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" 61fcebe456SJosef Bacik #include "qgroup.h" 621ec9a1aeSFilipe Manana #include "tree-log.h" 63ebb8765bSAnand Jain #include "compression.h" 64f46b5a66SChristoph Hellwig 65abccd00fSHugo Mills #ifdef CONFIG_64BIT 66abccd00fSHugo Mills /* If we have a 32-bit userspace and 64-bit kernel, then the UAPI 67abccd00fSHugo Mills * structures are incorrect, as the timespec structure from userspace 68abccd00fSHugo Mills * is 4 bytes too small. We define these alternatives here to teach 69abccd00fSHugo Mills * the kernel about the 32-bit struct packing. 70abccd00fSHugo Mills */ 71abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 { 72abccd00fSHugo Mills __u64 sec; 73abccd00fSHugo Mills __u32 nsec; 74abccd00fSHugo Mills } __attribute__ ((__packed__)); 75abccd00fSHugo Mills 76abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 { 77abccd00fSHugo Mills char uuid[BTRFS_UUID_SIZE]; /* in */ 78abccd00fSHugo Mills __u64 stransid; /* in */ 79abccd00fSHugo Mills __u64 rtransid; /* out */ 80abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 stime; /* in */ 81abccd00fSHugo Mills struct btrfs_ioctl_timespec_32 rtime; /* out */ 82abccd00fSHugo Mills __u64 flags; /* in */ 83abccd00fSHugo Mills __u64 reserved[16]; /* in */ 84abccd00fSHugo Mills } __attribute__ ((__packed__)); 85abccd00fSHugo Mills 86abccd00fSHugo Mills #define BTRFS_IOC_SET_RECEIVED_SUBVOL_32 _IOWR(BTRFS_IOCTL_MAGIC, 37, \ 87abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32) 88abccd00fSHugo Mills #endif 89abccd00fSHugo Mills 90abccd00fSHugo Mills 91416161dbSMark Fasheh static int btrfs_clone(struct inode *src, struct inode *inode, 921c919a5eSMark Fasheh u64 off, u64 olen, u64 olen_aligned, u64 destoff, 931c919a5eSMark Fasheh int no_time_update); 94416161dbSMark Fasheh 956cbff00fSChristoph Hellwig /* Mask out flags that are inappropriate for the given type of inode. */ 966cbff00fSChristoph Hellwig static inline __u32 btrfs_mask_flags(umode_t mode, __u32 flags) 976cbff00fSChristoph Hellwig { 986cbff00fSChristoph Hellwig if (S_ISDIR(mode)) 996cbff00fSChristoph Hellwig return flags; 1006cbff00fSChristoph Hellwig else if (S_ISREG(mode)) 1016cbff00fSChristoph Hellwig return flags & ~FS_DIRSYNC_FL; 1026cbff00fSChristoph Hellwig else 1036cbff00fSChristoph Hellwig return flags & (FS_NODUMP_FL | FS_NOATIME_FL); 1046cbff00fSChristoph Hellwig } 105f46b5a66SChristoph Hellwig 1066cbff00fSChristoph Hellwig /* 1076cbff00fSChristoph Hellwig * Export inode flags to the format expected by the FS_IOC_GETFLAGS ioctl. 1086cbff00fSChristoph Hellwig */ 1096cbff00fSChristoph Hellwig static unsigned int btrfs_flags_to_ioctl(unsigned int flags) 1106cbff00fSChristoph Hellwig { 1116cbff00fSChristoph Hellwig unsigned int iflags = 0; 1126cbff00fSChristoph Hellwig 1136cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_SYNC) 1146cbff00fSChristoph Hellwig iflags |= FS_SYNC_FL; 1156cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_IMMUTABLE) 1166cbff00fSChristoph Hellwig iflags |= FS_IMMUTABLE_FL; 1176cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_APPEND) 1186cbff00fSChristoph Hellwig iflags |= FS_APPEND_FL; 1196cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NODUMP) 1206cbff00fSChristoph Hellwig iflags |= FS_NODUMP_FL; 1216cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_NOATIME) 1226cbff00fSChristoph Hellwig iflags |= FS_NOATIME_FL; 1236cbff00fSChristoph Hellwig if (flags & BTRFS_INODE_DIRSYNC) 1246cbff00fSChristoph Hellwig iflags |= FS_DIRSYNC_FL; 125d0092bddSLi Zefan if (flags & BTRFS_INODE_NODATACOW) 126d0092bddSLi Zefan iflags |= FS_NOCOW_FL; 127d0092bddSLi Zefan 128d0092bddSLi Zefan if ((flags & BTRFS_INODE_COMPRESS) && !(flags & BTRFS_INODE_NOCOMPRESS)) 129d0092bddSLi Zefan iflags |= FS_COMPR_FL; 130d0092bddSLi Zefan else if (flags & BTRFS_INODE_NOCOMPRESS) 131d0092bddSLi Zefan iflags |= FS_NOCOMP_FL; 1326cbff00fSChristoph Hellwig 1336cbff00fSChristoph Hellwig return iflags; 1346cbff00fSChristoph Hellwig } 1356cbff00fSChristoph Hellwig 1366cbff00fSChristoph Hellwig /* 1376cbff00fSChristoph Hellwig * Update inode->i_flags based on the btrfs internal flags. 1386cbff00fSChristoph Hellwig */ 1396cbff00fSChristoph Hellwig void btrfs_update_iflags(struct inode *inode) 1406cbff00fSChristoph Hellwig { 1416cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 1423cc79392SFilipe Manana unsigned int new_fl = 0; 1436cbff00fSChristoph Hellwig 1446cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_SYNC) 1453cc79392SFilipe Manana new_fl |= S_SYNC; 1466cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_IMMUTABLE) 1473cc79392SFilipe Manana new_fl |= S_IMMUTABLE; 1486cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_APPEND) 1493cc79392SFilipe Manana new_fl |= S_APPEND; 1506cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_NOATIME) 1513cc79392SFilipe Manana new_fl |= S_NOATIME; 1526cbff00fSChristoph Hellwig if (ip->flags & BTRFS_INODE_DIRSYNC) 1533cc79392SFilipe Manana new_fl |= S_DIRSYNC; 1543cc79392SFilipe Manana 1553cc79392SFilipe Manana set_mask_bits(&inode->i_flags, 1563cc79392SFilipe Manana S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC, 1573cc79392SFilipe Manana new_fl); 1586cbff00fSChristoph Hellwig } 1596cbff00fSChristoph Hellwig 1606cbff00fSChristoph Hellwig /* 1616cbff00fSChristoph Hellwig * Inherit flags from the parent inode. 1626cbff00fSChristoph Hellwig * 163e27425d6SJosef Bacik * Currently only the compression flags and the cow flags are inherited. 1646cbff00fSChristoph Hellwig */ 1656cbff00fSChristoph Hellwig void btrfs_inherit_iflags(struct inode *inode, struct inode *dir) 1666cbff00fSChristoph Hellwig { 1670b4dcea5SChris Mason unsigned int flags; 1680b4dcea5SChris Mason 1690b4dcea5SChris Mason if (!dir) 1700b4dcea5SChris Mason return; 1710b4dcea5SChris Mason 1720b4dcea5SChris Mason flags = BTRFS_I(dir)->flags; 1736cbff00fSChristoph Hellwig 174e27425d6SJosef Bacik if (flags & BTRFS_INODE_NOCOMPRESS) { 175e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS; 176e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS; 177e27425d6SJosef Bacik } else if (flags & BTRFS_INODE_COMPRESS) { 178e27425d6SJosef Bacik BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS; 179e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS; 180e27425d6SJosef Bacik } 1816cbff00fSChristoph Hellwig 182213490b3SLiu Bo if (flags & BTRFS_INODE_NODATACOW) { 183e27425d6SJosef Bacik BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW; 184213490b3SLiu Bo if (S_ISREG(inode->i_mode)) 185213490b3SLiu Bo BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM; 186213490b3SLiu Bo } 187e27425d6SJosef Bacik 1886cbff00fSChristoph Hellwig btrfs_update_iflags(inode); 1896cbff00fSChristoph Hellwig } 1906cbff00fSChristoph Hellwig 1916cbff00fSChristoph Hellwig static int btrfs_ioctl_getflags(struct file *file, void __user *arg) 1926cbff00fSChristoph Hellwig { 193496ad9aaSAl Viro struct btrfs_inode *ip = BTRFS_I(file_inode(file)); 1946cbff00fSChristoph Hellwig unsigned int flags = btrfs_flags_to_ioctl(ip->flags); 1956cbff00fSChristoph Hellwig 1966cbff00fSChristoph Hellwig if (copy_to_user(arg, &flags, sizeof(flags))) 1976cbff00fSChristoph Hellwig return -EFAULT; 1986cbff00fSChristoph Hellwig return 0; 1996cbff00fSChristoph Hellwig } 2006cbff00fSChristoph Hellwig 20175e7cb7fSLiu Bo static int check_flags(unsigned int flags) 20275e7cb7fSLiu Bo { 20375e7cb7fSLiu Bo if (flags & ~(FS_IMMUTABLE_FL | FS_APPEND_FL | \ 20475e7cb7fSLiu Bo FS_NOATIME_FL | FS_NODUMP_FL | \ 20575e7cb7fSLiu Bo FS_SYNC_FL | FS_DIRSYNC_FL | \ 206e1e8fb6aSLi Zefan FS_NOCOMP_FL | FS_COMPR_FL | 207e1e8fb6aSLi Zefan FS_NOCOW_FL)) 20875e7cb7fSLiu Bo return -EOPNOTSUPP; 20975e7cb7fSLiu Bo 21075e7cb7fSLiu Bo if ((flags & FS_NOCOMP_FL) && (flags & FS_COMPR_FL)) 21175e7cb7fSLiu Bo return -EINVAL; 21275e7cb7fSLiu Bo 21375e7cb7fSLiu Bo return 0; 21475e7cb7fSLiu Bo } 21575e7cb7fSLiu Bo 2166cbff00fSChristoph Hellwig static int btrfs_ioctl_setflags(struct file *file, void __user *arg) 2176cbff00fSChristoph Hellwig { 218496ad9aaSAl Viro struct inode *inode = file_inode(file); 2196cbff00fSChristoph Hellwig struct btrfs_inode *ip = BTRFS_I(inode); 2206cbff00fSChristoph Hellwig struct btrfs_root *root = ip->root; 2216cbff00fSChristoph Hellwig struct btrfs_trans_handle *trans; 2226cbff00fSChristoph Hellwig unsigned int flags, oldflags; 2236cbff00fSChristoph Hellwig int ret; 224f062abf0SLi Zefan u64 ip_oldflags; 225f062abf0SLi Zefan unsigned int i_oldflags; 2267e97b8daSDavid Sterba umode_t mode; 2276cbff00fSChristoph Hellwig 228bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 229bd60ea0fSDavid Sterba return -EPERM; 230bd60ea0fSDavid Sterba 231b83cc969SLi Zefan if (btrfs_root_readonly(root)) 232b83cc969SLi Zefan return -EROFS; 233b83cc969SLi Zefan 2346cbff00fSChristoph Hellwig if (copy_from_user(&flags, arg, sizeof(flags))) 2356cbff00fSChristoph Hellwig return -EFAULT; 2366cbff00fSChristoph Hellwig 23775e7cb7fSLiu Bo ret = check_flags(flags); 23875e7cb7fSLiu Bo if (ret) 23975e7cb7fSLiu Bo return ret; 2406cbff00fSChristoph Hellwig 241e7848683SJan Kara ret = mnt_want_write_file(file); 242e7848683SJan Kara if (ret) 243e7848683SJan Kara return ret; 244e7848683SJan Kara 2455955102cSAl Viro inode_lock(inode); 2466cbff00fSChristoph Hellwig 247f062abf0SLi Zefan ip_oldflags = ip->flags; 248f062abf0SLi Zefan i_oldflags = inode->i_flags; 2497e97b8daSDavid Sterba mode = inode->i_mode; 250f062abf0SLi Zefan 2516cbff00fSChristoph Hellwig flags = btrfs_mask_flags(inode->i_mode, flags); 2526cbff00fSChristoph Hellwig oldflags = btrfs_flags_to_ioctl(ip->flags); 2536cbff00fSChristoph Hellwig if ((flags ^ oldflags) & (FS_APPEND_FL | FS_IMMUTABLE_FL)) { 2546cbff00fSChristoph Hellwig if (!capable(CAP_LINUX_IMMUTABLE)) { 2556cbff00fSChristoph Hellwig ret = -EPERM; 2566cbff00fSChristoph Hellwig goto out_unlock; 2576cbff00fSChristoph Hellwig } 2586cbff00fSChristoph Hellwig } 2596cbff00fSChristoph Hellwig 2606cbff00fSChristoph Hellwig if (flags & FS_SYNC_FL) 2616cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_SYNC; 2626cbff00fSChristoph Hellwig else 2636cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_SYNC; 2646cbff00fSChristoph Hellwig if (flags & FS_IMMUTABLE_FL) 2656cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_IMMUTABLE; 2666cbff00fSChristoph Hellwig else 2676cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_IMMUTABLE; 2686cbff00fSChristoph Hellwig if (flags & FS_APPEND_FL) 2696cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_APPEND; 2706cbff00fSChristoph Hellwig else 2716cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_APPEND; 2726cbff00fSChristoph Hellwig if (flags & FS_NODUMP_FL) 2736cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NODUMP; 2746cbff00fSChristoph Hellwig else 2756cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NODUMP; 2766cbff00fSChristoph Hellwig if (flags & FS_NOATIME_FL) 2776cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_NOATIME; 2786cbff00fSChristoph Hellwig else 2796cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_NOATIME; 2806cbff00fSChristoph Hellwig if (flags & FS_DIRSYNC_FL) 2816cbff00fSChristoph Hellwig ip->flags |= BTRFS_INODE_DIRSYNC; 2826cbff00fSChristoph Hellwig else 2836cbff00fSChristoph Hellwig ip->flags &= ~BTRFS_INODE_DIRSYNC; 2847e97b8daSDavid Sterba if (flags & FS_NOCOW_FL) { 2857e97b8daSDavid Sterba if (S_ISREG(mode)) { 2867e97b8daSDavid Sterba /* 2877e97b8daSDavid Sterba * It's safe to turn csums off here, no extents exist. 2887e97b8daSDavid Sterba * Otherwise we want the flag to reflect the real COW 2897e97b8daSDavid Sterba * status of the file and will not set it. 2907e97b8daSDavid Sterba */ 2917e97b8daSDavid Sterba if (inode->i_size == 0) 2927e97b8daSDavid Sterba ip->flags |= BTRFS_INODE_NODATACOW 2937e97b8daSDavid Sterba | BTRFS_INODE_NODATASUM; 2947e97b8daSDavid Sterba } else { 295e1e8fb6aSLi Zefan ip->flags |= BTRFS_INODE_NODATACOW; 2967e97b8daSDavid Sterba } 2977e97b8daSDavid Sterba } else { 2987e97b8daSDavid Sterba /* 2997e97b8daSDavid Sterba * Revert back under same assuptions as above 3007e97b8daSDavid Sterba */ 3017e97b8daSDavid Sterba if (S_ISREG(mode)) { 3027e97b8daSDavid Sterba if (inode->i_size == 0) 3037e97b8daSDavid Sterba ip->flags &= ~(BTRFS_INODE_NODATACOW 3047e97b8daSDavid Sterba | BTRFS_INODE_NODATASUM); 3057e97b8daSDavid Sterba } else { 306e1e8fb6aSLi Zefan ip->flags &= ~BTRFS_INODE_NODATACOW; 3077e97b8daSDavid Sterba } 3087e97b8daSDavid Sterba } 3096cbff00fSChristoph Hellwig 31075e7cb7fSLiu Bo /* 31175e7cb7fSLiu Bo * The COMPRESS flag can only be changed by users, while the NOCOMPRESS 31275e7cb7fSLiu Bo * flag may be changed automatically if compression code won't make 31375e7cb7fSLiu Bo * things smaller. 31475e7cb7fSLiu Bo */ 31575e7cb7fSLiu Bo if (flags & FS_NOCOMP_FL) { 31675e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_COMPRESS; 31775e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_NOCOMPRESS; 31863541927SFilipe David Borba Manana 31963541927SFilipe David Borba Manana ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0); 32063541927SFilipe David Borba Manana if (ret && ret != -ENODATA) 32163541927SFilipe David Borba Manana goto out_drop; 32275e7cb7fSLiu Bo } else if (flags & FS_COMPR_FL) { 32363541927SFilipe David Borba Manana const char *comp; 32463541927SFilipe David Borba Manana 32575e7cb7fSLiu Bo ip->flags |= BTRFS_INODE_COMPRESS; 32675e7cb7fSLiu Bo ip->flags &= ~BTRFS_INODE_NOCOMPRESS; 32763541927SFilipe David Borba Manana 32863541927SFilipe David Borba Manana if (root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 32963541927SFilipe David Borba Manana comp = "lzo"; 33063541927SFilipe David Borba Manana else 33163541927SFilipe David Borba Manana comp = "zlib"; 33263541927SFilipe David Borba Manana ret = btrfs_set_prop(inode, "btrfs.compression", 33363541927SFilipe David Borba Manana comp, strlen(comp), 0); 33463541927SFilipe David Borba Manana if (ret) 33563541927SFilipe David Borba Manana goto out_drop; 33663541927SFilipe David Borba Manana 337ebcb904dSLi Zefan } else { 33878a017a2SFilipe Manana ret = btrfs_set_prop(inode, "btrfs.compression", NULL, 0, 0); 33978a017a2SFilipe Manana if (ret && ret != -ENODATA) 34078a017a2SFilipe Manana goto out_drop; 341ebcb904dSLi Zefan ip->flags &= ~(BTRFS_INODE_COMPRESS | BTRFS_INODE_NOCOMPRESS); 34275e7cb7fSLiu Bo } 3436cbff00fSChristoph Hellwig 3444da6f1a3SLi Zefan trans = btrfs_start_transaction(root, 1); 345f062abf0SLi Zefan if (IS_ERR(trans)) { 346f062abf0SLi Zefan ret = PTR_ERR(trans); 347f062abf0SLi Zefan goto out_drop; 348f062abf0SLi Zefan } 3496cbff00fSChristoph Hellwig 350306424ccSLi Zefan btrfs_update_iflags(inode); 3510c4d2d95SJosef Bacik inode_inc_iversion(inode); 35204b285f3SDeepa Dinamani inode->i_ctime = current_fs_time(inode->i_sb); 3536cbff00fSChristoph Hellwig ret = btrfs_update_inode(trans, root, inode); 3546cbff00fSChristoph Hellwig 3556cbff00fSChristoph Hellwig btrfs_end_transaction(trans, root); 356f062abf0SLi Zefan out_drop: 357f062abf0SLi Zefan if (ret) { 358f062abf0SLi Zefan ip->flags = ip_oldflags; 359f062abf0SLi Zefan inode->i_flags = i_oldflags; 360f062abf0SLi Zefan } 3616cbff00fSChristoph Hellwig 3626cbff00fSChristoph Hellwig out_unlock: 3635955102cSAl Viro inode_unlock(inode); 364e7848683SJan Kara mnt_drop_write_file(file); 3652d4e6f6aSliubo return ret; 3666cbff00fSChristoph Hellwig } 3676cbff00fSChristoph Hellwig 3686cbff00fSChristoph Hellwig static int btrfs_ioctl_getversion(struct file *file, int __user *arg) 3696cbff00fSChristoph Hellwig { 370496ad9aaSAl Viro struct inode *inode = file_inode(file); 3716cbff00fSChristoph Hellwig 3726cbff00fSChristoph Hellwig return put_user(inode->i_generation, arg); 3736cbff00fSChristoph Hellwig } 374f46b5a66SChristoph Hellwig 375f7039b1dSLi Dongyang static noinline int btrfs_ioctl_fitrim(struct file *file, void __user *arg) 376f7039b1dSLi Dongyang { 37754563d41SAl Viro struct btrfs_fs_info *fs_info = btrfs_sb(file_inode(file)->i_sb); 378f7039b1dSLi Dongyang struct btrfs_device *device; 379f7039b1dSLi Dongyang struct request_queue *q; 380f7039b1dSLi Dongyang struct fstrim_range range; 381f7039b1dSLi Dongyang u64 minlen = ULLONG_MAX; 382f7039b1dSLi Dongyang u64 num_devices = 0; 383815745cfSAl Viro u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); 384f7039b1dSLi Dongyang int ret; 385f7039b1dSLi Dongyang 386f7039b1dSLi Dongyang if (!capable(CAP_SYS_ADMIN)) 387f7039b1dSLi Dongyang return -EPERM; 388f7039b1dSLi Dongyang 3891f78160cSXiao Guangrong rcu_read_lock(); 3901f78160cSXiao Guangrong list_for_each_entry_rcu(device, &fs_info->fs_devices->devices, 3911f78160cSXiao Guangrong dev_list) { 392f7039b1dSLi Dongyang if (!device->bdev) 393f7039b1dSLi Dongyang continue; 394f7039b1dSLi Dongyang q = bdev_get_queue(device->bdev); 395f7039b1dSLi Dongyang if (blk_queue_discard(q)) { 396f7039b1dSLi Dongyang num_devices++; 397f7039b1dSLi Dongyang minlen = min((u64)q->limits.discard_granularity, 398f7039b1dSLi Dongyang minlen); 399f7039b1dSLi Dongyang } 400f7039b1dSLi Dongyang } 4011f78160cSXiao Guangrong rcu_read_unlock(); 402f4c697e6SLukas Czerner 403f7039b1dSLi Dongyang if (!num_devices) 404f7039b1dSLi Dongyang return -EOPNOTSUPP; 405f7039b1dSLi Dongyang if (copy_from_user(&range, arg, sizeof(range))) 406f7039b1dSLi Dongyang return -EFAULT; 407e515c18bSLukas Czerner if (range.start > total_bytes || 408e515c18bSLukas Czerner range.len < fs_info->sb->s_blocksize) 409f4c697e6SLukas Czerner return -EINVAL; 410f7039b1dSLi Dongyang 411f4c697e6SLukas Czerner range.len = min(range.len, total_bytes - range.start); 412f7039b1dSLi Dongyang range.minlen = max(range.minlen, minlen); 413815745cfSAl Viro ret = btrfs_trim_fs(fs_info->tree_root, &range); 414f7039b1dSLi Dongyang if (ret < 0) 415f7039b1dSLi Dongyang return ret; 416f7039b1dSLi Dongyang 417f7039b1dSLi Dongyang if (copy_to_user(arg, &range, sizeof(range))) 418f7039b1dSLi Dongyang return -EFAULT; 419f7039b1dSLi Dongyang 420f7039b1dSLi Dongyang return 0; 421f7039b1dSLi Dongyang } 422f7039b1dSLi Dongyang 423dd5f9615SStefan Behrens int btrfs_is_empty_uuid(u8 *uuid) 424dd5f9615SStefan Behrens { 42546e0f66aSChris Mason int i; 42646e0f66aSChris Mason 42746e0f66aSChris Mason for (i = 0; i < BTRFS_UUID_SIZE; i++) { 42846e0f66aSChris Mason if (uuid[i]) 42946e0f66aSChris Mason return 0; 43046e0f66aSChris Mason } 43146e0f66aSChris Mason return 1; 432dd5f9615SStefan Behrens } 433dd5f9615SStefan Behrens 434d5c12070SMiao Xie static noinline int create_subvol(struct inode *dir, 435cb8e7090SChristoph Hellwig struct dentry *dentry, 43672fd032eSSage Weil char *name, int namelen, 4376f72c7e2SArne Jansen u64 *async_transid, 4388696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 439f46b5a66SChristoph Hellwig { 440f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 441f46b5a66SChristoph Hellwig struct btrfs_key key; 44249a3c4d9SDavid Sterba struct btrfs_root_item *root_item; 443f46b5a66SChristoph Hellwig struct btrfs_inode_item *inode_item; 444f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 445d5c12070SMiao Xie struct btrfs_root *root = BTRFS_I(dir)->root; 44676dda93cSYan, Zheng struct btrfs_root *new_root; 447d5c12070SMiao Xie struct btrfs_block_rsv block_rsv; 44804b285f3SDeepa Dinamani struct timespec cur_time = current_fs_time(dir->i_sb); 4495662344bSTsutomu Itoh struct inode *inode; 450f46b5a66SChristoph Hellwig int ret; 451f46b5a66SChristoph Hellwig int err; 452f46b5a66SChristoph Hellwig u64 objectid; 453f46b5a66SChristoph Hellwig u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID; 4543de4586cSChris Mason u64 index = 0; 455d5c12070SMiao Xie u64 qgroup_reserved; 4568ea05e3aSAlexander Block uuid_le new_uuid; 457f46b5a66SChristoph Hellwig 45849a3c4d9SDavid Sterba root_item = kzalloc(sizeof(*root_item), GFP_KERNEL); 45949a3c4d9SDavid Sterba if (!root_item) 46049a3c4d9SDavid Sterba return -ENOMEM; 46149a3c4d9SDavid Sterba 462581bb050SLi Zefan ret = btrfs_find_free_objectid(root->fs_info->tree_root, &objectid); 4632fbe8c8aSAl Viro if (ret) 46449a3c4d9SDavid Sterba goto fail_free; 4656a912213SJosef Bacik 466e09fe2d2SQu Wenruo /* 467e09fe2d2SQu Wenruo * Don't create subvolume whose level is not zero. Or qgroup will be 468e09fe2d2SQu Wenruo * screwed up since it assume subvolme qgroup's level to be 0. 469e09fe2d2SQu Wenruo */ 47049a3c4d9SDavid Sterba if (btrfs_qgroup_level(objectid)) { 47149a3c4d9SDavid Sterba ret = -ENOSPC; 47249a3c4d9SDavid Sterba goto fail_free; 47349a3c4d9SDavid Sterba } 474e09fe2d2SQu Wenruo 475d5c12070SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 4769ed74f2dSJosef Bacik /* 477d5c12070SMiao Xie * The same as the snapshot creation, please see the comment 478d5c12070SMiao Xie * of create_snapshot(). 4799ed74f2dSJosef Bacik */ 480d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 481dd5f9615SStefan Behrens 8, &qgroup_reserved, false); 482d5c12070SMiao Xie if (ret) 48349a3c4d9SDavid Sterba goto fail_free; 484f46b5a66SChristoph Hellwig 485d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 486d5c12070SMiao Xie if (IS_ERR(trans)) { 487d5c12070SMiao Xie ret = PTR_ERR(trans); 488de6e8200SLiu Bo btrfs_subvolume_release_metadata(root, &block_rsv, 489de6e8200SLiu Bo qgroup_reserved); 49049a3c4d9SDavid Sterba goto fail_free; 491d5c12070SMiao Xie } 492d5c12070SMiao Xie trans->block_rsv = &block_rsv; 493d5c12070SMiao Xie trans->bytes_reserved = block_rsv.size; 494f46b5a66SChristoph Hellwig 4958696c533SMiao Xie ret = btrfs_qgroup_inherit(trans, root->fs_info, 0, objectid, inherit); 4966f72c7e2SArne Jansen if (ret) 4976f72c7e2SArne Jansen goto fail; 4986f72c7e2SArne Jansen 4994d75f8a9SDavid Sterba leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0); 5008e8a1e31SJosef Bacik if (IS_ERR(leaf)) { 5018e8a1e31SJosef Bacik ret = PTR_ERR(leaf); 5028e8a1e31SJosef Bacik goto fail; 5038e8a1e31SJosef Bacik } 504f46b5a66SChristoph Hellwig 5055d4f98a2SYan Zheng memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 506f46b5a66SChristoph Hellwig btrfs_set_header_bytenr(leaf, leaf->start); 507f46b5a66SChristoph Hellwig btrfs_set_header_generation(leaf, trans->transid); 5085d4f98a2SYan Zheng btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 509f46b5a66SChristoph Hellwig btrfs_set_header_owner(leaf, objectid); 510f46b5a66SChristoph Hellwig 5110a4e5586SRoss Kirk write_extent_buffer(leaf, root->fs_info->fsid, btrfs_header_fsid(), 512f46b5a66SChristoph Hellwig BTRFS_FSID_SIZE); 5135d4f98a2SYan Zheng write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, 514b308bc2fSGeert Uytterhoeven btrfs_header_chunk_tree_uuid(leaf), 5155d4f98a2SYan Zheng BTRFS_UUID_SIZE); 516f46b5a66SChristoph Hellwig btrfs_mark_buffer_dirty(leaf); 517f46b5a66SChristoph Hellwig 51849a3c4d9SDavid Sterba inode_item = &root_item->inode; 5193cae210fSQu Wenruo btrfs_set_stack_inode_generation(inode_item, 1); 5203cae210fSQu Wenruo btrfs_set_stack_inode_size(inode_item, 3); 5213cae210fSQu Wenruo btrfs_set_stack_inode_nlink(inode_item, 1); 522707e8a07SDavid Sterba btrfs_set_stack_inode_nbytes(inode_item, root->nodesize); 5233cae210fSQu Wenruo btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 524f46b5a66SChristoph Hellwig 52549a3c4d9SDavid Sterba btrfs_set_root_flags(root_item, 0); 52649a3c4d9SDavid Sterba btrfs_set_root_limit(root_item, 0); 5273cae210fSQu Wenruo btrfs_set_stack_inode_flags(inode_item, BTRFS_INODE_ROOT_ITEM_INIT); 52808fe4db1SLi Zefan 52949a3c4d9SDavid Sterba btrfs_set_root_bytenr(root_item, leaf->start); 53049a3c4d9SDavid Sterba btrfs_set_root_generation(root_item, trans->transid); 53149a3c4d9SDavid Sterba btrfs_set_root_level(root_item, 0); 53249a3c4d9SDavid Sterba btrfs_set_root_refs(root_item, 1); 53349a3c4d9SDavid Sterba btrfs_set_root_used(root_item, leaf->len); 53449a3c4d9SDavid Sterba btrfs_set_root_last_snapshot(root_item, 0); 535f46b5a66SChristoph Hellwig 53649a3c4d9SDavid Sterba btrfs_set_root_generation_v2(root_item, 53749a3c4d9SDavid Sterba btrfs_root_generation(root_item)); 5388ea05e3aSAlexander Block uuid_le_gen(&new_uuid); 53949a3c4d9SDavid Sterba memcpy(root_item->uuid, new_uuid.b, BTRFS_UUID_SIZE); 54049a3c4d9SDavid Sterba btrfs_set_stack_timespec_sec(&root_item->otime, cur_time.tv_sec); 54149a3c4d9SDavid Sterba btrfs_set_stack_timespec_nsec(&root_item->otime, cur_time.tv_nsec); 54249a3c4d9SDavid Sterba root_item->ctime = root_item->otime; 54349a3c4d9SDavid Sterba btrfs_set_root_ctransid(root_item, trans->transid); 54449a3c4d9SDavid Sterba btrfs_set_root_otransid(root_item, trans->transid); 545f46b5a66SChristoph Hellwig 546925baeddSChris Mason btrfs_tree_unlock(leaf); 547f46b5a66SChristoph Hellwig free_extent_buffer(leaf); 548f46b5a66SChristoph Hellwig leaf = NULL; 549f46b5a66SChristoph Hellwig 55049a3c4d9SDavid Sterba btrfs_set_root_dirid(root_item, new_dirid); 551f46b5a66SChristoph Hellwig 552f46b5a66SChristoph Hellwig key.objectid = objectid; 5535d4f98a2SYan Zheng key.offset = 0; 554962a298fSDavid Sterba key.type = BTRFS_ROOT_ITEM_KEY; 555f46b5a66SChristoph Hellwig ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key, 55649a3c4d9SDavid Sterba root_item); 557f46b5a66SChristoph Hellwig if (ret) 558f46b5a66SChristoph Hellwig goto fail; 