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" 53f46b5a66SChristoph Hellwig 546cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 556cbff00fSChristoph Hellwig static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags) 566cbff00fSChristoph Hellwig { 576cbff00fSChristoph Hellwig if (S_ISDIR(mode)) 586cbff00fSChristoph Hellwig return flags; 596cbff00fSChristoph Hellwig else if (S_ISREG(mode)) 606cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 616cbff00fSChristoph Hellwig else 626cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 636cbff00fSChristoph Hellwig } 64f46b5a66SChristoph Hellwig 656cbff00fSChristoph Hellwig /* 666cbff00fSChristoph Hellwig * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl. 676cbff00fSChristoph Hellwig */ 686cbff00fSChristoph Hellwig static unsigned int btrfs_flags_to_ioctl(unsigned int flags) 696cbff00fSChristoph Hellwig { 706cbff00fSChristoph Hellwig unsigned int iflags = 0; 716cbff00fSChristoph Hellwig 726cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 736cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 746cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 756cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 766cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 776cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 786cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 796cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 806cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 816cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 826cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 836cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 84d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 85d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 86d0092bddSLi Zefan 87d0092bddSLi Zefan if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS)) 88d0092bddSLi Zefan iflags |= FS_COMPR_FL; 89d0092bddSLi Zefan else if (flags & BTRFS_INODE_NOCOMPRESS) 90d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 916cbff00fSChristoph Hellwig 926cbff00fSChristoph Hellwig return iflags; 936cbff00fSChristoph Hellwig } 946cbff00fSChristoph Hellwig 956cbff00fSChristoph Hellwig /* 966cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 976cbff00fSChristoph Hellwig */ 986cbff00fSChristoph Hellwig void btrfs_update_iflags(struct inode *inode) 996cbff00fSChristoph Hellwig { 1006cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1016cbff00fSChristoph Hellwig 1026cbff00fSChristoph Hellwig inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); 1036cbff00fSChristoph Hellwig 1046cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_SYNC) 1056cbff00fSChristoph Hellwig inode->i_flags |= S_SYNC; 1066cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_IMMUTABLE) 1076cbff00fSChristoph Hellwig inode->i_flags |= S_IMMUTABLE; 1086cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_APPEND) 1096cbff00fSChristoph Hellwig inode->i_flags |= S_APPEND; 1106cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_NOATIME) 1116cbff00fSChristoph Hellwig inode->i_flags |= S_NOATIME; 1126cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_DIRSYNC) 1136cbff00fSChristoph Hellwig inode->i_flags |= S_DIRSYNC; 1146cbff00fSChristoph Hellwig } 1156cbff00fSChristoph Hellwig 1166cbff00fSChristoph Hellwig /* 1176cbff00fSChristoph Hellwig * Inherit flags from the parent inode. 1186cbff00fSChristoph Hellwig * 1196cbff00fSChristoph Hellwig * Unlike extN we don't have any flags we don't want to inherit currently. 1206cbff00fSChristoph Hellwig */ 1216cbff00fSChristoph Hellwig void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) 1226cbff00fSChristoph Hellwig { 1230b4dcea5SChris Mason unsigned int flags; 1240b4dcea5SChris Mason 1250b4dcea5SChris Mason if (!dir) 1260b4dcea5SChris Mason return; 1270b4dcea5SChris Mason 1280b4dcea5SChris Mason flags = BTRFS_I(dir)->flags; 1296cbff00fSChristoph Hellwig 1306cbff00fSChristoph Hellwig if (S_ISREG(inode->i_mode)) 1316cbff00fSChristoph Hellwig flags &= ~BTRFS_INODE_DIRSYNC; 1326cbff00fSChristoph Hellwig else if (!S_ISDIR(inode->i_mode)) 1336cbff00fSChristoph Hellwig flags &= (BTRFS_INODE_NODUMP | BTRFS_INODE_NOATIME); 1346cbff00fSChristoph Hellwig 1356cbff00fSChristoph Hellwig BTRFS_I(inode)->flags = flags; 1366cbff00fSChristoph Hellwig btrfs_update_iflags(inode); 1376cbff00fSChristoph Hellwig } 1386cbff00fSChristoph Hellwig 1396cbff00fSChristoph Hellwig static int btrfs_ioctl_getflags(struct file *file, void __user *arg) 1406cbff00fSChristoph Hellwig { 1416cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(file->f_path.dentry->d_inode); 1426cbff00fSChristoph Hellwig unsigned int flags = btrfs_flags_to_ioctl(ip->flags); 1436cbff00fSChristoph Hellwig 1446cbff00fSChristoph Hellwig if (copy_to_user(arg, &flags, sizeof(flags))) 1456cbff00fSChristoph Hellwig return -EFAULT; 1466cbff00fSChristoph Hellwig return 0; 1476cbff00fSChristoph Hellwig } 1486cbff00fSChristoph Hellwig 14975e7cb7fSLiu Bo static int check_flags(unsigned int flags) 15075e7cb7fSLiu Bo { 15175e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 15275e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 15375e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 154e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 155e1e8fb6aSLi Zefan FS_NOCOW_FL)) 15675e7cb7fSLiu Bo return -EOPNOTSUPP; 15775e7cb7fSLiu Bo 15875e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 15975e7cb7fSLiu Bo return -EINVAL; 16075e7cb7fSLiu Bo 16175e7cb7fSLiu Bo return 0; 16275e7cb7fSLiu Bo } 16375e7cb7fSLiu Bo 1646cbff00fSChristoph Hellwig static int btrfs_ioctl_setflags(struct file *file, void __user *arg) 1656cbff00fSChristoph Hellwig { 1666cbff00fSChristoph Hellwig struct inode *inode = file->f_path.dentry->d_inode; 1676cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1686cbff00fSChristoph Hellwig struct btrfs_root *root = ip->root; 1696cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 1706cbff00fSChristoph Hellwig unsigned int flags, oldflags; 1716cbff00fSChristoph Hellwig int ret; 1726cbff00fSChristoph Hellwig 173b83cc969SLi Zefan if (btrfs_root_readonly(root)) 174b83cc969SLi Zefan return -EROFS; 175b83cc969SLi Zefan 1766cbff00fSChristoph Hellwig if (copy_from_user(&flags, arg, sizeof(flags))) 1776cbff00fSChristoph Hellwig return -EFAULT; 1786cbff00fSChristoph Hellwig 17975e7cb7fSLiu Bo ret = check_flags(flags); 18075e7cb7fSLiu Bo if (ret) 18175e7cb7fSLiu Bo return ret; 1826cbff00fSChristoph Hellwig 1832e149670SSerge E. Hallyn if (!inode_owner_or_capable(inode)) 1846cbff00fSChristoph Hellwig return -EACCES; 1856cbff00fSChristoph Hellwig 1866cbff00fSChristoph Hellwig mutex_lock(&inode->i_mutex); 1876cbff00fSChristoph Hellwig 1886cbff00fSChristoph Hellwig flags = btrfs_mask_flags(inode->i_mode, flags); 1896cbff00fSChristoph Hellwig oldflags = btrfs_flags_to_ioctl(ip->flags); 1906cbff00fSChristoph Hellwig if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 1916cbff00fSChristoph Hellwig if (!capable(CAP_LINUX_IMMUTABLE)) { 1926cbff00fSChristoph Hellwig ret = -EPERM; 1936cbff00fSChristoph Hellwig goto out_unlock; 1946cbff00fSChristoph Hellwig } 1956cbff00fSChristoph Hellwig } 1966cbff00fSChristoph Hellwig 1976cbff00fSChristoph Hellwig ret = mnt_want_write(file->f_path.mnt); 1986cbff00fSChristoph Hellwig if (ret) 1996cbff00fSChristoph Hellwig goto out_unlock; 2006cbff00fSChristoph Hellwig 2016cbff00fSChristoph Hellwig if (flags & FS_SYNC_FL) 2026cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_SYNC; 2036cbff00fSChristoph Hellwig else 2046cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_SYNC; 2056cbff00fSChristoph Hellwig if (flags & FS_IMMUTABLE_FL) 2066cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_IMMUTABLE; 2076cbff00fSChristoph Hellwig else 2086cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_IMMUTABLE; 2096cbff00fSChristoph Hellwig if (flags & FS_APPEND_FL) 2106cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_APPEND; 2116cbff00fSChristoph Hellwig else 2126cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_APPEND; 2136cbff00fSChristoph Hellwig if (flags & FS_NODUMP_FL) 2146cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NODUMP; 2156cbff00fSChristoph Hellwig else 2166cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NODUMP; 2176cbff00fSChristoph Hellwig if (flags & FS_NOATIME_FL) 2186cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NOATIME; 2196cbff00fSChristoph Hellwig else 2206cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NOATIME; 2216cbff00fSChristoph Hellwig if (flags & FS_DIRSYNC_FL) 2226cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_DIRSYNC; 2236cbff00fSChristoph Hellwig else 2246cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_DIRSYNC; 225e1e8fb6aSLi Zefan if (flags & FS_NOCOW_FL) 226e1e8fb6aSLi Zefan ip->flags |= BTRFS_INODE_NODATACOW; 227e1e8fb6aSLi Zefan else 228e1e8fb6aSLi Zefan ip->flags &= ~BTRFS_INODE_NODATACOW; 2296cbff00fSChristoph Hellwig 23075e7cb7fSLiu Bo /* 23175e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 23275e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 23375e7cb7fSLiu Bo * things smaller. 23475e7cb7fSLiu Bo */ 23575e7cb7fSLiu Bo if (flags & FS_NOCOMP_FL) { 23675e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_COMPRESS; 23775e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_NOCOMPRESS; 23875e7cb7fSLiu Bo } else if (flags & FS_COMPR_FL) { 23975e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_COMPRESS; 24075e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_NOCOMPRESS; 241ebcb904dSLi Zefan } else { 242ebcb904dSLi Zefan ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 24375e7cb7fSLiu Bo } 2446cbff00fSChristoph Hellwig 2456cbff00fSChristoph Hellwig trans = btrfs_join_transaction(root, 1); 2463612b495STsutomu Itoh BUG_ON(IS_ERR(trans)); 2476cbff00fSChristoph Hellwig 2486cbff00fSChristoph Hellwig ret = btrfs_update_inode(trans, root, inode); 2496cbff00fSChristoph Hellwig BUG_ON(ret); 2506cbff00fSChristoph Hellwig 2516cbff00fSChristoph Hellwig btrfs_update_iflags(inode); 2526cbff00fSChristoph Hellwig inode->i_ctime = CURRENT_TIME; 2536cbff00fSChristoph Hellwig btrfs_end_transaction(trans, root); 2546cbff00fSChristoph Hellwig 2556cbff00fSChristoph Hellwig mnt_drop_write(file->f_path.mnt); 2562d4e6f6aSliubo 2572d4e6f6aSliubo ret = 0; 2586cbff00fSChristoph Hellwig out_unlock: 2596cbff00fSChristoph Hellwig mutex_unlock(&inode->i_mutex); 2602d4e6f6aSliubo return ret; 2616cbff00fSChristoph Hellwig } 2626cbff00fSChristoph Hellwig 2636cbff00fSChristoph Hellwig static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 2646cbff00fSChristoph Hellwig { 2656cbff00fSChristoph Hellwig struct inode *inode = file->f_path.dentry->d_inode; 2666cbff00fSChristoph Hellwig 2676cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 2686cbff00fSChristoph Hellwig } 269f46b5a66SChristoph Hellwig 270f7039b1dSLi Dongyang static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg) 271f7039b1dSLi Dongyang { 272f7039b1dSLi Dongyang struct btrfs_root *root = fdentry(file)->d_sb->s_fs_info; 273f7039b1dSLi Dongyang struct btrfs_fs_info *fs_info = root->fs_info; 274f7039b1dSLi Dongyang struct btrfs_device *device; 275f7039b1dSLi Dongyang struct request_queue *q; 276f7039b1dSLi Dongyang struct fstrim_range range; 277f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 278f7039b1dSLi Dongyang u64 num_devices = 0; 279f7039b1dSLi Dongyang int ret; 280f7039b1dSLi Dongyang 281f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 282f7039b1dSLi Dongyang return -EPERM; 283f7039b1dSLi Dongyang 284f7039b1dSLi Dongyang mutex_lock(&fs_info->fs_devices->device_list_mutex); 285f7039b1dSLi Dongyang list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) { 286f7039b1dSLi Dongyang if (!device->bdev) 287f7039b1dSLi Dongyang continue; 288f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 289f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 290f7039b1dSLi Dongyang num_devices++; 291f7039b1dSLi Dongyang minlen = min((u64)q->limits.discard_granularity, 292f7039b1dSLi Dongyang minlen); 293f7039b1dSLi Dongyang } 294f7039b1dSLi Dongyang } 295f7039b1dSLi Dongyang mutex_unlock(&fs_info->fs_devices->device_list_mutex); 296f7039b1dSLi Dongyang if (!num_devices) 297f7039b1dSLi Dongyang return -EOPNOTSUPP; 298f7039b1dSLi Dongyang 299f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 300f7039b1dSLi Dongyang return -EFAULT; 301f7039b1dSLi Dongyang 302f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 303f7039b1dSLi Dongyang ret = btrfs_trim_fs(root, &range); 304f7039b1dSLi Dongyang if (ret < 0) 305f7039b1dSLi Dongyang return ret; 306f7039b1dSLi Dongyang 307f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 308f7039b1dSLi Dongyang return -EFAULT; 309f7039b1dSLi Dongyang 310f7039b1dSLi Dongyang return 0; 311f7039b1dSLi Dongyang } 312f7039b1dSLi Dongyang 313cb8e7090SChristoph Hellwig static noinline int create_subvol(struct btrfs_root *root, 314cb8e7090SChristoph Hellwig struct dentry *dentry, 31572fd032eSSage Weil char *name, int namelen, 31672fd032eSSage Weil u64 *async_transid) 317f46b5a66SChristoph Hellwig { 318f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 319f46b5a66SChristoph Hellwig struct btrfs_key key; 320f46b5a66SChristoph Hellwig struct btrfs_root_item root_item; 321f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 322f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 32376dda93cSYan, Zheng struct btrfs_root *new_root; 3246a912213SJosef Bacik struct dentry *parent = dget_parent(dentry); 3256a912213SJosef Bacik struct inode *dir; 326f46b5a66SChristoph Hellwig int ret; 327f46b5a66SChristoph Hellwig int err; 328f46b5a66SChristoph Hellwig u64 objectid; 329f46b5a66SChristoph Hellwig u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 3303de4586cSChris Mason u64 index = 0; 331f46b5a66SChristoph Hellwig 332a22285a6SYan, Zheng ret = btrfs_find_free_objectid(NULL, root->fs_info->tree_root, 333a22285a6SYan, Zheng 0, &objectid); 3346a912213SJosef Bacik if (ret) { 3356a912213SJosef Bacik dput(parent); 336a22285a6SYan, Zheng return ret; 3376a912213SJosef Bacik } 3386a912213SJosef Bacik 3396a912213SJosef Bacik dir = parent->d_inode; 3406a912213SJosef Bacik 3419ed74f2dSJosef Bacik /* 3429ed74f2dSJosef Bacik * 1 - inode item 3439ed74f2dSJosef Bacik * 2 - refs 3449ed74f2dSJosef Bacik * 1 - root item 3459ed74f2dSJosef Bacik * 2 - dir items 3469ed74f2dSJosef Bacik */ 347a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 3486a912213SJosef Bacik if (IS_ERR(trans)) { 3496a912213SJosef Bacik dput(parent); 350a22285a6SYan, Zheng return PTR_ERR(trans); 3516a912213SJosef Bacik } 352f46b5a66SChristoph Hellwig 