1f46b5a66SChristoph Hellwig /* 2f46b5a66SChristoph Hellwig * Copyright (C) 2007 Oracle. All rights reserved. 3f46b5a66SChristoph Hellwig * 4f46b5a66SChristoph Hellwig * This program is free software; you can redistribute it and/or 5f46b5a66SChristoph Hellwig * modify it under the terms of the GNU General Public 6f46b5a66SChristoph Hellwig * License v2 as published by the Free Software Foundation. 7f46b5a66SChristoph Hellwig * 8f46b5a66SChristoph Hellwig * This program is distributed in the hope that it will be useful, 9f46b5a66SChristoph Hellwig * but WITHOUT ANY WARRANTY; without even the implied warranty of 10f46b5a66SChristoph Hellwig * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11f46b5a66SChristoph Hellwig * General Public License for more details. 12f46b5a66SChristoph Hellwig * 13f46b5a66SChristoph Hellwig * You should have received a copy of the GNU General Public 14f46b5a66SChristoph Hellwig * License along with this program; if not, write to the 15f46b5a66SChristoph Hellwig * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16f46b5a66SChristoph Hellwig * Boston, MA 021110-1307, USA. 17f46b5a66SChristoph Hellwig */ 18f46b5a66SChristoph Hellwig 19f46b5a66SChristoph Hellwig #include <linux/kernel.h> 20f46b5a66SChristoph Hellwig #include <linux/bio.h> 21f46b5a66SChristoph Hellwig #include <linux/buffer_head.h> 22f46b5a66SChristoph Hellwig #include <linux/file.h> 23f46b5a66SChristoph Hellwig #include <linux/fs.h> 24cb8e7090SChristoph Hellwig #include <linux/fsnotify.h> 25f46b5a66SChristoph Hellwig #include <linux/pagemap.h> 26f46b5a66SChristoph Hellwig #include <linux/highmem.h> 27f46b5a66SChristoph Hellwig #include <linux/time.h> 28f46b5a66SChristoph Hellwig #include <linux/init.h> 29f46b5a66SChristoph Hellwig #include <linux/string.h> 30f46b5a66SChristoph Hellwig #include <linux/backing-dev.h> 31cb8e7090SChristoph Hellwig #include <linux/mount.h> 32f46b5a66SChristoph Hellwig #include <linux/mpage.h> 33cb8e7090SChristoph Hellwig #include <linux/namei.h> 34f46b5a66SChristoph Hellwig #include <linux/swap.h> 35f46b5a66SChristoph Hellwig #include <linux/writeback.h> 36f46b5a66SChristoph Hellwig #include <linux/statfs.h> 37f46b5a66SChristoph Hellwig #include <linux/compat.h> 38f46b5a66SChristoph Hellwig #include <linux/bit_spinlock.h> 39cb8e7090SChristoph Hellwig #include <linux/security.h> 40f46b5a66SChristoph Hellwig #include <linux/xattr.h> 417ea394f1SYan Zheng #include <linux/vmalloc.h> 425a0e3ad6STejun Heo #include <linux/slab.h> 43f7039b1dSLi Dongyang #include <linux/blkdev.h> 444b4e25f2SChris Mason #include "compat.h" 45f46b5a66SChristoph Hellwig #include "ctree.h" 46f46b5a66SChristoph Hellwig #include "disk-io.h" 47f46b5a66SChristoph Hellwig #include "transaction.h" 48f46b5a66SChristoph Hellwig #include "btrfs_inode.h" 49f46b5a66SChristoph Hellwig #include "ioctl.h" 50f46b5a66SChristoph Hellwig #include "print-tree.h" 51f46b5a66SChristoph Hellwig #include "volumes.h" 52925baeddSChris Mason #include "locking.h" 53581bb050SLi Zefan #include "inode-map.h" 54d7728c96SJan Schmidt #include "backref.h" 55606686eeSJosef Bacik #include "rcu-string.h" 56f46b5a66SChristoph Hellwig 576cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 586cbff00fSChristoph Hellwig static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags) 596cbff00fSChristoph Hellwig { 606cbff00fSChristoph Hellwig if (S_ISDIR(mode)) 616cbff00fSChristoph Hellwig return flags; 626cbff00fSChristoph Hellwig else if (S_ISREG(mode)) 636cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 646cbff00fSChristoph Hellwig else 656cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 666cbff00fSChristoph Hellwig } 67f46b5a66SChristoph Hellwig 686cbff00fSChristoph Hellwig /* 696cbff00fSChristoph Hellwig * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl. 706cbff00fSChristoph Hellwig */ 716cbff00fSChristoph Hellwig static unsigned int btrfs_flags_to_ioctl(unsigned int flags) 726cbff00fSChristoph Hellwig { 736cbff00fSChristoph Hellwig unsigned int iflags = 0; 746cbff00fSChristoph Hellwig 756cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 766cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 776cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 786cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 796cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 806cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 816cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 826cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 836cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 846cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 856cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 866cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 87d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 88d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 89d0092bddSLi Zefan 90d0092bddSLi Zefan if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS)) 91d0092bddSLi Zefan iflags |= FS_COMPR_FL; 92d0092bddSLi Zefan else if (flags & BTRFS_INODE_NOCOMPRESS) 93d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 946cbff00fSChristoph Hellwig 956cbff00fSChristoph Hellwig return iflags; 966cbff00fSChristoph Hellwig } 976cbff00fSChristoph Hellwig 986cbff00fSChristoph Hellwig /* 996cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 1006cbff00fSChristoph Hellwig */ 1016cbff00fSChristoph Hellwig void btrfs_update_iflags(struct inode *inode) 1026cbff00fSChristoph Hellwig { 1036cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1046cbff00fSChristoph Hellwig 1056cbff00fSChristoph Hellwig inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); 1066cbff00fSChristoph Hellwig 1076cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_SYNC) 1086cbff00fSChristoph Hellwig inode->i_flags |= S_SYNC; 1096cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_IMMUTABLE) 1106cbff00fSChristoph Hellwig inode->i_flags |= S_IMMUTABLE; 1116cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_APPEND) 1126cbff00fSChristoph Hellwig inode->i_flags |= S_APPEND; 1136cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_NOATIME) 1146cbff00fSChristoph Hellwig inode->i_flags |= S_NOATIME; 1156cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_DIRSYNC) 1166cbff00fSChristoph Hellwig inode->i_flags |= S_DIRSYNC; 1176cbff00fSChristoph Hellwig } 1186cbff00fSChristoph Hellwig 1196cbff00fSChristoph Hellwig /* 1206cbff00fSChristoph Hellwig * Inherit flags from the parent inode. 1216cbff00fSChristoph Hellwig * 122e27425d6SJosef Bacik * Currently only the compression flags and the cow flags are inherited. 1236cbff00fSChristoph Hellwig */ 1246cbff00fSChristoph Hellwig void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) 1256cbff00fSChristoph Hellwig { 1260b4dcea5SChris Mason unsigned int flags; 1270b4dcea5SChris Mason 1280b4dcea5SChris Mason if (!dir) 1290b4dcea5SChris Mason return; 1300b4dcea5SChris Mason 1310b4dcea5SChris Mason flags = BTRFS_I(dir)->flags; 1326cbff00fSChristoph Hellwig 133e27425d6SJosef Bacik if (flags & BTRFS_INODE_NOCOMPRESS) { 134e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; 135e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; 136e27425d6SJosef Bacik } else if (flags & BTRFS_INODE_COMPRESS) { 137e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; 138e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; 139e27425d6SJosef Bacik } 1406cbff00fSChristoph Hellwig 141e27425d6SJosef Bacik if (flags & BTRFS_INODE_NODATACOW) 142e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; 143e27425d6SJosef Bacik 1446cbff00fSChristoph Hellwig btrfs_update_iflags(inode); 1456cbff00fSChristoph Hellwig } 1466cbff00fSChristoph Hellwig 1476cbff00fSChristoph Hellwig static int btrfs_ioctl_getflags(struct file *file, void __user *arg) 1486cbff00fSChristoph Hellwig { 1496cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode); 1506cbff00fSChristoph Hellwig unsigned int flags = btrfs_flags_to_ioctl(ip->flags); 1516cbff00fSChristoph Hellwig 1526cbff00fSChristoph Hellwig if (copy_to_user(arg, &flags, sizeof(flags))) 1536cbff00fSChristoph Hellwig return -EFAULT; 1546cbff00fSChristoph Hellwig return 0; 1556cbff00fSChristoph Hellwig } 1566cbff00fSChristoph Hellwig 15775e7cb7fSLiu Bo static int check_flags(unsigned int flags) 15875e7cb7fSLiu Bo { 15975e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 16075e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 16175e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 162e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 163e1e8fb6aSLi Zefan FS_NOCOW_FL)) 16475e7cb7fSLiu Bo return -EOPNOTSUPP; 16575e7cb7fSLiu Bo 16675e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 16775e7cb7fSLiu Bo return -EINVAL; 16875e7cb7fSLiu Bo 16975e7cb7fSLiu Bo return 0; 17075e7cb7fSLiu Bo } 17175e7cb7fSLiu Bo 1726cbff00fSChristoph Hellwig static int btrfs_ioctl_setflags(struct file *file, void __user *arg) 1736cbff00fSChristoph Hellwig { 1746cbff00fSChristoph Hellwig struct inode *inode = file->f_path.dentry->d_inode; 1756cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1766cbff00fSChristoph Hellwig struct btrfs_root *root = ip->root; 1776cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 1786cbff00fSChristoph Hellwig unsigned int flags, oldflags; 1796cbff00fSChristoph Hellwig int ret; 180f062abf0SLi Zefan u64 ip_oldflags; 181f062abf0SLi Zefan unsigned int i_oldflags; 1826cbff00fSChristoph Hellwig 183b83cc969SLi Zefan if (btrfs_root_readonly(root)) 184b83cc969SLi Zefan return -EROFS; 185b83cc969SLi Zefan 1866cbff00fSChristoph Hellwig if (copy_from_user(&flags, arg, sizeof(flags))) 1876cbff00fSChristoph Hellwig return -EFAULT; 1886cbff00fSChristoph Hellwig 18975e7cb7fSLiu Bo ret = check_flags(flags); 19075e7cb7fSLiu Bo if (ret) 19175e7cb7fSLiu Bo return ret; 1926cbff00fSChristoph Hellwig 1932e149670SSerge E. Hallyn if (!inode_owner_or_capable(inode)) 1946cbff00fSChristoph Hellwig return -EACCES; 1956cbff00fSChristoph Hellwig 1966cbff00fSChristoph Hellwig mutex_lock(&inode->i_mutex); 1976cbff00fSChristoph Hellwig 198f062abf0SLi Zefan ip_oldflags = ip->flags; 199f062abf0SLi Zefan i_oldflags = inode->i_flags; 200f062abf0SLi Zefan 2016cbff00fSChristoph Hellwig flags = btrfs_mask_flags(inode->i_mode, flags); 2026cbff00fSChristoph Hellwig oldflags = btrfs_flags_to_ioctl(ip->flags); 2036cbff00fSChristoph Hellwig if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 2046cbff00fSChristoph Hellwig if (!capable(CAP_LINUX_IMMUTABLE)) { 2056cbff00fSChristoph Hellwig ret = -EPERM; 2066cbff00fSChristoph Hellwig goto out_unlock; 2076cbff00fSChristoph Hellwig } 2086cbff00fSChristoph Hellwig } 2096cbff00fSChristoph Hellwig 210a561be71SAl Viro ret = mnt_want_write_file(file); 2116cbff00fSChristoph Hellwig if (ret) 2126cbff00fSChristoph Hellwig goto out_unlock; 2136cbff00fSChristoph Hellwig 2146cbff00fSChristoph Hellwig if (flags & FS_SYNC_FL) 2156cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_SYNC; 2166cbff00fSChristoph Hellwig else 2176cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_SYNC; 2186cbff00fSChristoph Hellwig if (flags & FS_IMMUTABLE_FL) 2196cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_IMMUTABLE; 2206cbff00fSChristoph Hellwig else 2216cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_IMMUTABLE; 2226cbff00fSChristoph Hellwig if (flags & FS_APPEND_FL) 2236cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_APPEND; 2246cbff00fSChristoph Hellwig else 2256cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_APPEND; 2266cbff00fSChristoph Hellwig if (flags & FS_NODUMP_FL) 2276cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NODUMP; 2286cbff00fSChristoph Hellwig else 2296cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NODUMP; 2306cbff00fSChristoph Hellwig if (flags & FS_NOATIME_FL) 2316cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NOATIME; 2326cbff00fSChristoph Hellwig else 2336cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NOATIME; 2346cbff00fSChristoph Hellwig if (flags & FS_DIRSYNC_FL) 2356cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_DIRSYNC; 2366cbff00fSChristoph Hellwig else 2376cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_DIRSYNC; 238e1e8fb6aSLi Zefan if (flags & FS_NOCOW_FL) 239e1e8fb6aSLi Zefan ip->flags |= BTRFS_INODE_NODATACOW; 240e1e8fb6aSLi Zefan else 241e1e8fb6aSLi Zefan ip->flags &= ~BTRFS_INODE_NODATACOW; 2426cbff00fSChristoph Hellwig 24375e7cb7fSLiu Bo /* 24475e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 24575e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 24675e7cb7fSLiu Bo * things smaller. 24775e7cb7fSLiu Bo */ 24875e7cb7fSLiu Bo if (flags & FS_NOCOMP_FL) { 24975e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_COMPRESS; 25075e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_NOCOMPRESS; 25175e7cb7fSLiu Bo } else if (flags & FS_COMPR_FL) { 25275e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_COMPRESS; 25375e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_NOCOMPRESS; 254ebcb904dSLi Zefan } else { 255ebcb904dSLi Zefan ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 25675e7cb7fSLiu Bo } 2576cbff00fSChristoph Hellwig 2584da6f1a3SLi Zefan trans = btrfs_start_transaction(root, 1); 259f062abf0SLi Zefan if (IS_ERR(trans)) { 260f062abf0SLi Zefan ret = PTR_ERR(trans); 261f062abf0SLi Zefan goto out_drop; 262f062abf0SLi Zefan } 2636cbff00fSChristoph Hellwig 264306424ccSLi Zefan btrfs_update_iflags(inode); 2650c4d2d95SJosef Bacik inode_inc_iversion(inode); 266306424ccSLi Zefan inode->i_ctime = CURRENT_TIME; 2676cbff00fSChristoph Hellwig ret = btrfs_update_inode(trans, root, inode); 2686cbff00fSChristoph Hellwig 2696cbff00fSChristoph Hellwig btrfs_end_transaction(trans, root); 270f062abf0SLi Zefan out_drop: 271f062abf0SLi Zefan if (ret) { 272f062abf0SLi Zefan ip->flags = ip_oldflags; 273f062abf0SLi Zefan inode->i_flags = i_oldflags; 274f062abf0SLi Zefan } 2756cbff00fSChristoph Hellwig 2762a79f17eSAl Viro mnt_drop_write_file(file); 2776cbff00fSChristoph Hellwig out_unlock: 2786cbff00fSChristoph Hellwig mutex_unlock(&inode->i_mutex); 2792d4e6f6aSliubo return ret; 2806cbff00fSChristoph Hellwig } 2816cbff00fSChristoph Hellwig 2826cbff00fSChristoph Hellwig static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 2836cbff00fSChristoph Hellwig { 2846cbff00fSChristoph Hellwig struct inode *inode = file->f_path.dentry->d_inode; 2856cbff00fSChristoph Hellwig 2866cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 2876cbff00fSChristoph Hellwig } 288f46b5a66SChristoph Hellwig 289f7039b1dSLi Dongyang static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg) 290f7039b1dSLi Dongyang { 291815745cfSAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(fdentry(file)->d_sb); 292f7039b1dSLi Dongyang struct btrfs_device *device; 293f7039b1dSLi Dongyang struct request_queue *q; 294f7039b1dSLi Dongyang struct fstrim_range range; 295f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 296f7039b1dSLi Dongyang u64 num_devices = 0; 297815745cfSAl Viro u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 298f7039b1dSLi Dongyang int ret; 299f7039b1dSLi Dongyang 300f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 301f7039b1dSLi Dongyang return -EPERM; 302f7039b1dSLi Dongyang 3031f78160cSXiao Guangrong rcu_read_lock(); 3041f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 3051f78160cSXiao Guangrong dev_list) { 306f7039b1dSLi Dongyang if (!device->bdev) 307f7039b1dSLi Dongyang continue; 308f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 309f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 310f7039b1dSLi Dongyang num_devices++; 311f7039b1dSLi Dongyang minlen = min((u64)q->limits.discard_granularity, 312f7039b1dSLi Dongyang minlen); 313f7039b1dSLi Dongyang } 314f7039b1dSLi Dongyang } 3151f78160cSXiao Guangrong rcu_read_unlock(); 316f4c697e6SLukas Czerner 317f7039b1dSLi Dongyang if (!num_devices) 318f7039b1dSLi Dongyang return -EOPNOTSUPP; 319f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 320f7039b1dSLi Dongyang return -EFAULT; 321f4c697e6SLukas Czerner if (range.start > total_bytes) 322f4c697e6SLukas Czerner return -EINVAL; 323f7039b1dSLi Dongyang 324f4c697e6SLukas Czerner range.len = min(range.len, total_bytes - range.start); 325f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 326815745cfSAl Viro ret = btrfs_trim_fs(fs_info->tree_root, &range); 327f7039b1dSLi Dongyang if (ret < 0) 328f7039b1dSLi Dongyang return ret; 329f7039b1dSLi Dongyang 330f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 331f7039b1dSLi Dongyang return -EFAULT; 332f7039b1dSLi Dongyang 333f7039b1dSLi Dongyang return 0; 334f7039b1dSLi Dongyang } 335f7039b1dSLi Dongyang 336cb8e7090SChristoph Hellwig static noinline int create_subvol(struct btrfs_root *root, 337cb8e7090SChristoph Hellwig struct dentry *dentry, 33872fd032eSSage Weil char *name, int namelen, 3396f72c7e2SArne Jansen u64 *async_transid, 3406f72c7e2SArne Jansen struct btrfs_qgroup_inherit **inherit) 341f46b5a66SChristoph Hellwig { 342f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 343f46b5a66SChristoph Hellwig struct btrfs_key key; 344f46b5a66SChristoph Hellwig struct btrfs_root_item root_item; 345f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 346f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 34776dda93cSYan, Zheng struct btrfs_root *new_root; 3482fbe8c8aSAl Viro struct dentry *parent = dentry->d_parent; 3496a912213SJosef Bacik struct inode *dir; 350f46b5a66SChristoph Hellwig int ret; 351f46b5a66SChristoph Hellwig int err; 352f46b5a66SChristoph Hellwig u64 objectid; 353f46b5a66SChristoph Hellwig u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 3543de4586cSChris Mason u64 index = 0; 355f46b5a66SChristoph Hellwig 356581bb050SLi Zefan ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid); 3572fbe8c8aSAl Viro if (ret) 358a22285a6SYan, Zheng return ret; 3596a912213SJosef Bacik 3606a912213SJosef Bacik dir = parent->d_inode; 3616a912213SJosef Bacik 3629ed74f2dSJosef Bacik /* 3639ed74f2dSJosef Bacik * 1 - inode item 3649ed74f2dSJosef Bacik * 2 - refs 3659ed74f2dSJosef Bacik * 1 - root item 3669ed74f2dSJosef Bacik * 2 - dir items 3679ed74f2dSJosef Bacik */ 368a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 3692fbe8c8aSAl Viro if (IS_ERR(trans)) 370a22285a6SYan, Zheng return PTR_ERR(trans); 371f46b5a66SChristoph Hellwig 3726f72c7e2SArne Jansen ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, 3736f72c7e2SArne Jansen inherit ? *inherit : NULL); 3746f72c7e2SArne Jansen if (ret) 3756f72c7e2SArne Jansen goto fail; 3766f72c7e2SArne Jansen 3775d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 3785581a51aSJan Schmidt 0, objectid, NULL, 0, 0, 0); 3798e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 3808e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 3818e8a1e31SJosef Bacik goto fail; 3828e8a1e31SJosef Bacik } 383f46b5a66SChristoph Hellwig 3845d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 385f46b5a66SChristoph Hellwig btrfs_set_header_bytenr(leaf, leaf->start); 386f46b5a66SChristoph Hellwig btrfs_set_header_generation(leaf, trans->transid); 3875d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 388f46b5a66SChristoph Hellwig btrfs_set_header_owner(leaf, objectid); 389f46b5a66SChristoph Hellwig 390f46b5a66SChristoph Hellwig write_extent_buffer(leaf, root->fs_info->fsid, 391f46b5a66SChristoph Hellwig (unsigned long)btrfs_header_fsid(leaf), 392f46b5a66SChristoph Hellwig BTRFS_FSID_SIZE); 3935d4f98a2SYan Zheng write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, 3945d4f98a2SYan Zheng (unsigned long)btrfs_header_chunk_tree_uuid(leaf), 3955d4f98a2SYan Zheng BTRFS_UUID_SIZE); 396f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 397f46b5a66SChristoph Hellwig 398f46b5a66SChristoph Hellwig inode_item = &root_item.inode; 399f46b5a66SChristoph Hellwig memset(inode_item, 0, sizeof(*inode_item)); 400f46b5a66SChristoph Hellwig inode_item->generation = cpu_to_le64(1); 401f46b5a66SChristoph Hellwig inode_item->size = cpu_to_le64(3); 402f46b5a66SChristoph Hellwig inode_item->nlink = cpu_to_le32(1); 403a76a3cd4SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 404f46b5a66SChristoph Hellwig inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 405f46b5a66SChristoph Hellwig 40608fe4db1SLi Zefan root_item.flags = 0; 40708fe4db1SLi Zefan root_item.byte_limit = 0; 40808fe4db1SLi Zefan inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT); 40908fe4db1SLi Zefan 410f46b5a66SChristoph Hellwig btrfs_set_root_bytenr(&root_item, leaf->start); 41184234f3aSYan Zheng btrfs_set_root_generation(&root_item, trans->transid); 412f46b5a66SChristoph Hellwig btrfs_set_root_level(&root_item, 0); 413f46b5a66SChristoph Hellwig btrfs_set_root_refs(&root_item, 1); 41486b9f2ecSYan, Zheng btrfs_set_root_used(&root_item, leaf->len); 41580ff3856SYan Zheng btrfs_set_root_last_snapshot(&root_item, 0); 416f46b5a66SChristoph Hellwig 417f46b5a66SChristoph Hellwig memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress)); 418f46b5a66SChristoph Hellwig root_item.drop_level = 0; 419f46b5a66SChristoph Hellwig 420925baeddSChris Mason btrfs_tree_unlock(leaf); 421f46b5a66SChristoph Hellwig free_extent_buffer(leaf); 422f46b5a66SChristoph Hellwig leaf = NULL; 423f46b5a66SChristoph Hellwig 424f46b5a66SChristoph Hellwig btrfs_set_root_dirid(&root_item, new_dirid); 425f46b5a66SChristoph Hellwig 426f46b5a66SChristoph