559f46b5a66SChristoph Hellwig 56076dda93cSYan, Zheng key.offset = (u64)-1; 56176dda93cSYan, Zheng new_root = btrfs_read_fs_root_no_name(root->fs_info, &key); 56279787eaaSJeff Mahoney if (IS_ERR(new_root)) { 56379787eaaSJeff Mahoney ret = PTR_ERR(new_root); 5646d13f549SDavid Sterba btrfs_abort_transaction(trans, root, ret); 56579787eaaSJeff Mahoney goto fail; 56679787eaaSJeff Mahoney } 56776dda93cSYan, Zheng 56876dda93cSYan, Zheng btrfs_record_root_in_trans(trans, new_root); 56976dda93cSYan, Zheng 57063541927SFilipe David Borba Manana ret = btrfs_create_subvol_root(trans, new_root, root, new_dirid); 571ce598979SMark Fasheh if (ret) { 572ce598979SMark Fasheh /* We potentially lose an unused inode item here */ 57379787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 574ce598979SMark Fasheh goto fail; 575ce598979SMark Fasheh } 576ce598979SMark Fasheh 577f32e48e9SChandan Rajendra mutex_lock(&new_root->objectid_mutex); 578f32e48e9SChandan Rajendra new_root->highest_objectid = new_dirid; 579f32e48e9SChandan Rajendra mutex_unlock(&new_root->objectid_mutex); 580f32e48e9SChandan Rajendra 581f46b5a66SChristoph Hellwig /* 582f46b5a66SChristoph Hellwig * insert the directory item 583f46b5a66SChristoph Hellwig */ 5843de4586cSChris Mason ret = btrfs_set_inode_index(dir, &index); 58579787eaaSJeff Mahoney if (ret) { 58679787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 58779787eaaSJeff Mahoney goto fail; 58879787eaaSJeff Mahoney } 5893de4586cSChris Mason 5903de4586cSChris Mason ret = btrfs_insert_dir_item(trans, root, 59116cdcec7SMiao Xie name, namelen, dir, &key, 5923de4586cSChris Mason BTRFS_FT_DIR, index); 59379787eaaSJeff Mahoney if (ret) { 59479787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 595f46b5a66SChristoph Hellwig goto fail; 59679787eaaSJeff Mahoney } 5970660b5afSChris Mason 59852c26179SYan Zheng btrfs_i_size_write(dir, dir->i_size + namelen * 2); 59952c26179SYan Zheng ret = btrfs_update_inode(trans, root, dir); 60052c26179SYan Zheng BUG_ON(ret); 60152c26179SYan Zheng 6020660b5afSChris Mason ret = btrfs_add_root_ref(trans, root->fs_info->tree_root, 6034df27c4dSYan, Zheng objectid, root->root_key.objectid, 60433345d01SLi Zefan btrfs_ino(dir), index, name, namelen); 6050660b5afSChris Mason BUG_ON(ret); 6060660b5afSChris Mason 607dd5f9615SStefan Behrens ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root, 60849a3c4d9SDavid Sterba root_item->uuid, BTRFS_UUID_KEY_SUBVOL, 609dd5f9615SStefan Behrens objectid); 610dd5f9615SStefan Behrens if (ret) 611dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 612dd5f9615SStefan Behrens 613f46b5a66SChristoph Hellwig fail: 61449a3c4d9SDavid Sterba kfree(root_item); 615d5c12070SMiao Xie trans->block_rsv = NULL; 616d5c12070SMiao Xie trans->bytes_reserved = 0; 617de6e8200SLiu Bo btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved); 618de6e8200SLiu Bo 61972fd032eSSage Weil if (async_transid) { 62072fd032eSSage Weil *async_transid = trans->transid; 62172fd032eSSage Weil err = btrfs_commit_transaction_async(trans, root, 1); 62200d71c9cSMiao Xie if (err) 62300d71c9cSMiao Xie err = btrfs_commit_transaction(trans, root); 62472fd032eSSage Weil } else { 62576dda93cSYan, Zheng err = btrfs_commit_transaction(trans, root); 62672fd032eSSage Weil } 627f46b5a66SChristoph Hellwig if (err && !ret) 628f46b5a66SChristoph Hellwig ret = err; 6291a65e24bSChris Mason 6305662344bSTsutomu Itoh if (!ret) { 6315662344bSTsutomu Itoh inode = btrfs_lookup_dentry(dir, dentry); 632de6e8200SLiu Bo if (IS_ERR(inode)) 633de6e8200SLiu Bo return PTR_ERR(inode); 6345662344bSTsutomu Itoh d_instantiate(dentry, inode); 6355662344bSTsutomu Itoh } 636f46b5a66SChristoph Hellwig return ret; 63749a3c4d9SDavid Sterba 63849a3c4d9SDavid Sterba fail_free: 63949a3c4d9SDavid Sterba kfree(root_item); 64049a3c4d9SDavid Sterba return ret; 641f46b5a66SChristoph Hellwig } 642f46b5a66SChristoph Hellwig 6439ea24bbeSFilipe Manana static void btrfs_wait_for_no_snapshoting_writes(struct btrfs_root *root) 6448257b2dcSMiao Xie { 6458257b2dcSMiao Xie s64 writers; 6468257b2dcSMiao Xie DEFINE_WAIT(wait); 6478257b2dcSMiao Xie 6488257b2dcSMiao Xie do { 6498257b2dcSMiao Xie prepare_to_wait(&root->subv_writers->wait, &wait, 6508257b2dcSMiao Xie TASK_UNINTERRUPTIBLE); 6518257b2dcSMiao Xie 6528257b2dcSMiao Xie writers = percpu_counter_sum(&root->subv_writers->counter); 6538257b2dcSMiao Xie if (writers) 6548257b2dcSMiao Xie schedule(); 6558257b2dcSMiao Xie 6568257b2dcSMiao Xie finish_wait(&root->subv_writers->wait, &wait); 6578257b2dcSMiao Xie } while (writers); 6588257b2dcSMiao Xie } 6598257b2dcSMiao Xie 660e9662f70SMiao Xie static int create_snapshot(struct btrfs_root *root, struct inode *dir, 661e9662f70SMiao Xie struct dentry *dentry, char *name, int namelen, 662e9662f70SMiao Xie u64 *async_transid, bool readonly, 663e9662f70SMiao Xie struct btrfs_qgroup_inherit *inherit) 664f46b5a66SChristoph Hellwig { 6652e4bfab9SYan, Zheng struct inode *inode; 666f46b5a66SChristoph Hellwig struct btrfs_pending_snapshot *pending_snapshot; 667f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 6682e4bfab9SYan, Zheng int ret; 669f46b5a66SChristoph Hellwig 67027cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 671f46b5a66SChristoph Hellwig return -EINVAL; 672f46b5a66SChristoph Hellwig 673a1ee7362SDavid Sterba pending_snapshot = kzalloc(sizeof(*pending_snapshot), GFP_NOFS); 674a1ee7362SDavid Sterba if (!pending_snapshot) 675a1ee7362SDavid Sterba return -ENOMEM; 676a1ee7362SDavid Sterba 677b0c0ea63SDavid Sterba pending_snapshot->root_item = kzalloc(sizeof(struct btrfs_root_item), 678b0c0ea63SDavid Sterba GFP_NOFS); 6798546b570SDavid Sterba pending_snapshot->path = btrfs_alloc_path(); 6808546b570SDavid Sterba if (!pending_snapshot->root_item || !pending_snapshot->path) { 681b0c0ea63SDavid Sterba ret = -ENOMEM; 682b0c0ea63SDavid Sterba goto free_pending; 683b0c0ea63SDavid Sterba } 684b0c0ea63SDavid Sterba 6858257b2dcSMiao Xie atomic_inc(&root->will_be_snapshoted); 6864e857c58SPeter Zijlstra smp_mb__after_atomic(); 6879ea24bbeSFilipe Manana btrfs_wait_for_no_snapshoting_writes(root); 6888257b2dcSMiao Xie 6896a03843dSMiao Xie ret = btrfs_start_delalloc_inodes(root, 0); 6906a03843dSMiao Xie if (ret) 691a1ee7362SDavid Sterba goto dec_and_free; 6926a03843dSMiao Xie 693b0244199SMiao Xie btrfs_wait_ordered_extents(root, -1); 6946a03843dSMiao Xie 69566d8f3ddSMiao Xie btrfs_init_block_rsv(&pending_snapshot->block_rsv, 69666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 697d5c12070SMiao Xie /* 698d5c12070SMiao Xie * 1 - parent dir inode 699d5c12070SMiao Xie * 2 - dir entries 700d5c12070SMiao Xie * 1 - root item 701d5c12070SMiao Xie * 2 - root ref/backref 702d5c12070SMiao Xie * 1 - root of snapshot 703dd5f9615SStefan Behrens * 1 - UUID item 704d5c12070SMiao Xie */ 705d5c12070SMiao Xie ret = btrfs_subvolume_reserve_metadata(BTRFS_I(dir)->root, 706dd5f9615SStefan Behrens &pending_snapshot->block_rsv, 8, 707ee3441b4SJeff Mahoney &pending_snapshot->qgroup_reserved, 708ee3441b4SJeff Mahoney false); 709d5c12070SMiao Xie if (ret) 710a1ee7362SDavid Sterba goto dec_and_free; 711d5c12070SMiao Xie 712a22285a6SYan, Zheng pending_snapshot->dentry = dentry; 713a22285a6SYan, Zheng pending_snapshot->root = root; 714b83cc969SLi Zefan pending_snapshot->readonly = readonly; 715e9662f70SMiao Xie pending_snapshot->dir = dir; 7168696c533SMiao Xie pending_snapshot->inherit = inherit; 717a22285a6SYan, Zheng 718d5c12070SMiao Xie trans = btrfs_start_transaction(root, 0); 719a22285a6SYan, Zheng if (IS_ERR(trans)) { 720a22285a6SYan, Zheng ret = PTR_ERR(trans); 721a22285a6SYan, Zheng goto fail; 722a22285a6SYan, Zheng } 723a22285a6SYan, Zheng 7248351583eSJosef Bacik spin_lock(&root->fs_info->trans_lock); 725a22285a6SYan, Zheng list_add(&pending_snapshot->list, 726a22285a6SYan, Zheng &trans->transaction->pending_snapshots); 7278351583eSJosef Bacik spin_unlock(&root->fs_info->trans_lock); 72872fd032eSSage Weil if (async_transid) { 72972fd032eSSage Weil *async_transid = trans->transid; 73072fd032eSSage Weil ret = btrfs_commit_transaction_async(trans, 73172fd032eSSage Weil root->fs_info->extent_root, 1); 73200d71c9cSMiao Xie if (ret) 73300d71c9cSMiao Xie ret = btrfs_commit_transaction(trans, root); 73472fd032eSSage Weil } else { 73572fd032eSSage Weil ret = btrfs_commit_transaction(trans, 73672fd032eSSage Weil root->fs_info->extent_root); 73772fd032eSSage Weil } 738aec8030aSMiao Xie if (ret) 739c37b2b62SJosef Bacik goto fail; 740a22285a6SYan, Zheng 741a22285a6SYan, Zheng ret = pending_snapshot->error; 742f46b5a66SChristoph Hellwig if (ret) 7432e4bfab9SYan, Zheng goto fail; 744f46b5a66SChristoph Hellwig 745d3797308SChris Mason ret = btrfs_orphan_cleanup(pending_snapshot->snap); 746d3797308SChris Mason if (ret) 747d3797308SChris Mason goto fail; 748d3797308SChris Mason 7492b0143b5SDavid Howells inode = btrfs_lookup_dentry(d_inode(dentry->d_parent), dentry); 7502e4bfab9SYan, Zheng if (IS_ERR(inode)) { 7512e4bfab9SYan, Zheng ret = PTR_ERR(inode); 7522e4bfab9SYan, Zheng goto fail; 7532e4bfab9SYan, Zheng } 7545662344bSTsutomu Itoh 7552e4bfab9SYan, Zheng d_instantiate(dentry, inode); 7562e4bfab9SYan, Zheng ret = 0; 7572e4bfab9SYan, Zheng fail: 758d5c12070SMiao Xie btrfs_subvolume_release_metadata(BTRFS_I(dir)->root, 759d5c12070SMiao Xie &pending_snapshot->block_rsv, 760d5c12070SMiao Xie pending_snapshot->qgroup_reserved); 761a1ee7362SDavid Sterba dec_and_free: 7629ea24bbeSFilipe Manana if (atomic_dec_and_test(&root->will_be_snapshoted)) 7639ea24bbeSFilipe Manana wake_up_atomic_t(&root->will_be_snapshoted); 764b0c0ea63SDavid Sterba free_pending: 765b0c0ea63SDavid Sterba kfree(pending_snapshot->root_item); 7668546b570SDavid Sterba btrfs_free_path(pending_snapshot->path); 767a1ee7362SDavid Sterba kfree(pending_snapshot); 768a1ee7362SDavid Sterba 769f46b5a66SChristoph Hellwig return ret; 770f46b5a66SChristoph Hellwig } 771f46b5a66SChristoph Hellwig 7724260f7c7SSage Weil /* copy of may_delete in fs/namei.c() 7734260f7c7SSage Weil * Check whether we can remove a link victim from directory dir, check 7744260f7c7SSage Weil * whether the type of victim is right. 7754260f7c7SSage Weil * 1. We can't do it if dir is read-only (done in permission()) 7764260f7c7SSage Weil * 2. We should have write and exec permissions on dir 7774260f7c7SSage Weil * 3. We can't remove anything from append-only dir 7784260f7c7SSage Weil * 4. We can't do anything with immutable dir (done in permission()) 7794260f7c7SSage Weil * 5. If the sticky bit on dir is set we should either 7804260f7c7SSage Weil * a. be owner of dir, or 7814260f7c7SSage Weil * b. be owner of victim, or 7824260f7c7SSage Weil * c. have CAP_FOWNER capability 7834260f7c7SSage Weil * 6. If the victim is append-only or immutable we can't do antyhing with 7844260f7c7SSage Weil * links pointing to it. 7854260f7c7SSage Weil * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 7864260f7c7SSage Weil * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 7874260f7c7SSage Weil * 9. We can't remove a root or mountpoint. 7884260f7c7SSage Weil * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 7894260f7c7SSage Weil * nfs_async_unlink(). 7904260f7c7SSage Weil */ 7914260f7c7SSage Weil 7924260f7c7SSage Weil static int btrfs_may_delete(struct inode *dir, struct dentry *victim, int isdir) 7934260f7c7SSage Weil { 7944260f7c7SSage Weil int error; 7954260f7c7SSage Weil 7962b0143b5SDavid Howells if (d_really_is_negative(victim)) 7974260f7c7SSage Weil return -ENOENT; 7984260f7c7SSage Weil 7992b0143b5SDavid Howells BUG_ON(d_inode(victim->d_parent) != dir); 8004fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 8014260f7c7SSage Weil 8024260f7c7SSage Weil error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 8034260f7c7SSage Weil if (error) 8044260f7c7SSage Weil return error; 8054260f7c7SSage Weil if (IS_APPEND(dir)) 8064260f7c7SSage Weil return -EPERM; 8072b0143b5SDavid Howells if (check_sticky(dir, d_inode(victim)) || IS_APPEND(d_inode(victim)) || 8082b0143b5SDavid Howells IS_IMMUTABLE(d_inode(victim)) || IS_SWAPFILE(d_inode(victim))) 8094260f7c7SSage Weil return -EPERM; 8104260f7c7SSage Weil if (isdir) { 811e36cb0b8SDavid Howells if (!d_is_dir(victim)) 8124260f7c7SSage Weil return -ENOTDIR; 8134260f7c7SSage Weil if (IS_ROOT(victim)) 8144260f7c7SSage Weil return -EBUSY; 815e36cb0b8SDavid Howells } else if (d_is_dir(victim)) 8164260f7c7SSage Weil return -EISDIR; 8174260f7c7SSage Weil if (IS_DEADDIR(dir)) 8184260f7c7SSage Weil return -ENOENT; 8194260f7c7SSage Weil if (victim->d_flags & DCACHE_NFSFS_RENAMED) 8204260f7c7SSage Weil return -EBUSY; 8214260f7c7SSage Weil return 0; 8224260f7c7SSage Weil } 8234260f7c7SSage Weil 824cb8e7090SChristoph Hellwig /* copy of may_create in fs/namei.c() */ 825cb8e7090SChristoph Hellwig static inline int btrfs_may_create(struct inode *dir, struct dentry *child) 826cb8e7090SChristoph Hellwig { 8272b0143b5SDavid Howells if (d_really_is_positive(child)) 828cb8e7090SChristoph Hellwig return -EEXIST; 829cb8e7090SChristoph Hellwig if (IS_DEADDIR(dir)) 830cb8e7090SChristoph Hellwig return -ENOENT; 831cb8e7090SChristoph Hellwig return inode_permission(dir, MAY_WRITE | MAY_EXEC); 832cb8e7090SChristoph Hellwig } 833cb8e7090SChristoph Hellwig 834cb8e7090SChristoph Hellwig /* 835cb8e7090SChristoph Hellwig * Create a new subvolume below @parent. This is largely modeled after 836cb8e7090SChristoph Hellwig * sys_mkdirat and vfs_mkdir, but we only do a single component lookup 837cb8e7090SChristoph Hellwig * inside this filesystem so it's quite a bit simpler. 838cb8e7090SChristoph Hellwig */ 83976dda93cSYan, Zheng static noinline int btrfs_mksubvol(struct path *parent, 84076dda93cSYan, Zheng char *name, int namelen, 84172fd032eSSage Weil struct btrfs_root *snap_src, 8426f72c7e2SArne Jansen u64 *async_transid, bool readonly, 8438696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 844cb8e7090SChristoph Hellwig { 8452b0143b5SDavid Howells struct inode *dir = d_inode(parent->dentry); 846cb8e7090SChristoph Hellwig struct dentry *dentry; 847cb8e7090SChristoph Hellwig int error; 848cb8e7090SChristoph Hellwig 8495c50c9b8SDavid Sterba error = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT); 8505c50c9b8SDavid Sterba if (error == -EINTR) 8515c50c9b8SDavid Sterba return error; 852cb8e7090SChristoph Hellwig 853cb8e7090SChristoph Hellwig dentry = lookup_one_len(name, parent->dentry, namelen); 854cb8e7090SChristoph Hellwig error = PTR_ERR(dentry); 855cb8e7090SChristoph Hellwig if (IS_ERR(dentry)) 856cb8e7090SChristoph Hellwig goto out_unlock; 857cb8e7090SChristoph Hellwig 85876dda93cSYan, Zheng error = btrfs_may_create(dir, dentry); 859cb8e7090SChristoph Hellwig if (error) 860a874a63eSLiu Bo goto out_dput; 861cb8e7090SChristoph Hellwig 8629c52057cSChris Mason /* 8639c52057cSChris Mason * even if this name doesn't exist, we may get hash collisions. 8649c52057cSChris Mason * check for them now when we can safely fail 8659c52057cSChris Mason */ 8669c52057cSChris Mason error = btrfs_check_dir_item_collision(BTRFS_I(dir)->root, 8679c52057cSChris Mason dir->i_ino, name, 8689c52057cSChris Mason namelen); 8699c52057cSChris Mason if (error) 8709c52057cSChris Mason goto out_dput; 8719c52057cSChris Mason 87276dda93cSYan, Zheng down_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 87376dda93cSYan, Zheng 87476dda93cSYan, Zheng if (btrfs_root_refs(&BTRFS_I(dir)->root->root_item) == 0) 87576dda93cSYan, Zheng goto out_up_read; 87676dda93cSYan, Zheng 8773de4586cSChris Mason if (snap_src) { 878e9662f70SMiao Xie error = create_snapshot(snap_src, dir, dentry, name, namelen, 8796f72c7e2SArne Jansen async_transid, readonly, inherit); 8803de4586cSChris Mason } else { 881d5c12070SMiao Xie error = create_subvol(dir, dentry, name, namelen, 882d5c12070SMiao Xie async_transid, inherit); 8833de4586cSChris Mason } 88476dda93cSYan, Zheng if (!error) 88576dda93cSYan, Zheng fsnotify_mkdir(dir, dentry); 88676dda93cSYan, Zheng out_up_read: 88776dda93cSYan, Zheng up_read(&BTRFS_I(dir)->root->fs_info->subvol_sem); 888cb8e7090SChristoph Hellwig out_dput: 889cb8e7090SChristoph Hellwig dput(dentry); 890cb8e7090SChristoph Hellwig out_unlock: 8915955102cSAl Viro inode_unlock(dir); 892cb8e7090SChristoph Hellwig return error; 893cb8e7090SChristoph Hellwig } 894cb8e7090SChristoph Hellwig 8954cb5300bSChris Mason /* 8964cb5300bSChris Mason * When we're defragging a range, we don't want to kick it off again 8974cb5300bSChris Mason * if it is really just waiting for delalloc to send it down. 8984cb5300bSChris Mason * If we find a nice big extent or delalloc range for the bytes in the 8994cb5300bSChris Mason * file you want to defrag, we return 0 to let you know to skip this 9004cb5300bSChris Mason * part of the file 9014cb5300bSChris Mason */ 902aab110abSDavid Sterba static int check_defrag_in_cache(struct inode *inode, u64 offset, u32 thresh) 9034cb5300bSChris Mason { 9044cb5300bSChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 9054cb5300bSChris Mason struct extent_map *em = NULL; 9064cb5300bSChris Mason struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 9074cb5300bSChris Mason u64 end; 9084cb5300bSChris Mason 9094cb5300bSChris Mason read_lock(&em_tree->lock); 91009cbfeafSKirill A. Shutemov em = lookup_extent_mapping(em_tree, offset, PAGE_SIZE); 9114cb5300bSChris Mason read_unlock(&em_tree->lock); 9124cb5300bSChris Mason 9134cb5300bSChris Mason if (em) { 9144cb5300bSChris Mason end = extent_map_end(em); 9154cb5300bSChris Mason free_extent_map(em); 9164cb5300bSChris Mason if (end - offset > thresh) 9174cb5300bSChris Mason return 0; 9184cb5300bSChris Mason } 9194cb5300bSChris Mason /* if we already have a nice delalloc here, just stop */ 9204cb5300bSChris Mason thresh /= 2; 9214cb5300bSChris Mason end = count_range_bits(io_tree, &offset, offset + thresh, 9224cb5300bSChris Mason thresh, EXTENT_DELALLOC, 1); 9234cb5300bSChris Mason if (end >= thresh) 9244cb5300bSChris Mason return 0; 9254cb5300bSChris Mason return 1; 9264cb5300bSChris Mason } 9274cb5300bSChris Mason 9284cb5300bSChris Mason /* 9294cb5300bSChris Mason * helper function to walk through a file and find extents 9304cb5300bSChris Mason * newer than a specific transid, and smaller than thresh. 9314cb5300bSChris Mason * 9324cb5300bSChris Mason * This is used by the defragging code to find new and small 9334cb5300bSChris Mason * extents 9344cb5300bSChris Mason */ 9354cb5300bSChris Mason static int find_new_extents(struct btrfs_root *root, 9364cb5300bSChris Mason struct inode *inode, u64 newer_than, 937aab110abSDavid Sterba u64 *off, u32 thresh) 9384cb5300bSChris Mason { 9394cb5300bSChris Mason struct btrfs_path *path; 9404cb5300bSChris Mason struct btrfs_key min_key; 9414cb5300bSChris Mason struct extent_buffer *leaf; 9424cb5300bSChris Mason struct btrfs_file_extent_item *extent; 9434cb5300bSChris Mason int type; 9444cb5300bSChris Mason int ret; 945a4689d2bSDavid Sterba u64 ino = btrfs_ino(inode); 9464cb5300bSChris Mason 9474cb5300bSChris Mason path = btrfs_alloc_path(); 9484cb5300bSChris Mason if (!path) 9494cb5300bSChris Mason return -ENOMEM; 9504cb5300bSChris Mason 951a4689d2bSDavid Sterba min_key.objectid = ino; 9524cb5300bSChris Mason min_key.type = BTRFS_EXTENT_DATA_KEY; 9534cb5300bSChris Mason min_key.offset = *off; 9544cb5300bSChris Mason 9554cb5300bSChris Mason while (1) { 9566174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &min_key, path, newer_than); 9574cb5300bSChris Mason if (ret != 0) 9584cb5300bSChris Mason goto none; 959f094c9bdSFilipe Manana process_slot: 960a4689d2bSDavid Sterba if (min_key.objectid != ino) 9614cb5300bSChris Mason goto none; 9624cb5300bSChris Mason if (min_key.type != BTRFS_EXTENT_DATA_KEY) 9634cb5300bSChris Mason goto none; 9644cb5300bSChris Mason 9654cb5300bSChris Mason leaf = path->nodes[0]; 9664cb5300bSChris Mason extent = btrfs_item_ptr(leaf, path->slots[0], 9674cb5300bSChris Mason struct btrfs_file_extent_item); 9684cb5300bSChris Mason 9694cb5300bSChris Mason type = btrfs_file_extent_type(leaf, extent); 9704cb5300bSChris Mason if (type == BTRFS_FILE_EXTENT_REG && 9714cb5300bSChris Mason btrfs_file_extent_num_bytes(leaf, extent) < thresh && 9724cb5300bSChris Mason check_defrag_in_cache(inode, min_key.offset, thresh)) { 9734cb5300bSChris Mason *off = min_key.offset; 9744cb5300bSChris Mason btrfs_free_path(path); 9754cb5300bSChris Mason return 0; 9764cb5300bSChris Mason } 9774cb5300bSChris Mason 978f094c9bdSFilipe Manana path->slots[0]++; 979f094c9bdSFilipe Manana if (path->slots[0] < btrfs_header_nritems(leaf)) { 980f094c9bdSFilipe Manana btrfs_item_key_to_cpu(leaf, &min_key, path->slots[0]); 981f094c9bdSFilipe Manana goto process_slot; 982f094c9bdSFilipe Manana } 983f094c9bdSFilipe Manana 9844cb5300bSChris Mason if (min_key.offset == (u64)-1) 9854cb5300bSChris Mason goto none; 9864cb5300bSChris Mason 9874cb5300bSChris Mason min_key.offset++; 9884cb5300bSChris Mason btrfs_release_path(path); 9894cb5300bSChris Mason } 9904cb5300bSChris Mason none: 9914cb5300bSChris Mason btrfs_free_path(path); 9924cb5300bSChris Mason return -ENOENT; 9934cb5300bSChris Mason } 9944cb5300bSChris Mason 9956c282eb4SLi Zefan static struct extent_map *defrag_lookup_extent(struct inode *inode, u64 start) 99617ce6ef8SLiu Bo { 99717ce6ef8SLiu Bo struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 998940100a4SChris Mason struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree; 9996c282eb4SLi Zefan struct extent_map *em; 100009cbfeafSKirill A. Shutemov u64 len = PAGE_SIZE; 1001940100a4SChris Mason 1002940100a4SChris Mason /* 1003940100a4SChris Mason * hopefully we have this extent in the tree already, try without 1004940100a4SChris Mason * the full extent lock 1005940100a4SChris Mason */ 1006940100a4SChris Mason read_lock(&em_tree->lock); 1007940100a4SChris Mason em = lookup_extent_mapping(em_tree, start, len); 1008940100a4SChris Mason read_unlock(&em_tree->lock); 1009940100a4SChris Mason 1010940100a4SChris Mason if (!em) { 1011308d9800SFilipe Manana struct extent_state *cached = NULL; 1012308d9800SFilipe Manana u64 end = start + len - 1; 1013308d9800SFilipe Manana 1014940100a4SChris Mason /* get the big lock and read metadata off disk */ 1015ff13db41SDavid Sterba lock_extent_bits(io_tree, start, end, &cached); 1016940100a4SChris Mason em = btrfs_get_extent(inode, NULL, 0, start, len, 0); 1017308d9800SFilipe Manana unlock_extent_cached(io_tree, start, end, &cached, GFP_NOFS); 1018940100a4SChris Mason 10196cf8bfbfSDan Carpenter if (IS_ERR(em)) 10206c282eb4SLi Zefan return NULL; 1021940100a4SChris Mason } 1022940100a4SChris Mason 10236c282eb4SLi Zefan return em; 10246c282eb4SLi Zefan } 10256c282eb4SLi Zefan 10266c282eb4SLi Zefan static bool defrag_check_next_extent(struct inode *inode, struct extent_map *em) 10276c282eb4SLi Zefan { 10286c282eb4SLi Zefan struct extent_map *next; 10296c282eb4SLi Zefan bool ret = true; 10306c282eb4SLi Zefan 10316c282eb4SLi Zefan /* this is the last extent */ 10326c282eb4SLi Zefan if (em->start + em->len >= i_size_read(inode)) 10336c282eb4SLi Zefan return false; 10346c282eb4SLi Zefan 10356c282eb4SLi Zefan next = defrag_lookup_extent(inode, em->start + em->len); 1036e9512d72SChris Mason if (!next || next->block_start >= EXTENT_MAP_LAST_BYTE) 1037e9512d72SChris Mason ret = false; 1038e9512d72SChris Mason else if ((em->block_start + em->block_len == next->block_start) && 1039ee22184bSByongho Lee (em->block_len > SZ_128K && next->block_len > SZ_128K)) 10406c282eb4SLi Zefan ret = false; 10416c282eb4SLi Zefan 10426c282eb4SLi Zefan free_extent_map(next); 10436c282eb4SLi Zefan return ret; 10446c282eb4SLi Zefan } 10456c282eb4SLi Zefan 1046aab110abSDavid Sterba static int should_defrag_range(struct inode *inode, u64 start, u32 thresh, 1047a43a2111SAndrew Mahone u64 *last_len, u64 *skip, u64 *defrag_end, 1048a43a2111SAndrew Mahone int compress) 10496c282eb4SLi Zefan { 10506c282eb4SLi Zefan struct extent_map *em; 10516c282eb4SLi Zefan int ret = 1; 10526c282eb4SLi Zefan bool next_mergeable = true; 10534a3560c4SLiu Bo bool prev_mergeable = true; 10546c282eb4SLi Zefan 10556c282eb4SLi Zefan /* 10566c282eb4SLi Zefan * make sure that once we start defragging an extent, we keep on 10576c282eb4SLi Zefan * defragging it 10586c282eb4SLi Zefan */ 10596c282eb4SLi Zefan if (start < *defrag_end) 10606c282eb4SLi Zefan return 1; 10616c282eb4SLi Zefan 10626c282eb4SLi Zefan *skip = 0; 10636c282eb4SLi Zefan 10646c282eb4SLi Zefan em = defrag_lookup_extent(inode, start); 10656c282eb4SLi Zefan if (!em) 10666c282eb4SLi Zefan return 0; 10676c282eb4SLi Zefan 1068940100a4SChris Mason /* this will cover holes, and inline extents */ 106917ce6ef8SLiu Bo if (em->block_start >= EXTENT_MAP_LAST_BYTE) { 1070940100a4SChris Mason ret = 0; 107117ce6ef8SLiu Bo goto out; 107217ce6ef8SLiu Bo } 107317ce6ef8SLiu Bo 10744a3560c4SLiu Bo if (!*defrag_end) 10754a3560c4SLiu Bo prev_mergeable = false; 10764a3560c4SLiu Bo 10776c282eb4SLi Zefan next_mergeable = defrag_check_next_extent(inode, em); 1078940100a4SChris Mason /* 10796c282eb4SLi Zefan * we hit a real extent, if it is big or the next extent is not a 10806c282eb4SLi Zefan * real extent, don't bother defragging it 1081940100a4SChris Mason */ 1082a43a2111SAndrew Mahone if (!compress && (*last_len == 0 || *last_len >= thresh) && 10834a3560c4SLiu Bo (em->len >= thresh || (!next_mergeable && !prev_mergeable))) 1084940100a4SChris Mason ret = 0; 108517ce6ef8SLiu Bo out: 1086940100a4SChris Mason /* 1087940100a4SChris Mason * last_len ends up being a counter of how many bytes we've defragged. 