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> 448ea05e3aSAlexander Block #include <linux/uuid.h> 4555e301fdSFilipe Brandenburger #include <linux/btrfs.h> 46416161dbSMark Fasheh #include <linux/uaccess.h> 47f46b5a66SChristoph Hellwig #include "ctree.h" 48f46b5a66SChristoph Hellwig #include "disk-io.h" 49f46b5a66SChristoph Hellwig #include "transaction.h" 50f46b5a66SChristoph Hellwig #include "btrfs_inode.h" 51f46b5a66SChristoph Hellwig #include "print-tree.h" 52f46b5a66SChristoph Hellwig #include "volumes.h" 53925baeddSChris Mason #include "locking.h" 54581bb050SLi Zefan #include "inode-map.h" 55d7728c96SJan Schmidt #include "backref.h" 56606686eeSJosef Bacik #include "rcu-string.h" 5731db9f7cSAlexander Block #include "send.h" 583f6bcfbdSStefan Behrens #include "dev-replace.h" 5963541927SFilipe David Borba Manana #include "props.h" 603b02a68aSJeff Mahoney #include "sysfs.h" 61f46b5a66SChristoph Hellwig 62416161dbSMark Fasheh static int btrfs_clone(struct inode *src, struct inode *inode, 63416161dbSMark Fasheh u64 off, u64 olen, u64 olen_aligned, u64 destoff); 64416161dbSMark Fasheh 656cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 666cbff00fSChristoph Hellwig static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags) 676cbff00fSChristoph Hellwig { 686cbff00fSChristoph Hellwig if (S_ISDIR(mode)) 696cbff00fSChristoph Hellwig return flags; 706cbff00fSChristoph Hellwig else if (S_ISREG(mode)) 716cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 726cbff00fSChristoph Hellwig else 736cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 746cbff00fSChristoph Hellwig } 75f46b5a66SChristoph Hellwig 766cbff00fSChristoph Hellwig /* 776cbff00fSChristoph Hellwig * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl. 786cbff00fSChristoph Hellwig */ 796cbff00fSChristoph Hellwig static unsigned int btrfs_flags_to_ioctl(unsigned int flags) 806cbff00fSChristoph Hellwig { 816cbff00fSChristoph Hellwig unsigned int iflags = 0; 826cbff00fSChristoph Hellwig 836cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 846cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 856cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 866cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 876cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 886cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 896cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 906cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 916cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 926cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 936cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 946cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 95d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 96d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 97d0092bddSLi Zefan 98d0092bddSLi Zefan if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS)) 99d0092bddSLi Zefan iflags |= FS_COMPR_FL; 100d0092bddSLi Zefan else if (flags & BTRFS_INODE_NOCOMPRESS) 101d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 1026cbff00fSChristoph Hellwig 1036cbff00fSChristoph Hellwig return iflags; 1046cbff00fSChristoph Hellwig } 1056cbff00fSChristoph Hellwig 1066cbff00fSChristoph Hellwig /* 1076cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 1086cbff00fSChristoph Hellwig */ 1096cbff00fSChristoph Hellwig void btrfs_update_iflags(struct inode *inode) 1106cbff00fSChristoph Hellwig { 1116cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1126cbff00fSChristoph Hellwig 1136cbff00fSChristoph Hellwig inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); 1146cbff00fSChristoph Hellwig 1156cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_SYNC) 1166cbff00fSChristoph Hellwig inode->i_flags |= S_SYNC; 1176cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_IMMUTABLE) 1186cbff00fSChristoph Hellwig inode->i_flags |= S_IMMUTABLE; 1196cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_APPEND) 1206cbff00fSChristoph Hellwig inode->i_flags |= S_APPEND; 1216cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_NOATIME) 1226cbff00fSChristoph Hellwig inode->i_flags |= S_NOATIME; 1236cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_DIRSYNC) 1246cbff00fSChristoph Hellwig inode->i_flags |= S_DIRSYNC; 1256cbff00fSChristoph Hellwig } 1266cbff00fSChristoph Hellwig 1276cbff00fSChristoph Hellwig /* 1286cbff00fSChristoph Hellwig * Inherit flags from the parent inode. 1296cbff00fSChristoph Hellwig * 130e27425d6SJosef Bacik * Currently only the compression flags and the cow flags are inherited. 1316cbff00fSChristoph Hellwig */ 1326cbff00fSChristoph Hellwig void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) 1336cbff00fSChristoph Hellwig { 1340b4dcea5SChris Mason unsigned int flags; 1350b4dcea5SChris Mason 1360b4dcea5SChris Mason if (!dir) 1370b4dcea5SChris Mason return; 1380b4dcea5SChris Mason 1390b4dcea5SChris Mason flags = BTRFS_I(dir)->flags; 1406cbff00fSChristoph Hellwig 141e27425d6SJosef Bacik if (flags & BTRFS_INODE_NOCOMPRESS) { 142e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; 143e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; 144e27425d6SJosef Bacik } else if (flags & BTRFS_INODE_COMPRESS) { 145e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; 146e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; 147e27425d6SJosef Bacik } 1486cbff00fSChristoph Hellwig 149213490b3SLiu Bo if (flags & BTRFS_INODE_NODATACOW) { 150e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; 151213490b3SLiu Bo if (S_ISREG(inode->i_mode)) 152213490b3SLiu Bo BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; 153213490b3SLiu Bo } 154e27425d6SJosef Bacik 1556cbff00fSChristoph Hellwig btrfs_update_iflags(inode); 1566cbff00fSChristoph Hellwig } 1576cbff00fSChristoph Hellwig 1586cbff00fSChristoph Hellwig static int btrfs_ioctl_getflags(struct file *file, void __user *arg) 1596cbff00fSChristoph Hellwig { 160496ad9aaSAl Viro struct btrfs_inode *ip = BTRFS_I(file_inode(file)); 1616cbff00fSChristoph Hellwig unsigned int flags = btrfs_flags_to_ioctl(ip->flags); 1626cbff00fSChristoph Hellwig 1636cbff00fSChristoph Hellwig if (copy_to_user(arg, &flags, sizeof(flags))) 1646cbff00fSChristoph Hellwig return -EFAULT; 1656cbff00fSChristoph Hellwig return 0; 1666cbff00fSChristoph Hellwig } 1676cbff00fSChristoph Hellwig 16875e7cb7fSLiu Bo static int check_flags(unsigned int flags) 16975e7cb7fSLiu Bo { 17075e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 17175e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 17275e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 173e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 174e1e8fb6aSLi Zefan FS_NOCOW_FL)) 17575e7cb7fSLiu Bo return -EOPNOTSUPP; 17675e7cb7fSLiu Bo 17775e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 17875e7cb7fSLiu Bo return -EINVAL; 17975e7cb7fSLiu Bo 18075e7cb7fSLiu Bo return 0; 18175e7cb7fSLiu Bo } 18275e7cb7fSLiu Bo 1836cbff00fSChristoph Hellwig static int btrfs_ioctl_setflags(struct file *file, void __user *arg) 1846cbff00fSChristoph Hellwig { 185496ad9aaSAl Viro struct inode *inode = file_inode(file); 1866cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1876cbff00fSChristoph Hellwig struct btrfs_root *root = ip->root; 1886cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 1896cbff00fSChristoph Hellwig unsigned int flags, oldflags; 1906cbff00fSChristoph Hellwig int ret; 191f062abf0SLi Zefan u64 ip_oldflags; 192f062abf0SLi Zefan unsigned int i_oldflags; 1937e97b8daSDavid Sterba umode_t mode; 1946cbff00fSChristoph Hellwig 195bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 196bd60ea0fSDavid Sterba return -EPERM; 197bd60ea0fSDavid Sterba 198b83cc969SLi Zefan if (btrfs_root_readonly(root)) 199b83cc969SLi Zefan return -EROFS; 200b83cc969SLi Zefan 2016cbff00fSChristoph Hellwig if (copy_from_user(&flags, arg, sizeof(flags))) 2026cbff00fSChristoph Hellwig return -EFAULT; 2036cbff00fSChristoph Hellwig 20475e7cb7fSLiu Bo ret = check_flags(flags); 20575e7cb7fSLiu Bo if (ret) 20675e7cb7fSLiu Bo return ret; 2076cbff00fSChristoph Hellwig 208e7848683SJan Kara ret = mnt_want_write_file(file); 209e7848683SJan Kara if (ret) 210e7848683SJan Kara return ret; 211e7848683SJan Kara 2126cbff00fSChristoph Hellwig mutex_lock(&inode->i_mutex); 2136cbff00fSChristoph Hellwig 214f062abf0SLi Zefan ip_oldflags = ip->flags; 215f062abf0SLi Zefan i_oldflags = inode->i_flags; 2167e97b8daSDavid Sterba mode = inode->i_mode; 217f062abf0SLi Zefan 2186cbff00fSChristoph Hellwig flags = btrfs_mask_flags(inode->i_mode, flags); 2196cbff00fSChristoph Hellwig oldflags = btrfs_flags_to_ioctl(ip->flags); 2206cbff00fSChristoph Hellwig if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 2216cbff00fSChristoph Hellwig if (!capable(CAP_LINUX_IMMUTABLE)) { 2226cbff00fSChristoph Hellwig ret = -EPERM; 2236cbff00fSChristoph Hellwig goto out_unlock; 2246cbff00fSChristoph Hellwig } 2256cbff00fSChristoph Hellwig } 2266cbff00fSChristoph Hellwig 2276cbff00fSChristoph Hellwig if (flags & FS_SYNC_FL) 2286cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_SYNC; 2296cbff00fSChristoph Hellwig else 2306cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_SYNC; 2316cbff00fSChristoph Hellwig if (flags & FS_IMMUTABLE_FL) 2326cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_IMMUTABLE; 2336cbff00fSChristoph Hellwig else 2346cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_IMMUTABLE; 2356cbff00fSChristoph Hellwig if (flags & FS_APPEND_FL) 2366cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_APPEND; 2376cbff00fSChristoph Hellwig else 2386cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_APPEND; 2396cbff00fSChristoph Hellwig if (flags & FS_NODUMP_FL) 2406cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NODUMP; 2416cbff00fSChristoph Hellwig else 2426cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NODUMP; 2436cbff00fSChristoph Hellwig if (flags & FS_NOATIME_FL) 2446cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NOATIME; 2456cbff00fSChristoph Hellwig else 2466cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NOATIME; 2476cbff00fSChristoph Hellwig if (flags & FS_DIRSYNC_FL) 2486cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_DIRSYNC; 2496cbff00fSChristoph Hellwig else 2506cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_DIRSYNC; 2517e97b8daSDavid Sterba if (flags & FS_NOCOW_FL) { 2527e97b8daSDavid Sterba if (S_ISREG(mode)) { 2537e97b8daSDavid Sterba /* 2547e97b8daSDavid Sterba * It's safe to turn csums off here, no extents exist. 2557e97b8daSDavid Sterba * Otherwise we want the flag to reflect the real COW 2567e97b8daSDavid Sterba * status of the file and will not set it. 2577e97b8daSDavid Sterba */ 2587e97b8daSDavid Sterba if (inode->i_size == 0) 2597e97b8daSDavid Sterba ip->flags |= BTRFS_INODE_NODATACOW 2607e97b8daSDavid Sterba | BTRFS_INODE_NODATASUM; 2617e97b8daSDavid Sterba } else { 262e1e8fb6aSLi Zefan ip->flags |= BTRFS_INODE_NODATACOW; 2637e97b8daSDavid Sterba } 2647e97b8daSDavid Sterba } else { 2657e97b8daSDavid Sterba /* 2667e97b8daSDavid Sterba * Revert back under same assuptions as above 2677e97b8daSDavid Sterba */ 2687e97b8daSDavid Sterba if (S_ISREG(mode)) { 2697e97b8daSDavid Sterba if (inode->i_size == 0) 2707e97b8daSDavid Sterba ip->flags &= ~(BTRFS_INODE_NODATACOW 2717e97b8daSDavid Sterba | BTRFS_INODE_NODATASUM); 2727e97b8daSDavid Sterba } else { 273e1e8fb6aSLi Zefan ip->flags &= ~BTRFS_INODE_NODATACOW; 2747e97b8daSDavid Sterba } 2757e97b8daSDavid Sterba } 2766cbff00fSChristoph Hellwig 27775e7cb7fSLiu Bo /* 27875e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 27975e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 28075e7cb7fSLiu Bo * things smaller. 28175e7cb7fSLiu Bo */ 28275e7cb7fSLiu Bo if (flags & FS_NOCOMP_FL) { 28375e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_COMPRESS; 28475e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_NOCOMPRESS; 28563541927SFilipe David Borba Manana 28663541927SFilipe David Borba Manana ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0); 28763541927SFilipe David Borba Manana if (ret && ret != -ENODATA) 28863541927SFilipe David Borba Manana goto out_drop; 28975e7cb7fSLiu Bo } else if (flags & FS_COMPR_FL) { 29063541927SFilipe David Borba Manana const char *comp; 29163541927SFilipe David Borba Manana 29275e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_COMPRESS; 29375e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_NOCOMPRESS; 29463541927SFilipe David Borba Manana 29563541927SFilipe David Borba Manana if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 29663541927SFilipe David Borba Manana comp = "lzo"; 29763541927SFilipe David Borba Manana else 29863541927SFilipe David Borba Manana comp = "zlib"; 29963541927SFilipe David Borba Manana ret = btrfs_set_prop(inode, "btrfs.compression", 30063541927SFilipe David Borba Manana comp, strlen(comp), 0); 30163541927SFilipe David Borba Manana if (ret) 30263541927SFilipe David Borba Manana goto out_drop; 30363541927SFilipe David Borba Manana 304ebcb904dSLi Zefan } else { 305ebcb904dSLi Zefan ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 30675e7cb7fSLiu Bo } 3076cbff00fSChristoph Hellwig 3084da6f1a3SLi Zefan trans = btrfs_start_transaction(root, 1); 309f062abf0SLi Zefan if (IS_ERR(trans)) { 310f062abf0SLi Zefan ret = PTR_ERR(trans); 311f062abf0SLi Zefan goto out_drop; 312f062abf0SLi Zefan } 3136cbff00fSChristoph Hellwig 314306424ccSLi Zefan btrfs_update_iflags(inode); 3150c4d2d95SJosef Bacik inode_inc_iversion(inode); 316306424ccSLi Zefan inode->i_ctime = CURRENT_TIME; 3176cbff00fSChristoph Hellwig ret = btrfs_update_inode(trans, root, inode); 3186cbff00fSChristoph Hellwig 3196cbff00fSChristoph Hellwig btrfs_end_transaction(trans, root); 320f062abf0SLi Zefan out_drop: 321f062abf0SLi Zefan if (ret) { 322f062abf0SLi Zefan ip->flags = ip_oldflags; 323f062abf0SLi Zefan inode->i_flags = i_oldflags; 324f062abf0SLi Zefan } 3256cbff00fSChristoph Hellwig 3266cbff00fSChristoph Hellwig out_unlock: 3276cbff00fSChristoph Hellwig mutex_unlock(&inode->i_mutex); 328e7848683SJan Kara mnt_drop_write_file(file); 3292d4e6f6aSliubo return ret; 3306cbff00fSChristoph Hellwig } 3316cbff00fSChristoph Hellwig 3326cbff00fSChristoph Hellwig static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 3336cbff00fSChristoph Hellwig { 334496ad9aaSAl Viro struct inode *inode = file_inode(file); 3356cbff00fSChristoph Hellwig 3366cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 3376cbff00fSChristoph Hellwig } 338f46b5a66SChristoph Hellwig 339f7039b1dSLi Dongyang static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg) 340f7039b1dSLi Dongyang { 34154563d41SAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 342f7039b1dSLi Dongyang struct btrfs_device *device; 343f7039b1dSLi Dongyang struct request_queue *q; 344f7039b1dSLi Dongyang struct fstrim_range range; 345f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 346f7039b1dSLi Dongyang u64 num_devices = 0; 347815745cfSAl Viro u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 348f7039b1dSLi Dongyang int ret; 349f7039b1dSLi Dongyang 350f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 351f7039b1dSLi Dongyang return -EPERM; 352f7039b1dSLi Dongyang 3531f78160cSXiao Guangrong rcu_read_lock(); 3541f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 3551f78160cSXiao Guangrong dev_list) { 356f7039b1dSLi Dongyang if (!device->bdev) 357f7039b1dSLi Dongyang continue; 358f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 359f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 360f7039b1dSLi Dongyang num_devices++; 361f7039b1dSLi Dongyang minlen = min((u64)q->limits.discard_granularity, 362f7039b1dSLi Dongyang minlen); 363f7039b1dSLi Dongyang } 364f7039b1dSLi Dongyang } 3651f78160cSXiao Guangrong rcu_read_unlock(); 366f4c697e6SLukas Czerner 367f7039b1dSLi Dongyang if (!num_devices) 368f7039b1dSLi Dongyang return -EOPNOTSUPP; 369f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 370f7039b1dSLi Dongyang return -EFAULT; 371e515c18bSLukas Czerner if (range.start > total_bytes || 372e515c18bSLukas Czerner range.len < fs_info->sb->s_blocksize) 373f4c697e6SLukas Czerner return -EINVAL; 374f7039b1dSLi Dongyang 375f4c697e6SLukas Czerner range.len = min(range.len, total_bytes - range.start); 376f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 377815745cfSAl Viro ret = btrfs_trim_fs(fs_info->tree_root, &range); 378f7039b1dSLi Dongyang if (ret < 0) 379f7039b1dSLi Dongyang return ret; 380f7039b1dSLi Dongyang 381f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 382f7039b1dSLi Dongyang return -EFAULT; 383f7039b1dSLi Dongyang 384f7039b1dSLi Dongyang return 0; 385f7039b1dSLi Dongyang } 386f7039b1dSLi Dongyang 387dd5f9615SStefan Behrens int btrfs_is_empty_uuid(u8 *uuid) 388dd5f9615SStefan Behrens { 38946e0f66aSChris Mason int i; 39046e0f66aSChris Mason 39146e0f66aSChris Mason for (i = 0; i < BTRFS_UUID_SIZE; i++) { 39246e0f66aSChris Mason if (uuid[i]) 39346e0f66aSChris Mason return 0; 39446e0f66aSChris Mason } 39546e0f66aSChris Mason return 1; 396dd5f9615SStefan Behrens } 397dd5f9615SStefan Behrens 398d5c12070SMiao Xie static noinline int create_subvol(struct inode *dir, 399cb8e7090SChristoph Hellwig struct dentry *dentry, 40072fd032eSSage Weil char *name, int namelen, 4016f72c7e2SArne Jansen u64 *async_transid, 4028696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 403f46b5a66SChristoph Hellwig { 404f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 405f46b5a66SChristoph Hellwig struct btrfs_key key; 406f46b5a66SChristoph Hellwig struct btrfs_root_item root_item; 407f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 408f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 409d5c12070SMiao Xie struct btrfs_root *root = BTRFS_I(dir)->root; 41076dda93cSYan, Zheng struct btrfs_root *new_root; 411d5c12070SMiao Xie struct btrfs_block_rsv block_rsv; 4128ea05e3aSAlexander Block struct timespec cur_time = CURRENT_TIME; 4135662344bSTsutomu Itoh struct inode *inode; 414f46b5a66SChristoph Hellwig int ret; 415f46b5a66SChristoph Hellwig int err; 416f46b5a66SChristoph Hellwig u64 objectid; 417f46b5a66SChristoph Hellwig u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 4183de4586cSChris Mason u64 index = 0; 419d5c12070SMiao Xie u64 qgroup_reserved; 4208ea05e3aSAlexander Block uuid_le new_uuid; 421f46b5a66SChristoph Hellwig 422581bb050SLi Zefan ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid); 4232fbe8c8aSAl Viro if (ret) 424a22285a6SYan, Zheng return ret; 4256a912213SJosef Bacik 426d5c12070SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 4279ed74f2dSJosef Bacik /* 428d5c12070SMiao Xie * The same as the snapshot creation, please see the comment 429d5c12070SMiao Xie * of create_snapshot(). 4309ed74f2dSJosef Bacik */ 431d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 432dd5f9615SStefan Behrens 8, &qgroup_reserved, false); 433d5c12070SMiao Xie if (ret) 434d5c12070SMiao Xie return ret; 435f46b5a66SChristoph Hellwig 436d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 437d5c12070SMiao Xie if (IS_ERR(trans)) { 438d5c12070SMiao Xie ret = PTR_ERR(trans); 439de6e8200SLiu Bo btrfs_subvolume_release_metadata(root, &block_rsv, 440de6e8200SLiu Bo qgroup_reserved); 441de6e8200SLiu Bo return ret; 442d5c12070SMiao Xie } 443d5c12070SMiao Xie trans->block_rsv = &block_rsv; 444d5c12070SMiao Xie trans->bytes_reserved = block_rsv.size; 445f46b5a66SChristoph Hellwig 4468696c533SMiao Xie ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit); 4476f72c7e2SArne Jansen if (ret) 4486f72c7e2SArne Jansen goto fail; 4496f72c7e2SArne Jansen 4505d4f98a2SYan Zheng leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 4515581a51aSJan Schmidt 0, objectid, NULL, 0, 0, 0); 4528e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 4538e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 4548e8a1e31SJosef Bacik goto fail; 4558e8a1e31SJosef Bacik } 456f46b5a66SChristoph Hellwig 4575d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 458f46b5a66SChristoph Hellwig btrfs_set_header_bytenr(leaf, leaf->start); 459f46b5a66SChristoph Hellwig btrfs_set_header_generation(leaf, trans->transid); 4605d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 461f46b5a66SChristoph Hellwig btrfs_set_header_owner(leaf, objectid); 462f46b5a66SChristoph Hellwig 4630a4e5586SRoss Kirk write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(), 464f46b5a66SChristoph Hellwig BTRFS_FSID_SIZE); 4655d4f98a2SYan Zheng write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, 466b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 4675d4f98a2SYan Zheng BTRFS_UUID_SIZE); 468f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 469f46b5a66SChristoph Hellwig 4708ea05e3aSAlexander Block memset(&root_item, 0, sizeof(root_item)); 4718ea05e3aSAlexander Block 472f46b5a66SChristoph Hellwig inode_item = &root_item.inode; 4733cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 4743cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 4753cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 4763cae210fSQu Wenruo btrfs_set_stack_inode_nbytes(inode_item, root->leafsize); 4773cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 478f46b5a66SChristoph Hellwig 4793cae210fSQu Wenruo btrfs_set_root_flags(&root_item, 0); 4803cae210fSQu Wenruo btrfs_set_root_limit(&root_item, 0); 4813cae210fSQu Wenruo btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 48208fe4db1SLi Zefan 483f46b5a66SChristoph Hellwig btrfs_set_root_bytenr(&root_item, leaf->start); 48484234f3aSYan Zheng btrfs_set_root_generation(&root_item, trans->transid); 485f46b5a66SChristoph Hellwig btrfs_set_root_level(&root_item, 0); 486f46b5a66SChristoph Hellwig btrfs_set_root_refs(&root_item, 1); 48786b9f2ecSYan, Zheng btrfs_set_root_used(&root_item, leaf->len); 48880ff3856SYan Zheng btrfs_set_root_last_snapshot(&root_item, 0); 489f46b5a66SChristoph Hellwig 4908ea05e3aSAlexander Block btrfs_set_root_generation_v2(&root_item, 4918ea05e3aSAlexander Block btrfs_root_generation(&root_item)); 4928ea05e3aSAlexander Block uuid_le_gen(&new_uuid); 4938ea05e3aSAlexander Block memcpy(root_item.uuid, new_uuid.b, BTRFS_UUID_SIZE); 4943cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item.otime, cur_time.tv_sec); 4953cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item.otime, cur_time.tv_nsec); 4968ea05e3aSAlexander Block root_item.ctime = root_item.otime; 4978ea05e3aSAlexander Block btrfs_set_root_ctransid(&root_item, trans->transid); 4988ea05e3aSAlexander Block btrfs_set_root_otransid(&root_item, trans->transid); 499f46b5a66SChristoph Hellwig 500925baeddSChris Mason btrfs_tree_unlock(leaf); 501f46b5a66SChristoph Hellwig free_extent_buffer(leaf); 502f46b5a66SChristoph Hellwig leaf = NULL; 503f46b5a66SChristoph Hellwig 504f46b5a66SChristoph Hellwig btrfs_set_root_dirid(&root_item, new_dirid); 505f46b5a66SChristoph Hellwig 506f46b5a66SChristoph Hellwig key.objectid = objectid; 5075d4f98a2SYan Zheng key.offset = 0; 508f46b5a66SChristoph Hellwig btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY); 509f46b5a66SChristoph Hellwig ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, 510f46b5a66SChristoph Hellwig &root_item); 511f46b5a66SChristoph Hellwig if (ret) 512f46b5a66SChristoph Hellwig goto fail; 513f46b5a66SChristoph Hellwig 51476dda93cSYan, Zheng key.offset = (u64)-1; 51576dda93cSYan, Zheng new_root = btrfs_read_fs_root_no_name(root->fs_info, &key); 51679787eaaSJeff Mahoney if (IS_ERR(new_root)) { 51779787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, PTR_ERR(new_root)); 51879787eaaSJeff Mahoney ret = PTR_ERR(new_root); 51979787eaaSJeff Mahoney goto fail; 52079787eaaSJeff Mahoney } 52176dda93cSYan, Zheng 52276dda93cSYan, Zheng btrfs_record_root_in_trans(trans, new_root); 52376dda93cSYan, Zheng 52463541927SFilipe David Borba Manana ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid); 525ce598979SMark Fasheh if (ret) { 526ce598979SMark Fasheh /* We potentially lose an unused inode item here */ 52779787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 528ce598979SMark Fasheh goto fail; 529ce598979SMark Fasheh } 530ce598979SMark Fasheh 531f46b5a66SChristoph Hellwig /* 532f46b5a66SChristoph Hellwig * insert the directory item 533f46b5a66SChristoph Hellwig */ 5343de4586cSChris Mason ret = btrfs_set_inode_index(dir, &index); 53579787eaaSJeff Mahoney if (ret) { 53679787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 53779787eaaSJeff Mahoney goto fail; 53879787eaaSJeff Mahoney } 5393de4586cSChris Mason 5403de4586cSChris Mason ret = btrfs_insert_dir_item(trans, root, 54116cdcec7SMiao Xie name, namelen, dir, &key, 5423de4586cSChris Mason BTRFS_FT_DIR, index); 54379787eaaSJeff Mahoney if (ret) { 54479787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 545f46b5a66SChristoph Hellwig goto fail; 54679787eaaSJeff Mahoney } 5470660b5afSChris Mason 54852c26179SYan Zheng btrfs_i_size_write(dir, dir->i_size + namelen * 2); 54952c26179SYan Zheng ret = btrfs_update_inode(trans, root, dir); 55052c26179SYan Zheng BUG_ON(ret); 55152c26179SYan Zheng 5520660b5afSChris Mason ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 5534df27c4dSYan, Zheng objectid, root->root_key.objectid, 55433345d01SLi Zefan btrfs_ino(dir), index, name, namelen); 5550660b5afSChris Mason BUG_ON(ret); 5560660b5afSChris Mason 557dd5f9615SStefan Behrens ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root, 558dd5f9615SStefan Behrens root_item.uuid, BTRFS_UUID_KEY_SUBVOL, 559dd5f9615SStefan Behrens objectid); 560dd5f9615SStefan Behrens if (ret) 561dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 562dd5f9615SStefan Behrens 563f46b5a66SChristoph Hellwig fail: 564d5c12070SMiao Xie trans->block_rsv = NULL; 565d5c12070SMiao Xie trans->bytes_reserved = 0; 566de6e8200SLiu Bo btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved); 567de6e8200SLiu Bo 56872fd032eSSage Weil if (async_transid) { 56972fd032eSSage Weil *async_transid = trans->transid; 57072fd032eSSage Weil err = btrfs_commit_transaction_async(trans, root, 1); 57100d71c9cSMiao Xie if (err) 57200d71c9cSMiao Xie err = btrfs_commit_transaction(trans, root); 57372fd032eSSage Weil } else { 57476dda93cSYan, Zheng err = btrfs_commit_transaction(trans, root); 57572fd032eSSage Weil } 576f46b5a66SChristoph Hellwig if (err && !ret) 577f46b5a66SChristoph Hellwig ret = err; 5781a65e24bSChris Mason 5795662344bSTsutomu Itoh if (!ret) { 5805662344bSTsutomu Itoh inode = btrfs_lookup_dentry(dir, dentry); 581de6e8200SLiu Bo if (IS_ERR(inode)) 582de6e8200SLiu Bo return PTR_ERR(inode); 5835662344bSTsutomu Itoh d_instantiate(dentry, inode); 5845662344bSTsutomu Itoh } 585f46b5a66SChristoph Hellwig return ret; 586f46b5a66SChristoph Hellwig } 587f46b5a66SChristoph Hellwig 588e9662f70SMiao Xie static int create_snapshot(struct btrfs_root *root, struct inode *dir, 589e9662f70SMiao Xie struct dentry *dentry, char *name, int namelen, 590e9662f70SMiao Xie u64 *async_transid, bool readonly, 591e9662f70SMiao Xie struct btrfs_qgroup_inherit *inherit) 592f46b5a66SChristoph Hellwig { 5932e4bfab9SYan, Zheng struct inode *inode; 594f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 595f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 5962e4bfab9SYan, Zheng int ret; 597f46b5a66SChristoph Hellwig 598f46b5a66SChristoph Hellwig if (!root->ref_cows) 599f46b5a66SChristoph Hellwig return -EINVAL; 600f46b5a66SChristoph Hellwig 6016a03843dSMiao Xie ret = btrfs_start_delalloc_inodes(root, 0); 6026a03843dSMiao Xie if (ret) 6036a03843dSMiao Xie return ret; 6046a03843dSMiao Xie 605b0244199SMiao Xie btrfs_wait_ordered_extents(root, -1); 6066a03843dSMiao Xie 607a22285a6SYan, Zheng pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS); 608a22285a6SYan, Zheng if (!pending_snapshot) 609a22285a6SYan, Zheng return -ENOMEM; 610a22285a6SYan, Zheng 61166d8f3ddSMiao Xie btrfs_init_block_rsv(&pending_snapshot->block_rsv, 61266d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 613d5c12070SMiao Xie /* 614d5c12070SMiao Xie * 1 - parent dir inode 615d5c12070SMiao Xie * 2 - dir entries 616d5c12070SMiao Xie * 1 - root item 617d5c12070SMiao Xie * 2 - root ref/backref 618d5c12070SMiao Xie * 1 - root of snapshot 619dd5f9615SStefan Behrens * 1 - UUID item 620d5c12070SMiao Xie */ 621d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 622dd5f9615SStefan Behrens &pending_snapshot->block_rsv, 8, 623ee3441b4SJeff Mahoney &pending_snapshot->qgroup_reserved, 624ee3441b4SJeff Mahoney false); 625d5c12070SMiao Xie if (ret) 626d5c12070SMiao Xie goto out; 627d5c12070SMiao Xie 628a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 629a22285a6SYan, Zheng pending_snapshot->root = root; 630b83cc969SLi Zefan pending_snapshot->readonly = readonly; 631e9662f70SMiao Xie pending_snapshot->dir = dir; 6328696c533SMiao Xie pending_snapshot->inherit = inherit; 633a22285a6SYan, Zheng 634d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 635a22285a6SYan, Zheng if (IS_ERR(trans)) { 636a22285a6SYan, Zheng ret = PTR_ERR(trans); 637a22285a6SYan, Zheng goto fail; 638a22285a6SYan, Zheng } 639a22285a6SYan, Zheng 6408351583eSJosef Bacik spin_lock(&root->fs_info->trans_lock); 641a22285a6SYan, Zheng list_add(&pending_snapshot->list, 642a22285a6SYan, Zheng &trans->transaction->pending_snapshots); 6438351583eSJosef Bacik spin_unlock(&root->fs_info->trans_lock); 64472fd032eSSage Weil if (async_transid) { 64572fd032eSSage Weil *async_transid = trans->transid; 64672fd032eSSage Weil ret = btrfs_commit_transaction_async(trans, 64772fd032eSSage Weil root->fs_info->extent_root, 1); 64800d71c9cSMiao Xie if (ret) 64900d71c9cSMiao Xie ret = btrfs_commit_transaction(trans, root); 65072fd032eSSage Weil } else { 65172fd032eSSage Weil ret = btrfs_commit_transaction(trans, 65272fd032eSSage Weil root->fs_info->extent_root); 65372fd032eSSage Weil } 654aec8030aSMiao Xie if (ret) 655c37b2b62SJosef Bacik goto fail; 656a22285a6SYan, Zheng 657a22285a6SYan, Zheng ret = pending_snapshot->error; 658f46b5a66SChristoph Hellwig if (ret) 6592e4bfab9SYan, Zheng goto fail; 660f46b5a66SChristoph Hellwig 66166b4ffd1SJosef Bacik ret = btrfs_orphan_cleanup(pending_snapshot->snap); 66266b4ffd1SJosef Bacik if (ret) 66366b4ffd1SJosef Bacik goto fail; 664f46b5a66SChristoph Hellwig 6652fbe8c8aSAl Viro inode = btrfs_lookup_dentry(dentry->d_parent->d_inode, dentry); 6662e4bfab9SYan, Zheng if (IS_ERR(inode)) { 6672e4bfab9SYan, Zheng ret = PTR_ERR(inode); 6682e4bfab9SYan, Zheng goto fail; 6692e4bfab9SYan, Zheng } 6705662344bSTsutomu Itoh 6712e4bfab9SYan, Zheng d_instantiate(dentry, inode); 6722e4bfab9SYan, Zheng ret = 0; 6732e4bfab9SYan, Zheng fail: 674d5c12070SMiao Xie btrfs_subvolume_release_metadata(BTRFS_I(dir)->root, 675d5c12070SMiao Xie &pending_snapshot->block_rsv, 676d5c12070SMiao Xie pending_snapshot->qgroup_reserved); 677d5c12070SMiao Xie out: 678a22285a6SYan, Zheng kfree(pending_snapshot); 679f46b5a66SChristoph Hellwig return ret; 680f46b5a66SChristoph Hellwig } 681f46b5a66SChristoph Hellwig 6824260f7c7SSage Weil /* copy of check_sticky in fs/namei.c() 6834260f7c7SSage Weil * It's inline, so penalty for filesystems that don't use sticky bit is 6844260f7c7SSage Weil * minimal. 