block ret = -EINVAL; 44488ea05e3aSAlexander Block goto out; 44498ea05e3aSAlexander Block } 44508ea05e3aSAlexander Block 44518ea05e3aSAlexander Block if (btrfs_root_readonly(root)) { 44528ea05e3aSAlexander Block ret = -EROFS; 44538ea05e3aSAlexander Block goto out; 44548ea05e3aSAlexander Block } 44558ea05e3aSAlexander Block 4456dd5f9615SStefan Behrens /* 4457dd5f9615SStefan Behrens * 1 - root item 4458dd5f9615SStefan Behrens * 2 - uuid items (received uuid + subvol uuid) 4459dd5f9615SStefan Behrens */ 4460dd5f9615SStefan Behrens trans = btrfs_start_transaction(root, 3); 44618ea05e3aSAlexander Block if (IS_ERR(trans)) { 44628ea05e3aSAlexander Block ret = PTR_ERR(trans); 44638ea05e3aSAlexander Block trans = NULL; 44648ea05e3aSAlexander Block goto out; 44658ea05e3aSAlexander Block } 44668ea05e3aSAlexander Block 44678ea05e3aSAlexander Block sa->rtransid = trans->transid; 44688ea05e3aSAlexander Block sa->rtime.sec = ct.tv_sec; 44698ea05e3aSAlexander Block sa->rtime.nsec = ct.tv_nsec; 44708ea05e3aSAlexander Block 4471dd5f9615SStefan Behrens received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 4472dd5f9615SStefan Behrens BTRFS_UUID_SIZE); 4473dd5f9615SStefan Behrens if (received_uuid_changed && 4474d87ff758SNikolay Borisov !btrfs_is_empty_uuid(root_item->received_uuid)) { 4475d1957791SLu Fengqi ret = btrfs_uuid_tree_remove(trans, root_item->received_uuid, 4476dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4477dd5f9615SStefan Behrens root->root_key.objectid); 4478d87ff758SNikolay Borisov if (ret && ret != -ENOENT) { 4479d87ff758SNikolay Borisov btrfs_abort_transaction(trans, ret); 4480d87ff758SNikolay Borisov btrfs_end_transaction(trans); 4481d87ff758SNikolay Borisov goto out; 4482d87ff758SNikolay Borisov } 4483d87ff758SNikolay Borisov } 44848ea05e3aSAlexander Block memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 44858ea05e3aSAlexander Block btrfs_set_root_stransid(root_item, sa->stransid); 44868ea05e3aSAlexander Block btrfs_set_root_rtransid(root_item, sa->rtransid); 44873cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 44883cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 44893cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 44903cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 44918ea05e3aSAlexander Block 44920b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 44938ea05e3aSAlexander Block &root->root_key, &root->root_item); 44948ea05e3aSAlexander Block if (ret < 0) { 44953a45bb20SJeff Mahoney btrfs_end_transaction(trans); 44968ea05e3aSAlexander Block goto out; 4497dd5f9615SStefan Behrens } 4498dd5f9615SStefan Behrens if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 4499cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, sa->uuid, 4500dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4501dd5f9615SStefan Behrens root->root_key.objectid); 4502dd5f9615SStefan Behrens if (ret < 0 && ret != -EEXIST) { 450366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 4504efd38150SNikolay Borisov btrfs_end_transaction(trans); 4505dd5f9615SStefan Behrens goto out; 4506dd5f9615SStefan Behrens } 4507dd5f9615SStefan Behrens } 45083a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 4509abccd00fSHugo Mills out: 45100b246afaSJeff Mahoney up_write(&fs_info->subvol_sem); 4511abccd00fSHugo Mills mnt_drop_write_file(file); 4512abccd00fSHugo Mills return ret; 4513abccd00fSHugo Mills } 4514abccd00fSHugo Mills 4515abccd00fSHugo Mills #ifdef CONFIG_64BIT 4516abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol_32(struct file *file, 4517abccd00fSHugo Mills void __user *arg) 4518abccd00fSHugo Mills { 4519abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL; 4520abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *args64 = NULL; 4521abccd00fSHugo Mills int ret = 0; 