3535d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 3545d4f98a2SYan Zheng 0, objectid, NULL, 0, 0, 0); 3558e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 3568e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 3578e8a1e31SJosef Bacik goto fail; 3588e8a1e31SJosef Bacik } 359f46b5a66SChristoph Hellwig 3605d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 361f46b5a66SChristoph Hellwig btrfs_set_header_bytenr(leaf, leaf->start); 362f46b5a66SChristoph Hellwig btrfs_set_header_generation(leaf, trans->transid); 3635d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 364f46b5a66SChristoph Hellwig btrfs_set_header_owner(leaf, objectid); 365f46b5a66SChristoph Hellwig 366f46b5a66SChristoph Hellwig write_extent_buffer(leaf, root->fs_info->fsid, 367f46b5a66SChristoph Hellwig (unsigned long)btrfs_header_fsid(leaf), 368f46b5a66SChristoph Hellwig BTRFS_FSID_SIZE); 3695d4f98a2SYan Zheng write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, 3705d4f98a2SYan Zheng (unsigned long)btrfs_header_chunk_tree_uuid(leaf), 3715d4f98a2SYan Zheng BTRFS_UUID_SIZE); 372f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 373f46b5a66SChristoph Hellwig 374f46b5a66SChristoph Hellwig inode_item = &root_item.inode; 375f46b5a66SChristoph Hellwig memset(inode_item, 0, sizeof(*inode_item)); 376f46b5a66SChristoph Hellwig inode_item->generation = cpu_to_le64(1); 377f46b5a66SChristoph Hellwig inode_item->size = cpu_to_le64(3); 378f46b5a66SChristoph Hellwig inode_item->nlink = cpu_to_le32(1); 379a76a3cd4SYan Zheng inode_item->nbytes = cpu_to_le64(root->leafsize); 380f46b5a66SChristoph Hellwig inode_item->mode = cpu_to_le32(S_IFDIR | 0755); 381f46b5a66SChristoph Hellwig 38208fe4db1SLi Zefan root_item.flags = 0; 38308fe4db1SLi Zefan root_item.byte_limit = 0; 38408fe4db1SLi Zefan inode_item->flags = cpu_to_le64(BTRFS_INODE_ROOT_ITEM_INIT); 38508fe4db1SLi Zefan 386f46b5a66SChristoph Hellwig btrfs_set_root_bytenr(&root_item, leaf->start); 38784234f3aSYan Zheng btrfs_set_root_generation(&root_item, trans->transid); 388f46b5a66SChristoph Hellwig btrfs_set_root_level(&root_item, 0); 389f46b5a66SChristoph Hellwig btrfs_set_root_refs(&root_item, 1); 39086b9f2ecSYan, Zheng btrfs_set_root_used(&root_item, leaf->len); 39180ff3856SYan Zheng btrfs_set_root_last_snapshot(&root_item, 0); 392f46b5a66SChristoph Hellwig 393f46b5a66SChristoph Hellwig memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress)); 394f46b5a66SChristoph Hellwig root_item.drop_level = 0; 395f46b5a66SChristoph Hellwig 396925baeddSChris Mason btrfs_tree_unlock(leaf); 397f46b5a66SChristoph Hellwig free_extent_buffer(leaf); 398f46b5a66SChristoph Hellwig leaf = NULL; 399f46b5a66SChristoph Hellwig 400f46b5a66SChristoph Hellwig btrfs_set_root_dirid(&root_item, new_dirid); 401f46b5a66SChristoph Hellwig 402f46b5a66SChristoph Hellwig key.objectid = objectid; 4035d4f98a2SYan Zheng key.offset = 0; 404f46b5a66SChristoph Hellwig btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 405f46b5a66SChristoph Hellwig ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, 406f46b5a66SChristoph Hellwig &root_item); 407f46b5a66SChristoph Hellwig if (ret) 408f46b5a66SChristoph Hellwig goto fail; 409f46b5a66SChristoph Hellwig 41076dda93cSYan, Zheng key.offset = (u64)-1; 41176dda93cSYan, Zheng new_root = btrfs_read_fs_root_no_name(root->fs_info, &key); 41276dda93cSYan, Zheng BUG_ON(IS_ERR(new_root)); 41376dda93cSYan, Zheng 41476dda93cSYan, Zheng btrfs_record_root_in_trans(trans, new_root); 41576dda93cSYan, Zheng 41676dda93cSYan, Zheng ret = btrfs_create_subvol_root(trans, new_root, new_dirid, 41776dda93cSYan, Zheng BTRFS_I(dir)->block_group); 418f46b5a66SChristoph Hellwig /* 419f46b5a66SChristoph Hellwig * insert the directory item 420f46b5a66SChristoph Hellwig */ 4213de4586cSChris Mason ret = btrfs_set_inode_index(dir, &index); 4223de4586cSChris Mason BUG_ON(ret); 4233de4586cSChris Mason 4243de4586cSChris Mason ret = btrfs_insert_dir_item(trans, root, 425f46b5a66SChristoph Hellwig name, namelen, dir->i_ino, &key, 4263de4586cSChris Mason BTRFS_FT_DIR, index); 427f46b5a66SChristoph Hellwig if (ret) 428f46b5a66SChristoph Hellwig goto fail; 4290660b5afSChris Mason 43052c26179SYan Zheng btrfs_i_size_write(dir, dir->i_size + namelen * 2); 43152c26179SYan Zheng ret = btrfs_update_inode(trans, root, dir); 43252c26179SYan Zheng BUG_ON(ret); 43352c26179SYan Zheng 4340660b5afSChris Mason ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 4354df27c4dSYan, Zheng objectid, root->root_key.objectid, 4360660b5afSChris Mason dir->i_ino, index, name, namelen); 43776dda93cSYan, Zheng 4380660b5afSChris Mason BUG_ON(ret); 4390660b5afSChris Mason 44076dda93cSYan, Zheng d_instantiate(dentry, btrfs_lookup_dentry(dir, dentry)); 441f46b5a66SChristoph Hellwig fail: 4426a912213SJosef Bacik dput(parent); 44372fd032eSSage Weil if (async_transid) { 44472fd032eSSage Weil *async_transid = trans->transid; 44572fd032eSSage Weil err = btrfs_commit_transaction_async(trans, root, 1); 44672fd032eSSage Weil } else { 44776dda93cSYan, Zheng err = btrfs_commit_transaction(trans, root); 44872fd032eSSage Weil } 449f46b5a66SChristoph Hellwig if (err && !ret) 450f46b5a66SChristoph Hellwig ret = err; 451f46b5a66SChristoph Hellwig return ret; 452f46b5a66SChristoph Hellwig } 453f46b5a66SChristoph Hellwig 45472fd032eSSage Weil static int create_snapshot(struct btrfs_root *root, struct dentry *dentry, 455b83cc969SLi Zefan char *name, int namelen, u64 *async_transid, 456b83cc969SLi Zefan bool readonly) 457f46b5a66SChristoph Hellwig { 4582e4bfab9SYan, Zheng struct inode *inode; 4596a912213SJosef Bacik struct dentry *parent; 460f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 461f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 4622e4bfab9SYan, Zheng int ret; 463f46b5a66SChristoph Hellwig 464f46b5a66SChristoph Hellwig if (!root->ref_cows) 465f46b5a66SChristoph Hellwig return -EINVAL; 466f46b5a66SChristoph Hellwig 467a22285a6SYan, Zheng pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS); 468a22285a6SYan, Zheng if (!pending_snapshot) 469a22285a6SYan, Zheng return -ENOMEM; 470a22285a6SYan, Zheng 471a22285a6SYan, Zheng btrfs_init_block_rsv(&pending_snapshot->block_rsv); 472a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 473a22285a6SYan, Zheng pending_snapshot->root = root; 474b83cc969SLi Zefan pending_snapshot->readonly = readonly; 475a22285a6SYan, Zheng 476a22285a6SYan, Zheng trans = btrfs_start_transaction(root->fs_info->extent_root, 5); 477a22285a6SYan, Zheng if (IS_ERR(trans)) { 478a22285a6SYan, Zheng ret = PTR_ERR(trans); 479a22285a6SYan, Zheng goto fail; 480a22285a6SYan, Zheng } 481a22285a6SYan, Zheng 482a22285a6SYan, Zheng ret = btrfs_snap_reserve_metadata(trans, pending_snapshot); 483a22285a6SYan, Zheng BUG_ON(ret); 484a22285a6SYan, Zheng 485a22285a6SYan, Zheng list_add(&pending_snapshot->list, 486a22285a6SYan, Zheng &trans->transaction->pending_snapshots); 48772fd032eSSage Weil if (async_transid) { 48872fd032eSSage Weil *async_transid = trans->transid; 48972fd032eSSage Weil ret = btrfs_commit_transaction_async(trans, 49072fd032eSSage Weil root->fs_info->extent_root, 1); 49172fd032eSSage Weil } else { 49272fd032eSSage Weil ret = btrfs_commit_transaction(trans, 49372fd032eSSage Weil root->fs_info->extent_root); 49472fd032eSSage Weil } 495a22285a6SYan, Zheng BUG_ON(ret); 496a22285a6SYan, Zheng 497a22285a6SYan, Zheng ret = pending_snapshot->error; 498f46b5a66SChristoph Hellwig if (ret) 4992e4bfab9SYan, Zheng goto fail; 500f46b5a66SChristoph Hellwig 50166b4ffd1SJosef Bacik ret = btrfs_orphan_cleanup(pending_snapshot->snap); 50266b4ffd1SJosef Bacik if (ret) 50366b4ffd1SJosef Bacik goto fail; 504f46b5a66SChristoph Hellwig 5056a912213SJosef Bacik parent = dget_parent(dentry); 5066a912213SJosef Bacik inode = btrfs_lookup_dentry(parent->d_inode, dentry); 5076a912213SJosef Bacik dput(parent); 5082e4bfab9SYan, Zheng if (IS_ERR(inode)) { 5092e4bfab9SYan, Zheng ret = PTR_ERR(inode); 5102e4bfab9SYan, Zheng goto fail; 5112e4bfab9SYan, Zheng } 5122e4bfab9SYan, Zheng BUG_ON(!inode); 5132e4bfab9SYan, Zheng d_instantiate(dentry, inode); 5142e4bfab9SYan, Zheng ret = 0; 5152e4bfab9SYan, Zheng fail: 516a22285a6SYan, Zheng kfree(pending_snapshot); 517f46b5a66SChristoph Hellwig return ret; 518f46b5a66SChristoph Hellwig } 519f46b5a66SChristoph Hellwig 5204260f7c7SSage Weil /* copy of check_sticky in fs/namei.c() 5214260f7c7SSage Weil * It's inline, so penalty for filesystems that don't use sticky bit is 5224260f7c7SSage Weil * minimal. 5234260f7c7SSage Weil */ 5244260f7c7SSage Weil static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode) 5254260f7c7SSage Weil { 5264260f7c7SSage Weil uid_t fsuid = current_fsuid(); 5274260f7c7SSage Weil 5284260f7c7SSage Weil if (!(dir->i_mode & S_ISVTX)) 5294260f7c7SSage Weil return 0; 5304260f7c7SSage Weil if (inode->i_uid == fsuid) 5314260f7c7SSage Weil return 0; 5324260f7c7SSage Weil if (dir->i_uid == fsuid) 5334260f7c7SSage Weil return 0; 5344260f7c7SSage Weil return !capable(CAP_FOWNER); 5354260f7c7SSage Weil } 5364260f7c7SSage Weil 5374260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 5384260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 5394260f7c7SSage Weil * whether the type of victim is right. 5404260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 5414260f7c7SSage Weil * 2. We should have write and exec permissions on dir 5424260f7c7SSage Weil * 3. We can't remove anything from append-only dir 5434260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 5444260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 5454260f7c7SSage Weil * a. be owner of dir, or 5464260f7c7SSage Weil * b. be owner of victim, or 5474260f7c7SSage Weil * c. have CAP_FOWNER capability 5484260f7c7SSage Weil * 6. If the victim is append-only or immutable we can't do antyhing with 5494260f7c7SSage Weil * links pointing to it. 5504260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 5514260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 5524260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 5534260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 5544260f7c7SSage Weil * nfs_async_unlink(). 5554260f7c7SSage Weil */ 5564260f7c7SSage Weil 5574260f7c7SSage Weil static int btrfs_may_delete(struct inode *dir,struct dentry *victim,int isdir) 5584260f7c7SSage Weil { 5594260f7c7SSage Weil int error; 5604260f7c7SSage Weil 5614260f7c7SSage Weil if (!victim->d_inode) 5624260f7c7SSage Weil return -ENOENT; 5634260f7c7SSage Weil 5644260f7c7SSage Weil BUG_ON(victim->d_parent->d_inode != dir); 5654260f7c7SSage Weil audit_inode_child(victim, dir); 5664260f7c7SSage Weil 5674260f7c7SSage Weil error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 5684260f7c7SSage Weil if (error) 5694260f7c7SSage Weil return error; 5704260f7c7SSage Weil if (IS_APPEND(dir)) 5714260f7c7SSage Weil return -EPERM; 5724260f7c7SSage Weil if (btrfs_check_sticky(dir, victim->d_inode)|| 5734260f7c7SSage Weil IS_APPEND(victim->d_inode)|| 5744260f7c7SSage Weil IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 5754260f7c7SSage Weil return -EPERM; 5764260f7c7SSage Weil if (isdir) { 5774260f7c7SSage Weil if (!S_ISDIR(victim->d_inode->i_mode)) 5784260f7c7SSage Weil return -ENOTDIR; 5794260f7c7SSage Weil if (IS_ROOT(victim)) 5804260f7c7SSage Weil return -EBUSY; 5814260f7c7SSage Weil } else if (S_ISDIR(victim->d_inode->i_mode)) 5824260f7c7SSage Weil return -EISDIR; 5834260f7c7SSage Weil if (IS_DEADDIR(dir)) 5844260f7c7SSage Weil return -ENOENT; 5854260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 5864260f7c7SSage Weil return -EBUSY; 5874260f7c7SSage Weil return 0; 5884260f7c7SSage Weil } 5894260f7c7SSage Weil 590cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 591cb8e7090SChristoph Hellwig static inline int btrfs_may_create(struct inode *dir, struct dentry *child) 592cb8e7090SChristoph Hellwig { 593cb8e7090SChristoph Hellwig if (child->d_inode) 594cb8e7090SChristoph Hellwig return -EEXIST; 595cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 596cb8e7090SChristoph Hellwig return -ENOENT; 597cb8e7090SChristoph Hellwig return inode_permission(dir, MAY_WRITE | MAY_EXEC); 598cb8e7090SChristoph Hellwig } 599cb8e7090SChristoph Hellwig 600cb8e7090SChristoph Hellwig /* 601cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 602cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 603cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 604cb8e7090SChristoph Hellwig */ 60576dda93cSYan, Zheng static noinline int btrfs_mksubvol(struct path *parent, 60676dda93cSYan, Zheng char *name, int namelen, 60772fd032eSSage Weil struct btrfs_root *snap_src, 608b83cc969SLi Zefan u64 *async_transid, bool readonly) 609cb8e7090SChristoph Hellwig { 61076dda93cSYan, Zheng struct inode *dir = parent->dentry->d_inode; 611cb8e7090SChristoph Hellwig struct dentry *dentry; 612cb8e7090SChristoph Hellwig int error; 613cb8e7090SChristoph Hellwig 61476dda93cSYan, Zheng mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT); 615cb8e7090SChristoph Hellwig 616cb8e7090SChristoph Hellwig dentry = lookup_one_len(name, parent->dentry, namelen); 617cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 618cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 619cb8e7090SChristoph Hellwig goto out_unlock; 620cb8e7090SChristoph Hellwig 621cb8e7090SChristoph Hellwig error = -EEXIST; 622cb8e7090SChristoph Hellwig if (dentry->d_inode) 623cb8e7090SChristoph Hellwig goto out_dput; 624cb8e7090SChristoph Hellwig 625cb8e7090SChristoph Hellwig error = mnt_want_write(parent->mnt); 626cb8e7090SChristoph Hellwig if (error) 627cb8e7090SChristoph Hellwig goto out_dput; 628cb8e7090SChristoph Hellwig 62976dda93cSYan, Zheng error = btrfs_may_create(dir, dentry); 630cb8e7090SChristoph Hellwig if (error) 631cb8e7090SChristoph Hellwig goto out_drop_write; 632cb8e7090SChristoph Hellwig 63376dda93cSYan, Zheng down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 63476dda93cSYan, Zheng 63576dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 63676dda93cSYan, Zheng goto out_up_read; 63776dda93cSYan, Zheng 6383de4586cSChris Mason if (snap_src) { 63972fd032eSSage Weil error = create_snapshot(snap_src, dentry, 640b83cc969SLi Zefan name, namelen, async_transid, readonly); 6413de4586cSChris Mason } else { 64276dda93cSYan, Zheng error = create_subvol(BTRFS_I(dir)->root, dentry, 64372fd032eSSage Weil name, namelen, async_transid); 6443de4586cSChris Mason } 64576dda93cSYan, Zheng if (!error) 64676dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 64776dda93cSYan, Zheng out_up_read: 64876dda93cSYan, Zheng up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 649cb8e7090SChristoph Hellwig out_drop_write: 650cb8e7090SChristoph Hellwig