6854260f7c7SSage Weil */ 6864260f7c7SSage Weil static inline int btrfs_check_sticky(struct inode *dir, struct inode *inode) 6874260f7c7SSage Weil { 6882f2f43d3SEric W. Biederman kuid_t fsuid = current_fsuid(); 6894260f7c7SSage Weil 6904260f7c7SSage Weil if (!(dir->i_mode & S_ISVTX)) 6914260f7c7SSage Weil return 0; 6922f2f43d3SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 6934260f7c7SSage Weil return 0; 6942f2f43d3SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 6954260f7c7SSage Weil return 0; 6964260f7c7SSage Weil return !capable(CAP_FOWNER); 6974260f7c7SSage Weil } 6984260f7c7SSage Weil 6994260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 7004260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 7014260f7c7SSage Weil * whether the type of victim is right. 7024260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 7034260f7c7SSage Weil * 2. We should have write and exec permissions on dir 7044260f7c7SSage Weil * 3. We can't remove anything from append-only dir 7054260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 7064260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 7074260f7c7SSage Weil * a. be owner of dir, or 7084260f7c7SSage Weil * b. be owner of victim, or 7094260f7c7SSage Weil * c. have CAP_FOWNER capability 7104260f7c7SSage Weil * 6. If the victim is append-only or immutable we can't do antyhing with 7114260f7c7SSage Weil * links pointing to it. 7124260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 7134260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 7144260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 7154260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 7164260f7c7SSage Weil * nfs_async_unlink(). 7174260f7c7SSage Weil */ 7184260f7c7SSage Weil 7194260f7c7SSage Weil static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir) 7204260f7c7SSage Weil { 7214260f7c7SSage Weil int error; 7224260f7c7SSage Weil 7234260f7c7SSage Weil if (!victim->d_inode) 7244260f7c7SSage Weil return -ENOENT; 7254260f7c7SSage Weil 7264260f7c7SSage Weil BUG_ON(victim->d_parent->d_inode != dir); 7274fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 7284260f7c7SSage Weil 7294260f7c7SSage Weil error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 7304260f7c7SSage Weil if (error) 7314260f7c7SSage Weil return error; 7324260f7c7SSage Weil if (IS_APPEND(dir)) 7334260f7c7SSage Weil return -EPERM; 7344260f7c7SSage Weil if (btrfs_check_sticky(dir, victim->d_inode)|| 7354260f7c7SSage Weil IS_APPEND(victim->d_inode)|| 7364260f7c7SSage Weil IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 7374260f7c7SSage Weil return -EPERM; 7384260f7c7SSage Weil if (isdir) { 7394260f7c7SSage Weil if (!S_ISDIR(victim->d_inode->i_mode)) 7404260f7c7SSage Weil return -ENOTDIR; 7414260f7c7SSage Weil if (IS_ROOT(victim)) 7424260f7c7SSage Weil return -EBUSY; 7434260f7c7SSage Weil } else if (S_ISDIR(victim->d_inode->i_mode)) 7444260f7c7SSage Weil return -EISDIR; 7454260f7c7SSage Weil if (IS_DEADDIR(dir)) 7464260f7c7SSage Weil return -ENOENT; 7474260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 7484260f7c7SSage Weil return -EBUSY; 7494260f7c7SSage Weil return 0; 7504260f7c7SSage Weil } 7514260f7c7SSage Weil 752cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 753cb8e7090SChristoph Hellwig static inline int btrfs_may_create(struct inode *dir, struct dentry *child) 754cb8e7090SChristoph Hellwig { 755cb8e7090SChristoph Hellwig if (child->d_inode) 756cb8e7090SChristoph Hellwig return -EEXIST; 757cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 758cb8e7090SChristoph Hellwig return -ENOENT; 759cb8e7090SChristoph Hellwig return inode_permission(dir, MAY_WRITE | MAY_EXEC); 760cb8e7090SChristoph Hellwig } 761cb8e7090SChristoph Hellwig 762cb8e7090SChristoph Hellwig /* 763cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 764cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 765cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 766cb8e7090SChristoph Hellwig */ 76776dda93cSYan, Zheng static noinline int btrfs_mksubvol(struct path *parent, 76876dda93cSYan, Zheng char *name, int namelen, 76972fd032eSSage Weil struct btrfs_root *snap_src, 7706f72c7e2SArne Jansen u64 *async_transid, bool readonly, 7718696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 772cb8e7090SChristoph Hellwig { 77376dda93cSYan, Zheng struct inode *dir = parent->dentry->d_inode; 774cb8e7090SChristoph Hellwig struct dentry *dentry; 775cb8e7090SChristoph Hellwig int error; 776cb8e7090SChristoph Hellwig 7775c50c9b8SDavid Sterba error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT); 7785c50c9b8SDavid Sterba if (error == -EINTR) 7795c50c9b8SDavid Sterba return error; 780cb8e7090SChristoph Hellwig 781cb8e7090SChristoph Hellwig dentry = lookup_one_len(name, parent->dentry, namelen); 782cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 783cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 784cb8e7090SChristoph Hellwig goto out_unlock; 785cb8e7090SChristoph Hellwig 786cb8e7090SChristoph Hellwig error = -EEXIST; 787cb8e7090SChristoph Hellwig if (dentry->d_inode) 788cb8e7090SChristoph Hellwig goto out_dput; 789cb8e7090SChristoph Hellwig 79076dda93cSYan, Zheng error = btrfs_may_create(dir, dentry); 791cb8e7090SChristoph Hellwig if (error) 792a874a63eSLiu Bo goto out_dput; 793cb8e7090SChristoph Hellwig 7949c52057cSChris Mason /* 7959c52057cSChris Mason * even if this name doesn't exist, we may get hash collisions. 7969c52057cSChris Mason * check for them now when we can safely fail 7979c52057cSChris Mason */ 7989c52057cSChris Mason error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 7999c52057cSChris Mason dir->i_ino, name, 8009c52057cSChris Mason namelen); 8019c52057cSChris Mason if (error) 8029c52057cSChris Mason goto out_dput; 8039c52057cSChris Mason 80476dda93cSYan, Zheng down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 80576dda93cSYan, Zheng 80676dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 80776dda93cSYan, Zheng goto out_up_read; 80876dda93cSYan, Zheng 8093de4586cSChris Mason if (snap_src) { 810e9662f70SMiao Xie error = create_snapshot(snap_src, dir, dentry, name, namelen, 8116f72c7e2SArne Jansen async_transid, readonly, inherit); 8123de4586cSChris Mason } else { 813d5c12070SMiao Xie error = create_subvol(dir, dentry, name, namelen, 814d5c12070SMiao Xie async_transid, inherit); 8153de4586cSChris Mason } 81676dda93cSYan, Zheng if (!error) 81776dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 81876dda93cSYan, Zheng out_up_read: 81976dda93cSYan, Zheng up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 820cb8e7090SChristoph Hellwig out_dput: 821cb8e7090SChristoph Hellwig dput(dentry); 822cb8e7090SChristoph Hellwig out_unlock: 82376dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 824cb8e7090SChristoph Hellwig return error; 825cb8e7090SChristoph Hellwig } 826cb8e7090SChristoph Hellwig 8274cb5300bSChris Mason /* 8284cb5300bSChris Mason * When we're defragging a range, we don't want to kick it off again 8294cb5300bSChris Mason * if it is really just waiting for delalloc to send it down. 8304cb5300bSChris Mason * If we find a nice big extent or delalloc range for the bytes in the 8314cb5300bSChris Mason * file you want to defrag, we return 0 to let you know to skip this 8324cb5300bSChris Mason * part of the file 8334cb5300bSChris Mason */ 8344cb5300bSChris Mason static int check_defrag_in_cache(struct inode *inode, u64 offset, int thresh) 8354cb5300bSChris Mason { 8364cb5300bSChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 8374cb5300bSChris Mason struct extent_map *em = NULL; 8384cb5300bSChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 8394cb5300bSChris Mason u64 end; 8404cb5300bSChris Mason 8414cb5300bSChris Mason read_lock(&em_tree->lock); 8424cb5300bSChris Mason em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE); 8434cb5300bSChris Mason read_unlock(&em_tree->lock); 8444cb5300bSChris Mason 8454cb5300bSChris Mason if (em) { 8464cb5300bSChris Mason end = extent_map_end(em); 8474cb5300bSChris Mason free_extent_map(em); 8484cb5300bSChris Mason if (end - offset > thresh) 8494cb5300bSChris Mason return 0; 8504cb5300bSChris Mason } 8514cb5300bSChris Mason /* if we already have a nice delalloc here, just stop */ 8524cb5300bSChris Mason thresh /= 2; 8534cb5300bSChris Mason end = count_range_bits(io_tree, &offset, offset + thresh, 8544cb5300bSChris Mason thresh, EXTENT_DELALLOC, 1); 8554cb5300bSChris Mason if (end >= thresh) 8564cb5300bSChris Mason return 0; 8574cb5300bSChris Mason return 1; 8584cb5300bSChris Mason } 8594cb5300bSChris Mason 8604cb5300bSChris Mason /* 8614cb5300bSChris Mason * helper function to walk through a file and find extents 8624cb5300bSChris Mason * newer than a specific transid, and smaller than thresh. 8634cb5300bSChris Mason * 8644cb5300bSChris Mason * This is used by the defragging code to find new and small 8654cb5300bSChris Mason * extents 8664cb5300bSChris Mason */ 8674cb5300bSChris Mason static int find_new_extents(struct btrfs_root *root, 8684cb5300bSChris Mason struct inode *inode, u64 newer_than, 8694cb5300bSChris Mason u64 *off, int thresh) 8704cb5300bSChris Mason { 8714cb5300bSChris Mason struct btrfs_path *path; 8724cb5300bSChris Mason struct btrfs_key min_key; 8734cb5300bSChris Mason struct extent_buffer *leaf; 8744cb5300bSChris Mason struct btrfs_file_extent_item *extent; 8754cb5300bSChris Mason int type; 8764cb5300bSChris Mason int ret; 877a4689d2bSDavid Sterba u64 ino = btrfs_ino(inode); 8784cb5300bSChris Mason 8794cb5300bSChris Mason path = btrfs_alloc_path(); 8804cb5300bSChris Mason if (!path) 8814cb5300bSChris Mason return -ENOMEM; 8824cb5300bSChris Mason 883a4689d2bSDavid Sterba min_key.objectid = ino; 8844cb5300bSChris Mason min_key.type = BTRFS_EXTENT_DATA_KEY; 8854cb5300bSChris Mason min_key.offset = *off; 8864cb5300bSChris Mason 8874cb5300bSChris Mason path->keep_locks = 1; 8884cb5300bSChris Mason 8894cb5300bSChris Mason while (1) { 8906174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &min_key, path, newer_than); 8914cb5300bSChris Mason if (ret != 0) 8924cb5300bSChris Mason goto none; 893a4689d2bSDavid Sterba if (min_key.objectid != ino) 8944cb5300bSChris Mason goto none; 8954cb5300bSChris Mason if (min_key.type != BTRFS_EXTENT_DATA_KEY) 8964cb5300bSChris Mason goto none; 8974cb5300bSChris Mason 8984cb5300bSChris Mason leaf = path->nodes[0]; 8994cb5300bSChris Mason extent = btrfs_item_ptr(leaf, path->slots[0], 9004cb5300bSChris Mason struct btrfs_file_extent_item); 9014cb5300bSChris Mason 9024cb5300bSChris Mason type = btrfs_file_extent_type(leaf, extent); 9034cb5300bSChris Mason if (type == BTRFS_FILE_EXTENT_REG && 9044cb5300bSChris Mason btrfs_file_extent_num_bytes(leaf, extent) < thresh && 9054cb5300bSChris Mason check_defrag_in_cache(inode, min_key.offset, thresh)) { 9064cb5300bSChris Mason *off = min_key.offset; 9074cb5300bSChris Mason btrfs_free_path(path); 9084cb5300bSChris Mason return 0; 9094cb5300bSChris Mason } 9104cb5300bSChris Mason 9114cb5300bSChris Mason if (min_key.offset == (u64)-1) 9124cb5300bSChris Mason goto none; 9134cb5300bSChris Mason 9144cb5300bSChris Mason min_key.offset++; 9154cb5300bSChris Mason btrfs_release_path(path); 9164cb5300bSChris Mason } 9174cb5300bSChris Mason none: 9184cb5300bSChris Mason btrfs_free_path(path); 9194cb5300bSChris Mason return -ENOENT; 9204cb5300bSChris Mason } 9214cb5300bSChris Mason 9226c282eb4SLi Zefan static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start) 92317ce6ef8SLiu Bo { 92417ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 925940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 9266c282eb4SLi Zefan struct extent_map *em; 9276c282eb4SLi Zefan u64 len = PAGE_CACHE_SIZE; 928940100a4SChris Mason 929940100a4SChris Mason /* 930940100a4SChris Mason * hopefully we have this extent in the tree already, try without 931940100a4SChris Mason * the full extent lock 932940100a4SChris Mason */ 933940100a4SChris Mason read_lock(&em_tree->lock); 934940100a4SChris Mason em = lookup_extent_mapping(em_tree, start, len); 935940100a4SChris Mason read_unlock(&em_tree->lock); 936940100a4SChris Mason 937940100a4SChris Mason if (!em) { 938940100a4SChris Mason /* get the big lock and read metadata off disk */ 939d0082371SJeff Mahoney lock_extent(io_tree, start, start + len - 1); 940940100a4SChris Mason em = btrfs_get_extent(inode, NULL, 0, start, len, 0); 941d0082371SJeff Mahoney unlock_extent(io_tree, start, start + len - 1); 942940100a4SChris Mason 9436cf8bfbfSDan Carpenter if (IS_ERR(em)) 9446c282eb4SLi Zefan return NULL; 945940100a4SChris Mason } 946940100a4SChris Mason 9476c282eb4SLi Zefan return em; 9486c282eb4SLi Zefan } 9496c282eb4SLi Zefan 9506c282eb4SLi Zefan static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em) 9516c282eb4SLi Zefan { 9526c282eb4SLi Zefan struct extent_map *next; 9536c282eb4SLi Zefan bool ret = true; 9546c282eb4SLi Zefan 9556c282eb4SLi Zefan /* this is the last extent */ 9566c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 9576c282eb4SLi Zefan return false; 9586c282eb4SLi Zefan 9596c282eb4SLi Zefan next = defrag_lookup_extent(inode, em->start + em->len); 9606c282eb4SLi Zefan if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 9616c282eb4SLi Zefan ret = false; 9626c282eb4SLi Zefan 9636c282eb4SLi Zefan free_extent_map(next); 9646c282eb4SLi Zefan return ret; 9656c282eb4SLi Zefan } 9666c282eb4SLi Zefan 9676c282eb4SLi Zefan static int should_defrag_range(struct inode *inode, u64 start, int thresh, 968a43a2111SAndrew Mahone u64 *last_len, u64 *skip, u64 *defrag_end, 969a43a2111SAndrew Mahone int compress) 9706c282eb4SLi Zefan { 9716c282eb4SLi Zefan struct extent_map *em; 9726c282eb4SLi Zefan int ret = 1; 9736c282eb4SLi Zefan bool next_mergeable = true; 9746c282eb4SLi Zefan 9756c282eb4SLi Zefan /* 9766c282eb4SLi Zefan * make sure that once we start defragging an extent, we keep on 9776c282eb4SLi Zefan * defragging it 9786c282eb4SLi Zefan */ 9796c282eb4SLi Zefan if (start < *defrag_end) 9806c282eb4SLi Zefan return 1; 9816c282eb4SLi Zefan 9826c282eb4SLi Zefan *skip = 0; 9836c282eb4SLi Zefan 9846c282eb4SLi Zefan em = defrag_lookup_extent(inode, start); 9856c282eb4SLi Zefan if (!em) 9866c282eb4SLi Zefan return 0; 9876c282eb4SLi Zefan 988940100a4SChris Mason /* this will cover holes, and inline extents */ 98917ce6ef8SLiu Bo if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 990940100a4SChris Mason ret = 0; 99117ce6ef8SLiu Bo goto out; 99217ce6ef8SLiu Bo } 99317ce6ef8SLiu Bo 9946c282eb4SLi Zefan next_mergeable = defrag_check_next_extent(inode, em); 995940100a4SChris Mason 996940100a4SChris Mason /* 9976c282eb4SLi Zefan * we hit a real extent, if it is big or the next extent is not a 9986c282eb4SLi Zefan * real extent, don't bother defragging it 999940100a4SChris Mason */ 1000a43a2111SAndrew Mahone if (!compress && (*last_len == 0 || *last_len >= thresh) && 10016c282eb4SLi Zefan (em->len >= thresh || !next_mergeable)) 1002940100a4SChris Mason ret = 0; 100317ce6ef8SLiu Bo out: 1004940100a4SChris Mason /* 1005940100a4SChris Mason * last_len ends up being a counter of how many bytes we've defragged. 1006940100a4SChris Mason * every time we choose not to defrag an extent, we reset *last_len 1007940100a4SChris Mason * so that the next tiny extent will force a defrag. 1008940100a4SChris Mason * 1009940100a4SChris Mason * The end result of this is that tiny extents before a single big 1010940100a4SChris Mason * extent will force at least part of that big extent to be defragged. 1011940100a4SChris Mason */ 1012940100a4SChris Mason if (ret) { 1013940100a4SChris Mason *defrag_end = extent_map_end(em); 1014940100a4SChris Mason } else { 1015940100a4SChris Mason *last_len = 0; 1016940100a4SChris Mason *skip = extent_map_end(em); 1017940100a4SChris Mason *defrag_end = 0; 1018940100a4SChris Mason } 1019940100a4SChris Mason 1020940100a4SChris Mason free_extent_map(em); 1021940100a4SChris Mason return ret; 1022940100a4SChris Mason } 1023940100a4SChris Mason 10244cb5300bSChris Mason /* 10254cb5300bSChris Mason * it doesn't do much good to defrag one or two pages 10264cb5300bSChris Mason * at a time. This pulls in a nice chunk of pages 10274cb5300bSChris Mason * to COW and defrag. 10284cb5300bSChris Mason * 10294cb5300bSChris Mason * It also makes sure the delalloc code has enough 10304cb5300bSChris Mason * dirty data to avoid making new small extents as part 10314cb5300bSChris Mason * of the defrag 10324cb5300bSChris Mason * 10334cb5300bSChris Mason * It's a good idea to start RA on this range 10344cb5300bSChris Mason * before calling this. 10354cb5300bSChris Mason */ 10364cb5300bSChris Mason static int cluster_pages_for_defrag(struct inode *inode, 10374cb5300bSChris Mason struct page **pages, 10384cb5300bSChris Mason unsigned long start_index, 10394cb5300bSChris Mason int num_pages) 1040f46b5a66SChristoph Hellwig { 10414cb5300bSChris Mason unsigned long file_end; 10424cb5300bSChris Mason u64 isize = i_size_read(inode); 1043f46b5a66SChristoph Hellwig u64 page_start; 1044f46b5a66SChristoph Hellwig u64 page_end; 10451f12bd06SLiu Bo u64 page_cnt; 10464cb5300bSChris Mason int ret; 10474cb5300bSChris Mason int i; 10484cb5300bSChris Mason int i_done; 10494cb5300bSChris Mason struct btrfs_ordered_extent *ordered; 10504cb5300bSChris Mason struct extent_state *cached_state = NULL; 1051600a45e1SMiao Xie struct extent_io_tree *tree; 10523b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 10534cb5300bSChris Mason 10544cb5300bSChris Mason file_end = (isize - 1) >> PAGE_CACHE_SHIFT; 10551f12bd06SLiu Bo if (!isize || start_index > file_end) 10561f12bd06SLiu Bo return 0; 10571f12bd06SLiu Bo 10581f12bd06SLiu Bo page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1); 10594cb5300bSChris Mason 10604cb5300bSChris Mason ret = btrfs_delalloc_reserve_space(inode, 10611f12bd06SLiu Bo page_cnt << PAGE_CACHE_SHIFT); 10624cb5300bSChris Mason if (ret) 10634cb5300bSChris Mason return ret; 10644cb5300bSChris Mason i_done = 0; 1065600a45e1SMiao Xie tree = &BTRFS_I(inode)->io_tree; 10664cb5300bSChris Mason 10674cb5300bSChris Mason /* step one, lock all the pages */ 10681f12bd06SLiu Bo for (i = 0; i < page_cnt; i++) { 10694cb5300bSChris Mason struct page *page; 1070600a45e1SMiao Xie again: 1071a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, 10723b16a4e3SJosef Bacik start_index + i, mask); 10734cb5300bSChris Mason if (!page) 10744cb5300bSChris Mason break; 10754cb5300bSChris Mason 1076600a45e1SMiao Xie page_start = page_offset(page); 1077600a45e1SMiao Xie page_end = page_start + PAGE_CACHE_SIZE - 1; 1078600a45e1SMiao Xie while (1) { 1079d0082371SJeff Mahoney lock_extent(tree, page_start, page_end); 1080600a45e1SMiao Xie ordered = btrfs_lookup_ordered_extent(inode, 1081600a45e1SMiao Xie page_start); 1082d0082371SJeff Mahoney unlock_extent(tree, page_start, page_end); 1083600a45e1SMiao Xie if (!ordered) 1084600a45e1SMiao Xie break; 1085600a45e1SMiao Xie 1086600a45e1SMiao Xie unlock_page(page); 1087600a45e1SMiao Xie btrfs_start_ordered_extent(inode, ordered, 1); 1088600a45e1SMiao Xie btrfs_put_ordered_extent(ordered); 1089600a45e1SMiao Xie lock_page(page); 10901f12bd06SLiu Bo /* 10911f12bd06SLiu Bo * we unlocked the page above, so we need check if 10921f12bd06SLiu Bo * it was released or not. 10931f12bd06SLiu Bo */ 10941f12bd06SLiu Bo if (page->mapping != inode->i_mapping) { 10951f12bd06SLiu Bo unlock_page(page); 10961f12bd06SLiu Bo page_cache_release(page); 10971f12bd06SLiu Bo goto again; 10981f12bd06SLiu Bo } 1099600a45e1SMiao Xie } 1100600a45e1SMiao Xie 11014cb5300bSChris Mason if (!PageUptodate(page)) { 11024cb5300bSChris Mason btrfs_readpage(NULL, page); 11034cb5300bSChris Mason lock_page(page); 11044cb5300bSChris Mason if (!PageUptodate(page)) { 11054cb5300bSChris Mason unlock_page(page); 11064cb5300bSChris Mason page_cache_release(page); 11074cb5300bSChris Mason ret = -EIO; 11084cb5300bSChris Mason break; 11094cb5300bSChris Mason } 11104cb5300bSChris Mason } 1111600a45e1SMiao Xie 1112600a45e1SMiao Xie if (page->mapping != inode->i_mapping) { 1113600a45e1SMiao Xie unlock_page(page); 1114600a45e1SMiao Xie page_cache_release(page); 1115600a45e1SMiao Xie goto again; 1116600a45e1SMiao Xie } 1117600a45e1SMiao Xie 11184cb5300bSChris Mason pages[i] = page; 11194cb5300bSChris Mason i_done++; 11204cb5300bSChris Mason } 11214cb5300bSChris Mason if (!i_done || ret) 11224cb5300bSChris Mason goto out; 11234cb5300bSChris Mason 11244cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 11254cb5300bSChris Mason goto out; 11264cb5300bSChris Mason 11274cb5300bSChris Mason /* 11284cb5300bSChris Mason * so now we have a nice long stream of locked 11294cb5300bSChris Mason * and up to date pages, lets wait on them 11304cb5300bSChris Mason */ 11314cb5300bSChris Mason for (i = 0; i < i_done; i++) 11324cb5300bSChris Mason wait_on_page_writeback(pages[i]); 11334cb5300bSChris Mason 11344cb5300bSChris Mason page_start = page_offset(pages[0]); 11354cb5300bSChris Mason page_end = page_offset(pages[i_done - 1]) + PAGE_CACHE_SIZE; 11364cb5300bSChris Mason 11374cb5300bSChris Mason lock_extent_bits(&BTRFS_I(inode)->io_tree, 1138d0082371SJeff Mahoney page_start, page_end - 1, 0, &cached_state); 11394cb5300bSChris Mason clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, 11404cb5300bSChris Mason page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | 11419e8a4a8bSLiu Bo EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, 11429e8a4a8bSLiu Bo &cached_state, GFP_NOFS); 11434cb5300bSChris Mason 11441f12bd06SLiu Bo if (i_done != page_cnt) { 11459e0baf60SJosef Bacik spin_lock(&BTRFS_I(inode)->lock); 11469e0baf60SJosef Bacik BTRFS_I(inode)->outstanding_extents++; 11479e0baf60SJosef Bacik spin_unlock(&BTRFS_I(inode)->lock); 11484cb5300bSChris Mason btrfs_delalloc_release_space(inode, 11491f12bd06SLiu Bo (page_cnt - i_done) << PAGE_CACHE_SHIFT); 11504cb5300bSChris Mason } 11514cb5300bSChris Mason 11524cb5300bSChris Mason 11539e8a4a8bSLiu Bo set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, 11549e8a4a8bSLiu Bo &cached_state, GFP_NOFS); 11554cb5300bSChris Mason 11564cb5300bSChris Mason unlock_extent_cached(&BTRFS_I(inode)->io_tree, 11574cb5300bSChris Mason page_start, page_end - 1, &cached_state, 11584cb5300bSChris Mason GFP_NOFS); 11594cb5300bSChris Mason 11604cb5300bSChris Mason for (i = 0; i < i_done; i++) { 11614cb5300bSChris Mason clear_page_dirty_for_io(pages[i]); 11624cb5300bSChris Mason ClearPageChecked(pages[i]); 11634cb5300bSChris Mason set_page_extent_mapped(pages[i]); 11644cb5300bSChris Mason set_page_dirty(pages[i]); 11654cb5300bSChris Mason unlock_page(pages[i]); 11664cb5300bSChris Mason page_cache_release(pages[i]); 11674cb5300bSChris Mason } 11684cb5300bSChris Mason return i_done; 11694cb5300bSChris Mason out: 11704cb5300bSChris Mason for (i = 0; i < i_done; i++) { 11714cb5300bSChris Mason unlock_page(pages[i]); 11724cb5300bSChris Mason page_cache_release(pages[i]); 11734cb5300bSChris Mason } 11741f12bd06SLiu Bo btrfs_delalloc_release_space(inode, page_cnt << PAGE_CACHE_SHIFT); 11754cb5300bSChris Mason return ret; 11764cb5300bSChris Mason 11774cb5300bSChris Mason } 11784cb5300bSChris Mason 11794cb5300bSChris Mason int btrfs_defrag_file(struct inode *inode, struct file *file, 11804cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 11814cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 11824cb5300bSChris Mason { 11834cb5300bSChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 11844cb5300bSChris Mason struct file_ra_state *ra = NULL; 11854cb5300bSChris Mason unsigned long last_index; 1186151a31b2SLi Zefan u64 isize = i_size_read(inode); 1187940100a4SChris Mason u64 last_len = 0; 1188940100a4SChris Mason u64 skip = 0; 1189940100a4SChris Mason u64 defrag_end = 0; 11904cb5300bSChris Mason u64 newer_off = range->start; 1191f46b5a66SChristoph Hellwig unsigned long i; 1192008873eaSLi Zefan unsigned long ra_index = 0; 1193f46b5a66SChristoph Hellwig int ret; 11944cb5300bSChris Mason int defrag_count = 0; 11951a419d85SLi Zefan int compress_type = BTRFS_COMPRESS_ZLIB; 11964cb5300bSChris Mason int extent_thresh = range->extent_thresh; 1197008873eaSLi Zefan int max_cluster = (256 * 1024) >> PAGE_CACHE_SHIFT; 1198008873eaSLi Zefan int cluster = max_cluster; 11994cb5300bSChris Mason u64 new_align = ~((u64)128 * 1024 - 1); 12004cb5300bSChris Mason struct page **pages = NULL; 12014cb5300bSChris Mason 12020abd5b17SLiu Bo if (isize == 0) 12030abd5b17SLiu Bo return 0; 12040abd5b17SLiu Bo 12050abd5b17SLiu Bo if (range->start >= isize) 12060abd5b17SLiu Bo return -EINVAL; 12071a419d85SLi Zefan 12081a419d85SLi Zefan if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 12091a419d85SLi Zefan if (range->compress_type > BTRFS_COMPRESS_TYPES) 12101a419d85SLi Zefan return -EINVAL; 12111a419d85SLi Zefan if (range->compress_type) 12121a419d85SLi Zefan compress_type = range->compress_type; 12131a419d85SLi Zefan } 1214f46b5a66SChristoph Hellwig 12150abd5b17SLiu Bo if (extent_thresh == 0) 12160abd5b17SLiu Bo extent_thresh = 256 * 1024; 1217f46b5a66SChristoph Hellwig 12184cb5300bSChris Mason /* 12194cb5300bSChris Mason * if we were not given a file, allocate a readahead 12204cb5300bSChris Mason * context 12214cb5300bSChris Mason */ 12224cb5300bSChris Mason if (!file) { 12234cb5300bSChris Mason ra = kzalloc(sizeof(*ra), GFP_NOFS); 12244cb5300bSChris Mason if (!ra) 12254cb5300bSChris Mason return -ENOMEM; 12264cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 12274cb5300bSChris Mason } else { 12284cb5300bSChris Mason ra = &file->f_ra; 12294cb5300bSChris Mason } 12304cb5300bSChris Mason 1231d9b0d9baSDulshani Gunawardhana pages = kmalloc_array(max_cluster, sizeof(struct page *), 12324cb5300bSChris Mason GFP_NOFS); 12334cb5300bSChris Mason if (!pages) { 12344cb5300bSChris Mason ret = -ENOMEM; 12354cb5300bSChris Mason goto out_ra; 12364cb5300bSChris Mason } 12374cb5300bSChris Mason 12384cb5300bSChris Mason /* find the last page to defrag */ 12391e701a32SChris Mason if (range->start + range->len > range->start) { 1240151a31b2SLi Zefan last_index = min_t(u64, isize - 1, 12411e701a32SChris Mason range->start + range->len - 1) >> PAGE_CACHE_SHIFT; 12421e701a32SChris Mason } else { 1243151a31b2SLi Zefan last_index = (isize - 1) >> PAGE_CACHE_SHIFT; 12441e701a32SChris Mason } 12451e701a32SChris Mason 12464cb5300bSChris Mason if (newer_than) { 12474cb5300bSChris Mason ret = find_new_extents(root, inode, newer_than, 12484cb5300bSChris Mason &newer_off, 64 * 1024); 12494cb5300bSChris Mason if (!ret) { 12504cb5300bSChris Mason range->start = newer_off; 12514cb5300bSChris Mason /* 12524cb5300bSChris Mason * we always align our defrag to help keep 12534cb5300bSChris Mason * the extents in the file evenly spaced 12544cb5300bSChris Mason */ 12554cb5300bSChris Mason i = (newer_off & new_align) >> PAGE_CACHE_SHIFT; 12564cb5300bSChris Mason } else 12574cb5300bSChris Mason goto out_ra; 12584cb5300bSChris Mason } else { 12591e701a32SChris Mason i = range->start >> PAGE_CACHE_SHIFT; 12604cb5300bSChris Mason } 12614cb5300bSChris Mason if (!max_to_defrag) 12627ec31b54SLiu Bo max_to_defrag = last_index + 1; 12634cb5300bSChris Mason 12642a0f7f57SLi Zefan /* 12652a0f7f57SLi Zefan * make writeback starts from i, so the defrag range can be 12662a0f7f57SLi Zefan * written sequentially. 