Hellwig key.objectid = objectid; 4275d4f98a2SYan Zheng key.offset = 0; 428f46b5a66SChristoph Hellwig btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 429f46b5a66SChristoph Hellwig ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, 430f46b5a66SChristoph Hellwig &root_item); 431f46b5a66SChristoph Hellwig if (ret) 432f46b5a66SChristoph Hellwig goto fail; 433f46b5a66SChristoph Hellwig 43476dda93cSYan, Zheng key.offset = (u64)-1; 43576dda93cSYan, Zheng new_root = btrfs_read_fs_root_no_name(root->fs_info, &key); 43679787eaaSJeff Mahoney if (IS_ERR(new_root)) { 43779787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, PTR_ERR(new_root)); 43879787eaaSJeff Mahoney ret = PTR_ERR(new_root); 43979787eaaSJeff Mahoney goto fail; 44079787eaaSJeff Mahoney } 44176dda93cSYan, Zheng 44276dda93cSYan, Zheng btrfs_record_root_in_trans(trans, new_root); 44376dda93cSYan, Zheng 444d82a6f1dSJosef Bacik ret = btrfs_create_subvol_root(trans, new_root, new_dirid); 445ce598979SMark Fasheh if (ret) { 446ce598979SMark Fasheh /* We potentially lose an unused inode item here */ 44779787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 448ce598979SMark Fasheh goto fail; 449ce598979SMark Fasheh } 450ce598979SMark Fasheh 451f46b5a66SChristoph Hellwig /* 452f46b5a66SChristoph Hellwig * insert the directory item 453f46b5a66SChristoph Hellwig */ 4543de4586cSChris Mason ret = btrfs_set_inode_index(dir, &index); 45579787eaaSJeff Mahoney if (ret) { 45679787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 45779787eaaSJeff Mahoney goto fail; 45879787eaaSJeff Mahoney } 4593de4586cSChris Mason 4603de4586cSChris Mason ret = btrfs_insert_dir_item(trans, root, 46116cdcec7SMiao Xie name, namelen, dir, &key, 4623de4586cSChris Mason BTRFS_FT_DIR, index); 46379787eaaSJeff Mahoney if (ret) { 46479787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 465f46b5a66SChristoph Hellwig goto fail; 46679787eaaSJeff Mahoney } 4670660b5afSChris Mason 46852c26179SYan Zheng btrfs_i_size_write(dir, dir->i_size + namelen * 2); 46952c26179SYan Zheng ret = btrfs_update_inode(trans, root, dir); 47052c26179SYan Zheng BUG_ON(ret); 47152c26179SYan Zheng 4720660b5afSChris Mason ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 4734df27c4dSYan, Zheng objectid, root->root_key.objectid, 47433345d01SLi Zefan btrfs_ino(dir), index, name, namelen); 47576dda93cSYan, Zheng 4760660b5afSChris Mason BUG_ON(ret); 4770660b5afSChris Mason 47876dda93cSYan, Zheng d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry)); 479f46b5a66SChristoph Hellwig fail: 48072fd032eSSage Weil if (async_transid) { 48172fd032eSSage Weil *async_transid = trans->transid; 48272fd032eSSage Weil err = btrfs_commit_transaction_async(trans, root, 1); 48372fd032eSSage Weil } else { 48476dda93cSYan, Zheng err = btrfs_commit_transaction(trans, root); 48572fd032eSSage Weil } 486f46b5a66SChristoph Hellwig if (err && !ret) 487f46b5a66SChristoph Hellwig ret = err; 488f46b5a66SChristoph Hellwig return ret; 489f46b5a66SChristoph Hellwig } 490f46b5a66SChristoph Hellwig 49172fd032eSSage Weil static int create_snapshot(struct btrfs_root *root, struct dentry *dentry, 492b83cc969SLi Zefan char *name, int namelen, u64 *async_transid, 4936f72c7e2SArne Jansen bool readonly, struct btrfs_qgroup_inherit **inherit) 494f46b5a66SChristoph Hellwig { 4952e4bfab9SYan, Zheng struct inode *inode; 496f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 497f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 4982e4bfab9SYan, Zheng int ret; 499f46b5a66SChristoph Hellwig 500f46b5a66SChristoph Hellwig if (!root->ref_cows) 501f46b5a66SChristoph Hellwig return -EINVAL; 502f46b5a66SChristoph Hellwig 503a22285a6SYan, Zheng pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS); 504a22285a6SYan, Zheng if (!pending_snapshot) 505a22285a6SYan, Zheng return -ENOMEM; 506a22285a6SYan, Zheng 507a22285a6SYan, Zheng btrfs_init_block_rsv(&pending_snapshot->block_rsv); 508a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 509a22285a6SYan, Zheng pending_snapshot->root = root; 510b83cc969SLi Zefan pending_snapshot->readonly = readonly; 5116f72c7e2SArne Jansen if (inherit) { 5126f72c7e2SArne Jansen pending_snapshot->inherit = *inherit; 5136f72c7e2SArne Jansen *inherit = NULL; /* take responsibility to free it */ 5146f72c7e2SArne Jansen } 515a22285a6SYan, Zheng 516a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->extent_root, 5); 517a22285a6SYan, Zheng if (IS_ERR(trans)) { 518a22285a6SYan, Zheng ret = PTR_ERR(trans); 519a22285a6SYan, Zheng goto fail; 520a22285a6SYan, Zheng } 521a22285a6SYan, Zheng 522a22285a6SYan, Zheng ret = btrfs_snap_reserve_metadata(trans, pending_snapshot); 523a22285a6SYan, Zheng BUG_ON(ret); 524a22285a6SYan, Zheng 5258351583eSJosef Bacik spin_lock(&root->fs_info->trans_lock); 526a22285a6SYan, Zheng list_add(&pending_snapshot->list, 527a22285a6SYan, Zheng &trans->transaction->pending_snapshots); 5288351583eSJosef Bacik spin_unlock(&root->fs_info->trans_lock); 52972fd032eSSage Weil if (async_transid) { 53072fd032eSSage Weil *async_transid = trans->transid; 53172fd032eSSage Weil ret = btrfs_commit_transaction_async(trans, 53272fd032eSSage Weil root->fs_info->extent_root, 1); 53372fd032eSSage Weil } else { 53472fd032eSSage Weil ret = btrfs_commit_transaction(trans, 53572fd032eSSage Weil root->fs_info->extent_root); 53672fd032eSSage Weil } 537a22285a6SYan, Zheng BUG_ON(ret); 538a22285a6SYan, Zheng 539a22285a6SYan, Zheng ret = pending_snapshot->error; 540f46b5a66SChristoph Hellwig if (ret) 5412e4bfab9SYan, Zheng goto fail; 542f46b5a66SChristoph Hellwig 54366b4ffd1SJosef Bacik ret = btrfs_orphan_cleanup(pending_snapshot->snap); 54466b4ffd1SJosef Bacik if (ret) 54566b4ffd1SJosef Bacik goto fail; 546f46b5a66SChristoph Hellwig 5472fbe8c8aSAl Viro inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry); 5482e4bfab9SYan, Zheng if (IS_ERR(inode)) { 5492e4bfab9SYan, Zheng ret = PTR_ERR(inode); 5502e4bfab9SYan, Zheng goto fail; 5512e4bfab9SYan, Zheng } 5522e4bfab9SYan, Zheng BUG_ON(!inode); 5532e4bfab9SYan, Zheng d_instantiate(dentry, inode); 5542e4bfab9SYan, Zheng ret = 0; 5552e4bfab9SYan, Zheng fail: 556a22285a6SYan, Zheng kfree(pending_snapshot); 557f46b5a66SChristoph Hellwig return ret; 558f46b5a66SChristoph Hellwig } 559f46b5a66SChristoph Hellwig 5604260f7c7SSage Weil /* copy of check_sticky in fs/namei.c() 5614260f7c7SSage Weil * It's inline, so penalty for filesystems that don't use sticky bit is 5624260f7c7SSage Weil * minimal. 5634260f7c7SSage Weil */ 5644260f7c7SSage Weil static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode) 5654260f7c7SSage Weil { 5664260f7c7SSage Weil uid_t fsuid = current_fsuid(); 5674260f7c7SSage Weil 5684260f7c7SSage Weil if (!(dir->i_mode & S_ISVTX)) 5694260f7c7SSage Weil return 0; 5704260f7c7SSage Weil if (inode->i_uid == fsuid) 5714260f7c7SSage Weil return 0; 5724260f7c7SSage Weil if (dir->i_uid == fsuid) 5734260f7c7SSage Weil return 0; 5744260f7c7SSage Weil return !capable(CAP_FOWNER); 5754260f7c7SSage Weil } 5764260f7c7SSage Weil 5774260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 5784260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 5794260f7c7SSage Weil * whether the type of victim is right. 5804260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 5814260f7c7SSage Weil * 2. We should have write and exec permissions on dir 5824260f7c7SSage Weil * 3. We can't remove anything from append-only dir 5834260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 5844260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 5854260f7c7SSage Weil * a. be owner of dir, or 5864260f7c7SSage Weil * b. be owner of victim, or 5874260f7c7SSage Weil * c. have CAP_FOWNER capability 5884260f7c7SSage Weil * 6. If the victim is append-only or immutable we can't do antyhing with 5894260f7c7SSage Weil * links pointing to it. 5904260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 5914260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 5924260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 5934260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 5944260f7c7SSage Weil * nfs_async_unlink(). 5954260f7c7SSage Weil */ 5964260f7c7SSage Weil 5974260f7c7SSage Weil static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir) 5984260f7c7SSage Weil { 5994260f7c7SSage Weil int error; 6004260f7c7SSage Weil 6014260f7c7SSage Weil if (!victim->d_inode) 6024260f7c7SSage Weil return -ENOENT; 6034260f7c7SSage Weil 6044260f7c7SSage Weil BUG_ON(victim->d_parent->d_inode != dir); 6054260f7c7SSage Weil audit_inode_child(victim, dir); 6064260f7c7SSage Weil 6074260f7c7SSage Weil error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 6084260f7c7SSage Weil if (error) 6094260f7c7SSage Weil return error; 6104260f7c7SSage Weil if (IS_APPEND(dir)) 6114260f7c7SSage Weil return -EPERM; 6124260f7c7SSage Weil if (btrfs_check_sticky(dir, victim->d_inode)|| 6134260f7c7SSage Weil IS_APPEND(victim->d_inode)|| 6144260f7c7SSage Weil IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 6154260f7c7SSage Weil return -EPERM; 6164260f7c7SSage Weil if (isdir) { 6174260f7c7SSage Weil if (!S_ISDIR(victim->d_inode->i_mode)) 6184260f7c7SSage Weil return -ENOTDIR; 6194260f7c7SSage Weil if (IS_ROOT(victim)) 6204260f7c7SSage Weil return -EBUSY; 6214260f7c7SSage Weil } else if (S_ISDIR(victim->d_inode->i_mode)) 6224260f7c7SSage Weil return -EISDIR; 6234260f7c7SSage Weil if (IS_DEADDIR(dir)) 6244260f7c7SSage Weil return -ENOENT; 6254260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 6264260f7c7SSage Weil return -EBUSY; 6274260f7c7SSage Weil return 0; 6284260f7c7SSage Weil } 6294260f7c7SSage Weil 630cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 631cb8e7090SChristoph Hellwig static inline int btrfs_may_create(struct inode *dir, struct dentry *child) 632cb8e7090SChristoph Hellwig { 633cb8e7090SChristoph Hellwig if (child->d_inode) 634cb8e7090SChristoph Hellwig return -EEXIST; 635cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 636cb8e7090SChristoph Hellwig return -ENOENT; 637cb8e7090SChristoph Hellwig return inode_permission(dir, MAY_WRITE | MAY_EXEC); 638cb8e7090SChristoph Hellwig } 639cb8e7090SChristoph Hellwig 640cb8e7090SChristoph Hellwig /* 641cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 642cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 643cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 644cb8e7090SChristoph Hellwig */ 64576dda93cSYan, Zheng static noinline int btrfs_mksubvol(struct path *parent, 64676dda93cSYan, Zheng char *name, int namelen, 64772fd032eSSage Weil struct btrfs_root *snap_src, 6486f72c7e2SArne Jansen u64 *async_transid, bool readonly, 6496f72c7e2SArne Jansen struct btrfs_qgroup_inherit **inherit) 650cb8e7090SChristoph Hellwig { 65176dda93cSYan, Zheng struct inode *dir = parent->dentry->d_inode; 652cb8e7090SChristoph Hellwig struct dentry *dentry; 653cb8e7090SChristoph Hellwig int error; 654cb8e7090SChristoph Hellwig 65576dda93cSYan, Zheng mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT); 656cb8e7090SChristoph Hellwig 657cb8e7090SChristoph Hellwig dentry = lookup_one_len(name, parent->dentry, namelen); 658cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 659cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 660cb8e7090SChristoph Hellwig goto out_unlock; 661cb8e7090SChristoph Hellwig 662cb8e7090SChristoph Hellwig error = -EEXIST; 663cb8e7090SChristoph Hellwig if (dentry->d_inode) 664cb8e7090SChristoph Hellwig goto out_dput; 665cb8e7090SChristoph Hellwig 66676dda93cSYan, Zheng error = btrfs_may_create(dir, dentry); 667cb8e7090SChristoph Hellwig if (error) 668a874a63eSLiu Bo goto out_dput; 669cb8e7090SChristoph Hellwig 67076dda93cSYan, Zheng down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 67176dda93cSYan, Zheng 67276dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 67376dda93cSYan, Zheng goto out_up_read; 67476dda93cSYan, Zheng 6753de4586cSChris Mason if (snap_src) { 6766f72c7e2SArne Jansen error = create_snapshot(snap_src, dentry, name, namelen, 6776f72c7e2SArne Jansen async_transid, readonly, inherit); 6783de4586cSChris Mason } else { 67976dda93cSYan, Zheng error = create_subvol(BTRFS_I(dir)->root, dentry, 6806f72c7e2SArne Jansen name, namelen, async_transid, inherit); 6813de4586cSChris Mason } 68276dda93cSYan, Zheng if (!error) 68376dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 68476dda93cSYan, Zheng out_up_read: 68576dda93cSYan, Zheng up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 686cb8e7090SChristoph Hellwig out_dput: 687cb8e7090SChristoph Hellwig dput(dentry); 688cb8e7090SChristoph Hellwig out_unlock: 68976dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 690cb8e7090SChristoph Hellwig return error; 691cb8e7090SChristoph Hellwig } 692cb8e7090SChristoph Hellwig 6934cb5300bSChris Mason /* 6944cb5300bSChris Mason * When we're defragging a range, we don't want to kick it off again 6954cb5300bSChris Mason * if it is really just waiting for delalloc to send it down. 6964cb5300bSChris Mason * If we find a nice big extent or delalloc range for the bytes in the 6974cb5300bSChris Mason * file you want to defrag, we return 0 to let you know to skip this 6984cb5300bSChris Mason * part of the file 6994cb5300bSChris Mason */ 7004cb5300bSChris Mason static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh) 7014cb5300bSChris Mason { 7024cb5300bSChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 7034cb5300bSChris Mason struct extent_map *em = NULL; 7044cb5300bSChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 7054cb5300bSChris Mason u64 end; 7064cb5300bSChris Mason 7074cb5300bSChris Mason read_lock(&em_tree->lock); 7084cb5300bSChris Mason em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 7094cb5300bSChris Mason read_unlock(&em_tree->lock); 7104cb5300bSChris Mason 7114cb5300bSChris Mason if (em) { 7124cb5300bSChris Mason end = extent_map_end(em); 7134cb5300bSChris Mason free_extent_map(em); 7144cb5300bSChris Mason if (end - offset > thresh) 7154cb5300bSChris Mason return 0; 7164cb5300bSChris Mason } 7174cb5300bSChris Mason /* if we already have a nice delalloc here, just stop */ 7184cb5300bSChris Mason thresh /= 2; 7194cb5300bSChris Mason end = count_range_bits(io_tree, &offset, offset + thresh, 7204cb5300bSChris Mason thresh, EXTENT_DELALLOC, 1); 7214cb5300bSChris Mason if (end >= thresh) 7224cb5300bSChris Mason return 0; 7234cb5300bSChris Mason return 1; 7244cb5300bSChris Mason } 7254cb5300bSChris Mason 7264cb5300bSChris Mason /* 7274cb5300bSChris Mason * helper function to walk through a file and find extents 7284cb5300bSChris Mason * newer than a specific transid, and smaller than thresh. 7294cb5300bSChris Mason * 7304cb5300bSChris Mason * This is used by the defragging code to find new and small 7314cb5300bSChris Mason * extents 7324cb5300bSChris Mason */ 7334cb5300bSChris Mason static int find_new_extents(struct btrfs_root *root, 7344cb5300bSChris Mason struct inode *inode, u64 newer_than, 7354cb5300bSChris Mason u64 *off, int thresh) 7364cb5300bSChris Mason { 7374cb5300bSChris Mason struct btrfs_path *path; 7384cb5300bSChris Mason struct btrfs_key min_key; 7394cb5300bSChris Mason struct btrfs_key max_key; 7404cb5300bSChris Mason struct extent_buffer *leaf; 7414cb5300bSChris Mason struct btrfs_file_extent_item *extent; 7424cb5300bSChris Mason int type; 7434cb5300bSChris Mason int ret; 744a4689d2bSDavid Sterba u64 ino = btrfs_ino(inode); 7454cb5300bSChris Mason 7464cb5300bSChris Mason path = btrfs_alloc_path(); 7474cb5300bSChris Mason if (!path) 7484cb5300bSChris Mason return -ENOMEM; 7494cb5300bSChris Mason 750a4689d2bSDavid Sterba min_key.objectid = ino; 7514cb5300bSChris Mason min_key.type = BTRFS_EXTENT_DATA_KEY; 7524cb5300bSChris Mason min_key.offset = *off; 7534cb5300bSChris Mason 754a4689d2bSDavid Sterba max_key.objectid = ino; 7554cb5300bSChris Mason max_key.type = (u8)-1; 7564cb5300bSChris Mason max_key.offset = (u64)-1; 7574cb5300bSChris Mason 7584cb5300bSChris Mason path->keep_locks = 1; 7594cb5300bSChris Mason 7604cb5300bSChris Mason while(1) { 7614cb5300bSChris Mason ret = btrfs_search_forward(root, &min_key, &max_key, 7624cb5300bSChris Mason path, 0, newer_than); 7634cb5300bSChris Mason if (ret != 0) 7644cb5300bSChris Mason goto none; 765a4689d2bSDavid Sterba if (min_key.objectid != ino) 7664cb5300bSChris Mason goto none; 7674cb5300bSChris Mason if (min_key.type != BTRFS_EXTENT_DATA_KEY) 7684cb5300bSChris Mason goto none; 7694cb5300bSChris Mason 7704cb5300bSChris Mason leaf = path->nodes[0]; 7714cb5300bSChris Mason extent = btrfs_item_ptr(leaf, path->slots[0], 7724cb5300bSChris Mason struct btrfs_file_extent_item); 7734cb5300bSChris Mason 7744cb5300bSChris Mason type = btrfs_file_extent_type(leaf, extent); 7754cb5300bSChris Mason if (type == BTRFS_FILE_EXTENT_REG && 7764cb5300bSChris Mason btrfs_file_extent_num_bytes(leaf, extent) < thresh && 7774cb5300bSChris Mason check_defrag_in_cache(inode, min_key.offset, thresh)) { 7784cb5300bSChris Mason *off = min_key.offset; 7794cb5300bSChris Mason btrfs_free_path(path); 7804cb5300bSChris Mason return 0; 7814cb5300bSChris Mason } 7824cb5300bSChris Mason 7834cb5300bSChris Mason if (min_key.offset == (u64)-1) 7844cb5300bSChris Mason goto none; 7854cb5300bSChris Mason 7864cb5300bSChris Mason min_key.offset++; 7874cb5300bSChris Mason btrfs_release_path(path); 7884cb5300bSChris Mason } 7894cb5300bSChris Mason none: 7904cb5300bSChris Mason btrfs_free_path(path); 7914cb5300bSChris Mason return -ENOENT; 7924cb5300bSChris Mason } 7934cb5300bSChris Mason 7946c282eb4SLi Zefan static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start) 79517ce6ef8SLiu Bo { 79617ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 797940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 7986c282eb4SLi Zefan struct extent_map *em; 7996c282eb4SLi Zefan u64 len = PAGE_CACHE_SIZE; 800940100a4SChris Mason 801940100a4SChris Mason /* 802940100a4SChris Mason * hopefully we have this extent in the tree already, try without 803940100a4SChris Mason * the full extent lock 804940100a4SChris Mason */ 805940100a4SChris Mason read_lock(&em_tree->lock); 806940100a4SChris Mason em = lookup_extent_mapping(em_tree, start, len); 807940100a4SChris Mason read_unlock(&em_tree->lock); 808940100a4SChris Mason 809940100a4SChris Mason if (!em) { 810940100a4SChris Mason /* get the big lock and read metadata off disk */ 811d0082371SJeff Mahoney lock_extent(io_tree, start, start + len - 1); 812940100a4SChris Mason em = btrfs_get_extent(inode, NULL, 0, start, len, 0); 813d0082371SJeff Mahoney unlock_extent(io_tree, start, start + len - 1); 814940100a4SChris Mason 8156cf8bfbfSDan Carpenter if (IS_ERR(em)) 8166c282eb4SLi Zefan return NULL; 817940100a4SChris Mason } 818940100a4SChris Mason 8196c282eb4SLi Zefan return em; 8206c282eb4SLi Zefan } 8216c282eb4SLi Zefan 8226c282eb4SLi Zefan static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em) 8236c282eb4SLi Zefan { 8246c282eb4SLi Zefan struct extent_map *next; 8256c282eb4SLi Zefan bool ret = true; 8266c282eb4SLi Zefan 8276c282eb4SLi Zefan /* this is the last extent */ 8286c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 8296c282eb4SLi Zefan return false; 8306c282eb4SLi Zefan 8316c282eb4SLi Zefan next = defrag_lookup_extent(inode, em->start + em->len); 8326c282eb4SLi Zefan if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 8336c282eb4SLi Zefan ret = false; 8346c282eb4SLi Zefan 8356c282eb4SLi Zefan free_extent_map(next); 8366c282eb4SLi Zefan return ret; 8376c282eb4SLi Zefan } 8386c282eb4SLi Zefan 8396c282eb4SLi Zefan static int should_defrag_range(struct inode *inode, u64 start, int thresh, 840a43a2111SAndrew Mahone u64 *last_len, u64 *skip, u64 *defrag_end, 841a43a2111SAndrew Mahone int compress) 8426c282eb4SLi Zefan { 8436c282eb4SLi Zefan struct extent_map *em; 8446c282eb4SLi Zefan int ret = 1; 8456c282eb4SLi Zefan bool next_mergeable = true; 8466c282eb4SLi Zefan 8476c282eb4SLi Zefan /* 8486c282eb4SLi Zefan * make sure that once we start defragging an extent, we keep on 8496c282eb4SLi Zefan * defragging it 8506c282eb4SLi Zefan */ 8516c282eb4SLi Zefan if (start < *defrag_end) 8526c282eb4SLi Zefan return 1; 8536c282eb4SLi Zefan 8546c282eb4SLi Zefan *skip = 0; 8556c282eb4SLi Zefan 8566c282eb4SLi Zefan em = defrag_lookup_extent(inode, start); 8576c282eb4SLi Zefan if (!em) 8586c282eb4SLi Zefan return 0; 8596c282eb4SLi Zefan 860940100a4SChris Mason /* this will cover holes, and inline extents */ 86117ce6ef8SLiu Bo if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 862940100a4SChris Mason ret = 0; 86317ce6ef8SLiu Bo goto out; 86417ce6ef8SLiu Bo } 86517ce6ef8SLiu Bo 8666c282eb4SLi Zefan next_mergeable = defrag_check_next_extent(inode, em); 867940100a4SChris Mason 868940100a4SChris Mason /* 8696c282eb4SLi Zefan * we hit a real extent, if it is big or the next extent is not a 8706c282eb4SLi Zefan * real extent, don't bother defragging it 871940100a4SChris Mason */ 872a43a2111SAndrew Mahone if (!compress && (*last_len == 0 || *last_len >= thresh) && 8736c282eb4SLi Zefan (em->len >= thresh || !next_mergeable)) 874940100a4SChris Mason ret = 0; 87517ce6ef8SLiu Bo out: 876940100a4SChris Mason /* 877940100a4SChris Mason * last_len ends up being a counter of how many bytes we've defragged. 878940100a4SChris Mason * every time we choose not to defrag an extent, we reset *last_len 879940100a4SChris Mason * so that the next tiny extent will force a defrag. 