mnt_drop_write(parent->mnt); 651cb8e7090SChristoph Hellwig out_dput: 652cb8e7090SChristoph Hellwig dput(dentry); 653cb8e7090SChristoph Hellwig out_unlock: 65476dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 655cb8e7090SChristoph Hellwig return error; 656cb8e7090SChristoph Hellwig } 657cb8e7090SChristoph Hellwig 658940100a4SChris Mason static int should_defrag_range(struct inode *inode, u64 start, u64 len, 6591e701a32SChris Mason int thresh, u64 *last_len, u64 *skip, 6601e701a32SChris Mason u64 *defrag_end) 661940100a4SChris Mason { 662940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 663940100a4SChris Mason struct extent_map *em = NULL; 664940100a4SChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 665940100a4SChris Mason int ret = 1; 666940100a4SChris Mason 6671e701a32SChris Mason 6681e701a32SChris Mason if (thresh == 0) 6691e701a32SChris Mason thresh = 256 * 1024; 6701e701a32SChris Mason 671940100a4SChris Mason /* 672940100a4SChris Mason * make sure that once we start defragging and extent, we keep on 673940100a4SChris Mason * defragging it 674940100a4SChris Mason */ 675940100a4SChris Mason if (start < *defrag_end) 676940100a4SChris Mason return 1; 677940100a4SChris Mason 678940100a4SChris Mason *skip = 0; 679940100a4SChris Mason 680940100a4SChris Mason /* 681940100a4SChris Mason * hopefully we have this extent in the tree already, try without 682940100a4SChris Mason * the full extent lock 683940100a4SChris Mason */ 684940100a4SChris Mason read_lock(&em_tree->lock); 685940100a4SChris Mason em = lookup_extent_mapping(em_tree, start, len); 686940100a4SChris Mason read_unlock(&em_tree->lock); 687940100a4SChris Mason 688940100a4SChris Mason if (!em) { 689940100a4SChris Mason /* get the big lock and read metadata off disk */ 690940100a4SChris Mason lock_extent(io_tree, start, start + len - 1, GFP_NOFS); 691940100a4SChris Mason em = btrfs_get_extent(inode, NULL, 0, start, len, 0); 692940100a4SChris Mason unlock_extent(io_tree, start, start + len - 1, GFP_NOFS); 693940100a4SChris Mason 6946cf8bfbfSDan Carpenter if (IS_ERR(em)) 695940100a4SChris Mason return 0; 696940100a4SChris Mason } 697940100a4SChris Mason 698940100a4SChris Mason /* this will cover holes, and inline extents */ 699940100a4SChris Mason if (em->block_start >= EXTENT_MAP_LAST_BYTE) 700940100a4SChris Mason ret = 0; 701940100a4SChris Mason 702940100a4SChris Mason /* 703940100a4SChris Mason * we hit a real extent, if it is big don't bother defragging it again 704940100a4SChris Mason */ 7051e701a32SChris Mason if ((*last_len == 0 || *last_len >= thresh) && em->len >= thresh) 706940100a4SChris Mason ret = 0; 707940100a4SChris Mason 708940100a4SChris Mason /* 709940100a4SChris Mason * last_len ends up being a counter of how many bytes we've defragged. 710940100a4SChris Mason * every time we choose not to defrag an extent, we reset *last_len 711940100a4SChris Mason * so that the next tiny extent will force a defrag. 712940100a4SChris Mason * 713940100a4SChris Mason * The end result of this is that tiny extents before a single big 714940100a4SChris Mason * extent will force at least part of that big extent to be defragged. 715940100a4SChris Mason */ 716940100a4SChris Mason if (ret) { 717940100a4SChris Mason *last_len += len; 718940100a4SChris Mason *defrag_end = extent_map_end(em); 719940100a4SChris Mason } else { 720940100a4SChris Mason *last_len = 0; 721940100a4SChris Mason *skip = extent_map_end(em); 722940100a4SChris Mason *defrag_end = 0; 723940100a4SChris Mason } 724940100a4SChris Mason 725940100a4SChris Mason free_extent_map(em); 726940100a4SChris Mason return ret; 727940100a4SChris Mason } 728940100a4SChris Mason 7291e701a32SChris Mason static int btrfs_defrag_file(struct file *file, 7301e701a32SChris Mason struct btrfs_ioctl_defrag_range_args *range) 731f46b5a66SChristoph Hellwig { 732f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 733f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 734f46b5a66SChristoph Hellwig struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 7353eaa2885SChris Mason struct btrfs_ordered_extent *ordered; 736f46b5a66SChristoph Hellwig struct page *page; 7371a419d85SLi Zefan struct btrfs_super_block *disk_super; 738f46b5a66SChristoph Hellwig unsigned long last_index; 739f46b5a66SChristoph Hellwig unsigned long ra_pages = root->fs_info->bdi.ra_pages; 740f46b5a66SChristoph Hellwig unsigned long total_read = 0; 7411a419d85SLi Zefan u64 features; 742f46b5a66SChristoph Hellwig u64 page_start; 743f46b5a66SChristoph Hellwig u64 page_end; 744940100a4SChris Mason u64 last_len = 0; 745940100a4SChris Mason u64 skip = 0; 746940100a4SChris Mason u64 defrag_end = 0; 747f46b5a66SChristoph Hellwig unsigned long i; 748f46b5a66SChristoph Hellwig int ret; 7491a419d85SLi Zefan int compress_type = BTRFS_COMPRESS_ZLIB; 7501a419d85SLi Zefan 7511a419d85SLi Zefan if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 7521a419d85SLi Zefan if (range->compress_type > BTRFS_COMPRESS_TYPES) 7531a419d85SLi Zefan return -EINVAL; 7541a419d85SLi Zefan if (range->compress_type) 7551a419d85SLi Zefan compress_type = range->compress_type; 7561a419d85SLi Zefan } 757f46b5a66SChristoph Hellwig 758940100a4SChris Mason if (inode->i_size == 0) 759940100a4SChris Mason return 0; 760f46b5a66SChristoph Hellwig 7611e701a32SChris Mason if (range->start + range->len > range->start) { 7621e701a32SChris Mason last_index = min_t(u64, inode->i_size - 1, 7631e701a32SChris Mason range->start + range->len - 1) >> PAGE_CACHE_SHIFT; 7641e701a32SChris Mason } else { 765940100a4SChris Mason last_index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT; 7661e701a32SChris Mason } 7671e701a32SChris Mason 7681e701a32SChris Mason i = range->start >> PAGE_CACHE_SHIFT; 769940100a4SChris Mason while (i <= last_index) { 770940100a4SChris Mason if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT, 7711e701a32SChris Mason PAGE_CACHE_SIZE, 7721e701a32SChris Mason range->extent_thresh, 7731e701a32SChris Mason &last_len, &skip, 774940100a4SChris Mason &defrag_end)) { 775940100a4SChris Mason unsigned long next; 776940100a4SChris Mason /* 777940100a4SChris Mason * the should_defrag function tells us how much to skip 778940100a4SChris Mason * bump our counter by the suggested amount 779940100a4SChris Mason */ 780940100a4SChris Mason next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 781940100a4SChris Mason i = max(i + 1, next); 782940100a4SChris Mason continue; 783940100a4SChris Mason } 784940100a4SChris Mason 785f46b5a66SChristoph Hellwig if (total_read % ra_pages == 0) { 786f46b5a66SChristoph Hellwig btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i, 787f46b5a66SChristoph Hellwig min(last_index, i + ra_pages - 1)); 788f46b5a66SChristoph Hellwig } 789f46b5a66SChristoph Hellwig total_read++; 790940100a4SChris Mason mutex_lock(&inode->i_mutex); 7911e701a32SChris Mason if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) 7921a419d85SLi Zefan BTRFS_I(inode)->force_compress = compress_type; 793940100a4SChris Mason 7940ca1f7ceSYan, Zheng ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE); 7950ca1f7ceSYan, Zheng if (ret) 7960ca1f7ceSYan, Zheng goto err_unlock; 7973eaa2885SChris Mason again: 798940100a4SChris Mason if (inode->i_size == 0 || 799940100a4SChris Mason i > ((inode->i_size - 1) >> PAGE_CACHE_SHIFT)) { 800940100a4SChris Mason ret = 0; 801940100a4SChris Mason goto err_reservations; 802940100a4SChris Mason } 803940100a4SChris Mason 804f46b5a66SChristoph Hellwig page = grab_cache_page(inode->i_mapping, i); 8050ca1f7ceSYan, Zheng if (!page) { 8060ca1f7ceSYan, Zheng ret = -ENOMEM; 807940100a4SChris Mason goto err_reservations; 8080ca1f7ceSYan, Zheng } 809940100a4SChris Mason 810f46b5a66SChristoph Hellwig if (!PageUptodate(page)) { 811f46b5a66SChristoph Hellwig btrfs_readpage(NULL, page); 812f46b5a66SChristoph Hellwig lock_page(page); 813f46b5a66SChristoph Hellwig if (!PageUptodate(page)) { 814f46b5a66SChristoph Hellwig unlock_page(page); 815f46b5a66SChristoph Hellwig page_cache_release(page); 8160ca1f7ceSYan, Zheng ret = -EIO; 817940100a4SChris Mason goto err_reservations; 818f46b5a66SChristoph Hellwig } 819f46b5a66SChristoph Hellwig } 820f46b5a66SChristoph Hellwig 821940100a4SChris Mason if (page->mapping != inode->i_mapping) { 822940100a4SChris Mason unlock_page(page); 823940100a4SChris Mason page_cache_release(page); 824940100a4SChris Mason goto again; 825940100a4SChris Mason } 826940100a4SChris Mason 827f46b5a66SChristoph Hellwig wait_on_page_writeback(page); 828f46b5a66SChristoph Hellwig 829940100a4SChris Mason if (PageDirty(page)) { 8300ca1f7ceSYan, Zheng btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); 831940100a4SChris Mason goto loop_unlock; 832940100a4SChris Mason } 833940100a4SChris Mason 834f46b5a66SChristoph Hellwig page_start = (u64)page->index << PAGE_CACHE_SHIFT; 835f46b5a66SChristoph Hellwig page_end = page_start + PAGE_CACHE_SIZE - 1; 836f46b5a66SChristoph Hellwig lock_extent(io_tree, page_start, page_end, GFP_NOFS); 8373eaa2885SChris Mason 8383eaa2885SChris Mason ordered = btrfs_lookup_ordered_extent(inode, page_start); 8393eaa2885SChris Mason if (ordered) { 8403eaa2885SChris Mason unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 8413eaa2885SChris Mason unlock_page(page); 8423eaa2885SChris Mason page_cache_release(page); 8433eaa2885SChris Mason btrfs_start_ordered_extent(inode, ordered, 1); 8443eaa2885SChris Mason btrfs_put_ordered_extent(ordered); 8453eaa2885SChris Mason goto again; 8463eaa2885SChris Mason } 8473eaa2885SChris Mason set_page_extent_mapped(page); 8483eaa2885SChris Mason 849f87f057bSChris Mason /* 850f87f057bSChris Mason * this makes sure page_mkwrite is called on the 851f87f057bSChris Mason * page if it is dirtied again later 852f87f057bSChris Mason */ 853f87f057bSChris Mason clear_page_dirty_for_io(page); 854940100a4SChris Mason clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, 855940100a4SChris Mason page_end, EXTENT_DIRTY | EXTENT_DELALLOC | 856940100a4SChris Mason EXTENT_DO_ACCOUNTING, GFP_NOFS); 857f87f057bSChris Mason 8582ac55d41SJosef Bacik btrfs_set_extent_delalloc(inode, page_start, page_end, NULL); 859940100a4SChris Mason ClearPageChecked(page); 860f46b5a66SChristoph Hellwig set_page_dirty(page); 861a1ed835eSChris Mason unlock_extent(io_tree, page_start, page_end, GFP_NOFS); 862940100a4SChris Mason 863940100a4SChris Mason loop_unlock: 864f46b5a66SChristoph Hellwig unlock_page(page); 865f46b5a66SChristoph Hellwig page_cache_release(page); 866940100a4SChris Mason mutex_unlock(&inode->i_mutex); 867940100a4SChris Mason 868f46b5a66SChristoph Hellwig balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1); 869940100a4SChris Mason i++; 870f46b5a66SChristoph Hellwig } 871f46b5a66SChristoph Hellwig 8721e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) 8731e701a32SChris Mason filemap_flush(inode->i_mapping); 8741e701a32SChris Mason 8751e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 8761e701a32SChris Mason /* the filemap_flush will queue IO into the worker threads, but 8771e701a32SChris Mason * we have to make sure the IO is actually started and that 8781e701a32SChris Mason * ordered extents get created before we return 8791e701a32SChris Mason */ 8801e701a32SChris Mason atomic_inc(&root->fs_info->async_submit_draining); 8811e701a32SChris Mason while (atomic_read(&root->fs_info->nr_async_submits) || 8821e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages)) { 8831e701a32SChris Mason wait_event(root->fs_info->async_submit_wait, 8841e701a32SChris Mason (atomic_read(&root->fs_info->nr_async_submits) == 0 && 8851e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages) == 0)); 8861e701a32SChris Mason } 8871e701a32SChris Mason atomic_dec(&root->fs_info->async_submit_draining); 8881e701a32SChris Mason 8891e701a32SChris Mason mutex_lock(&inode->i_mutex); 890261507a0SLi Zefan BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE; 8911e701a32SChris Mason mutex_unlock(&inode->i_mutex); 8921e701a32SChris Mason } 8931e701a32SChris Mason 8941a419d85SLi Zefan disk_super = &root->fs_info->super_copy; 8951a419d85SLi Zefan features = btrfs_super_incompat_flags(disk_super); 8961a419d85SLi Zefan if (range->compress_type == BTRFS_COMPRESS_LZO) { 8971a419d85SLi Zefan features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 8981a419d85SLi Zefan btrfs_set_super_incompat_flags(disk_super, features); 8991a419d85SLi Zefan } 9001a419d85SLi Zefan 901f46b5a66SChristoph Hellwig return 0; 902940100a4SChris Mason 903940100a4SChris Mason err_reservations: 9040ca1f7ceSYan, Zheng btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE); 9050ca1f7ceSYan, Zheng err_unlock: 906940100a4SChris Mason mutex_unlock(&inode->i_mutex); 907940100a4SChris Mason return ret; 908f46b5a66SChristoph Hellwig } 909f46b5a66SChristoph Hellwig 91076dda93cSYan, Zheng static noinline int btrfs_ioctl_resize(struct btrfs_root *root, 91176dda93cSYan, Zheng void __user *arg) 912f46b5a66SChristoph Hellwig { 913f46b5a66SChristoph Hellwig u64 new_size; 914f46b5a66SChristoph Hellwig u64 old_size; 915f46b5a66SChristoph Hellwig u64 devid = 1; 916f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 917f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 918f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 919f46b5a66SChristoph Hellwig char *sizestr; 920f46b5a66SChristoph Hellwig char *devstr = NULL; 921f46b5a66SChristoph Hellwig int ret = 0; 922f46b5a66SChristoph Hellwig int mod = 0; 923f46b5a66SChristoph Hellwig 924c146afadSYan Zheng if (root->fs_info->sb->s_flags & MS_RDONLY) 925c146afadSYan Zheng return -EROFS; 926c146afadSYan Zheng 927e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 928e441d54dSChris Mason return -EPERM; 929e441d54dSChris Mason 930dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 931dae7b665SLi Zefan if (IS_ERR(vol_args)) 932dae7b665SLi Zefan return PTR_ERR(vol_args); 9335516e595SMark Fasheh 9345516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 935f46b5a66SChristoph Hellwig 9367d9eb12cSChris Mason mutex_lock(&root->fs_info->volume_mutex); 937f46b5a66SChristoph Hellwig sizestr = vol_args->name; 938f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 939f46b5a66SChristoph Hellwig if (devstr) { 940f46b5a66SChristoph Hellwig char *end; 941f46b5a66SChristoph Hellwig sizestr = devstr + 1; 942f46b5a66SChristoph Hellwig *devstr = '\0'; 943f46b5a66SChristoph Hellwig devstr = vol_args->name; 944f46b5a66SChristoph Hellwig devid = simple_strtoull(devstr, &end, 10); 94521380931SJoel Becker printk(KERN_INFO "resizing devid %llu\n", 94621380931SJoel Becker (unsigned long long)devid); 947f46b5a66SChristoph Hellwig } 9482b82032cSYan Zheng device = btrfs_find_device(root, devid, NULL, NULL); 949f46b5a66SChristoph Hellwig if (!device) { 95021380931SJoel Becker printk(KERN_INFO "resizer unable to find device %llu\n", 95121380931SJoel Becker (unsigned long long)devid); 952f46b5a66SChristoph Hellwig ret = -EINVAL; 953f46b5a66SChristoph Hellwig goto out_unlock; 954f46b5a66SChristoph Hellwig } 955f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 956f46b5a66SChristoph Hellwig new_size = device->bdev->bd_inode->i_size; 957f46b5a66SChristoph Hellwig else { 958f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 959f46b5a66SChristoph Hellwig mod = -1; 960f46b5a66SChristoph Hellwig sizestr++; 961f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 962f46b5a66SChristoph Hellwig mod = 1; 963f46b5a66SChristoph Hellwig sizestr++; 964f46b5a66SChristoph Hellwig } 96591748467SAkinobu Mita new_size = memparse(sizestr, NULL); 966f46b5a66SChristoph Hellwig if (new_size == 0) { 967f46b5a66SChristoph Hellwig ret = -EINVAL; 968f46b5a66SChristoph Hellwig goto out_unlock; 969f46b5a66SChristoph Hellwig } 970f46b5a66SChristoph Hellwig } 971f46b5a66SChristoph Hellwig 972f46b5a66SChristoph Hellwig old_size = device->total_bytes; 973f46b5a66SChristoph Hellwig 974f46b5a66SChristoph Hellwig if (mod < 0) { 975f46b5a66SChristoph Hellwig if (new_size > old_size) { 976f46b5a66SChristoph Hellwig ret = -EINVAL; 977f46b5a66SChristoph Hellwig goto out_unlock; 978f46b5a66SChristoph Hellwig } 979f46b5a66SChristoph Hellwig new_size = old_size - new_size; 980f46b5a66SChristoph Hellwig } else if (mod > 0) { 981f46b5a66SChristoph Hellwig new_size = old_size + new_size; 982f46b5a66SChristoph Hellwig } 983f46b5a66SChristoph Hellwig 984f46b5a66SChristoph Hellwig if (new_size < 256 * 1024 * 1024) { 985f46b5a66SChristoph Hellwig ret = -EINVAL; 986f46b5a66SChristoph Hellwig goto out_unlock; 987f46b5a66SChristoph Hellwig } 988f46b5a66SChristoph Hellwig if (new_size > device->bdev->bd_inode->i_size) { 989f46b5a66SChristoph Hellwig ret = -EFBIG; 990f46b5a66SChristoph Hellwig goto out_unlock; 991f46b5a66SChristoph Hellwig } 992f46b5a66SChristoph Hellwig 993f46b5a66SChristoph Hellwig do_div(new_size, root->sectorsize); 994f46b5a66SChristoph Hellwig new_size *= root->sectorsize; 995f46b5a66SChristoph Hellwig 996f46b5a66SChristoph Hellwig printk(KERN_INFO "new size for %s is %llu\n", 997f46b5a66SChristoph Hellwig device->name, (unsigned long long)new_size); 998f46b5a66SChristoph Hellwig 999f46b5a66SChristoph Hellwig if (new_size > old_size) { 1000a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 100198d5dc13STsutomu Itoh if (IS_ERR(trans)) { 100298d5dc13STsutomu Itoh ret = PTR_ERR(trans); 100398d5dc13STsutomu Itoh goto out_unlock; 100498d5dc13STsutomu Itoh } 1005f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 1006f46b5a66SChristoph Hellwig btrfs_commit_transaction(trans, root); 1007f46b5a66SChristoph Hellwig } else { 1008f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 1009f46b5a66SChristoph Hellwig } 1010f46b5a66SChristoph Hellwig 1011f46b5a66SChristoph Hellwig out_unlock: 10127d9eb12cSChris Mason mutex_unlock(&root->fs_info->volume_mutex); 1013f46b5a66SChristoph Hellwig kfree(vol_args); 1014f46b5a66SChristoph Hellwig return ret; 1015f46b5a66SChristoph Hellwig } 1016f46b5a66SChristoph Hellwig 101772fd032eSSage Weil static noinline int btrfs_ioctl_snap_create_transid(struct file *file, 101872fd032eSSage Weil char *name, 101972fd032eSSage Weil unsigned long fd, 102072fd032eSSage Weil int subvol, 1021b83cc969SLi Zefan u64 *transid, 1022b83cc969SLi Zefan bool readonly) 1023f46b5a66SChristoph Hellwig { 1024cb8e7090SChristoph Hellwig struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 10253de4586cSChris Mason struct file *src_file; 1026f46b5a66SChristoph Hellwig int namelen; 10273de4586cSChris Mason int ret = 0; 1028f46b5a66SChristoph Hellwig 1029c146afadSYan Zheng if (root->fs_info->sb->s_flags & MS_RDONLY) 1030c146afadSYan Zheng return -EROFS; 1031c146afadSYan Zheng 103272fd032eSSage Weil namelen = strlen(name); 103372fd032eSSage Weil if (strchr(name, '/')) { 1034f46b5a66SChristoph Hellwig ret = -EINVAL; 1035f46b5a66SChristoph Hellwig goto out; 1036f46b5a66SChristoph Hellwig } 1037f46b5a66SChristoph Hellwig 10383de4586cSChris Mason if (subvol) { 103972fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 1040b83cc969SLi Zefan NULL, transid, readonly); 1041cb8e7090SChristoph Hellwig } else { 10423de4586cSChris Mason struct inode *src_inode; 104372fd032eSSage Weil src_file = fget(fd); 10443de4586cSChris Mason if (!src_file) { 10453de4586cSChris Mason ret = -EINVAL; 10463de4586cSChris Mason goto out; 10473de4586cSChris Mason } 10483de4586cSChris Mason 10493de4586cSChris Mason src_inode = src_file->f_path.dentry->d_inode; 10503de4586cSChris Mason if (src_inode->i_sb != file->f_path.dentry->d_inode->i_sb) { 1051d397712bSChris Mason printk(KERN_INFO "btrfs: Snapshot src from " 1052d397712bSChris Mason "another FS\n"); 10533de4586cSChris Mason ret = -EINVAL; 10543de4586cSChris Mason fput(src_file); 10553de4586cSChris Mason goto out; 10563de4586cSChris Mason } 105772fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 105872fd032eSSage Weil BTRFS_I(src_inode)->root, 1059b83cc969SLi Zefan transid, readonly); 10603de4586cSChris Mason fput(src_file); 1061cb8e7090SChristoph Hellwig } 1062f46b5a66SChristoph Hellwig out: 106372fd032eSSage Weil return ret; 106472fd032eSSage Weil } 106572fd032eSSage Weil 106672fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 1067fa0d2b9bSLi Zefan void __user *arg, int subvol) 106872fd032eSSage Weil { 1069fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 107072fd032eSSage Weil int ret; 107172fd032eSSage Weil 1072fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1073fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1074fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1075fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1076fa0d2b9bSLi Zefan 1077fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1078b83cc969SLi Zefan vol_args->fd, subvol, 1079b83cc969SLi Zefan NULL, false); 1080fa0d2b9bSLi Zefan 1081fa0d2b9bSLi Zefan kfree(vol_args); 1082fa0d2b9bSLi Zefan return ret; 1083fa0d2b9bSLi Zefan } 1084fa0d2b9bSLi Zefan 1085fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1086fa0d2b9bSLi Zefan void __user *arg, int subvol) 1087fa0d2b9bSLi Zefan { 1088fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 1089fa0d2b9bSLi Zefan int ret; 1090fdfb1e4fSLi Zefan u64 transid = 0; 1091fdfb1e4fSLi Zefan u64 *ptr = NULL; 1092b83cc969SLi Zefan bool readonly = false; 109372fd032eSSage Weil 1094fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1095fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1096fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1097fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1098fdfb1e4fSLi Zefan 1099b83cc969SLi Zefan if (vol_args->flags & 1100b83cc969SLi Zefan ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY)) { 1101b83cc969SLi Zefan ret = -EOPNOTSUPP; 1102fdfb1e4fSLi Zefan goto out; 1103fdfb1e4fSLi Zefan } 1104fdfb1e4fSLi Zefan 1105fa0d2b9bSLi Zefan if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) 1106fdfb1e4fSLi Zefan ptr = &transid; 1107b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1108b83cc969SLi Zefan readonly = true; 110975eaa0e2SSage Weil 1110fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1111b83cc969SLi Zefan vol_args->fd, subvol, 1112b83cc969SLi Zefan ptr, readonly); 111375eaa0e2SSage Weil 1114fdfb1e4fSLi Zefan if (ret == 0 && ptr && 111575eaa0e2SSage Weil copy_to_user(arg + 1116fdfb1e4fSLi Zefan offsetof(struct btrfs_ioctl_vol_args_v2, 1117fdfb1e4fSLi Zefan transid), ptr, sizeof(*ptr))) 111875eaa0e2SSage Weil ret = -EFAULT; 1119fdfb1e4fSLi Zefan out: 1120f46b5a66SChristoph Hellwig kfree(vol_args); 1121f46b5a66SChristoph Hellwig return ret; 1122f46b5a66SChristoph Hellwig } 1123f46b5a66SChristoph Hellwig 11240caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 11250caa102dSLi Zefan void __user *arg) 11260caa102dSLi Zefan { 11270caa102dSLi Zefan struct inode *inode = fdentry(file)->d_inode; 11280caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 11290caa102dSLi Zefan int ret = 0; 11300caa102dSLi Zefan u64 flags = 0; 11310caa102dSLi Zefan 11320caa102dSLi Zefan if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) 11330caa102dSLi Zefan return -EINVAL; 11340caa102dSLi Zefan 11350caa102dSLi Zefan down_read(&root->fs_info->subvol_sem); 11360caa102dSLi Zefan if (btrfs_root_readonly(root)) 11370caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 11380caa102dSLi Zefan up_read(&root->fs_info->subvol_sem); 11390caa102dSLi Zefan 11400caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 11410caa102dSLi Zefan ret = -EFAULT; 11420caa102dSLi Zefan 11430caa102dSLi Zefan return ret; 11440caa102dSLi Zefan } 11450caa102dSLi Zefan 11460caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 11470caa102dSLi Zefan void __user *arg) 11480caa102dSLi Zefan { 11490caa102dSLi Zefan struct inode *inode = fdentry(file)->d_inode; 11500caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 11510caa102dSLi Zefan struct btrfs_trans_handle *trans; 11520caa102dSLi Zefan u64 root_flags; 11530caa102dSLi Zefan u64 flags; 11540caa102dSLi Zefan int ret = 0; 11550caa102dSLi Zefan 11560caa102dSLi Zefan if (root->fs_info->sb->s_flags & MS_RDONLY) 11570caa102dSLi Zefan return -EROFS; 11580caa102dSLi Zefan 11590caa102dSLi Zefan if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) 11600caa102dSLi Zefan return -EINVAL; 11610caa102dSLi Zefan 11620caa102dSLi Zefan if (copy_from_user(&flags, arg, sizeof(flags))) 11630caa102dSLi Zefan return -EFAULT; 11640caa102dSLi Zefan 1165b4dc2b8cSLi Zefan if (flags & BTRFS_SUBVOL_CREATE_ASYNC) 11660caa102dSLi Zefan return -EINVAL; 11670caa102dSLi Zefan 11680caa102dSLi Zefan if (flags & ~BTRFS_SUBVOL_RDONLY) 11690caa102dSLi Zefan return -EOPNOTSUPP; 11700caa102dSLi Zefan 11712e149670SSerge E. Hallyn if (!inode_owner_or_capable(inode)) 1172b4dc2b8cSLi Zefan return -EACCES; 1173b4dc2b8cSLi Zefan 11740caa102dSLi Zefan down_write(&root->fs_info->subvol_sem); 11750caa102dSLi Zefan 11760caa102dSLi Zefan /* nothing to do */ 11770caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 11780caa102dSLi Zefan goto out; 11790caa102dSLi Zefan 11800caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 11810caa102dSLi Zefan if (flags & BTRFS_SUBVOL_RDONLY) 11820caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 11830caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 11840caa102dSLi Zefan else 11850caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 11860caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 11870caa102dSLi Zefan 11880caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 11890caa102dSLi Zefan if (IS_ERR(trans)) { 11900caa102dSLi Zefan ret = PTR_ERR(trans); 11910caa102dSLi Zefan goto out_reset; 11920caa102dSLi Zefan } 11930caa102dSLi Zefan 1194b4dc2b8cSLi Zefan ret = btrfs_update_root(trans, root->fs_info->tree_root, 11950caa102dSLi Zefan &root->root_key, &root->root_item); 11960caa102dSLi Zefan 11970caa102dSLi Zefan btrfs_commit_transaction(trans, root); 11980caa102dSLi Zefan out_reset: 11990caa102dSLi Zefan if (ret) 12000caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 12010caa102dSLi Zefan out: 12020caa102dSLi Zefan up_write(&root->fs_info->subvol_sem); 12030caa102dSLi Zefan return ret; 12040caa102dSLi Zefan } 12050caa102dSLi Zefan 120676dda93cSYan, Zheng /* 120776dda93cSYan, Zheng * helper to check if the subvolume references other subvolumes 120876dda93cSYan, Zheng */ 120976dda93cSYan, Zheng static noinline int may_destroy_subvol(struct btrfs_root *root) 121076dda93cSYan, Zheng { 121176dda93cSYan, Zheng struct btrfs_path *path; 121276dda93cSYan, Zheng struct btrfs_key key; 121376dda93cSYan, Zheng int ret; 121476dda93cSYan, Zheng 121576dda93cSYan, Zheng path = btrfs_alloc_path(); 121676dda93cSYan, Zheng if (!path) 121776dda93cSYan, Zheng return -ENOMEM; 121876dda93cSYan, Zheng 121976dda93cSYan, Zheng key.objectid = root->root_key.objectid; 122076dda93cSYan, Zheng key.type = BTRFS_ROOT_REF_KEY; 122176dda93cSYan, Zheng key.offset = (u64)-1; 122276dda93cSYan, Zheng 122376dda93cSYan, Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, 122476dda93cSYan, Zheng &key, path, 0, 0); 122576dda93cSYan, Zheng if (ret < 0) 122676dda93cSYan, Zheng goto out; 122776dda93cSYan, Zheng BUG_ON(ret == 0); 122876dda93cSYan, Zheng 122976dda93cSYan, Zheng ret = 0; 123076dda93cSYan, Zheng if (path->slots[0] > 0) { 123176dda93cSYan, Zheng path->slots[0]--; 123276dda93cSYan, Zheng btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 123376dda93cSYan, Zheng if (key.objectid == root->root_key.objectid && 123476dda93cSYan, Zheng key.type == BTRFS_ROOT_REF_KEY) 123576dda93cSYan, Zheng ret = -ENOTEMPTY; 123676dda93cSYan, Zheng } 123776dda93cSYan, Zheng out: 123876dda93cSYan, Zheng btrfs_free_path(path); 123976dda93cSYan, Zheng return ret; 124076dda93cSYan, Zheng } 124176dda93cSYan, Zheng 1242ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 1243ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 1244ac8e9819SChris Mason { 1245abc6e134SChris Mason struct btrfs_key test; 1246abc6e134SChris Mason int ret; 1247abc6e134SChris Mason 1248abc6e134SChris Mason test.objectid = sk->min_objectid; 1249abc6e134SChris Mason test.type = sk->min_type; 1250abc6e134SChris Mason test.offset = sk->min_offset; 1251abc6e134SChris Mason 1252abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1253abc6e134SChris Mason if (ret < 0) 1254ac8e9819SChris Mason return 0; 1255abc6e134SChris Mason 1256abc6e134SChris Mason test.objectid = sk->max_objectid; 1257abc6e134SChris Mason test.type = sk->max_type; 1258abc6e134SChris Mason test.offset = sk->max_offset; 1259abc6e134SChris Mason 1260abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1261abc6e134SChris Mason if (ret > 0) 1262ac8e9819SChris Mason return 0; 1263ac8e9819SChris Mason return 1; 1264ac8e9819SChris Mason } 1265ac8e9819SChris Mason 1266ac8e9819SChris Mason static noinline int copy_to_sk(struct btrfs_root *root, 1267ac8e9819SChris Mason struct btrfs_path *path, 1268ac8e9819SChris Mason struct btrfs_key *key, 1269ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 1270ac8e9819SChris Mason char *buf, 1271ac8e9819SChris Mason unsigned long *sk_offset, 1272ac8e9819SChris Mason int *num_found) 1273ac8e9819SChris Mason { 1274ac8e9819SChris Mason u64 found_transid; 1275ac8e9819SChris Mason struct extent_buffer *leaf; 1276ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 1277ac8e9819SChris Mason unsigned long item_off; 1278ac8e9819SChris Mason unsigned long item_len; 1279ac8e9819SChris Mason int nritems; 