12672a0f7f57SLi Zefan */ 12682a0f7f57SLi Zefan if (i < inode->i_mapping->writeback_index) 12692a0f7f57SLi Zefan inode->i_mapping->writeback_index = i; 12702a0f7f57SLi Zefan 1271f7f43cc8SChris Mason while (i <= last_index && defrag_count < max_to_defrag && 1272f7f43cc8SChris Mason (i < (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> 1273f7f43cc8SChris Mason PAGE_CACHE_SHIFT)) { 12744cb5300bSChris Mason /* 12754cb5300bSChris Mason * make sure we stop running if someone unmounts 12764cb5300bSChris Mason * the FS 12774cb5300bSChris Mason */ 12784cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 12794cb5300bSChris Mason break; 12804cb5300bSChris Mason 1281210549ebSDavid Sterba if (btrfs_defrag_cancelled(root->fs_info)) { 1282efe120a0SFrank Holton printk(KERN_DEBUG "BTRFS: defrag_file cancelled\n"); 1283210549ebSDavid Sterba ret = -EAGAIN; 1284210549ebSDavid Sterba break; 1285210549ebSDavid Sterba } 1286210549ebSDavid Sterba 128766c26892SLiu Bo if (!should_defrag_range(inode, (u64)i << PAGE_CACHE_SHIFT, 12886c282eb4SLi Zefan extent_thresh, &last_len, &skip, 1289a43a2111SAndrew Mahone &defrag_end, range->flags & 1290a43a2111SAndrew Mahone BTRFS_DEFRAG_RANGE_COMPRESS)) { 1291940100a4SChris Mason unsigned long next; 1292940100a4SChris Mason /* 1293940100a4SChris Mason * the should_defrag function tells us how much to skip 1294940100a4SChris Mason * bump our counter by the suggested amount 1295940100a4SChris Mason */ 1296940100a4SChris Mason next = (skip + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1297940100a4SChris Mason i = max(i + 1, next); 1298940100a4SChris Mason continue; 1299940100a4SChris Mason } 1300008873eaSLi Zefan 1301008873eaSLi Zefan if (!newer_than) { 1302008873eaSLi Zefan cluster = (PAGE_CACHE_ALIGN(defrag_end) >> 1303008873eaSLi Zefan PAGE_CACHE_SHIFT) - i; 1304008873eaSLi Zefan cluster = min(cluster, max_cluster); 1305008873eaSLi Zefan } else { 1306008873eaSLi Zefan cluster = max_cluster; 1307008873eaSLi Zefan } 1308008873eaSLi Zefan 1309008873eaSLi Zefan if (i + cluster > ra_index) { 1310008873eaSLi Zefan ra_index = max(i, ra_index); 1311008873eaSLi Zefan btrfs_force_ra(inode->i_mapping, ra, file, ra_index, 1312008873eaSLi Zefan cluster); 1313008873eaSLi Zefan ra_index += max_cluster; 1314008873eaSLi Zefan } 13154cb5300bSChris Mason 1316ecb8bea8SLiu Bo mutex_lock(&inode->i_mutex); 1317633085c7SFilipe David Borba Manana if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) 1318633085c7SFilipe David Borba Manana BTRFS_I(inode)->force_compress = compress_type; 1319008873eaSLi Zefan ret = cluster_pages_for_defrag(inode, pages, i, cluster); 1320ecb8bea8SLiu Bo if (ret < 0) { 1321ecb8bea8SLiu Bo mutex_unlock(&inode->i_mutex); 13224cb5300bSChris Mason goto out_ra; 1323ecb8bea8SLiu Bo } 13244cb5300bSChris Mason 13254cb5300bSChris Mason defrag_count += ret; 1326d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited(inode->i_mapping); 1327ecb8bea8SLiu Bo mutex_unlock(&inode->i_mutex); 13284cb5300bSChris Mason 13294cb5300bSChris Mason if (newer_than) { 13304cb5300bSChris Mason if (newer_off == (u64)-1) 13314cb5300bSChris Mason break; 13324cb5300bSChris Mason 1333e1f041e1SLiu Bo if (ret > 0) 1334e1f041e1SLiu Bo i += ret; 1335e1f041e1SLiu Bo 13364cb5300bSChris Mason newer_off = max(newer_off + 1, 13374cb5300bSChris Mason (u64)i << PAGE_CACHE_SHIFT); 13384cb5300bSChris Mason 13394cb5300bSChris Mason ret = find_new_extents(root, inode, 13404cb5300bSChris Mason newer_than, &newer_off, 13414cb5300bSChris Mason 64 * 1024); 13424cb5300bSChris Mason if (!ret) { 13434cb5300bSChris Mason range->start = newer_off; 13444cb5300bSChris Mason i = (newer_off & new_align) >> PAGE_CACHE_SHIFT; 13454cb5300bSChris Mason } else { 13464cb5300bSChris Mason break; 1347940100a4SChris Mason } 13484cb5300bSChris Mason } else { 1349008873eaSLi Zefan if (ret > 0) { 1350cbcc8326SLi Zefan i += ret; 1351008873eaSLi Zefan last_len += ret << PAGE_CACHE_SHIFT; 1352008873eaSLi Zefan } else { 1353940100a4SChris Mason i++; 1354008873eaSLi Zefan last_len = 0; 1355008873eaSLi Zefan } 1356f46b5a66SChristoph Hellwig } 13574cb5300bSChris Mason } 1358f46b5a66SChristoph Hellwig 13591e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) 13601e701a32SChris Mason filemap_flush(inode->i_mapping); 13611e701a32SChris Mason 13621e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 13631e701a32SChris Mason /* the filemap_flush will queue IO into the worker threads, but 13641e701a32SChris Mason * we have to make sure the IO is actually started and that 13651e701a32SChris Mason * ordered extents get created before we return 13661e701a32SChris Mason */ 13671e701a32SChris Mason atomic_inc(&root->fs_info->async_submit_draining); 13681e701a32SChris Mason while (atomic_read(&root->fs_info->nr_async_submits) || 13691e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages)) { 13701e701a32SChris Mason wait_event(root->fs_info->async_submit_wait, 13711e701a32SChris Mason (atomic_read(&root->fs_info->nr_async_submits) == 0 && 13721e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages) == 0)); 13731e701a32SChris Mason } 13741e701a32SChris Mason atomic_dec(&root->fs_info->async_submit_draining); 13751e701a32SChris Mason } 13761e701a32SChris Mason 13771a419d85SLi Zefan if (range->compress_type == BTRFS_COMPRESS_LZO) { 13782b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO); 13791a419d85SLi Zefan } 13801a419d85SLi Zefan 138160ccf82fSDiego Calleja ret = defrag_count; 1382940100a4SChris Mason 13834cb5300bSChris Mason out_ra: 1384633085c7SFilipe David Borba Manana if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 1385633085c7SFilipe David Borba Manana mutex_lock(&inode->i_mutex); 1386633085c7SFilipe David Borba Manana BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE; 1387633085c7SFilipe David Borba Manana mutex_unlock(&inode->i_mutex); 1388633085c7SFilipe David Borba Manana } 13894cb5300bSChris Mason if (!file) 13904cb5300bSChris Mason kfree(ra); 13914cb5300bSChris Mason kfree(pages); 1392940100a4SChris Mason return ret; 1393f46b5a66SChristoph Hellwig } 1394f46b5a66SChristoph Hellwig 1395198605a8SMiao Xie static noinline int btrfs_ioctl_resize(struct file *file, 139676dda93cSYan, Zheng void __user *arg) 1397f46b5a66SChristoph Hellwig { 1398f46b5a66SChristoph Hellwig u64 new_size; 1399f46b5a66SChristoph Hellwig u64 old_size; 1400f46b5a66SChristoph Hellwig u64 devid = 1; 1401496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 1402f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1403f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1404f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1405f46b5a66SChristoph Hellwig char *sizestr; 1406f46b5a66SChristoph Hellwig char *devstr = NULL; 1407f46b5a66SChristoph Hellwig int ret = 0; 1408f46b5a66SChristoph Hellwig int mod = 0; 1409f46b5a66SChristoph Hellwig 1410e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1411e441d54dSChris Mason return -EPERM; 1412e441d54dSChris Mason 1413198605a8SMiao Xie ret = mnt_want_write_file(file); 1414198605a8SMiao Xie if (ret) 1415198605a8SMiao Xie return ret; 1416198605a8SMiao Xie 14175ac00addSStefan Behrens if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 14185ac00addSStefan Behrens 1)) { 141997547676SMiao Xie mnt_drop_write_file(file); 1420e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 1421c9e9f97bSIlya Dryomov } 1422c9e9f97bSIlya Dryomov 14235ac00addSStefan Behrens mutex_lock(&root->fs_info->volume_mutex); 1424dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1425c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1426c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1427c9e9f97bSIlya Dryomov goto out; 1428c9e9f97bSIlya Dryomov } 14295516e595SMark Fasheh 14305516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1431f46b5a66SChristoph Hellwig 1432f46b5a66SChristoph Hellwig sizestr = vol_args->name; 1433f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 1434f46b5a66SChristoph Hellwig if (devstr) { 1435f46b5a66SChristoph Hellwig char *end; 1436f46b5a66SChristoph Hellwig sizestr = devstr + 1; 1437f46b5a66SChristoph Hellwig *devstr = '\0'; 1438f46b5a66SChristoph Hellwig devstr = vol_args->name; 1439f46b5a66SChristoph Hellwig devid = simple_strtoull(devstr, &end, 10); 1440dfd79829SMiao Xie if (!devid) { 1441dfd79829SMiao Xie ret = -EINVAL; 1442dfd79829SMiao Xie goto out_free; 1443dfd79829SMiao Xie } 1444efe120a0SFrank Holton btrfs_info(root->fs_info, "resizing devid %llu", devid); 1445f46b5a66SChristoph Hellwig } 1446dba60f3fSMiao Xie 1447aa1b8cd4SStefan Behrens device = btrfs_find_device(root->fs_info, devid, NULL, NULL); 1448f46b5a66SChristoph Hellwig if (!device) { 1449efe120a0SFrank Holton btrfs_info(root->fs_info, "resizer unable to find device %llu", 1450c1c9ff7cSGeert Uytterhoeven devid); 1451dfd79829SMiao Xie ret = -ENODEV; 1452c9e9f97bSIlya Dryomov goto out_free; 1453f46b5a66SChristoph Hellwig } 1454dba60f3fSMiao Xie 1455dba60f3fSMiao Xie if (!device->writeable) { 1456efe120a0SFrank Holton btrfs_info(root->fs_info, 1457efe120a0SFrank Holton "resizer unable to apply on readonly device %llu", 1458c1c9ff7cSGeert Uytterhoeven devid); 1459dfd79829SMiao Xie ret = -EPERM; 14604e42ae1bSLiu Bo goto out_free; 14614e42ae1bSLiu Bo } 14624e42ae1bSLiu Bo 1463f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 1464f46b5a66SChristoph Hellwig new_size = device->bdev->bd_inode->i_size; 1465f46b5a66SChristoph Hellwig else { 1466f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 1467f46b5a66SChristoph Hellwig mod = -1; 1468f46b5a66SChristoph Hellwig sizestr++; 1469f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 1470f46b5a66SChristoph Hellwig mod = 1; 1471f46b5a66SChristoph Hellwig sizestr++; 1472f46b5a66SChristoph Hellwig } 147391748467SAkinobu Mita new_size = memparse(sizestr, NULL); 1474f46b5a66SChristoph Hellwig if (new_size == 0) { 1475f46b5a66SChristoph Hellwig ret = -EINVAL; 1476c9e9f97bSIlya Dryomov goto out_free; 1477f46b5a66SChristoph Hellwig } 1478f46b5a66SChristoph Hellwig } 1479f46b5a66SChristoph Hellwig 148063a212abSStefan Behrens if (device->is_tgtdev_for_dev_replace) { 1481dfd79829SMiao Xie ret = -EPERM; 148263a212abSStefan Behrens goto out_free; 148363a212abSStefan Behrens } 148463a212abSStefan Behrens 1485f46b5a66SChristoph Hellwig old_size = device->total_bytes; 1486f46b5a66SChristoph Hellwig 1487f46b5a66SChristoph Hellwig if (mod < 0) { 1488f46b5a66SChristoph Hellwig if (new_size > old_size) { 1489f46b5a66SChristoph Hellwig ret = -EINVAL; 1490c9e9f97bSIlya Dryomov goto out_free; 1491f46b5a66SChristoph Hellwig } 1492f46b5a66SChristoph Hellwig new_size = old_size - new_size; 1493f46b5a66SChristoph Hellwig } else if (mod > 0) { 1494eb8052e0SWenliang Fan if (new_size > ULLONG_MAX - old_size) { 1495eb8052e0SWenliang Fan ret = -EINVAL; 1496eb8052e0SWenliang Fan goto out_free; 1497eb8052e0SWenliang Fan } 1498f46b5a66SChristoph Hellwig new_size = old_size + new_size; 1499f46b5a66SChristoph Hellwig } 1500f46b5a66SChristoph Hellwig 1501f46b5a66SChristoph Hellwig if (new_size < 256 * 1024 * 1024) { 1502f46b5a66SChristoph Hellwig ret = -EINVAL; 1503c9e9f97bSIlya Dryomov goto out_free; 1504f46b5a66SChristoph Hellwig } 1505f46b5a66SChristoph Hellwig if (new_size > device->bdev->bd_inode->i_size) { 1506f46b5a66SChristoph Hellwig ret = -EFBIG; 1507c9e9f97bSIlya Dryomov goto out_free; 1508f46b5a66SChristoph Hellwig } 1509f46b5a66SChristoph Hellwig 1510f46b5a66SChristoph Hellwig do_div(new_size, root->sectorsize); 1511f46b5a66SChristoph Hellwig new_size *= root->sectorsize; 1512f46b5a66SChristoph Hellwig 1513efe120a0SFrank Holton printk_in_rcu(KERN_INFO "BTRFS: new size for %s is %llu\n", 1514c1c9ff7cSGeert Uytterhoeven rcu_str_deref(device->name), new_size); 1515f46b5a66SChristoph Hellwig 1516f46b5a66SChristoph Hellwig if (new_size > old_size) { 1517a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 151898d5dc13STsutomu Itoh if (IS_ERR(trans)) { 151998d5dc13STsutomu Itoh ret = PTR_ERR(trans); 1520c9e9f97bSIlya Dryomov goto out_free; 152198d5dc13STsutomu Itoh } 1522f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 1523f46b5a66SChristoph Hellwig btrfs_commit_transaction(trans, root); 1524ece7d20eSMike Fleetwood } else if (new_size < old_size) { 1525f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 15260253f40eSjeff.liu } /* equal, nothing need to do */ 1527f46b5a66SChristoph Hellwig 1528c9e9f97bSIlya Dryomov out_free: 1529f46b5a66SChristoph Hellwig kfree(vol_args); 1530c9e9f97bSIlya Dryomov out: 1531c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 15325ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 153318f39c41SIlya Dryomov mnt_drop_write_file(file); 1534f46b5a66SChristoph Hellwig return ret; 1535f46b5a66SChristoph Hellwig } 1536f46b5a66SChristoph Hellwig 153772fd032eSSage Weil static noinline int btrfs_ioctl_snap_create_transid(struct file *file, 15386f72c7e2SArne Jansen char *name, unsigned long fd, int subvol, 15396f72c7e2SArne Jansen u64 *transid, bool readonly, 15408696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 1541f46b5a66SChristoph Hellwig { 1542f46b5a66SChristoph Hellwig int namelen; 15433de4586cSChris Mason int ret = 0; 1544f46b5a66SChristoph Hellwig 1545a874a63eSLiu Bo ret = mnt_want_write_file(file); 1546a874a63eSLiu Bo if (ret) 1547a874a63eSLiu Bo goto out; 1548a874a63eSLiu Bo 154972fd032eSSage Weil namelen = strlen(name); 155072fd032eSSage Weil if (strchr(name, '/')) { 1551f46b5a66SChristoph Hellwig ret = -EINVAL; 1552a874a63eSLiu Bo goto out_drop_write; 1553f46b5a66SChristoph Hellwig } 1554f46b5a66SChristoph Hellwig 155516780cabSChris Mason if (name[0] == '.' && 155616780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 155716780cabSChris Mason ret = -EEXIST; 1558a874a63eSLiu Bo goto out_drop_write; 155916780cabSChris Mason } 156016780cabSChris Mason 15613de4586cSChris Mason if (subvol) { 156272fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 15636f72c7e2SArne Jansen NULL, transid, readonly, inherit); 1564cb8e7090SChristoph Hellwig } else { 15652903ff01SAl Viro struct fd src = fdget(fd); 15663de4586cSChris Mason struct inode *src_inode; 15672903ff01SAl Viro if (!src.file) { 15683de4586cSChris Mason ret = -EINVAL; 1569a874a63eSLiu Bo goto out_drop_write; 15703de4586cSChris Mason } 15713de4586cSChris Mason 1572496ad9aaSAl Viro src_inode = file_inode(src.file); 1573496ad9aaSAl Viro if (src_inode->i_sb != file_inode(file)->i_sb) { 1574efe120a0SFrank Holton btrfs_info(BTRFS_I(src_inode)->root->fs_info, 1575efe120a0SFrank Holton "Snapshot src from another FS"); 15763de4586cSChris Mason ret = -EINVAL; 1577d0242061SDavid Sterba } else if (!inode_owner_or_capable(src_inode)) { 1578d0242061SDavid Sterba /* 1579d0242061SDavid Sterba * Subvolume creation is not restricted, but snapshots 1580d0242061SDavid Sterba * are limited to own subvolumes only 1581d0242061SDavid Sterba */ 1582d0242061SDavid Sterba ret = -EPERM; 1583ecd18815SAl Viro } else { 158472fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 158572fd032eSSage Weil BTRFS_I(src_inode)->root, 15866f72c7e2SArne Jansen transid, readonly, inherit); 1587ecd18815SAl Viro } 15882903ff01SAl Viro fdput(src); 1589cb8e7090SChristoph Hellwig } 1590a874a63eSLiu Bo out_drop_write: 1591a874a63eSLiu Bo mnt_drop_write_file(file); 1592f46b5a66SChristoph Hellwig out: 159372fd032eSSage Weil return ret; 159472fd032eSSage Weil } 159572fd032eSSage Weil 159672fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 1597fa0d2b9bSLi Zefan void __user *arg, int subvol) 159872fd032eSSage Weil { 1599fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 160072fd032eSSage Weil int ret; 160172fd032eSSage Weil 1602fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1603fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1604fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1605fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1606fa0d2b9bSLi Zefan 1607fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1608b83cc969SLi Zefan vol_args->fd, subvol, 16096f72c7e2SArne Jansen NULL, false, NULL); 1610fa0d2b9bSLi Zefan 1611fa0d2b9bSLi Zefan kfree(vol_args); 1612fa0d2b9bSLi Zefan return ret; 1613fa0d2b9bSLi Zefan } 1614fa0d2b9bSLi Zefan 1615fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1616fa0d2b9bSLi Zefan void __user *arg, int subvol) 1617fa0d2b9bSLi Zefan { 1618fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 1619fa0d2b9bSLi Zefan int ret; 1620fdfb1e4fSLi Zefan u64 transid = 0; 1621fdfb1e4fSLi Zefan u64 *ptr = NULL; 1622b83cc969SLi Zefan bool readonly = false; 16236f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 162472fd032eSSage Weil 1625fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1626fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1627fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1628fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1629fdfb1e4fSLi Zefan 1630b83cc969SLi Zefan if (vol_args->flags & 16316f72c7e2SArne Jansen ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY | 16326f72c7e2SArne Jansen BTRFS_SUBVOL_QGROUP_INHERIT)) { 1633b83cc969SLi Zefan ret = -EOPNOTSUPP; 1634fdfb1e4fSLi Zefan goto out; 1635fdfb1e4fSLi Zefan } 1636fdfb1e4fSLi Zefan 1637fa0d2b9bSLi Zefan if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) 1638fdfb1e4fSLi Zefan ptr = &transid; 1639b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1640b83cc969SLi Zefan readonly = true; 16416f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 16426f72c7e2SArne Jansen if (vol_args->size > PAGE_CACHE_SIZE) { 16436f72c7e2SArne Jansen ret = -EINVAL; 16446f72c7e2SArne Jansen goto out; 16456f72c7e2SArne Jansen } 16466f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 16476f72c7e2SArne Jansen if (IS_ERR(inherit)) { 16486f72c7e2SArne Jansen ret = PTR_ERR(inherit); 16496f72c7e2SArne Jansen goto out; 16506f72c7e2SArne Jansen } 16516f72c7e2SArne Jansen } 165275eaa0e2SSage Weil 1653fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 16546f72c7e2SArne Jansen vol_args->fd, subvol, ptr, 16558696c533SMiao Xie readonly, inherit); 165675eaa0e2SSage Weil 1657fdfb1e4fSLi Zefan if (ret == 0 && ptr && 165875eaa0e2SSage Weil copy_to_user(arg + 1659fdfb1e4fSLi Zefan offsetof(struct btrfs_ioctl_vol_args_v2, 1660fdfb1e4fSLi Zefan transid), ptr, sizeof(*ptr))) 166175eaa0e2SSage Weil ret = -EFAULT; 1662fdfb1e4fSLi Zefan out: 1663f46b5a66SChristoph Hellwig kfree(vol_args); 16646f72c7e2SArne Jansen kfree(inherit); 1665f46b5a66SChristoph Hellwig return ret; 1666f46b5a66SChristoph Hellwig } 1667f46b5a66SChristoph Hellwig 16680caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 16690caa102dSLi Zefan void __user *arg) 16700caa102dSLi Zefan { 1671496ad9aaSAl Viro struct inode *inode = file_inode(file); 16720caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 16730caa102dSLi Zefan int ret = 0; 16740caa102dSLi Zefan u64 flags = 0; 16750caa102dSLi Zefan 167633345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) 16770caa102dSLi Zefan return -EINVAL; 16780caa102dSLi Zefan 16790caa102dSLi Zefan down_read(&root->fs_info->subvol_sem); 16800caa102dSLi Zefan if (btrfs_root_readonly(root)) 16810caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 16820caa102dSLi Zefan up_read(&root->fs_info->subvol_sem); 16830caa102dSLi Zefan 16840caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 16850caa102dSLi Zefan ret = -EFAULT; 16860caa102dSLi Zefan 16870caa102dSLi Zefan return ret; 16880caa102dSLi Zefan } 16890caa102dSLi Zefan 16900caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 16910caa102dSLi Zefan void __user *arg) 16920caa102dSLi Zefan { 1693496ad9aaSAl Viro struct inode *inode = file_inode(file); 16940caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 16950caa102dSLi Zefan struct btrfs_trans_handle *trans; 16960caa102dSLi Zefan u64 root_flags; 16970caa102dSLi Zefan u64 flags; 16980caa102dSLi Zefan int ret = 0; 16990caa102dSLi Zefan 1700bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 1701bd60ea0fSDavid Sterba return -EPERM; 1702bd60ea0fSDavid Sterba 1703b9ca0664SLiu Bo ret = mnt_want_write_file(file); 1704b9ca0664SLiu Bo if (ret) 1705b9ca0664SLiu Bo goto out; 17060caa102dSLi Zefan 1707b9ca0664SLiu Bo if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 1708b9ca0664SLiu Bo ret = -EINVAL; 1709b9ca0664SLiu Bo goto out_drop_write; 1710b9ca0664SLiu Bo } 17110caa102dSLi Zefan 1712b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 1713b9ca0664SLiu Bo ret = -EFAULT; 1714b9ca0664SLiu Bo goto out_drop_write; 1715b9ca0664SLiu Bo } 17160caa102dSLi Zefan 1717b9ca0664SLiu Bo if (flags & BTRFS_SUBVOL_CREATE_ASYNC) { 1718b9ca0664SLiu Bo ret = -EINVAL; 1719b9ca0664SLiu Bo goto out_drop_write; 1720b9ca0664SLiu Bo } 17210caa102dSLi Zefan 1722b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 1723b9ca0664SLiu Bo ret = -EOPNOTSUPP; 1724b9ca0664SLiu Bo goto out_drop_write; 1725b9ca0664SLiu Bo } 17260caa102dSLi Zefan 17270caa102dSLi Zefan down_write(&root->fs_info->subvol_sem); 17280caa102dSLi Zefan 17290caa102dSLi Zefan /* nothing to do */ 17300caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 1731b9ca0664SLiu Bo goto out_drop_sem; 17320caa102dSLi Zefan 17330caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 17342c686537SDavid Sterba if (flags & BTRFS_SUBVOL_RDONLY) { 17350caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 17360caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 17372c686537SDavid Sterba } else { 17382c686537SDavid Sterba /* 17392c686537SDavid Sterba * Block RO -> RW transition if this subvolume is involved in 17402c686537SDavid Sterba * send 17412c686537SDavid Sterba */ 17422c686537SDavid Sterba spin_lock(&root->root_item_lock); 17432c686537SDavid Sterba if (root->send_in_progress == 0) { 17440caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 17450caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 17462c686537SDavid Sterba spin_unlock(&root->root_item_lock); 17472c686537SDavid Sterba } else { 17482c686537SDavid Sterba spin_unlock(&root->root_item_lock); 17492c686537SDavid Sterba btrfs_warn(root->fs_info, 17502c686537SDavid Sterba "Attempt to set subvolume %llu read-write during send", 17512c686537SDavid Sterba root->root_key.objectid); 17522c686537SDavid Sterba ret = -EPERM; 17532c686537SDavid Sterba goto out_drop_sem; 17542c686537SDavid Sterba } 17552c686537SDavid Sterba } 17560caa102dSLi Zefan 17570caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 17580caa102dSLi Zefan if (IS_ERR(trans)) { 17590caa102dSLi Zefan ret = PTR_ERR(trans); 17600caa102dSLi Zefan goto out_reset; 17610caa102dSLi Zefan } 17620caa102dSLi Zefan 1763b4dc2b8cSLi Zefan ret = btrfs_update_root(trans, root->fs_info->tree_root, 17640caa102dSLi Zefan &root->root_key, &root->root_item); 17650caa102dSLi Zefan 17660caa102dSLi Zefan btrfs_commit_transaction(trans, root); 17670caa102dSLi Zefan out_reset: 17680caa102dSLi Zefan if (ret) 17690caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 1770b9ca0664SLiu Bo out_drop_sem: 17710caa102dSLi Zefan up_write(&root->fs_info->subvol_sem); 1772b9ca0664SLiu Bo out_drop_write: 1773b9ca0664SLiu Bo mnt_drop_write_file(file); 1774b9ca0664SLiu Bo out: 17750caa102dSLi Zefan return ret; 17760caa102dSLi Zefan } 17770caa102dSLi Zefan 177876dda93cSYan, Zheng /* 177976dda93cSYan, Zheng * helper to check if the subvolume references other subvolumes 178076dda93cSYan, Zheng */ 178176dda93cSYan, Zheng static noinline int may_destroy_subvol(struct btrfs_root *root) 178276dda93cSYan, Zheng { 178376dda93cSYan, Zheng struct btrfs_path *path; 1784175a2b87SJosef Bacik struct btrfs_dir_item *di; 178576dda93cSYan, Zheng struct btrfs_key key; 1786175a2b87SJosef Bacik u64 dir_id; 178776dda93cSYan, Zheng int ret; 178876dda93cSYan, Zheng 178976dda93cSYan, Zheng path = btrfs_alloc_path(); 179076dda93cSYan, Zheng if (!path) 179176dda93cSYan, Zheng return -ENOMEM; 179276dda93cSYan, Zheng 1793175a2b87SJosef Bacik /* Make sure this root isn't set as the default subvol */ 1794175a2b87SJosef Bacik dir_id = btrfs_super_root_dir(root->fs_info->super_copy); 1795175a2b87SJosef Bacik di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path, 1796175a2b87SJosef Bacik dir_id, "default", 7, 0); 1797175a2b87SJosef Bacik if (di && !IS_ERR(di)) { 1798175a2b87SJosef Bacik btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); 1799175a2b87SJosef Bacik if (key.objectid == root->root_key.objectid) { 1800175a2b87SJosef Bacik ret = -ENOTEMPTY; 1801175a2b87SJosef Bacik goto out; 1802175a2b87SJosef Bacik } 1803175a2b87SJosef Bacik btrfs_release_path(path); 1804175a2b87SJosef Bacik } 1805175a2b87SJosef Bacik 180676dda93cSYan, Zheng key.objectid = root->root_key.objectid; 180776dda93cSYan, Zheng key.type = BTRFS_ROOT_REF_KEY; 180876dda93cSYan, Zheng key.offset = (u64)-1; 180976dda93cSYan, Zheng 181076dda93cSYan, Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, 181176dda93cSYan, Zheng &key, path, 0, 0); 181276dda93cSYan, Zheng if (ret < 0) 181376dda93cSYan, Zheng goto out; 181476dda93cSYan, Zheng BUG_ON(ret == 0); 181576dda93cSYan, Zheng 181676dda93cSYan, Zheng ret = 0; 181776dda93cSYan, Zheng if (path->slots[0] > 0) { 181876dda93cSYan, Zheng path->slots[0]--; 181976dda93cSYan, Zheng btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 182076dda93cSYan, Zheng if (key.objectid == root->root_key.objectid && 182176dda93cSYan, Zheng key.type == BTRFS_ROOT_REF_KEY) 182276dda93cSYan, Zheng ret = -ENOTEMPTY; 182376dda93cSYan, Zheng } 182476dda93cSYan, Zheng out: 182576dda93cSYan, Zheng btrfs_free_path(path); 182676dda93cSYan, Zheng return ret; 182776dda93cSYan, Zheng } 182876dda93cSYan, Zheng 1829ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 1830ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 1831ac8e9819SChris