880940100a4SChris Mason * 881940100a4SChris Mason * The end result of this is that tiny extents before a single big 882940100a4SChris Mason * extent will force at least part of that big extent to be defragged. 883940100a4SChris Mason */ 884940100a4SChris Mason if (ret) { 885940100a4SChris Mason *defrag_end = extent_map_end(em); 886940100a4SChris Mason } else { 887940100a4SChris Mason *last_len = 0; 888940100a4SChris Mason *skip = extent_map_end(em); 889940100a4SChris Mason *defrag_end = 0; 890940100a4SChris Mason } 891940100a4SChris Mason 892940100a4SChris Mason free_extent_map(em); 893940100a4SChris Mason return ret; 894940100a4SChris Mason } 895940100a4SChris Mason 8964cb5300bSChris Mason /* 8974cb5300bSChris Mason * it doesn't do much good to defrag one or two pages 8984cb5300bSChris Mason * at a time. This pulls in a nice chunk of pages 8994cb5300bSChris Mason * to COW and defrag. 9004cb5300bSChris Mason * 9014cb5300bSChris Mason * It also makes sure the delalloc code has enough 9024cb5300bSChris Mason * dirty data to avoid making new small extents as part 9034cb5300bSChris Mason * of the defrag 9044cb5300bSChris Mason * 9054cb5300bSChris Mason * It's a good idea to start RA on this range 9064cb5300bSChris Mason * before calling this. 9074cb5300bSChris Mason */ 9084cb5300bSChris Mason static int cluster_pages_for_defrag(struct inode *inode, 9094cb5300bSChris Mason struct page **pages, 9104cb5300bSChris Mason unsigned long start_index, 9114cb5300bSChris Mason int num_pages) 912f46b5a66SChristoph Hellwig { 9134cb5300bSChris Mason unsigned long file_end; 9144cb5300bSChris Mason u64 isize = i_size_read(inode); 915f46b5a66SChristoph Hellwig u64 page_start; 916f46b5a66SChristoph Hellwig u64 page_end; 9171f12bd06SLiu Bo u64 page_cnt; 9184cb5300bSChris Mason int ret; 9194cb5300bSChris Mason int i; 9204cb5300bSChris Mason int i_done; 9214cb5300bSChris Mason struct btrfs_ordered_extent *ordered; 9224cb5300bSChris Mason struct extent_state *cached_state = NULL; 923600a45e1SMiao Xie struct extent_io_tree *tree; 9243b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 9254cb5300bSChris Mason 9264cb5300bSChris Mason file_end = (isize - 1) >> PAGE_CACHE_SHIFT; 9271f12bd06SLiu Bo if (!isize || start_index > file_end) 9281f12bd06SLiu Bo return 0; 9291f12bd06SLiu Bo 9301f12bd06SLiu Bo page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1); 9314cb5300bSChris Mason 9324cb5300bSChris Mason ret = btrfs_delalloc_reserve_space(inode, 9331f12bd06SLiu Bo page_cnt << PAGE_CACHE_SHIFT); 9344cb5300bSChris Mason if (ret) 9354cb5300bSChris Mason return ret; 9364cb5300bSChris Mason i_done = 0; 937600a45e1SMiao Xie tree = &BTRFS_I(inode)->io_tree; 9384cb5300bSChris Mason 9394cb5300bSChris Mason /* step one, lock all the pages */ 9401f12bd06SLiu Bo for (i = 0; i < page_cnt; i++) { 9414cb5300bSChris Mason struct page *page; 942600a45e1SMiao Xie again: 943a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, 9443b16a4e3SJosef Bacik start_index + i, mask); 9454cb5300bSChris Mason if (!page) 9464cb5300bSChris Mason break; 9474cb5300bSChris Mason 948600a45e1SMiao Xie page_start = page_offset(page); 949600a45e1SMiao Xie page_end = page_start + PAGE_CACHE_SIZE - 1; 950600a45e1SMiao Xie while (1) { 951d0082371SJeff Mahoney lock_extent(tree, page_start, page_end); 952600a45e1SMiao Xie ordered = btrfs_lookup_ordered_extent(inode, 953600a45e1SMiao Xie page_start); 954d0082371SJeff Mahoney unlock_extent(tree, page_start, page_end); 955600a45e1SMiao Xie if (!ordered) 956600a45e1SMiao Xie break; 957600a45e1SMiao Xie 958600a45e1SMiao Xie unlock_page(page); 959600a45e1SMiao Xie btrfs_start_ordered_extent(inode, ordered, 1); 960600a45e1SMiao Xie btrfs_put_ordered_extent(ordered); 961600a45e1SMiao Xie lock_page(page); 9621f12bd06SLiu Bo /* 9631f12bd06SLiu Bo * we unlocked the page above, so we need check if 9641f12bd06SLiu Bo * it was released or not. 9651f12bd06SLiu Bo */ 9661f12bd06SLiu Bo if (page->mapping != inode->i_mapping) { 9671f12bd06SLiu Bo unlock_page(page); 9681f12bd06SLiu Bo page_cache_release(page); 9691f12bd06SLiu Bo goto again; 9701f12bd06SLiu Bo } 971600a45e1SMiao Xie } 972600a45e1SMiao Xie 9734cb5300bSChris Mason if (!PageUptodate(page)) { 9744cb5300bSChris Mason btrfs_readpage(NULL, page); 9754cb5300bSChris Mason lock_page(page); 9764cb5300bSChris Mason if (!PageUptodate(page)) { 9774cb5300bSChris Mason unlock_page(page); 9784cb5300bSChris Mason page_cache_release(page); 9794cb5300bSChris Mason ret = -EIO; 9804cb5300bSChris Mason break; 9814cb5300bSChris Mason } 9824cb5300bSChris Mason } 983600a45e1SMiao Xie 984600a45e1SMiao Xie if (page->mapping != inode->i_mapping) { 985600a45e1SMiao Xie unlock_page(page); 986600a45e1SMiao Xie page_cache_release(page); 987600a45e1SMiao Xie goto again; 988600a45e1SMiao Xie } 989600a45e1SMiao Xie 9904cb5300bSChris Mason pages[i] = page; 9914cb5300bSChris Mason i_done++; 9924cb5300bSChris Mason } 9934cb5300bSChris Mason if (!i_done || ret) 9944cb5300bSChris Mason goto out; 9954cb5300bSChris Mason 9964cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 9974cb5300bSChris Mason goto out; 9984cb5300bSChris Mason 9994cb5300bSChris Mason /* 10004cb5300bSChris Mason * so now we have a nice long stream of locked 10014cb5300bSChris Mason * and up to date pages, lets wait on them 10024cb5300bSChris Mason */ 10034cb5300bSChris Mason for (i = 0; i < i_done; i++) 10044cb5300bSChris Mason wait_on_page_writeback(pages[i]); 10054cb5300bSChris Mason 10064cb5300bSChris Mason page_start = page_offset(pages[0]); 10074cb5300bSChris Mason page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE; 10084cb5300bSChris Mason 10094cb5300bSChris Mason lock_extent_bits(&BTRFS_I(inode)->io_tree, 1010d0082371SJeff Mahoney page_start, page_end - 1, 0, &cached_state); 10114cb5300bSChris Mason clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, 10124cb5300bSChris Mason page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | 10134cb5300bSChris Mason EXTENT_DO_ACCOUNTING, 0, 0, &cached_state, 10144cb5300bSChris Mason GFP_NOFS); 10154cb5300bSChris Mason 10161f12bd06SLiu Bo if (i_done != page_cnt) { 10179e0baf60SJosef Bacik spin_lock(&BTRFS_I(inode)->lock); 10189e0baf60SJosef Bacik BTRFS_I(inode)->outstanding_extents++; 10199e0baf60SJosef Bacik spin_unlock(&BTRFS_I(inode)->lock); 10204cb5300bSChris Mason btrfs_delalloc_release_space(inode, 10211f12bd06SLiu Bo (page_cnt - i_done) << PAGE_CACHE_SHIFT); 10224cb5300bSChris Mason } 10234cb5300bSChris Mason 10244cb5300bSChris Mason 10254cb5300bSChris Mason btrfs_set_extent_delalloc(inode, page_start, page_end - 1, 10264cb5300bSChris Mason &cached_state); 10274cb5300bSChris Mason 10284cb5300bSChris Mason unlock_extent_cached(&BTRFS_I(inode)->io_tree, 10294cb5300bSChris Mason page_start, page_end - 1, &cached_state, 10304cb5300bSChris Mason GFP_NOFS); 10314cb5300bSChris Mason 10324cb5300bSChris Mason for (i = 0; i < i_done; i++) { 10334cb5300bSChris Mason clear_page_dirty_for_io(pages[i]); 10344cb5300bSChris Mason ClearPageChecked(pages[i]); 10354cb5300bSChris Mason set_page_extent_mapped(pages[i]); 10364cb5300bSChris Mason set_page_dirty(pages[i]); 10374cb5300bSChris Mason unlock_page(pages[i]); 10384cb5300bSChris Mason page_cache_release(pages[i]); 10394cb5300bSChris Mason } 10404cb5300bSChris Mason return i_done; 10414cb5300bSChris Mason out: 10424cb5300bSChris Mason for (i = 0; i < i_done; i++) { 10434cb5300bSChris Mason unlock_page(pages[i]); 10444cb5300bSChris Mason page_cache_release(pages[i]); 10454cb5300bSChris Mason } 10461f12bd06SLiu Bo btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT); 10474cb5300bSChris Mason return ret; 10484cb5300bSChris Mason 10494cb5300bSChris Mason } 10504cb5300bSChris Mason 10514cb5300bSChris Mason int btrfs_defrag_file(struct inode *inode, struct file *file, 10524cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 10534cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 10544cb5300bSChris Mason { 10554cb5300bSChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 10564cb5300bSChris Mason struct file_ra_state *ra = NULL; 10574cb5300bSChris Mason unsigned long last_index; 1058151a31b2SLi Zefan u64 isize = i_size_read(inode); 1059940100a4SChris Mason u64 last_len = 0; 1060940100a4SChris Mason u64 skip = 0; 1061940100a4SChris Mason u64 defrag_end = 0; 10624cb5300bSChris Mason u64 newer_off = range->start; 1063f46b5a66SChristoph Hellwig unsigned long i; 1064008873eaSLi Zefan unsigned long ra_index = 0; 1065f46b5a66SChristoph Hellwig int ret; 10664cb5300bSChris Mason int defrag_count = 0; 10671a419d85SLi Zefan int compress_type = BTRFS_COMPRESS_ZLIB; 10684cb5300bSChris Mason int extent_thresh = range->extent_thresh; 1069008873eaSLi Zefan int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT; 1070008873eaSLi Zefan int cluster = max_cluster; 10714cb5300bSChris Mason u64 new_align = ~((u64)128 * 1024 - 1); 10724cb5300bSChris Mason struct page **pages = NULL; 10734cb5300bSChris Mason 10744cb5300bSChris Mason if (extent_thresh == 0) 10754cb5300bSChris Mason extent_thresh = 256 * 1024; 10761a419d85SLi Zefan 10771a419d85SLi Zefan if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 10781a419d85SLi Zefan if (range->compress_type > BTRFS_COMPRESS_TYPES) 10791a419d85SLi Zefan return -EINVAL; 10801a419d85SLi Zefan if (range->compress_type) 10811a419d85SLi Zefan compress_type = range->compress_type; 10821a419d85SLi Zefan } 1083f46b5a66SChristoph Hellwig 1084151a31b2SLi Zefan if (isize == 0) 1085940100a4SChris Mason return 0; 1086f46b5a66SChristoph Hellwig 10874cb5300bSChris Mason /* 10884cb5300bSChris Mason * if we were not given a file, allocate a readahead 10894cb5300bSChris Mason * context 10904cb5300bSChris Mason */ 10914cb5300bSChris Mason if (!file) { 10924cb5300bSChris Mason ra = kzalloc(sizeof(*ra), GFP_NOFS); 10934cb5300bSChris Mason if (!ra) 10944cb5300bSChris Mason return -ENOMEM; 10954cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 10964cb5300bSChris Mason } else { 10974cb5300bSChris Mason ra = &file->f_ra; 10984cb5300bSChris Mason } 10994cb5300bSChris Mason 1100008873eaSLi Zefan pages = kmalloc(sizeof(struct page *) * max_cluster, 11014cb5300bSChris Mason GFP_NOFS); 11024cb5300bSChris Mason if (!pages) { 11034cb5300bSChris Mason ret = -ENOMEM; 11044cb5300bSChris Mason goto out_ra; 11054cb5300bSChris Mason } 11064cb5300bSChris Mason 11074cb5300bSChris Mason /* find the last page to defrag */ 11081e701a32SChris Mason if (range->start + range->len > range->start) { 1109151a31b2SLi Zefan last_index = min_t(u64, isize - 1, 11101e701a32SChris Mason range->start + range->len - 1) >> PAGE_CACHE_SHIFT; 11111e701a32SChris Mason } else { 1112151a31b2SLi Zefan last_index = (isize - 1) >> PAGE_CACHE_SHIFT; 11131e701a32SChris Mason } 11141e701a32SChris Mason 11154cb5300bSChris Mason if (newer_than) { 11164cb5300bSChris Mason ret = find_new_extents(root, inode, newer_than, 11174cb5300bSChris Mason &newer_off, 64 * 1024); 11184cb5300bSChris Mason if (!ret) { 11194cb5300bSChris Mason range->start = newer_off; 11204cb5300bSChris Mason /* 11214cb5300bSChris Mason * we always align our defrag to help keep 11224cb5300bSChris Mason * the extents in the file evenly spaced 11234cb5300bSChris Mason */ 11244cb5300bSChris Mason i = (newer_off & new_align) >> PAGE_CACHE_SHIFT; 11254cb5300bSChris Mason } else 11264cb5300bSChris Mason goto out_ra; 11274cb5300bSChris Mason } else { 11281e701a32SChris Mason i = range->start >> PAGE_CACHE_SHIFT; 11294cb5300bSChris Mason } 11304cb5300bSChris Mason if (!max_to_defrag) 11317ec31b54SLiu Bo max_to_defrag = last_index + 1; 11324cb5300bSChris Mason 11332a0f7f57SLi Zefan /* 11342a0f7f57SLi Zefan * make writeback starts from i, so the defrag range can be 11352a0f7f57SLi Zefan * written sequentially. 11362a0f7f57SLi Zefan */ 11372a0f7f57SLi Zefan if (i < inode->i_mapping->writeback_index) 11382a0f7f57SLi Zefan inode->i_mapping->writeback_index = i; 11392a0f7f57SLi Zefan 1140f7f43cc8SChris Mason while (i <= last_index && defrag_count < max_to_defrag && 1141f7f43cc8SChris Mason (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> 1142f7f43cc8SChris Mason PAGE_CACHE_SHIFT)) { 11434cb5300bSChris Mason /* 11444cb5300bSChris Mason * make sure we stop running if someone unmounts 11454cb5300bSChris Mason * the FS 11464cb5300bSChris Mason */ 11474cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 11484cb5300bSChris Mason break; 11494cb5300bSChris Mason 115066c26892SLiu Bo if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT, 11516c282eb4SLi Zefan extent_thresh, &last_len, &skip, 1152a43a2111SAndrew Mahone &defrag_end, range->flags & 1153a43a2111SAndrew Mahone BTRFS_DEFRAG_RANGE_COMPRESS)) { 1154940100a4SChris Mason unsigned long next; 1155940100a4SChris Mason /* 1156940100a4SChris Mason * the should_defrag function tells us how much to skip 1157940100a4SChris Mason * bump our counter by the suggested amount 1158940100a4SChris Mason */ 1159940100a4SChris Mason next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1160940100a4SChris Mason i = max(i + 1, next); 1161940100a4SChris Mason continue; 1162940100a4SChris Mason } 1163008873eaSLi Zefan 1164008873eaSLi Zefan if (!newer_than) { 1165008873eaSLi Zefan cluster = (PAGE_CACHE_ALIGN(defrag_end) >> 1166008873eaSLi Zefan PAGE_CACHE_SHIFT) - i; 1167008873eaSLi Zefan cluster = min(cluster, max_cluster); 1168008873eaSLi Zefan } else { 1169008873eaSLi Zefan cluster = max_cluster; 1170008873eaSLi Zefan } 1171008873eaSLi Zefan 11721e701a32SChris Mason if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) 11731a419d85SLi Zefan BTRFS_I(inode)->force_compress = compress_type; 1174940100a4SChris Mason 1175008873eaSLi Zefan if (i + cluster > ra_index) { 1176008873eaSLi Zefan ra_index = max(i, ra_index); 1177008873eaSLi Zefan btrfs_force_ra(inode->i_mapping, ra, file, ra_index, 1178008873eaSLi Zefan cluster); 1179008873eaSLi Zefan ra_index += max_cluster; 1180008873eaSLi Zefan } 11814cb5300bSChris Mason 1182ecb8bea8SLiu Bo mutex_lock(&inode->i_mutex); 1183008873eaSLi Zefan ret = cluster_pages_for_defrag(inode, pages, i, cluster); 1184ecb8bea8SLiu Bo if (ret < 0) { 1185ecb8bea8SLiu Bo mutex_unlock(&inode->i_mutex); 11864cb5300bSChris Mason goto out_ra; 1187ecb8bea8SLiu Bo } 11884cb5300bSChris Mason 11894cb5300bSChris Mason defrag_count += ret; 11904cb5300bSChris Mason balance_dirty_pages_ratelimited_nr(inode->i_mapping, ret); 1191ecb8bea8SLiu Bo mutex_unlock(&inode->i_mutex); 11924cb5300bSChris Mason 11934cb5300bSChris Mason if (newer_than) { 11944cb5300bSChris Mason if (newer_off == (u64)-1) 11954cb5300bSChris Mason break; 11964cb5300bSChris Mason 1197e1f041e1SLiu Bo if (ret > 0) 1198e1f041e1SLiu Bo i += ret; 1199e1f041e1SLiu Bo 12004cb5300bSChris Mason newer_off = max(newer_off + 1, 12014cb5300bSChris Mason (u64)i << PAGE_CACHE_SHIFT); 12024cb5300bSChris Mason 12034cb5300bSChris Mason ret = find_new_extents(root, inode, 12044cb5300bSChris Mason newer_than, &newer_off, 12054cb5300bSChris Mason 64 * 1024); 12064cb5300bSChris Mason if (!ret) { 12074cb5300bSChris Mason range->start = newer_off; 12084cb5300bSChris Mason i = (newer_off & new_align) >> PAGE_CACHE_SHIFT; 12094cb5300bSChris Mason } else { 12104cb5300bSChris Mason break; 1211940100a4SChris Mason } 12124cb5300bSChris Mason } else { 1213008873eaSLi Zefan if (ret > 0) { 1214cbcc8326SLi Zefan i += ret; 1215008873eaSLi Zefan last_len += ret << PAGE_CACHE_SHIFT; 1216008873eaSLi Zefan } else { 1217940100a4SChris Mason i++; 1218008873eaSLi Zefan last_len = 0; 1219008873eaSLi Zefan } 1220f46b5a66SChristoph Hellwig } 12214cb5300bSChris Mason } 1222f46b5a66SChristoph Hellwig 12231e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) 12241e701a32SChris Mason filemap_flush(inode->i_mapping); 12251e701a32SChris Mason 12261e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 12271e701a32SChris Mason /* the filemap_flush will queue IO into the worker threads, but 12281e701a32SChris Mason * we have to make sure the IO is actually started and that 12291e701a32SChris Mason * ordered extents get created before we return 12301e701a32SChris Mason */ 12311e701a32SChris Mason atomic_inc(&root->fs_info->async_submit_draining); 12321e701a32SChris Mason while (atomic_read(&root->fs_info->nr_async_submits) || 12331e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages)) { 12341e701a32SChris Mason wait_event(root->fs_info->async_submit_wait, 12351e701a32SChris Mason (atomic_read(&root->fs_info->nr_async_submits) == 0 && 12361e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages) == 0)); 12371e701a32SChris Mason } 12381e701a32SChris Mason atomic_dec(&root->fs_info->async_submit_draining); 12391e701a32SChris Mason 12401e701a32SChris Mason mutex_lock(&inode->i_mutex); 1241261507a0SLi Zefan BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE; 12421e701a32SChris Mason mutex_unlock(&inode->i_mutex); 12431e701a32SChris Mason } 12441e701a32SChris Mason 12451a419d85SLi Zefan if (range->compress_type == BTRFS_COMPRESS_LZO) { 1246*2b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO); 12471a419d85SLi Zefan } 12481a419d85SLi Zefan 124960ccf82fSDiego Calleja ret = defrag_count; 1250940100a4SChris Mason 12514cb5300bSChris Mason out_ra: 12524cb5300bSChris Mason if (!file) 12534cb5300bSChris Mason kfree(ra); 12544cb5300bSChris Mason kfree(pages); 1255940100a4SChris Mason return ret; 1256f46b5a66SChristoph Hellwig } 1257f46b5a66SChristoph Hellwig 125876dda93cSYan, Zheng static noinline int btrfs_ioctl_resize(struct btrfs_root *root, 125976dda93cSYan, Zheng void __user *arg) 1260f46b5a66SChristoph Hellwig { 1261f46b5a66SChristoph Hellwig u64 new_size; 1262f46b5a66SChristoph Hellwig u64 old_size; 1263f46b5a66SChristoph Hellwig u64 devid = 1; 1264f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1265f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1266f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1267f46b5a66SChristoph Hellwig char *sizestr; 1268f46b5a66SChristoph Hellwig char *devstr = NULL; 1269f46b5a66SChristoph Hellwig int ret = 0; 1270f46b5a66SChristoph Hellwig int mod = 0; 1271f46b5a66SChristoph Hellwig 1272c146afadSYan Zheng if (root->fs_info->sb->s_flags & MS_RDONLY) 1273c146afadSYan Zheng return -EROFS; 1274c146afadSYan Zheng 1275e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1276e441d54dSChris Mason return -EPERM; 1277e441d54dSChris Mason 1278c9e9f97bSIlya Dryomov mutex_lock(&root->fs_info->volume_mutex); 1279c9e9f97bSIlya Dryomov if (root->fs_info->balance_ctl) { 1280c9e9f97bSIlya Dryomov printk(KERN_INFO "btrfs: balance in progress\n"); 1281c9e9f97bSIlya Dryomov ret = -EINVAL; 1282c9e9f97bSIlya Dryomov goto out; 1283c9e9f97bSIlya Dryomov } 1284c9e9f97bSIlya Dryomov 1285dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1286c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1287c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1288c9e9f97bSIlya Dryomov goto out; 1289c9e9f97bSIlya Dryomov } 12905516e595SMark Fasheh 12915516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1292f46b5a66SChristoph Hellwig 1293f46b5a66SChristoph Hellwig sizestr = vol_args->name; 1294f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 1295f46b5a66SChristoph Hellwig if (devstr) { 1296f46b5a66SChristoph Hellwig char *end; 1297f46b5a66SChristoph Hellwig sizestr = devstr + 1; 1298f46b5a66SChristoph Hellwig *devstr = '\0'; 1299f46b5a66SChristoph Hellwig devstr = vol_args->name; 1300f46b5a66SChristoph Hellwig devid = simple_strtoull(devstr, &end, 10); 13015bb14682SArnd Hannemann printk(KERN_INFO "btrfs: resizing devid %llu\n", 130221380931SJoel Becker (unsigned long long)devid); 1303f46b5a66SChristoph Hellwig } 13042b82032cSYan Zheng device = btrfs_find_device(root, devid, NULL, NULL); 1305f46b5a66SChristoph Hellwig if (!device) { 13065bb14682SArnd Hannemann printk(KERN_INFO "btrfs: resizer unable to find device %llu\n", 130721380931SJoel Becker (unsigned long long)devid); 1308f46b5a66SChristoph Hellwig ret = -EINVAL; 1309c9e9f97bSIlya Dryomov goto out_free; 1310f46b5a66SChristoph Hellwig } 13114e42ae1bSLiu Bo if (device->fs_devices && device->fs_devices->seeding) { 13124e42ae1bSLiu Bo printk(KERN_INFO "btrfs: resizer unable to apply on " 1313a8c4a33bSChris Mason "seeding device %llu\n", 1314a8c4a33bSChris Mason (unsigned long long)devid); 13154e42ae1bSLiu Bo ret = -EINVAL; 13164e42ae1bSLiu Bo goto out_free; 13174e42ae1bSLiu Bo } 13184e42ae1bSLiu Bo 1319f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 1320f46b5a66SChristoph Hellwig new_size = device->bdev->bd_inode->i_size; 1321f46b5a66SChristoph Hellwig else { 1322f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 1323f46b5a66SChristoph Hellwig mod = -1; 1324f46b5a66SChristoph Hellwig sizestr++; 1325f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 1326f46b5a66SChristoph Hellwig mod = 1; 1327f46b5a66SChristoph Hellwig sizestr++; 1328f46b5a66SChristoph Hellwig } 132991748467SAkinobu Mita new_size = memparse(sizestr, NULL); 1330f46b5a66SChristoph Hellwig if (new_size == 0) { 1331f46b5a66SChristoph Hellwig ret = -EINVAL; 1332c9e9f97bSIlya Dryomov goto out_free; 1333f46b5a66SChristoph Hellwig } 1334f46b5a66SChristoph Hellwig } 1335f46b5a66SChristoph Hellwig 1336f46b5a66SChristoph Hellwig old_size = device->total_bytes; 1337f46b5a66SChristoph Hellwig 1338f46b5a66SChristoph Hellwig if (mod < 0) { 1339f46b5a66SChristoph Hellwig if (new_size > old_size) { 