1280ac8e9819SChris Mason int i; 1281ac8e9819SChris Mason int slot; 1282ac8e9819SChris Mason int ret = 0; 1283ac8e9819SChris Mason 1284ac8e9819SChris Mason leaf = path->nodes[0]; 1285ac8e9819SChris Mason slot = path->slots[0]; 1286ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 1287ac8e9819SChris Mason 1288ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 1289ac8e9819SChris Mason i = nritems; 1290ac8e9819SChris Mason goto advance_key; 1291ac8e9819SChris Mason } 1292ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 1293ac8e9819SChris Mason 1294ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 1295ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 1296ac8e9819SChris Mason item_len = btrfs_item_size_nr(leaf, i); 1297ac8e9819SChris Mason 1298ac8e9819SChris Mason if (item_len > BTRFS_SEARCH_ARGS_BUFSIZE) 1299ac8e9819SChris Mason item_len = 0; 1300ac8e9819SChris Mason 1301ac8e9819SChris Mason if (sizeof(sh) + item_len + *sk_offset > 1302ac8e9819SChris Mason BTRFS_SEARCH_ARGS_BUFSIZE) { 1303ac8e9819SChris Mason ret = 1; 1304ac8e9819SChris Mason goto overflow; 1305ac8e9819SChris Mason } 1306ac8e9819SChris Mason 1307ac8e9819SChris Mason btrfs_item_key_to_cpu(leaf, key, i); 1308ac8e9819SChris Mason if (!key_in_sk(key, sk)) 1309ac8e9819SChris Mason continue; 1310ac8e9819SChris Mason 1311ac8e9819SChris Mason sh.objectid = key->objectid; 1312ac8e9819SChris Mason sh.offset = key->offset; 1313ac8e9819SChris Mason sh.type = key->type; 1314ac8e9819SChris Mason sh.len = item_len; 1315ac8e9819SChris Mason sh.transid = found_transid; 1316ac8e9819SChris Mason 1317ac8e9819SChris Mason /* copy search result header */ 1318ac8e9819SChris Mason memcpy(buf + *sk_offset, &sh, sizeof(sh)); 1319ac8e9819SChris Mason *sk_offset += sizeof(sh); 1320ac8e9819SChris Mason 1321ac8e9819SChris Mason if (item_len) { 1322ac8e9819SChris Mason char *p = buf + *sk_offset; 1323ac8e9819SChris Mason /* copy the item */ 1324ac8e9819SChris Mason read_extent_buffer(leaf, p, 1325ac8e9819SChris Mason item_off, item_len); 1326ac8e9819SChris Mason *sk_offset += item_len; 1327ac8e9819SChris Mason } 1328*e2156867SHugo Mills (*num_found)++; 1329ac8e9819SChris Mason 1330ac8e9819SChris Mason if (*num_found >= sk->nr_items) 1331ac8e9819SChris Mason break; 1332ac8e9819SChris Mason } 1333ac8e9819SChris Mason advance_key: 1334ac8e9819SChris Mason ret = 0; 1335abc6e134SChris Mason if (key->offset < (u64)-1 && key->offset < sk->max_offset) 1336abc6e134SChris Mason key->offset++; 1337abc6e134SChris Mason else if (key->type < (u8)-1 && key->type < sk->max_type) { 1338abc6e134SChris Mason key->offset = 0; 1339abc6e134SChris Mason key->type++; 1340abc6e134SChris Mason } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) { 1341abc6e134SChris Mason key->offset = 0; 1342abc6e134SChris Mason key->type = 0; 1343abc6e134SChris Mason key->objectid++; 1344abc6e134SChris Mason } else 1345abc6e134SChris Mason ret = 1; 1346ac8e9819SChris Mason overflow: 1347ac8e9819SChris Mason return ret; 1348ac8e9819SChris Mason } 1349ac8e9819SChris Mason 1350ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 1351ac8e9819SChris Mason struct btrfs_ioctl_search_args *args) 1352ac8e9819SChris Mason { 1353ac8e9819SChris Mason struct btrfs_root *root; 1354ac8e9819SChris Mason struct btrfs_key key; 1355ac8e9819SChris Mason struct btrfs_key max_key; 1356ac8e9819SChris Mason struct btrfs_path *path; 1357ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk = &args->key; 1358ac8e9819SChris Mason struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info; 1359ac8e9819SChris Mason int ret; 1360ac8e9819SChris Mason int num_found = 0; 1361ac8e9819SChris Mason unsigned long sk_offset = 0; 1362ac8e9819SChris Mason 1363ac8e9819SChris Mason path = btrfs_alloc_path(); 1364ac8e9819SChris Mason if (!path) 1365ac8e9819SChris Mason return -ENOMEM; 1366ac8e9819SChris Mason 1367ac8e9819SChris Mason if (sk->tree_id == 0) { 1368ac8e9819SChris Mason /* search the root of the inode that was passed */ 1369ac8e9819SChris Mason root = BTRFS_I(inode)->root; 1370ac8e9819SChris Mason } else { 1371ac8e9819SChris Mason key.objectid = sk->tree_id; 1372ac8e9819SChris Mason key.type = BTRFS_ROOT_ITEM_KEY; 1373ac8e9819SChris Mason key.offset = (u64)-1; 1374ac8e9819SChris Mason root = btrfs_read_fs_root_no_name(info, &key); 1375ac8e9819SChris Mason if (IS_ERR(root)) { 1376ac8e9819SChris Mason printk(KERN_ERR "could not find root %llu\n", 1377ac8e9819SChris Mason sk->tree_id); 1378ac8e9819SChris Mason btrfs_free_path(path); 1379ac8e9819SChris Mason return -ENOENT; 1380ac8e9819SChris Mason } 1381ac8e9819SChris Mason } 1382ac8e9819SChris Mason 1383ac8e9819SChris Mason key.objectid = sk->min_objectid; 1384ac8e9819SChris Mason key.type = sk->min_type; 1385ac8e9819SChris Mason key.offset = sk->min_offset; 1386ac8e9819SChris Mason 1387ac8e9819SChris Mason max_key.objectid = sk->max_objectid; 1388ac8e9819SChris Mason max_key.type = sk->max_type; 1389ac8e9819SChris Mason max_key.offset = sk->max_offset; 1390ac8e9819SChris Mason 1391ac8e9819SChris Mason path->keep_locks = 1; 1392ac8e9819SChris Mason 1393ac8e9819SChris Mason while(1) { 1394ac8e9819SChris Mason ret = btrfs_search_forward(root, &key, &max_key, path, 0, 1395ac8e9819SChris Mason sk->min_transid); 1396ac8e9819SChris Mason if (ret != 0) { 1397ac8e9819SChris Mason if (ret > 0) 1398ac8e9819SChris Mason ret = 0; 1399ac8e9819SChris Mason goto err; 1400ac8e9819SChris Mason } 1401ac8e9819SChris Mason ret = copy_to_sk(root, path, &key, sk, args->buf, 1402ac8e9819SChris Mason &sk_offset, &num_found); 1403ac8e9819SChris Mason btrfs_release_path(root, path); 1404ac8e9819SChris Mason if (ret || num_found >= sk->nr_items) 1405ac8e9819SChris Mason break; 1406ac8e9819SChris Mason 1407ac8e9819SChris Mason } 1408ac8e9819SChris Mason ret = 0; 1409ac8e9819SChris Mason err: 1410ac8e9819SChris Mason sk->nr_items = num_found; 1411ac8e9819SChris Mason btrfs_free_path(path); 1412ac8e9819SChris Mason return ret; 1413ac8e9819SChris Mason } 1414ac8e9819SChris Mason 1415ac8e9819SChris Mason static noinline int btrfs_ioctl_tree_search(struct file *file, 1416ac8e9819SChris Mason void __user *argp) 1417ac8e9819SChris Mason { 1418ac8e9819SChris Mason struct btrfs_ioctl_search_args *args; 1419ac8e9819SChris Mason struct inode *inode; 1420ac8e9819SChris Mason int ret; 1421ac8e9819SChris Mason 1422ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 1423ac8e9819SChris Mason return -EPERM; 1424ac8e9819SChris Mason 14252354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 14262354d08fSJulia Lawall if (IS_ERR(args)) 14272354d08fSJulia Lawall return PTR_ERR(args); 1428ac8e9819SChris Mason 1429ac8e9819SChris Mason inode = fdentry(file)->d_inode; 1430ac8e9819SChris Mason ret = search_ioctl(inode, args); 1431ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 1432ac8e9819SChris Mason ret = -EFAULT; 1433ac8e9819SChris Mason kfree(args); 1434ac8e9819SChris Mason return ret; 1435ac8e9819SChris Mason } 1436ac8e9819SChris Mason 143798d377a0STARUISI Hiroaki /* 1438ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 1439ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 144098d377a0STARUISI Hiroaki */ 144198d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 144298d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 144398d377a0STARUISI Hiroaki { 144498d377a0STARUISI Hiroaki struct btrfs_root *root; 144598d377a0STARUISI Hiroaki struct btrfs_key key; 1446ac8e9819SChris Mason char *ptr; 144798d377a0STARUISI Hiroaki int ret = -1; 144898d377a0STARUISI Hiroaki int slot; 144998d377a0STARUISI Hiroaki int len; 145098d377a0STARUISI Hiroaki int total_len = 0; 145198d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 145298d377a0STARUISI Hiroaki struct extent_buffer *l; 145398d377a0STARUISI Hiroaki struct btrfs_path *path; 145498d377a0STARUISI Hiroaki 145598d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 145698d377a0STARUISI Hiroaki name[0]='\0'; 145798d377a0STARUISI Hiroaki return 0; 145898d377a0STARUISI Hiroaki } 145998d377a0STARUISI Hiroaki 146098d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 146198d377a0STARUISI Hiroaki if (!path) 146298d377a0STARUISI Hiroaki return -ENOMEM; 146398d377a0STARUISI Hiroaki 1464ac8e9819SChris Mason ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX]; 146598d377a0STARUISI Hiroaki 146698d377a0STARUISI Hiroaki key.objectid = tree_id; 146798d377a0STARUISI Hiroaki key.type = BTRFS_ROOT_ITEM_KEY; 146898d377a0STARUISI Hiroaki key.offset = (u64)-1; 146998d377a0STARUISI Hiroaki root = btrfs_read_fs_root_no_name(info, &key); 147098d377a0STARUISI Hiroaki if (IS_ERR(root)) { 147198d377a0STARUISI Hiroaki printk(KERN_ERR "could not find root %llu\n", tree_id); 14728ad6fcabSChris Mason ret = -ENOENT; 14738ad6fcabSChris Mason goto out; 147498d377a0STARUISI Hiroaki } 147598d377a0STARUISI Hiroaki 147698d377a0STARUISI Hiroaki key.objectid = dirid; 147798d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 14788ad6fcabSChris Mason key.offset = (u64)-1; 147998d377a0STARUISI Hiroaki 148098d377a0STARUISI Hiroaki while(1) { 148198d377a0STARUISI Hiroaki ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 148298d377a0STARUISI Hiroaki if (ret < 0) 148398d377a0STARUISI Hiroaki goto out; 148498d377a0STARUISI Hiroaki 148598d377a0STARUISI Hiroaki l = path->nodes[0]; 148698d377a0STARUISI Hiroaki slot = path->slots[0]; 14878ad6fcabSChris Mason if (ret > 0 && slot > 0) 14888ad6fcabSChris Mason slot--; 148998d377a0STARUISI Hiroaki btrfs_item_key_to_cpu(l, &key, slot); 149098d377a0STARUISI Hiroaki 149198d377a0STARUISI Hiroaki if (ret > 0 && (key.objectid != dirid || 1492ac8e9819SChris Mason key.type != BTRFS_INODE_REF_KEY)) { 1493ac8e9819SChris Mason ret = -ENOENT; 149498d377a0STARUISI Hiroaki goto out; 1495ac8e9819SChris Mason } 149698d377a0STARUISI Hiroaki 149798d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 149898d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 149998d377a0STARUISI Hiroaki ptr -= len + 1; 150098d377a0STARUISI Hiroaki total_len += len + 1; 1501ac8e9819SChris Mason if (ptr < name) 150298d377a0STARUISI Hiroaki goto out; 150398d377a0STARUISI Hiroaki 150498d377a0STARUISI Hiroaki *(ptr + len) = '/'; 150598d377a0STARUISI Hiroaki read_extent_buffer(l, ptr,(unsigned long)(iref + 1), len); 150698d377a0STARUISI Hiroaki 150798d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 150898d377a0STARUISI Hiroaki break; 150998d377a0STARUISI Hiroaki 151098d377a0STARUISI Hiroaki btrfs_release_path(root, path); 151198d377a0STARUISI Hiroaki key.objectid = key.offset; 15128ad6fcabSChris Mason key.offset = (u64)-1; 151398d377a0STARUISI Hiroaki dirid = key.objectid; 151498d377a0STARUISI Hiroaki 151598d377a0STARUISI Hiroaki } 1516ac8e9819SChris Mason if (ptr < name) 151798d377a0STARUISI Hiroaki goto out; 1518ac8e9819SChris Mason memcpy(name, ptr, total_len); 151998d377a0STARUISI Hiroaki name[total_len]='\0'; 152098d377a0STARUISI Hiroaki ret = 0; 152198d377a0STARUISI Hiroaki out: 152298d377a0STARUISI Hiroaki btrfs_free_path(path); 1523ac8e9819SChris Mason return ret; 1524ac8e9819SChris Mason } 1525ac8e9819SChris Mason 1526ac8e9819SChris Mason static noinline int btrfs_ioctl_ino_lookup(struct file *file, 1527ac8e9819SChris Mason void __user *argp) 1528ac8e9819SChris Mason { 1529ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 1530ac8e9819SChris Mason struct inode *inode; 1531ac8e9819SChris Mason int ret; 1532ac8e9819SChris Mason 1533ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 1534ac8e9819SChris Mason return -EPERM; 1535ac8e9819SChris Mason 15362354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 15372354d08fSJulia Lawall if (IS_ERR(args)) 15382354d08fSJulia Lawall return PTR_ERR(args); 1539c2b96929SDan Carpenter 1540ac8e9819SChris Mason inode = fdentry(file)->d_inode; 1541ac8e9819SChris Mason 15421b53ac4dSChris Mason if (args->treeid == 0) 15431b53ac4dSChris Mason args->treeid = BTRFS_I(inode)->root->root_key.objectid; 15441b53ac4dSChris Mason 1545ac8e9819SChris Mason ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 1546ac8e9819SChris Mason args->treeid, args->objectid, 1547ac8e9819SChris Mason args->name); 1548ac8e9819SChris Mason 1549ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 1550ac8e9819SChris Mason ret = -EFAULT; 1551ac8e9819SChris Mason 1552ac8e9819SChris Mason kfree(args); 155398d377a0STARUISI Hiroaki return ret; 155498d377a0STARUISI Hiroaki } 155598d377a0STARUISI Hiroaki 155676dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 155776dda93cSYan, Zheng void __user *arg) 155876dda93cSYan, Zheng { 155976dda93cSYan, Zheng struct dentry *parent = fdentry(file); 156076dda93cSYan, Zheng struct dentry *dentry; 156176dda93cSYan, Zheng struct inode *dir = parent->d_inode; 156276dda93cSYan, Zheng struct inode *inode; 156376dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 156476dda93cSYan, Zheng struct btrfs_root *dest = NULL; 156576dda93cSYan, Zheng struct btrfs_ioctl_vol_args *vol_args; 156676dda93cSYan, Zheng struct btrfs_trans_handle *trans; 156776dda93cSYan, Zheng int namelen; 156876dda93cSYan, Zheng int ret; 156976dda93cSYan, Zheng int err = 0; 157076dda93cSYan, Zheng 157176dda93cSYan, Zheng vol_args = memdup_user(arg, sizeof(*vol_args)); 157276dda93cSYan, Zheng if (IS_ERR(vol_args)) 157376dda93cSYan, Zheng return PTR_ERR(vol_args); 157476dda93cSYan, Zheng 157576dda93cSYan, Zheng vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 157676dda93cSYan, Zheng namelen = strlen(vol_args->name); 157776dda93cSYan, Zheng if (strchr(vol_args->name, '/') || 157876dda93cSYan, Zheng strncmp(vol_args->name, "..", namelen) == 0) { 157976dda93cSYan, Zheng err = -EINVAL; 158076dda93cSYan, Zheng goto out; 158176dda93cSYan, Zheng } 158276dda93cSYan, Zheng 158376dda93cSYan, Zheng err = mnt_want_write(file->f_path.mnt); 158476dda93cSYan, Zheng if (err) 158576dda93cSYan, Zheng goto out; 158676dda93cSYan, Zheng 158776dda93cSYan, Zheng mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT); 158876dda93cSYan, Zheng dentry = lookup_one_len(vol_args->name, parent, namelen); 158976dda93cSYan, Zheng if (IS_ERR(dentry)) { 159076dda93cSYan, Zheng err = PTR_ERR(dentry); 159176dda93cSYan, Zheng goto out_unlock_dir; 159276dda93cSYan, Zheng } 159376dda93cSYan, Zheng 159476dda93cSYan, Zheng if (!dentry->d_inode) { 159576dda93cSYan, Zheng err = -ENOENT; 159676dda93cSYan, Zheng goto out_dput; 159776dda93cSYan, Zheng } 159876dda93cSYan, Zheng 159976dda93cSYan, Zheng inode = dentry->d_inode; 16004260f7c7SSage Weil dest = BTRFS_I(inode)->root; 16014260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)){ 16024260f7c7SSage Weil /* 16034260f7c7SSage Weil * Regular user. Only allow this with a special mount 16044260f7c7SSage Weil * option, when the user has write+exec access to the 16054260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 16064260f7c7SSage Weil * allowed. 