Mason { 1832abc6e134SChris Mason struct btrfs_key test; 1833abc6e134SChris Mason int ret; 1834abc6e134SChris Mason 1835abc6e134SChris Mason test.objectid = sk->min_objectid; 1836abc6e134SChris Mason test.type = sk->min_type; 1837abc6e134SChris Mason test.offset = sk->min_offset; 1838abc6e134SChris Mason 1839abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1840abc6e134SChris Mason if (ret < 0) 1841ac8e9819SChris Mason return 0; 1842abc6e134SChris Mason 1843abc6e134SChris Mason test.objectid = sk->max_objectid; 1844abc6e134SChris Mason test.type = sk->max_type; 1845abc6e134SChris Mason test.offset = sk->max_offset; 1846abc6e134SChris Mason 1847abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1848abc6e134SChris Mason if (ret > 0) 1849ac8e9819SChris Mason return 0; 1850ac8e9819SChris Mason return 1; 1851ac8e9819SChris Mason } 1852ac8e9819SChris Mason 1853ac8e9819SChris Mason static noinline int copy_to_sk(struct btrfs_root *root, 1854ac8e9819SChris Mason struct btrfs_path *path, 1855ac8e9819SChris Mason struct btrfs_key *key, 1856ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 1857ac8e9819SChris Mason char *buf, 1858ac8e9819SChris Mason unsigned long *sk_offset, 1859ac8e9819SChris Mason int *num_found) 1860ac8e9819SChris Mason { 1861ac8e9819SChris Mason u64 found_transid; 1862ac8e9819SChris Mason struct extent_buffer *leaf; 1863ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 1864ac8e9819SChris Mason unsigned long item_off; 1865ac8e9819SChris Mason unsigned long item_len; 1866ac8e9819SChris Mason int nritems; 1867ac8e9819SChris Mason int i; 1868ac8e9819SChris Mason int slot; 1869ac8e9819SChris Mason int ret = 0; 1870ac8e9819SChris Mason 1871ac8e9819SChris Mason leaf = path->nodes[0]; 1872ac8e9819SChris Mason slot = path->slots[0]; 1873ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 1874ac8e9819SChris Mason 1875ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 1876ac8e9819SChris Mason i = nritems; 1877ac8e9819SChris Mason goto advance_key; 1878ac8e9819SChris Mason } 1879ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 1880ac8e9819SChris Mason 1881ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 1882ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 1883ac8e9819SChris Mason item_len = btrfs_item_size_nr(leaf, i); 1884ac8e9819SChris Mason 188503b71c6cSGabriel de Perthuis btrfs_item_key_to_cpu(leaf, key, i); 188603b71c6cSGabriel de Perthuis if (!key_in_sk(key, sk)) 188703b71c6cSGabriel de Perthuis continue; 188803b71c6cSGabriel de Perthuis 188903b71c6cSGabriel de Perthuis if (sizeof(sh) + item_len > BTRFS_SEARCH_ARGS_BUFSIZE) 1890ac8e9819SChris Mason item_len = 0; 1891ac8e9819SChris Mason 1892ac8e9819SChris Mason if (sizeof(sh) + item_len + *sk_offset > 1893ac8e9819SChris Mason BTRFS_SEARCH_ARGS_BUFSIZE) { 1894ac8e9819SChris Mason ret = 1; 1895ac8e9819SChris Mason goto overflow; 1896ac8e9819SChris Mason } 1897ac8e9819SChris Mason 1898ac8e9819SChris Mason sh.objectid = key->objectid; 1899ac8e9819SChris Mason sh.offset = key->offset; 1900ac8e9819SChris Mason sh.type = key->type; 1901ac8e9819SChris Mason sh.len = item_len; 1902ac8e9819SChris Mason sh.transid = found_transid; 1903ac8e9819SChris Mason 1904ac8e9819SChris Mason /* copy search result header */ 1905ac8e9819SChris Mason memcpy(buf + *sk_offset, &sh, sizeof(sh)); 1906ac8e9819SChris Mason *sk_offset += sizeof(sh); 1907ac8e9819SChris Mason 1908ac8e9819SChris Mason if (item_len) { 1909ac8e9819SChris Mason char *p = buf + *sk_offset; 1910ac8e9819SChris Mason /* copy the item */ 1911ac8e9819SChris Mason read_extent_buffer(leaf, p, 1912ac8e9819SChris Mason item_off, item_len); 1913ac8e9819SChris Mason *sk_offset += item_len; 1914ac8e9819SChris Mason } 1915e2156867SHugo Mills (*num_found)++; 1916ac8e9819SChris Mason 1917ac8e9819SChris Mason if (*num_found >= sk->nr_items) 1918ac8e9819SChris Mason break; 1919ac8e9819SChris Mason } 1920ac8e9819SChris Mason advance_key: 1921ac8e9819SChris Mason ret = 0; 1922abc6e134SChris Mason if (key->offset < (u64)-1 && key->offset < sk->max_offset) 1923abc6e134SChris Mason key->offset++; 1924abc6e134SChris Mason else if (key->type < (u8)-1 && key->type < sk->max_type) { 1925abc6e134SChris Mason key->offset = 0; 1926abc6e134SChris Mason key->type++; 1927abc6e134SChris Mason } else if (key->objectid < (u64)-1 && key->objectid < sk->max_objectid) { 1928abc6e134SChris Mason key->offset = 0; 1929abc6e134SChris Mason key->type = 0; 1930abc6e134SChris Mason key->objectid++; 1931abc6e134SChris Mason } else 1932abc6e134SChris Mason ret = 1; 1933ac8e9819SChris Mason overflow: 1934ac8e9819SChris Mason return ret; 1935ac8e9819SChris Mason } 1936ac8e9819SChris Mason 1937ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 1938ac8e9819SChris Mason struct btrfs_ioctl_search_args *args) 1939ac8e9819SChris Mason { 1940ac8e9819SChris Mason struct btrfs_root *root; 1941ac8e9819SChris Mason struct btrfs_key key; 1942ac8e9819SChris Mason struct btrfs_path *path; 1943ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk = &args->key; 1944ac8e9819SChris Mason struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info; 1945ac8e9819SChris Mason int ret; 1946ac8e9819SChris Mason int num_found = 0; 1947ac8e9819SChris Mason unsigned long sk_offset = 0; 1948ac8e9819SChris Mason 1949ac8e9819SChris Mason path = btrfs_alloc_path(); 1950ac8e9819SChris Mason if (!path) 1951ac8e9819SChris Mason return -ENOMEM; 1952ac8e9819SChris Mason 1953ac8e9819SChris Mason if (sk->tree_id == 0) { 1954ac8e9819SChris Mason /* search the root of the inode that was passed */ 1955ac8e9819SChris Mason root = BTRFS_I(inode)->root; 1956ac8e9819SChris Mason } else { 1957ac8e9819SChris Mason key.objectid = sk->tree_id; 1958ac8e9819SChris Mason key.type = BTRFS_ROOT_ITEM_KEY; 1959ac8e9819SChris Mason key.offset = (u64)-1; 1960ac8e9819SChris Mason root = btrfs_read_fs_root_no_name(info, &key); 1961ac8e9819SChris Mason if (IS_ERR(root)) { 1962efe120a0SFrank Holton printk(KERN_ERR "BTRFS: could not find root %llu\n", 1963ac8e9819SChris Mason sk->tree_id); 1964ac8e9819SChris Mason btrfs_free_path(path); 1965ac8e9819SChris Mason return -ENOENT; 1966ac8e9819SChris Mason } 1967ac8e9819SChris Mason } 1968ac8e9819SChris Mason 1969ac8e9819SChris Mason key.objectid = sk->min_objectid; 1970ac8e9819SChris Mason key.type = sk->min_type; 1971ac8e9819SChris Mason key.offset = sk->min_offset; 1972ac8e9819SChris Mason 1973ac8e9819SChris Mason path->keep_locks = 1; 1974ac8e9819SChris Mason 1975ac8e9819SChris Mason while (1) { 19766174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &key, path, sk->min_transid); 1977ac8e9819SChris Mason if (ret != 0) { 1978ac8e9819SChris Mason if (ret > 0) 1979ac8e9819SChris Mason ret = 0; 1980ac8e9819SChris Mason goto err; 1981ac8e9819SChris Mason } 1982ac8e9819SChris Mason ret = copy_to_sk(root, path, &key, sk, args->buf, 1983ac8e9819SChris Mason &sk_offset, &num_found); 1984b3b4aa74SDavid Sterba btrfs_release_path(path); 1985ac8e9819SChris Mason if (ret || num_found >= sk->nr_items) 1986ac8e9819SChris Mason break; 1987ac8e9819SChris Mason 1988ac8e9819SChris Mason } 1989ac8e9819SChris Mason ret = 0; 1990ac8e9819SChris Mason err: 1991ac8e9819SChris Mason sk->nr_items = num_found; 1992ac8e9819SChris Mason btrfs_free_path(path); 1993ac8e9819SChris Mason return ret; 1994ac8e9819SChris Mason } 1995ac8e9819SChris Mason 1996ac8e9819SChris Mason static noinline int btrfs_ioctl_tree_search(struct file *file, 1997ac8e9819SChris Mason void __user *argp) 1998ac8e9819SChris Mason { 1999ac8e9819SChris Mason struct btrfs_ioctl_search_args *args; 2000ac8e9819SChris Mason struct inode *inode; 2001ac8e9819SChris Mason int ret; 2002ac8e9819SChris Mason 2003ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2004ac8e9819SChris Mason return -EPERM; 2005ac8e9819SChris Mason 20062354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 20072354d08fSJulia Lawall if (IS_ERR(args)) 20082354d08fSJulia Lawall return PTR_ERR(args); 2009ac8e9819SChris Mason 2010496ad9aaSAl Viro inode = file_inode(file); 2011ac8e9819SChris Mason ret = search_ioctl(inode, args); 2012ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2013ac8e9819SChris Mason ret = -EFAULT; 2014ac8e9819SChris Mason kfree(args); 2015ac8e9819SChris Mason return ret; 2016ac8e9819SChris Mason } 2017ac8e9819SChris Mason 201898d377a0STARUISI Hiroaki /* 2019ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 2020ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 202198d377a0STARUISI Hiroaki */ 202298d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 202398d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 202498d377a0STARUISI Hiroaki { 202598d377a0STARUISI Hiroaki struct btrfs_root *root; 202698d377a0STARUISI Hiroaki struct btrfs_key key; 2027ac8e9819SChris Mason char *ptr; 202898d377a0STARUISI Hiroaki int ret = -1; 202998d377a0STARUISI Hiroaki int slot; 203098d377a0STARUISI Hiroaki int len; 203198d377a0STARUISI Hiroaki int total_len = 0; 203298d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 203398d377a0STARUISI Hiroaki struct extent_buffer *l; 203498d377a0STARUISI Hiroaki struct btrfs_path *path; 203598d377a0STARUISI Hiroaki 203698d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 203798d377a0STARUISI Hiroaki name[0]='\0'; 203898d377a0STARUISI Hiroaki return 0; 203998d377a0STARUISI Hiroaki } 204098d377a0STARUISI Hiroaki 204198d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 204298d377a0STARUISI Hiroaki if (!path) 204398d377a0STARUISI Hiroaki return -ENOMEM; 204498d377a0STARUISI Hiroaki 2045ac8e9819SChris Mason ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX]; 204698d377a0STARUISI Hiroaki 204798d377a0STARUISI Hiroaki key.objectid = tree_id; 204898d377a0STARUISI Hiroaki key.type = BTRFS_ROOT_ITEM_KEY; 204998d377a0STARUISI Hiroaki key.offset = (u64)-1; 205098d377a0STARUISI Hiroaki root = btrfs_read_fs_root_no_name(info, &key); 205198d377a0STARUISI Hiroaki if (IS_ERR(root)) { 2052efe120a0SFrank Holton printk(KERN_ERR "BTRFS: could not find root %llu\n", tree_id); 20538ad6fcabSChris Mason ret = -ENOENT; 20548ad6fcabSChris Mason goto out; 205598d377a0STARUISI Hiroaki } 205698d377a0STARUISI Hiroaki 205798d377a0STARUISI Hiroaki key.objectid = dirid; 205898d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 20598ad6fcabSChris Mason key.offset = (u64)-1; 206098d377a0STARUISI Hiroaki 206198d377a0STARUISI Hiroaki while (1) { 206298d377a0STARUISI Hiroaki ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 206398d377a0STARUISI Hiroaki if (ret < 0) 206498d377a0STARUISI Hiroaki goto out; 206518674c6cSFilipe David Borba Manana else if (ret > 0) { 206618674c6cSFilipe David Borba Manana ret = btrfs_previous_item(root, path, dirid, 206718674c6cSFilipe David Borba Manana BTRFS_INODE_REF_KEY); 206818674c6cSFilipe David Borba Manana if (ret < 0) 206918674c6cSFilipe David Borba Manana goto out; 207018674c6cSFilipe David Borba Manana else if (ret > 0) { 2071ac8e9819SChris Mason ret = -ENOENT; 207298d377a0STARUISI Hiroaki goto out; 2073ac8e9819SChris Mason } 207418674c6cSFilipe David Borba Manana } 207518674c6cSFilipe David Borba Manana 207618674c6cSFilipe David Borba Manana l = path->nodes[0]; 207718674c6cSFilipe David Borba Manana slot = path->slots[0]; 207818674c6cSFilipe David Borba Manana btrfs_item_key_to_cpu(l, &key, slot); 207998d377a0STARUISI Hiroaki 208098d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 208198d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 208298d377a0STARUISI Hiroaki ptr -= len + 1; 208398d377a0STARUISI Hiroaki total_len += len + 1; 2084a696cf35SFilipe David Borba Manana if (ptr < name) { 2085a696cf35SFilipe David Borba Manana ret = -ENAMETOOLONG; 208698d377a0STARUISI Hiroaki goto out; 2087a696cf35SFilipe David Borba Manana } 208898d377a0STARUISI Hiroaki 208998d377a0STARUISI Hiroaki *(ptr + len) = '/'; 209098d377a0STARUISI Hiroaki read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 209198d377a0STARUISI Hiroaki 209298d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 209398d377a0STARUISI Hiroaki break; 209498d377a0STARUISI Hiroaki 2095b3b4aa74SDavid Sterba btrfs_release_path(path); 209698d377a0STARUISI Hiroaki key.objectid = key.offset; 20978ad6fcabSChris Mason key.offset = (u64)-1; 209898d377a0STARUISI Hiroaki dirid = key.objectid; 209998d377a0STARUISI Hiroaki } 210077906a50SLi Zefan memmove(name, ptr, total_len); 210198d377a0STARUISI Hiroaki name[total_len] = '\0'; 210298d377a0STARUISI Hiroaki ret = 0; 210398d377a0STARUISI Hiroaki out: 210498d377a0STARUISI Hiroaki btrfs_free_path(path); 2105ac8e9819SChris Mason return ret; 2106ac8e9819SChris Mason } 2107ac8e9819SChris Mason 2108ac8e9819SChris Mason static noinline int btrfs_ioctl_ino_lookup(struct file *file, 2109ac8e9819SChris Mason void __user *argp) 2110ac8e9819SChris Mason { 2111ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 2112ac8e9819SChris Mason struct inode *inode; 2113ac8e9819SChris Mason int ret; 2114ac8e9819SChris Mason 2115ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2116ac8e9819SChris Mason return -EPERM; 2117ac8e9819SChris Mason 21182354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 21192354d08fSJulia Lawall if (IS_ERR(args)) 21202354d08fSJulia Lawall return PTR_ERR(args); 2121c2b96929SDan Carpenter 2122496ad9aaSAl Viro inode = file_inode(file); 2123ac8e9819SChris Mason 21241b53ac4dSChris Mason if (args->treeid == 0) 21251b53ac4dSChris Mason args->treeid = BTRFS_I(inode)->root->root_key.objectid; 21261b53ac4dSChris Mason 2127ac8e9819SChris Mason ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 2128ac8e9819SChris Mason args->treeid, args->objectid, 2129ac8e9819SChris Mason args->name); 2130ac8e9819SChris Mason 2131ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2132ac8e9819SChris Mason ret = -EFAULT; 2133ac8e9819SChris Mason 2134ac8e9819SChris Mason kfree(args); 213598d377a0STARUISI Hiroaki return ret; 213698d377a0STARUISI Hiroaki } 213798d377a0STARUISI Hiroaki 213876dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 213976dda93cSYan, Zheng void __user *arg) 214076dda93cSYan, Zheng { 214154563d41SAl Viro struct dentry *parent = file->f_path.dentry; 214276dda93cSYan, Zheng struct dentry *dentry; 214376dda93cSYan, Zheng struct inode *dir = parent->d_inode; 214476dda93cSYan, Zheng struct inode *inode; 214576dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 214676dda93cSYan, Zheng struct btrfs_root *dest = NULL; 214776dda93cSYan, Zheng struct btrfs_ioctl_vol_args *vol_args; 214876dda93cSYan, Zheng struct btrfs_trans_handle *trans; 2149c58aaad2SMiao Xie struct btrfs_block_rsv block_rsv; 2150c58aaad2SMiao Xie u64 qgroup_reserved; 215176dda93cSYan, Zheng int namelen; 215276dda93cSYan, Zheng int ret; 215376dda93cSYan, Zheng int err = 0; 215476dda93cSYan, Zheng 215576dda93cSYan, Zheng vol_args = memdup_user(arg, sizeof(*vol_args)); 215676dda93cSYan, Zheng if (IS_ERR(vol_args)) 215776dda93cSYan, Zheng return PTR_ERR(vol_args); 215876dda93cSYan, Zheng 215976dda93cSYan, Zheng vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 216076dda93cSYan, Zheng namelen = strlen(vol_args->name); 216176dda93cSYan, Zheng if (strchr(vol_args->name, '/') || 216276dda93cSYan, Zheng strncmp(vol_args->name, "..", namelen) == 0) { 216376dda93cSYan, Zheng err = -EINVAL; 216476dda93cSYan, Zheng goto out; 216576dda93cSYan, Zheng } 216676dda93cSYan, Zheng 2167a561be71SAl Viro err = mnt_want_write_file(file); 216876dda93cSYan, Zheng if (err) 216976dda93cSYan, Zheng goto out; 217076dda93cSYan, Zheng 21715c50c9b8SDavid Sterba err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT); 21725c50c9b8SDavid Sterba if (err == -EINTR) 2173e43f998eSDavid Sterba goto out_drop_write; 217476dda93cSYan, Zheng dentry = lookup_one_len(vol_args->name, parent, namelen); 217576dda93cSYan, Zheng if (IS_ERR(dentry)) { 217676dda93cSYan, Zheng err = PTR_ERR(dentry); 217776dda93cSYan, Zheng goto out_unlock_dir; 217876dda93cSYan, Zheng } 217976dda93cSYan, Zheng 218076dda93cSYan, Zheng if (!dentry->d_inode) { 218176dda93cSYan, Zheng err = -ENOENT; 218276dda93cSYan, Zheng goto out_dput; 218376dda93cSYan, Zheng } 218476dda93cSYan, Zheng 218576dda93cSYan, Zheng inode = dentry->d_inode; 21864260f7c7SSage Weil dest = BTRFS_I(inode)->root; 21874260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)) { 21884260f7c7SSage Weil /* 21894260f7c7SSage Weil * Regular user. Only allow this with a special mount 21904260f7c7SSage Weil * option, when the user has write+exec access to the 21914260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 21924260f7c7SSage Weil * allowed. 21934260f7c7SSage Weil * 21944260f7c7SSage Weil * Note that this is _not_ check that the subvol is 21954260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 21964260f7c7SSage Weil * otherwise be able to delete. 21974260f7c7SSage Weil * 21984260f7c7SSage Weil * Users who want to delete empty subvols should try 21994260f7c7SSage Weil * rmdir(2). 22004260f7c7SSage Weil */ 22014260f7c7SSage Weil err = -EPERM; 22024260f7c7SSage Weil if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED)) 22034260f7c7SSage Weil goto out_dput; 22044260f7c7SSage Weil 22054260f7c7SSage Weil /* 22064260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 22074260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 22084260f7c7SSage Weil * must be called on the dentry referencing the root 22094260f7c7SSage Weil * of the subvol, not a random directory contained 22104260f7c7SSage Weil * within it. 22114260f7c7SSage Weil */ 22124260f7c7SSage Weil err = -EINVAL; 22134260f7c7SSage Weil if (root == dest) 22144260f7c7SSage Weil goto out_dput; 22154260f7c7SSage Weil 22164260f7c7SSage Weil err = inode_permission(inode, MAY_WRITE | MAY_EXEC); 22174260f7c7SSage Weil if (err) 22184260f7c7SSage Weil goto out_dput; 22195c39da5bSMiao Xie } 22204260f7c7SSage Weil 22215c39da5bSMiao Xie /* check if subvolume may be deleted by a user */ 22224260f7c7SSage Weil err = btrfs_may_delete(dir, dentry, 1); 22234260f7c7SSage Weil if (err) 22244260f7c7SSage Weil goto out_dput; 22254260f7c7SSage Weil 222633345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 222776dda93cSYan, Zheng err = -EINVAL; 222876dda93cSYan, Zheng goto out_dput; 222976dda93cSYan, Zheng } 223076dda93cSYan, Zheng 223176dda93cSYan, Zheng mutex_lock(&inode->i_mutex); 223276dda93cSYan, Zheng err = d_invalidate(dentry); 223376dda93cSYan, Zheng if (err) 223476dda93cSYan, Zheng goto out_unlock; 223576dda93cSYan, Zheng 223676dda93cSYan, Zheng down_write(&root->fs_info->subvol_sem); 223776dda93cSYan, Zheng 223876dda93cSYan, Zheng err = may_destroy_subvol(dest); 223976dda93cSYan, Zheng if (err) 224076dda93cSYan, Zheng goto out_up_write; 224176dda93cSYan, Zheng 2242c58aaad2SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 2243c58aaad2SMiao Xie /* 2244c58aaad2SMiao Xie * One for dir inode, two for dir entries, two for root 2245c58aaad2SMiao Xie * ref/backref. 2246c58aaad2SMiao Xie */ 2247c58aaad2SMiao Xie err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 2248ee3441b4SJeff Mahoney 5, &qgroup_reserved, true); 2249c58aaad2SMiao Xie if (err) 2250c58aaad2SMiao Xie goto out_up_write; 2251c58aaad2SMiao Xie 2252a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 2253a22285a6SYan, Zheng if (IS_ERR(trans)) { 2254a22285a6SYan, Zheng err = PTR_ERR(trans); 2255c58aaad2SMiao Xie goto out_release; 2256a22285a6SYan, Zheng } 2257c58aaad2SMiao Xie trans->block_rsv = &block_rsv; 2258c58aaad2SMiao Xie trans->bytes_reserved = block_rsv.size; 2259a22285a6SYan, Zheng 226076dda93cSYan, Zheng ret = btrfs_unlink_subvol(trans, root, dir, 226176dda93cSYan, Zheng dest->root_key.objectid, 226276dda93cSYan, Zheng dentry->d_name.name, 226376dda93cSYan, Zheng dentry->d_name.len); 226479787eaaSJeff Mahoney if (ret) { 226579787eaaSJeff Mahoney err = ret; 226679787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 226779787eaaSJeff Mahoney goto out_end_trans; 226879787eaaSJeff Mahoney } 226976dda93cSYan, Zheng 227076dda93cSYan, Zheng btrfs_record_root_in_trans(trans, dest); 227176dda93cSYan, Zheng 227276dda93cSYan, Zheng memset(&dest->root_item.drop_progress, 0, 227376dda93cSYan, Zheng sizeof(dest->root_item.drop_progress)); 227476dda93cSYan, Zheng dest->root_item.drop_level = 0; 227576dda93cSYan, Zheng btrfs_set_root_refs(&dest->root_item, 0); 227676dda93cSYan, Zheng 2277d68fc57bSYan, Zheng if (!xchg(&dest->orphan_item_inserted, 1)) { 227876dda93cSYan, Zheng ret = btrfs_insert_orphan_item(trans, 227976dda93cSYan, Zheng root->fs_info->tree_root, 228076dda93cSYan, Zheng dest->root_key.objectid); 228179787eaaSJeff Mahoney if (ret) { 228279787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 228379787eaaSJeff Mahoney err = ret; 228479787eaaSJeff Mahoney goto out_end_trans; 2285d68fc57bSYan, Zheng } 228679787eaaSJeff Mahoney } 2287dd5f9615SStefan Behrens 2288dd5f9615SStefan Behrens ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 2289dd5f9615SStefan Behrens dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL, 2290dd5f9615SStefan Behrens dest->root_key.objectid); 2291dd5f9615SStefan Behrens if (ret && ret != -ENOENT) { 2292dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 2293dd5f9615SStefan Behrens err = ret; 2294dd5f9615SStefan Behrens goto out_end_trans; 2295dd5f9615SStefan Behrens } 2296dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { 2297dd5f9615SStefan Behrens ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 2298dd5f9615SStefan Behrens dest->root_item.received_uuid, 2299dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 2300dd5f9615SStefan Behrens dest->root_key.objectid); 2301dd5f9615SStefan Behrens if (ret && ret != -ENOENT) { 2302dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 2303dd5f9615SStefan Behrens err = ret; 2304dd5f9615SStefan Behrens goto out_end_trans; 2305dd5f9615SStefan Behrens } 2306dd5f9615SStefan Behrens } 2307dd5f9615SStefan Behrens 230879787eaaSJeff Mahoney out_end_trans: 2309c58aaad2SMiao Xie trans->block_rsv = NULL; 2310c58aaad2SMiao Xie trans->bytes_reserved = 0; 2311531cb13fSSage Weil ret = btrfs_end_transaction(trans, root); 231279787eaaSJeff Mahoney if (ret && !err) 231379787eaaSJeff Mahoney err = ret; 231476dda93cSYan, Zheng inode->i_flags |= S_DEAD; 2315c58aaad2SMiao Xie out_release: 2316c58aaad2SMiao Xie btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved); 231776dda93cSYan, Zheng out_up_write: 231876dda93cSYan, Zheng up_write(&root->fs_info->subvol_sem); 231976dda93cSYan, Zheng out_unlock: 232076dda93cSYan, Zheng mutex_unlock(&inode->i_mutex); 232176dda93cSYan, Zheng if (!err) { 2322efefb143SYan, Zheng shrink_dcache_sb(root->fs_info->sb); 232376dda93cSYan, Zheng btrfs_invalidate_inodes(dest); 232476dda93cSYan, Zheng d_delete(dentry); 2325fa6ac876SLiu Bo 2326fa6ac876SLiu Bo /* the last ref */ 2327fa6ac876SLiu Bo if (dest->cache_inode) { 2328fa6ac876SLiu Bo iput(dest->cache_inode); 2329fa6ac876SLiu Bo dest->cache_inode = NULL; 2330fa6ac876SLiu Bo } 233176dda93cSYan, Zheng } 233276dda93cSYan, Zheng out_dput: 233376dda93cSYan, Zheng dput(dentry); 233476dda93cSYan, Zheng out_unlock_dir: 233576dda93cSYan, Zheng mutex_unlock(&dir->i_mutex); 2336e43f998eSDavid Sterba out_drop_write: 23372a79f17eSAl Viro mnt_drop_write_file(file); 233876dda93cSYan, Zheng out: 233976dda93cSYan, Zheng kfree(vol_args); 234076dda93cSYan, Zheng return err; 234176dda93cSYan, Zheng } 234276dda93cSYan, Zheng 23431e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 2344f46b5a66SChristoph Hellwig { 2345496ad9aaSAl Viro struct inode *inode = file_inode(file); 2346f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 23471e701a32SChris Mason struct btrfs_ioctl_defrag_range_args *range; 2348c146afadSYan Zheng int ret; 2349c146afadSYan Zheng 235025122d15SIlya Dryomov ret = mnt_want_write_file(file); 235125122d15SIlya Dryomov if (ret) 235225122d15SIlya Dryomov return ret; 2353b83cc969SLi Zefan 235425122d15SIlya Dryomov if (btrfs_root_readonly(root)) { 235525122d15SIlya Dryomov ret = -EROFS; 235625122d15SIlya Dryomov goto out; 23575ac00addSStefan Behrens } 2358f46b5a66SChristoph Hellwig 2359f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2360f46b5a66SChristoph Hellwig case S_IFDIR: 2361e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 2362e441d54dSChris Mason ret = -EPERM; 2363e441d54dSChris Mason goto out; 2364e441d54dSChris Mason } 2365de78b51aSEric Sandeen ret = btrfs_defrag_root(root); 23668929ecfaSYan, Zheng if (ret) 23678929ecfaSYan, Zheng goto out; 2368de78b51aSEric Sandeen ret = btrfs_defrag_root(root->fs_info->extent_root); 2369f46b5a66SChristoph Hellwig break; 2370f46b5a66SChristoph Hellwig case S_IFREG: 2371e441d54dSChris Mason if (!