1340f46b5a66SChristoph Hellwig ret = -EINVAL; 1341c9e9f97bSIlya Dryomov goto out_free; 1342f46b5a66SChristoph Hellwig } 1343f46b5a66SChristoph Hellwig new_size = old_size - new_size; 1344f46b5a66SChristoph Hellwig } else if (mod > 0) { 1345f46b5a66SChristoph Hellwig new_size = old_size + new_size; 1346f46b5a66SChristoph Hellwig } 1347f46b5a66SChristoph Hellwig 1348f46b5a66SChristoph Hellwig if (new_size < 256 * 1024 * 1024) { 1349f46b5a66SChristoph Hellwig ret = -EINVAL; 1350c9e9f97bSIlya Dryomov goto out_free; 1351f46b5a66SChristoph Hellwig } 1352f46b5a66SChristoph Hellwig if (new_size > device->bdev->bd_inode->i_size) { 1353f46b5a66SChristoph Hellwig ret = -EFBIG; 1354c9e9f97bSIlya Dryomov goto out_free; 1355f46b5a66SChristoph Hellwig } 1356f46b5a66SChristoph Hellwig 1357f46b5a66SChristoph Hellwig do_div(new_size, root->sectorsize); 1358f46b5a66SChristoph Hellwig new_size *= root->sectorsize; 1359f46b5a66SChristoph Hellwig 1360606686eeSJosef Bacik printk_in_rcu(KERN_INFO "btrfs: new size for %s is %llu\n", 1361606686eeSJosef Bacik rcu_str_deref(device->name), 1362606686eeSJosef Bacik (unsigned long long)new_size); 1363f46b5a66SChristoph Hellwig 1364f46b5a66SChristoph Hellwig if (new_size > old_size) { 1365a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 136698d5dc13STsutomu Itoh if (IS_ERR(trans)) { 136798d5dc13STsutomu Itoh ret = PTR_ERR(trans); 1368c9e9f97bSIlya Dryomov goto out_free; 136998d5dc13STsutomu Itoh } 1370f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 1371f46b5a66SChristoph Hellwig btrfs_commit_transaction(trans, root); 1372ece7d20eSMike Fleetwood } else if (new_size < old_size) { 1373f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 1374f46b5a66SChristoph Hellwig } 1375f46b5a66SChristoph Hellwig 1376c9e9f97bSIlya Dryomov out_free: 1377f46b5a66SChristoph Hellwig kfree(vol_args); 1378c9e9f97bSIlya Dryomov out: 1379c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 1380f46b5a66SChristoph Hellwig return ret; 1381f46b5a66SChristoph Hellwig } 1382f46b5a66SChristoph Hellwig 138372fd032eSSage Weil static noinline int btrfs_ioctl_snap_create_transid(struct file *file, 13846f72c7e2SArne Jansen char *name, unsigned long fd, int subvol, 13856f72c7e2SArne Jansen u64 *transid, bool readonly, 13866f72c7e2SArne Jansen struct btrfs_qgroup_inherit **inherit) 1387f46b5a66SChristoph Hellwig { 13883de4586cSChris Mason struct file *src_file; 1389f46b5a66SChristoph Hellwig int namelen; 13903de4586cSChris Mason int ret = 0; 1391f46b5a66SChristoph Hellwig 1392a874a63eSLiu Bo ret = mnt_want_write_file(file); 1393a874a63eSLiu Bo if (ret) 1394a874a63eSLiu Bo goto out; 1395a874a63eSLiu Bo 139672fd032eSSage Weil namelen = strlen(name); 139772fd032eSSage Weil if (strchr(name, '/')) { 1398f46b5a66SChristoph Hellwig ret = -EINVAL; 1399a874a63eSLiu Bo goto out_drop_write; 1400f46b5a66SChristoph Hellwig } 1401f46b5a66SChristoph Hellwig 140216780cabSChris Mason if (name[0] == '.' && 140316780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 140416780cabSChris Mason ret = -EEXIST; 1405a874a63eSLiu Bo goto out_drop_write; 140616780cabSChris Mason } 140716780cabSChris Mason 14083de4586cSChris Mason if (subvol) { 140972fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 14106f72c7e2SArne Jansen NULL, transid, readonly, inherit); 1411cb8e7090SChristoph Hellwig } else { 14123de4586cSChris Mason struct inode *src_inode; 141372fd032eSSage Weil src_file = fget(fd); 14143de4586cSChris Mason if (!src_file) { 14153de4586cSChris Mason ret = -EINVAL; 1416a874a63eSLiu Bo goto out_drop_write; 14173de4586cSChris Mason } 14183de4586cSChris Mason 14193de4586cSChris Mason src_inode = src_file->f_path.dentry->d_inode; 14203de4586cSChris Mason if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) { 1421d397712bSChris Mason printk(KERN_INFO "btrfs: Snapshot src from " 1422d397712bSChris Mason "another FS\n"); 14233de4586cSChris Mason ret = -EINVAL; 14243de4586cSChris Mason fput(src_file); 1425a874a63eSLiu Bo goto out_drop_write; 14263de4586cSChris Mason } 142772fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 142872fd032eSSage Weil BTRFS_I(src_inode)->root, 14296f72c7e2SArne Jansen transid, readonly, inherit); 14303de4586cSChris Mason fput(src_file); 1431cb8e7090SChristoph Hellwig } 1432a874a63eSLiu Bo out_drop_write: 1433a874a63eSLiu Bo mnt_drop_write_file(file); 1434f46b5a66SChristoph Hellwig out: 143572fd032eSSage Weil return ret; 143672fd032eSSage Weil } 143772fd032eSSage Weil 143872fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 1439fa0d2b9bSLi Zefan void __user *arg, int subvol) 144072fd032eSSage Weil { 1441fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 144272fd032eSSage Weil int ret; 144372fd032eSSage Weil 1444fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1445fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1446fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1447fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1448fa0d2b9bSLi Zefan 1449fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1450b83cc969SLi Zefan vol_args->fd, subvol, 14516f72c7e2SArne Jansen NULL, false, NULL); 1452fa0d2b9bSLi Zefan 1453fa0d2b9bSLi Zefan kfree(vol_args); 1454fa0d2b9bSLi Zefan return ret; 1455fa0d2b9bSLi Zefan } 1456fa0d2b9bSLi Zefan 1457fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1458fa0d2b9bSLi Zefan void __user *arg, int subvol) 1459fa0d2b9bSLi Zefan { 1460fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 1461fa0d2b9bSLi Zefan int ret; 1462fdfb1e4fSLi Zefan u64 transid = 0; 1463fdfb1e4fSLi Zefan u64 *ptr = NULL; 1464b83cc969SLi Zefan bool readonly = false; 14656f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 146672fd032eSSage Weil 1467fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1468fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1469fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1470fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1471fdfb1e4fSLi Zefan 1472b83cc969SLi Zefan if (vol_args->flags & 14736f72c7e2SArne Jansen ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY | 14746f72c7e2SArne Jansen BTRFS_SUBVOL_QGROUP_INHERIT)) { 1475b83cc969SLi Zefan ret = -EOPNOTSUPP; 1476fdfb1e4fSLi Zefan goto out; 1477fdfb1e4fSLi Zefan } 1478fdfb1e4fSLi Zefan 1479fa0d2b9bSLi Zefan if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) 1480fdfb1e4fSLi Zefan ptr = &transid; 1481b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1482b83cc969SLi Zefan readonly = true; 14836f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 14846f72c7e2SArne Jansen if (vol_args->size > PAGE_CACHE_SIZE) { 14856f72c7e2SArne Jansen ret = -EINVAL; 14866f72c7e2SArne Jansen goto out; 14876f72c7e2SArne Jansen } 14886f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 14896f72c7e2SArne Jansen if (IS_ERR(inherit)) { 14906f72c7e2SArne Jansen ret = PTR_ERR(inherit); 14916f72c7e2SArne Jansen goto out; 14926f72c7e2SArne Jansen } 14936f72c7e2SArne Jansen } 149475eaa0e2SSage Weil 1495fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 14966f72c7e2SArne Jansen vol_args->fd, subvol, ptr, 14976f72c7e2SArne Jansen readonly, &inherit); 149875eaa0e2SSage Weil 1499fdfb1e4fSLi Zefan if (ret == 0 && ptr && 150075eaa0e2SSage Weil copy_to_user(arg + 1501fdfb1e4fSLi Zefan offsetof(struct btrfs_ioctl_vol_args_v2, 1502fdfb1e4fSLi Zefan transid), ptr, sizeof(*ptr))) 150375eaa0e2SSage Weil ret = -EFAULT; 1504fdfb1e4fSLi Zefan out: 1505f46b5a66SChristoph Hellwig kfree(vol_args); 15066f72c7e2SArne Jansen kfree(inherit); 1507f46b5a66SChristoph Hellwig return ret; 1508f46b5a66SChristoph Hellwig } 1509f46b5a66SChristoph Hellwig 15100caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 15110caa102dSLi Zefan void __user *arg) 15120caa102dSLi Zefan { 15130caa102dSLi Zefan struct inode *inode = fdentry(file)->d_inode; 15140caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 15150caa102dSLi Zefan int ret = 0; 15160caa102dSLi Zefan u64 flags = 0; 15170caa102dSLi Zefan 151833345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) 15190caa102dSLi Zefan return -EINVAL; 15200caa102dSLi Zefan 15210caa102dSLi Zefan down_read(&root->fs_info->subvol_sem); 15220caa102dSLi Zefan if (btrfs_root_readonly(root)) 15230caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 15240caa102dSLi Zefan up_read(&root->fs_info->subvol_sem); 15250caa102dSLi Zefan 15260caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 15270caa102dSLi Zefan ret = -EFAULT; 15280caa102dSLi Zefan 15290caa102dSLi Zefan return ret; 15300caa102dSLi Zefan } 15310caa102dSLi Zefan 15320caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 15330caa102dSLi Zefan void __user *arg) 15340caa102dSLi Zefan { 15350caa102dSLi Zefan struct inode *inode = fdentry(file)->d_inode; 15360caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 15370caa102dSLi Zefan struct btrfs_trans_handle *trans; 15380caa102dSLi Zefan u64 root_flags; 15390caa102dSLi Zefan u64 flags; 15400caa102dSLi Zefan int ret = 0; 15410caa102dSLi Zefan 1542b9ca0664SLiu Bo ret = mnt_want_write_file(file); 1543b9ca0664SLiu Bo if (ret) 1544b9ca0664SLiu Bo goto out; 15450caa102dSLi Zefan 1546b9ca0664SLiu Bo if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 1547b9ca0664SLiu Bo ret = -EINVAL; 1548b9ca0664SLiu Bo goto out_drop_write; 1549b9ca0664SLiu Bo } 15500caa102dSLi Zefan 1551b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 1552b9ca0664SLiu Bo ret = -EFAULT; 1553b9ca0664SLiu Bo goto out_drop_write; 1554b9ca0664SLiu Bo } 15550caa102dSLi Zefan 1556b9ca0664SLiu Bo if (flags & BTRFS_SUBVOL_CREATE_ASYNC) { 1557b9ca0664SLiu Bo ret = -EINVAL; 1558b9ca0664SLiu Bo goto out_drop_write; 1559b9ca0664SLiu Bo } 15600caa102dSLi Zefan 1561b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 1562b9ca0664SLiu Bo ret = -EOPNOTSUPP; 1563b9ca0664SLiu Bo goto out_drop_write; 1564b9ca0664SLiu Bo } 15650caa102dSLi Zefan 1566b9ca0664SLiu Bo if (!inode_owner_or_capable(inode)) { 1567b9ca0664SLiu Bo ret = -EACCES; 1568b9ca0664SLiu Bo goto out_drop_write; 1569b9ca0664SLiu Bo } 1570b4dc2b8cSLi Zefan 15710caa102dSLi Zefan down_write(&root->fs_info->subvol_sem); 15720caa102dSLi Zefan 15730caa102dSLi Zefan /* nothing to do */ 15740caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 1575b9ca0664SLiu Bo goto out_drop_sem; 15760caa102dSLi Zefan 15770caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 15780caa102dSLi Zefan if (flags & BTRFS_SUBVOL_RDONLY) 15790caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 15800caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 15810caa102dSLi Zefan else 15820caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 15830caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 15840caa102dSLi Zefan 15850caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 15860caa102dSLi Zefan if (IS_ERR(trans)) { 15870caa102dSLi Zefan ret = PTR_ERR(trans); 15880caa102dSLi Zefan goto out_reset; 15890caa102dSLi Zefan } 15900caa102dSLi Zefan 1591b4dc2b8cSLi Zefan ret = btrfs_update_root(trans, root->fs_info->tree_root, 15920caa102dSLi Zefan &root->root_key, &root->root_item); 15930caa102dSLi Zefan 15940caa102dSLi Zefan btrfs_commit_transaction(trans, root); 15950caa102dSLi Zefan out_reset: 15960caa102dSLi Zefan if (ret) 15970caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 1598b9ca0664SLiu Bo out_drop_sem: 15990caa102dSLi Zefan up_write(&root->fs_info->subvol_sem); 1600b9ca0664SLiu Bo out_drop_write: 1601b9ca0664SLiu Bo mnt_drop_write_file(file); 1602b9ca0664SLiu Bo out: 16030caa102dSLi Zefan return ret; 16040caa102dSLi Zefan } 16050caa102dSLi Zefan 160676dda93cSYan, Zheng /* 160776dda93cSYan, Zheng * helper to check if the subvolume references other subvolumes 160876dda93cSYan, Zheng */ 160976dda93cSYan, Zheng static noinline int may_destroy_subvol(struct btrfs_root *root) 161076dda93cSYan, Zheng { 161176dda93cSYan, Zheng struct btrfs_path *path; 161276dda93cSYan, Zheng struct btrfs_key key; 161376dda93cSYan, Zheng int ret; 161476dda93cSYan, Zheng 161576dda93cSYan, Zheng path = btrfs_alloc_path(); 161676dda93cSYan, Zheng if (!path) 161776dda93cSYan, Zheng return -ENOMEM; 161876dda93cSYan, Zheng 161976dda93cSYan, Zheng key.objectid = root->root_key.objectid; 162076dda93cSYan, Zheng key.type = BTRFS_ROOT_REF_KEY; 162176dda93cSYan, Zheng key.offset = (u64)-1; 162276dda93cSYan, Zheng 162376dda93cSYan, Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, 162476dda93cSYan, Zheng &key, path, 0, 0); 162576dda93cSYan, Zheng if (ret < 0) 162676dda93cSYan, Zheng goto out; 162776dda93cSYan, Zheng BUG_ON(ret == 0); 162876dda93cSYan, Zheng 162976dda93cSYan, Zheng ret = 0; 163076dda93cSYan, Zheng if (path->slots[0] > 0) { 163176dda93cSYan, Zheng path->slots[0]--; 163276dda93cSYan, Zheng btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 163376dda93cSYan, Zheng if (key.objectid == root->root_key.objectid && 163476dda93cSYan, Zheng key.type == BTRFS_ROOT_REF_KEY) 163576dda93cSYan, Zheng ret = -ENOTEMPTY; 163676dda93cSYan, Zheng } 163776dda93cSYan, Zheng out: 163876dda93cSYan, Zheng btrfs_free_path(path); 163976dda93cSYan, Zheng return ret; 164076dda93cSYan, Zheng } 164176dda93cSYan, Zheng 1642ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 1643ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 1644ac8e9819SChris Mason { 1645abc6e134SChris Mason struct btrfs_key test; 1646abc6e134SChris Mason int ret; 1647abc6e134SChris Mason 1648abc6e134SChris Mason test.objectid = sk->min_objectid; 1649abc6e134SChris Mason test.type = sk->min_type; 1650abc6e134SChris Mason test.offset = sk->min_offset; 1651abc6e134SChris Mason 1652abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1653abc6e134SChris Mason if (ret < 0) 1654ac8e9819SChris Mason return 0; 1655abc6e134SChris Mason 1656abc6e134SChris Mason test.objectid = sk->max_objectid; 1657abc6e134SChris Mason test.type = sk->max_type; 1658abc6e134SChris Mason test.offset = sk->max_offset; 1659abc6e134SChris Mason 1660abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1661abc6e134SChris Mason if (ret > 0) 1662ac8e9819SChris Mason return 0; 1663ac8e9819SChris Mason return 1; 1664ac8e9819SChris Mason } 1665ac8e9819SChris Mason 1666ac8e9819SChris Mason static noinline int copy_to_sk(struct btrfs_root *root, 1667ac8e9819SChris Mason struct btrfs_path *path, 1668ac8e9819SChris Mason struct btrfs_key *key, 1669ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 1670ac8e9819SChris Mason char *buf, 1671ac8e9819SChris Mason unsigned long *sk_offset, 1672ac8e9819SChris Mason int *num_found) 1673ac8e9819SChris Mason { 1674ac8e9819SChris Mason u64 found_transid; 1675ac8e9819SChris Mason struct extent_buffer *leaf; 1676ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 1677ac8e9819SChris Mason unsigned long item_off; 1678ac8e9819SChris Mason unsigned long item_len; 1679ac8e9819SChris Mason int nritems; 1680ac8e9819SChris Mason int i; 1681ac8e9819SChris Mason int slot; 1682ac8e9819SChris Mason int ret = 0; 1683ac8e9819SChris Mason 1684ac8e9819SChris Mason leaf = path->nodes[0]; 1685ac8e9819SChris Mason slot = path->slots[0]; 1686ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 1687ac8e9819SChris Mason 1688ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 1689ac8e9819SChris Mason i = nritems; 1690ac8e9819SChris Mason goto advance_key; 1691ac8e9819SChris Mason } 1692ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 1693ac8e9819SChris Mason 1694ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 1695ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 1696ac8e9819SChris Mason item_len = btrfs_item_size_nr(leaf, i); 1697ac8e9819SChris Mason 1698ac8e9819SChris Mason if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE) 1699ac8e9819SChris Mason item_len = 0; 1700ac8e9819SChris Mason 1701ac8e9819SChris Mason if (sizeof(sh) + item_len + *sk_offset > 1702ac8e9819SChris Mason BTRFS_SEARCH_ARGS_BUFSIZE) { 1703ac8e9819SChris Mason ret = 1; 1704ac8e9819SChris Mason goto overflow; 1705ac8e9819SChris Mason } 1706ac8e9819SChris Mason 1707ac8e9819SChris Mason btrfs_item_key_to_cpu(leaf, key, i); 1708ac8e9819SChris Mason if (!key_in_sk(key, sk)) 1709ac8e9819SChris Mason continue; 1710ac8e9819SChris Mason 1711ac8e9819SChris Mason sh.objectid = key->objectid; 1712ac8e9819SChris Mason sh.offset = key->offset; 1713ac8e9819SChris Mason sh.type = key->type; 1714ac8e9819SChris Mason sh.len = item_len; 1715ac8e9819SChris Mason sh.transid = found_transid; 1716ac8e9819SChris Mason 1717ac8e9819SChris Mason /* copy search result header */ 1718ac8e9819SChris Mason memcpy(buf + *sk_offset, &sh, sizeof(sh)); 1719ac8e9819SChris Mason *sk_offset += sizeof(sh); 1720ac8e9819SChris Mason 1721ac8e9819SChris Mason if (item_len) { 1722ac8e9819SChris Mason char *p = buf + *sk_offset; 1723ac8e9819SChris Mason /* copy the item */ 1724ac8e9819SChris Mason read_extent_buffer(leaf, p, 1725ac8e9819SChris Mason item_off, item_len); 1726ac8e9819SChris Mason *sk_offset += item_len; 1727ac8e9819SChris Mason } 1728e2156867SHugo Mills (*num_found)++; 1729ac8e9819SChris Mason 1730ac8e9819SChris Mason if (*num_found >= sk->nr_items) 1731ac8e9819SChris Mason break; 1732ac8e9819SChris Mason } 1733ac8e9819SChris Mason advance_key: 1734ac8e9819SChris Mason ret = 0; 1735abc6e134SChris Mason if (key->offset < (u64)-1 && key->offset < sk->max_offset) 1736abc6e134SChris Mason key->offset++; 1737abc6e134SChris Mason else if (key->type < (u8)-1 && key->type < sk->max_type) { 1738abc6e134SChris Mason key->offset = 0; 1739abc6e134SChris Mason key->type++; 1740abc6e134SChris Mason } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) { 1741abc6e134SChris Mason key->offset = 0; 1742abc6e134SChris Mason key->type = 0; 1743abc6e134SChris Mason key->objectid++; 1744abc6e134SChris Mason } else 1745abc6e134SChris Mason ret = 1; 1746ac8e9819SChris Mason overflow: 1747ac8e9819SChris Mason return ret; 1748ac8e9819SChris Mason } 1749ac8e9819SChris Mason 1750ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 1751ac8e9819SChris Mason struct btrfs_ioctl_search_args *args) 1752ac8e9819SChris Mason { 1753ac8e9819SChris Mason struct btrfs_root *root; 1754ac8e9819SChris Mason struct btrfs_key key; 1755ac8e9819SChris Mason struct btrfs_key max_key; 1756ac8e9819SChris Mason struct btrfs_path *path; 1757ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk = &args->key; 1758ac8e9819SChris Mason struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info; 1759ac8e9819SChris Mason int ret; 1760ac8e9819SChris Mason int num_found = 0; 1761ac8e9819SChris Mason unsigned long sk_offset = 0; 1762ac8e9819SChris Mason 1763ac8e9819SChris Mason path = btrfs_alloc_path(); 1764ac8e9819SChris Mason if (!path) 1765ac8e9819SChris Mason return -ENOMEM; 1766ac8e9819SChris Mason 1767ac8e9819SChris Mason if (sk->tree_id == 0) { 1768ac8e9819SChris Mason /* search the root of the inode that was passed */ 1769ac8e9819SChris Mason root = BTRFS_I(inode)->root; 1770ac8e9819SChris Mason } else { 1771ac8e9819SChris Mason key.objectid = sk->tree_id; 1772ac8e9819SChris Mason key.type = BTRFS_ROOT_ITEM_KEY; 1773ac8e9819SChris Mason key.offset = (u64)-1; 1774ac8e9819SChris Mason root = btrfs_read_fs_root_no_name(info, &key); 1775ac8e9819SChris Mason if (IS_ERR(root)) { 1776ac8e9819SChris Mason printk(KERN_ERR "could not find root %llu\n", 1777ac8e9819SChris Mason sk->tree_id); 1778ac8e9819SChris Mason btrfs_free_path(path); 1779ac8e9819SChris Mason return -ENOENT; 1780ac8e9819SChris Mason } 1781ac8e9819SChris Mason } 1782ac8e9819SChris Mason 1783ac8e9819SChris Mason key.objectid = sk->min_objectid; 1784ac8e9819SChris Mason key.type = sk->min_type; 1785ac8e9819SChris Mason key.offset = sk->min_offset; 1786ac8e9819SChris Mason 1787ac8e9819SChris Mason max_key.objectid = sk->max_objectid; 1788ac8e9819SChris Mason max_key.type = sk->max_type; 1789ac8e9819SChris Mason max_key.offset = sk->max_offset; 1790ac8e9819SChris Mason 1791ac8e9819SChris Mason path->keep_locks = 1; 1792ac8e9819SChris Mason 1793ac8e9819SChris Mason while(1) { 1794ac8e9819SChris Mason ret = btrfs_search_forward(root, &key, &max_key, path, 0, 1795ac8e9819SChris Mason sk->min_transid); 1796ac8e9819SChris Mason if (ret != 0) { 1797ac8e9819SChris Mason if (ret > 0) 1798ac8e9819SChris Mason ret = 0; 1799ac8e9819SChris Mason goto err; 1800ac8e9819SChris Mason } 1801ac8e9819SChris Mason ret = copy_to_sk(root, path, &key, sk, args->buf, 1802ac8e9819SChris Mason &sk_offset, &num_found); 1803b3b4aa74SDavid Sterba btrfs_release_path(path); 1804ac8e9819SChris Mason if (ret || num_found >= sk->nr_items) 1805ac8e9819SChris Mason break; 1806ac8e9819SChris Mason 1807ac8e9819SChris Mason } 1808ac8e9819SChris Mason ret = 0; 1809ac8e9819SChris Mason err: 1810ac8e9819SChris Mason sk->nr_items = num_found; 1811ac8e9819SChris Mason btrfs_free_path(path); 1812ac8e9819SChris Mason return ret; 1813ac8e9819SChris Mason } 1814ac8e9819SChris Mason 1815ac8e9819SChris Mason static noinline int btrfs_ioctl_tree_search(struct file *file, 1816ac8e9819SChris Mason void __user *argp) 1817ac8e9819SChris Mason { 1818ac8e9819SChris Mason struct btrfs_ioctl_search_args *args; 1819ac8e9819SChris Mason struct inode *inode; 1820ac8e9819SChris Mason int ret; 1821ac8e9819SChris Mason 1822ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 1823ac8e9819SChris Mason return -EPERM; 1824ac8e9819SChris Mason 18252354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 18262354d08fSJulia Lawall if (IS_ERR(args)) 