1088940100a4SChris Mason * every time we choose not to defrag an extent, we reset *last_len 1089940100a4SChris Mason * so that the next tiny extent will force a defrag. 1090940100a4SChris Mason * 1091940100a4SChris Mason * The end result of this is that tiny extents before a single big 1092940100a4SChris Mason * extent will force at least part of that big extent to be defragged. 1093940100a4SChris Mason */ 1094940100a4SChris Mason if (ret) { 1095940100a4SChris Mason *defrag_end = extent_map_end(em); 1096940100a4SChris Mason } else { 1097940100a4SChris Mason *last_len = 0; 1098940100a4SChris Mason *skip = extent_map_end(em); 1099940100a4SChris Mason *defrag_end = 0; 1100940100a4SChris Mason } 1101940100a4SChris Mason 1102940100a4SChris Mason free_extent_map(em); 1103940100a4SChris Mason return ret; 1104940100a4SChris Mason } 1105940100a4SChris Mason 11064cb5300bSChris Mason /* 11074cb5300bSChris Mason * it doesn't do much good to defrag one or two pages 11084cb5300bSChris Mason * at a time. This pulls in a nice chunk of pages 11094cb5300bSChris Mason * to COW and defrag. 11104cb5300bSChris Mason * 11114cb5300bSChris Mason * It also makes sure the delalloc code has enough 11124cb5300bSChris Mason * dirty data to avoid making new small extents as part 11134cb5300bSChris Mason * of the defrag 11144cb5300bSChris Mason * 11154cb5300bSChris Mason * It's a good idea to start RA on this range 11164cb5300bSChris Mason * before calling this. 11174cb5300bSChris Mason */ 11184cb5300bSChris Mason static int cluster_pages_for_defrag(struct inode *inode, 11194cb5300bSChris Mason struct page **pages, 11204cb5300bSChris Mason unsigned long start_index, 1121c41570c9SJustin Maggard unsigned long num_pages) 1122f46b5a66SChristoph Hellwig { 11234cb5300bSChris Mason unsigned long file_end; 11244cb5300bSChris Mason u64 isize = i_size_read(inode); 1125f46b5a66SChristoph Hellwig u64 page_start; 1126f46b5a66SChristoph Hellwig u64 page_end; 11271f12bd06SLiu Bo u64 page_cnt; 11284cb5300bSChris Mason int ret; 11294cb5300bSChris Mason int i; 11304cb5300bSChris Mason int i_done; 11314cb5300bSChris Mason struct btrfs_ordered_extent *ordered; 11324cb5300bSChris Mason struct extent_state *cached_state = NULL; 1133600a45e1SMiao Xie struct extent_io_tree *tree; 11343b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 11354cb5300bSChris Mason 113609cbfeafSKirill A. Shutemov file_end = (isize - 1) >> PAGE_SHIFT; 11371f12bd06SLiu Bo if (!isize || start_index > file_end) 11381f12bd06SLiu Bo return 0; 11391f12bd06SLiu Bo 11401f12bd06SLiu Bo page_cnt = min_t(u64, (u64)num_pages, (u64)file_end - start_index + 1); 11414cb5300bSChris Mason 11424cb5300bSChris Mason ret = btrfs_delalloc_reserve_space(inode, 114309cbfeafSKirill A. Shutemov start_index << PAGE_SHIFT, 114409cbfeafSKirill A. Shutemov page_cnt << PAGE_SHIFT); 11454cb5300bSChris Mason if (ret) 11464cb5300bSChris Mason return ret; 11474cb5300bSChris Mason i_done = 0; 1148600a45e1SMiao Xie tree = &BTRFS_I(inode)->io_tree; 11494cb5300bSChris Mason 11504cb5300bSChris Mason /* step one, lock all the pages */ 11511f12bd06SLiu Bo for (i = 0; i < page_cnt; i++) { 11524cb5300bSChris Mason struct page *page; 1153600a45e1SMiao Xie again: 1154a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, 11553b16a4e3SJosef Bacik start_index + i, mask); 11564cb5300bSChris Mason if (!page) 11574cb5300bSChris Mason break; 11584cb5300bSChris Mason 1159600a45e1SMiao Xie page_start = page_offset(page); 116009cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 1161600a45e1SMiao Xie while (1) { 1162308d9800SFilipe Manana lock_extent_bits(tree, page_start, page_end, 1163ff13db41SDavid Sterba &cached_state); 1164600a45e1SMiao Xie ordered = btrfs_lookup_ordered_extent(inode, 1165600a45e1SMiao Xie page_start); 1166308d9800SFilipe Manana unlock_extent_cached(tree, page_start, page_end, 1167308d9800SFilipe Manana &cached_state, GFP_NOFS); 1168600a45e1SMiao Xie if (!ordered) 1169600a45e1SMiao Xie break; 1170600a45e1SMiao Xie 1171600a45e1SMiao Xie unlock_page(page); 1172600a45e1SMiao Xie btrfs_start_ordered_extent(inode, ordered, 1); 1173600a45e1SMiao Xie btrfs_put_ordered_extent(ordered); 1174600a45e1SMiao Xie lock_page(page); 11751f12bd06SLiu Bo /* 11761f12bd06SLiu Bo * we unlocked the page above, so we need check if 11771f12bd06SLiu Bo * it was released or not. 11781f12bd06SLiu Bo */ 11791f12bd06SLiu Bo if (page->mapping != inode->i_mapping) { 11801f12bd06SLiu Bo unlock_page(page); 118109cbfeafSKirill A. Shutemov put_page(page); 11821f12bd06SLiu Bo goto again; 11831f12bd06SLiu Bo } 1184600a45e1SMiao Xie } 1185600a45e1SMiao Xie 11864cb5300bSChris Mason if (!PageUptodate(page)) { 11874cb5300bSChris Mason btrfs_readpage(NULL, page); 11884cb5300bSChris Mason lock_page(page); 11894cb5300bSChris Mason if (!PageUptodate(page)) { 11904cb5300bSChris Mason unlock_page(page); 119109cbfeafSKirill A. Shutemov put_page(page); 11924cb5300bSChris Mason ret = -EIO; 11934cb5300bSChris Mason break; 11944cb5300bSChris Mason } 11954cb5300bSChris Mason } 1196600a45e1SMiao Xie 1197600a45e1SMiao Xie if (page->mapping != inode->i_mapping) { 1198600a45e1SMiao Xie unlock_page(page); 119909cbfeafSKirill A. Shutemov put_page(page); 1200600a45e1SMiao Xie goto again; 1201600a45e1SMiao Xie } 1202600a45e1SMiao Xie 12034cb5300bSChris Mason pages[i] = page; 12044cb5300bSChris Mason i_done++; 12054cb5300bSChris Mason } 12064cb5300bSChris Mason if (!i_done || ret) 12074cb5300bSChris Mason goto out; 12084cb5300bSChris Mason 12094cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 12104cb5300bSChris Mason goto out; 12114cb5300bSChris Mason 12124cb5300bSChris Mason /* 12134cb5300bSChris Mason * so now we have a nice long stream of locked 12144cb5300bSChris Mason * and up to date pages, lets wait on them 12154cb5300bSChris Mason */ 12164cb5300bSChris Mason for (i = 0; i < i_done; i++) 12174cb5300bSChris Mason wait_on_page_writeback(pages[i]); 12184cb5300bSChris Mason 12194cb5300bSChris Mason page_start = page_offset(pages[0]); 122009cbfeafSKirill A. Shutemov page_end = page_offset(pages[i_done - 1]) + PAGE_SIZE; 12214cb5300bSChris Mason 12224cb5300bSChris Mason lock_extent_bits(&BTRFS_I(inode)->io_tree, 1223ff13db41SDavid Sterba page_start, page_end - 1, &cached_state); 12244cb5300bSChris Mason clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, 12254cb5300bSChris Mason page_end - 1, EXTENT_DIRTY | EXTENT_DELALLOC | 12269e8a4a8bSLiu Bo EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 0, 0, 12279e8a4a8bSLiu Bo &cached_state, GFP_NOFS); 12284cb5300bSChris Mason 12291f12bd06SLiu Bo if (i_done != page_cnt) { 12309e0baf60SJosef Bacik spin_lock(&BTRFS_I(inode)->lock); 12319e0baf60SJosef Bacik BTRFS_I(inode)->outstanding_extents++; 12329e0baf60SJosef Bacik spin_unlock(&BTRFS_I(inode)->lock); 12334cb5300bSChris Mason btrfs_delalloc_release_space(inode, 123409cbfeafSKirill A. Shutemov start_index << PAGE_SHIFT, 123509cbfeafSKirill A. Shutemov (page_cnt - i_done) << PAGE_SHIFT); 12364cb5300bSChris Mason } 12374cb5300bSChris Mason 12384cb5300bSChris Mason 12399e8a4a8bSLiu Bo set_extent_defrag(&BTRFS_I(inode)->io_tree, page_start, page_end - 1, 12409e8a4a8bSLiu Bo &cached_state, GFP_NOFS); 12414cb5300bSChris Mason 12424cb5300bSChris Mason unlock_extent_cached(&BTRFS_I(inode)->io_tree, 12434cb5300bSChris Mason page_start, page_end - 1, &cached_state, 12444cb5300bSChris Mason GFP_NOFS); 12454cb5300bSChris Mason 12464cb5300bSChris Mason for (i = 0; i < i_done; i++) { 12474cb5300bSChris Mason clear_page_dirty_for_io(pages[i]); 12484cb5300bSChris Mason ClearPageChecked(pages[i]); 12494cb5300bSChris Mason set_page_extent_mapped(pages[i]); 12504cb5300bSChris Mason set_page_dirty(pages[i]); 12514cb5300bSChris Mason unlock_page(pages[i]); 125209cbfeafSKirill A. Shutemov put_page(pages[i]); 12534cb5300bSChris Mason } 12544cb5300bSChris Mason return i_done; 12554cb5300bSChris Mason out: 12564cb5300bSChris Mason for (i = 0; i < i_done; i++) { 12574cb5300bSChris Mason unlock_page(pages[i]); 125809cbfeafSKirill A. Shutemov put_page(pages[i]); 12594cb5300bSChris Mason } 12607cf5b976SQu Wenruo btrfs_delalloc_release_space(inode, 126109cbfeafSKirill A. Shutemov start_index << PAGE_SHIFT, 126209cbfeafSKirill A. Shutemov page_cnt << PAGE_SHIFT); 12634cb5300bSChris Mason return ret; 12644cb5300bSChris Mason 12654cb5300bSChris Mason } 12664cb5300bSChris Mason 12674cb5300bSChris Mason int btrfs_defrag_file(struct inode *inode, struct file *file, 12684cb5300bSChris Mason struct btrfs_ioctl_defrag_range_args *range, 12694cb5300bSChris Mason u64 newer_than, unsigned long max_to_defrag) 12704cb5300bSChris Mason { 12714cb5300bSChris Mason struct btrfs_root *root = BTRFS_I(inode)->root; 12724cb5300bSChris Mason struct file_ra_state *ra = NULL; 12734cb5300bSChris Mason unsigned long last_index; 1274151a31b2SLi Zefan u64 isize = i_size_read(inode); 1275940100a4SChris Mason u64 last_len = 0; 1276940100a4SChris Mason u64 skip = 0; 1277940100a4SChris Mason u64 defrag_end = 0; 12784cb5300bSChris Mason u64 newer_off = range->start; 1279f46b5a66SChristoph Hellwig unsigned long i; 1280008873eaSLi Zefan unsigned long ra_index = 0; 1281f46b5a66SChristoph Hellwig int ret; 12824cb5300bSChris Mason int defrag_count = 0; 12831a419d85SLi Zefan int compress_type = BTRFS_COMPRESS_ZLIB; 1284aab110abSDavid Sterba u32 extent_thresh = range->extent_thresh; 128509cbfeafSKirill A. Shutemov unsigned long max_cluster = SZ_256K >> PAGE_SHIFT; 1286c41570c9SJustin Maggard unsigned long cluster = max_cluster; 1287ee22184bSByongho Lee u64 new_align = ~((u64)SZ_128K - 1); 12884cb5300bSChris Mason struct page **pages = NULL; 12894cb5300bSChris Mason 12900abd5b17SLiu Bo if (isize == 0) 12910abd5b17SLiu Bo return 0; 12920abd5b17SLiu Bo 12930abd5b17SLiu Bo if (range->start >= isize) 12940abd5b17SLiu Bo return -EINVAL; 12951a419d85SLi Zefan 12961a419d85SLi Zefan if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 12971a419d85SLi Zefan if (range->compress_type > BTRFS_COMPRESS_TYPES) 12981a419d85SLi Zefan return -EINVAL; 12991a419d85SLi Zefan if (range->compress_type) 13001a419d85SLi Zefan compress_type = range->compress_type; 13011a419d85SLi Zefan } 1302f46b5a66SChristoph Hellwig 13030abd5b17SLiu Bo if (extent_thresh == 0) 1304ee22184bSByongho Lee extent_thresh = SZ_256K; 1305f46b5a66SChristoph Hellwig 13064cb5300bSChris Mason /* 13074cb5300bSChris Mason * if we were not given a file, allocate a readahead 13084cb5300bSChris Mason * context 13094cb5300bSChris Mason */ 13104cb5300bSChris Mason if (!file) { 13114cb5300bSChris Mason ra = kzalloc(sizeof(*ra), GFP_NOFS); 13124cb5300bSChris Mason if (!ra) 13134cb5300bSChris Mason return -ENOMEM; 13144cb5300bSChris Mason file_ra_state_init(ra, inode->i_mapping); 13154cb5300bSChris Mason } else { 13164cb5300bSChris Mason ra = &file->f_ra; 13174cb5300bSChris Mason } 13184cb5300bSChris Mason 1319d9b0d9baSDulshani Gunawardhana pages = kmalloc_array(max_cluster, sizeof(struct page *), 13204cb5300bSChris Mason GFP_NOFS); 13214cb5300bSChris Mason if (!pages) { 13224cb5300bSChris Mason ret = -ENOMEM; 13234cb5300bSChris Mason goto out_ra; 13244cb5300bSChris Mason } 13254cb5300bSChris Mason 13264cb5300bSChris Mason /* find the last page to defrag */ 13271e701a32SChris Mason if (range->start + range->len > range->start) { 1328151a31b2SLi Zefan last_index = min_t(u64, isize - 1, 132909cbfeafSKirill A. Shutemov range->start + range->len - 1) >> PAGE_SHIFT; 13301e701a32SChris Mason } else { 133109cbfeafSKirill A. Shutemov last_index = (isize - 1) >> PAGE_SHIFT; 13321e701a32SChris Mason } 13331e701a32SChris Mason 13344cb5300bSChris Mason if (newer_than) { 13354cb5300bSChris Mason ret = find_new_extents(root, inode, newer_than, 1336ee22184bSByongho Lee &newer_off, SZ_64K); 13374cb5300bSChris Mason if (!ret) { 13384cb5300bSChris Mason range->start = newer_off; 13394cb5300bSChris Mason /* 13404cb5300bSChris Mason * we always align our defrag to help keep 13414cb5300bSChris Mason * the extents in the file evenly spaced 13424cb5300bSChris Mason */ 134309cbfeafSKirill A. Shutemov i = (newer_off & new_align) >> PAGE_SHIFT; 13444cb5300bSChris Mason } else 13454cb5300bSChris Mason goto out_ra; 13464cb5300bSChris Mason } else { 134709cbfeafSKirill A. Shutemov i = range->start >> PAGE_SHIFT; 13484cb5300bSChris Mason } 13494cb5300bSChris Mason if (!max_to_defrag) 1350070034bdSchandan max_to_defrag = last_index - i + 1; 13514cb5300bSChris Mason 13522a0f7f57SLi Zefan /* 13532a0f7f57SLi Zefan * make writeback starts from i, so the defrag range can be 13542a0f7f57SLi Zefan * written sequentially. 13552a0f7f57SLi Zefan */ 13562a0f7f57SLi Zefan if (i < inode->i_mapping->writeback_index) 13572a0f7f57SLi Zefan inode->i_mapping->writeback_index = i; 13582a0f7f57SLi Zefan 1359f7f43cc8SChris Mason while (i <= last_index && defrag_count < max_to_defrag && 136009cbfeafSKirill A. Shutemov (i < DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE))) { 13614cb5300bSChris Mason /* 13624cb5300bSChris Mason * make sure we stop running if someone unmounts 13634cb5300bSChris Mason * the FS 13644cb5300bSChris Mason */ 13654cb5300bSChris Mason if (!(inode->i_sb->s_flags & MS_ACTIVE)) 13664cb5300bSChris Mason break; 13674cb5300bSChris Mason 1368210549ebSDavid Sterba if (btrfs_defrag_cancelled(root->fs_info)) { 1369f14d104dSDavid Sterba btrfs_debug(root->fs_info, "defrag_file cancelled"); 1370210549ebSDavid Sterba ret = -EAGAIN; 1371210549ebSDavid Sterba break; 1372210549ebSDavid Sterba } 1373210549ebSDavid Sterba 137409cbfeafSKirill A. Shutemov if (!should_defrag_range(inode, (u64)i << PAGE_SHIFT, 13756c282eb4SLi Zefan extent_thresh, &last_len, &skip, 1376a43a2111SAndrew Mahone &defrag_end, range->flags & 1377a43a2111SAndrew Mahone BTRFS_DEFRAG_RANGE_COMPRESS)) { 1378940100a4SChris Mason unsigned long next; 1379940100a4SChris Mason /* 1380940100a4SChris Mason * the should_defrag function tells us how much to skip 1381940100a4SChris Mason * bump our counter by the suggested amount 1382940100a4SChris Mason */ 138309cbfeafSKirill A. Shutemov next = DIV_ROUND_UP(skip, PAGE_SIZE); 1384940100a4SChris Mason i = max(i + 1, next); 1385940100a4SChris Mason continue; 1386940100a4SChris Mason } 1387008873eaSLi Zefan 1388008873eaSLi Zefan if (!newer_than) { 138909cbfeafSKirill A. Shutemov cluster = (PAGE_ALIGN(defrag_end) >> 139009cbfeafSKirill A. Shutemov PAGE_SHIFT) - i; 1391008873eaSLi Zefan cluster = min(cluster, max_cluster); 1392008873eaSLi Zefan } else { 1393008873eaSLi Zefan cluster = max_cluster; 1394008873eaSLi Zefan } 1395008873eaSLi Zefan 1396008873eaSLi Zefan if (i + cluster > ra_index) { 1397008873eaSLi Zefan ra_index = max(i, ra_index); 1398008873eaSLi Zefan btrfs_force_ra(inode->i_mapping, ra, file, ra_index, 1399008873eaSLi Zefan cluster); 1400e4826a5bSchandan ra_index += cluster; 1401008873eaSLi Zefan } 14024cb5300bSChris Mason 14035955102cSAl Viro inode_lock(inode); 1404633085c7SFilipe David Borba Manana if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) 1405633085c7SFilipe David Borba Manana BTRFS_I(inode)->force_compress = compress_type; 1406008873eaSLi Zefan ret = cluster_pages_for_defrag(inode, pages, i, cluster); 1407ecb8bea8SLiu Bo if (ret < 0) { 14085955102cSAl Viro inode_unlock(inode); 14094cb5300bSChris Mason goto out_ra; 1410ecb8bea8SLiu Bo } 14114cb5300bSChris Mason 14124cb5300bSChris Mason defrag_count += ret; 1413d0e1d66bSNamjae Jeon balance_dirty_pages_ratelimited(inode->i_mapping); 14145955102cSAl Viro inode_unlock(inode); 14154cb5300bSChris Mason 14164cb5300bSChris Mason if (newer_than) { 14174cb5300bSChris Mason if (newer_off == (u64)-1) 14184cb5300bSChris Mason break; 14194cb5300bSChris Mason 1420e1f041e1SLiu Bo if (ret > 0) 1421e1f041e1SLiu Bo i += ret; 1422e1f041e1SLiu Bo 14234cb5300bSChris Mason newer_off = max(newer_off + 1, 142409cbfeafSKirill A. Shutemov (u64)i << PAGE_SHIFT); 14254cb5300bSChris Mason 1426ee22184bSByongho Lee ret = find_new_extents(root, inode, newer_than, 1427ee22184bSByongho Lee &newer_off, SZ_64K); 14284cb5300bSChris Mason if (!ret) { 14294cb5300bSChris Mason range->start = newer_off; 143009cbfeafSKirill A. Shutemov i = (newer_off & new_align) >> PAGE_SHIFT; 14314cb5300bSChris Mason } else { 14324cb5300bSChris Mason break; 1433940100a4SChris Mason } 14344cb5300bSChris Mason } else { 1435008873eaSLi Zefan if (ret > 0) { 1436cbcc8326SLi Zefan i += ret; 143709cbfeafSKirill A. Shutemov last_len += ret << PAGE_SHIFT; 1438008873eaSLi Zefan } else { 1439940100a4SChris Mason i++; 1440008873eaSLi Zefan last_len = 0; 1441008873eaSLi Zefan } 1442f46b5a66SChristoph Hellwig } 14434cb5300bSChris Mason } 1444f46b5a66SChristoph Hellwig 1445dec8ef90SFilipe Manana if ((range->flags & BTRFS_DEFRAG_RANGE_START_IO)) { 14461e701a32SChris Mason filemap_flush(inode->i_mapping); 1447dec8ef90SFilipe Manana if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, 1448dec8ef90SFilipe Manana &BTRFS_I(inode)->runtime_flags)) 1449dec8ef90SFilipe Manana filemap_flush(inode->i_mapping); 1450dec8ef90SFilipe Manana } 14511e701a32SChris Mason 14521e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 14531e701a32SChris Mason /* the filemap_flush will queue IO into the worker threads, but 14541e701a32SChris Mason * we have to make sure the IO is actually started and that 14551e701a32SChris Mason * ordered extents get created before we return 14561e701a32SChris Mason */ 14571e701a32SChris Mason atomic_inc(&root->fs_info->async_submit_draining); 14581e701a32SChris Mason while (atomic_read(&root->fs_info->nr_async_submits) || 14591e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages)) { 14601e701a32SChris Mason wait_event(root->fs_info->async_submit_wait, 14611e701a32SChris Mason (atomic_read(&root->fs_info->nr_async_submits) == 0 && 14621e701a32SChris Mason atomic_read(&root->fs_info->async_delalloc_pages) == 0)); 14631e701a32SChris Mason } 14641e701a32SChris Mason atomic_dec(&root->fs_info->async_submit_draining); 14651e701a32SChris Mason } 14661e701a32SChris Mason 14671a419d85SLi Zefan if (range->compress_type == BTRFS_COMPRESS_LZO) { 14682b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, COMPRESS_LZO); 14691a419d85SLi Zefan } 14701a419d85SLi Zefan 147160ccf82fSDiego Calleja ret = defrag_count; 1472940100a4SChris Mason 14734cb5300bSChris Mason out_ra: 1474633085c7SFilipe David Borba Manana if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) { 14755955102cSAl Viro inode_lock(inode); 1476633085c7SFilipe David Borba Manana BTRFS_I(inode)->force_compress = BTRFS_COMPRESS_NONE; 14775955102cSAl Viro inode_unlock(inode); 1478633085c7SFilipe David Borba Manana } 14794cb5300bSChris Mason if (!file) 14804cb5300bSChris Mason kfree(ra); 14814cb5300bSChris Mason kfree(pages); 1482940100a4SChris Mason return ret; 1483f46b5a66SChristoph Hellwig } 1484f46b5a66SChristoph Hellwig 1485198605a8SMiao Xie static noinline int btrfs_ioctl_resize(struct file *file, 148676dda93cSYan, Zheng void __user *arg) 1487f46b5a66SChristoph Hellwig { 1488f46b5a66SChristoph Hellwig u64 new_size; 1489f46b5a66SChristoph Hellwig u64 old_size; 1490f46b5a66SChristoph Hellwig u64 devid = 1; 1491496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 1492f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 1493f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 1494f46b5a66SChristoph Hellwig struct btrfs_device *device = NULL; 1495f46b5a66SChristoph Hellwig char *sizestr; 14969a40f122SGui Hecheng char *retptr; 1497f46b5a66SChristoph Hellwig char *devstr = NULL; 1498f46b5a66SChristoph Hellwig int ret = 0; 1499f46b5a66SChristoph Hellwig int mod = 0; 1500f46b5a66SChristoph Hellwig 1501e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 1502e441d54dSChris Mason return -EPERM; 1503e441d54dSChris Mason 1504198605a8SMiao Xie ret = mnt_want_write_file(file); 1505198605a8SMiao Xie if (ret) 1506198605a8SMiao Xie return ret; 1507198605a8SMiao Xie 15085ac00addSStefan Behrens if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 15095ac00addSStefan Behrens 1)) { 151097547676SMiao Xie mnt_drop_write_file(file); 1511e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 1512c9e9f97bSIlya Dryomov } 1513c9e9f97bSIlya Dryomov 15145ac00addSStefan Behrens mutex_lock(&root->fs_info->volume_mutex); 1515dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1516c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 1517c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 1518c9e9f97bSIlya Dryomov goto out; 1519c9e9f97bSIlya Dryomov } 15205516e595SMark Fasheh 15215516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1522f46b5a66SChristoph Hellwig 1523f46b5a66SChristoph Hellwig sizestr = vol_args->name; 1524f46b5a66SChristoph Hellwig devstr = strchr(sizestr, ':'); 1525f46b5a66SChristoph Hellwig if (devstr) { 1526f46b5a66SChristoph Hellwig sizestr = devstr + 1; 1527f46b5a66SChristoph Hellwig *devstr = '\0'; 1528f46b5a66SChristoph Hellwig devstr = vol_args->name; 152958dfae63SZhangZhen ret = kstrtoull(devstr, 10, &devid); 153058dfae63SZhangZhen if (ret) 153158dfae63SZhangZhen goto out_free; 1532dfd79829SMiao Xie if (!devid) { 1533dfd79829SMiao Xie ret = -EINVAL; 1534dfd79829SMiao Xie goto out_free; 1535dfd79829SMiao Xie } 1536efe120a0SFrank Holton btrfs_info(root->fs_info, "resizing devid %llu", devid); 1537f46b5a66SChristoph Hellwig } 1538dba60f3fSMiao Xie 1539aa1b8cd4SStefan Behrens device = btrfs_find_device(root->fs_info, devid, NULL, NULL); 1540f46b5a66SChristoph Hellwig if (!device) { 1541efe120a0SFrank Holton btrfs_info(root->fs_info, "resizer unable to find device %llu", 1542c1c9ff7cSGeert Uytterhoeven devid); 1543dfd79829SMiao Xie ret = -ENODEV; 1544c9e9f97bSIlya Dryomov goto out_free; 1545f46b5a66SChristoph Hellwig } 1546dba60f3fSMiao Xie 1547dba60f3fSMiao Xie if (!device->writeable) { 1548efe120a0SFrank Holton btrfs_info(root->fs_info, 1549efe120a0SFrank Holton "resizer unable to apply on readonly device %llu", 1550c1c9ff7cSGeert Uytterhoeven devid); 1551dfd79829SMiao Xie ret = -EPERM; 15524e42ae1bSLiu Bo goto out_free; 15534e42ae1bSLiu Bo } 15544e42ae1bSLiu Bo 1555f46b5a66SChristoph Hellwig if (!strcmp(sizestr, "max")) 1556f46b5a66SChristoph Hellwig new_size = device->bdev->bd_inode->i_size; 1557f46b5a66SChristoph Hellwig else { 1558f46b5a66SChristoph Hellwig if (sizestr[0] == '-') { 1559f46b5a66SChristoph Hellwig mod = -1; 1560f46b5a66SChristoph Hellwig sizestr++; 1561f46b5a66SChristoph Hellwig } else if (sizestr[0] == '+') { 1562f46b5a66SChristoph Hellwig mod = 1; 1563f46b5a66SChristoph Hellwig sizestr++; 1564f46b5a66SChristoph Hellwig } 15659a40f122SGui Hecheng new_size = memparse(sizestr, &retptr); 15669a40f122SGui Hecheng if (*retptr != '\0' || new_size == 0) { 1567f46b5a66SChristoph Hellwig ret = -EINVAL; 1568c9e9f97bSIlya Dryomov goto out_free; 1569f46b5a66SChristoph Hellwig } 1570f46b5a66SChristoph Hellwig } 1571f46b5a66SChristoph Hellwig 157263a212abSStefan Behrens if (device->is_tgtdev_for_dev_replace) { 1573dfd79829SMiao Xie ret = -EPERM; 157463a212abSStefan Behrens goto out_free; 157563a212abSStefan Behrens } 157663a212abSStefan Behrens 15777cc8e58dSMiao Xie old_size = btrfs_device_get_total_bytes(device); 1578f46b5a66SChristoph Hellwig 1579f46b5a66SChristoph Hellwig if (mod < 0) { 1580f46b5a66SChristoph Hellwig if (new_size > old_size) { 1581f46b5a66SChristoph Hellwig ret = -EINVAL; 1582c9e9f97bSIlya Dryomov goto out_free; 1583f46b5a66SChristoph Hellwig } 1584f46b5a66SChristoph Hellwig new_size = old_size - new_size; 1585f46b5a66SChristoph Hellwig } else if (mod > 0) { 1586eb8052e0SWenliang Fan if (new_size > ULLONG_MAX - old_size) { 1587902c68a4SGui Hecheng ret = -ERANGE; 1588eb8052e0SWenliang Fan goto out_free; 1589eb8052e0SWenliang Fan } 1590f46b5a66SChristoph Hellwig new_size = old_size + new_size; 1591f46b5a66SChristoph Hellwig } 1592f46b5a66SChristoph Hellwig 1593ee22184bSByongho Lee if (new_size < SZ_256M) { 1594f46b5a66SChristoph Hellwig ret = -EINVAL; 1595c9e9f97bSIlya Dryomov goto out_free; 1596f46b5a66SChristoph Hellwig } 1597f46b5a66SChristoph Hellwig if (new_size > device->bdev->bd_inode->i_size) { 1598f46b5a66SChristoph Hellwig ret = -EFBIG; 1599c9e9f97bSIlya Dryomov goto out_free; 1600f46b5a66SChristoph Hellwig } 1601f46b5a66SChristoph Hellwig 1602b8b93addSDavid Sterba new_size = div_u64(new_size, root->sectorsize); 1603f46b5a66SChristoph Hellwig new_size *= root->sectorsize; 1604f46b5a66SChristoph Hellwig 1605ecaeb14bSDavid Sterba btrfs_info_in_rcu(root->fs_info, "new size for %s is %llu", 1606c1c9ff7cSGeert Uytterhoeven rcu_str_deref(device->name), new_size); 1607f46b5a66SChristoph Hellwig 1608f46b5a66SChristoph Hellwig if (new_size > old_size) { 1609a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 161098d5dc13STsutomu Itoh if (IS_ERR(trans)) { 161198d5dc13STsutomu Itoh ret = PTR_ERR(trans); 1612c9e9f97bSIlya Dryomov goto out_free; 161398d5dc13STsutomu