(file->f_mode & FMODE_WRITE)) { 2372e441d54dSChris Mason ret = -EINVAL; 2373e441d54dSChris Mason goto out; 2374e441d54dSChris Mason } 23751e701a32SChris Mason 23761e701a32SChris Mason range = kzalloc(sizeof(*range), GFP_KERNEL); 23771e701a32SChris Mason if (!range) { 23781e701a32SChris Mason ret = -ENOMEM; 23791e701a32SChris Mason goto out; 23801e701a32SChris Mason } 23811e701a32SChris Mason 23821e701a32SChris Mason if (argp) { 23831e701a32SChris Mason if (copy_from_user(range, argp, 23841e701a32SChris Mason sizeof(*range))) { 23851e701a32SChris Mason ret = -EFAULT; 23861e701a32SChris Mason kfree(range); 2387683be16eSDan Carpenter goto out; 23881e701a32SChris Mason } 23891e701a32SChris Mason /* compression requires us to start the IO */ 23901e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 23911e701a32SChris Mason range->flags |= BTRFS_DEFRAG_RANGE_START_IO; 23921e701a32SChris Mason range->extent_thresh = (u32)-1; 23931e701a32SChris Mason } 23941e701a32SChris Mason } else { 23951e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 23961e701a32SChris Mason range->len = (u64)-1; 23971e701a32SChris Mason } 2398496ad9aaSAl Viro ret = btrfs_defrag_file(file_inode(file), file, 23994cb5300bSChris Mason range, 0, 0); 24004cb5300bSChris Mason if (ret > 0) 24014cb5300bSChris Mason ret = 0; 24021e701a32SChris Mason kfree(range); 2403f46b5a66SChristoph Hellwig break; 24048929ecfaSYan, Zheng default: 24058929ecfaSYan, Zheng ret = -EINVAL; 2406f46b5a66SChristoph Hellwig } 2407e441d54dSChris Mason out: 240825122d15SIlya Dryomov mnt_drop_write_file(file); 2409e441d54dSChris Mason return ret; 2410f46b5a66SChristoph Hellwig } 2411f46b5a66SChristoph Hellwig 2412b2950863SChristoph Hellwig static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg) 2413f46b5a66SChristoph Hellwig { 2414f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2415f46b5a66SChristoph Hellwig int ret; 2416f46b5a66SChristoph Hellwig 2417e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2418e441d54dSChris Mason return -EPERM; 2419e441d54dSChris Mason 24205ac00addSStefan Behrens if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 24215ac00addSStefan Behrens 1)) { 2422e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 2423c9e9f97bSIlya Dryomov } 2424c9e9f97bSIlya Dryomov 24255ac00addSStefan Behrens mutex_lock(&root->fs_info->volume_mutex); 2426dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2427c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2428c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2429c9e9f97bSIlya Dryomov goto out; 2430c9e9f97bSIlya Dryomov } 2431f46b5a66SChristoph Hellwig 24325516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2433f46b5a66SChristoph Hellwig ret = btrfs_init_new_device(root, vol_args->name); 2434f46b5a66SChristoph Hellwig 2435f46b5a66SChristoph Hellwig kfree(vol_args); 2436c9e9f97bSIlya Dryomov out: 2437c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 24385ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 2439f46b5a66SChristoph Hellwig return ret; 2440f46b5a66SChristoph Hellwig } 2441f46b5a66SChristoph Hellwig 2442da24927bSMiao Xie static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 2443f46b5a66SChristoph Hellwig { 2444496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 2445f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2446f46b5a66SChristoph Hellwig int ret; 2447f46b5a66SChristoph Hellwig 2448e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2449e441d54dSChris Mason return -EPERM; 2450e441d54dSChris Mason 2451da24927bSMiao Xie ret = mnt_want_write_file(file); 2452da24927bSMiao Xie if (ret) 2453da24927bSMiao Xie return ret; 2454c146afadSYan Zheng 2455dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2456c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2457c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2458c9e9f97bSIlya Dryomov goto out; 2459c9e9f97bSIlya Dryomov } 2460f46b5a66SChristoph Hellwig 24615516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2462f46b5a66SChristoph Hellwig 2463183860f6SAnand Jain if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 2464183860f6SAnand Jain 1)) { 2465183860f6SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 2466183860f6SAnand Jain goto out; 2467183860f6SAnand Jain } 2468183860f6SAnand Jain 2469183860f6SAnand Jain mutex_lock(&root->fs_info->volume_mutex); 2470183860f6SAnand Jain ret = btrfs_rm_device(root, vol_args->name); 2471c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 24725ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 2473183860f6SAnand Jain 2474183860f6SAnand Jain out: 2475183860f6SAnand Jain kfree(vol_args); 24764ac20c70SIlya Dryomov mnt_drop_write_file(file); 2477f46b5a66SChristoph Hellwig return ret; 2478f46b5a66SChristoph Hellwig } 2479f46b5a66SChristoph Hellwig 2480475f6387SJan Schmidt static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg) 2481475f6387SJan Schmidt { 2482027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 2483475f6387SJan Schmidt struct btrfs_device *device; 2484475f6387SJan Schmidt struct btrfs_device *next; 2485475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2486027ed2f0SLi Zefan int ret = 0; 2487475f6387SJan Schmidt 2488475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 2489475f6387SJan Schmidt return -EPERM; 2490475f6387SJan Schmidt 2491027ed2f0SLi Zefan fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL); 2492027ed2f0SLi Zefan if (!fi_args) 2493027ed2f0SLi Zefan return -ENOMEM; 2494027ed2f0SLi Zefan 2495f7171750SFilipe David Borba Manana mutex_lock(&fs_devices->device_list_mutex); 2496027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 2497027ed2f0SLi Zefan memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid)); 2498475f6387SJan Schmidt 2499475f6387SJan Schmidt list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { 2500027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 2501027ed2f0SLi Zefan fi_args->max_id = device->devid; 2502475f6387SJan Schmidt } 2503475f6387SJan Schmidt mutex_unlock(&fs_devices->device_list_mutex); 2504475f6387SJan Schmidt 2505027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 2506027ed2f0SLi Zefan ret = -EFAULT; 2507475f6387SJan Schmidt 2508027ed2f0SLi Zefan kfree(fi_args); 2509027ed2f0SLi Zefan return ret; 2510475f6387SJan Schmidt } 2511475f6387SJan Schmidt 2512475f6387SJan Schmidt static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg) 2513475f6387SJan Schmidt { 2514475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 2515475f6387SJan Schmidt struct btrfs_device *dev; 2516475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2517475f6387SJan Schmidt int ret = 0; 2518475f6387SJan Schmidt char *s_uuid = NULL; 2519475f6387SJan Schmidt 2520475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 2521475f6387SJan Schmidt return -EPERM; 2522475f6387SJan Schmidt 2523475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 2524475f6387SJan Schmidt if (IS_ERR(di_args)) 2525475f6387SJan Schmidt return PTR_ERR(di_args); 2526475f6387SJan Schmidt 2527dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(di_args->uuid)) 2528475f6387SJan Schmidt s_uuid = di_args->uuid; 2529475f6387SJan Schmidt 2530475f6387SJan Schmidt mutex_lock(&fs_devices->device_list_mutex); 2531aa1b8cd4SStefan Behrens dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL); 2532475f6387SJan Schmidt 2533475f6387SJan Schmidt if (!dev) { 2534475f6387SJan Schmidt ret = -ENODEV; 2535475f6387SJan Schmidt goto out; 2536475f6387SJan Schmidt } 2537475f6387SJan Schmidt 2538475f6387SJan Schmidt di_args->devid = dev->devid; 2539475f6387SJan Schmidt di_args->bytes_used = dev->bytes_used; 2540475f6387SJan Schmidt di_args->total_bytes = dev->total_bytes; 2541475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 2542a27202fbSJim Meyering if (dev->name) { 2543606686eeSJosef Bacik struct rcu_string *name; 2544606686eeSJosef Bacik 2545606686eeSJosef Bacik rcu_read_lock(); 2546606686eeSJosef Bacik name = rcu_dereference(dev->name); 2547606686eeSJosef Bacik strncpy(di_args->path, name->str, sizeof(di_args->path)); 2548606686eeSJosef Bacik rcu_read_unlock(); 2549a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 2550a27202fbSJim Meyering } else { 255199ba55adSStefan Behrens di_args->path[0] = '\0'; 2552a27202fbSJim Meyering } 2553475f6387SJan Schmidt 2554475f6387SJan Schmidt out: 255555793c0dSDavid Sterba mutex_unlock(&fs_devices->device_list_mutex); 2556475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 2557475f6387SJan Schmidt ret = -EFAULT; 2558475f6387SJan Schmidt 2559475f6387SJan Schmidt kfree(di_args); 2560475f6387SJan Schmidt return ret; 2561475f6387SJan Schmidt } 2562475f6387SJan Schmidt 2563416161dbSMark Fasheh static struct page *extent_same_get_page(struct inode *inode, u64 off) 2564416161dbSMark Fasheh { 2565416161dbSMark Fasheh struct page *page; 2566416161dbSMark Fasheh pgoff_t index; 2567416161dbSMark Fasheh struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; 2568416161dbSMark Fasheh 2569416161dbSMark Fasheh index = off >> PAGE_CACHE_SHIFT; 2570416161dbSMark Fasheh 2571416161dbSMark Fasheh page = grab_cache_page(inode->i_mapping, index); 2572416161dbSMark Fasheh if (!page) 2573416161dbSMark Fasheh return NULL; 2574416161dbSMark Fasheh 2575416161dbSMark Fasheh if (!PageUptodate(page)) { 2576416161dbSMark Fasheh if (extent_read_full_page_nolock(tree, page, btrfs_get_extent, 2577416161dbSMark Fasheh 0)) 2578416161dbSMark Fasheh return NULL; 2579416161dbSMark Fasheh lock_page(page); 2580416161dbSMark Fasheh if (!PageUptodate(page)) { 2581416161dbSMark Fasheh unlock_page(page); 2582416161dbSMark Fasheh page_cache_release(page); 2583416161dbSMark Fasheh return NULL; 2584416161dbSMark Fasheh } 2585416161dbSMark Fasheh } 2586416161dbSMark Fasheh unlock_page(page); 2587416161dbSMark Fasheh 2588416161dbSMark Fasheh return page; 2589416161dbSMark Fasheh } 2590416161dbSMark Fasheh 259177fe20dcSMark Fasheh static inline void lock_extent_range(struct inode *inode, u64 off, u64 len) 259277fe20dcSMark Fasheh { 259377fe20dcSMark Fasheh /* do any pending delalloc/csum calc on src, one way or 259477fe20dcSMark Fasheh another, and lock file content */ 259577fe20dcSMark Fasheh while (1) { 259677fe20dcSMark Fasheh struct btrfs_ordered_extent *ordered; 259777fe20dcSMark Fasheh lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 259877fe20dcSMark Fasheh ordered = btrfs_lookup_first_ordered_extent(inode, 259977fe20dcSMark Fasheh off + len - 1); 260077fe20dcSMark Fasheh if (!ordered && 260177fe20dcSMark Fasheh !test_range_bit(&BTRFS_I(inode)->io_tree, off, 260277fe20dcSMark Fasheh off + len - 1, EXTENT_DELALLOC, 0, NULL)) 260377fe20dcSMark Fasheh break; 260477fe20dcSMark Fasheh unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 260577fe20dcSMark Fasheh if (ordered) 260677fe20dcSMark Fasheh btrfs_put_ordered_extent(ordered); 260777fe20dcSMark Fasheh btrfs_wait_ordered_range(inode, off, len); 260877fe20dcSMark Fasheh } 260977fe20dcSMark Fasheh } 261077fe20dcSMark Fasheh 2611416161dbSMark Fasheh static void btrfs_double_unlock(struct inode *inode1, u64 loff1, 2612416161dbSMark Fasheh struct inode *inode2, u64 loff2, u64 len) 2613416161dbSMark Fasheh { 2614416161dbSMark Fasheh unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 2615416161dbSMark Fasheh unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 2616416161dbSMark Fasheh 2617416161dbSMark Fasheh mutex_unlock(&inode1->i_mutex); 2618416161dbSMark Fasheh mutex_unlock(&inode2->i_mutex); 2619416161dbSMark Fasheh } 2620416161dbSMark Fasheh 2621416161dbSMark Fasheh static void btrfs_double_lock(struct inode *inode1, u64 loff1, 2622416161dbSMark Fasheh struct inode *inode2, u64 loff2, u64 len) 2623416161dbSMark Fasheh { 2624416161dbSMark Fasheh if (inode1 < inode2) { 2625416161dbSMark Fasheh swap(inode1, inode2); 2626416161dbSMark Fasheh swap(loff1, loff2); 2627416161dbSMark Fasheh } 2628416161dbSMark Fasheh 2629416161dbSMark Fasheh mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT); 2630416161dbSMark Fasheh lock_extent_range(inode1, loff1, len); 2631416161dbSMark Fasheh if (inode1 != inode2) { 2632416161dbSMark Fasheh mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD); 2633416161dbSMark Fasheh lock_extent_range(inode2, loff2, len); 2634416161dbSMark Fasheh } 2635416161dbSMark Fasheh } 2636416161dbSMark Fasheh 2637416161dbSMark Fasheh static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst, 2638416161dbSMark Fasheh u64 dst_loff, u64 len) 2639416161dbSMark Fasheh { 2640416161dbSMark Fasheh int ret = 0; 2641416161dbSMark Fasheh struct page *src_page, *dst_page; 2642416161dbSMark Fasheh unsigned int cmp_len = PAGE_CACHE_SIZE; 2643416161dbSMark Fasheh void *addr, *dst_addr; 2644416161dbSMark Fasheh 2645416161dbSMark Fasheh while (len) { 2646416161dbSMark Fasheh if (len < PAGE_CACHE_SIZE) 2647416161dbSMark Fasheh cmp_len = len; 2648416161dbSMark Fasheh 2649416161dbSMark Fasheh src_page = extent_same_get_page(src, loff); 2650416161dbSMark Fasheh if (!src_page) 2651416161dbSMark Fasheh return -EINVAL; 2652416161dbSMark Fasheh dst_page = extent_same_get_page(dst, dst_loff); 2653416161dbSMark Fasheh if (!dst_page) { 2654416161dbSMark Fasheh page_cache_release(src_page); 2655416161dbSMark Fasheh return -EINVAL; 2656416161dbSMark Fasheh } 2657416161dbSMark Fasheh addr = kmap_atomic(src_page); 2658416161dbSMark Fasheh dst_addr = kmap_atomic(dst_page); 2659416161dbSMark Fasheh 2660416161dbSMark Fasheh flush_dcache_page(src_page); 2661416161dbSMark Fasheh flush_dcache_page(dst_page); 2662416161dbSMark Fasheh 2663416161dbSMark Fasheh if (memcmp(addr, dst_addr, cmp_len)) 2664416161dbSMark Fasheh ret = BTRFS_SAME_DATA_DIFFERS; 2665416161dbSMark Fasheh 2666416161dbSMark Fasheh kunmap_atomic(addr); 2667416161dbSMark Fasheh kunmap_atomic(dst_addr); 2668416161dbSMark Fasheh page_cache_release(src_page); 2669416161dbSMark Fasheh page_cache_release(dst_page); 2670416161dbSMark Fasheh 2671416161dbSMark Fasheh if (ret) 2672416161dbSMark Fasheh break; 2673416161dbSMark Fasheh 2674416161dbSMark Fasheh loff += cmp_len; 2675416161dbSMark Fasheh dst_loff += cmp_len; 2676416161dbSMark Fasheh len -= cmp_len; 2677416161dbSMark Fasheh } 2678416161dbSMark Fasheh 2679416161dbSMark Fasheh return ret; 2680416161dbSMark Fasheh } 2681416161dbSMark Fasheh 2682416161dbSMark Fasheh static int extent_same_check_offsets(struct inode *inode, u64 off, u64 len) 2683416161dbSMark Fasheh { 2684416161dbSMark Fasheh u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize; 2685416161dbSMark Fasheh 2686416161dbSMark Fasheh if (off + len > inode->i_size || off + len < off) 2687416161dbSMark Fasheh return -EINVAL; 2688416161dbSMark Fasheh /* Check that we are block aligned - btrfs_clone() requires this */ 2689416161dbSMark Fasheh if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs)) 2690416161dbSMark Fasheh return -EINVAL; 2691416161dbSMark Fasheh 2692416161dbSMark Fasheh return 0; 2693416161dbSMark Fasheh } 2694416161dbSMark Fasheh 2695416161dbSMark Fasheh static int btrfs_extent_same(struct inode *src, u64 loff, u64 len, 2696416161dbSMark Fasheh struct inode *dst, u64 dst_loff) 2697416161dbSMark Fasheh { 2698416161dbSMark Fasheh int ret; 2699416161dbSMark Fasheh 2700416161dbSMark Fasheh /* 2701416161dbSMark Fasheh * btrfs_clone() can't handle extents in the same file 2702416161dbSMark Fasheh * yet. Once that works, we can drop this check and replace it 2703416161dbSMark Fasheh * with a check for the same inode, but overlapping extents. 2704416161dbSMark Fasheh */ 2705416161dbSMark Fasheh if (src == dst) 2706416161dbSMark Fasheh return -EINVAL; 2707416161dbSMark Fasheh 2708416161dbSMark Fasheh btrfs_double_lock(src, loff, dst, dst_loff, len); 2709416161dbSMark Fasheh 2710416161dbSMark Fasheh ret = extent_same_check_offsets(src, loff, len); 2711416161dbSMark Fasheh if (ret) 2712416161dbSMark Fasheh goto out_unlock; 2713416161dbSMark Fasheh 2714416161dbSMark Fasheh ret = extent_same_check_offsets(dst, dst_loff, len); 2715416161dbSMark Fasheh if (ret) 2716416161dbSMark Fasheh goto out_unlock; 2717416161dbSMark Fasheh 2718416161dbSMark Fasheh /* don't make the dst file partly checksummed */ 2719416161dbSMark Fasheh if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 2720416161dbSMark Fasheh (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) { 2721416161dbSMark Fasheh ret = -EINVAL; 2722416161dbSMark Fasheh goto out_unlock; 2723416161dbSMark Fasheh } 2724416161dbSMark Fasheh 2725416161dbSMark Fasheh ret = btrfs_cmp_data(src, loff, dst, dst_loff, len); 2726416161dbSMark Fasheh if (ret == 0) 2727416161dbSMark Fasheh ret = btrfs_clone(src, dst, loff, len, len, dst_loff); 2728416161dbSMark Fasheh 2729416161dbSMark Fasheh out_unlock: 2730416161dbSMark Fasheh btrfs_double_unlock(src, loff, dst, dst_loff, len); 2731416161dbSMark Fasheh 2732416161dbSMark Fasheh return ret; 2733416161dbSMark Fasheh } 2734416161dbSMark Fasheh 2735416161dbSMark Fasheh #define BTRFS_MAX_DEDUPE_LEN (16 * 1024 * 1024) 2736416161dbSMark Fasheh 2737416161dbSMark Fasheh static long btrfs_ioctl_file_extent_same(struct file *file, 2738416161dbSMark Fasheh void __user *argp) 2739416161dbSMark Fasheh { 2740cbf8b8caSMark Fasheh struct btrfs_ioctl_same_args tmp; 2741cbf8b8caSMark Fasheh struct btrfs_ioctl_same_args *same; 2742cbf8b8caSMark Fasheh struct btrfs_ioctl_same_extent_info *info; 2743416161dbSMark Fasheh struct inode *src = file->f_dentry->d_inode; 2744416161dbSMark Fasheh struct file *dst_file = NULL; 2745416161dbSMark Fasheh struct inode *dst; 2746416161dbSMark Fasheh u64 off; 2747416161dbSMark Fasheh u64 len; 2748416161dbSMark Fasheh int i; 2749416161dbSMark Fasheh int ret; 2750cbf8b8caSMark Fasheh unsigned long size; 2751416161dbSMark Fasheh u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize; 2752416161dbSMark Fasheh bool is_admin = capable(CAP_SYS_ADMIN); 2753416161dbSMark Fasheh 2754416161dbSMark Fasheh if (!(file->f_mode & FMODE_READ)) 2755416161dbSMark Fasheh return -EINVAL; 2756416161dbSMark Fasheh 2757416161dbSMark Fasheh ret = mnt_want_write_file(file); 2758416161dbSMark Fasheh if (ret) 2759416161dbSMark Fasheh return ret; 2760416161dbSMark Fasheh 2761cbf8b8caSMark Fasheh if (copy_from_user(&tmp, 2762416161dbSMark Fasheh (struct btrfs_ioctl_same_args __user *)argp, 2763cbf8b8caSMark Fasheh sizeof(tmp))) { 2764416161dbSMark Fasheh ret = -EFAULT; 2765416161dbSMark Fasheh goto out; 2766416161dbSMark Fasheh } 2767416161dbSMark Fasheh 2768cbf8b8caSMark Fasheh size = sizeof(tmp) + 2769cbf8b8caSMark Fasheh tmp.dest_count * sizeof(struct btrfs_ioctl_same_extent_info); 2770cbf8b8caSMark Fasheh 2771229eed43SGeyslan G. Bem same = memdup_user((struct btrfs_ioctl_same_args __user *)argp, size); 2772cbf8b8caSMark Fasheh 2773229eed43SGeyslan G. Bem if (IS_ERR(same)) { 2774229eed43SGeyslan G. Bem ret = PTR_ERR(same); 2775cbf8b8caSMark Fasheh goto out; 2776cbf8b8caSMark Fasheh } 2777cbf8b8caSMark Fasheh 2778cbf8b8caSMark Fasheh off = same->logical_offset; 2779cbf8b8caSMark Fasheh len = same->length; 2780416161dbSMark Fasheh 2781416161dbSMark Fasheh /* 2782416161dbSMark Fasheh * Limit the total length we will dedupe for each operation. 2783416161dbSMark Fasheh * This is intended to bound the total time spent in this 2784416161dbSMark Fasheh * ioctl to something sane. 2785416161dbSMark Fasheh */ 2786416161dbSMark Fasheh if (len > BTRFS_MAX_DEDUPE_LEN) 2787416161dbSMark Fasheh len = BTRFS_MAX_DEDUPE_LEN; 2788416161dbSMark Fasheh 2789416161dbSMark Fasheh if (WARN_ON_ONCE(bs < PAGE_CACHE_SIZE)) { 2790416161dbSMark Fasheh /* 2791416161dbSMark Fasheh * Btrfs does not support blocksize < page_size. As a 2792416161dbSMark Fasheh * result, btrfs_cmp_data() won't correctly handle 2793416161dbSMark Fasheh * this situation without an update. 2794416161dbSMark Fasheh */ 2795416161dbSMark Fasheh ret = -EINVAL; 2796416161dbSMark Fasheh goto out; 2797416161dbSMark Fasheh } 2798416161dbSMark Fasheh 2799416161dbSMark Fasheh ret = -EISDIR; 2800416161dbSMark Fasheh if (S_ISDIR(src->i_mode)) 2801416161dbSMark Fasheh goto out; 2802416161dbSMark Fasheh 2803416161dbSMark Fasheh ret = -EACCES; 2804416161dbSMark Fasheh if (!S_ISREG(src->i_mode)) 2805416161dbSMark Fasheh goto out; 2806416161dbSMark Fasheh 2807cbf8b8caSMark Fasheh /* pre-format output fields to sane values */ 2808cbf8b8caSMark Fasheh for (i = 0; i < same->dest_count; i++) { 2809cbf8b8caSMark Fasheh same->info[i].bytes_deduped = 0ULL; 2810cbf8b8caSMark Fasheh same->info[i].status = 0; 2811416161dbSMark Fasheh } 2812416161dbSMark Fasheh 2813cbf8b8caSMark Fasheh ret = 0; 2814cbf8b8caSMark Fasheh for (i = 0; i < same->dest_count; i++) { 2815cbf8b8caSMark Fasheh info = &same->info[i]; 2816416161dbSMark Fasheh 2817cbf8b8caSMark Fasheh dst_file = fget(info->fd); 2818416161dbSMark Fasheh if (!dst_file) { 2819cbf8b8caSMark Fasheh info->status = -EBADF; 2820416161dbSMark Fasheh goto next; 2821416161dbSMark Fasheh } 2822416161dbSMark Fasheh 2823416161dbSMark Fasheh if (!(is_admin || (dst_file->f_mode & FMODE_WRITE))) { 2824cbf8b8caSMark Fasheh info->status = -EINVAL; 2825416161dbSMark Fasheh goto next; 2826416161dbSMark Fasheh } 2827416161dbSMark Fasheh 2828cbf8b8caSMark Fasheh info->status = -EXDEV; 2829416161dbSMark Fasheh if (file->f_path.mnt != dst_file->f_path.mnt) 2830416161dbSMark Fasheh goto next; 2831416161dbSMark Fasheh 2832416161dbSMark Fasheh dst = dst_file->f_dentry->d_inode; 2833416161dbSMark Fasheh if (src->i_sb != dst->i_sb) 2834416161dbSMark Fasheh goto next; 2835416161dbSMark Fasheh 2836416161dbSMark Fasheh if (S_ISDIR(dst->i_mode)) { 2837cbf8b8caSMark Fasheh info->status = -EISDIR; 2838416161dbSMark Fasheh goto next; 2839416161dbSMark Fasheh } 2840416161dbSMark Fasheh 2841416161dbSMark Fasheh if (!S_ISREG(dst->i_mode)) { 2842cbf8b8caSMark Fasheh info->status = -EACCES; 2843416161dbSMark Fasheh goto next; 2844416161dbSMark Fasheh } 2845416161dbSMark Fasheh 2846cbf8b8caSMark Fasheh info->status = btrfs_extent_same(src, off, len, dst, 2847cbf8b8caSMark Fasheh info->logical_offset); 2848cbf8b8caSMark Fasheh if (info->status == 0) 2849cbf8b8caSMark Fasheh info->bytes_deduped += len; 2850416161dbSMark Fasheh 2851416161dbSMark Fasheh next: 2852416161dbSMark Fasheh if (dst_file) 2853416161dbSMark Fasheh fput(dst_file); 2854cbf8b8caSMark Fasheh } 2855416161dbSMark Fasheh 2856cbf8b8caSMark Fasheh ret = copy_to_user(argp, same, size); 2857cbf8b8caSMark Fasheh if (ret) 2858416161dbSMark Fasheh ret = -EFAULT; 2859416161dbSMark Fasheh 2860416161dbSMark Fasheh out: 2861416161dbSMark Fasheh mnt_drop_write_file(file); 2862416161dbSMark Fasheh return ret; 2863416161dbSMark Fasheh } 2864416161dbSMark Fasheh 286532b7c687SMark Fasheh /** 286632b7c687SMark Fasheh * btrfs_clone() - clone a range from inode file to another 286732b7c687SMark Fasheh * 286832b7c687SMark Fasheh * @src: Inode to clone from 286932b7c687SMark Fasheh * @inode: Inode to clone to 287032b7c687SMark Fasheh * @off: Offset within source to start clone from 287132b7c687SMark Fasheh * @olen: Original length, passed by user, of range to clone 287232b7c687SMark Fasheh * @olen_aligned: Block-aligned value of olen, extent_same uses 287332b7c687SMark Fasheh * identical values here 287432b7c687SMark Fasheh * @destoff: Offset within @inode to start clone 287532b7c687SMark Fasheh */ 287632b7c687SMark Fasheh static int btrfs_clone(struct inode *src, struct inode *inode, 287732b7c687SMark Fasheh u64 off, u64 olen, u64 olen_aligned, u64 destoff) 2878f46b5a66SChristoph Hellwig { 2879f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 288032b7c687SMark Fasheh struct btrfs_path *path = NULL; 2881f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 288232b7c687SMark Fasheh struct btrfs_trans_handle *trans; 288332b7c687SMark Fasheh char *buf = NULL; 2884ae01a0abSYan Zheng struct btrfs_key key; 2885f46b5a66SChristoph Hellwig u32 nritems; 2886f46b5a66SChristoph Hellwig int slot; 2887ae01a0abSYan Zheng int ret; 288832b7c687SMark Fasheh u64 len = olen_aligned; 2889ae01a0abSYan Zheng 2890ae01a0abSYan Zheng ret = -ENOMEM; 2891ae01a0abSYan Zheng buf = vmalloc(btrfs_level_size(root, 0)); 2892ae01a0abSYan Zheng if (!buf) 289332b7c687SMark Fasheh return ret; 2894ae01a0abSYan Zheng 2895ae01a0abSYan Zheng path = btrfs_alloc_path(); 2896ae01a0abSYan Zheng if (!path) { 2897ae01a0abSYan Zheng vfree(buf); 289832b7c687SMark Fasheh return ret; 2899ae01a0abSYan Zheng } 290032b7c687SMark Fasheh 2901ae01a0abSYan Zheng path->reada = 2; 2902c5c9cd4dSSage Weil /* clone data */ 290333345d01SLi Zefan key.objectid = btrfs_ino(src); 2904ae01a0abSYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 2905ae01a0abSYan Zheng key.offset = 0; 2906f46b5a66SChristoph Hellwig 2907f46b5a66SChristoph Hellwig while (1) { 2908f46b5a66SChristoph Hellwig /* 2909f46b5a66SChristoph Hellwig * note the key will change type as we walk through the 2910f46b5a66SChristoph Hellwig * tree. 2911f46b5a66SChristoph Hellwig */ 2912e4355f34SFilipe David Borba Manana path->leave_spinning = 1; 2913362a20c5SDavid Sterba ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 2914362a20c5SDavid Sterba 0, 0); 2915f46b5a66SChristoph Hellwig if (ret < 0) 2916f46b5a66SChristoph Hellwig goto out; 2917f46b5a66SChristoph Hellwig 2918ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 2919e4355f34SFilipe David Borba Manana process_slot: 2920ae01a0abSYan Zheng if (path->slots[0] >= nritems) { 2921362a20c5SDavid Sterba ret = btrfs_next_leaf(BTRFS_I(src)->root, path); 2922f46b5a66SChristoph Hellwig if (ret < 0) 2923f46b5a66SChristoph Hellwig goto out; 2924f46b5a66SChristoph Hellwig if (ret > 0) 2925f46b5a66SChristoph Hellwig break; 2926ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 2927f46b5a66SChristoph Hellwig } 2928f46b5a66SChristoph Hellwig leaf = path->nodes[0]; 2929f46b5a66SChristoph Hellwig slot = path->slots[0]; 2930f46b5a66SChristoph Hellwig 2931ae01a0abSYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 2932d20f7043SChris Mason if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY || 293333345d01SLi Zefan key.objectid != btrfs_ino(src)) 2934f46b5a66SChristoph Hellwig break; 2935f46b5a66SChristoph Hellwig 2936c5c9cd4dSSage Weil if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) { 2937c5c9cd4dSSage Weil struct btrfs_file_extent_item *extent; 2938c5c9cd4dSSage Weil int type; 293931840ae1SZheng Yan u32 size; 294031840ae1SZheng Yan struct btrfs_key new_key; 2941c5c9cd4dSSage Weil u64 disko = 0, diskl = 0; 2942c5c9cd4dSSage Weil u64 datao = 0, datal = 0; 2943c5c9cd4dSSage Weil u8 comp; 2944b5384d48SSage Weil u64 endoff; 294531840ae1SZheng Yan 2946c5c9cd4dSSage Weil extent = btrfs_item_ptr(leaf, slot, 2947c5c9cd4dSSage Weil struct btrfs_file_extent_item); 2948c5c9cd4dSSage Weil comp = btrfs_file_extent_compression(leaf, extent); 2949c5c9cd4dSSage Weil type = btrfs_file_extent_type(leaf, extent); 2950c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 2951c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 2952d397712bSChris Mason disko = btrfs_file_extent_disk_bytenr(leaf, 2953d397712bSChris Mason extent); 2954d397712bSChris Mason diskl = btrfs_file_extent_disk_num_bytes(leaf, 2955d397712bSChris Mason extent); 2956c5c9cd4dSSage Weil datao = btrfs_file_extent_offset(leaf, extent); 2957d397712bSChris Mason datal = btrfs_file_extent_num_bytes(leaf, 2958d397712bSChris Mason extent); 2959c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 2960c5c9cd4dSSage Weil /* take upper bound, may be compressed */ 2961c5c9cd4dSSage Weil datal = btrfs_file_extent_ram_bytes(leaf, 2962c5c9cd4dSSage Weil extent); 2963c5c9cd4dSSage Weil } 296431840ae1SZheng Yan 2965050006a7SSage Weil if (key.offset + datal <= off || 2966e4355f34SFilipe David Borba Manana key.offset >= off + len - 1) { 2967e4355f34SFilipe David Borba Manana path->slots[0]++; 2968e4355f34SFilipe David Borba Manana goto process_slot; 2969e4355f34SFilipe David Borba Manana } 2970e4355f34SFilipe David Borba Manana 2971e4355f34SFilipe David Borba Manana size = btrfs_item_size_nr(leaf, slot); 2972e4355f34SFilipe David Borba Manana read_extent_buffer(leaf, buf, 2973e4355f34SFilipe David Borba Manana btrfs_item_ptr_offset(leaf, slot), 2974e4355f34SFilipe David Borba Manana size); 2975e4355f34SFilipe David Borba Manana 2976e4355f34SFilipe David Borba Manana btrfs_release_path(path); 2977e4355f34SFilipe David Borba Manana path->leave_spinning = 0; 2978c5c9cd4dSSage Weil 297931840ae1SZheng Yan memcpy(&new_key, &key, sizeof(new_key)); 298033345d01SLi Zefan new_key.objectid = btrfs_ino(inode); 29814d728ec7SLi Zefan if (off <= key.offset) 2982c5c9cd4dSSage Weil new_key.offset = key.offset + destoff - off; 29834d728ec7SLi Zefan else 29844d728ec7SLi Zefan new_key.offset = destoff; 2985c5c9cd4dSSage Weil 2986b6f3409bSSage Weil /* 2987b6f3409bSSage Weil * 1 - adjusting old extent (we may have to split it) 2988b6f3409bSSage Weil * 1 - add new extent 2989b6f3409bSSage Weil * 1 - inode update 2990b6f3409bSSage Weil */ 2991b6f3409bSSage Weil trans = btrfs_start_transaction(root, 3); 2992a22285a6SYan, Zheng if (IS_ERR(trans)) { 2993a22285a6SYan, Zheng ret = PTR_ERR(trans); 2994a22285a6SYan, Zheng goto out; 2995a22285a6SYan, Zheng } 2996a22285a6SYan, Zheng 2997c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 2998c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 2999d72c0842SLi Zefan /* 3000d72c0842SLi Zefan * a | --- range to clone ---| b 3001d72c0842SLi Zefan * | ------------- extent ------------- | 3002d72c0842SLi Zefan */ 3003d72c0842SLi Zefan 3004d72c0842SLi Zefan /* substract range b */ 3005d72c0842SLi Zefan if (key.offset + datal > off + len) 3006d72c0842SLi Zefan datal = off + len - key.offset; 3007d72c0842SLi Zefan 3008d72c0842SLi Zefan /* substract range a */ 3009a22285a6SYan, Zheng if (off > key.offset) { 3010a22285a6SYan, Zheng datao += off - key.offset; 3011a22285a6SYan, Zheng datal -= off - key.offset; 3012a22285a6SYan, Zheng } 3013a22285a6SYan, Zheng 30145dc562c5SJosef Bacik ret = btrfs_drop_extents(trans, root, inode, 3015a22285a6SYan, Zheng new_key.offset, 3016a22285a6SYan, Zheng new_key.offset + datal, 30172671485dSJosef Bacik 1); 301879787eaaSJeff Mahoney if (ret) { 301979787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 302079787eaaSJeff Mahoney ret); 302179787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 302279787eaaSJeff Mahoney goto out; 302379787eaaSJeff Mahoney } 3024a22285a6SYan, Zheng 302531840ae1SZheng Yan ret = btrfs_insert_empty_item(trans, root, path, 302631840ae1SZheng Yan &new_key, size); 302779787eaaSJeff Mahoney if (ret) { 302879787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 302979787eaaSJeff Mahoney ret); 303079787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 303179787eaaSJeff Mahoney goto out; 303279787eaaSJeff Mahoney } 303331840ae1SZheng Yan 303431840ae1SZheng Yan leaf = path->nodes[0]; 303531840ae1SZheng Yan slot = path->slots[0]; 303631840ae1SZheng Yan write_extent_buffer(leaf, buf, 303731840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 303831840ae1SZheng Yan size); 3039ae01a0abSYan Zheng 3040f46b5a66SChristoph Hellwig extent = btrfs_item_ptr(leaf, slot, 3041f46b5a66SChristoph Hellwig struct btrfs_file_extent_item); 3042c5c9cd4dSSage Weil 3043c5c9cd4dSSage Weil /* disko == 0 means it's a hole */ 3044c5c9cd4dSSage Weil if (!disko) 3045c5c9cd4dSSage Weil datao = 0; 3046c5c9cd4dSSage Weil 3047c5c9cd4dSSage Weil btrfs_set_file_extent_offset(leaf, extent, 3048c5c9cd4dSSage Weil datao); 3049c5c9cd4dSSage Weil btrfs_set_file_extent_num_bytes(leaf, extent, 