18272354d08fSJulia Lawall return PTR_ERR(args); 1828ac8e9819SChris Mason 1829ac8e9819SChris Mason inode = fdentry(file)->d_inode; 1830ac8e9819SChris Mason ret = search_ioctl(inode, args); 1831ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 1832ac8e9819SChris Mason ret = -EFAULT; 1833ac8e9819SChris Mason kfree(args); 1834ac8e9819SChris Mason return ret; 1835ac8e9819SChris Mason } 1836ac8e9819SChris Mason 183798d377a0STARUISI Hiroaki /* 1838ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 1839ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 184098d377a0STARUISI Hiroaki */ 184198d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 184298d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 184398d377a0STARUISI Hiroaki { 184498d377a0STARUISI Hiroaki struct btrfs_root *root; 184598d377a0STARUISI Hiroaki struct btrfs_key key; 1846ac8e9819SChris Mason char *ptr; 184798d377a0STARUISI Hiroaki int ret = -1; 184898d377a0STARUISI Hiroaki int slot; 184998d377a0STARUISI Hiroaki int len; 185098d377a0STARUISI Hiroaki int total_len = 0; 185198d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 185298d377a0STARUISI Hiroaki struct extent_buffer *l; 185398d377a0STARUISI Hiroaki struct btrfs_path *path; 185498d377a0STARUISI Hiroaki 185598d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 185698d377a0STARUISI Hiroaki name[0]='\0'; 185798d377a0STARUISI Hiroaki return 0; 185898d377a0STARUISI Hiroaki } 185998d377a0STARUISI Hiroaki 186098d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 186198d377a0STARUISI Hiroaki if (!path) 186298d377a0STARUISI Hiroaki return -ENOMEM; 186398d377a0STARUISI Hiroaki 1864ac8e9819SChris Mason ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX]; 186598d377a0STARUISI Hiroaki 186698d377a0STARUISI Hiroaki key.objectid = tree_id; 186798d377a0STARUISI Hiroaki key.type = BTRFS_ROOT_ITEM_KEY; 186898d377a0STARUISI Hiroaki key.offset = (u64)-1; 186998d377a0STARUISI Hiroaki root = btrfs_read_fs_root_no_name(info, &key); 187098d377a0STARUISI Hiroaki if (IS_ERR(root)) { 187198d377a0STARUISI Hiroaki printk(KERN_ERR "could not find root %llu\n", tree_id); 18728ad6fcabSChris Mason ret = -ENOENT; 18738ad6fcabSChris Mason goto out; 187498d377a0STARUISI Hiroaki } 187598d377a0STARUISI Hiroaki 187698d377a0STARUISI Hiroaki key.objectid = dirid; 187798d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 18788ad6fcabSChris Mason key.offset = (u64)-1; 187998d377a0STARUISI Hiroaki 188098d377a0STARUISI Hiroaki while(1) { 188198d377a0STARUISI Hiroaki ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 188298d377a0STARUISI Hiroaki if (ret < 0) 188398d377a0STARUISI Hiroaki goto out; 188498d377a0STARUISI Hiroaki 188598d377a0STARUISI Hiroaki l = path->nodes[0]; 188698d377a0STARUISI Hiroaki slot = path->slots[0]; 18878ad6fcabSChris Mason if (ret > 0 && slot > 0) 18888ad6fcabSChris Mason slot--; 188998d377a0STARUISI Hiroaki btrfs_item_key_to_cpu(l, &key, slot); 189098d377a0STARUISI Hiroaki 189198d377a0STARUISI Hiroaki if (ret > 0 && (key.objectid != dirid || 1892ac8e9819SChris Mason key.type != BTRFS_INODE_REF_KEY)) { 1893ac8e9819SChris Mason ret = -ENOENT; 189498d377a0STARUISI Hiroaki goto out; 1895ac8e9819SChris Mason } 189698d377a0STARUISI Hiroaki 189798d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 189898d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 189998d377a0STARUISI Hiroaki ptr -= len + 1; 190098d377a0STARUISI Hiroaki total_len += len + 1; 1901ac8e9819SChris Mason if (ptr < name) 190298d377a0STARUISI Hiroaki goto out; 190398d377a0STARUISI Hiroaki 190498d377a0STARUISI Hiroaki *(ptr + len) = '/'; 190598d377a0STARUISI Hiroaki read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len); 190698d377a0STARUISI Hiroaki 190798d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 190898d377a0STARUISI Hiroaki break; 190998d377a0STARUISI Hiroaki 1910b3b4aa74SDavid Sterba btrfs_release_path(path); 191198d377a0STARUISI Hiroaki key.objectid = key.offset; 19128ad6fcabSChris Mason key.offset = (u64)-1; 191398d377a0STARUISI Hiroaki dirid = key.objectid; 191498d377a0STARUISI Hiroaki } 1915ac8e9819SChris Mason if (ptr < name) 191698d377a0STARUISI Hiroaki goto out; 191777906a50SLi Zefan memmove(name, ptr, total_len); 191898d377a0STARUISI Hiroaki name[total_len]='\0'; 191998d377a0STARUISI Hiroaki ret = 0; 192098d377a0STARUISI Hiroaki out: 192198d377a0STARUISI Hiroaki btrfs_free_path(path); 1922ac8e9819SChris Mason return ret; 1923ac8e9819SChris Mason } 1924ac8e9819SChris Mason 1925ac8e9819SChris Mason static noinline int btrfs_ioctl_ino_lookup(struct file *file, 1926ac8e9819SChris Mason void __user *argp) 1927ac8e9819SChris Mason { 1928ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 1929ac8e9819SChris Mason struct inode *inode; 1930ac8e9819SChris Mason int ret; 1931ac8e9819SChris Mason 1932ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 1933ac8e9819SChris Mason return -EPERM; 1934ac8e9819SChris Mason 19352354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 19362354d08fSJulia Lawall if (IS_ERR(args)) 19372354d08fSJulia Lawall return PTR_ERR(args); 1938c2b96929SDan Carpenter 1939ac8e9819SChris Mason inode = fdentry(file)->d_inode; 1940ac8e9819SChris Mason 19411b53ac4dSChris Mason if (args->treeid == 0) 19421b53ac4dSChris Mason args->treeid = BTRFS_I(inode)->root->root_key.objectid; 19431b53ac4dSChris Mason 1944ac8e9819SChris Mason ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 1945ac8e9819SChris Mason args->treeid, args->objectid, 1946ac8e9819SChris Mason args->name); 1947ac8e9819SChris Mason 1948ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 1949ac8e9819SChris Mason ret = -EFAULT; 1950ac8e9819SChris Mason 1951ac8e9819SChris Mason kfree(args); 195298d377a0STARUISI Hiroaki return ret; 195398d377a0STARUISI Hiroaki } 195498d377a0STARUISI Hiroaki 195576dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 195676dda93cSYan, Zheng void __user *arg) 195776dda93cSYan, Zheng { 195876dda93cSYan, Zheng struct dentry *parent = fdentry(file); 195976dda93cSYan, Zheng struct dentry *dentry; 196076dda93cSYan, Zheng struct inode *dir = parent->d_inode; 196176dda93cSYan, Zheng struct inode *inode; 196276dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 196376dda93cSYan, Zheng struct btrfs_root *dest = NULL; 196476dda93cSYan, Zheng struct btrfs_ioctl_vol_args *vol_args; 196576dda93cSYan, Zheng struct btrfs_trans_handle *trans; 196676dda93cSYan, Zheng int namelen; 196776dda93cSYan, Zheng int ret; 196876dda93cSYan, Zheng int err = 0; 196976dda93cSYan, Zheng 197076dda93cSYan, Zheng vol_args = memdup_user(arg, sizeof(*vol_args)); 197176dda93cSYan, Zheng if (IS_ERR(vol_args)) 197276dda93cSYan, Zheng return PTR_ERR(vol_args); 197376dda93cSYan, Zheng 197476dda93cSYan, Zheng vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 197576dda93cSYan, Zheng namelen = strlen(vol_args->name); 197676dda93cSYan, Zheng if (strchr(vol_args->name, '/') || 197776dda93cSYan, Zheng strncmp(vol_args->name, "..", namelen) == 0) { 197876dda93cSYan, Zheng err = -EINVAL; 197976dda93cSYan, Zheng goto out; 198076dda93cSYan, Zheng } 198176dda93cSYan, Zheng 1982a561be71SAl Viro err = mnt_want_write_file(file); 198376dda93cSYan, Zheng if (err) 198476dda93cSYan, Zheng goto out; 198576dda93cSYan, Zheng 198676dda93cSYan, Zheng mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT); 198776dda93cSYan, Zheng dentry = lookup_one_len(vol_args->name, parent, namelen); 198876dda93cSYan, Zheng if (IS_ERR(dentry)) { 198976dda93cSYan, Zheng err = PTR_ERR(dentry); 199076dda93cSYan, Zheng goto out_unlock_dir; 199176dda93cSYan, Zheng } 199276dda93cSYan, Zheng 199376dda93cSYan, Zheng if (!dentry->d_inode) { 199476dda93cSYan, Zheng err = -ENOENT; 199576dda93cSYan, Zheng goto out_dput; 199676dda93cSYan, Zheng } 199776dda93cSYan, Zheng 199876dda93cSYan, Zheng inode = dentry->d_inode; 19994260f7c7SSage Weil dest = BTRFS_I(inode)->root; 20004260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)){ 20014260f7c7SSage Weil /* 20024260f7c7SSage Weil * Regular user. Only allow this with a special mount 20034260f7c7SSage Weil * option, when the user has write+exec access to the 20044260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 20054260f7c7SSage Weil * allowed. 20064260f7c7SSage Weil * 20074260f7c7SSage Weil * Note that this is _not_ check that the subvol is 20084260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 20094260f7c7SSage Weil * otherwise be able to delete. 20104260f7c7SSage Weil * 20114260f7c7SSage Weil * Users who want to delete empty subvols should try 20124260f7c7SSage Weil * rmdir(2). 20134260f7c7SSage Weil */ 20144260f7c7SSage Weil err = -EPERM; 20154260f7c7SSage Weil if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED)) 20164260f7c7SSage Weil goto out_dput; 20174260f7c7SSage Weil 20184260f7c7SSage Weil /* 20194260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 20204260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 20214260f7c7SSage Weil * must be called on the dentry referencing the root 20224260f7c7SSage Weil * of the subvol, not a random directory contained 20234260f7c7SSage Weil * within it. 20244260f7c7SSage Weil */ 20254260f7c7SSage Weil err = -EINVAL; 20264260f7c7SSage Weil if (root == dest) 20274260f7c7SSage Weil goto out_dput; 20284260f7c7SSage Weil 20294260f7c7SSage Weil err = inode_permission(inode, MAY_WRITE | MAY_EXEC); 20304260f7c7SSage Weil if (err) 20314260f7c7SSage Weil goto out_dput; 20324260f7c7SSage Weil 20334260f7c7SSage Weil /* check if subvolume may be deleted by a non-root user */ 20344260f7c7SSage Weil err = btrfs_may_delete(dir, dentry, 1); 20354260f7c7SSage Weil if (err) 20364260f7c7SSage Weil goto out_dput; 20374260f7c7SSage Weil } 20384260f7c7SSage Weil 203933345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 204076dda93cSYan, Zheng err = -EINVAL; 204176dda93cSYan, Zheng goto out_dput; 204276dda93cSYan, Zheng } 204376dda93cSYan, Zheng 204476dda93cSYan, Zheng mutex_lock(&inode->i_mutex); 204576dda93cSYan, Zheng err = d_invalidate(dentry); 204676dda93cSYan, Zheng if (err) 204776dda93cSYan, Zheng goto out_unlock; 204876dda93cSYan, Zheng 204976dda93cSYan, Zheng down_write(&root->fs_info->subvol_sem); 205076dda93cSYan, Zheng 205176dda93cSYan, Zheng err = may_destroy_subvol(dest); 205276dda93cSYan, Zheng if (err) 205376dda93cSYan, Zheng goto out_up_write; 205476dda93cSYan, Zheng 2055a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 2056a22285a6SYan, Zheng if (IS_ERR(trans)) { 2057a22285a6SYan, Zheng err = PTR_ERR(trans); 2058d327099aSDan Carpenter goto out_up_write; 2059a22285a6SYan, Zheng } 2060a22285a6SYan, Zheng trans->block_rsv = &root->fs_info->global_block_rsv; 2061a22285a6SYan, Zheng 206276dda93cSYan, Zheng ret = btrfs_unlink_subvol(trans, root, dir, 206376dda93cSYan, Zheng dest->root_key.objectid, 206476dda93cSYan, Zheng dentry->d_name.name, 206576dda93cSYan, Zheng dentry->d_name.len); 206679787eaaSJeff Mahoney if (ret) { 206779787eaaSJeff Mahoney err = ret; 206879787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 206979787eaaSJeff Mahoney goto out_end_trans; 207079787eaaSJeff Mahoney } 207176dda93cSYan, Zheng 207276dda93cSYan, Zheng btrfs_record_root_in_trans(trans, dest); 207376dda93cSYan, Zheng 207476dda93cSYan, Zheng memset(&dest->root_item.drop_progress, 0, 207576dda93cSYan, Zheng sizeof(dest->root_item.drop_progress)); 207676dda93cSYan, Zheng dest->root_item.drop_level = 0; 207776dda93cSYan, Zheng btrfs_set_root_refs(&dest->root_item, 0); 207876dda93cSYan, Zheng 2079d68fc57bSYan, Zheng if (!xchg(&dest->orphan_item_inserted, 1)) { 208076dda93cSYan, Zheng ret = btrfs_insert_orphan_item(trans, 208176dda93cSYan, Zheng root->fs_info->tree_root, 208276dda93cSYan, Zheng dest->root_key.objectid); 208379787eaaSJeff Mahoney if (ret) { 208479787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 208579787eaaSJeff Mahoney err = ret; 208679787eaaSJeff Mahoney goto out_end_trans; 2087d68fc57bSYan, Zheng } 208879787eaaSJeff Mahoney } 208979787eaaSJeff Mahoney out_end_trans: 2090531cb13fSSage Weil ret = btrfs_end_transaction(trans, root); 209179787eaaSJeff Mahoney if (ret && !err) 209279787eaaSJeff Mahoney err = ret; 209376dda93cSYan, Zheng inode->i_flags |= S_DEAD; 209476dda93cSYan, Zheng out_up_write: 209576dda93cSYan, Zheng up_write(&root->fs_info->subvol_sem); 209676dda93cSYan, Zheng out_unlock: 209776dda93cSYan, Zheng mutex_unlock(&inode->i_mutex); 209876dda93cSYan, Zheng if (!err) { 2099efefb143SYan, Zheng shrink_dcache_sb(root->fs_info->sb); 210076dda93cSYan, Zheng btrfs_invalidate_inodes(dest); 210176dda93cSYan, Zheng d_delete(dentry); 210276dda93cSYan, Zheng } 210376dda93cSYan, Zheng out_dput: 210476dda93cSYan, Zheng dput(dentry); 210576dda93cSYan, Zheng out_unlock_dir: 210676dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 21072a79f17eSAl Viro mnt_drop_write_file(file); 210876dda93cSYan, Zheng out: 210976dda93cSYan, Zheng kfree(vol_args); 211076dda93cSYan, Zheng return err; 211176dda93cSYan, Zheng } 211276dda93cSYan, Zheng 21131e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 2114f46b5a66SChristoph Hellwig { 2115f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2116f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 21171e701a32SChris Mason struct btrfs_ioctl_defrag_range_args *range; 2118c146afadSYan Zheng int ret; 2119c146afadSYan Zheng 2120b83cc969SLi Zefan if (btrfs_root_readonly(root)) 2121b83cc969SLi Zefan return -EROFS; 2122b83cc969SLi Zefan 2123a561be71SAl Viro ret = mnt_want_write_file(file); 2124c146afadSYan Zheng if (ret) 2125c146afadSYan Zheng return ret; 2126f46b5a66SChristoph Hellwig 2127f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2128f46b5a66SChristoph Hellwig case S_IFDIR: 2129e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 2130e441d54dSChris Mason ret = -EPERM; 2131e441d54dSChris Mason goto out; 2132e441d54dSChris Mason } 21338929ecfaSYan, Zheng ret = btrfs_defrag_root(root, 0); 21348929ecfaSYan, Zheng if (ret) 21358929ecfaSYan, Zheng goto out; 21368929ecfaSYan, Zheng ret = btrfs_defrag_root(root->fs_info->extent_root, 0); 2137f46b5a66SChristoph Hellwig break; 2138f46b5a66SChristoph Hellwig case S_IFREG: 2139e441d54dSChris Mason if (!(file->f_mode & FMODE_WRITE)) { 2140e441d54dSChris Mason ret = -EINVAL; 2141e441d54dSChris Mason goto out; 2142e441d54dSChris Mason } 21431e701a32SChris Mason 21441e701a32SChris Mason range = kzalloc(sizeof(*range), GFP_KERNEL); 21451e701a32SChris Mason if (!range) { 21461e701a32SChris Mason ret = -ENOMEM; 21471e701a32SChris Mason goto out; 21481e701a32SChris Mason } 21491e701a32SChris Mason 21501e701a32SChris Mason if (argp) { 21511e701a32SChris Mason if (copy_from_user(range, argp, 21521e701a32SChris Mason sizeof(*range))) { 21531e701a32SChris Mason ret = -EFAULT; 21541e701a32SChris Mason kfree(range); 2155683be16eSDan Carpenter goto out; 21561e701a32SChris Mason } 21571e701a32SChris Mason /* compression requires us to start the IO */ 21581e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 21591e701a32SChris Mason range->flags |= BTRFS_DEFRAG_RANGE_START_IO; 21601e701a32SChris Mason range->extent_thresh = (u32)-1; 21611e701a32SChris Mason } 21621e701a32SChris Mason } else { 21631e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 21641e701a32SChris Mason range->len = (u64)-1; 21651e701a32SChris Mason } 21664cb5300bSChris Mason ret = btrfs_defrag_file(fdentry(file)->d_inode, file, 21674cb5300bSChris Mason range, 0, 0); 21684cb5300bSChris Mason if (ret > 0) 21694cb5300bSChris Mason ret = 0; 21701e701a32SChris Mason kfree(range); 2171f46b5a66SChristoph Hellwig break; 21728929ecfaSYan, Zheng default: 21738929ecfaSYan, Zheng ret = -EINVAL; 2174f46b5a66SChristoph Hellwig } 2175e441d54dSChris Mason out: 21762a79f17eSAl Viro mnt_drop_write_file(file); 2177e441d54dSChris Mason return ret; 2178f46b5a66SChristoph Hellwig } 2179f46b5a66SChristoph Hellwig 2180b2950863SChristoph Hellwig static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg) 2181f46b5a66SChristoph Hellwig { 2182f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2183f46b5a66SChristoph Hellwig int ret; 2184f46b5a66SChristoph Hellwig 2185e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2186e441d54dSChris Mason return -EPERM; 2187e441d54dSChris Mason 2188c9e9f97bSIlya Dryomov mutex_lock(&root->fs_info->volume_mutex); 2189c9e9f97bSIlya Dryomov if (root->fs_info->balance_ctl) { 2190c9e9f97bSIlya Dryomov printk(KERN_INFO "btrfs: balance in progress\n"); 2191c9e9f97bSIlya Dryomov ret = -EINVAL; 2192c9e9f97bSIlya Dryomov goto out; 2193c9e9f97bSIlya Dryomov } 2194c9e9f97bSIlya Dryomov 2195dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2196c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2197c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2198c9e9f97bSIlya Dryomov goto out; 2199c9e9f97bSIlya Dryomov } 2200f46b5a66SChristoph Hellwig 22015516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2202f46b5a66SChristoph Hellwig ret = btrfs_init_new_device(root, vol_args->name); 2203f46b5a66SChristoph Hellwig 2204f46b5a66SChristoph Hellwig kfree(vol_args); 2205c9e9f97bSIlya Dryomov out: 2206c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 2207f46b5a66SChristoph Hellwig return ret; 2208f46b5a66SChristoph Hellwig } 2209f46b5a66SChristoph Hellwig 2210b2950863SChristoph Hellwig static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg) 2211f46b5a66SChristoph Hellwig { 2212f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2213f46b5a66SChristoph Hellwig int ret; 2214f46b5a66SChristoph Hellwig 2215e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2216e441d54dSChris Mason return -EPERM; 2217e441d54dSChris Mason 2218c146afadSYan Zheng if (root->fs_info->sb->s_flags & MS_RDONLY) 2219c146afadSYan Zheng return -EROFS; 2220c146afadSYan Zheng 2221c9e9f97bSIlya Dryomov mutex_lock(&root->fs_info->volume_mutex); 2222c9e9f97bSIlya Dryomov if (root->fs_info->balance_ctl) { 2223c9e9f97bSIlya Dryomov printk(KERN_INFO "btrfs: balance in progress\n"); 2224c9e9f97bSIlya Dryomov ret = -EINVAL; 2225c9e9f97bSIlya Dryomov goto out; 2226c9e9f97bSIlya Dryomov } 2227c9e9f97bSIlya Dryomov 2228dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2229c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2230c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2231c9e9f97bSIlya Dryomov goto out; 2232c9e9f97bSIlya Dryomov } 2233f46b5a66SChristoph Hellwig 22345516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2235f46b5a66SChristoph Hellwig ret = btrfs_rm_device(root, vol_args->name); 2236f46b5a66SChristoph Hellwig 2237f46b5a66SChristoph Hellwig kfree(vol_args); 2238c9e9f97bSIlya Dryomov out: 2239c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 2240f46b5a66SChristoph Hellwig return ret; 2241f46b5a66SChristoph Hellwig } 2242f46b5a66SChristoph Hellwig 2243475f6387SJan Schmidt static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg) 2244475f6387SJan Schmidt { 2245027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 2246475f6387SJan Schmidt struct btrfs_device *device; 2247475f6387SJan Schmidt struct btrfs_device *next; 2248475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2249027ed2f0SLi Zefan int ret = 0; 2250475f6387SJan Schmidt 2251475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 2252475f6387SJan Schmidt return -EPERM; 2253475f6387SJan Schmidt 2254027ed2f0SLi Zefan fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL); 2255027ed2f0SLi Zefan if (!fi_args) 2256027ed2f0SLi Zefan return -ENOMEM; 2257027ed2f0SLi Zefan 2258027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 2259027ed2f0SLi Zefan memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid)); 2260475f6387SJan Schmidt 2261475f6387SJan Schmidt mutex_lock(&fs_devices->device_list_mutex); 2262475f6387SJan Schmidt list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 2263027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 2264027ed2f0SLi Zefan fi_args->max_id = device->devid; 2265475f6387SJan Schmidt } 2266475f6387SJan Schmidt mutex_unlock(&fs_devices->device_list_mutex); 2267475f6387SJan Schmidt 2268027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 2269027ed2f0SLi Zefan ret = -EFAULT; 2270475f6387SJan Schmidt 2271027ed2f0SLi Zefan kfree(fi_args); 2272027ed2f0SLi Zefan return ret; 2273475f6387SJan Schmidt } 2274475f6387SJan Schmidt 2275475f6387SJan Schmidt static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg) 2276475f6387SJan Schmidt { 2277475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 2278475f6387SJan Schmidt struct btrfs_device *dev; 2279475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2280475f6387SJan Schmidt int ret = 0; 2281475f6387SJan Schmidt char *s_uuid = NULL; 2282475f6387SJan Schmidt char empty_uuid[BTRFS_UUID_SIZE] = {0}; 2283475f6387SJan Schmidt 2284475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 2285475f6387SJan Schmidt return -EPERM; 2286475f6387SJan Schmidt 2287475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 2288475f6387SJan Schmidt if (IS_ERR(di_args)) 2289475f6387SJan Schmidt return PTR_ERR(di_args); 2290475f6387SJan Schmidt 2291475f6387SJan Schmidt if (memcmp(empty_uuid, di_args->uuid, BTRFS_UUID_SIZE) != 0) 2292475f6387SJan Schmidt s_uuid = di_args->uuid; 2293475f6387SJan Schmidt 2294475f6387SJan Schmidt mutex_lock(&fs_devices->device_list_mutex); 2295475f6387SJan Schmidt dev = btrfs_find_device(root, di_args->devid, s_uuid, NULL); 2296475f6387SJan Schmidt mutex_unlock(&fs_devices->device_list_mutex); 2297475f6387SJan Schmidt 2298475f6387SJan Schmidt if (!dev) { 2299475f6387SJan Schmidt ret = -ENODEV; 2300475f6387SJan Schmidt goto out; 2301475f6387SJan Schmidt } 2302475f6387SJan Schmidt 2303475f6387SJan Schmidt di_args->devid = dev->devid; 2304475f6387SJan Schmidt di_args->bytes_used = dev->bytes_used; 2305475f6387SJan Schmidt di_args->total_bytes = dev->total_bytes; 2306475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 2307a27202fbSJim Meyering if (dev->name) { 2308606686eeSJosef Bacik struct rcu_string *name; 2309606686eeSJosef Bacik 2310606686eeSJosef Bacik rcu_read_lock(); 2311606686eeSJosef Bacik name = rcu_dereference(dev->name); 2312606686eeSJosef Bacik strncpy(di_args->path, name->str, sizeof(di_args->path)); 2313606686eeSJosef Bacik rcu_read_unlock(); 2314a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 2315a27202fbSJim Meyering } else { 231699ba55adSStefan Behrens di_args->path[0] = '\0'; 2317a27202fbSJim Meyering } 2318475f6387SJan Schmidt 2319475f6387SJan Schmidt out: 2320475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 2321475f6387SJan Schmidt ret = -EFAULT; 2322475f6387SJan Schmidt 2323475f6387SJan Schmidt kfree(di_args); 2324475f6387SJan Schmidt return ret; 2325475f6387SJan Schmidt } 2326475f6387SJan Schmidt 232776dda93cSYan, Zheng static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, 2328b2950863SChristoph Hellwig u64 off, u64 olen, u64 destoff) 2329f46b5a66SChristoph Hellwig { 2330f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2331f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 2332f46b5a66SChristoph Hellwig struct file *src_file; 2333f46b5a66SChristoph Hellwig struct inode *src; 2334f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 2335f46b5a66SChristoph Hellwig struct btrfs_path *path; 2336f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 2337ae01a0abSYan Zheng char *buf; 2338ae01a0abSYan Zheng struct btrfs_key key; 2339f46b5a66SChristoph Hellwig u32 nritems; 2340f46b5a66SChristoph Hellwig int slot; 2341ae01a0abSYan Zheng int ret; 2342c5c9cd4dSSage Weil u64 len = olen; 2343c5c9cd4dSSage Weil u64 bs = root->fs_info->sb->s_blocksize; 2344c5c9cd4dSSage Weil u64 hint_byte; 2345d20f7043SChris Mason 2346c5c9cd4dSSage Weil /* 2347c5c9cd4dSSage Weil * TODO: 2348c5c9cd4dSSage Weil * - split compressed inline extents. annoying: we need to 2349c5c9cd4dSSage Weil * decompress into destination's address_space (the file offset 2350c5c9cd4dSSage Weil * may change, so source mapping won't do), then recompress (or 2351c5c9cd4dSSage Weil * otherwise reinsert) a subrange. 