4522abccd00fSHugo Mills 4523abccd00fSHugo Mills args32 = memdup_user(arg, sizeof(*args32)); 45247b9ea627SShailendra Verma if (IS_ERR(args32)) 45257b9ea627SShailendra Verma return PTR_ERR(args32); 4526abccd00fSHugo Mills 45278d2db785SDavid Sterba args64 = kmalloc(sizeof(*args64), GFP_KERNEL); 452884dbeb87SDan Carpenter if (!args64) { 452984dbeb87SDan Carpenter ret = -ENOMEM; 4530abccd00fSHugo Mills goto out; 4531abccd00fSHugo Mills } 4532abccd00fSHugo Mills 4533abccd00fSHugo Mills memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE); 4534abccd00fSHugo Mills args64->stransid = args32->stransid; 4535abccd00fSHugo Mills args64->rtransid = args32->rtransid; 4536abccd00fSHugo Mills args64->stime.sec = args32->stime.sec; 4537abccd00fSHugo Mills args64->stime.nsec = args32->stime.nsec; 4538abccd00fSHugo Mills args64->rtime.sec = args32->rtime.sec; 4539abccd00fSHugo Mills args64->rtime.nsec = args32->rtime.nsec; 4540abccd00fSHugo Mills args64->flags = args32->flags; 4541abccd00fSHugo Mills 4542e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), args64); 4543abccd00fSHugo Mills if (ret) 4544abccd00fSHugo Mills goto out; 4545abccd00fSHugo Mills 4546abccd00fSHugo Mills memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE); 4547abccd00fSHugo Mills args32->stransid = args64->stransid; 4548abccd00fSHugo Mills args32->rtransid = args64->rtransid; 4549abccd00fSHugo Mills args32->stime.sec = args64->stime.sec; 4550abccd00fSHugo Mills args32->stime.nsec = args64->stime.nsec; 4551abccd00fSHugo Mills args32->rtime.sec = args64->rtime.sec; 4552abccd00fSHugo Mills args32->rtime.nsec = args64->rtime.nsec; 4553abccd00fSHugo Mills args32->flags = args64->flags; 4554abccd00fSHugo Mills 4555abccd00fSHugo Mills ret = copy_to_user(arg, args32, sizeof(*args32)); 4556abccd00fSHugo Mills if (ret) 4557abccd00fSHugo Mills ret = -EFAULT; 4558abccd00fSHugo Mills 4559abccd00fSHugo Mills out: 4560abccd00fSHugo Mills kfree(args32); 4561abccd00fSHugo Mills kfree(args64); 4562abccd00fSHugo Mills return ret; 4563abccd00fSHugo Mills } 4564abccd00fSHugo Mills #endif 4565abccd00fSHugo Mills 4566abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol(struct file *file, 4567abccd00fSHugo Mills void __user *arg) 4568abccd00fSHugo Mills { 4569abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa = NULL; 4570abccd00fSHugo Mills int ret = 0; 4571abccd00fSHugo Mills 4572abccd00fSHugo Mills sa = memdup_user(arg, sizeof(*sa)); 45737b9ea627SShailendra Verma if (IS_ERR(sa)) 45747b9ea627SShailendra Verma return PTR_ERR(sa); 4575abccd00fSHugo Mills 4576e4fed17aSChristian Brauner ret = _btrfs_ioctl_set_received_subvol(file, file_mnt_user_ns(file), sa); 4577abccd00fSHugo Mills 4578abccd00fSHugo Mills if (ret) 4579abccd00fSHugo Mills goto out; 4580abccd00fSHugo Mills 45818ea05e3aSAlexander Block ret = copy_to_user(arg, sa, sizeof(*sa)); 45828ea05e3aSAlexander Block if (ret) 45838ea05e3aSAlexander Block ret = -EFAULT; 45848ea05e3aSAlexander Block 45858ea05e3aSAlexander Block out: 45868ea05e3aSAlexander Block kfree(sa); 45878ea05e3aSAlexander Block return ret; 45888ea05e3aSAlexander Block } 45898ea05e3aSAlexander Block 4590b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_fslabel(struct btrfs_fs_info *fs_info, 4591b929c1d8SMarcos Paulo de Souza void __user *arg) 4592867ab667Sjeff.liu { 4593a1b83ac5SAnand Jain size_t len; 4594867ab667Sjeff.liu int ret; 4595a1b83ac5SAnand Jain char label[BTRFS_LABEL_SIZE]; 4596a1b83ac5SAnand Jain 45970b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 45980b246afaSJeff Mahoney memcpy(label, fs_info->super_copy->label, BTRFS_LABEL_SIZE); 45990b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 4600a1b83ac5SAnand Jain 4601a1b83ac5SAnand Jain len = strnlen(label, BTRFS_LABEL_SIZE); 4602867ab667Sjeff.liu 4603867ab667Sjeff.liu