16074260f7c7SSage Weil * 16084260f7c7SSage Weil * Note that this is _not_ check that the subvol is 16094260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 16104260f7c7SSage Weil * otherwise be able to delete. 16114260f7c7SSage Weil * 16124260f7c7SSage Weil * Users who want to delete empty subvols should try 16134260f7c7SSage Weil * rmdir(2). 16144260f7c7SSage Weil */ 16154260f7c7SSage Weil err = -EPERM; 16164260f7c7SSage Weil if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED)) 16174260f7c7SSage Weil goto out_dput; 16184260f7c7SSage Weil 16194260f7c7SSage Weil /* 16204260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 16214260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 16224260f7c7SSage Weil * must be called on the dentry referencing the root 16234260f7c7SSage Weil * of the subvol, not a random directory contained 16244260f7c7SSage Weil * within it. 16254260f7c7SSage Weil */ 16264260f7c7SSage Weil err = -EINVAL; 16274260f7c7SSage Weil if (root == dest) 16284260f7c7SSage Weil goto out_dput; 16294260f7c7SSage Weil 16304260f7c7SSage Weil err = inode_permission(inode, MAY_WRITE | MAY_EXEC); 16314260f7c7SSage Weil if (err) 16324260f7c7SSage Weil goto out_dput; 16334260f7c7SSage Weil 16344260f7c7SSage Weil /* check if subvolume may be deleted by a non-root user */ 16354260f7c7SSage Weil err = btrfs_may_delete(dir, dentry, 1); 16364260f7c7SSage Weil if (err) 16374260f7c7SSage Weil goto out_dput; 16384260f7c7SSage Weil } 16394260f7c7SSage Weil 164076dda93cSYan, Zheng if (inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) { 164176dda93cSYan, Zheng err = -EINVAL; 164276dda93cSYan, Zheng goto out_dput; 164376dda93cSYan, Zheng } 164476dda93cSYan, Zheng 164576dda93cSYan, Zheng mutex_lock(&inode->i_mutex); 164676dda93cSYan, Zheng err = d_invalidate(dentry); 164776dda93cSYan, Zheng if (err) 164876dda93cSYan, Zheng goto out_unlock; 164976dda93cSYan, Zheng 165076dda93cSYan, Zheng down_write(&root->fs_info->subvol_sem); 165176dda93cSYan, Zheng 165276dda93cSYan, Zheng err = may_destroy_subvol(dest); 165376dda93cSYan, Zheng if (err) 165476dda93cSYan, Zheng goto out_up_write; 165576dda93cSYan, Zheng 1656a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 1657a22285a6SYan, Zheng if (IS_ERR(trans)) { 1658a22285a6SYan, Zheng err = PTR_ERR(trans); 1659d327099aSDan Carpenter goto out_up_write; 1660a22285a6SYan, Zheng } 1661a22285a6SYan, Zheng trans->block_rsv = &root->fs_info->global_block_rsv; 1662a22285a6SYan, Zheng 166376dda93cSYan, Zheng ret = btrfs_unlink_subvol(trans, root, dir, 166476dda93cSYan, Zheng dest->root_key.objectid, 166576dda93cSYan, Zheng dentry->d_name.name, 166676dda93cSYan, Zheng dentry->d_name.len); 166776dda93cSYan, Zheng BUG_ON(ret); 166876dda93cSYan, Zheng 166976dda93cSYan, Zheng btrfs_record_root_in_trans(trans, dest); 167076dda93cSYan, Zheng 167176dda93cSYan, Zheng memset(&dest->root_item.drop_progress, 0, 167276dda93cSYan, Zheng sizeof(dest->root_item.drop_progress)); 167376dda93cSYan, Zheng dest->root_item.drop_level = 0; 167476dda93cSYan, Zheng btrfs_set_root_refs(&dest->root_item, 0); 167576dda93cSYan, Zheng 1676d68fc57bSYan, Zheng if (!xchg(&dest->orphan_item_inserted, 1)) { 167776dda93cSYan, Zheng ret = btrfs_insert_orphan_item(trans, 167876dda93cSYan, Zheng root->fs_info->tree_root, 167976dda93cSYan, Zheng dest->root_key.objectid); 168076dda93cSYan, Zheng BUG_ON(ret); 1681d68fc57bSYan, Zheng } 168276dda93cSYan, Zheng 1683531cb13fSSage Weil ret = btrfs_end_transaction(trans, root); 168476dda93cSYan, Zheng BUG_ON(ret); 168576dda93cSYan, Zheng inode->i_flags |= S_DEAD; 168676dda93cSYan, Zheng out_up_write: 168776dda93cSYan, Zheng up_write(&root->fs_info->subvol_sem); 168876dda93cSYan, Zheng out_unlock: 168976dda93cSYan, Zheng mutex_unlock(&inode->i_mutex); 169076dda93cSYan, Zheng if (!err) { 1691efefb143SYan, Zheng shrink_dcache_sb(root->fs_info->sb); 169276dda93cSYan, Zheng btrfs_invalidate_inodes(dest); 169376dda93cSYan, Zheng d_delete(dentry); 169476dda93cSYan, Zheng } 169576dda93cSYan, Zheng out_dput: 169676dda93cSYan, Zheng dput(dentry); 169776dda93cSYan, Zheng out_unlock_dir: 169876dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 169976dda93cSYan, Zheng mnt_drop_write(file->f_path.mnt); 170076dda93cSYan, Zheng out: 170176dda93cSYan, Zheng kfree(vol_args); 170276dda93cSYan, Zheng return err; 170376dda93cSYan, Zheng } 170476dda93cSYan, Zheng 17051e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 1706f46b5a66SChristoph Hellwig { 1707f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 1708f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 17091e701a32SChris Mason struct btrfs_ioctl_defrag_range_args *range; 1710c146afadSYan Zheng int ret; 1711c146afadSYan Zheng 1712b83cc969SLi Zefan if (btrfs_root_readonly(root)) 1713b83cc969SLi Zefan return -EROFS; 1714b83cc969SLi Zefan 1715c146afadSYan Zheng ret = mnt_want_write(file->f_path.mnt); 1716c146afadSYan Zheng if (ret) 1717c146afadSYan Zheng return ret; 1718f46b5a66SChristoph Hellwig 1719f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1720f46b5a66SChristoph Hellwig case S_IFDIR: 1721e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 1722e441d54dSChris Mason ret = -EPERM; 1723e441d54dSChris Mason goto out; 1724e441d54dSChris Mason } 17258929ecfaSYan, Zheng ret = btrfs_defrag_root(root, 0); 17268929ecfaSYan, Zheng if (ret) 17278929ecfaSYan, Zheng goto out; 17288929ecfaSYan, Zheng ret = btrfs_defrag_root(root->fs_info->extent_root, 0); 1729f46b5a66SChristoph Hellwig break; 1730f46b5a66SChristoph Hellwig case S_IFREG: 1731e441d54dSChris Mason if (!(file->f_mode & FMODE_WRITE)) { 1732e441d54dSChris Mason ret = -EINVAL; 1733e441d54dSChris Mason goto out; 1734e441d54dSChris Mason } 17351e701a32SChris Mason 17361e701a32SChris Mason range = kzalloc(sizeof(*range), GFP_KERNEL); 17371e701a32SChris Mason if (!range) { 17381e701a32SChris Mason ret = -ENOMEM; 17391e701a32SChris Mason goto out; 17401e701a32SChris Mason } 17411e701a32SChris Mason 17421e701a32SChris Mason if (argp) { 17431e701a32SChris Mason if (copy_from_user(range, argp, 17441e701a32SChris Mason sizeof(*range))) { 17451e701a32SChris Mason ret = -EFAULT; 17461e701a32SChris Mason kfree(range); 1747683be16eSDan Carpenter goto out; 17481e701a32SChris Mason } 17491e701a32SChris Mason /* compression requires us to start the IO */ 17501e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 17511e701a32SChris Mason range->flags |= BTRFS_DEFRAG_RANGE_START_IO; 17521e701a32SChris Mason range->extent_thresh = (u32)-1; 17531e701a32SChris Mason } 17541e701a32SChris Mason } else { 17551e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 17561e701a32SChris Mason range->len = (u64)-1; 17571e701a32SChris Mason } 17588929ecfaSYan, Zheng ret = btrfs_defrag_file(file, range); 17591e701a32SChris Mason kfree(range); 1760f46b5a66SChristoph Hellwig break; 17618929ecfaSYan, Zheng default: 17628929ecfaSYan, Zheng ret = -EINVAL; 1763f46b5a66SChristoph Hellwig } 1764e441d54dSChris Mason out: 1765ab67b7c1SYan Zheng mnt_drop_write(file->f_path.mnt); 1766e441d54dSChris Mason return ret; 1767f46b5a66SChristoph Hellwig } 1768f46b5a66SChristoph Hellwig 1769b2950863SChristoph Hellwig static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg) 1770f46b5a66SChristoph Hellwig { 1771f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1772f46b5a66SChristoph Hellwig int ret; 1773f46b5a66SChristoph Hellwig 1774e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1775e441d54dSChris Mason return -EPERM; 1776e441d54dSChris Mason 1777dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1778dae7b665SLi Zefan if (IS_ERR(vol_args)) 1779dae7b665SLi Zefan return PTR_ERR(vol_args); 1780f46b5a66SChristoph Hellwig 17815516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1782f46b5a66SChristoph Hellwig ret = btrfs_init_new_device(root, vol_args->name); 1783f46b5a66SChristoph Hellwig 1784f46b5a66SChristoph Hellwig kfree(vol_args); 1785f46b5a66SChristoph Hellwig return ret; 1786f46b5a66SChristoph Hellwig } 1787f46b5a66SChristoph Hellwig 1788b2950863SChristoph Hellwig static long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg) 1789f46b5a66SChristoph Hellwig { 1790f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1791f46b5a66SChristoph Hellwig int ret; 1792f46b5a66SChristoph Hellwig 1793e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1794e441d54dSChris Mason return -EPERM; 1795e441d54dSChris Mason 1796c146afadSYan Zheng if (root->fs_info->sb->s_flags & MS_RDONLY) 1797c146afadSYan Zheng return -EROFS; 1798c146afadSYan Zheng 1799dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1800dae7b665SLi Zefan if (IS_ERR(vol_args)) 1801dae7b665SLi Zefan return PTR_ERR(vol_args); 1802f46b5a66SChristoph Hellwig 18035516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1804f46b5a66SChristoph Hellwig ret = btrfs_rm_device(root, vol_args->name); 1805f46b5a66SChristoph Hellwig 1806f46b5a66SChristoph Hellwig kfree(vol_args); 1807f46b5a66SChristoph Hellwig return ret; 1808f46b5a66SChristoph Hellwig } 1809f46b5a66SChristoph Hellwig 181076dda93cSYan, Zheng static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, 1811b2950863SChristoph Hellwig u64 off, u64 olen, u64 destoff) 1812f46b5a66SChristoph Hellwig { 1813f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 1814f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 1815f46b5a66SChristoph Hellwig struct file *src_file; 1816f46b5a66SChristoph Hellwig struct inode *src; 1817f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1818f46b5a66SChristoph Hellwig struct btrfs_path *path; 1819f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 1820ae01a0abSYan Zheng char *buf; 1821ae01a0abSYan Zheng struct btrfs_key key; 1822f46b5a66SChristoph Hellwig u32 nritems; 1823f46b5a66SChristoph Hellwig int slot; 1824ae01a0abSYan Zheng int ret; 1825c5c9cd4dSSage Weil u64 len = olen; 1826c5c9cd4dSSage Weil u64 bs = root->fs_info->sb->s_blocksize; 1827c5c9cd4dSSage Weil u64 hint_byte; 1828d20f7043SChris Mason 1829c5c9cd4dSSage Weil /* 1830c5c9cd4dSSage Weil * TODO: 1831c5c9cd4dSSage Weil * - split compressed inline extents. annoying: we need to 1832c5c9cd4dSSage Weil * decompress into destination's address_space (the file offset 1833c5c9cd4dSSage Weil * may change, so source mapping won't do), then recompress (or 1834c5c9cd4dSSage Weil * otherwise reinsert) a subrange. 1835c5c9cd4dSSage Weil * - allow ranges within the same file to be cloned (provided 1836c5c9cd4dSSage Weil * they don't overlap)? 1837c5c9cd4dSSage Weil */ 1838c5c9cd4dSSage Weil 1839e441d54dSChris Mason /* the destination must be opened for writing */ 18402ebc3464SDan Rosenberg if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) 1841e441d54dSChris Mason return -EINVAL; 1842e441d54dSChris Mason 1843b83cc969SLi Zefan if (btrfs_root_readonly(root)) 1844b83cc969SLi Zefan return -EROFS; 1845b83cc969SLi Zefan 1846c146afadSYan Zheng ret = mnt_want_write(file->f_path.mnt); 1847c146afadSYan Zheng if (ret) 1848c146afadSYan Zheng return ret; 1849c146afadSYan Zheng 1850c5c9cd4dSSage Weil src_file = fget(srcfd); 1851ab67b7c1SYan Zheng if (!src_file) { 1852ab67b7c1SYan Zheng ret = -EBADF; 1853ab67b7c1SYan Zheng goto out_drop_write; 1854ab67b7c1SYan Zheng } 18555dc64164SDan Rosenberg 1856f46b5a66SChristoph Hellwig src = src_file->f_dentry->d_inode; 1857f46b5a66SChristoph Hellwig 1858c5c9cd4dSSage Weil ret = -EINVAL; 1859c5c9cd4dSSage Weil if (src == inode) 1860c5c9cd4dSSage Weil goto out_fput; 1861c5c9cd4dSSage Weil 18625dc64164SDan Rosenberg /* the src must be open for reading */ 18635dc64164SDan Rosenberg if (!(src_file->f_mode & FMODE_READ)) 18645dc64164SDan Rosenberg goto out_fput; 18655dc64164SDan Rosenberg 1866ae01a0abSYan Zheng ret = -EISDIR; 1867ae01a0abSYan Zheng if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode)) 1868f46b5a66SChristoph Hellwig goto out_fput; 1869f46b5a66SChristoph Hellwig 1870ae01a0abSYan Zheng ret = -EXDEV; 1871ae01a0abSYan Zheng if (src->i_sb != inode->i_sb || BTRFS_I(src)->root != root) 1872ae01a0abSYan Zheng goto out_fput; 1873ae01a0abSYan Zheng 1874ae01a0abSYan Zheng ret = -ENOMEM; 1875ae01a0abSYan Zheng buf = vmalloc(btrfs_level_size(root, 0)); 1876ae01a0abSYan Zheng if (!buf) 1877ae01a0abSYan Zheng goto out_fput; 1878ae01a0abSYan Zheng 1879ae01a0abSYan Zheng path = btrfs_alloc_path(); 1880ae01a0abSYan Zheng if (!path) { 1881ae01a0abSYan Zheng vfree(buf); 1882ae01a0abSYan Zheng goto out_fput; 1883ae01a0abSYan Zheng } 1884ae01a0abSYan Zheng path->reada = 2; 1885ae01a0abSYan Zheng 1886f46b5a66SChristoph Hellwig if (inode < src) { 1887fccdae43SSage Weil mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT); 1888fccdae43SSage Weil mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD); 1889f46b5a66SChristoph Hellwig } else { 1890fccdae43SSage Weil mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT); 1891fccdae43SSage Weil mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); 1892f46b5a66SChristoph Hellwig } 1893f46b5a66SChristoph Hellwig 1894c5c9cd4dSSage Weil /* determine range to clone */ 1895c5c9cd4dSSage Weil ret = -EINVAL; 18962ebc3464SDan Rosenberg if (off + len > src->i_size || off + len < off) 1897f46b5a66SChristoph Hellwig goto out_unlock; 1898c5c9cd4dSSage Weil if (len == 0) 1899c5c9cd4dSSage Weil olen = len = src->i_size - off; 1900c5c9cd4dSSage Weil /* if we extend to eof, continue to block boundary */ 1901c5c9cd4dSSage Weil if (off + len == src->i_size) 19022a6b8daeSLi Zefan len = ALIGN(src->i_size, bs) - off; 1903c5c9cd4dSSage Weil 1904c5c9cd4dSSage Weil /* verify the end result is block aligned */ 19052a6b8daeSLi Zefan if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) || 19062a6b8daeSLi Zefan !IS_ALIGNED(destoff, bs)) 1907c5c9cd4dSSage Weil goto out_unlock; 1908c5c9cd4dSSage Weil 1909f46b5a66SChristoph Hellwig /* do any pending delalloc/csum calc on src, one way or 1910f46b5a66SChristoph Hellwig another, and lock file content */ 1911f46b5a66SChristoph Hellwig while (1) { 191231840ae1SZheng Yan struct btrfs_ordered_extent *ordered; 1913c5c9cd4dSSage Weil lock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS); 19149a019196SSage Weil ordered = btrfs_lookup_first_ordered_extent(src, off+len); 19159a019196SSage Weil if (!ordered && 19169a019196SSage Weil !test_range_bit(&BTRFS_I(src)->io_tree, off, off+len, 19179a019196SSage Weil EXTENT_DELALLOC, 0, NULL)) 1918f46b5a66SChristoph Hellwig break; 1919c5c9cd4dSSage Weil unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS); 1920ae01a0abSYan Zheng if (ordered) 1921ae01a0abSYan Zheng btrfs_put_ordered_extent(ordered); 