Itoh } 1614f46b5a66SChristoph Hellwig ret = btrfs_grow_device(trans, device, new_size); 1615f46b5a66SChristoph Hellwig btrfs_commit_transaction(trans, root); 1616ece7d20eSMike Fleetwood } else if (new_size < old_size) { 1617f46b5a66SChristoph Hellwig ret = btrfs_shrink_device(device, new_size); 16180253f40eSjeff.liu } /* equal, nothing need to do */ 1619f46b5a66SChristoph Hellwig 1620c9e9f97bSIlya Dryomov out_free: 1621f46b5a66SChristoph Hellwig kfree(vol_args); 1622c9e9f97bSIlya Dryomov out: 1623c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 16245ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 162518f39c41SIlya Dryomov mnt_drop_write_file(file); 1626f46b5a66SChristoph Hellwig return ret; 1627f46b5a66SChristoph Hellwig } 1628f46b5a66SChristoph Hellwig 162972fd032eSSage Weil static noinline int btrfs_ioctl_snap_create_transid(struct file *file, 16306f72c7e2SArne Jansen char *name, unsigned long fd, int subvol, 16316f72c7e2SArne Jansen u64 *transid, bool readonly, 16328696c533SMiao Xie struct btrfs_qgroup_inherit *inherit) 1633f46b5a66SChristoph Hellwig { 1634f46b5a66SChristoph Hellwig int namelen; 16353de4586cSChris Mason int ret = 0; 1636f46b5a66SChristoph Hellwig 1637a874a63eSLiu Bo ret = mnt_want_write_file(file); 1638a874a63eSLiu Bo if (ret) 1639a874a63eSLiu Bo goto out; 1640a874a63eSLiu Bo 164172fd032eSSage Weil namelen = strlen(name); 164272fd032eSSage Weil if (strchr(name, '/')) { 1643f46b5a66SChristoph Hellwig ret = -EINVAL; 1644a874a63eSLiu Bo goto out_drop_write; 1645f46b5a66SChristoph Hellwig } 1646f46b5a66SChristoph Hellwig 164716780cabSChris Mason if (name[0] == '.' && 164816780cabSChris Mason (namelen == 1 || (name[1] == '.' && namelen == 2))) { 164916780cabSChris Mason ret = -EEXIST; 1650a874a63eSLiu Bo goto out_drop_write; 165116780cabSChris Mason } 165216780cabSChris Mason 16533de4586cSChris Mason if (subvol) { 165472fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 16556f72c7e2SArne Jansen NULL, transid, readonly, inherit); 1656cb8e7090SChristoph Hellwig } else { 16572903ff01SAl Viro struct fd src = fdget(fd); 16583de4586cSChris Mason struct inode *src_inode; 16592903ff01SAl Viro if (!src.file) { 16603de4586cSChris Mason ret = -EINVAL; 1661a874a63eSLiu Bo goto out_drop_write; 16623de4586cSChris Mason } 16633de4586cSChris Mason 1664496ad9aaSAl Viro src_inode = file_inode(src.file); 1665496ad9aaSAl Viro if (src_inode->i_sb != file_inode(file)->i_sb) { 1666c79b4713SJosef Bacik btrfs_info(BTRFS_I(file_inode(file))->root->fs_info, 1667efe120a0SFrank Holton "Snapshot src from another FS"); 166823ad5b17SKusanagi Kouichi ret = -EXDEV; 1669d0242061SDavid Sterba } else if (!inode_owner_or_capable(src_inode)) { 1670d0242061SDavid Sterba /* 1671d0242061SDavid Sterba * Subvolume creation is not restricted, but snapshots 1672d0242061SDavid Sterba * are limited to own subvolumes only 1673d0242061SDavid Sterba */ 1674d0242061SDavid Sterba ret = -EPERM; 1675ecd18815SAl Viro } else { 167672fd032eSSage Weil ret = btrfs_mksubvol(&file->f_path, name, namelen, 167772fd032eSSage Weil BTRFS_I(src_inode)->root, 16786f72c7e2SArne Jansen transid, readonly, inherit); 1679ecd18815SAl Viro } 16802903ff01SAl Viro fdput(src); 1681cb8e7090SChristoph Hellwig } 1682a874a63eSLiu Bo out_drop_write: 1683a874a63eSLiu Bo mnt_drop_write_file(file); 1684f46b5a66SChristoph Hellwig out: 168572fd032eSSage Weil return ret; 168672fd032eSSage Weil } 168772fd032eSSage Weil 168872fd032eSSage Weil static noinline int btrfs_ioctl_snap_create(struct file *file, 1689fa0d2b9bSLi Zefan void __user *arg, int subvol) 169072fd032eSSage Weil { 1691fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args *vol_args; 169272fd032eSSage Weil int ret; 169372fd032eSSage Weil 1694fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1695fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1696fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1697fa0d2b9bSLi Zefan vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 1698fa0d2b9bSLi Zefan 1699fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 1700b83cc969SLi Zefan vol_args->fd, subvol, 17016f72c7e2SArne Jansen NULL, false, NULL); 1702fa0d2b9bSLi Zefan 1703fa0d2b9bSLi Zefan kfree(vol_args); 1704fa0d2b9bSLi Zefan return ret; 1705fa0d2b9bSLi Zefan } 1706fa0d2b9bSLi Zefan 1707fa0d2b9bSLi Zefan static noinline int btrfs_ioctl_snap_create_v2(struct file *file, 1708fa0d2b9bSLi Zefan void __user *arg, int subvol) 1709fa0d2b9bSLi Zefan { 1710fa0d2b9bSLi Zefan struct btrfs_ioctl_vol_args_v2 *vol_args; 1711fa0d2b9bSLi Zefan int ret; 1712fdfb1e4fSLi Zefan u64 transid = 0; 1713fdfb1e4fSLi Zefan u64 *ptr = NULL; 1714b83cc969SLi Zefan bool readonly = false; 17156f72c7e2SArne Jansen struct btrfs_qgroup_inherit *inherit = NULL; 171672fd032eSSage Weil 1717fa0d2b9bSLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 1718fa0d2b9bSLi Zefan if (IS_ERR(vol_args)) 1719fa0d2b9bSLi Zefan return PTR_ERR(vol_args); 1720fa0d2b9bSLi Zefan vol_args->name[BTRFS_SUBVOL_NAME_MAX] = '\0'; 1721fdfb1e4fSLi Zefan 1722b83cc969SLi Zefan if (vol_args->flags & 17236f72c7e2SArne Jansen ~(BTRFS_SUBVOL_CREATE_ASYNC | BTRFS_SUBVOL_RDONLY | 17246f72c7e2SArne Jansen BTRFS_SUBVOL_QGROUP_INHERIT)) { 1725b83cc969SLi Zefan ret = -EOPNOTSUPP; 1726c47ca32dSDan Carpenter goto free_args; 1727fdfb1e4fSLi Zefan } 1728fdfb1e4fSLi Zefan 1729fa0d2b9bSLi Zefan if (vol_args->flags & BTRFS_SUBVOL_CREATE_ASYNC) 1730fdfb1e4fSLi Zefan ptr = &transid; 1731b83cc969SLi Zefan if (vol_args->flags & BTRFS_SUBVOL_RDONLY) 1732b83cc969SLi Zefan readonly = true; 17336f72c7e2SArne Jansen if (vol_args->flags & BTRFS_SUBVOL_QGROUP_INHERIT) { 173409cbfeafSKirill A. Shutemov if (vol_args->size > PAGE_SIZE) { 17356f72c7e2SArne Jansen ret = -EINVAL; 1736c47ca32dSDan Carpenter goto free_args; 17376f72c7e2SArne Jansen } 17386f72c7e2SArne Jansen inherit = memdup_user(vol_args->qgroup_inherit, vol_args->size); 17396f72c7e2SArne Jansen if (IS_ERR(inherit)) { 17406f72c7e2SArne Jansen ret = PTR_ERR(inherit); 1741c47ca32dSDan Carpenter goto free_args; 17426f72c7e2SArne Jansen } 17436f72c7e2SArne Jansen } 174475eaa0e2SSage Weil 1745fa0d2b9bSLi Zefan ret = btrfs_ioctl_snap_create_transid(file, vol_args->name, 17466f72c7e2SArne Jansen vol_args->fd, subvol, ptr, 17478696c533SMiao Xie readonly, inherit); 1748c47ca32dSDan Carpenter if (ret) 1749c47ca32dSDan Carpenter goto free_inherit; 175075eaa0e2SSage Weil 1751c47ca32dSDan Carpenter if (ptr && copy_to_user(arg + 1752fdfb1e4fSLi Zefan offsetof(struct btrfs_ioctl_vol_args_v2, 1753c47ca32dSDan Carpenter transid), 1754c47ca32dSDan Carpenter ptr, sizeof(*ptr))) 175575eaa0e2SSage Weil ret = -EFAULT; 1756c47ca32dSDan Carpenter 1757c47ca32dSDan Carpenter free_inherit: 17586f72c7e2SArne Jansen kfree(inherit); 1759c47ca32dSDan Carpenter free_args: 1760c47ca32dSDan Carpenter kfree(vol_args); 1761f46b5a66SChristoph Hellwig return ret; 1762f46b5a66SChristoph Hellwig } 1763f46b5a66SChristoph Hellwig 17640caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_getflags(struct file *file, 17650caa102dSLi Zefan void __user *arg) 17660caa102dSLi Zefan { 1767496ad9aaSAl Viro struct inode *inode = file_inode(file); 17680caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 17690caa102dSLi Zefan int ret = 0; 17700caa102dSLi Zefan u64 flags = 0; 17710caa102dSLi Zefan 177233345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) 17730caa102dSLi Zefan return -EINVAL; 17740caa102dSLi Zefan 17750caa102dSLi Zefan down_read(&root->fs_info->subvol_sem); 17760caa102dSLi Zefan if (btrfs_root_readonly(root)) 17770caa102dSLi Zefan flags |= BTRFS_SUBVOL_RDONLY; 17780caa102dSLi Zefan up_read(&root->fs_info->subvol_sem); 17790caa102dSLi Zefan 17800caa102dSLi Zefan if (copy_to_user(arg, &flags, sizeof(flags))) 17810caa102dSLi Zefan ret = -EFAULT; 17820caa102dSLi Zefan 17830caa102dSLi Zefan return ret; 17840caa102dSLi Zefan } 17850caa102dSLi Zefan 17860caa102dSLi Zefan static noinline int btrfs_ioctl_subvol_setflags(struct file *file, 17870caa102dSLi Zefan void __user *arg) 17880caa102dSLi Zefan { 1789496ad9aaSAl Viro struct inode *inode = file_inode(file); 17900caa102dSLi Zefan struct btrfs_root *root = BTRFS_I(inode)->root; 17910caa102dSLi Zefan struct btrfs_trans_handle *trans; 17920caa102dSLi Zefan u64 root_flags; 17930caa102dSLi Zefan u64 flags; 17940caa102dSLi Zefan int ret = 0; 17950caa102dSLi Zefan 1796bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 1797bd60ea0fSDavid Sterba return -EPERM; 1798bd60ea0fSDavid Sterba 1799b9ca0664SLiu Bo ret = mnt_want_write_file(file); 1800b9ca0664SLiu Bo if (ret) 1801b9ca0664SLiu Bo goto out; 18020caa102dSLi Zefan 1803b9ca0664SLiu Bo if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 1804b9ca0664SLiu Bo ret = -EINVAL; 1805b9ca0664SLiu Bo goto out_drop_write; 1806b9ca0664SLiu Bo } 18070caa102dSLi Zefan 1808b9ca0664SLiu Bo if (copy_from_user(&flags, arg, sizeof(flags))) { 1809b9ca0664SLiu Bo ret = -EFAULT; 1810b9ca0664SLiu Bo goto out_drop_write; 1811b9ca0664SLiu Bo } 18120caa102dSLi Zefan 1813b9ca0664SLiu Bo if (flags & BTRFS_SUBVOL_CREATE_ASYNC) { 1814b9ca0664SLiu Bo ret = -EINVAL; 1815b9ca0664SLiu Bo goto out_drop_write; 1816b9ca0664SLiu Bo } 18170caa102dSLi Zefan 1818b9ca0664SLiu Bo if (flags & ~BTRFS_SUBVOL_RDONLY) { 1819b9ca0664SLiu Bo ret = -EOPNOTSUPP; 1820b9ca0664SLiu Bo goto out_drop_write; 1821b9ca0664SLiu Bo } 18220caa102dSLi Zefan 18230caa102dSLi Zefan down_write(&root->fs_info->subvol_sem); 18240caa102dSLi Zefan 18250caa102dSLi Zefan /* nothing to do */ 18260caa102dSLi Zefan if (!!(flags & BTRFS_SUBVOL_RDONLY) == btrfs_root_readonly(root)) 1827b9ca0664SLiu Bo goto out_drop_sem; 18280caa102dSLi Zefan 18290caa102dSLi Zefan root_flags = btrfs_root_flags(&root->root_item); 18302c686537SDavid Sterba if (flags & BTRFS_SUBVOL_RDONLY) { 18310caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 18320caa102dSLi Zefan root_flags | BTRFS_ROOT_SUBVOL_RDONLY); 18332c686537SDavid Sterba } else { 18342c686537SDavid Sterba /* 18352c686537SDavid Sterba * Block RO -> RW transition if this subvolume is involved in 18362c686537SDavid Sterba * send 18372c686537SDavid Sterba */ 18382c686537SDavid Sterba spin_lock(&root->root_item_lock); 18392c686537SDavid Sterba if (root->send_in_progress == 0) { 18400caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, 18410caa102dSLi Zefan root_flags & ~BTRFS_ROOT_SUBVOL_RDONLY); 18422c686537SDavid Sterba spin_unlock(&root->root_item_lock); 18432c686537SDavid Sterba } else { 18442c686537SDavid Sterba spin_unlock(&root->root_item_lock); 18452c686537SDavid Sterba btrfs_warn(root->fs_info, 18462c686537SDavid Sterba "Attempt to set subvolume %llu read-write during send", 18472c686537SDavid Sterba root->root_key.objectid); 18482c686537SDavid Sterba ret = -EPERM; 18492c686537SDavid Sterba goto out_drop_sem; 18502c686537SDavid Sterba } 18512c686537SDavid Sterba } 18520caa102dSLi Zefan 18530caa102dSLi Zefan trans = btrfs_start_transaction(root, 1); 18540caa102dSLi Zefan if (IS_ERR(trans)) { 18550caa102dSLi Zefan ret = PTR_ERR(trans); 18560caa102dSLi Zefan goto out_reset; 18570caa102dSLi Zefan } 18580caa102dSLi Zefan 1859b4dc2b8cSLi Zefan ret = btrfs_update_root(trans, root->fs_info->tree_root, 18600caa102dSLi Zefan &root->root_key, &root->root_item); 18610caa102dSLi Zefan 18620caa102dSLi Zefan btrfs_commit_transaction(trans, root); 18630caa102dSLi Zefan out_reset: 18640caa102dSLi Zefan if (ret) 18650caa102dSLi Zefan btrfs_set_root_flags(&root->root_item, root_flags); 1866b9ca0664SLiu Bo out_drop_sem: 18670caa102dSLi Zefan up_write(&root->fs_info->subvol_sem); 1868b9ca0664SLiu Bo out_drop_write: 1869b9ca0664SLiu Bo mnt_drop_write_file(file); 1870b9ca0664SLiu Bo out: 18710caa102dSLi Zefan return ret; 18720caa102dSLi Zefan } 18730caa102dSLi Zefan 187476dda93cSYan, Zheng /* 187576dda93cSYan, Zheng * helper to check if the subvolume references other subvolumes 187676dda93cSYan, Zheng */ 187776dda93cSYan, Zheng static noinline int may_destroy_subvol(struct btrfs_root *root) 187876dda93cSYan, Zheng { 187976dda93cSYan, Zheng struct btrfs_path *path; 1880175a2b87SJosef Bacik struct btrfs_dir_item *di; 188176dda93cSYan, Zheng struct btrfs_key key; 1882175a2b87SJosef Bacik u64 dir_id; 188376dda93cSYan, Zheng int ret; 188476dda93cSYan, Zheng 188576dda93cSYan, Zheng path = btrfs_alloc_path(); 188676dda93cSYan, Zheng if (!path) 188776dda93cSYan, Zheng return -ENOMEM; 188876dda93cSYan, Zheng 1889175a2b87SJosef Bacik /* Make sure this root isn't set as the default subvol */ 1890175a2b87SJosef Bacik dir_id = btrfs_super_root_dir(root->fs_info->super_copy); 1891175a2b87SJosef Bacik di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root, path, 1892175a2b87SJosef Bacik dir_id, "default", 7, 0); 1893175a2b87SJosef Bacik if (di && !IS_ERR(di)) { 1894175a2b87SJosef Bacik btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key); 1895175a2b87SJosef Bacik if (key.objectid == root->root_key.objectid) { 189672de6b53SGuangyu Sun ret = -EPERM; 189772de6b53SGuangyu Sun btrfs_err(root->fs_info, "deleting default subvolume " 189872de6b53SGuangyu Sun "%llu is not allowed", key.objectid); 1899175a2b87SJosef Bacik goto out; 1900175a2b87SJosef Bacik } 1901175a2b87SJosef Bacik btrfs_release_path(path); 1902175a2b87SJosef Bacik } 1903175a2b87SJosef Bacik 190476dda93cSYan, Zheng key.objectid = root->root_key.objectid; 190576dda93cSYan, Zheng key.type = BTRFS_ROOT_REF_KEY; 190676dda93cSYan, Zheng key.offset = (u64)-1; 190776dda93cSYan, Zheng 190876dda93cSYan, Zheng ret = btrfs_search_slot(NULL, root->fs_info->tree_root, 190976dda93cSYan, Zheng &key, path, 0, 0); 191076dda93cSYan, Zheng if (ret < 0) 191176dda93cSYan, Zheng goto out; 191276dda93cSYan, Zheng BUG_ON(ret == 0); 191376dda93cSYan, Zheng 191476dda93cSYan, Zheng ret = 0; 191576dda93cSYan, Zheng if (path->slots[0] > 0) { 191676dda93cSYan, Zheng path->slots[0]--; 191776dda93cSYan, Zheng btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 191876dda93cSYan, Zheng if (key.objectid == root->root_key.objectid && 191976dda93cSYan, Zheng key.type == BTRFS_ROOT_REF_KEY) 192076dda93cSYan, Zheng ret = -ENOTEMPTY; 192176dda93cSYan, Zheng } 192276dda93cSYan, Zheng out: 192376dda93cSYan, Zheng btrfs_free_path(path); 192476dda93cSYan, Zheng return ret; 192576dda93cSYan, Zheng } 192676dda93cSYan, Zheng 1927ac8e9819SChris Mason static noinline int key_in_sk(struct btrfs_key *key, 1928ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk) 1929ac8e9819SChris Mason { 1930abc6e134SChris Mason struct btrfs_key test; 1931abc6e134SChris Mason int ret; 1932abc6e134SChris Mason 1933abc6e134SChris Mason test.objectid = sk->min_objectid; 1934abc6e134SChris Mason test.type = sk->min_type; 1935abc6e134SChris Mason test.offset = sk->min_offset; 1936abc6e134SChris Mason 1937abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1938abc6e134SChris Mason if (ret < 0) 1939ac8e9819SChris Mason return 0; 1940abc6e134SChris Mason 1941abc6e134SChris Mason test.objectid = sk->max_objectid; 1942abc6e134SChris Mason test.type = sk->max_type; 1943abc6e134SChris Mason test.offset = sk->max_offset; 1944abc6e134SChris Mason 1945abc6e134SChris Mason ret = btrfs_comp_cpu_keys(key, &test); 1946abc6e134SChris Mason if (ret > 0) 1947ac8e9819SChris Mason return 0; 1948ac8e9819SChris Mason return 1; 1949ac8e9819SChris Mason } 1950ac8e9819SChris Mason 1951ac8e9819SChris Mason static noinline int copy_to_sk(struct btrfs_root *root, 1952ac8e9819SChris Mason struct btrfs_path *path, 1953ac8e9819SChris Mason struct btrfs_key *key, 1954ac8e9819SChris Mason struct btrfs_ioctl_search_key *sk, 19559b6e817dSGerhard Heift size_t *buf_size, 1956ba346b35SGerhard Heift char __user *ubuf, 1957ac8e9819SChris Mason unsigned long *sk_offset, 1958ac8e9819SChris Mason int *num_found) 1959ac8e9819SChris Mason { 1960ac8e9819SChris Mason u64 found_transid; 1961ac8e9819SChris Mason struct extent_buffer *leaf; 1962ac8e9819SChris Mason struct btrfs_ioctl_search_header sh; 1963dd81d459SNaohiro Aota struct btrfs_key test; 1964ac8e9819SChris Mason unsigned long item_off; 1965ac8e9819SChris Mason unsigned long item_len; 1966ac8e9819SChris Mason int nritems; 1967ac8e9819SChris Mason int i; 1968ac8e9819SChris Mason int slot; 1969ac8e9819SChris Mason int ret = 0; 1970ac8e9819SChris Mason 1971ac8e9819SChris Mason leaf = path->nodes[0]; 1972ac8e9819SChris Mason slot = path->slots[0]; 1973ac8e9819SChris Mason nritems = btrfs_header_nritems(leaf); 1974ac8e9819SChris Mason 1975ac8e9819SChris Mason if (btrfs_header_generation(leaf) > sk->max_transid) { 1976ac8e9819SChris Mason i = nritems; 1977ac8e9819SChris Mason goto advance_key; 1978ac8e9819SChris Mason } 1979ac8e9819SChris Mason found_transid = btrfs_header_generation(leaf); 1980ac8e9819SChris Mason 1981ac8e9819SChris Mason for (i = slot; i < nritems; i++) { 1982ac8e9819SChris Mason item_off = btrfs_item_ptr_offset(leaf, i); 1983ac8e9819SChris Mason item_len = btrfs_item_size_nr(leaf, i); 1984ac8e9819SChris Mason 198503b71c6cSGabriel de Perthuis btrfs_item_key_to_cpu(leaf, key, i); 198603b71c6cSGabriel de Perthuis if (!key_in_sk(key, sk)) 198703b71c6cSGabriel de Perthuis continue; 198803b71c6cSGabriel de Perthuis 19899b6e817dSGerhard Heift if (sizeof(sh) + item_len > *buf_size) { 19908f5f6178SGerhard Heift if (*num_found) { 1991ac8e9819SChris Mason ret = 1; 19928f5f6178SGerhard Heift goto out; 19938f5f6178SGerhard Heift } 19948f5f6178SGerhard Heift 19958f5f6178SGerhard Heift /* 19968f5f6178SGerhard Heift * return one empty item back for v1, which does not 19978f5f6178SGerhard Heift * handle -EOVERFLOW 19988f5f6178SGerhard Heift */ 19998f5f6178SGerhard Heift 20009b6e817dSGerhard Heift *buf_size = sizeof(sh) + item_len; 2001ac8e9819SChris Mason item_len = 0; 20028f5f6178SGerhard Heift ret = -EOVERFLOW; 20038f5f6178SGerhard Heift } 2004ac8e9819SChris Mason 20059b6e817dSGerhard Heift if (sizeof(sh) + item_len + *sk_offset > *buf_size) { 2006ac8e9819SChris Mason ret = 1; 200725c9bc2eSGerhard Heift goto out; 2008ac8e9819SChris Mason } 2009ac8e9819SChris Mason 2010ac8e9819SChris Mason sh.objectid = key->objectid; 2011ac8e9819SChris Mason sh.offset = key->offset; 2012ac8e9819SChris Mason sh.type = key->type; 2013ac8e9819SChris Mason sh.len = item_len; 2014ac8e9819SChris Mason sh.transid = found_transid; 2015ac8e9819SChris Mason 2016ac8e9819SChris Mason /* copy search result header */ 2017ba346b35SGerhard Heift if (copy_to_user(ubuf + *sk_offset, &sh, sizeof(sh))) { 2018ba346b35SGerhard Heift ret = -EFAULT; 2019ba346b35SGerhard Heift goto out; 2020ba346b35SGerhard Heift } 2021ba346b35SGerhard Heift 2022ac8e9819SChris Mason *sk_offset += sizeof(sh); 2023ac8e9819SChris Mason 2024ac8e9819SChris Mason if (item_len) { 2025ba346b35SGerhard Heift char __user *up = ubuf + *sk_offset; 2026ac8e9819SChris Mason /* copy the item */ 2027ba346b35SGerhard Heift if (read_extent_buffer_to_user(leaf, up, 2028ba346b35SGerhard Heift item_off, item_len)) { 2029ba346b35SGerhard Heift ret = -EFAULT; 2030ba346b35SGerhard Heift goto out; 2031ba346b35SGerhard Heift } 2032ba346b35SGerhard Heift 2033ac8e9819SChris Mason *sk_offset += item_len; 2034ac8e9819SChris Mason } 2035e2156867SHugo Mills (*num_found)++; 2036ac8e9819SChris Mason 20378f5f6178SGerhard Heift if (ret) /* -EOVERFLOW from above */ 20388f5f6178SGerhard Heift goto out; 20398f5f6178SGerhard Heift 204025c9bc2eSGerhard Heift if (*num_found >= sk->nr_items) { 204125c9bc2eSGerhard Heift ret = 1; 204225c9bc2eSGerhard Heift goto out; 204325c9bc2eSGerhard Heift } 2044ac8e9819SChris Mason } 2045ac8e9819SChris Mason advance_key: 2046ac8e9819SChris Mason ret = 0; 2047dd81d459SNaohiro Aota test.objectid = sk->max_objectid; 2048dd81d459SNaohiro Aota test.type = sk->max_type; 2049dd81d459SNaohiro Aota test.offset = sk->max_offset; 2050dd81d459SNaohiro Aota if (btrfs_comp_cpu_keys(key, &test) >= 0) 2051dd81d459SNaohiro Aota ret = 1; 2052dd81d459SNaohiro Aota else if (key->offset < (u64)-1) 2053abc6e134SChris Mason key->offset++; 2054dd81d459SNaohiro Aota else if (key->type < (u8)-1) { 2055abc6e134SChris Mason key->offset = 0; 2056abc6e134SChris Mason key->type++; 2057dd81d459SNaohiro Aota } else if (key->objectid < (u64)-1) { 2058abc6e134SChris Mason key->offset = 0; 2059abc6e134SChris Mason key->type = 0; 2060abc6e134SChris Mason key->objectid++; 2061abc6e134SChris Mason } else 2062abc6e134SChris Mason ret = 1; 206325c9bc2eSGerhard Heift out: 2064ba346b35SGerhard Heift /* 2065ba346b35SGerhard Heift * 0: all items from this leaf copied, continue with next 2066ba346b35SGerhard Heift * 1: * more items can be copied, but unused buffer is too small 2067ba346b35SGerhard Heift * * all items were found 2068ba346b35SGerhard Heift * Either way, it will stops the loop which iterates to the next 2069ba346b35SGerhard Heift * leaf 2070ba346b35SGerhard Heift * -EOVERFLOW: item was to large for buffer 2071ba346b35SGerhard Heift * -EFAULT: could not copy extent buffer back to userspace 2072ba346b35SGerhard Heift */ 2073ac8e9819SChris Mason return ret; 2074ac8e9819SChris Mason } 2075ac8e9819SChris Mason 2076ac8e9819SChris Mason static noinline int search_ioctl(struct inode *inode, 207712544442SGerhard Heift struct btrfs_ioctl_search_key *sk, 20789b6e817dSGerhard Heift size_t *buf_size, 2079ba346b35SGerhard Heift char __user *ubuf) 2080ac8e9819SChris Mason { 2081ac8e9819SChris Mason struct btrfs_root *root; 2082ac8e9819SChris Mason struct btrfs_key key; 2083ac8e9819SChris Mason struct btrfs_path *path; 2084ac8e9819SChris Mason struct btrfs_fs_info *info = BTRFS_I(inode)->root->fs_info; 2085ac8e9819SChris Mason int ret; 2086ac8e9819SChris Mason int num_found = 0; 2087ac8e9819SChris Mason unsigned long sk_offset = 0; 2088ac8e9819SChris Mason 20899b6e817dSGerhard Heift if (*buf_size < sizeof(struct btrfs_ioctl_search_header)) { 20909b6e817dSGerhard Heift *buf_size = sizeof(struct btrfs_ioctl_search_header); 209112544442SGerhard Heift return -EOVERFLOW; 20929b6e817dSGerhard Heift } 209312544442SGerhard Heift 2094ac8e9819SChris Mason path = btrfs_alloc_path(); 2095ac8e9819SChris Mason if (!path) 2096ac8e9819SChris Mason return -ENOMEM; 2097ac8e9819SChris Mason 2098ac8e9819SChris Mason if (sk->tree_id == 0) { 2099ac8e9819SChris Mason /* search the root of the inode that was passed */ 2100ac8e9819SChris Mason root = BTRFS_I(inode)->root; 2101ac8e9819SChris Mason } else { 2102ac8e9819SChris Mason key.objectid = sk->tree_id; 2103ac8e9819SChris Mason key.type = BTRFS_ROOT_ITEM_KEY; 2104ac8e9819SChris Mason key.offset = (u64)-1; 2105ac8e9819SChris Mason root = btrfs_read_fs_root_no_name(info, &key); 2106ac8e9819SChris Mason if (IS_ERR(root)) { 2107ac8e9819SChris Mason btrfs_free_path(path); 2108ac8e9819SChris Mason return -ENOENT; 2109ac8e9819SChris Mason } 2110ac8e9819SChris Mason } 2111ac8e9819SChris Mason 2112ac8e9819SChris Mason key.objectid = sk->min_objectid; 2113ac8e9819SChris Mason key.type = sk->min_type; 2114ac8e9819SChris Mason key.offset = sk->min_offset; 2115ac8e9819SChris Mason 2116ac8e9819SChris Mason while (1) { 21176174d3cbSFilipe David Borba Manana ret = btrfs_search_forward(root, &key, path, sk->min_transid); 2118ac8e9819SChris Mason if (ret != 0) { 2119ac8e9819SChris Mason if (ret > 0) 2120ac8e9819SChris Mason ret = 0; 2121ac8e9819SChris Mason goto err; 2122ac8e9819SChris Mason } 2123ba346b35SGerhard Heift ret = copy_to_sk(root, path, &key, sk, buf_size, ubuf, 2124ac8e9819SChris Mason &sk_offset, &num_found); 2125b3b4aa74SDavid Sterba btrfs_release_path(path); 212625c9bc2eSGerhard Heift if (ret) 2127ac8e9819SChris Mason break; 2128ac8e9819SChris Mason 2129ac8e9819SChris Mason } 21308f5f6178SGerhard Heift if (ret > 0) 2131ac8e9819SChris Mason ret = 0; 2132ac8e9819SChris Mason err: 2133ac8e9819SChris Mason sk->nr_items = num_found; 2134ac8e9819SChris Mason btrfs_free_path(path); 2135ac8e9819SChris Mason return ret; 2136ac8e9819SChris Mason } 2137ac8e9819SChris Mason 2138ac8e9819SChris Mason static noinline int btrfs_ioctl_tree_search(struct file *file, 2139ac8e9819SChris Mason void __user *argp) 2140ac8e9819SChris Mason { 2141ba346b35SGerhard Heift struct btrfs_ioctl_search_args __user *uargs; 2142ba346b35SGerhard Heift struct btrfs_ioctl_search_key sk; 2143ac8e9819SChris Mason struct inode *inode; 2144ac8e9819SChris Mason int ret; 21459b6e817dSGerhard Heift size_t buf_size; 2146ac8e9819SChris Mason 2147ac8e9819SChris Mason if (!capable(CAP_SYS_ADMIN)) 2148ac8e9819SChris Mason return -EPERM; 2149ac8e9819SChris Mason 2150ba346b35SGerhard Heift uargs = (struct btrfs_ioctl_search_args __user *)argp; 2151ac8e9819SChris Mason 2152ba346b35SGerhard Heift if (copy_from_user(&sk, &uargs->key, sizeof(sk))) 2153ba346b35SGerhard Heift return -EFAULT; 2154ba346b35SGerhard Heift 2155ba346b35SGerhard Heift buf_size = sizeof(uargs->buf); 2156ac8e9819SChris Mason 2157496ad9aaSAl Viro inode = file_inode(file); 2158ba346b35SGerhard Heift ret = search_ioctl(inode, &sk, &buf_size, uargs->buf); 21598f5f6178SGerhard Heift 21608f5f6178SGerhard Heift /* 21618f5f6178SGerhard Heift * In the origin implementation an overflow is handled by returning a 21628f5f6178SGerhard Heift * search header with a len of zero, so reset ret. 21638f5f6178SGerhard Heift */ 21648f5f6178SGerhard Heift if (ret == -EOVERFLOW) 21658f5f6178SGerhard Heift ret = 0; 21668f5f6178SGerhard Heift 2167ba346b35SGerhard Heift if (ret == 0 && copy_to_user(&uargs->key, &sk, sizeof(sk))) 2168ac8e9819SChris Mason ret = -EFAULT; 2169ac8e9819SChris Mason return ret; 2170ac8e9819SChris Mason } 2171ac8e9819SChris Mason 2172cc68a8a5SGerhard Heift static noinline int btrfs_ioctl_tree_search_v2(struct file *file, 2173cc68a8a5SGerhard Heift void __user *argp) 2174cc68a8a5SGerhard Heift { 2175cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 __user *uarg; 2176cc68a8a5SGerhard Heift struct btrfs_ioctl_search_args_v2 args; 2177cc68a8a5SGerhard Heift struct inode *inode; 2178cc68a8a5SGerhard Heift int ret; 2179cc68a8a5SGerhard Heift size_t buf_size; 2180ee22184bSByongho Lee const size_t buf_limit = SZ_16M; 2181cc68a8a5SGerhard Heift 2182cc68a8a5SGerhard Heift if (!capable(CAP_SYS_ADMIN)) 2183cc68a8a5SGerhard Heift return -EPERM; 2184cc68a8a5SGerhard Heift 2185cc68a8a5SGerhard Heift /* copy search header and buffer size */ 2186cc68a8a5SGerhard Heift uarg = (struct btrfs_ioctl_search_args_v2 __user *)argp; 2187cc68a8a5SGerhard Heift if (copy_from_user(&args, uarg, sizeof(args))) 2188cc68a8a5SGerhard Heift return -EFAULT; 2189cc68a8a5SGerhard Heift 2190cc68a8a5SGerhard Heift buf_size = args.buf_size; 2191cc68a8a5SGerhard Heift 2192cc68a8a5SGerhard Heift if (buf_size < sizeof(struct btrfs_ioctl_search_header)) 2193cc68a8a5SGerhard Heift return -EOVERFLOW; 2194cc68a8a5SGerhard Heift 2195cc68a8a5SGerhard Heift /* limit result size to 16MB */ 2196cc68a8a5SGerhard Heift if (buf_size > buf_limit) 2197cc68a8a5SGerhard Heift buf_size = buf_limit; 2198cc68a8a5SGerhard Heift 2199cc68a8a5SGerhard Heift inode = file_inode(file); 2200cc68a8a5SGerhard Heift ret = search_ioctl(inode, &args.key, &buf_size, 2201cc68a8a5SGerhard Heift (char *)(&uarg->buf[0])); 2202cc68a8a5SGerhard Heift if (ret == 0 && copy_to_user(&uarg->key, &args.key, sizeof(args.key))) 2203cc68a8a5SGerhard Heift ret = -EFAULT; 2204cc68a8a5SGerhard Heift else if (ret == -EOVERFLOW && 2205cc68a8a5SGerhard Heift copy_to_user(&uarg->buf_size, &buf_size, sizeof(buf_size))) 2206cc68a8a5SGerhard Heift ret = -EFAULT; 2207cc68a8a5SGerhard Heift 220876dda93cSYan, Zheng return ret; 220976dda93cSYan, Zheng } 221076dda93cSYan, Zheng 221198d377a0STARUISI Hiroaki /* 2212ac8e9819SChris Mason * Search INODE_REFs to identify path name of 'dirid' directory 2213ac8e9819SChris Mason * in a 'tree_id' tree. and sets path name to 'name'. 