3050c5c9cd4dSSage Weil datal); 3051c5c9cd4dSSage Weil if (disko) { 3052c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 3053ae01a0abSYan Zheng ret = btrfs_inc_extent_ref(trans, root, 30545d4f98a2SYan Zheng disko, diskl, 0, 3055f46b5a66SChristoph Hellwig root->root_key.objectid, 305633345d01SLi Zefan btrfs_ino(inode), 305766d7e7f0SArne Jansen new_key.offset - datao, 305866d7e7f0SArne Jansen 0); 305979787eaaSJeff Mahoney if (ret) { 306079787eaaSJeff Mahoney btrfs_abort_transaction(trans, 306179787eaaSJeff Mahoney root, 306279787eaaSJeff Mahoney ret); 306379787eaaSJeff Mahoney btrfs_end_transaction(trans, 306479787eaaSJeff Mahoney root); 306579787eaaSJeff Mahoney goto out; 306679787eaaSJeff Mahoney 306779787eaaSJeff Mahoney } 3068f46b5a66SChristoph Hellwig } 3069c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 3070c5c9cd4dSSage Weil u64 skip = 0; 3071c5c9cd4dSSage Weil u64 trim = 0; 3072c5c9cd4dSSage Weil if (off > key.offset) { 3073c5c9cd4dSSage Weil skip = off - key.offset; 3074c5c9cd4dSSage Weil new_key.offset += skip; 307531840ae1SZheng Yan } 3076d397712bSChris Mason 3077c5c9cd4dSSage Weil if (key.offset + datal > off + len) 3078c5c9cd4dSSage Weil trim = key.offset + datal - (off + len); 3079d397712bSChris Mason 3080c5c9cd4dSSage Weil if (comp && (skip || trim)) { 3081c5c9cd4dSSage Weil ret = -EINVAL; 3082a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 3083c5c9cd4dSSage Weil goto out; 308431840ae1SZheng Yan } 3085c5c9cd4dSSage Weil size -= skip + trim; 3086c5c9cd4dSSage Weil datal -= skip + trim; 3087a22285a6SYan, Zheng 30885dc562c5SJosef Bacik ret = btrfs_drop_extents(trans, root, inode, 3089a22285a6SYan, Zheng new_key.offset, 3090a22285a6SYan, Zheng new_key.offset + datal, 30912671485dSJosef Bacik 1); 309279787eaaSJeff Mahoney if (ret) { 309379787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 309479787eaaSJeff Mahoney ret); 309579787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 309679787eaaSJeff Mahoney goto out; 309779787eaaSJeff Mahoney } 3098a22285a6SYan, Zheng 3099c5c9cd4dSSage Weil ret = btrfs_insert_empty_item(trans, root, path, 3100c5c9cd4dSSage Weil &new_key, size); 310179787eaaSJeff Mahoney if (ret) { 310279787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 310379787eaaSJeff Mahoney ret); 310479787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 310579787eaaSJeff Mahoney goto out; 310679787eaaSJeff Mahoney } 3107c5c9cd4dSSage Weil 3108c5c9cd4dSSage Weil if (skip) { 3109d397712bSChris Mason u32 start = 3110d397712bSChris Mason btrfs_file_extent_calc_inline_size(0); 3111c5c9cd4dSSage Weil memmove(buf+start, buf+start+skip, 3112c5c9cd4dSSage Weil datal); 3113c5c9cd4dSSage Weil } 3114c5c9cd4dSSage Weil 3115c5c9cd4dSSage Weil leaf = path->nodes[0]; 3116c5c9cd4dSSage Weil slot = path->slots[0]; 3117c5c9cd4dSSage Weil write_extent_buffer(leaf, buf, 3118c5c9cd4dSSage Weil btrfs_item_ptr_offset(leaf, slot), 3119c5c9cd4dSSage Weil size); 3120c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 3121c5c9cd4dSSage Weil } 3122c5c9cd4dSSage Weil 3123c5c9cd4dSSage Weil btrfs_mark_buffer_dirty(leaf); 3124b3b4aa74SDavid Sterba btrfs_release_path(path); 3125c5c9cd4dSSage Weil 31260c4d2d95SJosef Bacik inode_inc_iversion(inode); 3127a22285a6SYan, Zheng inode->i_mtime = inode->i_ctime = CURRENT_TIME; 3128b5384d48SSage Weil 3129b5384d48SSage Weil /* 3130b5384d48SSage Weil * we round up to the block size at eof when 3131b5384d48SSage Weil * determining which extents to clone above, 3132b5384d48SSage Weil * but shouldn't round up the file size 3133b5384d48SSage Weil */ 3134b5384d48SSage Weil endoff = new_key.offset + datal; 31355f3888ffSLi Zefan if (endoff > destoff+olen) 31365f3888ffSLi Zefan endoff = destoff+olen; 3137b5384d48SSage Weil if (endoff > inode->i_size) 3138b5384d48SSage Weil btrfs_i_size_write(inode, endoff); 3139b5384d48SSage Weil 3140a22285a6SYan, Zheng ret = btrfs_update_inode(trans, root, inode); 314179787eaaSJeff Mahoney if (ret) { 314279787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 3143a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 314479787eaaSJeff Mahoney goto out; 314579787eaaSJeff Mahoney } 314679787eaaSJeff Mahoney ret = btrfs_end_transaction(trans, root); 3147a22285a6SYan, Zheng } 3148b3b4aa74SDavid Sterba btrfs_release_path(path); 3149ae01a0abSYan Zheng key.offset++; 3150ae01a0abSYan Zheng } 3151f46b5a66SChristoph Hellwig ret = 0; 315232b7c687SMark Fasheh 3153f46b5a66SChristoph Hellwig out: 3154b3b4aa74SDavid Sterba btrfs_release_path(path); 315532b7c687SMark Fasheh btrfs_free_path(path); 315632b7c687SMark Fasheh vfree(buf); 315732b7c687SMark Fasheh return ret; 315832b7c687SMark Fasheh } 315932b7c687SMark Fasheh 316032b7c687SMark Fasheh static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd, 316132b7c687SMark Fasheh u64 off, u64 olen, u64 destoff) 316232b7c687SMark Fasheh { 316354563d41SAl Viro struct inode *inode = file_inode(file); 316432b7c687SMark Fasheh struct btrfs_root *root = BTRFS_I(inode)->root; 316532b7c687SMark Fasheh struct fd src_file; 316632b7c687SMark Fasheh struct inode *src; 316732b7c687SMark Fasheh int ret; 316832b7c687SMark Fasheh u64 len = olen; 316932b7c687SMark Fasheh u64 bs = root->fs_info->sb->s_blocksize; 317032b7c687SMark Fasheh int same_inode = 0; 317132b7c687SMark Fasheh 317232b7c687SMark Fasheh /* 317332b7c687SMark Fasheh * TODO: 317432b7c687SMark Fasheh * - split compressed inline extents. annoying: we need to 317532b7c687SMark Fasheh * decompress into destination's address_space (the file offset 317632b7c687SMark Fasheh * may change, so source mapping won't do), then recompress (or 317732b7c687SMark Fasheh * otherwise reinsert) a subrange. 317832b7c687SMark Fasheh * - allow ranges within the same file to be cloned (provided 317932b7c687SMark Fasheh * they don't overlap)? 318032b7c687SMark Fasheh */ 318132b7c687SMark Fasheh 318232b7c687SMark Fasheh /* the destination must be opened for writing */ 318332b7c687SMark Fasheh if (!(file->f_mode & FMODE_WRITE) || (file->f_flags & O_APPEND)) 318432b7c687SMark Fasheh return -EINVAL; 318532b7c687SMark Fasheh 318632b7c687SMark Fasheh if (btrfs_root_readonly(root)) 318732b7c687SMark Fasheh return -EROFS; 318832b7c687SMark Fasheh 318932b7c687SMark Fasheh ret = mnt_want_write_file(file); 319032b7c687SMark Fasheh if (ret) 319132b7c687SMark Fasheh return ret; 319232b7c687SMark Fasheh 319332b7c687SMark Fasheh src_file = fdget(srcfd); 319432b7c687SMark Fasheh if (!src_file.file) { 319532b7c687SMark Fasheh ret = -EBADF; 319632b7c687SMark Fasheh goto out_drop_write; 319732b7c687SMark Fasheh } 319832b7c687SMark Fasheh 319932b7c687SMark Fasheh ret = -EXDEV; 320032b7c687SMark Fasheh if (src_file.file->f_path.mnt != file->f_path.mnt) 320132b7c687SMark Fasheh goto out_fput; 320232b7c687SMark Fasheh 320332b7c687SMark Fasheh src = file_inode(src_file.file); 320432b7c687SMark Fasheh 320532b7c687SMark Fasheh ret = -EINVAL; 320632b7c687SMark Fasheh if (src == inode) 320732b7c687SMark Fasheh same_inode = 1; 320832b7c687SMark Fasheh 320932b7c687SMark Fasheh /* the src must be open for reading */ 321032b7c687SMark Fasheh if (!(src_file.file->f_mode & FMODE_READ)) 321132b7c687SMark Fasheh goto out_fput; 321232b7c687SMark Fasheh 321332b7c687SMark Fasheh /* don't make the dst file partly checksummed */ 321432b7c687SMark Fasheh if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 321532b7c687SMark Fasheh (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) 321632b7c687SMark Fasheh goto out_fput; 321732b7c687SMark Fasheh 321832b7c687SMark Fasheh ret = -EISDIR; 321932b7c687SMark Fasheh if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode)) 322032b7c687SMark Fasheh goto out_fput; 322132b7c687SMark Fasheh 322232b7c687SMark Fasheh ret = -EXDEV; 322332b7c687SMark Fasheh if (src->i_sb != inode->i_sb) 322432b7c687SMark Fasheh goto out_fput; 322532b7c687SMark Fasheh 322632b7c687SMark Fasheh if (!same_inode) { 322732b7c687SMark Fasheh if (inode < src) { 322832b7c687SMark Fasheh mutex_lock_nested(&inode->i_mutex, I_MUTEX_PARENT); 322932b7c687SMark Fasheh mutex_lock_nested(&src->i_mutex, I_MUTEX_CHILD); 323032b7c687SMark Fasheh } else { 323132b7c687SMark Fasheh mutex_lock_nested(&src->i_mutex, I_MUTEX_PARENT); 323232b7c687SMark Fasheh mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); 323332b7c687SMark Fasheh } 323432b7c687SMark Fasheh } else { 323532b7c687SMark Fasheh mutex_lock(&src->i_mutex); 323632b7c687SMark Fasheh } 323732b7c687SMark Fasheh 323832b7c687SMark Fasheh /* determine range to clone */ 323932b7c687SMark Fasheh ret = -EINVAL; 324032b7c687SMark Fasheh if (off + len > src->i_size || off + len < off) 324132b7c687SMark Fasheh goto out_unlock; 324232b7c687SMark Fasheh if (len == 0) 324332b7c687SMark Fasheh olen = len = src->i_size - off; 324432b7c687SMark Fasheh /* if we extend to eof, continue to block boundary */ 324532b7c687SMark Fasheh if (off + len == src->i_size) 324632b7c687SMark Fasheh len = ALIGN(src->i_size, bs) - off; 324732b7c687SMark Fasheh 324832b7c687SMark Fasheh /* verify the end result is block aligned */ 324932b7c687SMark Fasheh if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) || 325032b7c687SMark Fasheh !IS_ALIGNED(destoff, bs)) 325132b7c687SMark Fasheh goto out_unlock; 325232b7c687SMark Fasheh 325332b7c687SMark Fasheh /* verify if ranges are overlapped within the same file */ 325432b7c687SMark Fasheh if (same_inode) { 325532b7c687SMark Fasheh if (destoff + len > off && destoff < off + len) 325632b7c687SMark Fasheh goto out_unlock; 325732b7c687SMark Fasheh } 325832b7c687SMark Fasheh 325932b7c687SMark Fasheh if (destoff > inode->i_size) { 326032b7c687SMark Fasheh ret = btrfs_cont_expand(inode, inode->i_size, destoff); 326132b7c687SMark Fasheh if (ret) 326232b7c687SMark Fasheh goto out_unlock; 326332b7c687SMark Fasheh } 326432b7c687SMark Fasheh 326532b7c687SMark Fasheh /* truncate page cache pages from target inode range */ 326632b7c687SMark Fasheh truncate_inode_pages_range(&inode->i_data, destoff, 326732b7c687SMark Fasheh PAGE_CACHE_ALIGN(destoff + len) - 1); 326832b7c687SMark Fasheh 326932b7c687SMark Fasheh lock_extent_range(src, off, len); 327032b7c687SMark Fasheh 327132b7c687SMark Fasheh ret = btrfs_clone(src, inode, off, olen, len, destoff); 327232b7c687SMark Fasheh 3273aa42ffd9SLiu Bo unlock_extent(&BTRFS_I(src)->io_tree, off, off + len - 1); 3274f46b5a66SChristoph Hellwig out_unlock: 3275c57c2b3eSFilipe David Borba Manana if (!same_inode) { 3276c57c2b3eSFilipe David Borba Manana if (inode < src) { 3277f46b5a66SChristoph Hellwig mutex_unlock(&src->i_mutex); 3278f46b5a66SChristoph Hellwig mutex_unlock(&inode->i_mutex); 3279c57c2b3eSFilipe David Borba Manana } else { 3280c57c2b3eSFilipe David Borba Manana mutex_unlock(&inode->i_mutex); 3281c57c2b3eSFilipe David Borba Manana mutex_unlock(&src->i_mutex); 3282c57c2b3eSFilipe David Borba Manana } 3283c57c2b3eSFilipe David Borba Manana } else { 3284c57c2b3eSFilipe David Borba Manana mutex_unlock(&src->i_mutex); 3285c57c2b3eSFilipe David Borba Manana } 3286f46b5a66SChristoph Hellwig out_fput: 32872903ff01SAl Viro fdput(src_file); 3288ab67b7c1SYan Zheng out_drop_write: 32892a79f17eSAl Viro mnt_drop_write_file(file); 3290f46b5a66SChristoph Hellwig return ret; 3291f46b5a66SChristoph Hellwig } 3292f46b5a66SChristoph Hellwig 32937a865e8aSChristoph Hellwig static long btrfs_ioctl_clone_range(struct file *file, void __user *argp) 3294c5c9cd4dSSage Weil { 3295c5c9cd4dSSage Weil struct btrfs_ioctl_clone_range_args args; 3296c5c9cd4dSSage Weil 32977a865e8aSChristoph Hellwig if (copy_from_user(&args, argp, sizeof(args))) 3298c5c9cd4dSSage Weil return -EFAULT; 3299c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, args.src_fd, args.src_offset, 3300c5c9cd4dSSage Weil args.src_length, args.dest_offset); 3301c5c9cd4dSSage Weil } 3302c5c9cd4dSSage Weil 3303f46b5a66SChristoph Hellwig /* 3304f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 3305f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 3306f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 3307f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 3308f46b5a66SChristoph Hellwig */ 3309b2950863SChristoph Hellwig static long btrfs_ioctl_trans_start(struct file *file) 3310f46b5a66SChristoph Hellwig { 3311496ad9aaSAl Viro struct inode *inode = file_inode(file); 3312f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3313f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 33141ab86aedSSage Weil int ret; 3315f46b5a66SChristoph Hellwig 33161ab86aedSSage Weil ret = -EPERM; 3317df5b5520SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 3318f46b5a66SChristoph Hellwig goto out; 33191ab86aedSSage Weil 33201ab86aedSSage Weil ret = -EINPROGRESS; 33211ab86aedSSage Weil if (file->private_data) 33221ab86aedSSage Weil goto out; 33239ca9ee09SSage Weil 3324b83cc969SLi Zefan ret = -EROFS; 3325b83cc969SLi Zefan if (btrfs_root_readonly(root)) 3326b83cc969SLi Zefan goto out; 3327b83cc969SLi Zefan 3328a561be71SAl Viro ret = mnt_want_write_file(file); 3329c146afadSYan Zheng if (ret) 3330c146afadSYan Zheng goto out; 3331c146afadSYan Zheng 3332a4abeea4SJosef Bacik atomic_inc(&root->fs_info->open_ioctl_trans); 33339ca9ee09SSage Weil 3334f46b5a66SChristoph Hellwig ret = -ENOMEM; 33357a7eaa40SJosef Bacik trans = btrfs_start_ioctl_transaction(root); 3336abd30bb0STsutomu Itoh if (IS_ERR(trans)) 33371ab86aedSSage Weil goto out_drop; 33381ab86aedSSage Weil 33391ab86aedSSage Weil file->private_data = trans; 33401ab86aedSSage Weil return 0; 33411ab86aedSSage Weil 33421ab86aedSSage Weil out_drop: 3343a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 33442a79f17eSAl Viro mnt_drop_write_file(file); 3345f46b5a66SChristoph Hellwig out: 3346f46b5a66SChristoph Hellwig return ret; 3347f46b5a66SChristoph Hellwig } 3348f46b5a66SChristoph Hellwig 33496ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 33506ef5ed0dSJosef Bacik { 3351496ad9aaSAl Viro struct inode *inode = file_inode(file); 33526ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 33536ef5ed0dSJosef Bacik struct btrfs_root *new_root; 33546ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 33556ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 33566ef5ed0dSJosef Bacik struct btrfs_path *path; 33576ef5ed0dSJosef Bacik struct btrfs_key location; 33586ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 33596ef5ed0dSJosef Bacik u64 objectid = 0; 33606ef5ed0dSJosef Bacik u64 dir_id; 33613c04ce01SMiao Xie int ret; 33626ef5ed0dSJosef Bacik 33636ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 33646ef5ed0dSJosef Bacik return -EPERM; 33656ef5ed0dSJosef Bacik 33663c04ce01SMiao Xie ret = mnt_want_write_file(file); 33673c04ce01SMiao Xie if (ret) 33683c04ce01SMiao Xie return ret; 33693c04ce01SMiao Xie 33703c04ce01SMiao Xie if (copy_from_user(&objectid, argp, sizeof(objectid))) { 33713c04ce01SMiao Xie ret = -EFAULT; 33723c04ce01SMiao Xie goto out; 33733c04ce01SMiao Xie } 33746ef5ed0dSJosef Bacik 33756ef5ed0dSJosef Bacik if (!objectid) 33761cecf579Schandan objectid = BTRFS_FS_TREE_OBJECTID; 33776ef5ed0dSJosef Bacik 33786ef5ed0dSJosef Bacik location.objectid = objectid; 33796ef5ed0dSJosef Bacik location.type = BTRFS_ROOT_ITEM_KEY; 33806ef5ed0dSJosef Bacik location.offset = (u64)-1; 33816ef5ed0dSJosef Bacik 33826ef5ed0dSJosef Bacik new_root = btrfs_read_fs_root_no_name(root->fs_info, &location); 33833c04ce01SMiao Xie if (IS_ERR(new_root)) { 33843c04ce01SMiao Xie ret = PTR_ERR(new_root); 33853c04ce01SMiao Xie goto out; 33863c04ce01SMiao Xie } 33876ef5ed0dSJosef Bacik 33886ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 33893c04ce01SMiao Xie if (!path) { 33903c04ce01SMiao Xie ret = -ENOMEM; 33913c04ce01SMiao Xie goto out; 33923c04ce01SMiao Xie } 33936ef5ed0dSJosef Bacik path->leave_spinning = 1; 33946ef5ed0dSJosef Bacik 33956ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 339698d5dc13STsutomu Itoh if (IS_ERR(trans)) { 33976ef5ed0dSJosef Bacik btrfs_free_path(path); 33983c04ce01SMiao Xie ret = PTR_ERR(trans); 33993c04ce01SMiao Xie goto out; 34006ef5ed0dSJosef Bacik } 34016ef5ed0dSJosef Bacik 34026c41761fSDavid Sterba dir_id = btrfs_super_root_dir(root->fs_info->super_copy); 34036ef5ed0dSJosef Bacik di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path, 34046ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 3405cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 34066ef5ed0dSJosef Bacik btrfs_free_path(path); 34076ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 3408efe120a0SFrank Holton btrfs_err(new_root->fs_info, "Umm, you don't have the default dir" 3409efe120a0SFrank Holton "item, this isn't going to work"); 34103c04ce01SMiao Xie ret = -ENOENT; 34113c04ce01SMiao Xie goto out; 34126ef5ed0dSJosef Bacik } 34136ef5ed0dSJosef Bacik 34146ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 34156ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 34166ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 34176ef5ed0dSJosef Bacik btrfs_free_path(path); 34186ef5ed0dSJosef Bacik 34192b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL); 34206ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 34213c04ce01SMiao Xie out: 34223c04ce01SMiao Xie mnt_drop_write_file(file); 34233c04ce01SMiao Xie return ret; 34246ef5ed0dSJosef Bacik } 34256ef5ed0dSJosef Bacik 34265af3e8ccSStefan Behrens void btrfs_get_block_group_info(struct list_head *groups_list, 3427bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 3428bf5fc093SJosef Bacik { 3429bf5fc093SJosef Bacik struct btrfs_block_group_cache *block_group; 3430bf5fc093SJosef Bacik 3431bf5fc093SJosef Bacik space->total_bytes = 0; 3432bf5fc093SJosef Bacik space->used_bytes = 0; 3433bf5fc093SJosef Bacik space->flags = 0; 3434bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 3435bf5fc093SJosef Bacik space->flags = block_group->flags; 3436bf5fc093SJosef Bacik space->total_bytes += block_group->key.offset; 3437bf5fc093SJosef Bacik space->used_bytes += 3438bf5fc093SJosef Bacik btrfs_block_group_used(&block_group->item); 3439bf5fc093SJosef Bacik } 3440bf5fc093SJosef Bacik } 3441bf5fc093SJosef Bacik 344248a3b636SEric Sandeen static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg) 34431406e432SJosef Bacik { 34441406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 34451406e432SJosef Bacik struct btrfs_ioctl_space_info space; 34461406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 34477fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 344813f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 34491406e432SJosef Bacik struct btrfs_space_info *info; 3450bf5fc093SJosef Bacik u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 3451bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 3452bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 3453bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 3454bf5fc093SJosef Bacik int num_types = 4; 34557fde62bfSChris Mason int alloc_size; 34561406e432SJosef Bacik int ret = 0; 345751788b1bSDan Rosenberg u64 slot_count = 0; 3458bf5fc093SJosef Bacik int i, c; 34591406e432SJosef Bacik 34601406e432SJosef Bacik if (copy_from_user(&space_args, 34611406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 34621406e432SJosef Bacik sizeof(space_args))) 34631406e432SJosef Bacik return -EFAULT; 34641406e432SJosef Bacik 3465bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3466bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3467bf5fc093SJosef Bacik 3468bf5fc093SJosef Bacik info = NULL; 34697fde62bfSChris Mason rcu_read_lock(); 3470bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 3471bf5fc093SJosef Bacik list) { 3472bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3473bf5fc093SJosef Bacik info = tmp; 3474bf5fc093SJosef Bacik break; 3475bf5fc093SJosef Bacik } 3476bf5fc093SJosef Bacik } 34777fde62bfSChris Mason rcu_read_unlock(); 34781406e432SJosef Bacik 3479bf5fc093SJosef Bacik if (!info) 3480bf5fc093SJosef Bacik continue; 3481bf5fc093SJosef Bacik 3482bf5fc093SJosef Bacik down_read(&info->groups_sem); 3483bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3484bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 3485bf5fc093SJosef Bacik slot_count++; 3486bf5fc093SJosef Bacik } 3487bf5fc093SJosef Bacik up_read(&info->groups_sem); 3488bf5fc093SJosef Bacik } 3489bf5fc093SJosef Bacik 34907fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 34917fde62bfSChris Mason if (space_args.space_slots == 0) { 34927fde62bfSChris Mason space_args.total_spaces = slot_count; 34937fde62bfSChris Mason goto out; 34947fde62bfSChris Mason } 3495bf5fc093SJosef Bacik 349651788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 3497bf5fc093SJosef Bacik 34987fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 3499bf5fc093SJosef Bacik 35007fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 35017fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 35027fde62bfSChris Mason */ 35037fde62bfSChris Mason if (alloc_size > PAGE_CACHE_SIZE) 35047fde62bfSChris Mason return -ENOMEM; 35057fde62bfSChris Mason 35067fde62bfSChris Mason space_args.total_spaces = 0; 35077fde62bfSChris Mason dest = kmalloc(alloc_size, GFP_NOFS); 35087fde62bfSChris Mason if (!dest) 35097fde62bfSChris Mason return -ENOMEM; 35107fde62bfSChris Mason dest_orig = dest; 35117fde62bfSChris Mason 35127fde62bfSChris Mason /* now we have a buffer to copy into */ 3513bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 3514bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 3515bf5fc093SJosef Bacik 351651788b1bSDan Rosenberg if (!slot_count) 351751788b1bSDan Rosenberg break; 351851788b1bSDan Rosenberg 3519bf5fc093SJosef Bacik info = NULL; 35201406e432SJosef Bacik rcu_read_lock(); 3521bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 3522bf5fc093SJosef Bacik list) { 3523bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 3524bf5fc093SJosef Bacik info = tmp; 35257fde62bfSChris Mason break; 3526bf5fc093SJosef Bacik } 3527bf5fc093SJosef Bacik } 3528bf5fc093SJosef Bacik rcu_read_unlock(); 35297fde62bfSChris Mason 3530bf5fc093SJosef Bacik if (!info) 3531bf5fc093SJosef Bacik continue; 3532bf5fc093SJosef Bacik down_read(&info->groups_sem); 3533bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3534bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 35355af3e8ccSStefan Behrens btrfs_get_block_group_info( 35365af3e8ccSStefan Behrens &info->block_groups[c], &space); 35377fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 35381406e432SJosef Bacik dest++; 35391406e432SJosef Bacik space_args.total_spaces++; 354051788b1bSDan Rosenberg slot_count--; 35411406e432SJosef Bacik } 354251788b1bSDan Rosenberg if (!slot_count) 354351788b1bSDan Rosenberg break; 3544bf5fc093SJosef Bacik } 3545bf5fc093SJosef Bacik up_read(&info->groups_sem); 3546bf5fc093SJosef Bacik } 35471406e432SJosef Bacik 35482eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 35497fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 35507fde62bfSChris Mason 35517fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 35527fde62bfSChris Mason ret = -EFAULT; 35537fde62bfSChris Mason 35547fde62bfSChris Mason kfree(dest_orig); 35557fde62bfSChris Mason out: 35567fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 35571406e432SJosef Bacik ret = -EFAULT; 35581406e432SJosef Bacik 35591406e432SJosef Bacik return ret; 35601406e432SJosef Bacik } 35611406e432SJosef Bacik 356201e219e8SJeff Mahoney static long btrfs_ioctl_global_rsv(struct btrfs_root *root, void __user *arg) 356301e219e8SJeff Mahoney { 356401e219e8SJeff Mahoney struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv; 356501e219e8SJeff Mahoney u64 reserved; 356601e219e8SJeff Mahoney 356701e219e8SJeff Mahoney spin_lock(&block_rsv->lock); 356801e219e8SJeff Mahoney reserved = block_rsv->reserved; 356901e219e8SJeff Mahoney spin_unlock(&block_rsv->lock); 357001e219e8SJeff Mahoney 357101e219e8SJeff Mahoney if (arg && copy_to_user(arg, &reserved, sizeof(reserved))) 357201e219e8SJeff Mahoney return -EFAULT; 357301e219e8SJeff Mahoney return 0; 357401e219e8SJeff Mahoney } 357501e219e8SJeff Mahoney 3576f46b5a66SChristoph Hellwig /* 3577f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 3578f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 3579f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 3580f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 3581f46b5a66SChristoph Hellwig */ 3582f46b5a66SChristoph Hellwig long btrfs_ioctl_trans_end(struct file *file) 3583f46b5a66SChristoph Hellwig { 3584496ad9aaSAl Viro struct inode *inode = file_inode(file); 3585f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3586f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 3587f46b5a66SChristoph Hellwig 3588f46b5a66SChristoph Hellwig trans = file->private_data; 35891ab86aedSSage Weil if (!trans) 35901ab86aedSSage Weil return -EINVAL; 3591b214107eSChristoph Hellwig file->private_data = NULL; 35929ca9ee09SSage Weil 35931ab86aedSSage Weil btrfs_end_transaction(trans, root); 35941ab86aedSSage Weil 3595a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 35969ca9ee09SSage Weil 35972a79f17eSAl Viro mnt_drop_write_file(file); 35981ab86aedSSage Weil return 0; 3599f46b5a66SChristoph Hellwig } 3600f46b5a66SChristoph Hellwig 36019a8c28beSMiao Xie static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 36029a8c28beSMiao Xie void __user *argp) 360346204592SSage Weil { 360446204592SSage Weil struct btrfs_trans_handle *trans; 360546204592SSage Weil u64 transid; 3606db5b493aSTsutomu Itoh int ret; 360746204592SSage Weil 3608d4edf39bSMiao Xie trans = btrfs_attach_transaction_barrier(root); 3609ff7c1d33SMiao Xie if (IS_ERR(trans)) { 3610ff7c1d33SMiao Xie if (PTR_ERR(trans) != -ENOENT) 361198d5dc13STsutomu Itoh return PTR_ERR(trans); 3612ff7c1d33SMiao Xie 3613ff7c1d33SMiao Xie /* No running transaction, don't bother */ 3614ff7c1d33SMiao Xie transid = root->fs_info->last_trans_committed; 3615ff7c1d33SMiao Xie goto out; 3616ff7c1d33SMiao Xie } 361746204592SSage Weil transid = trans->transid; 3618db5b493aSTsutomu Itoh ret = btrfs_commit_transaction_async(trans, root, 0); 36198b2b2d3cSTsutomu Itoh if (ret) { 36208b2b2d3cSTsutomu Itoh btrfs_end_transaction(trans, root); 3621db5b493aSTsutomu Itoh return ret; 36228b2b2d3cSTsutomu Itoh } 3623ff7c1d33SMiao Xie out: 362446204592SSage Weil if (argp) 362546204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 362646204592SSage Weil return -EFAULT; 362746204592SSage Weil return 0; 362846204592SSage Weil } 362946204592SSage Weil 36309a8c28beSMiao Xie static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root, 36319a8c28beSMiao Xie void __user *argp) 363246204592SSage Weil { 363346204592SSage Weil u64 transid; 363446204592SSage Weil 363546204592SSage Weil if (argp) { 363646204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 363746204592SSage Weil return -EFAULT; 363846204592SSage Weil } else { 363946204592SSage Weil transid = 0; /* current trans */ 364046204592SSage Weil } 364146204592SSage Weil return btrfs_wait_for_commit(root, transid); 364246204592SSage Weil } 364346204592SSage Weil 3644b8e95489SMiao Xie static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 3645475f6387SJan Schmidt { 3646496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 3647475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3648b8e95489SMiao Xie int ret; 3649475f6387SJan Schmidt 3650475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3651475f6387SJan Schmidt return -EPERM; 3652475f6387SJan Schmidt 3653475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3654475f6387SJan Schmidt if (IS_ERR(sa)) 3655475f6387SJan Schmidt return PTR_ERR(sa); 3656475f6387SJan Schmidt 3657b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 3658b8e95489SMiao Xie ret = mnt_want_write_file(file); 3659b8e95489SMiao Xie if (ret) 3660b8e95489SMiao Xie goto out; 3661b8e95489SMiao Xie } 3662b8e95489SMiao Xie 3663aa1b8cd4SStefan Behrens ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end, 366463a212abSStefan Behrens &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 366563a212abSStefan Behrens 0); 3666475f6387SJan Schmidt 