if (len == BTRFS_LABEL_SIZE) { 46040b246afaSJeff Mahoney btrfs_warn(fs_info, 46050b246afaSJeff Mahoney "label is too long, return the first %zu bytes", 46060b246afaSJeff Mahoney --len); 4607867ab667Sjeff.liu } 4608867ab667Sjeff.liu 4609867ab667Sjeff.liu ret = copy_to_user(arg, label, len); 4610867ab667Sjeff.liu 4611867ab667Sjeff.liu return ret ? -EFAULT : 0; 4612867ab667Sjeff.liu } 4613867ab667Sjeff.liu 4614a8bfd4abSjeff.liu static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 4615a8bfd4abSjeff.liu { 46160b246afaSJeff Mahoney struct inode *inode = file_inode(file); 46170b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46180b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 46190b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 4620a8bfd4abSjeff.liu struct btrfs_trans_handle *trans; 4621a8bfd4abSjeff.liu char label[BTRFS_LABEL_SIZE]; 4622a8bfd4abSjeff.liu int ret; 4623a8bfd4abSjeff.liu 4624a8bfd4abSjeff.liu if (!capable(CAP_SYS_ADMIN)) 4625a8bfd4abSjeff.liu return -EPERM; 4626a8bfd4abSjeff.liu 4627a8bfd4abSjeff.liu if (copy_from_user(label, arg, sizeof(label))) 4628a8bfd4abSjeff.liu return -EFAULT; 4629a8bfd4abSjeff.liu 4630a8bfd4abSjeff.liu if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 46310b246afaSJeff Mahoney btrfs_err(fs_info, 46325d163e0eSJeff Mahoney "unable to set label with more than %d bytes", 4633a8bfd4abSjeff.liu BTRFS_LABEL_SIZE - 1); 4634a8bfd4abSjeff.liu return -EINVAL; 4635a8bfd4abSjeff.liu } 4636a8bfd4abSjeff.liu 4637a8bfd4abSjeff.liu ret = mnt_want_write_file(file); 4638a8bfd4abSjeff.liu if (ret) 4639a8bfd4abSjeff.liu return ret; 4640a8bfd4abSjeff.liu 4641a8bfd4abSjeff.liu trans = btrfs_start_transaction(root, 0); 4642a8bfd4abSjeff.liu if (IS_ERR(trans)) { 4643a8bfd4abSjeff.liu ret = PTR_ERR(trans); 4644a8bfd4abSjeff.liu goto out_unlock; 4645a8bfd4abSjeff.liu } 4646a8bfd4abSjeff.liu 46470b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 4648a8bfd4abSjeff.liu strcpy(super_block->label, label); 46490b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 46503a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 4651a8bfd4abSjeff.liu 4652a8bfd4abSjeff.liu out_unlock: 4653a8bfd4abSjeff.liu mnt_drop_write_file(file); 4654a8bfd4abSjeff.liu return ret; 4655a8bfd4abSjeff.liu } 4656a8bfd4abSjeff.liu 46572eaa055fSJeff Mahoney #define INIT_FEATURE_FLAGS(suffix) \ 46582eaa055fSJeff Mahoney { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 46592eaa055fSJeff Mahoney .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 46602eaa055fSJeff Mahoney .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 46612eaa055fSJeff Mahoney 4662d5131b65SDavid Sterba int btrfs_ioctl_get_supported_features(void __user *arg) 46632eaa055fSJeff Mahoney { 46644d4ab6d6SDavid Sterba static const struct btrfs_ioctl_feature_flags features[3] = { 46652eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SUPP), 46662eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_SET), 46672eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_CLEAR) 46682eaa055fSJeff Mahoney }; 46692eaa055fSJeff Mahoney 46702eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 46712eaa055fSJeff Mahoney return -EFAULT; 46722eaa055fSJeff Mahoney 46732eaa055fSJeff Mahoney return 0; 46742eaa055fSJeff Mahoney } 46752eaa055fSJeff Mahoney 4676b929c1d8SMarcos Paulo de Souza static int btrfs_ioctl_get_features(struct btrfs_fs_info *fs_info, 4677b929c1d8SMarcos Paulo de Souza void __user *arg) 46782eaa055fSJeff Mahoney { 46790b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 46802eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags features; 46812eaa055fSJeff Mahoney 46822eaa055fSJeff Mahoney features.compat_flags = btrfs_super_compat_flags(super_block); 46832eaa055fSJeff Mahoney features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 