221498d377a0STARUISI Hiroaki */ 221598d377a0STARUISI Hiroaki static noinline int btrfs_search_path_in_tree(struct btrfs_fs_info *info, 221698d377a0STARUISI Hiroaki u64 tree_id, u64 dirid, char *name) 221798d377a0STARUISI Hiroaki { 221898d377a0STARUISI Hiroaki struct btrfs_root *root; 221998d377a0STARUISI Hiroaki struct btrfs_key key; 2220ac8e9819SChris Mason char *ptr; 222198d377a0STARUISI Hiroaki int ret = -1; 222298d377a0STARUISI Hiroaki int slot; 222398d377a0STARUISI Hiroaki int len; 222498d377a0STARUISI Hiroaki int total_len = 0; 222598d377a0STARUISI Hiroaki struct btrfs_inode_ref *iref; 222698d377a0STARUISI Hiroaki struct extent_buffer *l; 222798d377a0STARUISI Hiroaki struct btrfs_path *path; 222898d377a0STARUISI Hiroaki 222998d377a0STARUISI Hiroaki if (dirid == BTRFS_FIRST_FREE_OBJECTID) { 223098d377a0STARUISI Hiroaki name[0]='\0'; 223198d377a0STARUISI Hiroaki return 0; 223298d377a0STARUISI Hiroaki } 223398d377a0STARUISI Hiroaki 223498d377a0STARUISI Hiroaki path = btrfs_alloc_path(); 223598d377a0STARUISI Hiroaki if (!path) 223698d377a0STARUISI Hiroaki return -ENOMEM; 223798d377a0STARUISI Hiroaki 2238ac8e9819SChris Mason ptr = &name[BTRFS_INO_LOOKUP_PATH_MAX]; 223998d377a0STARUISI Hiroaki 224098d377a0STARUISI Hiroaki key.objectid = tree_id; 224198d377a0STARUISI Hiroaki key.type = BTRFS_ROOT_ITEM_KEY; 224298d377a0STARUISI Hiroaki key.offset = (u64)-1; 224398d377a0STARUISI Hiroaki root = btrfs_read_fs_root_no_name(info, &key); 224498d377a0STARUISI Hiroaki if (IS_ERR(root)) { 2245f14d104dSDavid Sterba btrfs_err(info, "could not find root %llu", tree_id); 22468ad6fcabSChris Mason ret = -ENOENT; 22478ad6fcabSChris Mason goto out; 224898d377a0STARUISI Hiroaki } 224998d377a0STARUISI Hiroaki 225098d377a0STARUISI Hiroaki key.objectid = dirid; 225198d377a0STARUISI Hiroaki key.type = BTRFS_INODE_REF_KEY; 22528ad6fcabSChris Mason key.offset = (u64)-1; 225398d377a0STARUISI Hiroaki 225498d377a0STARUISI Hiroaki while (1) { 225598d377a0STARUISI Hiroaki ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 225698d377a0STARUISI Hiroaki if (ret < 0) 225798d377a0STARUISI Hiroaki goto out; 225818674c6cSFilipe David Borba Manana else if (ret > 0) { 225918674c6cSFilipe David Borba Manana ret = btrfs_previous_item(root, path, dirid, 226018674c6cSFilipe David Borba Manana BTRFS_INODE_REF_KEY); 226118674c6cSFilipe David Borba Manana if (ret < 0) 226218674c6cSFilipe David Borba Manana goto out; 226318674c6cSFilipe David Borba Manana else if (ret > 0) { 2264ac8e9819SChris Mason ret = -ENOENT; 226598d377a0STARUISI Hiroaki goto out; 2266ac8e9819SChris Mason } 226718674c6cSFilipe David Borba Manana } 226818674c6cSFilipe David Borba Manana 226918674c6cSFilipe David Borba Manana l = path->nodes[0]; 227018674c6cSFilipe David Borba Manana slot = path->slots[0]; 227118674c6cSFilipe David Borba Manana btrfs_item_key_to_cpu(l, &key, slot); 227298d377a0STARUISI Hiroaki 227398d377a0STARUISI Hiroaki iref = btrfs_item_ptr(l, slot, struct btrfs_inode_ref); 227498d377a0STARUISI Hiroaki len = btrfs_inode_ref_name_len(l, iref); 227598d377a0STARUISI Hiroaki ptr -= len + 1; 227698d377a0STARUISI Hiroaki total_len += len + 1; 2277a696cf35SFilipe David Borba Manana if (ptr < name) { 2278a696cf35SFilipe David Borba Manana ret = -ENAMETOOLONG; 227998d377a0STARUISI Hiroaki goto out; 2280a696cf35SFilipe David Borba Manana } 228198d377a0STARUISI Hiroaki 228298d377a0STARUISI Hiroaki *(ptr + len) = '/'; 228398d377a0STARUISI Hiroaki read_extent_buffer(l, ptr, (unsigned long)(iref + 1), len); 228498d377a0STARUISI Hiroaki 228598d377a0STARUISI Hiroaki if (key.offset == BTRFS_FIRST_FREE_OBJECTID) 228698d377a0STARUISI Hiroaki break; 228798d377a0STARUISI Hiroaki 2288b3b4aa74SDavid Sterba btrfs_release_path(path); 228998d377a0STARUISI Hiroaki key.objectid = key.offset; 22908ad6fcabSChris Mason key.offset = (u64)-1; 229198d377a0STARUISI Hiroaki dirid = key.objectid; 229298d377a0STARUISI Hiroaki } 229377906a50SLi Zefan memmove(name, ptr, total_len); 229498d377a0STARUISI Hiroaki name[total_len] = '\0'; 229598d377a0STARUISI Hiroaki ret = 0; 229698d377a0STARUISI Hiroaki out: 229798d377a0STARUISI Hiroaki btrfs_free_path(path); 2298ac8e9819SChris Mason return ret; 2299ac8e9819SChris Mason } 2300ac8e9819SChris Mason 2301ac8e9819SChris Mason static noinline int btrfs_ioctl_ino_lookup(struct file *file, 2302ac8e9819SChris Mason void __user *argp) 2303ac8e9819SChris Mason { 2304ac8e9819SChris Mason struct btrfs_ioctl_ino_lookup_args *args; 2305ac8e9819SChris Mason struct inode *inode; 230601b810b8SDavid Sterba int ret = 0; 2307ac8e9819SChris Mason 23082354d08fSJulia Lawall args = memdup_user(argp, sizeof(*args)); 23092354d08fSJulia Lawall if (IS_ERR(args)) 23102354d08fSJulia Lawall return PTR_ERR(args); 2311c2b96929SDan Carpenter 2312496ad9aaSAl Viro inode = file_inode(file); 2313ac8e9819SChris Mason 231401b810b8SDavid Sterba /* 231501b810b8SDavid Sterba * Unprivileged query to obtain the containing subvolume root id. The 231601b810b8SDavid Sterba * path is reset so it's consistent with btrfs_search_path_in_tree. 231701b810b8SDavid Sterba */ 23181b53ac4dSChris Mason if (args->treeid == 0) 23191b53ac4dSChris Mason args->treeid = BTRFS_I(inode)->root->root_key.objectid; 23201b53ac4dSChris Mason 232101b810b8SDavid Sterba if (args->objectid == BTRFS_FIRST_FREE_OBJECTID) { 232201b810b8SDavid Sterba args->name[0] = 0; 232301b810b8SDavid Sterba goto out; 232401b810b8SDavid Sterba } 232501b810b8SDavid Sterba 232601b810b8SDavid Sterba if (!capable(CAP_SYS_ADMIN)) { 232701b810b8SDavid Sterba ret = -EPERM; 232801b810b8SDavid Sterba goto out; 232901b810b8SDavid Sterba } 233001b810b8SDavid Sterba 2331ac8e9819SChris Mason ret = btrfs_search_path_in_tree(BTRFS_I(inode)->root->fs_info, 2332ac8e9819SChris Mason args->treeid, args->objectid, 2333ac8e9819SChris Mason args->name); 2334ac8e9819SChris Mason 233501b810b8SDavid Sterba out: 2336ac8e9819SChris Mason if (ret == 0 && copy_to_user(argp, args, sizeof(*args))) 2337ac8e9819SChris Mason ret = -EFAULT; 2338ac8e9819SChris Mason 2339ac8e9819SChris Mason kfree(args); 234098d377a0STARUISI Hiroaki return ret; 234198d377a0STARUISI Hiroaki } 234298d377a0STARUISI Hiroaki 234376dda93cSYan, Zheng static noinline int btrfs_ioctl_snap_destroy(struct file *file, 234476dda93cSYan, Zheng void __user *arg) 234576dda93cSYan, Zheng { 234654563d41SAl Viro struct dentry *parent = file->f_path.dentry; 234776dda93cSYan, Zheng struct dentry *dentry; 23482b0143b5SDavid Howells struct inode *dir = d_inode(parent); 234976dda93cSYan, Zheng struct inode *inode; 235076dda93cSYan, Zheng struct btrfs_root *root = BTRFS_I(dir)->root; 235176dda93cSYan, Zheng struct btrfs_root *dest = NULL; 235276dda93cSYan, Zheng struct btrfs_ioctl_vol_args *vol_args; 235376dda93cSYan, Zheng struct btrfs_trans_handle *trans; 2354c58aaad2SMiao Xie struct btrfs_block_rsv block_rsv; 2355521e0546SDavid Sterba u64 root_flags; 2356c58aaad2SMiao Xie u64 qgroup_reserved; 235776dda93cSYan, Zheng int namelen; 235876dda93cSYan, Zheng int ret; 235976dda93cSYan, Zheng int err = 0; 236076dda93cSYan, Zheng 236176dda93cSYan, Zheng vol_args = memdup_user(arg, sizeof(*vol_args)); 236276dda93cSYan, Zheng if (IS_ERR(vol_args)) 236376dda93cSYan, Zheng return PTR_ERR(vol_args); 236476dda93cSYan, Zheng 236576dda93cSYan, Zheng vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 236676dda93cSYan, Zheng namelen = strlen(vol_args->name); 236776dda93cSYan, Zheng if (strchr(vol_args->name, '/') || 236876dda93cSYan, Zheng strncmp(vol_args->name, "..", namelen) == 0) { 236976dda93cSYan, Zheng err = -EINVAL; 237076dda93cSYan, Zheng goto out; 237176dda93cSYan, Zheng } 237276dda93cSYan, Zheng 2373a561be71SAl Viro err = mnt_want_write_file(file); 237476dda93cSYan, Zheng if (err) 237576dda93cSYan, Zheng goto out; 237676dda93cSYan, Zheng 2377521e0546SDavid Sterba 23785c50c9b8SDavid Sterba err = mutex_lock_killable_nested(&dir->i_mutex, I_MUTEX_PARENT); 23795c50c9b8SDavid Sterba if (err == -EINTR) 2380e43f998eSDavid Sterba goto out_drop_write; 238176dda93cSYan, Zheng dentry = lookup_one_len(vol_args->name, parent, namelen); 238276dda93cSYan, Zheng if (IS_ERR(dentry)) { 238376dda93cSYan, Zheng err = PTR_ERR(dentry); 238476dda93cSYan, Zheng goto out_unlock_dir; 238576dda93cSYan, Zheng } 238676dda93cSYan, Zheng 23872b0143b5SDavid Howells if (d_really_is_negative(dentry)) { 238876dda93cSYan, Zheng err = -ENOENT; 238976dda93cSYan, Zheng goto out_dput; 239076dda93cSYan, Zheng } 239176dda93cSYan, Zheng 23922b0143b5SDavid Howells inode = d_inode(dentry); 23934260f7c7SSage Weil dest = BTRFS_I(inode)->root; 23944260f7c7SSage Weil if (!capable(CAP_SYS_ADMIN)) { 23954260f7c7SSage Weil /* 23964260f7c7SSage Weil * Regular user. Only allow this with a special mount 23974260f7c7SSage Weil * option, when the user has write+exec access to the 23984260f7c7SSage Weil * subvol root, and when rmdir(2) would have been 23994260f7c7SSage Weil * allowed. 24004260f7c7SSage Weil * 24014260f7c7SSage Weil * Note that this is _not_ check that the subvol is 24024260f7c7SSage Weil * empty or doesn't contain data that we wouldn't 24034260f7c7SSage Weil * otherwise be able to delete. 24044260f7c7SSage Weil * 24054260f7c7SSage Weil * Users who want to delete empty subvols should try 24064260f7c7SSage Weil * rmdir(2). 24074260f7c7SSage Weil */ 24084260f7c7SSage Weil err = -EPERM; 24094260f7c7SSage Weil if (!btrfs_test_opt(root, USER_SUBVOL_RM_ALLOWED)) 24104260f7c7SSage Weil goto out_dput; 24114260f7c7SSage Weil 24124260f7c7SSage Weil /* 24134260f7c7SSage Weil * Do not allow deletion if the parent dir is the same 24144260f7c7SSage Weil * as the dir to be deleted. That means the ioctl 24154260f7c7SSage Weil * must be called on the dentry referencing the root 24164260f7c7SSage Weil * of the subvol, not a random directory contained 24174260f7c7SSage Weil * within it. 24184260f7c7SSage Weil */ 24194260f7c7SSage Weil err = -EINVAL; 24204260f7c7SSage Weil if (root == dest) 24214260f7c7SSage Weil goto out_dput; 24224260f7c7SSage Weil 24234260f7c7SSage Weil err = inode_permission(inode, MAY_WRITE | MAY_EXEC); 24244260f7c7SSage Weil if (err) 24254260f7c7SSage Weil goto out_dput; 24265c39da5bSMiao Xie } 24274260f7c7SSage Weil 24285c39da5bSMiao Xie /* check if subvolume may be deleted by a user */ 24294260f7c7SSage Weil err = btrfs_may_delete(dir, dentry, 1); 24304260f7c7SSage Weil if (err) 24314260f7c7SSage Weil goto out_dput; 24324260f7c7SSage Weil 243333345d01SLi Zefan if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 243476dda93cSYan, Zheng err = -EINVAL; 243576dda93cSYan, Zheng goto out_dput; 243676dda93cSYan, Zheng } 243776dda93cSYan, Zheng 24385955102cSAl Viro inode_lock(inode); 2439521e0546SDavid Sterba 2440521e0546SDavid Sterba /* 2441521e0546SDavid Sterba * Don't allow to delete a subvolume with send in progress. This is 2442521e0546SDavid Sterba * inside the i_mutex so the error handling that has to drop the bit 2443521e0546SDavid Sterba * again is not run concurrently. 2444521e0546SDavid Sterba */ 2445521e0546SDavid Sterba spin_lock(&dest->root_item_lock); 2446c55bfa67SFilipe Manana root_flags = btrfs_root_flags(&dest->root_item); 2447c55bfa67SFilipe Manana if (dest->send_in_progress == 0) { 2448c55bfa67SFilipe Manana btrfs_set_root_flags(&dest->root_item, 2449521e0546SDavid Sterba root_flags | BTRFS_ROOT_SUBVOL_DEAD); 2450521e0546SDavid Sterba spin_unlock(&dest->root_item_lock); 2451521e0546SDavid Sterba } else { 2452521e0546SDavid Sterba spin_unlock(&dest->root_item_lock); 2453521e0546SDavid Sterba btrfs_warn(root->fs_info, 2454521e0546SDavid Sterba "Attempt to delete subvolume %llu during send", 2455c55bfa67SFilipe Manana dest->root_key.objectid); 2456521e0546SDavid Sterba err = -EPERM; 2457909e26dcSOmar Sandoval goto out_unlock_inode; 2458521e0546SDavid Sterba } 2459521e0546SDavid Sterba 246076dda93cSYan, Zheng down_write(&root->fs_info->subvol_sem); 246176dda93cSYan, Zheng 246276dda93cSYan, Zheng err = may_destroy_subvol(dest); 246376dda93cSYan, Zheng if (err) 246476dda93cSYan, Zheng goto out_up_write; 246576dda93cSYan, Zheng 2466c58aaad2SMiao Xie btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP); 2467c58aaad2SMiao Xie /* 2468c58aaad2SMiao Xie * One for dir inode, two for dir entries, two for root 2469c58aaad2SMiao Xie * ref/backref. 2470c58aaad2SMiao Xie */ 2471c58aaad2SMiao Xie err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 2472ee3441b4SJeff Mahoney 5, &qgroup_reserved, true); 2473c58aaad2SMiao Xie if (err) 2474c58aaad2SMiao Xie goto out_up_write; 2475c58aaad2SMiao Xie 2476a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 0); 2477a22285a6SYan, Zheng if (IS_ERR(trans)) { 2478a22285a6SYan, Zheng err = PTR_ERR(trans); 2479c58aaad2SMiao Xie goto out_release; 2480a22285a6SYan, Zheng } 2481c58aaad2SMiao Xie trans->block_rsv = &block_rsv; 2482c58aaad2SMiao Xie trans->bytes_reserved = block_rsv.size; 2483a22285a6SYan, Zheng 24842be63d5cSFilipe Manana btrfs_record_snapshot_destroy(trans, dir); 24852be63d5cSFilipe Manana 248676dda93cSYan, Zheng ret = btrfs_unlink_subvol(trans, root, dir, 248776dda93cSYan, Zheng dest->root_key.objectid, 248876dda93cSYan, Zheng dentry->d_name.name, 248976dda93cSYan, Zheng dentry->d_name.len); 249079787eaaSJeff Mahoney if (ret) { 249179787eaaSJeff Mahoney err = ret; 249279787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 249379787eaaSJeff Mahoney goto out_end_trans; 249479787eaaSJeff Mahoney } 249576dda93cSYan, Zheng 249676dda93cSYan, Zheng btrfs_record_root_in_trans(trans, dest); 249776dda93cSYan, Zheng 249876dda93cSYan, Zheng memset(&dest->root_item.drop_progress, 0, 249976dda93cSYan, Zheng sizeof(dest->root_item.drop_progress)); 250076dda93cSYan, Zheng dest->root_item.drop_level = 0; 250176dda93cSYan, Zheng btrfs_set_root_refs(&dest->root_item, 0); 250276dda93cSYan, Zheng 250327cdeb70SMiao Xie if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) { 250476dda93cSYan, Zheng ret = btrfs_insert_orphan_item(trans, 250576dda93cSYan, Zheng root->fs_info->tree_root, 250676dda93cSYan, Zheng dest->root_key.objectid); 250779787eaaSJeff Mahoney if (ret) { 250879787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, ret); 250979787eaaSJeff Mahoney err = ret; 251079787eaaSJeff Mahoney goto out_end_trans; 2511d68fc57bSYan, Zheng } 251279787eaaSJeff Mahoney } 2513dd5f9615SStefan Behrens 2514dd5f9615SStefan Behrens ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 2515dd5f9615SStefan Behrens dest->root_item.uuid, BTRFS_UUID_KEY_SUBVOL, 2516dd5f9615SStefan Behrens dest->root_key.objectid); 2517dd5f9615SStefan Behrens if (ret && ret != -ENOENT) { 2518dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 2519dd5f9615SStefan Behrens err = ret; 2520dd5f9615SStefan Behrens goto out_end_trans; 2521dd5f9615SStefan Behrens } 2522dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) { 2523dd5f9615SStefan Behrens ret = btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 2524dd5f9615SStefan Behrens dest->root_item.received_uuid, 2525dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 2526dd5f9615SStefan Behrens dest->root_key.objectid); 2527dd5f9615SStefan Behrens if (ret && ret != -ENOENT) { 2528dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 2529dd5f9615SStefan Behrens err = ret; 2530dd5f9615SStefan Behrens goto out_end_trans; 2531dd5f9615SStefan Behrens } 2532dd5f9615SStefan Behrens } 2533dd5f9615SStefan Behrens 253479787eaaSJeff Mahoney out_end_trans: 2535c58aaad2SMiao Xie trans->block_rsv = NULL; 2536c58aaad2SMiao Xie trans->bytes_reserved = 0; 2537531cb13fSSage Weil ret = btrfs_end_transaction(trans, root); 253879787eaaSJeff Mahoney if (ret && !err) 253979787eaaSJeff Mahoney err = ret; 254076dda93cSYan, Zheng inode->i_flags |= S_DEAD; 2541c58aaad2SMiao Xie out_release: 2542c58aaad2SMiao Xie btrfs_subvolume_release_metadata(root, &block_rsv, qgroup_reserved); 254376dda93cSYan, Zheng out_up_write: 254476dda93cSYan, Zheng up_write(&root->fs_info->subvol_sem); 2545521e0546SDavid Sterba if (err) { 2546521e0546SDavid Sterba spin_lock(&dest->root_item_lock); 2547c55bfa67SFilipe Manana root_flags = btrfs_root_flags(&dest->root_item); 2548c55bfa67SFilipe Manana btrfs_set_root_flags(&dest->root_item, 2549521e0546SDavid Sterba root_flags & ~BTRFS_ROOT_SUBVOL_DEAD); 2550521e0546SDavid Sterba spin_unlock(&dest->root_item_lock); 2551521e0546SDavid Sterba } 2552909e26dcSOmar Sandoval out_unlock_inode: 25535955102cSAl Viro inode_unlock(inode); 255476dda93cSYan, Zheng if (!err) { 255564ad6c48SOmar Sandoval d_invalidate(dentry); 255676dda93cSYan, Zheng btrfs_invalidate_inodes(dest); 255776dda93cSYan, Zheng d_delete(dentry); 255861155aa0SDavid Sterba ASSERT(dest->send_in_progress == 0); 2559fa6ac876SLiu Bo 2560fa6ac876SLiu Bo /* the last ref */ 256157cdc8dbSDavid Sterba if (dest->ino_cache_inode) { 256257cdc8dbSDavid Sterba iput(dest->ino_cache_inode); 256357cdc8dbSDavid Sterba dest->ino_cache_inode = NULL; 2564fa6ac876SLiu Bo } 256576dda93cSYan, Zheng } 256676dda93cSYan, Zheng out_dput: 256776dda93cSYan, Zheng dput(dentry); 256876dda93cSYan, Zheng out_unlock_dir: 25695955102cSAl Viro inode_unlock(dir); 2570e43f998eSDavid Sterba out_drop_write: 25712a79f17eSAl Viro mnt_drop_write_file(file); 257276dda93cSYan, Zheng out: 257376dda93cSYan, Zheng kfree(vol_args); 257476dda93cSYan, Zheng return err; 257576dda93cSYan, Zheng } 257676dda93cSYan, Zheng 25771e701a32SChris Mason static int btrfs_ioctl_defrag(struct file *file, void __user *argp) 2578f46b5a66SChristoph Hellwig { 2579496ad9aaSAl Viro struct inode *inode = file_inode(file); 2580f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 25811e701a32SChris Mason struct btrfs_ioctl_defrag_range_args *range; 2582c146afadSYan Zheng int ret; 2583c146afadSYan Zheng 258425122d15SIlya Dryomov ret = mnt_want_write_file(file); 258525122d15SIlya Dryomov if (ret) 258625122d15SIlya Dryomov return ret; 2587b83cc969SLi Zefan 258825122d15SIlya Dryomov if (btrfs_root_readonly(root)) { 258925122d15SIlya Dryomov ret = -EROFS; 259025122d15SIlya Dryomov goto out; 25915ac00addSStefan Behrens } 2592f46b5a66SChristoph Hellwig 2593f46b5a66SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2594f46b5a66SChristoph Hellwig case S_IFDIR: 2595e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) { 2596e441d54dSChris Mason ret = -EPERM; 2597e441d54dSChris Mason goto out; 2598e441d54dSChris Mason } 2599de78b51aSEric Sandeen ret = btrfs_defrag_root(root); 26008929ecfaSYan, Zheng if (ret) 26018929ecfaSYan, Zheng goto out; 2602de78b51aSEric Sandeen ret = btrfs_defrag_root(root->fs_info->extent_root); 2603f46b5a66SChristoph Hellwig break; 2604f46b5a66SChristoph Hellwig case S_IFREG: 2605e441d54dSChris Mason if (!(file->f_mode & FMODE_WRITE)) { 2606e441d54dSChris Mason ret = -EINVAL; 2607e441d54dSChris Mason goto out; 2608e441d54dSChris Mason } 26091e701a32SChris Mason 26101e701a32SChris Mason range = kzalloc(sizeof(*range), GFP_KERNEL); 26111e701a32SChris Mason if (!range) { 26121e701a32SChris Mason ret = -ENOMEM; 26131e701a32SChris Mason goto out; 26141e701a32SChris Mason } 26151e701a32SChris Mason 26161e701a32SChris Mason if (argp) { 26171e701a32SChris Mason if (copy_from_user(range, argp, 26181e701a32SChris Mason sizeof(*range))) { 26191e701a32SChris Mason ret = -EFAULT; 26201e701a32SChris Mason kfree(range); 2621683be16eSDan Carpenter goto out; 26221e701a32SChris Mason } 26231e701a32SChris Mason /* compression requires us to start the IO */ 26241e701a32SChris Mason if ((range->flags & BTRFS_DEFRAG_RANGE_COMPRESS)) { 26251e701a32SChris Mason range->flags |= BTRFS_DEFRAG_RANGE_START_IO; 26261e701a32SChris Mason range->extent_thresh = (u32)-1; 26271e701a32SChris Mason } 26281e701a32SChris Mason } else { 26291e701a32SChris Mason /* the rest are all set to zero by kzalloc */ 26301e701a32SChris Mason range->len = (u64)-1; 26311e701a32SChris Mason } 2632496ad9aaSAl Viro ret = btrfs_defrag_file(file_inode(file), file, 26334cb5300bSChris Mason range, 0, 0); 26344cb5300bSChris Mason if (ret > 0) 26354cb5300bSChris Mason ret = 0; 26361e701a32SChris Mason kfree(range); 2637f46b5a66SChristoph Hellwig break; 26388929ecfaSYan, Zheng default: 26398929ecfaSYan, Zheng ret = -EINVAL; 2640f46b5a66SChristoph Hellwig } 2641e441d54dSChris Mason out: 264225122d15SIlya Dryomov mnt_drop_write_file(file); 2643e441d54dSChris Mason return ret; 2644f46b5a66SChristoph Hellwig } 2645f46b5a66SChristoph Hellwig 2646b2950863SChristoph Hellwig static long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg) 2647f46b5a66SChristoph Hellwig { 2648f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2649f46b5a66SChristoph Hellwig int ret; 2650f46b5a66SChristoph Hellwig 2651e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2652e441d54dSChris Mason return -EPERM; 2653e441d54dSChris Mason 26545ac00addSStefan Behrens if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 26555ac00addSStefan Behrens 1)) { 2656e57138b3SAnand Jain return BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 2657c9e9f97bSIlya Dryomov } 2658c9e9f97bSIlya Dryomov 26595ac00addSStefan Behrens mutex_lock(&root->fs_info->volume_mutex); 2660dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2661c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2662c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2663c9e9f97bSIlya Dryomov goto out; 2664c9e9f97bSIlya Dryomov } 2665f46b5a66SChristoph Hellwig 26665516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2667f46b5a66SChristoph Hellwig ret = btrfs_init_new_device(root, vol_args->name); 2668f46b5a66SChristoph Hellwig 266943d20761SAnand Jain if (!ret) 267043d20761SAnand Jain btrfs_info(root->fs_info, "disk added %s",vol_args->name); 267143d20761SAnand Jain 2672f46b5a66SChristoph Hellwig kfree(vol_args); 2673c9e9f97bSIlya Dryomov out: 2674c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 26755ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 2676f46b5a66SChristoph Hellwig return ret; 2677f46b5a66SChristoph Hellwig } 2678f46b5a66SChristoph Hellwig 2679da24927bSMiao Xie static long btrfs_ioctl_rm_dev(struct file *file, void __user *arg) 2680f46b5a66SChristoph Hellwig { 2681496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 2682f46b5a66SChristoph Hellwig struct btrfs_ioctl_vol_args *vol_args; 2683f46b5a66SChristoph Hellwig int ret; 2684f46b5a66SChristoph Hellwig 2685e441d54dSChris Mason if (!capable(CAP_SYS_ADMIN)) 2686e441d54dSChris Mason return -EPERM; 2687e441d54dSChris Mason 2688da24927bSMiao Xie ret = mnt_want_write_file(file); 2689da24927bSMiao Xie if (ret) 2690da24927bSMiao Xie return ret; 2691c146afadSYan Zheng 2692dae7b665SLi Zefan vol_args = memdup_user(arg, sizeof(*vol_args)); 2693c9e9f97bSIlya Dryomov if (IS_ERR(vol_args)) { 2694c9e9f97bSIlya