3667475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 3668475f6387SJan Schmidt ret = -EFAULT; 3669475f6387SJan Schmidt 3670b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) 3671b8e95489SMiao Xie mnt_drop_write_file(file); 3672b8e95489SMiao Xie out: 3673475f6387SJan Schmidt kfree(sa); 3674475f6387SJan Schmidt return ret; 3675475f6387SJan Schmidt } 3676475f6387SJan Schmidt 3677475f6387SJan Schmidt static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg) 3678475f6387SJan Schmidt { 3679475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3680475f6387SJan Schmidt return -EPERM; 3681475f6387SJan Schmidt 3682aa1b8cd4SStefan Behrens return btrfs_scrub_cancel(root->fs_info); 3683475f6387SJan Schmidt } 3684475f6387SJan Schmidt 3685475f6387SJan Schmidt static long btrfs_ioctl_scrub_progress(struct btrfs_root *root, 3686475f6387SJan Schmidt void __user *arg) 3687475f6387SJan Schmidt { 3688475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 3689475f6387SJan Schmidt int ret; 3690475f6387SJan Schmidt 3691475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3692475f6387SJan Schmidt return -EPERM; 3693475f6387SJan Schmidt 3694475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 3695475f6387SJan Schmidt if (IS_ERR(sa)) 3696475f6387SJan Schmidt return PTR_ERR(sa); 3697475f6387SJan Schmidt 3698475f6387SJan Schmidt ret = btrfs_scrub_progress(root, sa->devid, &sa->progress); 3699475f6387SJan Schmidt 3700475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 3701475f6387SJan Schmidt ret = -EFAULT; 3702475f6387SJan Schmidt 3703475f6387SJan Schmidt kfree(sa); 3704475f6387SJan Schmidt return ret; 3705475f6387SJan Schmidt } 3706475f6387SJan Schmidt 3707c11d2c23SStefan Behrens static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root, 3708b27f7c0cSDavid Sterba void __user *arg) 3709c11d2c23SStefan Behrens { 3710c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 3711c11d2c23SStefan Behrens int ret; 3712c11d2c23SStefan Behrens 3713c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 3714c11d2c23SStefan Behrens if (IS_ERR(sa)) 3715c11d2c23SStefan Behrens return PTR_ERR(sa); 3716c11d2c23SStefan Behrens 3717b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 3718b27f7c0cSDavid Sterba kfree(sa); 3719b27f7c0cSDavid Sterba return -EPERM; 3720b27f7c0cSDavid Sterba } 3721b27f7c0cSDavid Sterba 3722b27f7c0cSDavid Sterba ret = btrfs_get_dev_stats(root, sa); 3723c11d2c23SStefan Behrens 3724c11d2c23SStefan Behrens if (copy_to_user(arg, sa, sizeof(*sa))) 3725c11d2c23SStefan Behrens ret = -EFAULT; 3726c11d2c23SStefan Behrens 3727c11d2c23SStefan Behrens kfree(sa); 3728c11d2c23SStefan Behrens return ret; 3729c11d2c23SStefan Behrens } 3730c11d2c23SStefan Behrens 37313f6bcfbdSStefan Behrens static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg) 37323f6bcfbdSStefan Behrens { 37333f6bcfbdSStefan Behrens struct btrfs_ioctl_dev_replace_args *p; 37343f6bcfbdSStefan Behrens int ret; 37353f6bcfbdSStefan Behrens 37363f6bcfbdSStefan Behrens if (!capable(CAP_SYS_ADMIN)) 37373f6bcfbdSStefan Behrens return -EPERM; 37383f6bcfbdSStefan Behrens 37393f6bcfbdSStefan Behrens p = memdup_user(arg, sizeof(*p)); 37403f6bcfbdSStefan Behrens if (IS_ERR(p)) 37413f6bcfbdSStefan Behrens return PTR_ERR(p); 37423f6bcfbdSStefan Behrens 37433f6bcfbdSStefan Behrens switch (p->cmd) { 37443f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 3745adfa97cbSIlya Dryomov if (root->fs_info->sb->s_flags & MS_RDONLY) { 3746adfa97cbSIlya Dryomov ret = -EROFS; 3747adfa97cbSIlya Dryomov goto out; 3748adfa97cbSIlya Dryomov } 37493f6bcfbdSStefan Behrens if (atomic_xchg( 37503f6bcfbdSStefan Behrens &root->fs_info->mutually_exclusive_operation_running, 37513f6bcfbdSStefan Behrens 1)) { 3752e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 37533f6bcfbdSStefan Behrens } else { 37543f6bcfbdSStefan Behrens ret = btrfs_dev_replace_start(root, p); 37553f6bcfbdSStefan Behrens atomic_set( 37563f6bcfbdSStefan Behrens &root->fs_info->mutually_exclusive_operation_running, 37573f6bcfbdSStefan Behrens 0); 37583f6bcfbdSStefan Behrens } 37593f6bcfbdSStefan Behrens break; 37603f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 37613f6bcfbdSStefan Behrens btrfs_dev_replace_status(root->fs_info, p); 37623f6bcfbdSStefan Behrens ret = 0; 37633f6bcfbdSStefan Behrens break; 37643f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 37653f6bcfbdSStefan Behrens ret = btrfs_dev_replace_cancel(root->fs_info, p); 37663f6bcfbdSStefan Behrens break; 37673f6bcfbdSStefan Behrens default: 37683f6bcfbdSStefan Behrens ret = -EINVAL; 37693f6bcfbdSStefan Behrens break; 37703f6bcfbdSStefan Behrens } 37713f6bcfbdSStefan Behrens 37723f6bcfbdSStefan Behrens if (copy_to_user(arg, p, sizeof(*p))) 37733f6bcfbdSStefan Behrens ret = -EFAULT; 3774adfa97cbSIlya Dryomov out: 37753f6bcfbdSStefan Behrens kfree(p); 37763f6bcfbdSStefan Behrens return ret; 37773f6bcfbdSStefan Behrens } 37783f6bcfbdSStefan Behrens 3779d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 3780d7728c96SJan Schmidt { 3781d7728c96SJan Schmidt int ret = 0; 3782d7728c96SJan Schmidt int i; 3783740c3d22SChris Mason u64 rel_ptr; 3784d7728c96SJan Schmidt int size; 3785806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 3786d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 3787d7728c96SJan Schmidt struct btrfs_path *path; 3788d7728c96SJan Schmidt 378982b22ac8SKusanagi Kouichi if (!capable(CAP_DAC_READ_SEARCH)) 3790d7728c96SJan Schmidt return -EPERM; 3791d7728c96SJan Schmidt 3792d7728c96SJan Schmidt path = btrfs_alloc_path(); 3793d7728c96SJan Schmidt if (!path) { 3794d7728c96SJan Schmidt ret = -ENOMEM; 3795d7728c96SJan Schmidt goto out; 3796d7728c96SJan Schmidt } 3797d7728c96SJan Schmidt 3798d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 3799d7728c96SJan Schmidt if (IS_ERR(ipa)) { 3800d7728c96SJan Schmidt ret = PTR_ERR(ipa); 3801d7728c96SJan Schmidt ipa = NULL; 3802d7728c96SJan Schmidt goto out; 3803d7728c96SJan Schmidt } 3804d7728c96SJan Schmidt 3805d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 3806d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 3807d7728c96SJan Schmidt if (IS_ERR(ipath)) { 3808d7728c96SJan Schmidt ret = PTR_ERR(ipath); 3809d7728c96SJan Schmidt ipath = NULL; 3810d7728c96SJan Schmidt goto out; 3811d7728c96SJan Schmidt } 3812d7728c96SJan Schmidt 3813d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 3814d7728c96SJan Schmidt if (ret < 0) 3815d7728c96SJan Schmidt goto out; 3816d7728c96SJan Schmidt 3817d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 3818745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 3819745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 3820740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 3821d7728c96SJan Schmidt } 3822d7728c96SJan Schmidt 3823745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)ipa->fspath, 3824745c4d8eSJeff Mahoney (void *)(unsigned long)ipath->fspath, size); 3825d7728c96SJan Schmidt if (ret) { 3826d7728c96SJan Schmidt ret = -EFAULT; 3827d7728c96SJan Schmidt goto out; 3828d7728c96SJan Schmidt } 3829d7728c96SJan Schmidt 3830d7728c96SJan Schmidt out: 3831d7728c96SJan Schmidt btrfs_free_path(path); 3832d7728c96SJan Schmidt free_ipath(ipath); 3833d7728c96SJan Schmidt kfree(ipa); 3834d7728c96SJan Schmidt 3835d7728c96SJan Schmidt return ret; 3836d7728c96SJan Schmidt } 3837d7728c96SJan Schmidt 3838d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 3839d7728c96SJan Schmidt { 3840d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 3841d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 3842d7728c96SJan Schmidt 3843d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 3844d7728c96SJan Schmidt inodes->bytes_left -= c; 3845d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 3846d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 3847d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 3848d7728c96SJan Schmidt inodes->elem_cnt += 3; 3849d7728c96SJan Schmidt } else { 3850d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 3851d7728c96SJan Schmidt inodes->bytes_left = 0; 3852d7728c96SJan Schmidt inodes->elem_missed += 3; 3853d7728c96SJan Schmidt } 3854d7728c96SJan Schmidt 3855d7728c96SJan Schmidt return 0; 3856d7728c96SJan Schmidt } 3857d7728c96SJan Schmidt 3858d7728c96SJan Schmidt static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root, 3859d7728c96SJan Schmidt void __user *arg) 3860d7728c96SJan Schmidt { 3861d7728c96SJan Schmidt int ret = 0; 3862d7728c96SJan Schmidt int size; 3863d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 3864d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 3865d7728c96SJan Schmidt struct btrfs_path *path = NULL; 3866d7728c96SJan Schmidt 3867d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 3868d7728c96SJan Schmidt return -EPERM; 3869d7728c96SJan Schmidt 3870d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 3871d7728c96SJan Schmidt if (IS_ERR(loi)) { 3872d7728c96SJan Schmidt ret = PTR_ERR(loi); 3873d7728c96SJan Schmidt loi = NULL; 3874d7728c96SJan Schmidt goto out; 3875d7728c96SJan Schmidt } 3876d7728c96SJan Schmidt 3877d7728c96SJan Schmidt path = btrfs_alloc_path(); 3878d7728c96SJan Schmidt if (!path) { 3879d7728c96SJan Schmidt ret = -ENOMEM; 3880d7728c96SJan Schmidt goto out; 3881d7728c96SJan Schmidt } 3882d7728c96SJan Schmidt 3883425d17a2SLiu Bo size = min_t(u32, loi->size, 64 * 1024); 3884d7728c96SJan Schmidt inodes = init_data_container(size); 3885d7728c96SJan Schmidt if (IS_ERR(inodes)) { 3886d7728c96SJan Schmidt ret = PTR_ERR(inodes); 3887d7728c96SJan Schmidt inodes = NULL; 3888d7728c96SJan Schmidt goto out; 3889d7728c96SJan Schmidt } 3890d7728c96SJan Schmidt 3891df031f07SLiu Bo ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path, 3892df031f07SLiu Bo build_ino_list, inodes); 3893df031f07SLiu Bo if (ret == -EINVAL) 3894d7728c96SJan Schmidt ret = -ENOENT; 3895d7728c96SJan Schmidt if (ret < 0) 3896d7728c96SJan Schmidt goto out; 3897d7728c96SJan Schmidt 3898745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)loi->inodes, 3899745c4d8eSJeff Mahoney (void *)(unsigned long)inodes, size); 3900d7728c96SJan Schmidt if (ret) 3901d7728c96SJan Schmidt ret = -EFAULT; 3902d7728c96SJan Schmidt 3903d7728c96SJan Schmidt out: 3904d7728c96SJan Schmidt btrfs_free_path(path); 3905425d17a2SLiu Bo vfree(inodes); 3906d7728c96SJan Schmidt kfree(loi); 3907d7728c96SJan Schmidt 3908d7728c96SJan Schmidt return ret; 3909d7728c96SJan Schmidt } 3910d7728c96SJan Schmidt 391119a39dceSIlya Dryomov void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, 3912c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 3913c9e9f97bSIlya Dryomov { 3914c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 3915c9e9f97bSIlya Dryomov 3916c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 3917c9e9f97bSIlya Dryomov 3918837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_running)) 3919837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 3920837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 3921837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 3922a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 3923a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 3924837d5b6eSIlya Dryomov 3925c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 3926c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 3927c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 392819a39dceSIlya Dryomov 392919a39dceSIlya Dryomov if (lock) { 393019a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 393119a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 393219a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 393319a39dceSIlya Dryomov } else { 393419a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 393519a39dceSIlya Dryomov } 3936c9e9f97bSIlya Dryomov } 3937c9e9f97bSIlya Dryomov 39389ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 3939c9e9f97bSIlya Dryomov { 3940496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 3941c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 3942c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 3943c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 3944ed0fb78fSIlya Dryomov bool need_unlock; /* for mut. excl. ops lock */ 3945c9e9f97bSIlya Dryomov int ret; 3946c9e9f97bSIlya Dryomov 3947c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 3948c9e9f97bSIlya Dryomov return -EPERM; 3949c9e9f97bSIlya Dryomov 3950e54bfa31SLiu Bo ret = mnt_want_write_file(file); 39519ba1f6e4SLiu Bo if (ret) 39529ba1f6e4SLiu Bo return ret; 39539ba1f6e4SLiu Bo 3954ed0fb78fSIlya Dryomov again: 3955ed0fb78fSIlya Dryomov if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) { 3956c9e9f97bSIlya Dryomov mutex_lock(&fs_info->volume_mutex); 3957c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 3958ed0fb78fSIlya Dryomov need_unlock = true; 3959ed0fb78fSIlya Dryomov goto locked; 3960ed0fb78fSIlya Dryomov } 3961ed0fb78fSIlya Dryomov 3962ed0fb78fSIlya Dryomov /* 3963ed0fb78fSIlya Dryomov * mut. excl. ops lock is locked. Three possibilites: 3964ed0fb78fSIlya Dryomov * (1) some other op is running 3965ed0fb78fSIlya Dryomov * (2) balance is running 3966ed0fb78fSIlya Dryomov * (3) balance is paused -- special case (think resume) 3967ed0fb78fSIlya Dryomov */ 3968ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 3969ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 3970ed0fb78fSIlya Dryomov /* this is either (2) or (3) */ 3971ed0fb78fSIlya Dryomov if (!atomic_read(&fs_info->balance_running)) { 3972ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 3973ed0fb78fSIlya Dryomov if (!mutex_trylock(&fs_info->volume_mutex)) 3974ed0fb78fSIlya Dryomov goto again; 3975ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 3976ed0fb78fSIlya Dryomov 3977ed0fb78fSIlya Dryomov if (fs_info->balance_ctl && 3978ed0fb78fSIlya Dryomov !atomic_read(&fs_info->balance_running)) { 3979ed0fb78fSIlya Dryomov /* this is (3) */ 3980ed0fb78fSIlya Dryomov need_unlock = false; 3981ed0fb78fSIlya Dryomov goto locked; 3982ed0fb78fSIlya Dryomov } 3983ed0fb78fSIlya Dryomov 3984ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 3985ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->volume_mutex); 3986ed0fb78fSIlya Dryomov goto again; 3987ed0fb78fSIlya Dryomov } else { 3988ed0fb78fSIlya Dryomov /* this is (2) */ 3989ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 3990ed0fb78fSIlya Dryomov ret = -EINPROGRESS; 3991ed0fb78fSIlya Dryomov goto out; 3992ed0fb78fSIlya Dryomov } 3993ed0fb78fSIlya Dryomov } else { 3994ed0fb78fSIlya Dryomov /* this is (1) */ 3995ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 3996e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 3997ed0fb78fSIlya Dryomov goto out; 3998ed0fb78fSIlya Dryomov } 3999ed0fb78fSIlya Dryomov 4000ed0fb78fSIlya Dryomov locked: 4001ed0fb78fSIlya Dryomov BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running)); 4002c9e9f97bSIlya Dryomov 4003c9e9f97bSIlya Dryomov if (arg) { 4004c9e9f97bSIlya Dryomov bargs = memdup_user(arg, sizeof(*bargs)); 4005c9e9f97bSIlya Dryomov if (IS_ERR(bargs)) { 4006c9e9f97bSIlya Dryomov ret = PTR_ERR(bargs); 4007ed0fb78fSIlya Dryomov goto out_unlock; 4008c9e9f97bSIlya Dryomov } 4009de322263SIlya Dryomov 4010de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 4011de322263SIlya Dryomov if (!fs_info->balance_ctl) { 4012de322263SIlya Dryomov ret = -ENOTCONN; 4013de322263SIlya Dryomov goto out_bargs; 4014de322263SIlya Dryomov } 4015de322263SIlya Dryomov 4016de322263SIlya Dryomov bctl = fs_info->balance_ctl; 4017de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 4018de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 4019de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 4020de322263SIlya Dryomov 4021de322263SIlya Dryomov goto do_balance; 4022de322263SIlya Dryomov } 4023c9e9f97bSIlya Dryomov } else { 4024c9e9f97bSIlya Dryomov bargs = NULL; 4025c9e9f97bSIlya Dryomov } 4026c9e9f97bSIlya Dryomov 4027ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4028837d5b6eSIlya Dryomov ret = -EINPROGRESS; 4029837d5b6eSIlya Dryomov goto out_bargs; 4030837d5b6eSIlya Dryomov } 4031837d5b6eSIlya Dryomov 4032c9e9f97bSIlya Dryomov bctl = kzalloc(sizeof(*bctl), GFP_NOFS); 4033c9e9f97bSIlya Dryomov if (!bctl) { 4034c9e9f97bSIlya Dryomov ret = -ENOMEM; 4035c9e9f97bSIlya Dryomov goto out_bargs; 4036c9e9f97bSIlya Dryomov } 4037c9e9f97bSIlya Dryomov 4038c9e9f97bSIlya Dryomov bctl->fs_info = fs_info; 4039c9e9f97bSIlya Dryomov if (arg) { 4040c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 4041c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 4042c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 4043c9e9f97bSIlya Dryomov 4044c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 4045f43ffb60SIlya Dryomov } else { 4046f43ffb60SIlya Dryomov /* balance everything - no filters */ 4047f43ffb60SIlya Dryomov bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 4048c9e9f97bSIlya Dryomov } 4049c9e9f97bSIlya Dryomov 4050de322263SIlya Dryomov do_balance: 4051c9e9f97bSIlya Dryomov /* 4052ed0fb78fSIlya Dryomov * Ownership of bctl and mutually_exclusive_operation_running 4053ed0fb78fSIlya Dryomov * goes to to btrfs_balance. bctl is freed in __cancel_balance, 4054ed0fb78fSIlya Dryomov * or, if restriper was paused all the way until unmount, in 4055ed0fb78fSIlya Dryomov * free_fs_info. mutually_exclusive_operation_running is 4056ed0fb78fSIlya Dryomov * cleared in __cancel_balance. 4057c9e9f97bSIlya Dryomov */ 4058ed0fb78fSIlya Dryomov need_unlock = false; 4059ed0fb78fSIlya Dryomov 4060ed0fb78fSIlya Dryomov ret = btrfs_balance(bctl, bargs); 4061ed0fb78fSIlya Dryomov 4062c9e9f97bSIlya Dryomov if (arg) { 4063c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 4064c9e9f97bSIlya Dryomov ret = -EFAULT; 4065c9e9f97bSIlya Dryomov } 4066c9e9f97bSIlya Dryomov 4067c9e9f97bSIlya Dryomov out_bargs: 4068c9e9f97bSIlya Dryomov kfree(bargs); 4069ed0fb78fSIlya Dryomov out_unlock: 4070c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4071c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->volume_mutex); 4072ed0fb78fSIlya Dryomov if (need_unlock) 4073ed0fb78fSIlya Dryomov atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 4074ed0fb78fSIlya Dryomov out: 4075e54bfa31SLiu Bo mnt_drop_write_file(file); 4076c9e9f97bSIlya Dryomov return ret; 4077c9e9f97bSIlya Dryomov } 4078c9e9f97bSIlya Dryomov 4079837d5b6eSIlya Dryomov static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd) 4080837d5b6eSIlya Dryomov { 4081837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4082837d5b6eSIlya Dryomov return -EPERM; 4083837d5b6eSIlya Dryomov 4084837d5b6eSIlya Dryomov switch (cmd) { 4085837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 4086837d5b6eSIlya Dryomov return btrfs_pause_balance(root->fs_info); 4087a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 4088a7e99c69SIlya Dryomov return btrfs_cancel_balance(root->fs_info); 4089837d5b6eSIlya Dryomov } 4090837d5b6eSIlya Dryomov 4091837d5b6eSIlya Dryomov return -EINVAL; 4092837d5b6eSIlya Dryomov } 4093837d5b6eSIlya Dryomov 409419a39dceSIlya Dryomov static long btrfs_ioctl_balance_progress(struct btrfs_root *root, 409519a39dceSIlya Dryomov void __user *arg) 409619a39dceSIlya Dryomov { 409719a39dceSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 409819a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 409919a39dceSIlya Dryomov int ret = 0; 410019a39dceSIlya Dryomov 410119a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 410219a39dceSIlya Dryomov return -EPERM; 410319a39dceSIlya Dryomov 410419a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 410519a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 410619a39dceSIlya Dryomov ret = -ENOTCONN; 410719a39dceSIlya Dryomov goto out; 410819a39dceSIlya Dryomov } 410919a39dceSIlya Dryomov 411019a39dceSIlya Dryomov bargs = kzalloc(sizeof(*bargs), GFP_NOFS); 411119a39dceSIlya Dryomov if (!bargs) { 411219a39dceSIlya Dryomov ret = -ENOMEM; 411319a39dceSIlya Dryomov goto out; 411419a39dceSIlya Dryomov } 411519a39dceSIlya Dryomov 411619a39dceSIlya Dryomov update_ioctl_balance_args(fs_info, 1, bargs); 411719a39dceSIlya Dryomov 411819a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 411919a39dceSIlya Dryomov ret = -EFAULT; 412019a39dceSIlya Dryomov 412119a39dceSIlya Dryomov kfree(bargs); 412219a39dceSIlya Dryomov out: 412319a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 412419a39dceSIlya Dryomov return ret; 412519a39dceSIlya Dryomov } 412619a39dceSIlya Dryomov 4127905b0ddaSMiao Xie static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 41285d13a37bSArne Jansen { 4129496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 41305d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 41315d13a37bSArne Jansen struct btrfs_trans_handle *trans = NULL; 41325d13a37bSArne Jansen int ret; 41335d13a37bSArne Jansen int err; 41345d13a37bSArne Jansen 41355d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 41365d13a37bSArne Jansen return -EPERM; 41375d13a37bSArne Jansen 4138905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4139905b0ddaSMiao Xie if (ret) 4140905b0ddaSMiao Xie return ret; 41415d13a37bSArne Jansen 41425d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4143905b0ddaSMiao Xie if (IS_ERR(sa)) { 4144905b0ddaSMiao Xie ret = PTR_ERR(sa); 4145905b0ddaSMiao Xie goto drop_write; 4146905b0ddaSMiao Xie } 41475d13a37bSArne Jansen 41487708f029SWang Shilong down_write(&root->fs_info->subvol_sem); 414992f183aaSWang Shilong trans = btrfs_start_transaction(root->fs_info->tree_root, 2); 41505d13a37bSArne Jansen if (IS_ERR(trans)) { 41515d13a37bSArne Jansen ret = PTR_ERR(trans); 41525d13a37bSArne Jansen goto out; 41535d13a37bSArne Jansen } 41545d13a37bSArne Jansen 41555d13a37bSArne Jansen switch (sa->cmd) { 41565d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 41575d13a37bSArne Jansen ret = btrfs_quota_enable(trans, root->fs_info); 41585d13a37bSArne Jansen break; 41595d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 41605d13a37bSArne Jansen ret = btrfs_quota_disable(trans, root->fs_info); 41615d13a37bSArne Jansen break; 41625d13a37bSArne Jansen default: 41635d13a37bSArne Jansen ret = -EINVAL; 41645d13a37bSArne Jansen break; 41655d13a37bSArne Jansen } 41665d13a37bSArne Jansen 416792f183aaSWang Shilong err = btrfs_commit_transaction(trans, root->fs_info->tree_root); 41685d13a37bSArne Jansen if (err && !ret) 41695d13a37bSArne Jansen ret = err; 41705d13a37bSArne Jansen out: 41715d13a37bSArne Jansen kfree(sa); 41727708f029SWang Shilong up_write(&root->fs_info->subvol_sem); 4173905b0ddaSMiao Xie drop_write: 4174905b0ddaSMiao Xie mnt_drop_write_file(file); 41755d13a37bSArne Jansen return ret; 41765d13a37bSArne Jansen } 41775d13a37bSArne Jansen 4178905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 41795d13a37bSArne Jansen { 4180496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 41815d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 41825d13a37bSArne Jansen struct btrfs_trans_handle *trans; 41835d13a37bSArne Jansen int ret; 41845d13a37bSArne Jansen int err; 41855d13a37bSArne Jansen 41865d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 41875d13a37bSArne Jansen return -EPERM; 41885d13a37bSArne Jansen 4189905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4190905b0ddaSMiao Xie if (ret) 4191905b0ddaSMiao Xie return ret; 41925d13a37bSArne Jansen 41935d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4194905b0ddaSMiao Xie if (IS_ERR(sa)) { 4195905b0ddaSMiao Xie ret = PTR_ERR(sa); 4196905b0ddaSMiao Xie goto drop_write; 4197905b0ddaSMiao Xie } 41985d13a37bSArne Jansen 41995d13a37bSArne Jansen trans = btrfs_join_transaction(root); 42005d13a37bSArne Jansen if (IS_ERR(trans)) { 42015d13a37bSArne Jansen ret = PTR_ERR(trans); 42025d13a37bSArne Jansen goto out; 42035d13a37bSArne Jansen } 42045d13a37bSArne Jansen 42055d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 42065d13a37bSArne Jansen if (sa->assign) { 42075d13a37bSArne Jansen ret = btrfs_add_qgroup_relation(trans, root->fs_info, 42085d13a37bSArne Jansen sa->src, sa->dst); 42095d13a37bSArne Jansen } else { 42105d13a37bSArne Jansen ret = btrfs_del_qgroup_relation(trans, root->fs_info, 42115d13a37bSArne Jansen sa->src, sa->dst); 42125d13a37bSArne Jansen } 42135d13a37bSArne Jansen 42145d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 42155d13a37bSArne Jansen if (err && !ret) 42165d13a37bSArne Jansen ret = err; 42175d13a37bSArne Jansen 42185d13a37bSArne Jansen out: 42195d13a37bSArne Jansen kfree(sa); 4220905b0ddaSMiao Xie drop_write: 4221905b0ddaSMiao Xie mnt_drop_write_file(file); 42225d13a37bSArne Jansen return ret; 42235d13a37bSArne Jansen } 42245d13a37bSArne Jansen 4225905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 42265d13a37bSArne Jansen { 4227496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 42285d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 42295d13a37bSArne Jansen struct btrfs_trans_handle *trans; 42305d13a37bSArne Jansen int ret; 42315d13a37bSArne Jansen int err; 42325d13a37bSArne Jansen 42335d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 42345d13a37bSArne Jansen return -EPERM; 42355d13a37bSArne Jansen 4236905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4237905b0ddaSMiao Xie if (ret) 4238905b0ddaSMiao Xie return ret; 42395d13a37bSArne Jansen 42405d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4241905b0ddaSMiao Xie if (IS_ERR(sa)) { 4242905b0ddaSMiao Xie ret = PTR_ERR(sa); 4243905b0ddaSMiao Xie goto drop_write; 4244905b0ddaSMiao Xie } 42455d13a37bSArne Jansen 4246d86e56cfSMiao Xie if (!sa->qgroupid) { 4247d86e56cfSMiao Xie ret = -EINVAL; 4248d86e56cfSMiao Xie goto out; 4249d86e56cfSMiao Xie } 4250d86e56cfSMiao Xie 42515d13a37bSArne Jansen trans = btrfs_join_transaction(root); 42525d13a37bSArne Jansen if (IS_ERR(trans)) { 42535d13a37bSArne Jansen ret = PTR_ERR(trans); 42545d13a37bSArne Jansen goto out; 42555d13a37bSArne Jansen } 42565d13a37bSArne Jansen 42575d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 42585d13a37bSArne Jansen if (sa->create) { 42595d13a37bSArne Jansen ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid, 42605d13a37bSArne Jansen NULL); 42615d13a37bSArne Jansen } else { 42625d13a37bSArne Jansen ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid); 42635d13a37bSArne Jansen } 42645d13a37bSArne Jansen 42655d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 42665d13a37bSArne Jansen if (err && !ret) 42675d13a37bSArne Jansen ret = err; 42685d13a37bSArne Jansen 42695d13a37bSArne Jansen out: 42705d13a37bSArne Jansen kfree(sa); 4271905b0ddaSMiao Xie drop_write: 4272905b0ddaSMiao Xie mnt_drop_write_file(file); 42735d13a37bSArne Jansen return ret; 42745d13a37bSArne Jansen } 42755d13a37bSArne Jansen 4276905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 42775d13a37bSArne Jansen { 4278496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 42795d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 42805d13a37bSArne Jansen struct btrfs_trans_handle *trans; 42815d13a37bSArne Jansen int ret; 42825d13a37bSArne Jansen int err; 42835d13a37bSArne Jansen u64 qgroupid; 42845d13a37bSArne Jansen 42855d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 42865d13a37bSArne Jansen return -EPERM; 42875d13a37bSArne Jansen 4288905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4289905b0ddaSMiao Xie if (ret) 4290905b0ddaSMiao Xie return ret; 42915d13a37bSArne Jansen 42925d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4293905b0ddaSMiao Xie if (IS_ERR(sa)) { 4294905b0ddaSMiao Xie ret = PTR_ERR(sa); 4295905b0ddaSMiao Xie goto drop_write; 4296905b0ddaSMiao Xie } 42975d13a37bSArne Jansen 42985d13a37bSArne Jansen trans = btrfs_join_transaction(root); 42995d13a37bSArne Jansen if (IS_ERR(trans)) { 43005d13a37bSArne Jansen ret = PTR_ERR(trans); 43015d13a37bSArne Jansen goto out; 43025d13a37bSArne Jansen } 43035d13a37bSArne Jansen 43045d13a37bSArne Jansen qgroupid = sa->qgroupid; 43055d13a37bSArne Jansen if (!qgroupid) { 43065d13a37bSArne Jansen /* take the current subvol as qgroup */ 43075d13a37bSArne Jansen qgroupid = root->root_key.objectid; 