46842eaa055fSJeff Mahoney features.incompat_flags = btrfs_super_incompat_flags(super_block); 46852eaa055fSJeff Mahoney 46862eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 46872eaa055fSJeff Mahoney return -EFAULT; 46882eaa055fSJeff Mahoney 46892eaa055fSJeff Mahoney return 0; 46902eaa055fSJeff Mahoney } 46912eaa055fSJeff Mahoney 46922ff7e61eSJeff Mahoney static int check_feature_bits(struct btrfs_fs_info *fs_info, 46933b02a68aSJeff Mahoney enum btrfs_feature_set set, 46942eaa055fSJeff Mahoney u64 change_mask, u64 flags, u64 supported_flags, 46952eaa055fSJeff Mahoney u64 safe_set, u64 safe_clear) 46962eaa055fSJeff Mahoney { 4697f10152bcSDavid Sterba const char *type = btrfs_feature_set_name(set); 46983b02a68aSJeff Mahoney char *names; 46992eaa055fSJeff Mahoney u64 disallowed, unsupported; 47002eaa055fSJeff Mahoney u64 set_mask = flags & change_mask; 47012eaa055fSJeff Mahoney u64 clear_mask = ~flags & change_mask; 47022eaa055fSJeff Mahoney 47032eaa055fSJeff Mahoney unsupported = set_mask & ~supported_flags; 47042eaa055fSJeff Mahoney if (unsupported) { 47053b02a68aSJeff Mahoney names = btrfs_printable_features(set, unsupported); 47063b02a68aSJeff Mahoney if (names) { 47070b246afaSJeff Mahoney btrfs_warn(fs_info, 47083b02a68aSJeff Mahoney "this kernel does not support the %s feature bit%s", 47093b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 47103b02a68aSJeff Mahoney kfree(names); 47113b02a68aSJeff Mahoney } else 47120b246afaSJeff Mahoney btrfs_warn(fs_info, 47132eaa055fSJeff Mahoney "this kernel does not support %s bits 0x%llx", 47142eaa055fSJeff Mahoney type, unsupported); 47152eaa055fSJeff Mahoney return -EOPNOTSUPP; 47162eaa055fSJeff Mahoney } 47172eaa055fSJeff Mahoney 47182eaa055fSJeff Mahoney disallowed = set_mask & ~safe_set; 47192eaa055fSJeff Mahoney if (disallowed) { 47203b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 47213b02a68aSJeff Mahoney if (names) { 47220b246afaSJeff Mahoney btrfs_warn(fs_info, 47233b02a68aSJeff Mahoney "can't set the %s feature bit%s while mounted", 47243b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 47253b02a68aSJeff Mahoney kfree(names); 47263b02a68aSJeff Mahoney } else 47270b246afaSJeff Mahoney btrfs_warn(fs_info, 47282eaa055fSJeff Mahoney "can't set %s bits 0x%llx while mounted", 47292eaa055fSJeff Mahoney type, disallowed); 47302eaa055fSJeff Mahoney return -EPERM; 47312eaa055fSJeff Mahoney } 47322eaa055fSJeff Mahoney 47332eaa055fSJeff Mahoney disallowed = clear_mask & ~safe_clear; 47342eaa055fSJeff Mahoney if (disallowed) { 47353b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 47363b02a68aSJeff Mahoney if (names) { 47370b246afaSJeff Mahoney btrfs_warn(fs_info, 47383b02a68aSJeff Mahoney "can't clear the %s feature bit%s while mounted", 47393b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 47403b02a68aSJeff Mahoney kfree(names); 47413b02a68aSJeff Mahoney } else 47420b246afaSJeff Mahoney btrfs_warn(fs_info, 47432eaa055fSJeff Mahoney "can't clear %s bits 0x%llx while mounted", 47442eaa055fSJeff Mahoney type, disallowed); 47452eaa055fSJeff Mahoney return -EPERM; 47462eaa055fSJeff Mahoney } 47472eaa055fSJeff Mahoney 47482eaa055fSJeff Mahoney return 0; 47492eaa055fSJeff Mahoney } 47502eaa055fSJeff Mahoney 47512ff7e61eSJeff Mahoney #define check_feature(fs_info, change_mask, flags, mask_base) \ 47522ff7e61eSJeff Mahoney check_feature_bits(fs_info, FEAT_##mask_base, change_mask, flags, \ 47532eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 47542eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 47552eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 47562eaa055fSJeff Mahoney 47572eaa055fSJeff Mahoney static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 47582eaa055fSJeff Mahoney { 47590b246afaSJeff