Dryomov ret = PTR_ERR(vol_args); 2695c47ca32dSDan Carpenter goto err_drop; 2696c9e9f97bSIlya Dryomov } 2697f46b5a66SChristoph Hellwig 26985516e595SMark Fasheh vol_args->name[BTRFS_PATH_NAME_MAX] = '\0'; 2699f46b5a66SChristoph Hellwig 2700183860f6SAnand Jain if (atomic_xchg(&root->fs_info->mutually_exclusive_operation_running, 2701183860f6SAnand Jain 1)) { 2702183860f6SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 2703183860f6SAnand Jain goto out; 2704183860f6SAnand Jain } 2705183860f6SAnand Jain 2706183860f6SAnand Jain mutex_lock(&root->fs_info->volume_mutex); 2707183860f6SAnand Jain ret = btrfs_rm_device(root, vol_args->name); 2708c9e9f97bSIlya Dryomov mutex_unlock(&root->fs_info->volume_mutex); 27095ac00addSStefan Behrens atomic_set(&root->fs_info->mutually_exclusive_operation_running, 0); 2710183860f6SAnand Jain 2711ec95d491SAnand Jain if (!ret) 2712ec95d491SAnand Jain btrfs_info(root->fs_info, "disk deleted %s",vol_args->name); 2713ec95d491SAnand Jain 2714183860f6SAnand Jain out: 2715183860f6SAnand Jain kfree(vol_args); 2716c47ca32dSDan Carpenter err_drop: 27174ac20c70SIlya Dryomov mnt_drop_write_file(file); 2718f46b5a66SChristoph Hellwig return ret; 2719f46b5a66SChristoph Hellwig } 2720f46b5a66SChristoph Hellwig 2721475f6387SJan Schmidt static long btrfs_ioctl_fs_info(struct btrfs_root *root, void __user *arg) 2722475f6387SJan Schmidt { 2723027ed2f0SLi Zefan struct btrfs_ioctl_fs_info_args *fi_args; 2724475f6387SJan Schmidt struct btrfs_device *device; 2725475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2726027ed2f0SLi Zefan int ret = 0; 2727475f6387SJan Schmidt 2728027ed2f0SLi Zefan fi_args = kzalloc(sizeof(*fi_args), GFP_KERNEL); 2729027ed2f0SLi Zefan if (!fi_args) 2730027ed2f0SLi Zefan return -ENOMEM; 2731027ed2f0SLi Zefan 2732f7171750SFilipe David Borba Manana mutex_lock(&fs_devices->device_list_mutex); 2733027ed2f0SLi Zefan fi_args->num_devices = fs_devices->num_devices; 2734027ed2f0SLi Zefan memcpy(&fi_args->fsid, root->fs_info->fsid, sizeof(fi_args->fsid)); 2735475f6387SJan Schmidt 2736d7641a49SByongho Lee list_for_each_entry(device, &fs_devices->devices, dev_list) { 2737027ed2f0SLi Zefan if (device->devid > fi_args->max_id) 2738027ed2f0SLi Zefan fi_args->max_id = device->devid; 2739475f6387SJan Schmidt } 2740475f6387SJan Schmidt mutex_unlock(&fs_devices->device_list_mutex); 2741475f6387SJan Schmidt 274280a773fbSDavid Sterba fi_args->nodesize = root->fs_info->super_copy->nodesize; 274380a773fbSDavid Sterba fi_args->sectorsize = root->fs_info->super_copy->sectorsize; 274480a773fbSDavid Sterba fi_args->clone_alignment = root->fs_info->super_copy->sectorsize; 274580a773fbSDavid Sterba 2746027ed2f0SLi Zefan if (copy_to_user(arg, fi_args, sizeof(*fi_args))) 2747027ed2f0SLi Zefan ret = -EFAULT; 2748475f6387SJan Schmidt 2749027ed2f0SLi Zefan kfree(fi_args); 2750027ed2f0SLi Zefan return ret; 2751475f6387SJan Schmidt } 2752475f6387SJan Schmidt 2753475f6387SJan Schmidt static long btrfs_ioctl_dev_info(struct btrfs_root *root, void __user *arg) 2754475f6387SJan Schmidt { 2755475f6387SJan Schmidt struct btrfs_ioctl_dev_info_args *di_args; 2756475f6387SJan Schmidt struct btrfs_device *dev; 2757475f6387SJan Schmidt struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 2758475f6387SJan Schmidt int ret = 0; 2759475f6387SJan Schmidt char *s_uuid = NULL; 2760475f6387SJan Schmidt 2761475f6387SJan Schmidt di_args = memdup_user(arg, sizeof(*di_args)); 2762475f6387SJan Schmidt if (IS_ERR(di_args)) 2763475f6387SJan Schmidt return PTR_ERR(di_args); 2764475f6387SJan Schmidt 2765dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(di_args->uuid)) 2766475f6387SJan Schmidt s_uuid = di_args->uuid; 2767475f6387SJan Schmidt 2768475f6387SJan Schmidt mutex_lock(&fs_devices->device_list_mutex); 2769aa1b8cd4SStefan Behrens dev = btrfs_find_device(root->fs_info, di_args->devid, s_uuid, NULL); 2770475f6387SJan Schmidt 2771475f6387SJan Schmidt if (!dev) { 2772475f6387SJan Schmidt ret = -ENODEV; 2773475f6387SJan Schmidt goto out; 2774475f6387SJan Schmidt } 2775475f6387SJan Schmidt 2776475f6387SJan Schmidt di_args->devid = dev->devid; 27777cc8e58dSMiao Xie di_args->bytes_used = btrfs_device_get_bytes_used(dev); 27787cc8e58dSMiao Xie di_args->total_bytes = btrfs_device_get_total_bytes(dev); 2779475f6387SJan Schmidt memcpy(di_args->uuid, dev->uuid, sizeof(di_args->uuid)); 2780a27202fbSJim Meyering if (dev->name) { 2781606686eeSJosef Bacik struct rcu_string *name; 2782606686eeSJosef Bacik 2783606686eeSJosef Bacik rcu_read_lock(); 2784606686eeSJosef Bacik name = rcu_dereference(dev->name); 2785606686eeSJosef Bacik strncpy(di_args->path, name->str, sizeof(di_args->path)); 2786606686eeSJosef Bacik rcu_read_unlock(); 2787a27202fbSJim Meyering di_args->path[sizeof(di_args->path) - 1] = 0; 2788a27202fbSJim Meyering } else { 278999ba55adSStefan Behrens di_args->path[0] = '\0'; 2790a27202fbSJim Meyering } 2791475f6387SJan Schmidt 2792475f6387SJan Schmidt out: 279355793c0dSDavid Sterba mutex_unlock(&fs_devices->device_list_mutex); 2794475f6387SJan Schmidt if (ret == 0 && copy_to_user(arg, di_args, sizeof(*di_args))) 2795475f6387SJan Schmidt ret = -EFAULT; 2796475f6387SJan Schmidt 2797475f6387SJan Schmidt kfree(di_args); 2798475f6387SJan Schmidt return ret; 2799475f6387SJan Schmidt } 2800475f6387SJan Schmidt 2801f4414602SMark Fasheh static struct page *extent_same_get_page(struct inode *inode, pgoff_t index) 2802416161dbSMark Fasheh { 2803416161dbSMark Fasheh struct page *page; 2804416161dbSMark Fasheh 2805416161dbSMark Fasheh page = grab_cache_page(inode->i_mapping, index); 2806416161dbSMark Fasheh if (!page) 280731314002SFilipe Manana return ERR_PTR(-ENOMEM); 2808416161dbSMark Fasheh 2809416161dbSMark Fasheh if (!PageUptodate(page)) { 281031314002SFilipe Manana int ret; 281131314002SFilipe Manana 281231314002SFilipe Manana ret = btrfs_readpage(NULL, page); 281331314002SFilipe Manana if (ret) 281431314002SFilipe Manana return ERR_PTR(ret); 2815416161dbSMark Fasheh lock_page(page); 2816416161dbSMark Fasheh if (!PageUptodate(page)) { 2817416161dbSMark Fasheh unlock_page(page); 281809cbfeafSKirill A. Shutemov put_page(page); 281931314002SFilipe Manana return ERR_PTR(-EIO); 2820416161dbSMark Fasheh } 282131314002SFilipe Manana if (page->mapping != inode->i_mapping) { 2822416161dbSMark Fasheh unlock_page(page); 282309cbfeafSKirill A. Shutemov put_page(page); 282431314002SFilipe Manana return ERR_PTR(-EAGAIN); 2825416161dbSMark Fasheh } 2826416161dbSMark Fasheh } 2827416161dbSMark Fasheh 2828416161dbSMark Fasheh return page; 2829416161dbSMark Fasheh } 2830416161dbSMark Fasheh 2831f4414602SMark Fasheh static int gather_extent_pages(struct inode *inode, struct page **pages, 2832f4414602SMark Fasheh int num_pages, u64 off) 2833f4414602SMark Fasheh { 2834f4414602SMark Fasheh int i; 283509cbfeafSKirill A. Shutemov pgoff_t index = off >> PAGE_SHIFT; 2836f4414602SMark Fasheh 2837f4414602SMark Fasheh for (i = 0; i < num_pages; i++) { 283831314002SFilipe Manana again: 2839f4414602SMark Fasheh pages[i] = extent_same_get_page(inode, index + i); 284031314002SFilipe Manana if (IS_ERR(pages[i])) { 284131314002SFilipe Manana int err = PTR_ERR(pages[i]); 284231314002SFilipe Manana 284331314002SFilipe Manana if (err == -EAGAIN) 284431314002SFilipe Manana goto again; 284531314002SFilipe Manana pages[i] = NULL; 284631314002SFilipe Manana return err; 284731314002SFilipe Manana } 2848f4414602SMark Fasheh } 2849f4414602SMark Fasheh return 0; 2850f4414602SMark Fasheh } 2851f4414602SMark Fasheh 2852e0bd70c6SFilipe Manana static int lock_extent_range(struct inode *inode, u64 off, u64 len, 2853e0bd70c6SFilipe Manana bool retry_range_locking) 285477fe20dcSMark Fasheh { 2855e0bd70c6SFilipe Manana /* 2856e0bd70c6SFilipe Manana * Do any pending delalloc/csum calculations on inode, one way or 2857e0bd70c6SFilipe Manana * another, and lock file content. 2858e0bd70c6SFilipe Manana * The locking order is: 2859e0bd70c6SFilipe Manana * 2860e0bd70c6SFilipe Manana * 1) pages 2861e0bd70c6SFilipe Manana * 2) range in the inode's io tree 2862e0bd70c6SFilipe Manana */ 286377fe20dcSMark Fasheh while (1) { 286477fe20dcSMark Fasheh struct btrfs_ordered_extent *ordered; 286577fe20dcSMark Fasheh lock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 286677fe20dcSMark Fasheh ordered = btrfs_lookup_first_ordered_extent(inode, 286777fe20dcSMark Fasheh off + len - 1); 2868ff5df9b8SFilipe Manana if ((!ordered || 2869ff5df9b8SFilipe Manana ordered->file_offset + ordered->len <= off || 2870ff5df9b8SFilipe Manana ordered->file_offset >= off + len) && 287177fe20dcSMark Fasheh !test_range_bit(&BTRFS_I(inode)->io_tree, off, 2872ff5df9b8SFilipe Manana off + len - 1, EXTENT_DELALLOC, 0, NULL)) { 2873ff5df9b8SFilipe Manana if (ordered) 2874ff5df9b8SFilipe Manana btrfs_put_ordered_extent(ordered); 287577fe20dcSMark Fasheh break; 2876ff5df9b8SFilipe Manana } 287777fe20dcSMark Fasheh unlock_extent(&BTRFS_I(inode)->io_tree, off, off + len - 1); 287877fe20dcSMark Fasheh if (ordered) 287977fe20dcSMark Fasheh btrfs_put_ordered_extent(ordered); 2880e0bd70c6SFilipe Manana if (!retry_range_locking) 2881e0bd70c6SFilipe Manana return -EAGAIN; 288277fe20dcSMark Fasheh btrfs_wait_ordered_range(inode, off, len); 288377fe20dcSMark Fasheh } 2884e0bd70c6SFilipe Manana return 0; 288577fe20dcSMark Fasheh } 288677fe20dcSMark Fasheh 2887f4414602SMark Fasheh static void btrfs_double_inode_unlock(struct inode *inode1, struct inode *inode2) 2888416161dbSMark Fasheh { 28895955102cSAl Viro inode_unlock(inode1); 28905955102cSAl Viro inode_unlock(inode2); 2891416161dbSMark Fasheh } 2892416161dbSMark Fasheh 2893f4414602SMark Fasheh static void btrfs_double_inode_lock(struct inode *inode1, struct inode *inode2) 2894f4414602SMark Fasheh { 2895f4414602SMark Fasheh if (inode1 < inode2) 2896f4414602SMark Fasheh swap(inode1, inode2); 2897f4414602SMark Fasheh 28985955102cSAl Viro inode_lock_nested(inode1, I_MUTEX_PARENT); 28995955102cSAl Viro inode_lock_nested(inode2, I_MUTEX_CHILD); 2900f4414602SMark Fasheh } 2901f4414602SMark Fasheh 2902f4414602SMark Fasheh static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1, 2903f4414602SMark Fasheh struct inode *inode2, u64 loff2, u64 len) 2904f4414602SMark Fasheh { 2905f4414602SMark Fasheh unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1); 2906f4414602SMark Fasheh unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1); 2907f4414602SMark Fasheh } 2908f4414602SMark Fasheh 2909e0bd70c6SFilipe Manana static int btrfs_double_extent_lock(struct inode *inode1, u64 loff1, 2910e0bd70c6SFilipe Manana struct inode *inode2, u64 loff2, u64 len, 2911e0bd70c6SFilipe Manana bool retry_range_locking) 2912416161dbSMark Fasheh { 2913e0bd70c6SFilipe Manana int ret; 2914e0bd70c6SFilipe Manana 2915416161dbSMark Fasheh if (inode1 < inode2) { 2916416161dbSMark Fasheh swap(inode1, inode2); 2917416161dbSMark Fasheh swap(loff1, loff2); 2918416161dbSMark Fasheh } 2919e0bd70c6SFilipe Manana ret = lock_extent_range(inode1, loff1, len, retry_range_locking); 2920e0bd70c6SFilipe Manana if (ret) 2921e0bd70c6SFilipe Manana return ret; 2922e0bd70c6SFilipe Manana ret = lock_extent_range(inode2, loff2, len, retry_range_locking); 2923e0bd70c6SFilipe Manana if (ret) 2924e0bd70c6SFilipe Manana unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, 2925e0bd70c6SFilipe Manana loff1 + len - 1); 2926e0bd70c6SFilipe Manana return ret; 2927416161dbSMark Fasheh } 2928f4414602SMark Fasheh 2929f4414602SMark Fasheh struct cmp_pages { 2930f4414602SMark Fasheh int num_pages; 2931f4414602SMark Fasheh struct page **src_pages; 2932f4414602SMark Fasheh struct page **dst_pages; 2933f4414602SMark Fasheh }; 2934f4414602SMark Fasheh 2935f4414602SMark Fasheh static void btrfs_cmp_data_free(struct cmp_pages *cmp) 2936f4414602SMark Fasheh { 2937f4414602SMark Fasheh int i; 2938f4414602SMark Fasheh struct page *pg; 2939f4414602SMark Fasheh 2940f4414602SMark Fasheh for (i = 0; i < cmp->num_pages; i++) { 2941f4414602SMark Fasheh pg = cmp->src_pages[i]; 2942e0bd70c6SFilipe Manana if (pg) { 2943e0bd70c6SFilipe Manana unlock_page(pg); 294409cbfeafSKirill A. Shutemov put_page(pg); 2945e0bd70c6SFilipe Manana } 2946f4414602SMark Fasheh pg = cmp->dst_pages[i]; 2947e0bd70c6SFilipe Manana if (pg) { 2948e0bd70c6SFilipe Manana unlock_page(pg); 294909cbfeafSKirill A. Shutemov put_page(pg); 2950f4414602SMark Fasheh } 2951e0bd70c6SFilipe Manana } 2952f4414602SMark Fasheh kfree(cmp->src_pages); 2953f4414602SMark Fasheh kfree(cmp->dst_pages); 2954f4414602SMark Fasheh } 2955f4414602SMark Fasheh 2956f4414602SMark Fasheh static int btrfs_cmp_data_prepare(struct inode *src, u64 loff, 2957f4414602SMark Fasheh struct inode *dst, u64 dst_loff, 2958f4414602SMark Fasheh u64 len, struct cmp_pages *cmp) 2959f4414602SMark Fasheh { 2960f4414602SMark Fasheh int ret; 296109cbfeafSKirill A. Shutemov int num_pages = PAGE_ALIGN(len) >> PAGE_SHIFT; 2962f4414602SMark Fasheh struct page **src_pgarr, **dst_pgarr; 2963f4414602SMark Fasheh 2964f4414602SMark Fasheh /* 2965f4414602SMark Fasheh * We must gather up all the pages before we initiate our 2966f4414602SMark Fasheh * extent locking. We use an array for the page pointers. Size 2967f4414602SMark Fasheh * of the array is bounded by len, which is in turn bounded by 2968f4414602SMark Fasheh * BTRFS_MAX_DEDUPE_LEN. 2969f4414602SMark Fasheh */ 297066722f7cSDavid Sterba src_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL); 297166722f7cSDavid Sterba dst_pgarr = kcalloc(num_pages, sizeof(struct page *), GFP_KERNEL); 2972f4414602SMark Fasheh if (!src_pgarr || !dst_pgarr) { 2973f4414602SMark Fasheh kfree(src_pgarr); 2974f4414602SMark Fasheh kfree(dst_pgarr); 2975f4414602SMark Fasheh return -ENOMEM; 2976f4414602SMark Fasheh } 2977f4414602SMark Fasheh cmp->num_pages = num_pages; 2978f4414602SMark Fasheh cmp->src_pages = src_pgarr; 2979f4414602SMark Fasheh cmp->dst_pages = dst_pgarr; 2980f4414602SMark Fasheh 2981f4414602SMark Fasheh ret = gather_extent_pages(src, cmp->src_pages, cmp->num_pages, loff); 2982f4414602SMark Fasheh if (ret) 2983f4414602SMark Fasheh goto out; 2984f4414602SMark Fasheh 2985f4414602SMark Fasheh ret = gather_extent_pages(dst, cmp->dst_pages, cmp->num_pages, dst_loff); 2986f4414602SMark Fasheh 2987f4414602SMark Fasheh out: 2988f4414602SMark Fasheh if (ret) 2989f4414602SMark Fasheh btrfs_cmp_data_free(cmp); 2990f4414602SMark Fasheh return 0; 2991416161dbSMark Fasheh } 2992416161dbSMark Fasheh 2993416161dbSMark Fasheh static int btrfs_cmp_data(struct inode *src, u64 loff, struct inode *dst, 2994f4414602SMark Fasheh u64 dst_loff, u64 len, struct cmp_pages *cmp) 2995416161dbSMark Fasheh { 2996416161dbSMark Fasheh int ret = 0; 2997f4414602SMark Fasheh int i; 2998416161dbSMark Fasheh struct page *src_page, *dst_page; 299909cbfeafSKirill A. Shutemov unsigned int cmp_len = PAGE_SIZE; 3000416161dbSMark Fasheh void *addr, *dst_addr; 3001416161dbSMark Fasheh 3002f4414602SMark Fasheh i = 0; 3003416161dbSMark Fasheh while (len) { 300409cbfeafSKirill A. Shutemov if (len < PAGE_SIZE) 3005416161dbSMark Fasheh cmp_len = len; 3006416161dbSMark Fasheh 3007f4414602SMark Fasheh BUG_ON(i >= cmp->num_pages); 3008f4414602SMark Fasheh 3009f4414602SMark Fasheh src_page = cmp->src_pages[i]; 3010f4414602SMark Fasheh dst_page = cmp->dst_pages[i]; 3011e0bd70c6SFilipe Manana ASSERT(PageLocked(src_page)); 3012e0bd70c6SFilipe Manana ASSERT(PageLocked(dst_page)); 3013f4414602SMark Fasheh 3014416161dbSMark Fasheh addr = kmap_atomic(src_page); 3015416161dbSMark Fasheh dst_addr = kmap_atomic(dst_page); 3016416161dbSMark Fasheh 3017416161dbSMark Fasheh flush_dcache_page(src_page); 3018416161dbSMark Fasheh flush_dcache_page(dst_page); 3019416161dbSMark Fasheh 3020416161dbSMark Fasheh if (memcmp(addr, dst_addr, cmp_len)) 30212b3909f8SDarrick J. Wong ret = -EBADE; 3022416161dbSMark Fasheh 3023416161dbSMark Fasheh kunmap_atomic(addr); 3024416161dbSMark Fasheh kunmap_atomic(dst_addr); 3025416161dbSMark Fasheh 3026416161dbSMark Fasheh if (ret) 3027416161dbSMark Fasheh break; 3028416161dbSMark Fasheh 3029416161dbSMark Fasheh len -= cmp_len; 3030f4414602SMark Fasheh i++; 3031416161dbSMark Fasheh } 3032416161dbSMark Fasheh 3033416161dbSMark Fasheh return ret; 3034416161dbSMark Fasheh } 3035416161dbSMark Fasheh 3036e1d227a4SMark Fasheh static int extent_same_check_offsets(struct inode *inode, u64 off, u64 *plen, 3037e1d227a4SMark Fasheh u64 olen) 3038416161dbSMark Fasheh { 3039e1d227a4SMark Fasheh u64 len = *plen; 3040416161dbSMark Fasheh u64 bs = BTRFS_I(inode)->root->fs_info->sb->s_blocksize; 3041416161dbSMark Fasheh 3042e1d227a4SMark Fasheh if (off + olen > inode->i_size || off + olen < off) 3043416161dbSMark Fasheh return -EINVAL; 3044e1d227a4SMark Fasheh 3045e1d227a4SMark Fasheh /* if we extend to eof, continue to block boundary */ 3046e1d227a4SMark Fasheh if (off + len == inode->i_size) 3047e1d227a4SMark Fasheh *plen = len = ALIGN(inode->i_size, bs) - off; 3048e1d227a4SMark Fasheh 3049416161dbSMark Fasheh /* Check that we are block aligned - btrfs_clone() requires this */ 3050416161dbSMark Fasheh if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs)) 3051416161dbSMark Fasheh return -EINVAL; 3052416161dbSMark Fasheh 3053416161dbSMark Fasheh return 0; 3054416161dbSMark Fasheh } 3055416161dbSMark Fasheh 3056e1d227a4SMark Fasheh static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen, 3057416161dbSMark Fasheh struct inode *dst, u64 dst_loff) 3058416161dbSMark Fasheh { 3059416161dbSMark Fasheh int ret; 3060e1d227a4SMark Fasheh u64 len = olen; 3061f4414602SMark Fasheh struct cmp_pages cmp; 30620efa9f48SMark Fasheh int same_inode = 0; 30630efa9f48SMark Fasheh u64 same_lock_start = 0; 30640efa9f48SMark Fasheh u64 same_lock_len = 0; 3065416161dbSMark Fasheh 3066416161dbSMark Fasheh if (src == dst) 30670efa9f48SMark Fasheh same_inode = 1; 3068416161dbSMark Fasheh 3069113e8283SFilipe Manana if (len == 0) 3070113e8283SFilipe Manana return 0; 3071113e8283SFilipe Manana 30720efa9f48SMark Fasheh if (same_inode) { 30735955102cSAl Viro inode_lock(src); 30740efa9f48SMark Fasheh 30750efa9f48SMark Fasheh ret = extent_same_check_offsets(src, loff, &len, olen); 30760efa9f48SMark Fasheh if (ret) 30770efa9f48SMark Fasheh goto out_unlock; 3078f4dfe687SFilipe Manana ret = extent_same_check_offsets(src, dst_loff, &len, olen); 3079f4dfe687SFilipe Manana if (ret) 3080f4dfe687SFilipe Manana goto out_unlock; 30810efa9f48SMark Fasheh 30820efa9f48SMark Fasheh /* 30830efa9f48SMark Fasheh * Single inode case wants the same checks, except we 30840efa9f48SMark Fasheh * don't want our length pushed out past i_size as 30850efa9f48SMark Fasheh * comparing that data range makes no sense. 30860efa9f48SMark Fasheh * 30870efa9f48SMark Fasheh * extent_same_check_offsets() will do this for an 30880efa9f48SMark Fasheh * unaligned length at i_size, so catch it here and 30890efa9f48SMark Fasheh * reject the request. 30900efa9f48SMark Fasheh * 30910efa9f48SMark Fasheh * This effectively means we require aligned extents 30920efa9f48SMark Fasheh * for the single-inode case, whereas the other cases 30930efa9f48SMark Fasheh * allow an unaligned length so long as it ends at 30940efa9f48SMark Fasheh * i_size. 30950efa9f48SMark Fasheh */ 30960efa9f48SMark Fasheh if (len != olen) { 30970efa9f48SMark Fasheh ret = -EINVAL; 30980efa9f48SMark Fasheh goto out_unlock; 30990efa9f48SMark Fasheh } 31000efa9f48SMark Fasheh 31010efa9f48SMark Fasheh /* Check for overlapping ranges */ 31020efa9f48SMark Fasheh if (dst_loff + len > loff && dst_loff < loff + len) { 31030efa9f48SMark Fasheh ret = -EINVAL; 31040efa9f48SMark Fasheh goto out_unlock; 31050efa9f48SMark Fasheh } 31060efa9f48SMark Fasheh 31070efa9f48SMark Fasheh same_lock_start = min_t(u64, loff, dst_loff); 31080efa9f48SMark Fasheh same_lock_len = max_t(u64, loff, dst_loff) + len - same_lock_start; 31090efa9f48SMark Fasheh } else { 3110f4414602SMark Fasheh btrfs_double_inode_lock(src, dst); 3111416161dbSMark Fasheh 3112e1d227a4SMark Fasheh ret = extent_same_check_offsets(src, loff, &len, olen); 3113416161dbSMark Fasheh if (ret) 3114416161dbSMark Fasheh goto out_unlock; 3115416161dbSMark Fasheh 3116e1d227a4SMark Fasheh ret = extent_same_check_offsets(dst, dst_loff, &len, olen); 3117416161dbSMark Fasheh if (ret) 3118416161dbSMark Fasheh goto out_unlock; 31190efa9f48SMark Fasheh } 3120416161dbSMark Fasheh 3121416161dbSMark Fasheh /* don't make the dst file partly checksummed */ 3122416161dbSMark Fasheh if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 3123416161dbSMark Fasheh (BTRFS_I(dst)->flags & BTRFS_INODE_NODATASUM)) { 3124416161dbSMark Fasheh ret = -EINVAL; 3125416161dbSMark Fasheh goto out_unlock; 3126416161dbSMark Fasheh } 3127416161dbSMark Fasheh 3128e0bd70c6SFilipe Manana again: 3129f4414602SMark Fasheh ret = btrfs_cmp_data_prepare(src, loff, dst, dst_loff, olen, &cmp); 3130f4414602SMark Fasheh if (ret) 3131f4414602SMark Fasheh goto out_unlock; 3132f4414602SMark Fasheh 31330efa9f48SMark Fasheh if (same_inode) 3134e0bd70c6SFilipe Manana ret = lock_extent_range(src, same_lock_start, same_lock_len, 3135e0bd70c6SFilipe Manana false); 31360efa9f48SMark Fasheh else 3137e0bd70c6SFilipe Manana ret = btrfs_double_extent_lock(src, loff, dst, dst_loff, len, 3138e0bd70c6SFilipe Manana false); 3139e0bd70c6SFilipe Manana /* 3140e0bd70c6SFilipe Manana * If one of the inodes has dirty pages in the respective range or 3141e0bd70c6SFilipe Manana * ordered extents, we need to flush dellaloc and wait for all ordered 3142e0bd70c6SFilipe Manana * extents in the range. We must unlock the pages and the ranges in the 3143e0bd70c6SFilipe Manana * io trees to avoid deadlocks when flushing delalloc (requires locking 3144e0bd70c6SFilipe Manana * pages) and when waiting for ordered extents to complete (they require 3145e0bd70c6SFilipe Manana * range locking). 3146e0bd70c6SFilipe Manana */ 3147e0bd70c6SFilipe Manana if (ret == -EAGAIN) { 3148e0bd70c6SFilipe Manana /* 3149e0bd70c6SFilipe Manana * Ranges in the io trees already unlocked. Now unlock all 3150e0bd70c6SFilipe Manana * pages before waiting for all IO to complete. 3151e0bd70c6SFilipe Manana */ 3152e0bd70c6SFilipe Manana btrfs_cmp_data_free(&cmp); 3153e0bd70c6SFilipe Manana if (same_inode) { 3154e0bd70c6SFilipe Manana btrfs_wait_ordered_range(src, same_lock_start, 3155e0bd70c6SFilipe Manana same_lock_len); 3156e0bd70c6SFilipe Manana } else { 3157e0bd70c6SFilipe Manana btrfs_wait_ordered_range(src, loff, len); 3158e0bd70c6SFilipe Manana btrfs_wait_ordered_range(dst, dst_loff, len); 3159e0bd70c6SFilipe Manana } 3160e0bd70c6SFilipe Manana goto again; 3161e0bd70c6SFilipe Manana } 3162e0bd70c6SFilipe Manana ASSERT(ret == 0); 3163e0bd70c6SFilipe Manana if (WARN_ON(ret)) { 3164e0bd70c6SFilipe Manana /* ranges in the io trees already unlocked */ 3165e0bd70c6SFilipe Manana btrfs_cmp_data_free(&cmp); 3166e0bd70c6SFilipe Manana return ret; 3167e0bd70c6SFilipe Manana } 3168f4414602SMark Fasheh 3169207910ddSMark Fasheh /* pass original length for comparison so we stay within i_size */ 3170f4414602SMark Fasheh ret = btrfs_cmp_data(src, loff, dst, dst_loff, olen, &cmp); 3171416161dbSMark Fasheh if (ret == 0) 31721c919a5eSMark Fasheh ret = btrfs_clone(src, dst, loff, olen, len, dst_loff, 1); 3173416161dbSMark Fasheh 31740efa9f48SMark Fasheh if (same_inode) 31750efa9f48SMark Fasheh unlock_extent(&BTRFS_I(src)->io_tree, same_lock_start, 31760efa9f48SMark Fasheh same_lock_start + same_lock_len - 1); 31770efa9f48SMark Fasheh else 3178f4414602SMark Fasheh btrfs_double_extent_unlock(src, loff, dst, dst_loff, len); 3179f4414602SMark Fasheh 3180f4414602SMark Fasheh btrfs_cmp_data_free(&cmp); 3181416161dbSMark Fasheh out_unlock: 31820efa9f48SMark Fasheh if (same_inode) 31835955102cSAl Viro inode_unlock(src); 31840efa9f48SMark Fasheh else 3185f4414602SMark Fasheh btrfs_double_inode_unlock(src, dst); 3186416161dbSMark Fasheh 3187416161dbSMark Fasheh return ret; 3188416161dbSMark Fasheh } 3189416161dbSMark Fasheh 3190ee22184bSByongho Lee #define BTRFS_MAX_DEDUPE_LEN SZ_16M 3191416161dbSMark Fasheh 31922b3909f8SDarrick J. Wong ssize_t btrfs_dedupe_file_range(struct file *src_file, u64 loff, u64 olen, 31932b3909f8SDarrick J. Wong struct file *dst_file, u64 dst_loff) 3194416161dbSMark Fasheh { 31952b3909f8SDarrick J. Wong struct inode *src = file_inode(src_file); 31962b3909f8SDarrick J. Wong struct inode *dst = file_inode(dst_file); 3197416161dbSMark Fasheh u64 bs = BTRFS_I(src)->root->fs_info->sb->s_blocksize; 31982b3909f8SDarrick J. Wong ssize_t res; 3199416161dbSMark Fasheh 32002b3909f8SDarrick J. Wong if (olen > BTRFS_MAX_DEDUPE_LEN) 32012b3909f8SDarrick J. Wong olen = BTRFS_MAX_DEDUPE_LEN; 3202416161dbSMark Fasheh 320309cbfeafSKirill A. Shutemov if (WARN_ON_ONCE(bs < PAGE_SIZE)) { 3204416161dbSMark Fasheh /* 3205416161dbSMark Fasheh * Btrfs does not support blocksize < page_size. As a 3206416161dbSMark Fasheh * result, btrfs_cmp_data() won't correctly handle 3207416161dbSMark Fasheh * this situation without an update. 