43085d13a37bSArne Jansen } 43095d13a37bSArne Jansen 43105d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 43115d13a37bSArne Jansen ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim); 43125d13a37bSArne Jansen 43135d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 43145d13a37bSArne Jansen if (err && !ret) 43155d13a37bSArne Jansen ret = err; 43165d13a37bSArne Jansen 43175d13a37bSArne Jansen out: 43185d13a37bSArne Jansen kfree(sa); 4319905b0ddaSMiao Xie drop_write: 4320905b0ddaSMiao Xie mnt_drop_write_file(file); 43215d13a37bSArne Jansen return ret; 43225d13a37bSArne Jansen } 43235d13a37bSArne Jansen 43242f232036SJan Schmidt static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 43252f232036SJan Schmidt { 43266d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 43272f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 43282f232036SJan Schmidt int ret; 43292f232036SJan Schmidt 43302f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 43312f232036SJan Schmidt return -EPERM; 43322f232036SJan Schmidt 43332f232036SJan Schmidt ret = mnt_want_write_file(file); 43342f232036SJan Schmidt if (ret) 43352f232036SJan Schmidt return ret; 43362f232036SJan Schmidt 43372f232036SJan Schmidt qsa = memdup_user(arg, sizeof(*qsa)); 43382f232036SJan Schmidt if (IS_ERR(qsa)) { 43392f232036SJan Schmidt ret = PTR_ERR(qsa); 43402f232036SJan Schmidt goto drop_write; 43412f232036SJan Schmidt } 43422f232036SJan Schmidt 43432f232036SJan Schmidt if (qsa->flags) { 43442f232036SJan Schmidt ret = -EINVAL; 43452f232036SJan Schmidt goto out; 43462f232036SJan Schmidt } 43472f232036SJan Schmidt 43482f232036SJan Schmidt ret = btrfs_qgroup_rescan(root->fs_info); 43492f232036SJan Schmidt 43502f232036SJan Schmidt out: 43512f232036SJan Schmidt kfree(qsa); 43522f232036SJan Schmidt drop_write: 43532f232036SJan Schmidt mnt_drop_write_file(file); 43542f232036SJan Schmidt return ret; 43552f232036SJan Schmidt } 43562f232036SJan Schmidt 43572f232036SJan Schmidt static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg) 43582f232036SJan Schmidt { 43596d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 43602f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 43612f232036SJan Schmidt int ret = 0; 43622f232036SJan Schmidt 43632f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 43642f232036SJan Schmidt return -EPERM; 43652f232036SJan Schmidt 43662f232036SJan Schmidt qsa = kzalloc(sizeof(*qsa), GFP_NOFS); 43672f232036SJan Schmidt if (!qsa) 43682f232036SJan Schmidt return -ENOMEM; 43692f232036SJan Schmidt 43702f232036SJan Schmidt if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 43712f232036SJan Schmidt qsa->flags = 1; 43722f232036SJan Schmidt qsa->progress = root->fs_info->qgroup_rescan_progress.objectid; 43732f232036SJan Schmidt } 43742f232036SJan Schmidt 43752f232036SJan Schmidt if (copy_to_user(arg, qsa, sizeof(*qsa))) 43762f232036SJan Schmidt ret = -EFAULT; 43772f232036SJan Schmidt 43782f232036SJan Schmidt kfree(qsa); 43792f232036SJan Schmidt return ret; 43802f232036SJan Schmidt } 43812f232036SJan Schmidt 438257254b6eSJan Schmidt static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg) 438357254b6eSJan Schmidt { 438454563d41SAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 438557254b6eSJan Schmidt 438657254b6eSJan Schmidt if (!capable(CAP_SYS_ADMIN)) 438757254b6eSJan Schmidt return -EPERM; 438857254b6eSJan Schmidt 438957254b6eSJan Schmidt return btrfs_qgroup_wait_for_completion(root->fs_info); 439057254b6eSJan Schmidt } 439157254b6eSJan Schmidt 43928ea05e3aSAlexander Block static long btrfs_ioctl_set_received_subvol(struct file *file, 43938ea05e3aSAlexander Block void __user *arg) 43948ea05e3aSAlexander Block { 43958ea05e3aSAlexander Block struct btrfs_ioctl_received_subvol_args *sa = NULL; 4396496ad9aaSAl Viro struct inode *inode = file_inode(file); 43978ea05e3aSAlexander Block struct btrfs_root *root = BTRFS_I(inode)->root; 43988ea05e3aSAlexander Block struct btrfs_root_item *root_item = &root->root_item; 43998ea05e3aSAlexander Block struct btrfs_trans_handle *trans; 44008ea05e3aSAlexander Block struct timespec ct = CURRENT_TIME; 44018ea05e3aSAlexander Block int ret = 0; 4402dd5f9615SStefan Behrens int received_uuid_changed; 44038ea05e3aSAlexander Block 4404bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 4405bd60ea0fSDavid Sterba return -EPERM; 4406bd60ea0fSDavid Sterba 44078ea05e3aSAlexander Block ret = mnt_want_write_file(file); 44088ea05e3aSAlexander Block if (ret < 0) 44098ea05e3aSAlexander Block return ret; 44108ea05e3aSAlexander Block 44118ea05e3aSAlexander Block down_write(&root->fs_info->subvol_sem); 44128ea05e3aSAlexander Block 44138ea05e3aSAlexander Block if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 44148ea05e3aSAlexander Block ret = -EINVAL; 44158ea05e3aSAlexander Block goto out; 44168ea05e3aSAlexander Block } 44178ea05e3aSAlexander Block 44188ea05e3aSAlexander Block if (btrfs_root_readonly(root)) { 44198ea05e3aSAlexander Block ret = -EROFS; 44208ea05e3aSAlexander Block goto out; 44218ea05e3aSAlexander Block } 44228ea05e3aSAlexander Block 44238ea05e3aSAlexander Block sa = memdup_user(arg, sizeof(*sa)); 44248ea05e3aSAlexander Block if (IS_ERR(sa)) { 44258ea05e3aSAlexander Block ret = PTR_ERR(sa); 44268ea05e3aSAlexander Block sa = NULL; 44278ea05e3aSAlexander Block goto out; 44288ea05e3aSAlexander Block } 44298ea05e3aSAlexander Block 4430dd5f9615SStefan Behrens /* 4431dd5f9615SStefan Behrens * 1 - root item 4432dd5f9615SStefan Behrens * 2 - uuid items (received uuid + subvol uuid) 4433dd5f9615SStefan Behrens */ 4434dd5f9615SStefan Behrens trans = btrfs_start_transaction(root, 3); 44358ea05e3aSAlexander Block if (IS_ERR(trans)) { 44368ea05e3aSAlexander Block ret = PTR_ERR(trans); 44378ea05e3aSAlexander Block trans = NULL; 44388ea05e3aSAlexander Block goto out; 44398ea05e3aSAlexander Block } 44408ea05e3aSAlexander Block 44418ea05e3aSAlexander Block sa->rtransid = trans->transid; 44428ea05e3aSAlexander Block sa->rtime.sec = ct.tv_sec; 44438ea05e3aSAlexander Block sa->rtime.nsec = ct.tv_nsec; 44448ea05e3aSAlexander Block 4445dd5f9615SStefan Behrens received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 4446dd5f9615SStefan Behrens BTRFS_UUID_SIZE); 4447dd5f9615SStefan Behrens if (received_uuid_changed && 4448dd5f9615SStefan Behrens !btrfs_is_empty_uuid(root_item->received_uuid)) 4449dd5f9615SStefan Behrens btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 4450dd5f9615SStefan Behrens root_item->received_uuid, 4451dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4452dd5f9615SStefan Behrens root->root_key.objectid); 44538ea05e3aSAlexander Block memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 44548ea05e3aSAlexander Block btrfs_set_root_stransid(root_item, sa->stransid); 44558ea05e3aSAlexander Block btrfs_set_root_rtransid(root_item, sa->rtransid); 44563cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 44573cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 44583cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 44593cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 44608ea05e3aSAlexander Block 44618ea05e3aSAlexander Block ret = btrfs_update_root(trans, root->fs_info->tree_root, 44628ea05e3aSAlexander Block &root->root_key, &root->root_item); 44638ea05e3aSAlexander Block if (ret < 0) { 44648ea05e3aSAlexander Block btrfs_end_transaction(trans, root); 44658ea05e3aSAlexander Block goto out; 4466dd5f9615SStefan Behrens } 4467dd5f9615SStefan Behrens if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 4468dd5f9615SStefan Behrens ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root, 4469dd5f9615SStefan Behrens sa->uuid, 4470dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 4471dd5f9615SStefan Behrens root->root_key.objectid); 4472dd5f9615SStefan Behrens if (ret < 0 && ret != -EEXIST) { 4473dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 4474dd5f9615SStefan Behrens goto out; 4475dd5f9615SStefan Behrens } 4476dd5f9615SStefan Behrens } 44778ea05e3aSAlexander Block ret = btrfs_commit_transaction(trans, root); 4478dd5f9615SStefan Behrens if (ret < 0) { 4479dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 44808ea05e3aSAlexander Block goto out; 44818ea05e3aSAlexander Block } 44828ea05e3aSAlexander Block 44838ea05e3aSAlexander Block ret = copy_to_user(arg, sa, sizeof(*sa)); 44848ea05e3aSAlexander Block if (ret) 44858ea05e3aSAlexander Block ret = -EFAULT; 44868ea05e3aSAlexander Block 44878ea05e3aSAlexander Block out: 44888ea05e3aSAlexander Block kfree(sa); 44898ea05e3aSAlexander Block up_write(&root->fs_info->subvol_sem); 44908ea05e3aSAlexander Block mnt_drop_write_file(file); 44918ea05e3aSAlexander Block return ret; 44928ea05e3aSAlexander Block } 44938ea05e3aSAlexander Block 4494867ab667Sjeff.liu static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg) 4495867ab667Sjeff.liu { 44966d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4497a1b83ac5SAnand Jain size_t len; 4498867ab667Sjeff.liu int ret; 4499a1b83ac5SAnand Jain char label[BTRFS_LABEL_SIZE]; 4500a1b83ac5SAnand Jain 4501a1b83ac5SAnand Jain spin_lock(&root->fs_info->super_lock); 4502a1b83ac5SAnand Jain memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE); 4503a1b83ac5SAnand Jain spin_unlock(&root->fs_info->super_lock); 4504a1b83ac5SAnand Jain 4505a1b83ac5SAnand Jain len = strnlen(label, BTRFS_LABEL_SIZE); 4506867ab667Sjeff.liu 4507867ab667Sjeff.liu if (len == BTRFS_LABEL_SIZE) { 4508efe120a0SFrank Holton btrfs_warn(root->fs_info, 4509efe120a0SFrank Holton "label is too long, return the first %zu bytes", --len); 4510867ab667Sjeff.liu } 4511867ab667Sjeff.liu 4512867ab667Sjeff.liu ret = copy_to_user(arg, label, len); 4513867ab667Sjeff.liu 4514867ab667Sjeff.liu return ret ? -EFAULT : 0; 4515867ab667Sjeff.liu } 4516867ab667Sjeff.liu 4517a8bfd4abSjeff.liu static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 4518a8bfd4abSjeff.liu { 45196d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4520a8bfd4abSjeff.liu struct btrfs_super_block *super_block = root->fs_info->super_copy; 4521a8bfd4abSjeff.liu struct btrfs_trans_handle *trans; 4522a8bfd4abSjeff.liu char label[BTRFS_LABEL_SIZE]; 4523a8bfd4abSjeff.liu int ret; 4524a8bfd4abSjeff.liu 4525a8bfd4abSjeff.liu if (!capable(CAP_SYS_ADMIN)) 4526a8bfd4abSjeff.liu return -EPERM; 4527a8bfd4abSjeff.liu 4528a8bfd4abSjeff.liu if (copy_from_user(label, arg, sizeof(label))) 4529a8bfd4abSjeff.liu return -EFAULT; 4530a8bfd4abSjeff.liu 4531a8bfd4abSjeff.liu if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 4532efe120a0SFrank Holton btrfs_err(root->fs_info, "unable to set label with more than %d bytes", 4533a8bfd4abSjeff.liu BTRFS_LABEL_SIZE - 1); 4534a8bfd4abSjeff.liu return -EINVAL; 4535a8bfd4abSjeff.liu } 4536a8bfd4abSjeff.liu 4537a8bfd4abSjeff.liu ret = mnt_want_write_file(file); 4538a8bfd4abSjeff.liu if (ret) 4539a8bfd4abSjeff.liu return ret; 4540a8bfd4abSjeff.liu 4541a8bfd4abSjeff.liu trans = btrfs_start_transaction(root, 0); 4542a8bfd4abSjeff.liu if (IS_ERR(trans)) { 4543a8bfd4abSjeff.liu ret = PTR_ERR(trans); 4544a8bfd4abSjeff.liu goto out_unlock; 4545a8bfd4abSjeff.liu } 4546a8bfd4abSjeff.liu 4547a1b83ac5SAnand Jain spin_lock(&root->fs_info->super_lock); 4548a8bfd4abSjeff.liu strcpy(super_block->label, label); 4549a1b83ac5SAnand Jain spin_unlock(&root->fs_info->super_lock); 4550a8bfd4abSjeff.liu ret = btrfs_end_transaction(trans, root); 4551a8bfd4abSjeff.liu 4552a8bfd4abSjeff.liu out_unlock: 4553a8bfd4abSjeff.liu mnt_drop_write_file(file); 4554a8bfd4abSjeff.liu return ret; 4555a8bfd4abSjeff.liu } 4556a8bfd4abSjeff.liu 45572eaa055fSJeff Mahoney #define INIT_FEATURE_FLAGS(suffix) \ 45582eaa055fSJeff Mahoney { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 45592eaa055fSJeff Mahoney .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 45602eaa055fSJeff Mahoney .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 45612eaa055fSJeff Mahoney 45622eaa055fSJeff Mahoney static int btrfs_ioctl_get_supported_features(struct file *file, 45632eaa055fSJeff Mahoney void __user *arg) 45642eaa055fSJeff Mahoney { 45652eaa055fSJeff Mahoney static struct btrfs_ioctl_feature_flags features[3] = { 45662eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SUPP), 45672eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_SET), 45682eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_CLEAR) 45692eaa055fSJeff Mahoney }; 45702eaa055fSJeff Mahoney 45712eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 45722eaa055fSJeff Mahoney return -EFAULT; 45732eaa055fSJeff Mahoney 45742eaa055fSJeff Mahoney return 0; 45752eaa055fSJeff Mahoney } 45762eaa055fSJeff Mahoney 45772eaa055fSJeff Mahoney static int btrfs_ioctl_get_features(struct file *file, void __user *arg) 45782eaa055fSJeff Mahoney { 45792eaa055fSJeff Mahoney struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 45802eaa055fSJeff Mahoney struct btrfs_super_block *super_block = root->fs_info->super_copy; 45812eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags features; 45822eaa055fSJeff Mahoney 45832eaa055fSJeff Mahoney features.compat_flags = btrfs_super_compat_flags(super_block); 45842eaa055fSJeff Mahoney features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 45852eaa055fSJeff Mahoney features.incompat_flags = btrfs_super_incompat_flags(super_block); 45862eaa055fSJeff Mahoney 45872eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 45882eaa055fSJeff Mahoney return -EFAULT; 45892eaa055fSJeff Mahoney 45902eaa055fSJeff Mahoney return 0; 45912eaa055fSJeff Mahoney } 45922eaa055fSJeff Mahoney 45933b02a68aSJeff Mahoney static int check_feature_bits(struct btrfs_root *root, 45943b02a68aSJeff Mahoney enum btrfs_feature_set set, 45952eaa055fSJeff Mahoney u64 change_mask, u64 flags, u64 supported_flags, 45962eaa055fSJeff Mahoney u64 safe_set, u64 safe_clear) 45972eaa055fSJeff Mahoney { 45983b02a68aSJeff Mahoney const char *type = btrfs_feature_set_names[set]; 45993b02a68aSJeff Mahoney char *names; 46002eaa055fSJeff Mahoney u64 disallowed, unsupported; 46012eaa055fSJeff Mahoney u64 set_mask = flags & change_mask; 46022eaa055fSJeff Mahoney u64 clear_mask = ~flags & change_mask; 46032eaa055fSJeff Mahoney 46042eaa055fSJeff Mahoney unsupported = set_mask & ~supported_flags; 46052eaa055fSJeff Mahoney if (unsupported) { 46063b02a68aSJeff Mahoney names = btrfs_printable_features(set, unsupported); 46073b02a68aSJeff Mahoney if (names) { 46083b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 46093b02a68aSJeff Mahoney "this kernel does not support the %s feature bit%s", 46103b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 46113b02a68aSJeff Mahoney kfree(names); 46123b02a68aSJeff Mahoney } else 46132eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 46142eaa055fSJeff Mahoney "this kernel does not support %s bits 0x%llx", 46152eaa055fSJeff Mahoney type, unsupported); 46162eaa055fSJeff Mahoney return -EOPNOTSUPP; 46172eaa055fSJeff Mahoney } 46182eaa055fSJeff Mahoney 46192eaa055fSJeff Mahoney disallowed = set_mask & ~safe_set; 46202eaa055fSJeff Mahoney if (disallowed) { 46213b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 46223b02a68aSJeff Mahoney if (names) { 46233b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 46243b02a68aSJeff Mahoney "can't set the %s feature bit%s while mounted", 46253b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 46263b02a68aSJeff Mahoney kfree(names); 46273b02a68aSJeff Mahoney } else 46282eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 46292eaa055fSJeff Mahoney "can't set %s bits 0x%llx while mounted", 46302eaa055fSJeff Mahoney type, disallowed); 46312eaa055fSJeff Mahoney return -EPERM; 46322eaa055fSJeff Mahoney } 46332eaa055fSJeff Mahoney 46342eaa055fSJeff Mahoney disallowed = clear_mask & ~safe_clear; 46352eaa055fSJeff Mahoney if (disallowed) { 46363b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 46373b02a68aSJeff Mahoney if (names) { 46383b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 46393b02a68aSJeff Mahoney "can't clear the %s feature bit%s while mounted", 46403b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 46413b02a68aSJeff Mahoney kfree(names); 46423b02a68aSJeff Mahoney } else 46432eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 46442eaa055fSJeff Mahoney "can't clear %s bits 0x%llx while mounted", 46452eaa055fSJeff Mahoney type, disallowed); 46462eaa055fSJeff Mahoney return -EPERM; 46472eaa055fSJeff Mahoney } 46482eaa055fSJeff Mahoney 46492eaa055fSJeff Mahoney return 0; 46502eaa055fSJeff Mahoney } 46512eaa055fSJeff Mahoney 46522eaa055fSJeff Mahoney #define check_feature(root, change_mask, flags, mask_base) \ 46533b02a68aSJeff Mahoney check_feature_bits(root, FEAT_##mask_base, change_mask, flags, \ 46542eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 46552eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 46562eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 46572eaa055fSJeff Mahoney 46582eaa055fSJeff Mahoney static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 46592eaa055fSJeff Mahoney { 46602eaa055fSJeff Mahoney struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 46612eaa055fSJeff Mahoney struct btrfs_super_block *super_block = root->fs_info->super_copy; 46622eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags flags[2]; 46632eaa055fSJeff Mahoney struct btrfs_trans_handle *trans; 46642eaa055fSJeff Mahoney u64 newflags; 46652eaa055fSJeff Mahoney int ret; 46662eaa055fSJeff Mahoney 46672eaa055fSJeff Mahoney if (!capable(CAP_SYS_ADMIN)) 46682eaa055fSJeff Mahoney return -EPERM; 46692eaa055fSJeff Mahoney 46702eaa055fSJeff Mahoney if (copy_from_user(flags, arg, sizeof(flags))) 46712eaa055fSJeff Mahoney return -EFAULT; 46722eaa055fSJeff Mahoney 46732eaa055fSJeff Mahoney /* Nothing to do */ 46742eaa055fSJeff Mahoney if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 46752eaa055fSJeff Mahoney !flags[0].incompat_flags) 46762eaa055fSJeff Mahoney return 0; 46772eaa055fSJeff Mahoney 46782eaa055fSJeff Mahoney ret = check_feature(root, flags[0].compat_flags, 46792eaa055fSJeff Mahoney flags[1].compat_flags, COMPAT); 46802eaa055fSJeff Mahoney if (ret) 46812eaa055fSJeff Mahoney return ret; 46822eaa055fSJeff Mahoney 46832eaa055fSJeff Mahoney ret = check_feature(root, flags[0].compat_ro_flags, 46842eaa055fSJeff Mahoney flags[1].compat_ro_flags, COMPAT_RO); 46852eaa055fSJeff Mahoney if (ret) 46862eaa055fSJeff Mahoney return ret; 46872eaa055fSJeff Mahoney 46882eaa055fSJeff Mahoney ret = check_feature(root, flags[0].incompat_flags, 46892eaa055fSJeff Mahoney flags[1].incompat_flags, INCOMPAT); 46902eaa055fSJeff Mahoney if (ret) 46912eaa055fSJeff Mahoney return ret; 46922eaa055fSJeff Mahoney 46932eaa055fSJeff Mahoney trans = btrfs_start_transaction(root, 1); 46942eaa055fSJeff Mahoney if (IS_ERR(trans)) 46952eaa055fSJeff Mahoney return PTR_ERR(trans); 46962eaa055fSJeff Mahoney 46972eaa055fSJeff Mahoney spin_lock(&root->fs_info->super_lock); 46982eaa055fSJeff Mahoney newflags = btrfs_super_compat_flags(super_block); 46992eaa055fSJeff Mahoney newflags |= flags[0].compat_flags & flags[1].compat_flags; 47002eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 47012eaa055fSJeff Mahoney btrfs_set_super_compat_flags(super_block, newflags); 47022eaa055fSJeff Mahoney 47032eaa055fSJeff Mahoney newflags = btrfs_super_compat_ro_flags(super_block); 47042eaa055fSJeff Mahoney newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 47052eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 47062eaa055fSJeff Mahoney btrfs_set_super_compat_ro_flags(super_block, newflags); 47072eaa055fSJeff Mahoney 47082eaa055fSJeff Mahoney newflags = btrfs_super_incompat_flags(super_block); 47092eaa055fSJeff Mahoney newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 47102eaa055fSJeff Mahoney newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 47112eaa055fSJeff Mahoney btrfs_set_super_incompat_flags(super_block, newflags); 47122eaa055fSJeff Mahoney spin_unlock(&root->fs_info->super_lock); 47132eaa055fSJeff Mahoney 47142eaa055fSJeff Mahoney return btrfs_end_transaction(trans, root); 47152eaa055fSJeff Mahoney } 47162eaa055fSJeff Mahoney 4717f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 4718f46b5a66SChristoph Hellwig cmd, unsigned long arg) 4719f46b5a66SChristoph Hellwig { 4720496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 47214bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 4722f46b5a66SChristoph Hellwig 4723f46b5a66SChristoph Hellwig switch (cmd) { 47246cbff00fSChristoph Hellwig case FS_IOC_GETFLAGS: 47256cbff00fSChristoph Hellwig return btrfs_ioctl_getflags(file, argp); 47266cbff00fSChristoph Hellwig case FS_IOC_SETFLAGS: 47276cbff00fSChristoph Hellwig return btrfs_ioctl_setflags(file, argp); 47286cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 47296cbff00fSChristoph Hellwig return btrfs_ioctl_getversion(file, argp); 4730f7039b1dSLi Dongyang case FITRIM: 4731f7039b1dSLi Dongyang return btrfs_ioctl_fitrim(file, argp); 4732f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 4733fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 4734fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 4735fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 47363de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 4737fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 47386f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 47396f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 474076dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 474176dda93cSYan, Zheng return btrfs_ioctl_snap_destroy(file, argp); 47420caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 47430caa102dSLi Zefan return btrfs_ioctl_subvol_getflags(file, argp); 47440caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 47450caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 47466ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 47476ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 4748f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 47491e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 47501e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 47511e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 4752f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 4753198605a8SMiao Xie return btrfs_ioctl_resize(file, argp); 4754f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 47554bcabaa3SChristoph Hellwig return btrfs_ioctl_add_dev(root, argp); 4756f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 4757da24927bSMiao Xie return btrfs_ioctl_rm_dev(file, argp); 4758475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 4759475f6387SJan Schmidt return btrfs_ioctl_fs_info(root, argp); 4760475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 4761475f6387SJan Schmidt return btrfs_ioctl_dev_info(root, argp); 4762f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 47639ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, NULL); 4764f46b5a66SChristoph Hellwig case BTRFS_IOC_CLONE: 4765c5c9cd4dSSage Weil return btrfs_ioctl_clone(file, arg, 0, 0, 0); 4766c5c9cd4dSSage Weil case BTRFS_IOC_CLONE_RANGE: 47677a865e8aSChristoph Hellwig return btrfs_ioctl_clone_range(file, argp); 4768f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_START: 4769f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_start(file); 4770f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_END: 4771f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_end(file); 4772ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 4773ac8e9819SChris Mason return btrfs_ioctl_tree_search(file, argp); 4774ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 4775ac8e9819SChris Mason return btrfs_ioctl_ino_lookup(file, argp); 4776d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 4777d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 4778d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 4779d7728c96SJan Schmidt return btrfs_ioctl_logical_to_ino(root, argp); 47801406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 47811406e432SJosef Bacik return btrfs_ioctl_space_info(root, argp); 478201e219e8SJeff Mahoney case BTRFS_IOC_GLOBAL_RSV: 478301e219e8SJeff Mahoney return btrfs_ioctl_global_rsv(root, argp); 47849b199859SFilipe David Borba Manana case BTRFS_IOC_SYNC: { 47859b199859SFilipe David Borba Manana int ret; 47869b199859SFilipe David Borba Manana 478791aef86fSMiao Xie ret = btrfs_start_delalloc_roots(root->fs_info, 0); 47889b199859SFilipe David Borba Manana if (ret) 47899b199859SFilipe David Borba Manana return ret; 47909b199859SFilipe David Borba Manana ret = btrfs_sync_fs(file->f_dentry->d_sb, 1); 47919b199859SFilipe David Borba Manana return ret; 47929b199859SFilipe David Borba Manana } 479346204592SSage Weil case BTRFS_IOC_START_SYNC: 47949a8c28beSMiao Xie return btrfs_ioctl_start_sync(root, argp); 479546204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 47969a8c28beSMiao Xie return btrfs_ioctl_wait_sync(root, argp); 4797475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 4798b8e95489SMiao Xie return btrfs_ioctl_scrub(file, argp); 4799475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 4800475f6387SJan Schmidt return btrfs_ioctl_scrub_cancel(root, argp); 4801475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 4802475f6387SJan Schmidt return btrfs_ioctl_scrub_progress(root, argp); 4803c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 48049ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 4805837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 4806837d5b6eSIlya Dryomov return btrfs_ioctl_balance_ctl(root, arg); 480719a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 480819a39dceSIlya Dryomov return btrfs_ioctl_balance_progress(root, argp); 48098ea05e3aSAlexander Block case BTRFS_IOC_SET_RECEIVED_SUBVOL: 48108ea05e3aSAlexander Block return btrfs_ioctl_set_received_subvol(file, argp); 481131db9f7cSAlexander Block case BTRFS_IOC_SEND: 481231db9f7cSAlexander Block return btrfs_ioctl_send(file, argp); 4813c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 4814b27f7c0cSDavid Sterba return btrfs_ioctl_get_dev_stats(root, argp); 48155d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 4816905b0ddaSMiao Xie return btrfs_ioctl_quota_ctl(file, argp); 48175d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 4818905b0ddaSMiao Xie return btrfs_ioctl_qgroup_assign(file, argp); 48195d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 4820905b0ddaSMiao Xie return btrfs_ioctl_qgroup_create(file, argp); 48215d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 4822905b0ddaSMiao Xie return btrfs_ioctl_qgroup_limit(file, argp); 48232f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN: 48242f232036SJan Schmidt return btrfs_ioctl_quota_rescan(file, argp); 48252f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_STATUS: 48262f232036SJan Schmidt return btrfs_ioctl_quota_rescan_status(file, argp); 482757254b6eSJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_WAIT: 482857254b6eSJan Schmidt return btrfs_ioctl_quota_rescan_wait(file, argp); 48293f6bcfbdSStefan Behrens case BTRFS_IOC_DEV_REPLACE: 48303f6bcfbdSStefan Behrens return btrfs_ioctl_dev_replace(root, argp); 4831867ab667Sjeff.liu case BTRFS_IOC_GET_FSLABEL: 4832867ab667Sjeff.liu return btrfs_ioctl_get_fslabel(file, argp); 4833a8bfd4abSjeff.liu case BTRFS_IOC_SET_FSLABEL: 4834a8bfd4abSjeff.liu return btrfs_ioctl_set_fslabel(file, argp); 4835416161dbSMark Fasheh case BTRFS_IOC_FILE_EXTENT_SAME: 4836416161dbSMark Fasheh return btrfs_ioctl_file_extent_same(file, argp); 48372eaa055fSJeff Mahoney case BTRFS_IOC_GET_SUPPORTED_FEATURES: 48382eaa055fSJeff Mahoney return btrfs_ioctl_get_supported_features(file, argp); 48392eaa055fSJeff Mahoney case BTRFS_IOC_GET_FEATURES: 48402eaa055fSJeff Mahoney return btrfs_ioctl_get_features(file, argp); 48412eaa055fSJeff Mahoney case BTRFS_IOC_SET_FEATURES: 48422eaa055fSJeff Mahoney return btrfs_ioctl_set_features(file, argp); 4843f46b5a66SChristoph Hellwig } 4844f46b5a66SChristoph Hellwig 4845f46b5a66SChristoph Hellwig return -ENOTTY; 4846f46b5a66SChristoph Hellwig } 4847