Mahoney struct inode *inode = file_inode(file); 47600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 47610b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 47620b246afaSJeff Mahoney struct btrfs_super_block *super_block = fs_info->super_copy; 47632eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags flags[2]; 47642eaa055fSJeff Mahoney struct btrfs_trans_handle *trans; 47652eaa055fSJeff Mahoney u64 newflags; 47662eaa055fSJeff Mahoney int ret; 47672eaa055fSJeff Mahoney 47682eaa055fSJeff Mahoney if (!capable(CAP_SYS_ADMIN)) 47692eaa055fSJeff Mahoney return -EPERM; 47702eaa055fSJeff Mahoney 47712eaa055fSJeff Mahoney if (copy_from_user(flags, arg, sizeof(flags))) 47722eaa055fSJeff Mahoney return -EFAULT; 47732eaa055fSJeff Mahoney 47742eaa055fSJeff Mahoney /* Nothing to do */ 47752eaa055fSJeff Mahoney if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 47762eaa055fSJeff Mahoney !flags[0].incompat_flags) 47772eaa055fSJeff Mahoney return 0; 47782eaa055fSJeff Mahoney 47792ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_flags, 47802eaa055fSJeff Mahoney flags[1].compat_flags, COMPAT); 47812eaa055fSJeff Mahoney if (ret) 47822eaa055fSJeff Mahoney return ret; 47832eaa055fSJeff Mahoney 47842ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].compat_ro_flags, 47852eaa055fSJeff Mahoney flags[1].compat_ro_flags, COMPAT_RO); 47862eaa055fSJeff Mahoney if (ret) 47872eaa055fSJeff Mahoney return ret; 47882eaa055fSJeff Mahoney 47892ff7e61eSJeff Mahoney ret = check_feature(fs_info, flags[0].incompat_flags, 47902eaa055fSJeff Mahoney flags[1].incompat_flags, INCOMPAT); 47912eaa055fSJeff Mahoney if (ret) 47922eaa055fSJeff Mahoney return ret; 47932eaa055fSJeff Mahoney 47947ab19625SDavid Sterba ret = mnt_want_write_file(file); 47957ab19625SDavid Sterba if (ret) 47967ab19625SDavid Sterba return ret; 47977ab19625SDavid Sterba 47988051aa1aSDavid Sterba trans = btrfs_start_transaction(root, 0); 47997ab19625SDavid Sterba if (IS_ERR(trans)) { 48007ab19625SDavid Sterba ret = PTR_ERR(trans); 48017ab19625SDavid Sterba goto out_drop_write; 48027ab19625SDavid Sterba } 48032eaa055fSJeff Mahoney 48040b246afaSJeff Mahoney spin_lock(&fs_info->super_lock); 48052eaa055fSJeff Mahoney newflags = btrfs_super_compat_flags(super_block); 48062eaa055fSJeff Mahoney newflags |= flags[0].compat_flags & flags[1].compat_flags; 48072eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 48082eaa055fSJeff Mahoney btrfs_set_super_compat_flags(super_block, newflags); 48092eaa055fSJeff Mahoney 48102eaa055fSJeff Mahoney newflags = btrfs_super_compat_ro_flags(super_block); 48112eaa055fSJeff Mahoney newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 48122eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 48132eaa055fSJeff Mahoney btrfs_set_super_compat_ro_flags(super_block, newflags); 48142eaa055fSJeff Mahoney 48152eaa055fSJeff Mahoney newflags = btrfs_super_incompat_flags(super_block); 48162eaa055fSJeff Mahoney newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 48172eaa055fSJeff Mahoney newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 48182eaa055fSJeff Mahoney btrfs_set_super_incompat_flags(super_block, newflags); 48190b246afaSJeff Mahoney spin_unlock(&fs_info->super_lock); 48202eaa055fSJeff Mahoney 48213a45bb20SJeff Mahoney ret = btrfs_commit_transaction(trans); 48227ab19625SDavid Sterba out_drop_write: 48237ab19625SDavid Sterba mnt_drop_write_file(file); 48247ab19625SDavid Sterba 48257ab19625SDavid Sterba return ret; 48262eaa055fSJeff Mahoney } 48272eaa055fSJeff Mahoney 48282351f431SJosef Bacik static int _btrfs_ioctl_send(struct file *file, void __user *argp, bool compat) 48292351f431SJosef Bacik { 48302351f431SJosef Bacik struct btrfs_ioctl_send_args *arg; 48312351f431SJosef Bacik int ret; 48322351f431SJosef Bacik 