2352c5c9cd4dSSage Weil * - allow ranges within the same file to be cloned (provided 2353c5c9cd4dSSage Weil * they don't overlap)? 2354c5c9cd4dSSage Weil */ 2355c5c9cd4dSSage Weil 2356e441d54dSChris Mason /* the destination must be opened for writing */ 23572ebc3464SDan Rosenberg if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) 2358e441d54dSChris Mason return -EINVAL; 2359e441d54dSChris Mason 2360b83cc969SLi Zefan if (btrfs_root_readonly(root)) 2361b83cc969SLi Zefan return -EROFS; 2362b83cc969SLi Zefan 2363a561be71SAl Viro ret = mnt_want_write_file(file); 2364c146afadSYan Zheng if (ret) 2365c146afadSYan Zheng return ret; 2366c146afadSYan Zheng 2367c5c9cd4dSSage Weil src_file = fget(srcfd); 2368ab67b7c1SYan Zheng if (!src_file) { 2369ab67b7c1SYan Zheng ret = -EBADF; 2370ab67b7c1SYan Zheng goto out_drop_write; 2371ab67b7c1SYan Zheng } 23725dc64164SDan Rosenberg 2373f46b5a66SChristoph Hellwig src = src_file->f_dentry->d_inode; 2374f46b5a66SChristoph Hellwig 2375c5c9cd4dSSage Weil ret = -EINVAL; 2376c5c9cd4dSSage Weil if (src == inode) 2377c5c9cd4dSSage Weil goto out_fput; 2378c5c9cd4dSSage Weil 23795dc64164SDan Rosenberg /* the src must be open for reading */ 23805dc64164SDan Rosenberg if (!(src_file->f_mode & FMODE_READ)) 23815dc64164SDan Rosenberg goto out_fput; 23825dc64164SDan Rosenberg 23830e7b824cSLi Zefan /* don't make the dst file partly checksummed */ 23840e7b824cSLi Zefan if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 23850e7b824cSLi Zefan (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) 23860e7b824cSLi Zefan goto out_fput; 23870e7b824cSLi Zefan 2388ae01a0abSYan Zheng ret = -EISDIR; 2389ae01a0abSYan Zheng if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode)) 2390f46b5a66SChristoph Hellwig goto out_fput; 2391f46b5a66SChristoph Hellwig 2392ae01a0abSYan Zheng ret = -EXDEV; 2393ae01a0abSYan Zheng if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root) 2394ae01a0abSYan Zheng goto out_fput; 2395ae01a0abSYan Zheng 2396ae01a0abSYan Zheng ret = -ENOMEM; 2397ae01a0abSYan Zheng buf = vmalloc(btrfs_level_size(root, 0)); 2398ae01a0abSYan Zheng if (!buf) 2399ae01a0abSYan Zheng goto out_fput; 2400ae01a0abSYan Zheng 2401ae01a0abSYan Zheng path = btrfs_alloc_path(); 2402ae01a0abSYan Zheng if (!path) { 2403ae01a0abSYan Zheng vfree(buf); 2404ae01a0abSYan Zheng goto out_fput; 2405ae01a0abSYan Zheng } 2406ae01a0abSYan Zheng path->reada = 2; 2407ae01a0abSYan Zheng 2408f46b5a66SChristoph Hellwig if (inode < src) { 2409fccdae43SSage Weil mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT); 2410fccdae43SSage Weil mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD); 2411f46b5a66SChristoph Hellwig } else { 2412fccdae43SSage Weil mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT); 2413fccdae43SSage Weil mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); 2414f46b5a66SChristoph Hellwig } 2415f46b5a66SChristoph Hellwig 2416c5c9cd4dSSage Weil /* determine range to clone */ 2417c5c9cd4dSSage Weil ret = -EINVAL; 24182ebc3464SDan Rosenberg if (off + len > src->i_size || off + len < off) 2419f46b5a66SChristoph Hellwig goto out_unlock; 2420c5c9cd4dSSage Weil if (len == 0) 2421c5c9cd4dSSage Weil olen = len = src->i_size - off; 2422c5c9cd4dSSage Weil /* if we extend to eof, continue to block boundary */ 2423c5c9cd4dSSage Weil if (off + len == src->i_size) 24242a6b8daeSLi Zefan len = ALIGN(src->i_size, bs) - off; 2425c5c9cd4dSSage Weil 2426c5c9cd4dSSage Weil /* verify the end result is block aligned */ 24272a6b8daeSLi Zefan if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) || 24282a6b8daeSLi Zefan !IS_ALIGNED(destoff, bs)) 2429c5c9cd4dSSage Weil goto out_unlock; 2430c5c9cd4dSSage Weil 2431d525e8abSLi Zefan if (destoff > inode->i_size) { 2432d525e8abSLi Zefan ret = btrfs_cont_expand(inode, inode->i_size, destoff); 2433d525e8abSLi Zefan if (ret) 2434d525e8abSLi Zefan goto out_unlock; 2435d525e8abSLi Zefan } 2436d525e8abSLi Zefan 243771ef0786SLi Zefan /* truncate page cache pages from target inode range */ 243871ef0786SLi Zefan truncate_inode_pages_range(&inode->i_data, destoff, 243971ef0786SLi Zefan PAGE_CACHE_ALIGN(destoff + len) - 1); 244071ef0786SLi Zefan 2441f46b5a66SChristoph Hellwig /* do any pending delalloc/csum calc on src, one way or 2442f46b5a66SChristoph Hellwig another, and lock file content */ 2443f46b5a66SChristoph Hellwig while (1) { 244431840ae1SZheng Yan struct btrfs_ordered_extent *ordered; 2445d0082371SJeff Mahoney lock_extent(&BTRFS_I(src)->io_tree, off, off+len); 24469a019196SSage Weil ordered = btrfs_lookup_first_ordered_extent(src, off+len); 24479a019196SSage Weil if (!ordered && 24489a019196SSage Weil !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len, 24499a019196SSage Weil EXTENT_DELALLOC, 0, NULL)) 2450f46b5a66SChristoph Hellwig break; 2451d0082371SJeff Mahoney unlock_extent(&BTRFS_I(src)->io_tree, off, off+len); 2452ae01a0abSYan Zheng if (ordered) 2453ae01a0abSYan Zheng btrfs_put_ordered_extent(ordered); 24549a019196SSage Weil btrfs_wait_ordered_range(src, off, len); 2455f46b5a66SChristoph Hellwig } 2456f46b5a66SChristoph Hellwig 2457c5c9cd4dSSage Weil /* clone data */ 245833345d01SLi Zefan key.objectid = btrfs_ino(src); 2459ae01a0abSYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 2460ae01a0abSYan Zheng key.offset = 0; 2461f46b5a66SChristoph Hellwig 2462f46b5a66SChristoph Hellwig while (1) { 2463f46b5a66SChristoph Hellwig /* 2464f46b5a66SChristoph Hellwig * note the key will change type as we walk through the 2465f46b5a66SChristoph Hellwig * tree. 2466f46b5a66SChristoph Hellwig */ 2467a22285a6SYan, Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2468f46b5a66SChristoph Hellwig if (ret < 0) 2469f46b5a66SChristoph Hellwig goto out; 2470f46b5a66SChristoph Hellwig 2471ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 2472ae01a0abSYan Zheng if (path->slots[0] >= nritems) { 2473f46b5a66SChristoph Hellwig ret = btrfs_next_leaf(root, path); 2474f46b5a66SChristoph Hellwig if (ret < 0) 2475f46b5a66SChristoph Hellwig goto out; 2476f46b5a66SChristoph Hellwig if (ret > 0) 2477f46b5a66SChristoph Hellwig break; 2478ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 2479f46b5a66SChristoph Hellwig } 2480f46b5a66SChristoph Hellwig leaf = path->nodes[0]; 2481f46b5a66SChristoph Hellwig slot = path->slots[0]; 2482f46b5a66SChristoph Hellwig 2483ae01a0abSYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 2484d20f7043SChris Mason if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY || 248533345d01SLi Zefan key.objectid != btrfs_ino(src)) 2486f46b5a66SChristoph Hellwig break; 2487f46b5a66SChristoph Hellwig 2488c5c9cd4dSSage Weil if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) { 2489c5c9cd4dSSage Weil struct btrfs_file_extent_item *extent; 2490c5c9cd4dSSage Weil int type; 249131840ae1SZheng Yan u32 size; 249231840ae1SZheng Yan struct btrfs_key new_key; 2493c5c9cd4dSSage Weil u64 disko = 0, diskl = 0; 2494c5c9cd4dSSage Weil u64 datao = 0, datal = 0; 2495c5c9cd4dSSage Weil u8 comp; 2496b5384d48SSage Weil u64 endoff; 249731840ae1SZheng Yan 249831840ae1SZheng Yan size = btrfs_item_size_nr(leaf, slot); 249931840ae1SZheng Yan read_extent_buffer(leaf, buf, 250031840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 250131840ae1SZheng Yan size); 2502c5c9cd4dSSage Weil 2503c5c9cd4dSSage Weil extent = btrfs_item_ptr(leaf, slot, 2504c5c9cd4dSSage Weil struct btrfs_file_extent_item); 2505c5c9cd4dSSage Weil comp = btrfs_file_extent_compression(leaf, extent); 2506c5c9cd4dSSage Weil type = btrfs_file_extent_type(leaf, extent); 2507c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 2508c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 2509d397712bSChris Mason disko = btrfs_file_extent_disk_bytenr(leaf, 2510d397712bSChris Mason extent); 2511d397712bSChris Mason diskl = btrfs_file_extent_disk_num_bytes(leaf, 2512d397712bSChris Mason extent); 2513c5c9cd4dSSage Weil datao = btrfs_file_extent_offset(leaf, extent); 2514d397712bSChris Mason datal = btrfs_file_extent_num_bytes(leaf, 2515d397712bSChris Mason extent); 2516c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 2517c5c9cd4dSSage Weil /* take upper bound, may be compressed */ 2518c5c9cd4dSSage Weil datal = btrfs_file_extent_ram_bytes(leaf, 2519c5c9cd4dSSage Weil extent); 2520c5c9cd4dSSage Weil } 2521b3b4aa74SDavid Sterba btrfs_release_path(path); 252231840ae1SZheng Yan 2523050006a7SSage Weil if (key.offset + datal <= off || 2524c5c9cd4dSSage Weil key.offset >= off+len) 2525c5c9cd4dSSage Weil goto next; 2526c5c9cd4dSSage Weil 252731840ae1SZheng Yan memcpy(&new_key, &key, sizeof(new_key)); 252833345d01SLi Zefan new_key.objectid = btrfs_ino(inode); 25294d728ec7SLi Zefan if (off <= key.offset) 2530c5c9cd4dSSage Weil new_key.offset = key.offset + destoff - off; 25314d728ec7SLi Zefan else 25324d728ec7SLi Zefan new_key.offset = destoff; 2533c5c9cd4dSSage Weil 2534b6f3409bSSage Weil /* 2535b6f3409bSSage Weil * 1 - adjusting old extent (we may have to split it) 2536b6f3409bSSage Weil * 1 - add new extent 2537b6f3409bSSage Weil * 1 - inode update 2538b6f3409bSSage Weil */ 2539b6f3409bSSage Weil trans = btrfs_start_transaction(root, 3); 2540a22285a6SYan, Zheng if (IS_ERR(trans)) { 2541a22285a6SYan, Zheng ret = PTR_ERR(trans); 2542a22285a6SYan, Zheng goto out; 2543a22285a6SYan, Zheng } 2544a22285a6SYan, Zheng 2545c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 2546c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 2547d72c0842SLi Zefan /* 2548d72c0842SLi Zefan * a | --- range to clone ---| b 2549d72c0842SLi Zefan * | ------------- extent ------------- | 2550d72c0842SLi Zefan */ 2551d72c0842SLi Zefan 2552d72c0842SLi Zefan /* substract range b */ 2553d72c0842SLi Zefan if (key.offset + datal > off + len) 2554d72c0842SLi Zefan datal = off + len - key.offset; 2555d72c0842SLi Zefan 2556d72c0842SLi Zefan /* substract range a */ 2557a22285a6SYan, Zheng if (off > key.offset) { 2558a22285a6SYan, Zheng datao += off - key.offset; 2559a22285a6SYan, Zheng datal -= off - key.offset; 2560a22285a6SYan, Zheng } 2561a22285a6SYan, Zheng 2562a22285a6SYan, Zheng ret = btrfs_drop_extents(trans, inode, 2563a22285a6SYan, Zheng new_key.offset, 2564a22285a6SYan, Zheng new_key.offset + datal, 2565a22285a6SYan, Zheng &hint_byte, 1); 256679787eaaSJeff Mahoney if (ret) { 256779787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 256879787eaaSJeff Mahoney ret); 256979787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 257079787eaaSJeff Mahoney goto out; 257179787eaaSJeff Mahoney } 2572a22285a6SYan, Zheng 257331840ae1SZheng Yan ret = btrfs_insert_empty_item(trans, root, path, 257431840ae1SZheng Yan &new_key, size); 257579787eaaSJeff Mahoney if (ret) { 257679787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 257779787eaaSJeff Mahoney ret); 257879787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 257979787eaaSJeff Mahoney goto out; 258079787eaaSJeff Mahoney } 258131840ae1SZheng Yan 258231840ae1SZheng Yan leaf = path->nodes[0]; 258331840ae1SZheng Yan slot = path->slots[0]; 258431840ae1SZheng Yan write_extent_buffer(leaf, buf, 258531840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 258631840ae1SZheng Yan size); 2587ae01a0abSYan Zheng 2588f46b5a66SChristoph Hellwig extent = btrfs_item_ptr(leaf, slot, 2589f46b5a66SChristoph Hellwig struct btrfs_file_extent_item); 2590c5c9cd4dSSage Weil 2591c5c9cd4dSSage Weil /* disko == 0 means it's a hole */ 2592c5c9cd4dSSage Weil if (!disko) 2593c5c9cd4dSSage Weil datao = 0; 2594c5c9cd4dSSage Weil 2595c5c9cd4dSSage Weil btrfs_set_file_extent_offset(leaf, extent, 2596c5c9cd4dSSage Weil datao); 2597c5c9cd4dSSage Weil btrfs_set_file_extent_num_bytes(leaf, extent, 2598c5c9cd4dSSage Weil datal); 2599c5c9cd4dSSage Weil if (disko) { 2600c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 2601ae01a0abSYan Zheng ret = btrfs_inc_extent_ref(trans, root, 26025d4f98a2SYan Zheng disko, diskl, 0, 2603f46b5a66SChristoph Hellwig root->root_key.objectid, 260433345d01SLi Zefan btrfs_ino(inode), 260566d7e7f0SArne Jansen new_key.offset - datao, 260666d7e7f0SArne Jansen 0); 260779787eaaSJeff Mahoney if (ret) { 260879787eaaSJeff Mahoney btrfs_abort_transaction(trans, 260979787eaaSJeff Mahoney root, 261079787eaaSJeff Mahoney ret); 261179787eaaSJeff Mahoney btrfs_end_transaction(trans, 261279787eaaSJeff Mahoney root); 261379787eaaSJeff Mahoney goto out; 261479787eaaSJeff Mahoney 261579787eaaSJeff Mahoney } 2616f46b5a66SChristoph Hellwig } 2617c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 2618c5c9cd4dSSage Weil u64 skip = 0; 2619c5c9cd4dSSage Weil u64 trim = 0; 2620c5c9cd4dSSage Weil if (off > key.offset) { 2621c5c9cd4dSSage Weil skip = off - key.offset; 2622c5c9cd4dSSage Weil new_key.offset += skip; 262331840ae1SZheng Yan } 2624d397712bSChris Mason 2625c5c9cd4dSSage Weil if (key.offset + datal > off+len) 2626c5c9cd4dSSage Weil trim = key.offset + datal - (off+len); 2627d397712bSChris Mason 2628c5c9cd4dSSage Weil if (comp && (skip || trim)) { 2629c5c9cd4dSSage Weil ret = -EINVAL; 2630a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 2631c5c9cd4dSSage Weil goto out; 263231840ae1SZheng Yan } 2633c5c9cd4dSSage Weil size -= skip + trim; 2634c5c9cd4dSSage Weil datal -= skip + trim; 2635a22285a6SYan, Zheng 2636a22285a6SYan, Zheng ret = btrfs_drop_extents(trans, inode, 2637a22285a6SYan, Zheng new_key.offset, 2638a22285a6SYan, Zheng new_key.offset + datal, 2639a22285a6SYan, Zheng &hint_byte, 1); 264079787eaaSJeff Mahoney if (ret) { 264179787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 264279787eaaSJeff Mahoney ret); 264379787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 264479787eaaSJeff Mahoney goto out; 264579787eaaSJeff Mahoney } 2646a22285a6SYan, Zheng 2647c5c9cd4dSSage Weil ret = btrfs_insert_empty_item(trans, root, path, 2648c5c9cd4dSSage Weil &new_key, size); 264979787eaaSJeff Mahoney if (ret) { 265079787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 265179787eaaSJeff Mahoney ret); 265279787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 265379787eaaSJeff Mahoney goto out; 265479787eaaSJeff Mahoney } 2655c5c9cd4dSSage Weil 2656c5c9cd4dSSage Weil if (skip) { 2657d397712bSChris Mason u32 start = 2658d397712bSChris Mason btrfs_file_extent_calc_inline_size(0); 2659c5c9cd4dSSage Weil memmove(buf+start, buf+start+skip, 2660c5c9cd4dSSage Weil datal); 2661c5c9cd4dSSage Weil } 2662c5c9cd4dSSage Weil 2663c5c9cd4dSSage Weil leaf = path->nodes[0]; 2664c5c9cd4dSSage Weil slot = path->slots[0]; 2665c5c9cd4dSSage Weil write_extent_buffer(leaf, buf, 2666c5c9cd4dSSage Weil btrfs_item_ptr_offset(leaf, slot), 2667c5c9cd4dSSage Weil size); 2668c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 2669c5c9cd4dSSage Weil } 2670c5c9cd4dSSage Weil 2671c5c9cd4dSSage Weil btrfs_mark_buffer_dirty(leaf); 2672b3b4aa74SDavid Sterba btrfs_release_path(path); 2673c5c9cd4dSSage Weil 26740c4d2d95SJosef Bacik inode_inc_iversion(inode); 2675a22285a6SYan, Zheng inode->i_mtime = inode->i_ctime = CURRENT_TIME; 2676b5384d48SSage Weil 2677b5384d48SSage Weil /* 2678b5384d48SSage Weil * we round up to the block size at eof when 2679b5384d48SSage Weil * determining which extents to clone above, 2680b5384d48SSage Weil * but shouldn't round up the file size 2681b5384d48SSage Weil */ 2682b5384d48SSage Weil endoff = new_key.offset + datal; 26835f3888ffSLi Zefan if (endoff > destoff+olen) 26845f3888ffSLi Zefan endoff = destoff+olen; 2685b5384d48SSage Weil if (endoff > inode->i_size) 2686b5384d48SSage Weil btrfs_i_size_write(inode, endoff); 2687b5384d48SSage Weil 2688a22285a6SYan, Zheng ret = btrfs_update_inode(trans, root, inode); 268979787eaaSJeff Mahoney if (ret) { 269079787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 2691a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 269279787eaaSJeff Mahoney goto out; 269379787eaaSJeff Mahoney } 269479787eaaSJeff Mahoney ret = btrfs_end_transaction(trans, root); 2695a22285a6SYan, Zheng } 2696c5c9cd4dSSage Weil next: 2697b3b4aa74SDavid Sterba btrfs_release_path(path); 2698ae01a0abSYan Zheng key.offset++; 2699ae01a0abSYan Zheng } 2700f46b5a66SChristoph Hellwig ret = 0; 2701f46b5a66SChristoph Hellwig out: 2702b3b4aa74SDavid Sterba btrfs_release_path(path); 2703d0082371SJeff Mahoney unlock_extent(&BTRFS_I(src)->io_tree, off, off+len); 2704f46b5a66SChristoph Hellwig out_unlock: 2705f46b5a66SChristoph Hellwig mutex_unlock(&src->i_mutex); 2706f46b5a66SChristoph Hellwig mutex_unlock(&inode->i_mutex); 2707ae01a0abSYan Zheng vfree(buf); 2708ae01a0abSYan Zheng btrfs_free_path(path); 2709f46b5a66SChristoph Hellwig out_fput: 2710f46b5a66SChristoph Hellwig fput(src_file); 2711ab67b7c1SYan Zheng out_drop_write: 27122a79f17eSAl Viro mnt_drop_write_file(file); 2713f46b5a66SChristoph Hellwig return ret; 2714f46b5a66SChristoph Hellwig } 2715f46b5a66SChristoph Hellwig 27167a865e8aSChristoph Hellwig static long btrfs_ioctl_clone_range(struct file *file, void __user *argp) 2717c5c9cd4dSSage Weil { 2718c5c9cd4dSSage Weil struct btrfs_ioctl_clone_range_args args; 2719c5c9cd4dSSage Weil 27207a865e8aSChristoph Hellwig if (copy_from_user(&args, argp, sizeof(args))) 2721c5c9cd4dSSage Weil return -EFAULT; 2722c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, args.src_fd, args.src_offset, 2723c5c9cd4dSSage Weil args.src_length, args.dest_offset); 2724c5c9cd4dSSage Weil } 2725c5c9cd4dSSage Weil 2726f46b5a66SChristoph Hellwig /* 2727f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 2728f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 2729f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 2730f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 2731f46b5a66SChristoph Hellwig */ 2732b2950863SChristoph Hellwig static long btrfs_ioctl_trans_start(struct file *file) 2733f46b5a66SChristoph Hellwig { 2734f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2735f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 2736f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 27371ab86aedSSage Weil int ret; 2738f46b5a66SChristoph Hellwig 27391ab86aedSSage Weil ret = -EPERM; 2740df5b5520SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 2741f46b5a66SChristoph Hellwig goto out; 27421ab86aedSSage Weil 27431ab86aedSSage Weil ret = -EINPROGRESS; 27441ab86aedSSage Weil if (file->private_data) 27451ab86aedSSage Weil goto out; 27469ca9ee09SSage Weil 2747b83cc969SLi Zefan ret = -EROFS; 2748b83cc969SLi Zefan if (btrfs_root_readonly(root)) 2749b83cc969SLi Zefan goto out; 2750b83cc969SLi Zefan 2751a561be71SAl Viro ret = mnt_want_write_file(file); 2752c146afadSYan Zheng if (ret) 2753c146afadSYan Zheng goto out; 2754c146afadSYan Zheng 2755a4abeea4SJosef Bacik atomic_inc(&root->fs_info->open_ioctl_trans); 27569ca9ee09SSage Weil 2757f46b5a66SChristoph Hellwig ret = -ENOMEM; 27587a7eaa40SJosef Bacik