19229a019196SSage Weil btrfs_wait_ordered_range(src, off, len); 1923f46b5a66SChristoph Hellwig } 1924f46b5a66SChristoph Hellwig 1925c5c9cd4dSSage Weil /* clone data */ 1926f46b5a66SChristoph Hellwig key.objectid = src->i_ino; 1927ae01a0abSYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 1928ae01a0abSYan Zheng key.offset = 0; 1929f46b5a66SChristoph Hellwig 1930f46b5a66SChristoph Hellwig while (1) { 1931f46b5a66SChristoph Hellwig /* 1932f46b5a66SChristoph Hellwig * note the key will change type as we walk through the 1933f46b5a66SChristoph Hellwig * tree. 1934f46b5a66SChristoph Hellwig */ 1935a22285a6SYan, Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1936f46b5a66SChristoph Hellwig if (ret < 0) 1937f46b5a66SChristoph Hellwig goto out; 1938f46b5a66SChristoph Hellwig 1939ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 1940ae01a0abSYan Zheng if (path->slots[0] >= nritems) { 1941f46b5a66SChristoph Hellwig ret = btrfs_next_leaf(root, path); 1942f46b5a66SChristoph Hellwig if (ret < 0) 1943f46b5a66SChristoph Hellwig goto out; 1944f46b5a66SChristoph Hellwig if (ret > 0) 1945f46b5a66SChristoph Hellwig break; 1946ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 1947f46b5a66SChristoph Hellwig } 1948f46b5a66SChristoph Hellwig leaf = path->nodes[0]; 1949f46b5a66SChristoph Hellwig slot = path->slots[0]; 1950f46b5a66SChristoph Hellwig 1951ae01a0abSYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 1952d20f7043SChris Mason if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY || 1953f46b5a66SChristoph Hellwig key.objectid != src->i_ino) 1954f46b5a66SChristoph Hellwig break; 1955f46b5a66SChristoph Hellwig 1956c5c9cd4dSSage Weil if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) { 1957c5c9cd4dSSage Weil struct btrfs_file_extent_item *extent; 1958c5c9cd4dSSage Weil int type; 195931840ae1SZheng Yan u32 size; 196031840ae1SZheng Yan struct btrfs_key new_key; 1961c5c9cd4dSSage Weil u64 disko = 0, diskl = 0; 1962c5c9cd4dSSage Weil u64 datao = 0, datal = 0; 1963c5c9cd4dSSage Weil u8 comp; 1964b5384d48SSage Weil u64 endoff; 196531840ae1SZheng Yan 196631840ae1SZheng Yan size = btrfs_item_size_nr(leaf, slot); 196731840ae1SZheng Yan read_extent_buffer(leaf, buf, 196831840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 196931840ae1SZheng Yan size); 1970c5c9cd4dSSage Weil 1971c5c9cd4dSSage Weil extent = btrfs_item_ptr(leaf, slot, 1972c5c9cd4dSSage Weil struct btrfs_file_extent_item); 1973c5c9cd4dSSage Weil comp = btrfs_file_extent_compression(leaf, extent); 1974c5c9cd4dSSage Weil type = btrfs_file_extent_type(leaf, extent); 1975c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 1976c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 1977d397712bSChris Mason disko = btrfs_file_extent_disk_bytenr(leaf, 1978d397712bSChris Mason extent); 1979d397712bSChris Mason diskl = btrfs_file_extent_disk_num_bytes(leaf, 1980d397712bSChris Mason extent); 1981c5c9cd4dSSage Weil datao = btrfs_file_extent_offset(leaf, extent); 1982d397712bSChris Mason datal = btrfs_file_extent_num_bytes(leaf, 1983d397712bSChris Mason extent); 1984c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 1985c5c9cd4dSSage Weil /* take upper bound, may be compressed */ 1986c5c9cd4dSSage Weil datal = btrfs_file_extent_ram_bytes(leaf, 1987c5c9cd4dSSage Weil extent); 1988c5c9cd4dSSage Weil } 198931840ae1SZheng Yan btrfs_release_path(root, path); 199031840ae1SZheng Yan 1991050006a7SSage Weil if (key.offset + datal <= off || 1992c5c9cd4dSSage Weil key.offset >= off+len) 1993c5c9cd4dSSage Weil goto next; 1994c5c9cd4dSSage Weil 199531840ae1SZheng Yan memcpy(&new_key, &key, sizeof(new_key)); 199631840ae1SZheng Yan new_key.objectid = inode->i_ino; 19974d728ec7SLi Zefan if (off <= key.offset) 1998c5c9cd4dSSage Weil new_key.offset = key.offset + destoff - off; 19994d728ec7SLi Zefan else 20004d728ec7SLi Zefan new_key.offset = destoff; 2001c5c9cd4dSSage Weil 2002a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 1); 2003a22285a6SYan, Zheng if (IS_ERR(trans)) { 2004a22285a6SYan, Zheng ret = PTR_ERR(trans); 2005a22285a6SYan, Zheng goto out; 2006a22285a6SYan, Zheng } 2007a22285a6SYan, Zheng 2008c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 2009c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 2010a22285a6SYan, Zheng if (off > key.offset) { 2011a22285a6SYan, Zheng datao += off - key.offset; 2012a22285a6SYan, Zheng datal -= off - key.offset; 2013a22285a6SYan, Zheng } 2014a22285a6SYan, Zheng 2015a22285a6SYan, Zheng if (key.offset + datal > off + len) 2016a22285a6SYan, Zheng datal = off + len - key.offset; 2017a22285a6SYan, Zheng 2018a22285a6SYan, Zheng ret = btrfs_drop_extents(trans, inode, 2019a22285a6SYan, Zheng new_key.offset, 2020a22285a6SYan, Zheng new_key.offset + datal, 2021a22285a6SYan, Zheng &hint_byte, 1); 2022a22285a6SYan, Zheng BUG_ON(ret); 2023a22285a6SYan, Zheng 202431840ae1SZheng Yan ret = btrfs_insert_empty_item(trans, root, path, 202531840ae1SZheng Yan &new_key, size); 2026a22285a6SYan, Zheng BUG_ON(ret); 202731840ae1SZheng Yan 202831840ae1SZheng Yan leaf = path->nodes[0]; 202931840ae1SZheng Yan slot = path->slots[0]; 203031840ae1SZheng Yan write_extent_buffer(leaf, buf, 203131840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 203231840ae1SZheng Yan size); 2033ae01a0abSYan Zheng 2034f46b5a66SChristoph Hellwig extent = btrfs_item_ptr(leaf, slot, 2035f46b5a66SChristoph Hellwig struct btrfs_file_extent_item); 2036c5c9cd4dSSage Weil 2037c5c9cd4dSSage Weil /* disko == 0 means it's a hole */ 2038c5c9cd4dSSage Weil if (!disko) 2039c5c9cd4dSSage Weil datao = 0; 2040c5c9cd4dSSage Weil 2041c5c9cd4dSSage Weil btrfs_set_file_extent_offset(leaf, extent, 2042c5c9cd4dSSage Weil datao); 2043c5c9cd4dSSage Weil btrfs_set_file_extent_num_bytes(leaf, extent, 2044c5c9cd4dSSage Weil datal); 2045c5c9cd4dSSage Weil if (disko) { 2046c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 2047ae01a0abSYan Zheng ret = btrfs_inc_extent_ref(trans, root, 20485d4f98a2SYan Zheng disko, diskl, 0, 2049f46b5a66SChristoph Hellwig root->root_key.objectid, 20505d4f98a2SYan Zheng inode->i_ino, 20515d4f98a2SYan Zheng new_key.offset - datao); 2052ae01a0abSYan Zheng BUG_ON(ret); 2053f46b5a66SChristoph Hellwig } 2054c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 2055c5c9cd4dSSage Weil u64 skip = 0; 2056c5c9cd4dSSage Weil u64 trim = 0; 2057c5c9cd4dSSage Weil if (off > key.offset) { 2058c5c9cd4dSSage Weil skip = off - key.offset; 2059c5c9cd4dSSage Weil new_key.offset += skip; 206031840ae1SZheng Yan } 2061d397712bSChris Mason 2062c5c9cd4dSSage Weil if (key.offset + datal > off+len) 2063c5c9cd4dSSage Weil trim = key.offset + datal - (off+len); 2064d397712bSChris Mason 2065c5c9cd4dSSage Weil if (comp && (skip || trim)) { 2066c5c9cd4dSSage Weil ret = -EINVAL; 2067a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 2068c5c9cd4dSSage Weil goto out; 206931840ae1SZheng Yan } 2070c5c9cd4dSSage Weil size -= skip + trim; 2071c5c9cd4dSSage Weil datal -= skip + trim; 2072a22285a6SYan, Zheng 2073a22285a6SYan, Zheng ret = btrfs_drop_extents(trans, inode, 2074a22285a6SYan, Zheng new_key.offset, 2075a22285a6SYan, Zheng new_key.offset + datal, 2076a22285a6SYan, Zheng &hint_byte, 1); 2077a22285a6SYan, Zheng BUG_ON(ret); 2078a22285a6SYan, Zheng 2079c5c9cd4dSSage Weil ret = btrfs_insert_empty_item(trans, root, path, 2080c5c9cd4dSSage Weil &new_key, size); 2081a22285a6SYan, Zheng BUG_ON(ret); 2082c5c9cd4dSSage Weil 2083c5c9cd4dSSage Weil if (skip) { 2084d397712bSChris Mason u32 start = 2085d397712bSChris Mason btrfs_file_extent_calc_inline_size(0); 2086c5c9cd4dSSage Weil memmove(buf+start, buf+start+skip, 2087c5c9cd4dSSage Weil datal); 2088c5c9cd4dSSage Weil } 2089c5c9cd4dSSage Weil 2090c5c9cd4dSSage Weil leaf = path->nodes[0]; 2091c5c9cd4dSSage Weil slot = path->slots[0]; 2092c5c9cd4dSSage Weil write_extent_buffer(leaf, buf, 2093c5c9cd4dSSage Weil btrfs_item_ptr_offset(leaf, slot), 2094c5c9cd4dSSage Weil size); 2095c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 2096c5c9cd4dSSage Weil } 2097c5c9cd4dSSage Weil 2098c5c9cd4dSSage Weil btrfs_mark_buffer_dirty(leaf); 2099a22285a6SYan, Zheng btrfs_release_path(root, path); 2100c5c9cd4dSSage Weil 2101a22285a6SYan, Zheng inode->i_mtime = inode->i_ctime = CURRENT_TIME; 2102b5384d48SSage Weil 2103b5384d48SSage Weil /* 2104b5384d48SSage Weil * we round up to the block size at eof when 2105b5384d48SSage Weil * determining which extents to clone above, 2106b5384d48SSage Weil * but shouldn't round up the file size 2107b5384d48SSage Weil */ 2108b5384d48SSage Weil endoff = new_key.offset + datal; 21095f3888ffSLi Zefan if (endoff > destoff+olen) 21105f3888ffSLi Zefan endoff = destoff+olen; 2111b5384d48SSage Weil if (endoff > inode->i_size) 2112b5384d48SSage Weil btrfs_i_size_write(inode, endoff); 2113b5384d48SSage Weil 2114a22285a6SYan, Zheng BTRFS_I(inode)->flags = BTRFS_I(src)->flags; 2115a22285a6SYan, Zheng ret = btrfs_update_inode(trans, root, inode); 2116a22285a6SYan, Zheng BUG_ON(ret); 2117a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 2118a22285a6SYan, Zheng } 2119c5c9cd4dSSage Weil next: 212031840ae1SZheng Yan btrfs_release_path(root, path); 2121ae01a0abSYan Zheng key.offset++; 2122ae01a0abSYan Zheng } 2123f46b5a66SChristoph Hellwig ret = 0; 2124f46b5a66SChristoph Hellwig out: 2125ae01a0abSYan Zheng btrfs_release_path(root, path); 2126c5c9cd4dSSage Weil unlock_extent(&BTRFS_I(src)->io_tree, off, off+len, GFP_NOFS); 2127f46b5a66SChristoph Hellwig out_unlock: 2128f46b5a66SChristoph Hellwig mutex_unlock(&src->i_mutex); 2129f46b5a66SChristoph Hellwig mutex_unlock(&inode->i_mutex); 2130ae01a0abSYan Zheng vfree(buf); 2131ae01a0abSYan Zheng btrfs_free_path(path); 2132f46b5a66SChristoph Hellwig out_fput: 2133f46b5a66SChristoph Hellwig fput(src_file); 2134ab67b7c1SYan Zheng out_drop_write: 2135ab67b7c1SYan Zheng mnt_drop_write(file->f_path.mnt); 2136f46b5a66SChristoph Hellwig return ret; 2137f46b5a66SChristoph Hellwig } 2138f46b5a66SChristoph Hellwig 21397a865e8aSChristoph Hellwig static long btrfs_ioctl_clone_range(struct file *file, void __user *argp) 2140c5c9cd4dSSage Weil { 2141c5c9cd4dSSage Weil struct btrfs_ioctl_clone_range_args args; 2142c5c9cd4dSSage Weil 21437a865e8aSChristoph Hellwig if (copy_from_user(&args, argp, sizeof(args))) 2144c5c9cd4dSSage Weil return -EFAULT; 2145c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, args.src_fd, args.src_offset, 2146c5c9cd4dSSage Weil args.src_length, args.dest_offset); 2147c5c9cd4dSSage Weil } 2148c5c9cd4dSSage Weil 2149f46b5a66SChristoph Hellwig /* 2150f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 2151f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 2152f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 2153f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 2154f46b5a66SChristoph Hellwig */ 2155b2950863SChristoph Hellwig static long btrfs_ioctl_trans_start(struct file *file) 2156f46b5a66SChristoph Hellwig { 2157f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2158f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 2159f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 21601ab86aedSSage Weil int ret; 2161f46b5a66SChristoph Hellwig 21621ab86aedSSage Weil ret = -EPERM; 2163df5b5520SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 2164f46b5a66SChristoph Hellwig goto out; 21651ab86aedSSage Weil 21661ab86aedSSage Weil ret = -EINPROGRESS; 21671ab86aedSSage Weil if (file->private_data) 21681ab86aedSSage Weil goto out; 21699ca9ee09SSage Weil 2170b83cc969SLi Zefan ret = -EROFS; 2171b83cc969SLi Zefan if (btrfs_root_readonly(root)) 2172b83cc969SLi Zefan goto out; 2173b83cc969SLi Zefan 2174c146afadSYan Zheng ret = mnt_want_write(file->f_path.mnt); 2175c146afadSYan Zheng if (ret) 2176c146afadSYan Zheng goto out; 2177c146afadSYan Zheng 21789ca9ee09SSage Weil mutex_lock(&root->fs_info->trans_mutex); 21799ca9ee09SSage Weil root->fs_info->open_ioctl_trans++; 21809ca9ee09SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 21819ca9ee09SSage Weil 2182f46b5a66SChristoph Hellwig ret = -ENOMEM; 21831ab86aedSSage Weil trans = btrfs_start_ioctl_transaction(root, 0); 2184abd30bb0STsutomu Itoh if (IS_ERR(trans)) 21851ab86aedSSage Weil goto out_drop; 21861ab86aedSSage Weil 21871ab86aedSSage Weil file->private_data = trans; 21881ab86aedSSage Weil return 0; 21891ab86aedSSage Weil 21901ab86aedSSage Weil out_drop: 21911ab86aedSSage Weil mutex_lock(&root->fs_info->trans_mutex); 21921ab86aedSSage Weil root->fs_info->open_ioctl_trans--; 21931ab86aedSSage Weil mutex_unlock(&root->fs_info->trans_mutex); 21941ab86aedSSage Weil mnt_drop_write(file->f_path.mnt); 2195f46b5a66SChristoph Hellwig out: 2196f46b5a66SChristoph Hellwig return ret; 2197f46b5a66SChristoph Hellwig } 2198f46b5a66SChristoph Hellwig 21996ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 22006ef5ed0dSJosef Bacik { 22016ef5ed0dSJosef Bacik struct inode *inode = fdentry(file)->d_inode; 22026ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 22036ef5ed0dSJosef Bacik struct btrfs_root *new_root; 22046ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 22056ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 22066ef5ed0dSJosef Bacik struct btrfs_path *path; 22076ef5ed0dSJosef Bacik struct btrfs_key location; 22086ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 22096ef5ed0dSJosef Bacik struct btrfs_super_block *disk_super; 22106ef5ed0dSJosef Bacik u64 features; 22116ef5ed0dSJosef Bacik u64 objectid = 0; 22126ef5ed0dSJosef Bacik u64 dir_id; 22136ef5ed0dSJosef Bacik 22146ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 22156ef5ed0dSJosef Bacik return -EPERM; 22166ef5ed0dSJosef Bacik 22176ef5ed0dSJosef Bacik if (copy_from_user(&objectid, argp, sizeof(objectid))) 22186ef5ed0dSJosef Bacik return -EFAULT; 22196ef5ed0dSJosef Bacik 22206ef5ed0dSJosef Bacik if (!objectid) 22216ef5ed0dSJosef Bacik objectid = root->root_key.objectid; 22226ef5ed0dSJosef Bacik 22236ef5ed0dSJosef Bacik location.objectid = objectid; 22246ef5ed0dSJosef Bacik location.type = BTRFS_ROOT_ITEM_KEY; 22256ef5ed0dSJosef Bacik location.offset = (u64)-1; 22266ef5ed0dSJosef Bacik 22276ef5ed0dSJosef Bacik new_root = btrfs_read_fs_root_no_name(root->fs_info, &location); 22286ef5ed0dSJosef Bacik if (IS_ERR(new_root)) 22296ef5ed0dSJosef Bacik return PTR_ERR(new_root); 22306ef5ed0dSJosef Bacik 22316ef5ed0dSJosef Bacik if (btrfs_root_refs(&new_root->root_item) == 0) 22326ef5ed0dSJosef Bacik return -ENOENT; 22336ef5ed0dSJosef Bacik 22346ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 22356ef5ed0dSJosef Bacik if (!path) 22366ef5ed0dSJosef Bacik return -ENOMEM; 22376ef5ed0dSJosef Bacik path->leave_spinning = 1; 22386ef5ed0dSJosef Bacik 22396ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 224098d5dc13STsutomu Itoh if (IS_ERR(trans)) { 22416ef5ed0dSJosef Bacik btrfs_free_path(path); 224298d5dc13STsutomu Itoh return PTR_ERR(trans); 22436ef5ed0dSJosef Bacik } 22446ef5ed0dSJosef Bacik 22456ef5ed0dSJosef Bacik dir_id = btrfs_super_root_dir(&root->fs_info->super_copy); 22466ef5ed0dSJosef Bacik di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path, 22476ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 2248cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 22496ef5ed0dSJosef Bacik btrfs_free_path(path); 22506ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 22516ef5ed0dSJosef Bacik printk(KERN_ERR "Umm, you don't have the default dir item, " 22526ef5ed0dSJosef Bacik "this isn't going to work\n"); 22536ef5ed0dSJosef Bacik return -ENOENT; 22546ef5ed0dSJosef Bacik } 22556ef5ed0dSJosef Bacik 22566ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 22576ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 22586ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 22596ef5ed0dSJosef Bacik btrfs_free_path(path); 22606ef5ed0dSJosef Bacik 22616ef5ed0dSJosef Bacik disk_super = &root->fs_info->super_copy; 22626ef5ed0dSJosef Bacik features = btrfs_super_incompat_flags(disk_super); 22636ef5ed0dSJosef Bacik if (!