48332351f431SJosef Bacik if (compat) { 48342351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 48352351f431SJosef Bacik struct btrfs_ioctl_send_args_32 args32; 48362351f431SJosef Bacik 48372351f431SJosef Bacik ret = copy_from_user(&args32, argp, sizeof(args32)); 48382351f431SJosef Bacik if (ret) 48392351f431SJosef Bacik return -EFAULT; 48402351f431SJosef Bacik arg = kzalloc(sizeof(*arg), GFP_KERNEL); 48412351f431SJosef Bacik if (!arg) 48422351f431SJosef Bacik return -ENOMEM; 48432351f431SJosef Bacik arg->send_fd = args32.send_fd; 48442351f431SJosef Bacik arg->clone_sources_count = args32.clone_sources_count; 48452351f431SJosef Bacik arg->clone_sources = compat_ptr(args32.clone_sources); 48462351f431SJosef Bacik arg->parent_root = args32.parent_root; 48472351f431SJosef Bacik arg->flags = args32.flags; 48482351f431SJosef Bacik memcpy(arg->reserved, args32.reserved, 48492351f431SJosef Bacik sizeof(args32.reserved)); 48502351f431SJosef Bacik #else 48512351f431SJosef Bacik return -ENOTTY; 48522351f431SJosef Bacik #endif 48532351f431SJosef Bacik } else { 48542351f431SJosef Bacik arg = memdup_user(argp, sizeof(*arg)); 48552351f431SJosef Bacik if (IS_ERR(arg)) 48562351f431SJosef Bacik return PTR_ERR(arg); 48572351f431SJosef Bacik } 48582351f431SJosef Bacik ret = btrfs_ioctl_send(file, arg); 48592351f431SJosef Bacik kfree(arg); 48602351f431SJosef Bacik return ret; 48612351f431SJosef Bacik } 48622351f431SJosef Bacik 4863f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 4864f46b5a66SChristoph Hellwig cmd, unsigned long arg) 4865f46b5a66SChristoph Hellwig { 48660b246afaSJeff Mahoney struct inode *inode = file_inode(file); 48670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 48680b246afaSJeff Mahoney struct btrfs_root *root = BTRFS_I(inode)->root; 48694bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 4870f46b5a66SChristoph Hellwig 4871f46b5a66SChristoph Hellwig switch (cmd) { 48726cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 48736cbff00fSChristoph Hellwig return btrfs_ioctl_getversion(file, argp); 487440cf931fSEric Sandeen case FS_IOC_GETFSLABEL: 4875b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_fslabel(fs_info, argp); 487640cf931fSEric Sandeen case FS_IOC_SETFSLABEL: 487740cf931fSEric Sandeen return btrfs_ioctl_set_fslabel(file, argp); 4878f7039b1dSLi Dongyang case FITRIM: 4879b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_fitrim(fs_info, argp); 4880f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 4881fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 4882fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 4883fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 48843de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 4885fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 48866f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 48876f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 488876dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 4889949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, false); 4890949964c9SMarcos Paulo de Souza case BTRFS_IOC_SNAP_DESTROY_V2: 4891949964c9SMarcos Paulo de Souza return btrfs_ioctl_snap_destroy(file, argp, true); 48920caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 48930caa102dSLi Zefan return btrfs_ioctl_subvol_getflags(file, argp); 48940caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 48950caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 48966ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 48976ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 4898f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 48991e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 49001e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 49011e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 