trans = btrfs_start_ioctl_transaction(root); 2759abd30bb0STsutomu Itoh if (IS_ERR(trans)) 27601ab86aedSSage Weil goto out_drop; 27611ab86aedSSage Weil 27621ab86aedSSage Weil file->private_data = trans; 27631ab86aedSSage Weil return 0; 27641ab86aedSSage Weil 27651ab86aedSSage Weil out_drop: 2766a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 27672a79f17eSAl Viro mnt_drop_write_file(file); 2768f46b5a66SChristoph Hellwig out: 2769f46b5a66SChristoph Hellwig return ret; 2770f46b5a66SChristoph Hellwig } 2771f46b5a66SChristoph Hellwig 27726ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 27736ef5ed0dSJosef Bacik { 27746ef5ed0dSJosef Bacik struct inode *inode = fdentry(file)->d_inode; 27756ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 27766ef5ed0dSJosef Bacik struct btrfs_root *new_root; 27776ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 27786ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 27796ef5ed0dSJosef Bacik struct btrfs_path *path; 27806ef5ed0dSJosef Bacik struct btrfs_key location; 27816ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 27826ef5ed0dSJosef Bacik u64 objectid = 0; 27836ef5ed0dSJosef Bacik u64 dir_id; 27846ef5ed0dSJosef Bacik 27856ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 27866ef5ed0dSJosef Bacik return -EPERM; 27876ef5ed0dSJosef Bacik 27886ef5ed0dSJosef Bacik if (copy_from_user(&objectid, argp, sizeof(objectid))) 27896ef5ed0dSJosef Bacik return -EFAULT; 27906ef5ed0dSJosef Bacik 27916ef5ed0dSJosef Bacik if (!objectid) 27926ef5ed0dSJosef Bacik objectid = root->root_key.objectid; 27936ef5ed0dSJosef Bacik 27946ef5ed0dSJosef Bacik location.objectid = objectid; 27956ef5ed0dSJosef Bacik location.type = BTRFS_ROOT_ITEM_KEY; 27966ef5ed0dSJosef Bacik location.offset = (u64)-1; 27976ef5ed0dSJosef Bacik 27986ef5ed0dSJosef Bacik new_root = btrfs_read_fs_root_no_name(root->fs_info, &location); 27996ef5ed0dSJosef Bacik if (IS_ERR(new_root)) 28006ef5ed0dSJosef Bacik return PTR_ERR(new_root); 28016ef5ed0dSJosef Bacik 28026ef5ed0dSJosef Bacik if (btrfs_root_refs(&new_root->root_item) == 0) 28036ef5ed0dSJosef Bacik return -ENOENT; 28046ef5ed0dSJosef Bacik 28056ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 28066ef5ed0dSJosef Bacik if (!path) 28076ef5ed0dSJosef Bacik return -ENOMEM; 28086ef5ed0dSJosef Bacik path->leave_spinning = 1; 28096ef5ed0dSJosef Bacik 28106ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 281198d5dc13STsutomu Itoh if (IS_ERR(trans)) { 28126ef5ed0dSJosef Bacik btrfs_free_path(path); 281398d5dc13STsutomu Itoh return PTR_ERR(trans); 28146ef5ed0dSJosef Bacik } 28156ef5ed0dSJosef Bacik 28166c41761fSDavid Sterba dir_id = btrfs_super_root_dir(root->fs_info->super_copy); 28176ef5ed0dSJosef Bacik di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path, 28186ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 2819cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 28206ef5ed0dSJosef Bacik btrfs_free_path(path); 28216ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 28226ef5ed0dSJosef Bacik printk(KERN_ERR "Umm, you don't have the default dir item, " 28236ef5ed0dSJosef Bacik "this isn't going to work\n"); 28246ef5ed0dSJosef Bacik return -ENOENT; 28256ef5ed0dSJosef Bacik } 28266ef5ed0dSJosef Bacik 28276ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 28286ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 28296ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 28306ef5ed0dSJosef Bacik btrfs_free_path(path); 28316ef5ed0dSJosef Bacik 2832*2b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL); 28336ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 28346ef5ed0dSJosef Bacik 28356ef5ed0dSJosef Bacik return 0; 28366ef5ed0dSJosef Bacik } 28376ef5ed0dSJosef Bacik 2838bf5fc093SJosef Bacik static void get_block_group_info(struct list_head *groups_list, 2839bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 2840bf5fc093SJosef Bacik { 2841bf5fc093SJosef Bacik struct btrfs_block_group_cache *block_group; 2842bf5fc093SJosef Bacik 2843bf5fc093SJosef Bacik space->total_bytes = 0; 2844bf5fc093SJosef Bacik space->used_bytes = 0; 2845bf5fc093SJosef Bacik space->flags = 0; 2846bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 2847bf5fc093SJosef Bacik space->flags = block_group->flags; 2848bf5fc093SJosef Bacik space->total_bytes += block_group->key.offset; 2849bf5fc093SJosef Bacik space->used_bytes += 2850bf5fc093SJosef Bacik btrfs_block_group_used(&block_group->item); 2851bf5fc093SJosef Bacik } 2852bf5fc093SJosef Bacik } 2853bf5fc093SJosef Bacik 28541406e432SJosef Bacik long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg) 28551406e432SJosef Bacik { 28561406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 28571406e432SJosef Bacik struct btrfs_ioctl_space_info space; 28581406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 28597fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 286013f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 28611406e432SJosef Bacik struct btrfs_space_info *info; 2862bf5fc093SJosef Bacik u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 2863bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 2864bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 2865bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 2866bf5fc093SJosef Bacik int num_types = 4; 28677fde62bfSChris Mason int alloc_size; 28681406e432SJosef Bacik int ret = 0; 286951788b1bSDan Rosenberg u64 slot_count = 0; 2870bf5fc093SJosef Bacik int i, c; 28711406e432SJosef Bacik 28721406e432SJosef Bacik if (copy_from_user(&space_args, 28731406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 28741406e432SJosef Bacik sizeof(space_args))) 28751406e432SJosef Bacik return -EFAULT; 28761406e432SJosef Bacik 2877bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 2878bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 2879bf5fc093SJosef Bacik 2880bf5fc093SJosef Bacik info = NULL; 28817fde62bfSChris Mason rcu_read_lock(); 2882bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 2883bf5fc093SJosef Bacik list) { 2884bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 2885bf5fc093SJosef Bacik info = tmp; 2886bf5fc093SJosef Bacik break; 2887bf5fc093SJosef Bacik } 2888bf5fc093SJosef Bacik } 28897fde62bfSChris Mason rcu_read_unlock(); 28901406e432SJosef Bacik 2891bf5fc093SJosef Bacik if (!info) 2892bf5fc093SJosef Bacik continue; 2893bf5fc093SJosef Bacik 2894bf5fc093SJosef Bacik down_read(&info->groups_sem); 2895bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 2896bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 2897bf5fc093SJosef Bacik slot_count++; 2898bf5fc093SJosef Bacik } 2899bf5fc093SJosef Bacik up_read(&info->groups_sem); 2900bf5fc093SJosef Bacik } 2901bf5fc093SJosef Bacik 29027fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 29037fde62bfSChris Mason if (space_args.space_slots == 0) { 29047fde62bfSChris Mason space_args.total_spaces = slot_count; 29057fde62bfSChris Mason goto out; 29067fde62bfSChris Mason } 2907bf5fc093SJosef Bacik 290851788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 2909bf5fc093SJosef Bacik 29107fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 2911bf5fc093SJosef Bacik 29127fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 29137fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 29147fde62bfSChris Mason */ 29157fde62bfSChris Mason if (alloc_size > PAGE_CACHE_SIZE) 29167fde62bfSChris Mason return -ENOMEM; 29177fde62bfSChris Mason 29187fde62bfSChris Mason space_args.total_spaces = 0; 29197fde62bfSChris Mason dest = kmalloc(alloc_size, GFP_NOFS); 29207fde62bfSChris Mason if (!dest) 29217fde62bfSChris Mason return -ENOMEM; 29227fde62bfSChris Mason dest_orig = dest; 29237fde62bfSChris Mason 29247fde62bfSChris Mason /* now we have a buffer to copy into */ 2925bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 2926bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 2927bf5fc093SJosef Bacik 292851788b1bSDan Rosenberg if (!slot_count) 292951788b1bSDan Rosenberg break; 293051788b1bSDan Rosenberg 2931bf5fc093SJosef Bacik info = NULL; 29321406e432SJosef Bacik rcu_read_lock(); 2933bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 2934bf5fc093SJosef Bacik list) { 2935bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 2936bf5fc093SJosef Bacik info = tmp; 29377fde62bfSChris Mason break; 2938bf5fc093SJosef Bacik } 2939bf5fc093SJosef Bacik } 2940bf5fc093SJosef Bacik rcu_read_unlock(); 29417fde62bfSChris Mason 2942bf5fc093SJosef Bacik if (!info) 2943bf5fc093SJosef Bacik continue; 2944bf5fc093SJosef Bacik down_read(&info->groups_sem); 2945bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 2946bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 2947bf5fc093SJosef Bacik get_block_group_info(&info->block_groups[c], 2948bf5fc093SJosef Bacik &space); 29497fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 29501406e432SJosef Bacik dest++; 29511406e432SJosef Bacik space_args.total_spaces++; 295251788b1bSDan Rosenberg slot_count--; 29531406e432SJosef Bacik } 295451788b1bSDan Rosenberg if (!slot_count) 295551788b1bSDan Rosenberg break; 2956bf5fc093SJosef Bacik } 2957bf5fc093SJosef Bacik up_read(&info->groups_sem); 2958bf5fc093SJosef Bacik } 29591406e432SJosef Bacik 29602eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 29617fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 29627fde62bfSChris Mason 29637fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 29647fde62bfSChris Mason ret = -EFAULT; 29657fde62bfSChris Mason 29667fde62bfSChris Mason kfree(dest_orig); 29677fde62bfSChris Mason out: 29687fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 29691406e432SJosef Bacik ret = -EFAULT; 29701406e432SJosef Bacik 29711406e432SJosef Bacik return ret; 29721406e432SJosef Bacik } 29731406e432SJosef Bacik 2974f46b5a66SChristoph Hellwig /* 2975f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 2976f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 2977f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 2978f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 2979f46b5a66SChristoph Hellwig */ 2980f46b5a66SChristoph Hellwig long btrfs_ioctl_trans_end(struct file *file) 2981f46b5a66SChristoph Hellwig { 2982f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2983f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 2984f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 2985f46b5a66SChristoph Hellwig 2986f46b5a66SChristoph Hellwig trans = file->private_data; 29871ab86aedSSage Weil if (!trans) 29881ab86aedSSage Weil return -EINVAL; 2989b214107eSChristoph Hellwig file->private_data = NULL; 29909ca9ee09SSage Weil 29911ab86aedSSage Weil btrfs_end_transaction(trans, root); 29921ab86aedSSage Weil 2993a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 29949ca9ee09SSage Weil 29952a79f17eSAl Viro mnt_drop_write_file(file); 29961ab86aedSSage Weil return 0; 2997f46b5a66SChristoph Hellwig } 2998f46b5a66SChristoph Hellwig 299946204592SSage Weil static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp) 300046204592SSage Weil { 300146204592SSage Weil struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root; 300246204592SSage Weil struct btrfs_trans_handle *trans; 300346204592SSage Weil u64 transid; 3004db5b493aSTsutomu Itoh int ret; 300546204592SSage Weil 300646204592SSage Weil trans = btrfs_start_transaction(root, 0); 300798d5dc13STsutomu Itoh if (IS_ERR(trans)) 300898d5dc13STsutomu Itoh return PTR_ERR(trans); 300946204592SSage Weil transid = trans->transid; 3010db5b493aSTsutomu Itoh ret = btrfs_commit_transaction_async(trans, root, 0); 30118b2b2d3cSTsutomu Itoh if (ret) { 30128b2b2d3cSTsutomu Itoh btrfs_end_transaction(trans, root); 3013db5b493aSTsutomu Itoh return ret; 30148b2b2d3cSTsutomu Itoh } 301546204592SSage Weil 301646204592SSage Weil if (argp) 301746204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 301846204592SSage Weil return -EFAULT; 301946204592SSage Weil return 0; 302046204592SSage Weil } 302146204592SSage Weil 302246204592SSage Weil static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp) 302346204592SSage Weil { 302446204592SSage Weil struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root; 302546204592SSage Weil u64 transid; 302646204592SSage Weil 302746204592SSage Weil if (argp) { 302846204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 302946204592SSage Weil return -EFAULT; 303046204592SSage Weil } else { 303146204592SSage Weil transid = 0; /* current trans */ 303246204592SSage Weil } 303346204592SSage Weil return btrfs_wait_for_commit(root, transid); 303446204592SSage Weil } 303546204592SSage Weil 3036475f6387SJan Schmidt static long btrfs_ioctl_scrub(struct btrfs_root *root, void __user *arg) 3037475f6387SJan Schmidt { 3038475f6387SJan Schmidt int ret; 3039475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3040475f6387SJan Schmidt 3041475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3042475f6387SJan Schmidt return -EPERM; 3043475f6387SJan Schmidt 3044475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3045475f6387SJan Schmidt if (IS_ERR(sa)) 3046475f6387SJan Schmidt return PTR_ERR(sa); 3047475f6387SJan Schmidt 3048475f6387SJan Schmidt ret = btrfs_scrub_dev(root, sa->devid, sa->start, sa->end, 30498628764eSArne Jansen &sa->progress, sa->flags & BTRFS_SCRUB_READONLY); 3050475f6387SJan Schmidt 3051475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 3052475f6387SJan Schmidt ret = -EFAULT; 3053475f6387SJan Schmidt 3054475f6387SJan Schmidt kfree(sa); 3055475f6387SJan Schmidt return ret; 3056475f6387SJan Schmidt } 3057475f6387SJan Schmidt 3058475f6387SJan Schmidt static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg) 3059475f6387SJan Schmidt { 3060475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3061475f6387SJan Schmidt return -EPERM; 3062475f6387SJan Schmidt 3063475f6387SJan Schmidt return btrfs_scrub_cancel(root); 3064475f6387SJan Schmidt } 3065475f6387SJan Schmidt 3066475f6387SJan Schmidt static long btrfs_ioctl_scrub_progress(struct btrfs_root *root, 3067475f6387SJan Schmidt void __user *arg) 3068475f6387SJan Schmidt { 3069475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3070475f6387SJan Schmidt int ret; 3071475f6387SJan Schmidt 3072475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3073475f6387SJan Schmidt return -EPERM; 3074475f6387SJan Schmidt 3075475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3076475f6387SJan Schmidt if (IS_ERR(sa)) 3077475f6387SJan Schmidt return PTR_ERR(sa); 3078475f6387SJan Schmidt 3079475f6387SJan Schmidt ret = btrfs_scrub_progress(root, sa->devid, &sa->progress); 3080475f6387SJan Schmidt 3081475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 3082475f6387SJan Schmidt ret = -EFAULT; 3083475f6387SJan Schmidt 3084475f6387SJan Schmidt kfree(sa); 3085475f6387SJan Schmidt return ret; 3086475f6387SJan Schmidt } 3087475f6387SJan Schmidt 3088c11d2c23SStefan Behrens static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root, 3089b27f7c0cSDavid Sterba void __user *arg) 3090c11d2c23SStefan Behrens { 3091c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 3092c11d2c23SStefan Behrens int ret; 3093c11d2c23SStefan Behrens 3094c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 3095c11d2c23SStefan Behrens if (IS_ERR(sa)) 3096c11d2c23SStefan Behrens return PTR_ERR(sa); 3097c11d2c23SStefan Behrens 3098b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 3099b27f7c0cSDavid Sterba kfree(sa); 3100b27f7c0cSDavid Sterba return -EPERM; 3101b27f7c0cSDavid Sterba } 3102b27f7c0cSDavid Sterba 3103b27f7c0cSDavid Sterba ret = btrfs_get_dev_stats(root, sa); 3104c11d2c23SStefan Behrens 3105c11d2c23SStefan Behrens if (copy_to_user(arg, sa, sizeof(*sa))) 3106c11d2c23SStefan Behrens ret = -EFAULT; 3107c11d2c23SStefan Behrens 3108c11d2c23SStefan Behrens kfree(sa); 3109c11d2c23SStefan Behrens return ret; 3110c11d2c23SStefan Behrens } 3111c11d2c23SStefan Behrens 3112d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 3113d7728c96SJan Schmidt { 3114d7728c96SJan Schmidt int ret = 0; 3115d7728c96SJan Schmidt int i; 3116740c3d22SChris Mason u64 rel_ptr; 3117d7728c96SJan Schmidt int size; 3118806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 3119d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 3120d7728c96SJan Schmidt struct btrfs_path *path; 3121d7728c96SJan Schmidt 3122d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3123d7728c96SJan Schmidt return -EPERM; 3124d7728c96SJan Schmidt 3125d7728c96SJan Schmidt path = btrfs_alloc_path(); 3126d7728c96SJan Schmidt if (!path) { 3127d7728c96SJan Schmidt ret = -ENOMEM; 3128d7728c96SJan Schmidt goto out; 3129d7728c96SJan Schmidt } 3130d7728c96SJan Schmidt 3131d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 3132d7728c96SJan Schmidt if (IS_ERR(ipa)) { 3133d7728c96SJan Schmidt ret = PTR_ERR(ipa); 3134d7728c96SJan Schmidt ipa = NULL; 3135d7728c96SJan Schmidt goto out; 3136d7728c96SJan Schmidt } 3137d7728c96SJan Schmidt 3138d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 3139d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 3140d7728c96SJan Schmidt if (IS_ERR(ipath)) { 3141d7728c96SJan Schmidt ret = PTR_ERR(ipath); 3142d7728c96SJan Schmidt ipath = NULL; 3143d7728c96SJan Schmidt goto out; 3144d7728c96SJan Schmidt } 3145d7728c96SJan Schmidt 3146d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 3147d7728c96SJan Schmidt if (ret < 0) 3148d7728c96SJan Schmidt goto out; 3149d7728c96SJan Schmidt 3150d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 3151745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 3152745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 3153740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 3154d7728c96SJan Schmidt } 3155d7728c96SJan Schmidt 3156745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)ipa->fspath, 3157745c4d8eSJeff Mahoney (void *)(unsigned long)ipath->fspath, size); 3158d7728c96SJan Schmidt if (ret) { 3159d7728c96SJan Schmidt ret = -EFAULT; 3160d7728c96SJan Schmidt goto out; 3161d7728c96SJan Schmidt } 3162d7728c96SJan Schmidt 3163d7728c96SJan Schmidt out: 3164d7728c96SJan Schmidt btrfs_free_path(path); 3165d7728c96SJan Schmidt free_ipath(ipath); 3166d7728c96SJan Schmidt kfree(ipa); 3167d7728c96SJan Schmidt 3168d7728c96SJan Schmidt return ret; 3169d7728c96SJan Schmidt } 3170d7728c96SJan Schmidt 3171d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 3172d7728c96SJan Schmidt { 3173d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 3174d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 3175d7728c96SJan Schmidt 3176d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 3177d7728c96SJan Schmidt inodes->bytes_left -= c; 3178d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 3179d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 3180d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 3181d7728c96SJan Schmidt inodes->elem_cnt += 3; 3182d7728c96SJan Schmidt } else { 3183d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 3184d7728c96SJan Schmidt inodes->bytes_left = 0; 3185d7728c96SJan Schmidt inodes->elem_missed += 3; 3186d7728c96SJan Schmidt } 3187d7728c96SJan Schmidt 3188d7728c96SJan Schmidt return 0; 3189d7728c96SJan Schmidt } 3190d7728c96SJan Schmidt 3191d7728c96SJan Schmidt static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root, 3192d7728c96SJan Schmidt void __user *arg) 3193d7728c96SJan Schmidt { 3194d7728c96SJan Schmidt int ret = 0; 3195d7728c96SJan Schmidt int size; 31964692cf58SJan Schmidt u64 extent_item_pos; 3197d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 3198d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 3199d7728c96SJan Schmidt struct btrfs_path *path = NULL; 3200d7728c96SJan Schmidt struct btrfs_key key; 3201d7728c96SJan Schmidt 3202d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3203d7728c96SJan Schmidt return -EPERM; 3204d7728c96SJan Schmidt 3205d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 3206d7728c96SJan Schmidt if (IS_ERR(loi)) { 3207d7728c96SJan Schmidt ret = PTR_ERR(loi); 3208d7728c96SJan Schmidt loi = NULL; 3209d7728c96SJan Schmidt goto out; 3210d7728c96SJan Schmidt } 3211d7728c96SJan Schmidt 3212d7728c96SJan Schmidt path = btrfs_alloc_path(); 3213d7728c96SJan Schmidt if (!path) { 3214d7728c96SJan Schmidt ret = -ENOMEM; 3215d7728c96SJan Schmidt goto out; 3216d7728c96SJan Schmidt } 3217d7728c96SJan Schmidt 3218d7728c96SJan Schmidt size = min_t(u32, loi->size, 4096); 3219d7728c96SJan Schmidt inodes = init_data_container(size); 3220d7728c96SJan Schmidt if (IS_ERR(inodes)) { 3221d7728c96SJan Schmidt ret = PTR_ERR(inodes); 3222d7728c96SJan Schmidt inodes = NULL; 3223d7728c96SJan Schmidt goto out; 3224d7728c96SJan Schmidt } 3225d7728c96SJan Schmidt 3226d7728c96SJan Schmidt ret = extent_from_logical(root->fs_info, loi->logical, path, &key); 32274692cf58SJan Schmidt btrfs_release_path(path); 3228d7728c96SJan Schmidt 3229d7728c96SJan Schmidt if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) 3230d7728c96SJan Schmidt ret = -ENOENT; 3231d7728c96SJan Schmidt if (ret < 0) 3232d7728c96SJan Schmidt goto out; 3233d7728c96SJan Schmidt 32344692cf58SJan Schmidt extent_item_pos = loi->logical - key.objectid; 32357a3ae2f8SJan Schmidt ret = iterate_extent_inodes(root->fs_info, key.objectid, 32367a3ae2f8SJan Schmidt extent_item_pos, 0, build_ino_list, 32374692cf58SJan Schmidt inodes); 3238d7728c96SJan Schmidt 3239d7728c96SJan Schmidt if (ret < 0) 3240d7728c96SJan