(features & BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)) { 22646ef5ed0dSJosef Bacik features |= BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL; 22656ef5ed0dSJosef Bacik btrfs_set_super_incompat_flags(disk_super, features); 22666ef5ed0dSJosef Bacik } 22676ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 22686ef5ed0dSJosef Bacik 22696ef5ed0dSJosef Bacik return 0; 22706ef5ed0dSJosef Bacik } 22716ef5ed0dSJosef Bacik 2272bf5fc093SJosef Bacik static void get_block_group_info(struct list_head *groups_list, 2273bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 2274bf5fc093SJosef Bacik { 2275bf5fc093SJosef Bacik struct btrfs_block_group_cache *block_group; 2276bf5fc093SJosef Bacik 2277bf5fc093SJosef Bacik space->total_bytes = 0; 2278bf5fc093SJosef Bacik space->used_bytes = 0; 2279bf5fc093SJosef Bacik space->flags = 0; 2280bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 2281bf5fc093SJosef Bacik space->flags = block_group->flags; 2282bf5fc093SJosef Bacik space->total_bytes += block_group->key.offset; 2283bf5fc093SJosef Bacik space->used_bytes += 2284bf5fc093SJosef Bacik btrfs_block_group_used(&block_group->item); 2285bf5fc093SJosef Bacik } 2286bf5fc093SJosef Bacik } 2287bf5fc093SJosef Bacik 22881406e432SJosef Bacik long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg) 22891406e432SJosef Bacik { 22901406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 22911406e432SJosef Bacik struct btrfs_ioctl_space_info space; 22921406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 22937fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 229413f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 22951406e432SJosef Bacik struct btrfs_space_info *info; 2296bf5fc093SJosef Bacik u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 2297bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 2298bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 2299bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 2300bf5fc093SJosef Bacik int num_types = 4; 23017fde62bfSChris Mason int alloc_size; 23021406e432SJosef Bacik int ret = 0; 230351788b1bSDan Rosenberg u64 slot_count = 0; 2304bf5fc093SJosef Bacik int i, c; 23051406e432SJosef Bacik 23061406e432SJosef Bacik if (copy_from_user(&space_args, 23071406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 23081406e432SJosef Bacik sizeof(space_args))) 23091406e432SJosef Bacik return -EFAULT; 23101406e432SJosef Bacik 2311bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 2312bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 2313bf5fc093SJosef Bacik 2314bf5fc093SJosef Bacik info = NULL; 23157fde62bfSChris Mason rcu_read_lock(); 2316bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 2317bf5fc093SJosef Bacik list) { 2318bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 2319bf5fc093SJosef Bacik info = tmp; 2320bf5fc093SJosef Bacik break; 2321bf5fc093SJosef Bacik } 2322bf5fc093SJosef Bacik } 23237fde62bfSChris Mason rcu_read_unlock(); 23241406e432SJosef Bacik 2325bf5fc093SJosef Bacik if (!info) 2326bf5fc093SJosef Bacik continue; 2327bf5fc093SJosef Bacik 2328bf5fc093SJosef Bacik down_read(&info->groups_sem); 2329bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 2330bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 2331bf5fc093SJosef Bacik slot_count++; 2332bf5fc093SJosef Bacik } 2333bf5fc093SJosef Bacik up_read(&info->groups_sem); 2334bf5fc093SJosef Bacik } 2335bf5fc093SJosef Bacik 23367fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 23377fde62bfSChris Mason if (space_args.space_slots == 0) { 23387fde62bfSChris Mason space_args.total_spaces = slot_count; 23397fde62bfSChris Mason goto out; 23407fde62bfSChris Mason } 2341bf5fc093SJosef Bacik 234251788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 2343bf5fc093SJosef Bacik 23447fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 2345bf5fc093SJosef Bacik 23467fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 23477fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 23487fde62bfSChris Mason */ 23497fde62bfSChris Mason if (alloc_size > PAGE_CACHE_SIZE) 23507fde62bfSChris Mason return -ENOMEM; 23517fde62bfSChris Mason 23527fde62bfSChris Mason space_args.total_spaces = 0; 23537fde62bfSChris Mason dest = kmalloc(alloc_size, GFP_NOFS); 23547fde62bfSChris Mason if (!dest) 23557fde62bfSChris Mason return -ENOMEM; 23567fde62bfSChris Mason dest_orig = dest; 23577fde62bfSChris Mason 23587fde62bfSChris Mason /* now we have a buffer to copy into */ 2359bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 2360bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 2361bf5fc093SJosef Bacik 236251788b1bSDan Rosenberg if (!slot_count) 236351788b1bSDan Rosenberg break; 236451788b1bSDan Rosenberg 2365bf5fc093SJosef Bacik info = NULL; 23661406e432SJosef Bacik rcu_read_lock(); 2367bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 2368bf5fc093SJosef Bacik list) { 2369bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 2370bf5fc093SJosef Bacik info = tmp; 23717fde62bfSChris Mason break; 2372bf5fc093SJosef Bacik } 2373bf5fc093SJosef Bacik } 2374bf5fc093SJosef Bacik rcu_read_unlock(); 23757fde62bfSChris Mason 2376bf5fc093SJosef Bacik if (!info) 2377bf5fc093SJosef Bacik continue; 2378bf5fc093SJosef Bacik down_read(&info->groups_sem); 2379bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 2380bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 2381bf5fc093SJosef Bacik get_block_group_info(&info->block_groups[c], 2382bf5fc093SJosef Bacik &space); 23837fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 23841406e432SJosef Bacik dest++; 23851406e432SJosef Bacik space_args.total_spaces++; 238651788b1bSDan Rosenberg slot_count--; 23871406e432SJosef Bacik } 238851788b1bSDan Rosenberg if (!slot_count) 238951788b1bSDan Rosenberg break; 2390bf5fc093SJosef Bacik } 2391bf5fc093SJosef Bacik up_read(&info->groups_sem); 2392bf5fc093SJosef Bacik } 23931406e432SJosef Bacik 23947fde62bfSChris Mason user_dest = (struct btrfs_ioctl_space_info *) 23957fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 23967fde62bfSChris Mason 23977fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 23987fde62bfSChris Mason ret = -EFAULT; 23997fde62bfSChris Mason 24007fde62bfSChris Mason kfree(dest_orig); 24017fde62bfSChris Mason out: 24027fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 24031406e432SJosef Bacik ret = -EFAULT; 24041406e432SJosef Bacik 24051406e432SJosef Bacik return ret; 24061406e432SJosef Bacik } 24071406e432SJosef Bacik 2408f46b5a66SChristoph Hellwig /* 2409f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 2410f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 2411f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 2412f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 2413f46b5a66SChristoph Hellwig */ 2414f46b5a66SChristoph Hellwig long btrfs_ioctl_trans_end(struct file *file) 2415f46b5a66SChristoph Hellwig { 2416f46b5a66SChristoph Hellwig struct inode *inode = fdentry(file)->d_inode; 2417f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 2418f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 2419f46b5a66SChristoph Hellwig 2420f46b5a66SChristoph Hellwig trans = file->private_data; 24211ab86aedSSage Weil if (!trans) 24221ab86aedSSage Weil return -EINVAL; 2423b214107eSChristoph Hellwig file->private_data = NULL; 24249ca9ee09SSage Weil 24251ab86aedSSage Weil btrfs_end_transaction(trans, root); 24261ab86aedSSage Weil 24279ca9ee09SSage Weil mutex_lock(&root->fs_info->trans_mutex); 24289ca9ee09SSage Weil root->fs_info->open_ioctl_trans--; 24299ca9ee09SSage Weil mutex_unlock(&root->fs_info->trans_mutex); 24309ca9ee09SSage Weil 2431cfc8ea87SSage Weil mnt_drop_write(file->f_path.mnt); 24321ab86aedSSage Weil return 0; 2433f46b5a66SChristoph Hellwig } 2434f46b5a66SChristoph Hellwig 243546204592SSage Weil static noinline long btrfs_ioctl_start_sync(struct file *file, void __user *argp) 243646204592SSage Weil { 243746204592SSage Weil struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root; 243846204592SSage Weil struct btrfs_trans_handle *trans; 243946204592SSage Weil u64 transid; 2440db5b493aSTsutomu Itoh int ret; 244146204592SSage Weil 244246204592SSage Weil trans = btrfs_start_transaction(root, 0); 244398d5dc13STsutomu Itoh if (IS_ERR(trans)) 244498d5dc13STsutomu Itoh return PTR_ERR(trans); 244546204592SSage Weil transid = trans->transid; 2446db5b493aSTsutomu Itoh ret = btrfs_commit_transaction_async(trans, root, 0); 24478b2b2d3cSTsutomu Itoh if (ret) { 24488b2b2d3cSTsutomu Itoh btrfs_end_transaction(trans, root); 2449db5b493aSTsutomu Itoh return ret; 24508b2b2d3cSTsutomu Itoh } 245146204592SSage Weil 245246204592SSage Weil if (argp) 245346204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 245446204592SSage Weil return -EFAULT; 245546204592SSage Weil return 0; 245646204592SSage Weil } 245746204592SSage Weil 245846204592SSage Weil static noinline long btrfs_ioctl_wait_sync(struct file *file, void __user *argp) 245946204592SSage Weil { 246046204592SSage Weil struct btrfs_root *root = BTRFS_I(file->f_dentry->d_inode)->root; 246146204592SSage Weil u64 transid; 246246204592SSage Weil 246346204592SSage Weil if (argp) { 246446204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 246546204592SSage Weil return -EFAULT; 246646204592SSage Weil } else { 246746204592SSage Weil transid = 0; /* current trans */ 246846204592SSage Weil } 246946204592SSage Weil return btrfs_wait_for_commit(root, transid); 247046204592SSage Weil } 247146204592SSage Weil 2472f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 2473f46b5a66SChristoph Hellwig cmd, unsigned long arg) 2474f46b5a66SChristoph Hellwig { 2475f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root; 24764bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 2477f46b5a66SChristoph Hellwig 2478f46b5a66SChristoph Hellwig switch (cmd) { 24796cbff00fSChristoph Hellwig case FS_IOC_GETFLAGS: 24806cbff00fSChristoph Hellwig return btrfs_ioctl_getflags(file, argp); 24816cbff00fSChristoph Hellwig case FS_IOC_SETFLAGS: 24826cbff00fSChristoph Hellwig return btrfs_ioctl_setflags(file, argp); 24836cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 24846cbff00fSChristoph Hellwig return btrfs_ioctl_getversion(file, argp); 2485f7039b1dSLi Dongyang case FITRIM: 2486f7039b1dSLi Dongyang return btrfs_ioctl_fitrim(file, argp); 2487f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 2488fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 2489fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 2490fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 24913de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 2492fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 249376dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 249476dda93cSYan, Zheng return btrfs_ioctl_snap_destroy(file, argp); 24950caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 24960caa102dSLi Zefan return btrfs_ioctl_subvol_getflags(file, argp); 24970caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 24980caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 24996ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 25006ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 2501f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 25021e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 25031e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 25041e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 2505f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 25064bcabaa3SChristoph Hellwig return btrfs_ioctl_resize(root, argp); 2507f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 25084bcabaa3SChristoph Hellwig return btrfs_ioctl_add_dev(root, argp); 2509f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 25104bcabaa3SChristoph Hellwig return btrfs_ioctl_rm_dev(root, argp); 2511f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 2512f46b5a66SChristoph Hellwig return btrfs_balance(root->fs_info->dev_root); 2513f46b5a66SChristoph Hellwig case BTRFS_IOC_CLONE: 2514c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, arg, 0, 0, 0); 2515c5c9cd4dSSage Weil case BTRFS_IOC_CLONE_RANGE: 25167a865e8aSChristoph Hellwig return btrfs_ioctl_clone_range(file, argp); 2517f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_START: 2518f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_start(file); 2519f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_END: 2520f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_end(file); 2521ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 2522ac8e9819SChris Mason return btrfs_ioctl_tree_search(file, argp); 2523ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 2524ac8e9819SChris Mason return btrfs_ioctl_ino_lookup(file, argp); 25251406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 25261406e432SJosef Bacik return btrfs_ioctl_space_info(root, argp); 2527f46b5a66SChristoph Hellwig case BTRFS_IOC_SYNC: 2528f46b5a66SChristoph Hellwig btrfs_sync_fs(file->f_dentry->d_sb, 1); 2529f46b5a66SChristoph Hellwig return 0; 253046204592SSage Weil case BTRFS_IOC_START_SYNC: 253146204592SSage Weil return btrfs_ioctl_start_sync(file, argp); 253246204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 253346204592SSage Weil return btrfs_ioctl_wait_sync(file, argp); 2534f46b5a66SChristoph Hellwig } 2535f46b5a66SChristoph Hellwig 2536f46b5a66SChristoph Hellwig return -ENOTTY; 2537f46b5a66SChristoph Hellwig } 2538