4902f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 4903198605a8SMiao Xie return btrfs_ioctl_resize(file, argp); 4904f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 49052ff7e61eSJeff Mahoney return btrfs_ioctl_add_dev(fs_info, argp); 4906f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 4907da24927bSMiao Xie return btrfs_ioctl_rm_dev(file, argp); 49086b526ed7SAnand Jain case BTRFS_IOC_RM_DEV_V2: 49096b526ed7SAnand Jain return btrfs_ioctl_rm_dev_v2(file, argp); 4910475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 49112ff7e61eSJeff Mahoney return btrfs_ioctl_fs_info(fs_info, argp); 4912475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 49132ff7e61eSJeff Mahoney return btrfs_ioctl_dev_info(fs_info, argp); 4914f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 49159ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, NULL); 4916ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 4917ac8e9819SChris Mason return btrfs_ioctl_tree_search(file, argp); 4918cc68a8a5SGerhard Heift case BTRFS_IOC_TREE_SEARCH_V2: 4919cc68a8a5SGerhard Heift return btrfs_ioctl_tree_search_v2(file, argp); 4920ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 4921ac8e9819SChris Mason return btrfs_ioctl_ino_lookup(file, argp); 4922d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 4923d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 4924d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 4925d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 1); 4926d24a67b2SZygo Blaxell case BTRFS_IOC_LOGICAL_INO_V2: 4927d24a67b2SZygo Blaxell return btrfs_ioctl_logical_to_ino(fs_info, argp, 2); 49281406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 49292ff7e61eSJeff Mahoney return btrfs_ioctl_space_info(fs_info, argp); 49309b199859SFilipe David Borba Manana case BTRFS_IOC_SYNC: { 49319b199859SFilipe David Borba Manana int ret; 49329b199859SFilipe David Borba Manana 49339db4dc24SNikolay Borisov ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false); 49349b199859SFilipe David Borba Manana if (ret) 49359b199859SFilipe David Borba Manana return ret; 49360b246afaSJeff Mahoney ret = btrfs_sync_fs(inode->i_sb, 1); 49372fad4e83SDavid Sterba /* 49382fad4e83SDavid Sterba * The transaction thread may want to do more work, 493901327610SNicholas D Steeves * namely it pokes the cleaner kthread that will start 49402fad4e83SDavid Sterba * processing uncleaned subvols. 49412fad4e83SDavid Sterba */ 49420b246afaSJeff Mahoney wake_up_process(fs_info->transaction_kthread); 49439b199859SFilipe David Borba Manana return ret; 49449b199859SFilipe David Borba Manana } 494546204592SSage Weil case BTRFS_IOC_START_SYNC: 49469a8c28beSMiao Xie return btrfs_ioctl_start_sync(root, argp); 494746204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 49482ff7e61eSJeff Mahoney return btrfs_ioctl_wait_sync(fs_info, argp); 4949475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 4950b8e95489SMiao Xie return btrfs_ioctl_scrub(file, argp); 4951475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 49522ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_cancel(fs_info); 4953475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 49542ff7e61eSJeff Mahoney return btrfs_ioctl_scrub_progress(fs_info, argp); 4955c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 49569ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 4957837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 49582ff7e61eSJeff Mahoney return btrfs_ioctl_balance_ctl(fs_info, arg); 495919a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 49602ff7e61eSJeff Mahoney return btrfs_ioctl_balance_progress(fs_info, argp); 49618ea05e3aSAlexander Block case BTRFS_IOC_SET_RECEIVED_SUBVOL: 49628ea05e3aSAlexander Block return btrfs_ioctl_set_received_subvol(file, argp); 