Schmidt goto out; 3241d7728c96SJan Schmidt 3242745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)loi->inodes, 3243745c4d8eSJeff Mahoney (void *)(unsigned long)inodes, size); 3244d7728c96SJan Schmidt if (ret) 3245d7728c96SJan Schmidt ret = -EFAULT; 3246d7728c96SJan Schmidt 3247d7728c96SJan Schmidt out: 3248d7728c96SJan Schmidt btrfs_free_path(path); 3249d7728c96SJan Schmidt kfree(inodes); 3250d7728c96SJan Schmidt kfree(loi); 3251d7728c96SJan Schmidt 3252d7728c96SJan Schmidt return ret; 3253d7728c96SJan Schmidt } 3254d7728c96SJan Schmidt 325519a39dceSIlya Dryomov void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, 3256c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 3257c9e9f97bSIlya Dryomov { 3258c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3259c9e9f97bSIlya Dryomov 3260c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 3261c9e9f97bSIlya Dryomov 3262837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_running)) 3263837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 3264837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 3265837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 3266a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 3267a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 3268837d5b6eSIlya Dryomov 3269c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 3270c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 3271c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 327219a39dceSIlya Dryomov 327319a39dceSIlya Dryomov if (lock) { 327419a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 327519a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 327619a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 327719a39dceSIlya Dryomov } else { 327819a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 327919a39dceSIlya Dryomov } 3280c9e9f97bSIlya Dryomov } 3281c9e9f97bSIlya Dryomov 32829ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 3283c9e9f97bSIlya Dryomov { 32849ba1f6e4SLiu Bo struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 3285c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 3286c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 3287c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 3288c9e9f97bSIlya Dryomov int ret; 3289c9e9f97bSIlya Dryomov 3290c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 3291c9e9f97bSIlya Dryomov return -EPERM; 3292c9e9f97bSIlya Dryomov 3293e54bfa31SLiu Bo ret = mnt_want_write_file(file); 32949ba1f6e4SLiu Bo if (ret) 32959ba1f6e4SLiu Bo return ret; 32969ba1f6e4SLiu Bo 3297c9e9f97bSIlya Dryomov mutex_lock(&fs_info->volume_mutex); 3298c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 3299c9e9f97bSIlya Dryomov 3300c9e9f97bSIlya Dryomov if (arg) { 3301c9e9f97bSIlya Dryomov bargs = memdup_user(arg, sizeof(*bargs)); 3302c9e9f97bSIlya Dryomov if (IS_ERR(bargs)) { 3303c9e9f97bSIlya Dryomov ret = PTR_ERR(bargs); 3304c9e9f97bSIlya Dryomov goto out; 3305c9e9f97bSIlya Dryomov } 3306de322263SIlya Dryomov 3307de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 3308de322263SIlya Dryomov if (!fs_info->balance_ctl) { 3309de322263SIlya Dryomov ret = -ENOTCONN; 3310de322263SIlya Dryomov goto out_bargs; 3311de322263SIlya Dryomov } 3312de322263SIlya Dryomov 3313de322263SIlya Dryomov bctl = fs_info->balance_ctl; 3314de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 3315de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 3316de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 3317de322263SIlya Dryomov 3318de322263SIlya Dryomov goto do_balance; 3319de322263SIlya Dryomov } 3320c9e9f97bSIlya Dryomov } else { 3321c9e9f97bSIlya Dryomov bargs = NULL; 3322c9e9f97bSIlya Dryomov } 3323c9e9f97bSIlya Dryomov 3324837d5b6eSIlya Dryomov if (fs_info->balance_ctl) { 3325837d5b6eSIlya Dryomov ret = -EINPROGRESS; 3326837d5b6eSIlya Dryomov goto out_bargs; 3327837d5b6eSIlya Dryomov } 3328837d5b6eSIlya Dryomov 3329c9e9f97bSIlya Dryomov bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 3330c9e9f97bSIlya Dryomov if (!bctl) { 3331c9e9f97bSIlya Dryomov ret = -ENOMEM; 3332c9e9f97bSIlya Dryomov goto out_bargs; 3333c9e9f97bSIlya Dryomov } 3334c9e9f97bSIlya Dryomov 3335c9e9f97bSIlya Dryomov bctl->fs_info = fs_info; 3336c9e9f97bSIlya Dryomov if (arg) { 3337c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 3338c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 3339c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 3340c9e9f97bSIlya Dryomov 3341c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 3342f43ffb60SIlya Dryomov } else { 3343f43ffb60SIlya Dryomov /* balance everything - no filters */ 3344f43ffb60SIlya Dryomov bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 3345c9e9f97bSIlya Dryomov } 3346c9e9f97bSIlya Dryomov 3347de322263SIlya Dryomov do_balance: 3348c9e9f97bSIlya Dryomov ret = btrfs_balance(bctl, bargs); 3349c9e9f97bSIlya Dryomov /* 3350837d5b6eSIlya Dryomov * bctl is freed in __cancel_balance or in free_fs_info if 3351837d5b6eSIlya Dryomov * restriper was paused all the way until unmount 3352c9e9f97bSIlya Dryomov */ 3353c9e9f97bSIlya Dryomov if (arg) { 3354c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 3355c9e9f97bSIlya Dryomov ret = -EFAULT; 3356c9e9f97bSIlya Dryomov } 3357c9e9f97bSIlya Dryomov 3358c9e9f97bSIlya Dryomov out_bargs: 3359c9e9f97bSIlya Dryomov kfree(bargs); 3360c9e9f97bSIlya Dryomov out: 3361c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 3362c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->volume_mutex); 3363e54bfa31SLiu Bo mnt_drop_write_file(file); 3364c9e9f97bSIlya Dryomov return ret; 3365c9e9f97bSIlya Dryomov } 3366c9e9f97bSIlya Dryomov 3367837d5b6eSIlya Dryomov static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd) 3368837d5b6eSIlya Dryomov { 3369837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 3370837d5b6eSIlya Dryomov return -EPERM; 3371837d5b6eSIlya Dryomov 3372837d5b6eSIlya Dryomov switch (cmd) { 3373837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 3374837d5b6eSIlya Dryomov return btrfs_pause_balance(root->fs_info); 3375a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 3376a7e99c69SIlya Dryomov return btrfs_cancel_balance(root->fs_info); 3377837d5b6eSIlya Dryomov } 3378837d5b6eSIlya Dryomov 3379837d5b6eSIlya Dryomov return -EINVAL; 3380837d5b6eSIlya Dryomov } 3381837d5b6eSIlya Dryomov 338219a39dceSIlya Dryomov static long btrfs_ioctl_balance_progress(struct btrfs_root *root, 338319a39dceSIlya Dryomov void __user *arg) 338419a39dceSIlya Dryomov { 338519a39dceSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 338619a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 338719a39dceSIlya Dryomov int ret = 0; 338819a39dceSIlya Dryomov 338919a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 339019a39dceSIlya Dryomov return -EPERM; 339119a39dceSIlya Dryomov 339219a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 339319a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 339419a39dceSIlya Dryomov ret = -ENOTCONN; 339519a39dceSIlya Dryomov goto out; 339619a39dceSIlya Dryomov } 339719a39dceSIlya Dryomov 339819a39dceSIlya Dryomov bargs = kzalloc(sizeof(*bargs), GFP_NOFS); 339919a39dceSIlya Dryomov if (!bargs) { 340019a39dceSIlya Dryomov ret = -ENOMEM; 340119a39dceSIlya Dryomov goto out; 340219a39dceSIlya Dryomov } 340319a39dceSIlya Dryomov 340419a39dceSIlya Dryomov update_ioctl_balance_args(fs_info, 1, bargs); 340519a39dceSIlya Dryomov 340619a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 340719a39dceSIlya Dryomov ret = -EFAULT; 340819a39dceSIlya Dryomov 340919a39dceSIlya Dryomov kfree(bargs); 341019a39dceSIlya Dryomov out: 341119a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 341219a39dceSIlya Dryomov return ret; 341319a39dceSIlya Dryomov } 341419a39dceSIlya Dryomov 34155d13a37bSArne Jansen static long btrfs_ioctl_quota_ctl(struct btrfs_root *root, void __user *arg) 34165d13a37bSArne Jansen { 34175d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 34185d13a37bSArne Jansen struct btrfs_trans_handle *trans = NULL; 34195d13a37bSArne Jansen int ret; 34205d13a37bSArne Jansen int err; 34215d13a37bSArne Jansen 34225d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 34235d13a37bSArne Jansen return -EPERM; 34245d13a37bSArne Jansen 34255d13a37bSArne Jansen if (root->fs_info->sb->s_flags & MS_RDONLY) 34265d13a37bSArne Jansen return -EROFS; 34275d13a37bSArne Jansen 34285d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 34295d13a37bSArne Jansen if (IS_ERR(sa)) 34305d13a37bSArne Jansen return PTR_ERR(sa); 34315d13a37bSArne Jansen 34325d13a37bSArne Jansen if (sa->cmd != BTRFS_QUOTA_CTL_RESCAN) { 34335d13a37bSArne Jansen trans = btrfs_start_transaction(root, 2); 34345d13a37bSArne Jansen if (IS_ERR(trans)) { 34355d13a37bSArne Jansen ret = PTR_ERR(trans); 34365d13a37bSArne Jansen goto out; 34375d13a37bSArne Jansen } 34385d13a37bSArne Jansen } 34395d13a37bSArne Jansen 34405d13a37bSArne Jansen switch (sa->cmd) { 34415d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 34425d13a37bSArne Jansen ret = btrfs_quota_enable(trans, root->fs_info); 34435d13a37bSArne Jansen break; 34445d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 34455d13a37bSArne Jansen ret = btrfs_quota_disable(trans, root->fs_info); 34465d13a37bSArne Jansen break; 34475d13a37bSArne Jansen case BTRFS_QUOTA_CTL_RESCAN: 34485d13a37bSArne Jansen ret = btrfs_quota_rescan(root->fs_info); 34495d13a37bSArne Jansen break; 34505d13a37bSArne Jansen default: 34515d13a37bSArne Jansen ret = -EINVAL; 34525d13a37bSArne Jansen break; 34535d13a37bSArne Jansen } 34545d13a37bSArne Jansen 34555d13a37bSArne Jansen if (copy_to_user(arg, sa, sizeof(*sa))) 34565d13a37bSArne Jansen ret = -EFAULT; 34575d13a37bSArne Jansen 34585d13a37bSArne Jansen if (trans) { 34595d13a37bSArne Jansen err = btrfs_commit_transaction(trans, root); 34605d13a37bSArne Jansen if (err && !ret) 34615d13a37bSArne Jansen ret = err; 34625d13a37bSArne Jansen } 34635d13a37bSArne Jansen 34645d13a37bSArne Jansen out: 34655d13a37bSArne Jansen kfree(sa); 34665d13a37bSArne Jansen return ret; 34675d13a37bSArne Jansen } 34685d13a37bSArne Jansen 34695d13a37bSArne Jansen static long btrfs_ioctl_qgroup_assign(struct btrfs_root *root, void __user *arg) 34705d13a37bSArne Jansen { 34715d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 34725d13a37bSArne Jansen struct btrfs_trans_handle *trans; 34735d13a37bSArne Jansen int ret; 34745d13a37bSArne Jansen int err; 34755d13a37bSArne Jansen 34765d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 34775d13a37bSArne Jansen return -EPERM; 34785d13a37bSArne Jansen 34795d13a37bSArne Jansen if (root->fs_info->sb->s_flags & MS_RDONLY) 34805d13a37bSArne Jansen return -EROFS; 34815d13a37bSArne Jansen 34825d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 34835d13a37bSArne Jansen if (IS_ERR(sa)) 34845d13a37bSArne Jansen return PTR_ERR(sa); 34855d13a37bSArne Jansen 34865d13a37bSArne Jansen trans = btrfs_join_transaction(root); 34875d13a37bSArne Jansen if (IS_ERR(trans)) { 34885d13a37bSArne Jansen ret = PTR_ERR(trans); 34895d13a37bSArne Jansen goto out; 34905d13a37bSArne Jansen } 34915d13a37bSArne Jansen 34925d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 34935d13a37bSArne Jansen if (sa->assign) { 34945d13a37bSArne Jansen ret = btrfs_add_qgroup_relation(trans, root->fs_info, 34955d13a37bSArne Jansen sa->src, sa->dst); 34965d13a37bSArne Jansen } else { 34975d13a37bSArne Jansen ret = btrfs_del_qgroup_relation(trans, root->fs_info, 34985d13a37bSArne Jansen sa->src, sa->dst); 34995d13a37bSArne Jansen } 35005d13a37bSArne Jansen 35015d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 35025d13a37bSArne Jansen if (err && !ret) 35035d13a37bSArne Jansen ret = err; 35045d13a37bSArne Jansen 35055d13a37bSArne Jansen out: 35065d13a37bSArne Jansen kfree(sa); 35075d13a37bSArne Jansen return ret; 35085d13a37bSArne Jansen } 35095d13a37bSArne Jansen 35105d13a37bSArne Jansen static long btrfs_ioctl_qgroup_create(struct btrfs_root *root, void __user *arg) 35115d13a37bSArne Jansen { 35125d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 35135d13a37bSArne Jansen struct btrfs_trans_handle *trans; 35145d13a37bSArne Jansen int ret; 35155d13a37bSArne Jansen int err; 35165d13a37bSArne Jansen 35175d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 35185d13a37bSArne Jansen return -EPERM; 35195d13a37bSArne Jansen 35205d13a37bSArne Jansen if (root->fs_info->sb->s_flags & MS_RDONLY) 35215d13a37bSArne Jansen return -EROFS; 35225d13a37bSArne Jansen 35235d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 35245d13a37bSArne Jansen if (IS_ERR(sa)) 35255d13a37bSArne Jansen return PTR_ERR(sa); 35265d13a37bSArne Jansen 35275d13a37bSArne Jansen trans = btrfs_join_transaction(root); 35285d13a37bSArne Jansen if (IS_ERR(trans)) { 35295d13a37bSArne Jansen ret = PTR_ERR(trans); 35305d13a37bSArne Jansen goto out; 35315d13a37bSArne Jansen } 35325d13a37bSArne Jansen 35335d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 35345d13a37bSArne Jansen if (sa->create) { 35355d13a37bSArne Jansen ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid, 35365d13a37bSArne Jansen NULL); 35375d13a37bSArne Jansen } else { 35385d13a37bSArne Jansen ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid); 35395d13a37bSArne Jansen } 35405d13a37bSArne Jansen 35415d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 35425d13a37bSArne Jansen if (err && !ret) 35435d13a37bSArne Jansen ret = err; 35445d13a37bSArne Jansen 35455d13a37bSArne Jansen out: 35465d13a37bSArne Jansen kfree(sa); 35475d13a37bSArne Jansen return ret; 35485d13a37bSArne Jansen } 35495d13a37bSArne Jansen 35505d13a37bSArne Jansen static long btrfs_ioctl_qgroup_limit(struct btrfs_root *root, void __user *arg) 35515d13a37bSArne Jansen { 35525d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 35535d13a37bSArne Jansen struct btrfs_trans_handle *trans; 35545d13a37bSArne Jansen int ret; 35555d13a37bSArne Jansen int err; 35565d13a37bSArne Jansen u64 qgroupid; 35575d13a37bSArne Jansen 35585d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 35595d13a37bSArne Jansen return -EPERM; 35605d13a37bSArne Jansen 35615d13a37bSArne Jansen if (root->fs_info->sb->s_flags & MS_RDONLY) 35625d13a37bSArne Jansen return -EROFS; 35635d13a37bSArne Jansen 35645d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 35655d13a37bSArne Jansen if (IS_ERR(sa)) 35665d13a37bSArne Jansen return PTR_ERR(sa); 35675d13a37bSArne Jansen 35685d13a37bSArne Jansen trans = btrfs_join_transaction(root); 35695d13a37bSArne Jansen if (IS_ERR(trans)) { 35705d13a37bSArne Jansen ret = PTR_ERR(trans); 35715d13a37bSArne Jansen goto out; 35725d13a37bSArne Jansen } 35735d13a37bSArne Jansen 35745d13a37bSArne Jansen qgroupid = sa->qgroupid; 35755d13a37bSArne Jansen if (!qgroupid) { 35765d13a37bSArne Jansen /* take the current subvol as qgroup */ 35775d13a37bSArne Jansen qgroupid = root->root_key.objectid; 35785d13a37bSArne Jansen } 35795d13a37bSArne Jansen 35805d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 35815d13a37bSArne Jansen ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim); 35825d13a37bSArne Jansen 35835d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 35845d13a37bSArne Jansen if (err && !ret) 35855d13a37bSArne Jansen ret = err; 35865d13a37bSArne Jansen 35875d13a37bSArne Jansen out: 35885d13a37bSArne Jansen kfree(sa); 35895d13a37bSArne Jansen return ret; 35905d13a37bSArne Jansen } 35915d13a37bSArne Jansen 3592f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 3593f46b5a66SChristoph Hellwig cmd, unsigned long arg) 3594f46b5a66SChristoph Hellwig { 3595f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 35964bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 3597f46b5a66SChristoph Hellwig 3598f46b5a66SChristoph Hellwig switch (cmd) { 35996cbff00fSChristoph Hellwig case FS_IOC_GETFLAGS: 36006cbff00fSChristoph Hellwig return btrfs_ioctl_getflags(file, argp); 36016cbff00fSChristoph Hellwig case FS_IOC_SETFLAGS: 36026cbff00fSChristoph Hellwig return btrfs_ioctl_setflags(file, argp); 36036cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 36046cbff00fSChristoph Hellwig return btrfs_ioctl_getversion(file, argp); 3605f7039b1dSLi Dongyang case FITRIM: 3606f7039b1dSLi Dongyang return btrfs_ioctl_fitrim(file, argp); 3607f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 3608fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 3609fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 3610fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 36113de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 3612fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 36136f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 36146f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 361576dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 361676dda93cSYan, Zheng return btrfs_ioctl_snap_destroy(file, argp); 36170caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 36180caa102dSLi Zefan return btrfs_ioctl_subvol_getflags(file, argp); 36190caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 36200caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 36216ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 36226ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 3623f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 36241e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 36251e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 36261e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 3627f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 36284bcabaa3SChristoph Hellwig return btrfs_ioctl_resize(root, argp); 3629f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 36304bcabaa3SChristoph Hellwig return btrfs_ioctl_add_dev(root, argp); 3631f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 36324bcabaa3SChristoph Hellwig return btrfs_ioctl_rm_dev(root, argp); 3633475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 3634475f6387SJan Schmidt return btrfs_ioctl_fs_info(root, argp); 3635475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 3636475f6387SJan Schmidt return btrfs_ioctl_dev_info(root, argp); 3637f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 36389ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, NULL); 3639f46b5a66SChristoph Hellwig case BTRFS_IOC_CLONE: 3640c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, arg, 0, 0, 0); 3641c5c9cd4dSSage Weil case BTRFS_IOC_CLONE_RANGE: 36427a865e8aSChristoph Hellwig return btrfs_ioctl_clone_range(file, argp); 3643f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_START: 3644f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_start(file); 3645f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_END: 3646f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_end(file); 3647ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 3648ac8e9819SChris Mason return btrfs_ioctl_tree_search(file, argp); 3649ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 3650ac8e9819SChris Mason return btrfs_ioctl_ino_lookup(file, argp); 3651d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 3652d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 3653d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 3654d7728c96SJan Schmidt return btrfs_ioctl_logical_to_ino(root, argp); 36551406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 36561406e432SJosef Bacik return btrfs_ioctl_space_info(root, argp); 3657f46b5a66SChristoph Hellwig case BTRFS_IOC_SYNC: 3658f46b5a66SChristoph Hellwig btrfs_sync_fs(file->f_dentry->d_sb, 1); 3659f46b5a66SChristoph Hellwig return 0; 366046204592SSage Weil case BTRFS_IOC_START_SYNC: 366146204592SSage Weil return btrfs_ioctl_start_sync(file, argp); 366246204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 366346204592SSage Weil return btrfs_ioctl_wait_sync(file, argp); 3664475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 3665475f6387SJan Schmidt return btrfs_ioctl_scrub(root, argp); 3666475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 3667475f6387SJan Schmidt return btrfs_ioctl_scrub_cancel(root, argp); 3668475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 3669475f6387SJan Schmidt return btrfs_ioctl_scrub_progress(root, argp); 3670c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 36719ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 3672837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 3673837d5b6eSIlya Dryomov return btrfs_ioctl_balance_ctl(root, arg); 367419a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 367519a39dceSIlya Dryomov return btrfs_ioctl_balance_progress(root, argp); 3676c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 3677b27f7c0cSDavid Sterba return btrfs_ioctl_get_dev_stats(root, argp); 36785d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 36795d13a37bSArne Jansen return btrfs_ioctl_quota_ctl(root, argp); 36805d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 36815d13a37bSArne Jansen return btrfs_ioctl_qgroup_assign(root, argp); 36825d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 36835d13a37bSArne Jansen return btrfs_ioctl_qgroup_create(root, argp); 36845d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 36855d13a37bSArne Jansen return btrfs_ioctl_qgroup_limit(root, argp); 3686f46b5a66SChristoph Hellwig } 3687f46b5a66SChristoph Hellwig 3688f46b5a66SChristoph Hellwig return -ENOTTY; 3689f46b5a66SChristoph Hellwig } 3690