4963abccd00fSHugo Mills #ifdef CONFIG_64BIT 4964abccd00fSHugo Mills case BTRFS_IOC_SET_RECEIVED_SUBVOL_32: 4965abccd00fSHugo Mills return btrfs_ioctl_set_received_subvol_32(file, argp); 4966abccd00fSHugo Mills #endif 496731db9f7cSAlexander Block case BTRFS_IOC_SEND: 49682351f431SJosef Bacik return _btrfs_ioctl_send(file, argp, false); 49692351f431SJosef Bacik #if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT) 49702351f431SJosef Bacik case BTRFS_IOC_SEND_32: 49712351f431SJosef Bacik return _btrfs_ioctl_send(file, argp, true); 49722351f431SJosef Bacik #endif 4973c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 49742ff7e61eSJeff Mahoney return btrfs_ioctl_get_dev_stats(fs_info, argp); 49755d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 4976905b0ddaSMiao Xie return btrfs_ioctl_quota_ctl(file, argp); 49775d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 4978905b0ddaSMiao Xie return btrfs_ioctl_qgroup_assign(file, argp); 49795d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 4980905b0ddaSMiao Xie return btrfs_ioctl_qgroup_create(file, argp); 49815d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 4982905b0ddaSMiao Xie return btrfs_ioctl_qgroup_limit(file, argp); 49832f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN: 49842f232036SJan Schmidt return btrfs_ioctl_quota_rescan(file, argp); 49852f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_STATUS: 4986b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_status(fs_info, argp); 498757254b6eSJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_WAIT: 4988b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_quota_rescan_wait(fs_info, argp); 49893f6bcfbdSStefan Behrens case BTRFS_IOC_DEV_REPLACE: 49902ff7e61eSJeff Mahoney return btrfs_ioctl_dev_replace(fs_info, argp); 49912eaa055fSJeff Mahoney case BTRFS_IOC_GET_SUPPORTED_FEATURES: 4992d5131b65SDavid Sterba return btrfs_ioctl_get_supported_features(argp); 49932eaa055fSJeff Mahoney case BTRFS_IOC_GET_FEATURES: 4994b929c1d8SMarcos Paulo de Souza return btrfs_ioctl_get_features(fs_info, argp); 49952eaa055fSJeff Mahoney case BTRFS_IOC_SET_FEATURES: 49962eaa055fSJeff Mahoney return btrfs_ioctl_set_features(file, argp); 4997b64ec075STomohiro Misono case BTRFS_IOC_GET_SUBVOL_INFO: 4998b64ec075STomohiro Misono return btrfs_ioctl_get_subvol_info(file, argp); 499942e4b520STomohiro Misono case BTRFS_IOC_GET_SUBVOL_ROOTREF: 500042e4b520STomohiro Misono return btrfs_ioctl_get_subvol_rootref(file, argp); 500123d0b79dSTomohiro Misono case BTRFS_IOC_INO_LOOKUP_USER: 500223d0b79dSTomohiro Misono return btrfs_ioctl_ino_lookup_user(file, argp); 500314605409SBoris Burkov case FS_IOC_ENABLE_VERITY: 500414605409SBoris Burkov return fsverity_ioctl_enable(file, (const void __user *)argp); 500514605409SBoris Burkov case FS_IOC_MEASURE_VERITY: 500614605409SBoris Burkov return fsverity_ioctl_measure(file, argp); 5007f46b5a66SChristoph Hellwig } 5008f46b5a66SChristoph Hellwig 5009f46b5a66SChristoph Hellwig return -ENOTTY; 5010f46b5a66SChristoph Hellwig } 50114c63c245SLuke Dashjr 50124c63c245SLuke Dashjr #ifdef CONFIG_COMPAT 50134c63c245SLuke Dashjr long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 50144c63c245SLuke Dashjr { 50152a362249SJeff Mahoney /* 50162a362249SJeff Mahoney * These all access 32-bit values anyway so no further 50172a362249SJeff Mahoney * handling is necessary. 50182a362249SJeff Mahoney */ 50194c63c245SLuke Dashjr switch (cmd) { 50204c63c245SLuke Dashjr case FS_IOC32_GETVERSION: 50214c63c245SLuke Dashjr cmd = FS_IOC_GETVERSION; 50224c63c245SLuke Dashjr break; 50234c63c245SLuke Dashjr } 50244c63c245SLuke Dashjr 50254c63c245SLuke Dashjr return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 50264c63c245SLuke Dashjr } 50274c63c245SLuke Dashjr #endif 5028