3208416161dbSMark Fasheh */ 32092b3909f8SDarrick J. Wong return -EINVAL; 3210416161dbSMark Fasheh } 3211416161dbSMark Fasheh 32122b3909f8SDarrick J. Wong res = btrfs_extent_same(src, loff, olen, dst, dst_loff); 32132b3909f8SDarrick J. Wong if (res) 32142b3909f8SDarrick J. Wong return res; 32152b3909f8SDarrick J. Wong return olen; 3216416161dbSMark Fasheh } 3217416161dbSMark Fasheh 3218f82a9901SFilipe Manana static int clone_finish_inode_update(struct btrfs_trans_handle *trans, 3219f82a9901SFilipe Manana struct inode *inode, 3220f82a9901SFilipe Manana u64 endoff, 3221f82a9901SFilipe Manana const u64 destoff, 32221c919a5eSMark Fasheh const u64 olen, 32231c919a5eSMark Fasheh int no_time_update) 3224f82a9901SFilipe Manana { 3225f82a9901SFilipe Manana struct btrfs_root *root = BTRFS_I(inode)->root; 3226f82a9901SFilipe Manana int ret; 3227f82a9901SFilipe Manana 3228f82a9901SFilipe Manana inode_inc_iversion(inode); 32291c919a5eSMark Fasheh if (!no_time_update) 323004b285f3SDeepa Dinamani inode->i_mtime = inode->i_ctime = current_fs_time(inode->i_sb); 3231f82a9901SFilipe Manana /* 3232f82a9901SFilipe Manana * We round up to the block size at eof when determining which 3233f82a9901SFilipe Manana * extents to clone above, but shouldn't round up the file size. 3234f82a9901SFilipe Manana */ 3235f82a9901SFilipe Manana if (endoff > destoff + olen) 3236f82a9901SFilipe Manana endoff = destoff + olen; 3237f82a9901SFilipe Manana if (endoff > inode->i_size) 3238f82a9901SFilipe Manana btrfs_i_size_write(inode, endoff); 3239f82a9901SFilipe Manana 3240f82a9901SFilipe Manana ret = btrfs_update_inode(trans, root, inode); 3241f82a9901SFilipe Manana if (ret) { 3242f82a9901SFilipe Manana btrfs_abort_transaction(trans, root, ret); 3243f82a9901SFilipe Manana btrfs_end_transaction(trans, root); 3244f82a9901SFilipe Manana goto out; 3245f82a9901SFilipe Manana } 3246f82a9901SFilipe Manana ret = btrfs_end_transaction(trans, root); 3247f82a9901SFilipe Manana out: 3248f82a9901SFilipe Manana return ret; 3249f82a9901SFilipe Manana } 3250f82a9901SFilipe Manana 32517ffbb598SFilipe Manana static void clone_update_extent_map(struct inode *inode, 32527ffbb598SFilipe Manana const struct btrfs_trans_handle *trans, 32537ffbb598SFilipe Manana const struct btrfs_path *path, 32547ffbb598SFilipe Manana const u64 hole_offset, 32557ffbb598SFilipe Manana const u64 hole_len) 32567ffbb598SFilipe Manana { 32577ffbb598SFilipe Manana struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 32587ffbb598SFilipe Manana struct extent_map *em; 32597ffbb598SFilipe Manana int ret; 32607ffbb598SFilipe Manana 32617ffbb598SFilipe Manana em = alloc_extent_map(); 32627ffbb598SFilipe Manana if (!em) { 32637ffbb598SFilipe Manana set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, 32647ffbb598SFilipe Manana &BTRFS_I(inode)->runtime_flags); 32657ffbb598SFilipe Manana return; 32667ffbb598SFilipe Manana } 32677ffbb598SFilipe Manana 326814f59796SFilipe Manana if (path) { 326914f59796SFilipe Manana struct btrfs_file_extent_item *fi; 327014f59796SFilipe Manana 327114f59796SFilipe Manana fi = btrfs_item_ptr(path->nodes[0], path->slots[0], 327214f59796SFilipe Manana struct btrfs_file_extent_item); 32737ffbb598SFilipe Manana btrfs_extent_item_to_extent_map(inode, path, fi, false, em); 32747ffbb598SFilipe Manana em->generation = -1; 32757ffbb598SFilipe Manana if (btrfs_file_extent_type(path->nodes[0], fi) == 32767ffbb598SFilipe Manana BTRFS_FILE_EXTENT_INLINE) 32777ffbb598SFilipe Manana set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, 32787ffbb598SFilipe Manana &BTRFS_I(inode)->runtime_flags); 32797ffbb598SFilipe Manana } else { 32807ffbb598SFilipe Manana em->start = hole_offset; 32817ffbb598SFilipe Manana em->len = hole_len; 32827ffbb598SFilipe Manana em->ram_bytes = em->len; 32837ffbb598SFilipe Manana em->orig_start = hole_offset; 32847ffbb598SFilipe Manana em->block_start = EXTENT_MAP_HOLE; 32857ffbb598SFilipe Manana em->block_len = 0; 32867ffbb598SFilipe Manana em->orig_block_len = 0; 32877ffbb598SFilipe Manana em->compress_type = BTRFS_COMPRESS_NONE; 32887ffbb598SFilipe Manana em->generation = trans->transid; 32897ffbb598SFilipe Manana } 32907ffbb598SFilipe Manana 32917ffbb598SFilipe Manana while (1) { 32927ffbb598SFilipe Manana write_lock(&em_tree->lock); 32937ffbb598SFilipe Manana ret = add_extent_mapping(em_tree, em, 1); 32947ffbb598SFilipe Manana write_unlock(&em_tree->lock); 32957ffbb598SFilipe Manana if (ret != -EEXIST) { 32967ffbb598SFilipe Manana free_extent_map(em); 32977ffbb598SFilipe Manana break; 32987ffbb598SFilipe Manana } 32997ffbb598SFilipe Manana btrfs_drop_extent_cache(inode, em->start, 33007ffbb598SFilipe Manana em->start + em->len - 1, 0); 33017ffbb598SFilipe Manana } 33027ffbb598SFilipe Manana 3303ee39b432SDavid Sterba if (ret) 33047ffbb598SFilipe Manana set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, 33057ffbb598SFilipe Manana &BTRFS_I(inode)->runtime_flags); 33067ffbb598SFilipe Manana } 33077ffbb598SFilipe Manana 33088039d87dSFilipe Manana /* 33098039d87dSFilipe Manana * Make sure we do not end up inserting an inline extent into a file that has 33108039d87dSFilipe Manana * already other (non-inline) extents. If a file has an inline extent it can 33118039d87dSFilipe Manana * not have any other extents and the (single) inline extent must start at the 33128039d87dSFilipe Manana * file offset 0. Failing to respect these rules will lead to file corruption, 33138039d87dSFilipe Manana * resulting in EIO errors on read/write operations, hitting BUG_ON's in mm, etc 33148039d87dSFilipe Manana * 33158039d87dSFilipe Manana * We can have extents that have been already written to disk or we can have 33168039d87dSFilipe Manana * dirty ranges still in delalloc, in which case the extent maps and items are 33178039d87dSFilipe Manana * created only when we run delalloc, and the delalloc ranges might fall outside 33188039d87dSFilipe Manana * the range we are currently locking in the inode's io tree. So we check the 33198039d87dSFilipe Manana * inode's i_size because of that (i_size updates are done while holding the 33208039d87dSFilipe Manana * i_mutex, which we are holding here). 33218039d87dSFilipe Manana * We also check to see if the inode has a size not greater than "datal" but has 33228039d87dSFilipe Manana * extents beyond it, due to an fallocate with FALLOC_FL_KEEP_SIZE (and we are 33238039d87dSFilipe Manana * protected against such concurrent fallocate calls by the i_mutex). 33248039d87dSFilipe Manana * 33258039d87dSFilipe Manana * If the file has no extents but a size greater than datal, do not allow the 33268039d87dSFilipe Manana * copy because we would need turn the inline extent into a non-inline one (even 33278039d87dSFilipe Manana * with NO_HOLES enabled). If we find our destination inode only has one inline 33288039d87dSFilipe Manana * extent, just overwrite it with the source inline extent if its size is less 33298039d87dSFilipe Manana * than the source extent's size, or we could copy the source inline extent's 33308039d87dSFilipe Manana * data into the destination inode's inline extent if the later is greater then 33318039d87dSFilipe Manana * the former. 33328039d87dSFilipe Manana */ 33338039d87dSFilipe Manana static int clone_copy_inline_extent(struct inode *src, 33348039d87dSFilipe Manana struct inode *dst, 33358039d87dSFilipe Manana struct btrfs_trans_handle *trans, 33368039d87dSFilipe Manana struct btrfs_path *path, 33378039d87dSFilipe Manana struct btrfs_key *new_key, 33388039d87dSFilipe Manana const u64 drop_start, 33398039d87dSFilipe Manana const u64 datal, 33408039d87dSFilipe Manana const u64 skip, 33418039d87dSFilipe Manana const u64 size, 33428039d87dSFilipe Manana char *inline_data) 33438039d87dSFilipe Manana { 33448039d87dSFilipe Manana struct btrfs_root *root = BTRFS_I(dst)->root; 33458039d87dSFilipe Manana const u64 aligned_end = ALIGN(new_key->offset + datal, 33468039d87dSFilipe Manana root->sectorsize); 33478039d87dSFilipe Manana int ret; 33488039d87dSFilipe Manana struct btrfs_key key; 33498039d87dSFilipe Manana 33508039d87dSFilipe Manana if (new_key->offset > 0) 33518039d87dSFilipe Manana return -EOPNOTSUPP; 33528039d87dSFilipe Manana 33538039d87dSFilipe Manana key.objectid = btrfs_ino(dst); 33548039d87dSFilipe Manana key.type = BTRFS_EXTENT_DATA_KEY; 33558039d87dSFilipe Manana key.offset = 0; 33568039d87dSFilipe Manana ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 33578039d87dSFilipe Manana if (ret < 0) { 33588039d87dSFilipe Manana return ret; 33598039d87dSFilipe Manana } else if (ret > 0) { 33608039d87dSFilipe Manana if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 33618039d87dSFilipe Manana ret = btrfs_next_leaf(root, path); 33628039d87dSFilipe Manana if (ret < 0) 33638039d87dSFilipe Manana return ret; 33648039d87dSFilipe Manana else if (ret > 0) 33658039d87dSFilipe Manana goto copy_inline_extent; 33668039d87dSFilipe Manana } 33678039d87dSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); 33688039d87dSFilipe Manana if (key.objectid == btrfs_ino(dst) && 33698039d87dSFilipe Manana key.type == BTRFS_EXTENT_DATA_KEY) { 33708039d87dSFilipe Manana ASSERT(key.offset > 0); 33718039d87dSFilipe Manana return -EOPNOTSUPP; 33728039d87dSFilipe Manana } 33738039d87dSFilipe Manana } else if (i_size_read(dst) <= datal) { 33748039d87dSFilipe Manana struct btrfs_file_extent_item *ei; 33758039d87dSFilipe Manana u64 ext_len; 33768039d87dSFilipe Manana 33778039d87dSFilipe Manana /* 33788039d87dSFilipe Manana * If the file size is <= datal, make sure there are no other 33798039d87dSFilipe Manana * extents following (can happen do to an fallocate call with 33808039d87dSFilipe Manana * the flag FALLOC_FL_KEEP_SIZE). 33818039d87dSFilipe Manana */ 33828039d87dSFilipe Manana ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 33838039d87dSFilipe Manana struct btrfs_file_extent_item); 33848039d87dSFilipe Manana /* 33858039d87dSFilipe Manana * If it's an inline extent, it can not have other extents 33868039d87dSFilipe Manana * following it. 33878039d87dSFilipe Manana */ 33888039d87dSFilipe Manana if (btrfs_file_extent_type(path->nodes[0], ei) == 33898039d87dSFilipe Manana BTRFS_FILE_EXTENT_INLINE) 33908039d87dSFilipe Manana goto copy_inline_extent; 33918039d87dSFilipe Manana 33928039d87dSFilipe Manana ext_len = btrfs_file_extent_num_bytes(path->nodes[0], ei); 33938039d87dSFilipe Manana if (ext_len > aligned_end) 33948039d87dSFilipe Manana return -EOPNOTSUPP; 33958039d87dSFilipe Manana 33968039d87dSFilipe Manana ret = btrfs_next_item(root, path); 33978039d87dSFilipe Manana if (ret < 0) { 33988039d87dSFilipe Manana return ret; 33998039d87dSFilipe Manana } else if (ret == 0) { 34008039d87dSFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &key, 34018039d87dSFilipe Manana path->slots[0]); 34028039d87dSFilipe Manana if (key.objectid == btrfs_ino(dst) && 34038039d87dSFilipe Manana key.type == BTRFS_EXTENT_DATA_KEY) 34048039d87dSFilipe Manana return -EOPNOTSUPP; 34058039d87dSFilipe Manana } 34068039d87dSFilipe Manana } 34078039d87dSFilipe Manana 34088039d87dSFilipe Manana copy_inline_extent: 34098039d87dSFilipe Manana /* 34108039d87dSFilipe Manana * We have no extent items, or we have an extent at offset 0 which may 34118039d87dSFilipe Manana * or may not be inlined. All these cases are dealt the same way. 34128039d87dSFilipe Manana */ 34138039d87dSFilipe Manana if (i_size_read(dst) > datal) { 34148039d87dSFilipe Manana /* 34158039d87dSFilipe Manana * If the destination inode has an inline extent... 34168039d87dSFilipe Manana * This would require copying the data from the source inline 34178039d87dSFilipe Manana * extent into the beginning of the destination's inline extent. 34188039d87dSFilipe Manana * But this is really complex, both extents can be compressed 34198039d87dSFilipe Manana * or just one of them, which would require decompressing and 34208039d87dSFilipe Manana * re-compressing data (which could increase the new compressed 34218039d87dSFilipe Manana * size, not allowing the compressed data to fit anymore in an 34228039d87dSFilipe Manana * inline extent). 34238039d87dSFilipe Manana * So just don't support this case for now (it should be rare, 34248039d87dSFilipe Manana * we are not really saving space when cloning inline extents). 34258039d87dSFilipe Manana */ 34268039d87dSFilipe Manana return -EOPNOTSUPP; 34278039d87dSFilipe Manana } 34288039d87dSFilipe Manana 34298039d87dSFilipe Manana btrfs_release_path(path); 34308039d87dSFilipe Manana ret = btrfs_drop_extents(trans, root, dst, drop_start, aligned_end, 1); 34318039d87dSFilipe Manana if (ret) 34328039d87dSFilipe Manana return ret; 34338039d87dSFilipe Manana ret = btrfs_insert_empty_item(trans, root, path, new_key, size); 34348039d87dSFilipe Manana if (ret) 34358039d87dSFilipe Manana return ret; 34368039d87dSFilipe Manana 34378039d87dSFilipe Manana if (skip) { 34388039d87dSFilipe Manana const u32 start = btrfs_file_extent_calc_inline_size(0); 34398039d87dSFilipe Manana 34408039d87dSFilipe Manana memmove(inline_data + start, inline_data + start + skip, datal); 34418039d87dSFilipe Manana } 34428039d87dSFilipe Manana 34438039d87dSFilipe Manana write_extent_buffer(path->nodes[0], inline_data, 34448039d87dSFilipe Manana btrfs_item_ptr_offset(path->nodes[0], 34458039d87dSFilipe Manana path->slots[0]), 34468039d87dSFilipe Manana size); 34478039d87dSFilipe Manana inode_add_bytes(dst, datal); 34488039d87dSFilipe Manana 34498039d87dSFilipe Manana return 0; 34508039d87dSFilipe Manana } 34518039d87dSFilipe Manana 345232b7c687SMark Fasheh /** 345332b7c687SMark Fasheh * btrfs_clone() - clone a range from inode file to another 345432b7c687SMark Fasheh * 345532b7c687SMark Fasheh * @src: Inode to clone from 345632b7c687SMark Fasheh * @inode: Inode to clone to 345732b7c687SMark Fasheh * @off: Offset within source to start clone from 345832b7c687SMark Fasheh * @olen: Original length, passed by user, of range to clone 34591c919a5eSMark Fasheh * @olen_aligned: Block-aligned value of olen 346032b7c687SMark Fasheh * @destoff: Offset within @inode to start clone 34611c919a5eSMark Fasheh * @no_time_update: Whether to update mtime/ctime on the target inode 346232b7c687SMark Fasheh */ 346332b7c687SMark Fasheh static int btrfs_clone(struct inode *src, struct inode *inode, 3464f82a9901SFilipe Manana const u64 off, const u64 olen, const u64 olen_aligned, 34651c919a5eSMark Fasheh const u64 destoff, int no_time_update) 3466f46b5a66SChristoph Hellwig { 3467f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 346832b7c687SMark Fasheh struct btrfs_path *path = NULL; 3469f46b5a66SChristoph Hellwig struct extent_buffer *leaf; 347032b7c687SMark Fasheh struct btrfs_trans_handle *trans; 347132b7c687SMark Fasheh char *buf = NULL; 3472ae01a0abSYan Zheng struct btrfs_key key; 3473f46b5a66SChristoph Hellwig u32 nritems; 3474f46b5a66SChristoph Hellwig int slot; 3475ae01a0abSYan Zheng int ret; 3476f82a9901SFilipe Manana const u64 len = olen_aligned; 3477f82a9901SFilipe Manana u64 last_dest_end = destoff; 3478ae01a0abSYan Zheng 3479ae01a0abSYan Zheng ret = -ENOMEM; 348015351955SDavid Sterba buf = kmalloc(root->nodesize, GFP_KERNEL | __GFP_NOWARN); 348115351955SDavid Sterba if (!buf) { 3482707e8a07SDavid Sterba buf = vmalloc(root->nodesize); 3483ae01a0abSYan Zheng if (!buf) 348432b7c687SMark Fasheh return ret; 348515351955SDavid Sterba } 3486ae01a0abSYan Zheng 3487ae01a0abSYan Zheng path = btrfs_alloc_path(); 3488ae01a0abSYan Zheng if (!path) { 348915351955SDavid Sterba kvfree(buf); 349032b7c687SMark Fasheh return ret; 3491ae01a0abSYan Zheng } 349232b7c687SMark Fasheh 3493e4058b54SDavid Sterba path->reada = READA_FORWARD; 3494c5c9cd4dSSage Weil /* clone data */ 349533345d01SLi Zefan key.objectid = btrfs_ino(src); 3496ae01a0abSYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 34972c463823SFilipe Manana key.offset = off; 3498f46b5a66SChristoph Hellwig 3499f46b5a66SChristoph Hellwig while (1) { 3500de249e66SChris Mason u64 next_key_min_offset = key.offset + 1; 3501df858e76SFilipe Manana 3502f46b5a66SChristoph Hellwig /* 3503f46b5a66SChristoph Hellwig * note the key will change type as we walk through the 3504f46b5a66SChristoph Hellwig * tree. 3505f46b5a66SChristoph Hellwig */ 3506e4355f34SFilipe David Borba Manana path->leave_spinning = 1; 3507362a20c5SDavid Sterba ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path, 3508362a20c5SDavid Sterba 0, 0); 3509f46b5a66SChristoph Hellwig if (ret < 0) 3510f46b5a66SChristoph Hellwig goto out; 35112c463823SFilipe Manana /* 35122c463823SFilipe Manana * First search, if no extent item that starts at offset off was 35132c463823SFilipe Manana * found but the previous item is an extent item, it's possible 35142c463823SFilipe Manana * it might overlap our target range, therefore process it. 35152c463823SFilipe Manana */ 35162c463823SFilipe Manana if (key.offset == off && ret > 0 && path->slots[0] > 0) { 35172c463823SFilipe Manana btrfs_item_key_to_cpu(path->nodes[0], &key, 35182c463823SFilipe Manana path->slots[0] - 1); 35192c463823SFilipe Manana if (key.type == BTRFS_EXTENT_DATA_KEY) 35202c463823SFilipe Manana path->slots[0]--; 35212c463823SFilipe Manana } 3522f46b5a66SChristoph Hellwig 3523ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 3524e4355f34SFilipe David Borba Manana process_slot: 3525ae01a0abSYan Zheng if (path->slots[0] >= nritems) { 3526362a20c5SDavid Sterba ret = btrfs_next_leaf(BTRFS_I(src)->root, path); 3527f46b5a66SChristoph Hellwig if (ret < 0) 3528f46b5a66SChristoph Hellwig goto out; 3529f46b5a66SChristoph Hellwig if (ret > 0) 3530f46b5a66SChristoph Hellwig break; 3531ae01a0abSYan Zheng nritems = btrfs_header_nritems(path->nodes[0]); 3532f46b5a66SChristoph Hellwig } 3533f46b5a66SChristoph Hellwig leaf = path->nodes[0]; 3534f46b5a66SChristoph Hellwig slot = path->slots[0]; 3535f46b5a66SChristoph Hellwig 3536ae01a0abSYan Zheng btrfs_item_key_to_cpu(leaf, &key, slot); 3537962a298fSDavid Sterba if (key.type > BTRFS_EXTENT_DATA_KEY || 353833345d01SLi Zefan key.objectid != btrfs_ino(src)) 3539f46b5a66SChristoph Hellwig break; 3540f46b5a66SChristoph Hellwig 3541962a298fSDavid Sterba if (key.type == BTRFS_EXTENT_DATA_KEY) { 3542c5c9cd4dSSage Weil struct btrfs_file_extent_item *extent; 3543c5c9cd4dSSage Weil int type; 354431840ae1SZheng Yan u32 size; 354531840ae1SZheng Yan struct btrfs_key new_key; 3546c5c9cd4dSSage Weil u64 disko = 0, diskl = 0; 3547c5c9cd4dSSage Weil u64 datao = 0, datal = 0; 3548c5c9cd4dSSage Weil u8 comp; 3549f82a9901SFilipe Manana u64 drop_start; 355031840ae1SZheng Yan 3551c5c9cd4dSSage Weil extent = btrfs_item_ptr(leaf, slot, 3552c5c9cd4dSSage Weil struct btrfs_file_extent_item); 3553c5c9cd4dSSage Weil comp = btrfs_file_extent_compression(leaf, extent); 3554c5c9cd4dSSage Weil type = btrfs_file_extent_type(leaf, extent); 3555c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 3556c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 3557d397712bSChris Mason disko = btrfs_file_extent_disk_bytenr(leaf, 3558d397712bSChris Mason extent); 3559d397712bSChris Mason diskl = btrfs_file_extent_disk_num_bytes(leaf, 3560d397712bSChris Mason extent); 3561c5c9cd4dSSage Weil datao = btrfs_file_extent_offset(leaf, extent); 3562d397712bSChris Mason datal = btrfs_file_extent_num_bytes(leaf, 3563d397712bSChris Mason extent); 3564c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 3565c5c9cd4dSSage Weil /* take upper bound, may be compressed */ 3566c5c9cd4dSSage Weil datal = btrfs_file_extent_ram_bytes(leaf, 3567c5c9cd4dSSage Weil extent); 3568c5c9cd4dSSage Weil } 356931840ae1SZheng Yan 35702c463823SFilipe Manana /* 35712c463823SFilipe Manana * The first search might have left us at an extent 35722c463823SFilipe Manana * item that ends before our target range's start, can 35732c463823SFilipe Manana * happen if we have holes and NO_HOLES feature enabled. 35742c463823SFilipe Manana */ 35752c463823SFilipe Manana if (key.offset + datal <= off) { 3576e4355f34SFilipe David Borba Manana path->slots[0]++; 3577e4355f34SFilipe David Borba Manana goto process_slot; 35782c463823SFilipe Manana } else if (key.offset >= off + len) { 35792c463823SFilipe Manana break; 3580e4355f34SFilipe David Borba Manana } 3581df858e76SFilipe Manana next_key_min_offset = key.offset + datal; 3582e4355f34SFilipe David Borba Manana size = btrfs_item_size_nr(leaf, slot); 3583e4355f34SFilipe David Borba Manana read_extent_buffer(leaf, buf, 3584e4355f34SFilipe David Borba Manana btrfs_item_ptr_offset(leaf, slot), 3585e4355f34SFilipe David Borba Manana size); 3586e4355f34SFilipe David Borba Manana 3587e4355f34SFilipe David Borba Manana btrfs_release_path(path); 3588e4355f34SFilipe David Borba Manana path->leave_spinning = 0; 3589c5c9cd4dSSage Weil 359031840ae1SZheng Yan memcpy(&new_key, &key, sizeof(new_key)); 359133345d01SLi Zefan new_key.objectid = btrfs_ino(inode); 35924d728ec7SLi Zefan if (off <= key.offset) 3593c5c9cd4dSSage Weil new_key.offset = key.offset + destoff - off; 35944d728ec7SLi Zefan else 35954d728ec7SLi Zefan new_key.offset = destoff; 3596c5c9cd4dSSage Weil 3597b6f3409bSSage Weil /* 3598f82a9901SFilipe Manana * Deal with a hole that doesn't have an extent item 3599f82a9901SFilipe Manana * that represents it (NO_HOLES feature enabled). 3600f82a9901SFilipe Manana * This hole is either in the middle of the cloning 3601f82a9901SFilipe Manana * range or at the beginning (fully overlaps it or 3602f82a9901SFilipe Manana * partially overlaps it). 3603f82a9901SFilipe Manana */ 3604f82a9901SFilipe Manana if (new_key.offset != last_dest_end) 3605f82a9901SFilipe Manana drop_start = last_dest_end; 3606f82a9901SFilipe Manana else 3607f82a9901SFilipe Manana drop_start = new_key.offset; 3608f82a9901SFilipe Manana 3609f82a9901SFilipe Manana /* 3610b6f3409bSSage Weil * 1 - adjusting old extent (we may have to split it) 3611b6f3409bSSage Weil * 1 - add new extent 3612b6f3409bSSage Weil * 1 - inode update 3613b6f3409bSSage Weil */ 3614b6f3409bSSage Weil trans = btrfs_start_transaction(root, 3); 3615a22285a6SYan, Zheng if (IS_ERR(trans)) { 3616a22285a6SYan, Zheng ret = PTR_ERR(trans); 3617a22285a6SYan, Zheng goto out; 3618a22285a6SYan, Zheng } 3619a22285a6SYan, Zheng 3620c8a894d7SChris Mason if (type == BTRFS_FILE_EXTENT_REG || 3621c8a894d7SChris Mason type == BTRFS_FILE_EXTENT_PREALLOC) { 3622d72c0842SLi Zefan /* 3623d72c0842SLi Zefan * a | --- range to clone ---| b 3624d72c0842SLi Zefan * | ------------- extent ------------- | 3625d72c0842SLi Zefan */ 3626d72c0842SLi Zefan 362793915584SAntonio Ospite /* subtract range b */ 3628d72c0842SLi Zefan if (key.offset + datal > off + len) 3629d72c0842SLi Zefan datal = off + len - key.offset; 3630d72c0842SLi Zefan 363193915584SAntonio Ospite /* subtract range a */ 3632a22285a6SYan, Zheng if (off > key.offset) { 3633a22285a6SYan, Zheng datao += off - key.offset; 3634a22285a6SYan, Zheng datal -= off - key.offset; 3635a22285a6SYan, Zheng } 3636a22285a6SYan, Zheng 36375dc562c5SJosef Bacik ret = btrfs_drop_extents(trans, root, inode, 3638f82a9901SFilipe Manana drop_start, 3639a22285a6SYan, Zheng new_key.offset + datal, 36402671485dSJosef Bacik 1); 364179787eaaSJeff Mahoney if (ret) { 36423f9e3df8SDavid Sterba if (ret != -EOPNOTSUPP) 364300fdf13aSLiu Bo btrfs_abort_transaction(trans, 364400fdf13aSLiu Bo root, ret); 364579787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 364679787eaaSJeff Mahoney goto out; 364779787eaaSJeff Mahoney } 3648a22285a6SYan, Zheng 364931840ae1SZheng Yan ret = btrfs_insert_empty_item(trans, root, path, 365031840ae1SZheng Yan &new_key, size); 365179787eaaSJeff Mahoney if (ret) { 365279787eaaSJeff Mahoney btrfs_abort_transaction(trans, root, 365379787eaaSJeff Mahoney ret); 365479787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 365579787eaaSJeff Mahoney goto out; 365679787eaaSJeff Mahoney } 365731840ae1SZheng Yan 365831840ae1SZheng Yan leaf = path->nodes[0]; 365931840ae1SZheng Yan slot = path->slots[0]; 366031840ae1SZheng Yan write_extent_buffer(leaf, buf, 366131840ae1SZheng Yan btrfs_item_ptr_offset(leaf, slot), 366231840ae1SZheng Yan size); 3663ae01a0abSYan Zheng 3664f46b5a66SChristoph Hellwig extent = btrfs_item_ptr(leaf, slot, 3665f46b5a66SChristoph Hellwig struct btrfs_file_extent_item); 3666c5c9cd4dSSage Weil 3667c5c9cd4dSSage Weil /* disko == 0 means it's a hole */ 3668c5c9cd4dSSage Weil if (!disko) 3669c5c9cd4dSSage Weil datao = 0; 3670c5c9cd4dSSage Weil 3671c5c9cd4dSSage Weil btrfs_set_file_extent_offset(leaf, extent, 3672c5c9cd4dSSage Weil datao); 3673c5c9cd4dSSage Weil btrfs_set_file_extent_num_bytes(leaf, extent, 3674c5c9cd4dSSage Weil datal); 3675fcebe456SJosef Bacik 3676c5c9cd4dSSage Weil if (disko) { 3677c5c9cd4dSSage Weil inode_add_bytes(inode, datal); 3678ae01a0abSYan Zheng ret = btrfs_inc_extent_ref(trans, root, 36795d4f98a2SYan Zheng disko, diskl, 0, 3680f46b5a66SChristoph Hellwig root->root_key.objectid, 368133345d01SLi Zefan btrfs_ino(inode), 3682b06c4bf5SFilipe Manana new_key.offset - datao); 368379787eaaSJeff Mahoney if (ret) { 368479787eaaSJeff Mahoney btrfs_abort_transaction(trans, 368579787eaaSJeff Mahoney root, 368679787eaaSJeff Mahoney ret); 368779787eaaSJeff Mahoney btrfs_end_transaction(trans, 368879787eaaSJeff Mahoney root); 368979787eaaSJeff Mahoney goto out; 369079787eaaSJeff Mahoney 369179787eaaSJeff Mahoney } 3692f46b5a66SChristoph Hellwig } 3693c5c9cd4dSSage Weil } else if (type == BTRFS_FILE_EXTENT_INLINE) { 3694c5c9cd4dSSage Weil u64 skip = 0; 3695c5c9cd4dSSage Weil u64 trim = 0; 3696ed958762SFilipe Manana 3697c5c9cd4dSSage Weil if (off > key.offset) { 3698c5c9cd4dSSage Weil skip = off - key.offset; 3699c5c9cd4dSSage Weil new_key.offset += skip; 370031840ae1SZheng Yan } 3701d397712bSChris Mason 3702c5c9cd4dSSage Weil if (key.offset + datal > off + len) 3703c5c9cd4dSSage Weil trim = key.offset + datal - (off + len); 3704d397712bSChris Mason 3705c5c9cd4dSSage Weil if (comp && (skip || trim)) { 3706c5c9cd4dSSage Weil ret = -EINVAL; 3707a22285a6SYan, Zheng btrfs_end_transaction(trans, root); 3708c5c9cd4dSSage Weil goto out; 370931840ae1SZheng Yan } 3710c5c9cd4dSSage Weil size -= skip + trim; 3711c5c9cd4dSSage Weil datal -= skip + trim; 3712a22285a6SYan, Zheng 37138039d87dSFilipe Manana ret = clone_copy_inline_extent(src, inode, 37148039d87dSFilipe Manana trans, path, 37158039d87dSFilipe Manana &new_key, 3716f82a9901SFilipe Manana drop_start, 37178039d87dSFilipe Manana datal, 37188039d87dSFilipe Manana skip, size, buf); 371979787eaaSJeff Mahoney if (ret) { 37203f9e3df8SDavid Sterba if (ret != -EOPNOTSUPP) 37213a29bc09SChris Mason btrfs_abort_transaction(trans, 37228039d87dSFilipe Manana root, 372379787eaaSJeff Mahoney ret); 372479787eaaSJeff Mahoney btrfs_end_transaction(trans, root); 372579787eaaSJeff Mahoney goto out; 372679787eaaSJeff Mahoney } 3727c5c9cd4dSSage Weil leaf = path->nodes[0]; 3728c5c9cd4dSSage Weil slot = path->slots[0]; 3729c5c9cd4dSSage Weil } 3730c5c9cd4dSSage Weil 37317ffbb598SFilipe Manana /* If we have an implicit hole (NO_HOLES feature). */ 37327ffbb598SFilipe Manana if (drop_start < new_key.offset) 37337ffbb598SFilipe Manana clone_update_extent_map(inode, trans, 373414f59796SFilipe Manana NULL, drop_start, 37357ffbb598SFilipe Manana new_key.offset - drop_start); 37367ffbb598SFilipe Manana 373714f59796SFilipe Manana clone_update_extent_map(inode, trans, path, 0, 0); 37387ffbb598SFilipe Manana 3739c5c9cd4dSSage Weil btrfs_mark_buffer_dirty(leaf); 3740b3b4aa74SDavid Sterba btrfs_release_path(path); 3741c5c9cd4dSSage Weil 374262e2390eSFilipe Manana last_dest_end = ALIGN(new_key.offset + datal, 374362e2390eSFilipe Manana root->sectorsize); 3744f82a9901SFilipe Manana ret = clone_finish_inode_update(trans, inode, 3745f82a9901SFilipe Manana last_dest_end, 37461c919a5eSMark Fasheh destoff, olen, 37471c919a5eSMark Fasheh no_time_update); 3748f82a9901SFilipe Manana if (ret) 374979787eaaSJeff Mahoney goto out; 37502c463823SFilipe Manana if (new_key.offset + datal >= destoff + len) 37512c463823SFilipe Manana break; 3752a22285a6SYan, Zheng } 3753b3b4aa74SDavid Sterba btrfs_release_path(path); 3754df858e76SFilipe Manana key.offset = next_key_min_offset; 3755ae01a0abSYan Zheng } 3756f46b5a66SChristoph Hellwig ret = 0; 375732b7c687SMark Fasheh 3758f82a9901SFilipe Manana if (last_dest_end < destoff + len) { 3759f82a9901SFilipe Manana /* 3760f82a9901SFilipe Manana * We have an implicit hole (NO_HOLES feature is enabled) that 3761f82a9901SFilipe Manana * fully or partially overlaps our cloning range at its end. 3762f82a9901SFilipe Manana */ 3763b3b4aa74SDavid Sterba btrfs_release_path(path); 3764f82a9901SFilipe Manana 3765f82a9901SFilipe Manana /* 3766f82a9901SFilipe Manana * 1 - remove extent(s) 3767f82a9901SFilipe Manana * 1 - inode update 3768f82a9901SFilipe Manana */ 3769f82a9901SFilipe Manana trans = btrfs_start_transaction(root, 2); 3770f82a9901SFilipe Manana if (IS_ERR(trans)) { 3771f82a9901SFilipe Manana ret = PTR_ERR(trans); 3772f82a9901SFilipe Manana goto out; 3773f82a9901SFilipe Manana } 3774f82a9901SFilipe Manana ret = btrfs_drop_extents(trans, root, inode, 3775f82a9901SFilipe Manana last_dest_end, destoff + len, 1); 3776f82a9901SFilipe Manana if (ret) { 3777f82a9901SFilipe Manana if (ret != -EOPNOTSUPP) 3778f82a9901SFilipe Manana btrfs_abort_transaction(trans, root, ret); 3779f82a9901SFilipe Manana btrfs_end_transaction(trans, root); 3780f82a9901SFilipe Manana goto out; 3781f82a9901SFilipe Manana } 378214f59796SFilipe Manana clone_update_extent_map(inode, trans, NULL, last_dest_end, 378314f59796SFilipe Manana destoff + len - last_dest_end); 3784f82a9901SFilipe Manana ret = clone_finish_inode_update(trans, inode, destoff + len, 37851c919a5eSMark Fasheh destoff, olen, no_time_update); 3786f82a9901SFilipe Manana } 3787f82a9901SFilipe Manana 3788f46b5a66SChristoph Hellwig out: 378932b7c687SMark Fasheh btrfs_free_path(path); 379015351955SDavid Sterba kvfree(buf); 379132b7c687SMark Fasheh return ret; 379232b7c687SMark Fasheh } 379332b7c687SMark Fasheh 37943db11b2eSZach Brown static noinline int btrfs_clone_files(struct file *file, struct file *file_src, 379532b7c687SMark Fasheh u64 off, u64 olen, u64 destoff) 379632b7c687SMark Fasheh { 379754563d41SAl Viro struct inode *inode = file_inode(file); 37983db11b2eSZach Brown struct inode *src = file_inode(file_src); 379932b7c687SMark Fasheh struct btrfs_root *root = BTRFS_I(inode)->root; 380032b7c687SMark Fasheh int ret; 380132b7c687SMark Fasheh u64 len = olen; 380232b7c687SMark Fasheh u64 bs = root->fs_info->sb->s_blocksize; 38033db11b2eSZach Brown int same_inode = src == inode; 380432b7c687SMark Fasheh 380532b7c687SMark Fasheh /* 380632b7c687SMark Fasheh * TODO: 380732b7c687SMark Fasheh * - split compressed inline extents. annoying: we need to 380832b7c687SMark Fasheh * decompress into destination's address_space (the file offset 380932b7c687SMark Fasheh * may change, so source mapping won't do), then recompress (or 381032b7c687SMark Fasheh * otherwise reinsert) a subrange. 381100fdf13aSLiu Bo * 381200fdf13aSLiu Bo * - split destination inode's inline extents. The inline extents can 381300fdf13aSLiu Bo * be either compressed or non-compressed. 381432b7c687SMark Fasheh */ 381532b7c687SMark Fasheh 381632b7c687SMark Fasheh if (btrfs_root_readonly(root)) 381732b7c687SMark Fasheh return -EROFS; 381832b7c687SMark Fasheh 38193db11b2eSZach Brown if (file_src->f_path.mnt != file->f_path.mnt || 38203db11b2eSZach Brown src->i_sb != inode->i_sb) 38213db11b2eSZach Brown return -EXDEV; 382232b7c687SMark Fasheh 382332b7c687SMark Fasheh /* don't make the dst file partly checksummed */ 382432b7c687SMark Fasheh if ((BTRFS_I(src)->flags & BTRFS_INODE_NODATASUM) != 382532b7c687SMark Fasheh (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) 38263db11b2eSZach Brown return -EINVAL; 382732b7c687SMark Fasheh 382832b7c687SMark Fasheh if (S_ISDIR(src->i_mode) || S_ISDIR(inode->i_mode)) 38293db11b2eSZach Brown return -EISDIR; 383032b7c687SMark Fasheh 383132b7c687SMark Fasheh if (!same_inode) { 3832293a8489SMark Fasheh btrfs_double_inode_lock(src, inode); 383332b7c687SMark Fasheh } else { 38345955102cSAl Viro inode_lock(src); 383532b7c687SMark Fasheh } 383632b7c687SMark Fasheh 383732b7c687SMark Fasheh /* determine range to clone */ 383832b7c687SMark Fasheh ret = -EINVAL; 383932b7c687SMark Fasheh if (off + len > src->i_size || off + len < off) 384032b7c687SMark Fasheh goto out_unlock; 384132b7c687SMark Fasheh if (len == 0) 384232b7c687SMark Fasheh olen = len = src->i_size - off; 384332b7c687SMark Fasheh /* if we extend to eof, continue to block boundary */ 384432b7c687SMark Fasheh if (off + len == src->i_size) 384532b7c687SMark Fasheh len = ALIGN(src->i_size, bs) - off; 384632b7c687SMark Fasheh 3847ccccf3d6SFilipe Manana if (len == 0) { 3848ccccf3d6SFilipe Manana ret = 0; 3849ccccf3d6SFilipe Manana goto out_unlock; 3850ccccf3d6SFilipe Manana } 3851ccccf3d6SFilipe Manana 385232b7c687SMark Fasheh /* verify the end result is block aligned */ 385332b7c687SMark Fasheh if (!IS_ALIGNED(off, bs) || !IS_ALIGNED(off + len, bs) || 385432b7c687SMark Fasheh !IS_ALIGNED(destoff, bs)) 385532b7c687SMark Fasheh goto out_unlock; 385632b7c687SMark Fasheh 385732b7c687SMark Fasheh /* verify if ranges are overlapped within the same file */ 385832b7c687SMark Fasheh if (same_inode) { 385932b7c687SMark Fasheh if (destoff + len > off && destoff < off + len) 386032b7c687SMark Fasheh goto out_unlock; 386132b7c687SMark Fasheh } 386232b7c687SMark Fasheh 386332b7c687SMark Fasheh if (destoff > inode->i_size) { 386432b7c687SMark Fasheh ret = btrfs_cont_expand(inode, inode->i_size, destoff); 386532b7c687SMark Fasheh if (ret) 386632b7c687SMark Fasheh goto out_unlock; 386732b7c687SMark Fasheh } 386832b7c687SMark Fasheh 3869c125b8bfSFilipe Manana /* 3870c125b8bfSFilipe Manana * Lock the target range too. Right after we replace the file extent 3871c125b8bfSFilipe Manana * items in the fs tree (which now point to the cloned data), we might 3872c125b8bfSFilipe Manana * have a worker replace them with extent items relative to a write 3873c125b8bfSFilipe Manana * operation that was issued before this clone operation (i.e. confront 3874c125b8bfSFilipe Manana * with inode.c:btrfs_finish_ordered_io). 3875c125b8bfSFilipe Manana */ 3876c125b8bfSFilipe Manana if (same_inode) { 3877c125b8bfSFilipe Manana u64 lock_start = min_t(u64, off, destoff); 3878c125b8bfSFilipe Manana u64 lock_len = max_t(u64, off, destoff) + len - lock_start; 387932b7c687SMark Fasheh 3880e0bd70c6SFilipe Manana ret = lock_extent_range(src, lock_start, lock_len, true); 3881c125b8bfSFilipe Manana } else { 3882e0bd70c6SFilipe Manana ret = btrfs_double_extent_lock(src, off, inode, destoff, len, 3883e0bd70c6SFilipe Manana true); 3884e0bd70c6SFilipe Manana } 3885e0bd70c6SFilipe Manana ASSERT(ret == 0); 3886e0bd70c6SFilipe Manana if (WARN_ON(ret)) { 3887e0bd70c6SFilipe Manana /* ranges in the io trees already unlocked */ 3888e0bd70c6SFilipe Manana goto out_unlock; 3889c125b8bfSFilipe Manana } 389032b7c687SMark Fasheh 38911c919a5eSMark Fasheh ret = btrfs_clone(src, inode, off, olen, len, destoff, 0); 389232b7c687SMark Fasheh 3893c125b8bfSFilipe Manana if (same_inode) { 3894c125b8bfSFilipe Manana u64 lock_start = min_t(u64, off, destoff); 3895c125b8bfSFilipe Manana u64 lock_end = max_t(u64, off, destoff) + len - 1; 3896c125b8bfSFilipe Manana 3897c125b8bfSFilipe Manana unlock_extent(&BTRFS_I(src)->io_tree, lock_start, lock_end); 3898c125b8bfSFilipe Manana } else { 3899293a8489SMark Fasheh btrfs_double_extent_unlock(src, off, inode, destoff, len); 3900c125b8bfSFilipe Manana } 3901c125b8bfSFilipe Manana /* 3902c125b8bfSFilipe Manana * Truncate page cache pages so that future reads will see the cloned 3903c125b8bfSFilipe Manana * data immediately and not the previous data. 3904c125b8bfSFilipe Manana */ 390565bfa658SChandan Rajendra truncate_inode_pages_range(&inode->i_data, 390609cbfeafSKirill A. Shutemov round_down(destoff, PAGE_SIZE), 390709cbfeafSKirill A. Shutemov round_up(destoff + len, PAGE_SIZE) - 1); 3908f46b5a66SChristoph Hellwig out_unlock: 3909293a8489SMark Fasheh if (!same_inode) 3910293a8489SMark Fasheh btrfs_double_inode_unlock(src, inode); 3911293a8489SMark Fasheh else 39125955102cSAl Viro inode_unlock(src); 39133db11b2eSZach Brown return ret; 39143db11b2eSZach Brown } 39153db11b2eSZach Brown 39163db11b2eSZach Brown ssize_t btrfs_copy_file_range(struct file *file_in, loff_t pos_in, 39173db11b2eSZach Brown struct file *file_out, loff_t pos_out, 39183db11b2eSZach Brown size_t len, unsigned int flags) 39193db11b2eSZach Brown { 39203db11b2eSZach Brown ssize_t ret; 39213db11b2eSZach Brown 39223db11b2eSZach Brown ret = btrfs_clone_files(file_out, file_in, pos_in, len, pos_out); 39233db11b2eSZach Brown if (ret == 0) 39243db11b2eSZach Brown ret = len; 39253db11b2eSZach Brown return ret; 39263db11b2eSZach Brown } 39273db11b2eSZach Brown 392804b38d60SChristoph Hellwig int btrfs_clone_file_range(struct file *src_file, loff_t off, 392904b38d60SChristoph Hellwig struct file *dst_file, loff_t destoff, u64 len) 39303db11b2eSZach Brown { 393104b38d60SChristoph Hellwig return btrfs_clone_files(dst_file, src_file, off, len, destoff); 3932c5c9cd4dSSage Weil } 3933c5c9cd4dSSage Weil 3934f46b5a66SChristoph Hellwig /* 3935f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 3936f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 3937f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 3938f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 3939f46b5a66SChristoph Hellwig */ 3940b2950863SChristoph Hellwig static long btrfs_ioctl_trans_start(struct file *file) 3941f46b5a66SChristoph Hellwig { 3942496ad9aaSAl Viro struct inode *inode = file_inode(file); 3943f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 3944f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 39451ab86aedSSage Weil int ret; 3946f46b5a66SChristoph Hellwig 39471ab86aedSSage Weil ret = -EPERM; 3948df5b5520SChristoph Hellwig if (!capable(CAP_SYS_ADMIN)) 3949f46b5a66SChristoph Hellwig goto out; 39501ab86aedSSage Weil 39511ab86aedSSage Weil ret = -EINPROGRESS; 39521ab86aedSSage Weil if (file->private_data) 39531ab86aedSSage Weil goto out; 39549ca9ee09SSage Weil 3955b83cc969SLi Zefan ret = -EROFS; 3956b83cc969SLi Zefan if (btrfs_root_readonly(root)) 3957b83cc969SLi Zefan goto out; 3958b83cc969SLi Zefan 3959a561be71SAl Viro ret = mnt_want_write_file(file); 3960c146afadSYan Zheng if (ret) 3961c146afadSYan Zheng goto out; 3962c146afadSYan Zheng 3963a4abeea4SJosef Bacik atomic_inc(&root->fs_info->open_ioctl_trans); 39649ca9ee09SSage Weil 3965f46b5a66SChristoph Hellwig ret = -ENOMEM; 39667a7eaa40SJosef Bacik trans = btrfs_start_ioctl_transaction(root); 3967abd30bb0STsutomu Itoh if (IS_ERR(trans)) 39681ab86aedSSage Weil goto out_drop; 39691ab86aedSSage Weil 39701ab86aedSSage Weil file->private_data = trans; 39711ab86aedSSage Weil return 0; 39721ab86aedSSage Weil 39731ab86aedSSage Weil out_drop: 3974a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 39752a79f17eSAl Viro mnt_drop_write_file(file); 3976f46b5a66SChristoph Hellwig out: 3977f46b5a66SChristoph Hellwig return ret; 3978f46b5a66SChristoph Hellwig } 3979f46b5a66SChristoph Hellwig 39806ef5ed0dSJosef Bacik static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp) 39816ef5ed0dSJosef Bacik { 3982496ad9aaSAl Viro struct inode *inode = file_inode(file); 39836ef5ed0dSJosef Bacik struct btrfs_root *root = BTRFS_I(inode)->root; 39846ef5ed0dSJosef Bacik struct btrfs_root *new_root; 39856ef5ed0dSJosef Bacik struct btrfs_dir_item *di; 39866ef5ed0dSJosef Bacik struct btrfs_trans_handle *trans; 39876ef5ed0dSJosef Bacik struct btrfs_path *path; 39886ef5ed0dSJosef Bacik struct btrfs_key location; 39896ef5ed0dSJosef Bacik struct btrfs_disk_key disk_key; 39906ef5ed0dSJosef Bacik u64 objectid = 0; 39916ef5ed0dSJosef Bacik u64 dir_id; 39923c04ce01SMiao Xie int ret; 39936ef5ed0dSJosef Bacik 39946ef5ed0dSJosef Bacik if (!capable(CAP_SYS_ADMIN)) 39956ef5ed0dSJosef Bacik return -EPERM; 39966ef5ed0dSJosef Bacik 39973c04ce01SMiao Xie ret = mnt_want_write_file(file); 39983c04ce01SMiao Xie if (ret) 39993c04ce01SMiao Xie return ret; 40003c04ce01SMiao Xie 40013c04ce01SMiao Xie if (copy_from_user(&objectid, argp, sizeof(objectid))) { 40023c04ce01SMiao Xie ret = -EFAULT; 40033c04ce01SMiao Xie goto out; 40043c04ce01SMiao Xie } 40056ef5ed0dSJosef Bacik 40066ef5ed0dSJosef Bacik if (!objectid) 40071cecf579Schandan objectid = BTRFS_FS_TREE_OBJECTID; 40086ef5ed0dSJosef Bacik 40096ef5ed0dSJosef Bacik location.objectid = objectid; 40106ef5ed0dSJosef Bacik location.type = BTRFS_ROOT_ITEM_KEY; 40116ef5ed0dSJosef Bacik location.offset = (u64)-1; 40126ef5ed0dSJosef Bacik 40136ef5ed0dSJosef Bacik new_root = btrfs_read_fs_root_no_name(root->fs_info, &location); 40143c04ce01SMiao Xie if (IS_ERR(new_root)) { 40153c04ce01SMiao Xie ret = PTR_ERR(new_root); 40163c04ce01SMiao Xie goto out; 40173c04ce01SMiao Xie } 40186ef5ed0dSJosef Bacik 40196ef5ed0dSJosef Bacik path = btrfs_alloc_path(); 40203c04ce01SMiao Xie if (!path) { 40213c04ce01SMiao Xie ret = -ENOMEM; 40223c04ce01SMiao Xie goto out; 40233c04ce01SMiao Xie } 40246ef5ed0dSJosef Bacik path->leave_spinning = 1; 40256ef5ed0dSJosef Bacik 40266ef5ed0dSJosef Bacik trans = btrfs_start_transaction(root, 1); 402798d5dc13STsutomu Itoh if (IS_ERR(trans)) { 40286ef5ed0dSJosef Bacik btrfs_free_path(path); 40293c04ce01SMiao Xie ret = PTR_ERR(trans); 40303c04ce01SMiao Xie goto out; 40316ef5ed0dSJosef Bacik } 40326ef5ed0dSJosef Bacik 40336c41761fSDavid Sterba dir_id = btrfs_super_root_dir(root->fs_info->super_copy); 40346ef5ed0dSJosef Bacik di = btrfs_lookup_dir_item(trans, root->fs_info->tree_root, path, 40356ef5ed0dSJosef Bacik dir_id, "default", 7, 1); 4036cf1e99a4SDan Carpenter if (IS_ERR_OR_NULL(di)) { 40376ef5ed0dSJosef Bacik btrfs_free_path(path); 40386ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 4039efe120a0SFrank Holton btrfs_err(new_root->fs_info, "Umm, you don't have the default dir" 4040efe120a0SFrank Holton "item, this isn't going to work"); 40413c04ce01SMiao Xie ret = -ENOENT; 40423c04ce01SMiao Xie goto out; 40436ef5ed0dSJosef Bacik } 40446ef5ed0dSJosef Bacik 40456ef5ed0dSJosef Bacik btrfs_cpu_key_to_disk(&disk_key, &new_root->root_key); 40466ef5ed0dSJosef Bacik btrfs_set_dir_item_key(path->nodes[0], di, &disk_key); 40476ef5ed0dSJosef Bacik btrfs_mark_buffer_dirty(path->nodes[0]); 40486ef5ed0dSJosef Bacik btrfs_free_path(path); 40496ef5ed0dSJosef Bacik 40502b0ce2c2SMitch Harder btrfs_set_fs_incompat(root->fs_info, DEFAULT_SUBVOL); 40516ef5ed0dSJosef Bacik btrfs_end_transaction(trans, root); 40523c04ce01SMiao Xie out: 40533c04ce01SMiao Xie mnt_drop_write_file(file); 40543c04ce01SMiao Xie return ret; 40556ef5ed0dSJosef Bacik } 40566ef5ed0dSJosef Bacik 40575af3e8ccSStefan Behrens void btrfs_get_block_group_info(struct list_head *groups_list, 4058bf5fc093SJosef Bacik struct btrfs_ioctl_space_info *space) 4059bf5fc093SJosef Bacik { 4060bf5fc093SJosef Bacik struct btrfs_block_group_cache *block_group; 4061bf5fc093SJosef Bacik 4062bf5fc093SJosef Bacik space->total_bytes = 0; 4063bf5fc093SJosef Bacik space->used_bytes = 0; 4064bf5fc093SJosef Bacik space->flags = 0; 4065bf5fc093SJosef Bacik list_for_each_entry(block_group, groups_list, list) { 4066bf5fc093SJosef Bacik space->flags = block_group->flags; 4067bf5fc093SJosef Bacik space->total_bytes += block_group->key.offset; 4068bf5fc093SJosef Bacik space->used_bytes += 4069bf5fc093SJosef Bacik btrfs_block_group_used(&block_group->item); 4070bf5fc093SJosef Bacik } 4071bf5fc093SJosef Bacik } 4072bf5fc093SJosef Bacik 407348a3b636SEric Sandeen static long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg) 40741406e432SJosef Bacik { 40751406e432SJosef Bacik struct btrfs_ioctl_space_args space_args; 40761406e432SJosef Bacik struct btrfs_ioctl_space_info space; 40771406e432SJosef Bacik struct btrfs_ioctl_space_info *dest; 40787fde62bfSChris Mason struct btrfs_ioctl_space_info *dest_orig; 407913f2696fSDaniel J Blueman struct btrfs_ioctl_space_info __user *user_dest; 40801406e432SJosef Bacik struct btrfs_space_info *info; 4081bf5fc093SJosef Bacik u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 4082bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_SYSTEM, 4083bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_METADATA, 4084bf5fc093SJosef Bacik BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 4085bf5fc093SJosef Bacik int num_types = 4; 40867fde62bfSChris Mason int alloc_size; 40871406e432SJosef Bacik int ret = 0; 408851788b1bSDan Rosenberg u64 slot_count = 0; 4089bf5fc093SJosef Bacik int i, c; 40901406e432SJosef Bacik 40911406e432SJosef Bacik if (copy_from_user(&space_args, 40921406e432SJosef Bacik (struct btrfs_ioctl_space_args __user *)arg, 40931406e432SJosef Bacik sizeof(space_args))) 40941406e432SJosef Bacik return -EFAULT; 40951406e432SJosef Bacik 4096bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 4097bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 4098bf5fc093SJosef Bacik 4099bf5fc093SJosef Bacik info = NULL; 41007fde62bfSChris Mason rcu_read_lock(); 4101bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 4102bf5fc093SJosef Bacik list) { 4103bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 4104bf5fc093SJosef Bacik info = tmp; 4105bf5fc093SJosef Bacik break; 4106bf5fc093SJosef Bacik } 4107bf5fc093SJosef Bacik } 41087fde62bfSChris Mason rcu_read_unlock(); 41091406e432SJosef Bacik 4110bf5fc093SJosef Bacik if (!info) 4111bf5fc093SJosef Bacik continue; 4112bf5fc093SJosef Bacik 4113bf5fc093SJosef Bacik down_read(&info->groups_sem); 4114bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 4115bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) 4116bf5fc093SJosef Bacik slot_count++; 4117bf5fc093SJosef Bacik } 4118bf5fc093SJosef Bacik up_read(&info->groups_sem); 4119bf5fc093SJosef Bacik } 4120bf5fc093SJosef Bacik 412136523e95SDavid Sterba /* 412236523e95SDavid Sterba * Global block reserve, exported as a space_info 412336523e95SDavid Sterba */ 412436523e95SDavid Sterba slot_count++; 412536523e95SDavid Sterba 41267fde62bfSChris Mason /* space_slots == 0 means they are asking for a count */ 41277fde62bfSChris Mason if (space_args.space_slots == 0) { 41287fde62bfSChris Mason space_args.total_spaces = slot_count; 41297fde62bfSChris Mason goto out; 41307fde62bfSChris Mason } 4131bf5fc093SJosef Bacik 413251788b1bSDan Rosenberg slot_count = min_t(u64, space_args.space_slots, slot_count); 4133bf5fc093SJosef Bacik 41347fde62bfSChris Mason alloc_size = sizeof(*dest) * slot_count; 4135bf5fc093SJosef Bacik 41367fde62bfSChris Mason /* we generally have at most 6 or so space infos, one for each raid 41377fde62bfSChris Mason * level. So, a whole page should be more than enough for everyone 41387fde62bfSChris Mason */ 413909cbfeafSKirill A. Shutemov if (alloc_size > PAGE_SIZE) 41407fde62bfSChris Mason return -ENOMEM; 41417fde62bfSChris Mason 41427fde62bfSChris Mason space_args.total_spaces = 0; 41438d2db785SDavid Sterba dest = kmalloc(alloc_size, GFP_KERNEL); 41447fde62bfSChris Mason if (!dest) 41457fde62bfSChris Mason return -ENOMEM; 41467fde62bfSChris Mason dest_orig = dest; 41477fde62bfSChris Mason 41487fde62bfSChris Mason /* now we have a buffer to copy into */ 4149bf5fc093SJosef Bacik for (i = 0; i < num_types; i++) { 4150bf5fc093SJosef Bacik struct btrfs_space_info *tmp; 4151bf5fc093SJosef Bacik 415251788b1bSDan Rosenberg if (!slot_count) 415351788b1bSDan Rosenberg break; 415451788b1bSDan Rosenberg 4155bf5fc093SJosef Bacik info = NULL; 41561406e432SJosef Bacik rcu_read_lock(); 4157bf5fc093SJosef Bacik list_for_each_entry_rcu(tmp, &root->fs_info->space_info, 4158bf5fc093SJosef Bacik list) { 4159bf5fc093SJosef Bacik if (tmp->flags == types[i]) { 4160bf5fc093SJosef Bacik info = tmp; 41617fde62bfSChris Mason break; 4162bf5fc093SJosef Bacik } 4163bf5fc093SJosef Bacik } 4164bf5fc093SJosef Bacik rcu_read_unlock(); 41657fde62bfSChris Mason 4166bf5fc093SJosef Bacik if (!info) 4167bf5fc093SJosef Bacik continue; 4168bf5fc093SJosef Bacik down_read(&info->groups_sem); 4169bf5fc093SJosef Bacik for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 4170bf5fc093SJosef Bacik if (!list_empty(&info->block_groups[c])) { 41715af3e8ccSStefan Behrens btrfs_get_block_group_info( 41725af3e8ccSStefan Behrens &info->block_groups[c], &space); 41737fde62bfSChris Mason memcpy(dest, &space, sizeof(space)); 41741406e432SJosef Bacik dest++; 41751406e432SJosef Bacik space_args.total_spaces++; 417651788b1bSDan Rosenberg slot_count--; 41771406e432SJosef Bacik } 417851788b1bSDan Rosenberg if (!slot_count) 417951788b1bSDan Rosenberg break; 4180bf5fc093SJosef Bacik } 4181bf5fc093SJosef Bacik up_read(&info->groups_sem); 4182bf5fc093SJosef Bacik } 41831406e432SJosef Bacik 418436523e95SDavid Sterba /* 418536523e95SDavid Sterba * Add global block reserve 418636523e95SDavid Sterba */ 418736523e95SDavid Sterba if (slot_count) { 418836523e95SDavid Sterba struct btrfs_block_rsv *block_rsv = &root->fs_info->global_block_rsv; 418936523e95SDavid Sterba 419036523e95SDavid Sterba spin_lock(&block_rsv->lock); 419136523e95SDavid Sterba space.total_bytes = block_rsv->size; 419236523e95SDavid Sterba space.used_bytes = block_rsv->size - block_rsv->reserved; 419336523e95SDavid Sterba spin_unlock(&block_rsv->lock); 419436523e95SDavid Sterba space.flags = BTRFS_SPACE_INFO_GLOBAL_RSV; 419536523e95SDavid Sterba memcpy(dest, &space, sizeof(space)); 419636523e95SDavid Sterba space_args.total_spaces++; 419736523e95SDavid Sterba } 419836523e95SDavid Sterba 41992eec6c81SDaniel J Blueman user_dest = (struct btrfs_ioctl_space_info __user *) 42007fde62bfSChris Mason (arg + sizeof(struct btrfs_ioctl_space_args)); 42017fde62bfSChris Mason 42027fde62bfSChris Mason if (copy_to_user(user_dest, dest_orig, alloc_size)) 42037fde62bfSChris Mason ret = -EFAULT; 42047fde62bfSChris Mason 42057fde62bfSChris Mason kfree(dest_orig); 42067fde62bfSChris Mason out: 42077fde62bfSChris Mason if (ret == 0 && copy_to_user(arg, &space_args, sizeof(space_args))) 42081406e432SJosef Bacik ret = -EFAULT; 42091406e432SJosef Bacik 42101406e432SJosef Bacik return ret; 42111406e432SJosef Bacik } 42121406e432SJosef Bacik 4213f46b5a66SChristoph Hellwig /* 4214f46b5a66SChristoph Hellwig * there are many ways the trans_start and trans_end ioctls can lead 4215f46b5a66SChristoph Hellwig * to deadlocks. They should only be used by applications that 4216f46b5a66SChristoph Hellwig * basically own the machine, and have a very in depth understanding 4217f46b5a66SChristoph Hellwig * of all the possible deadlocks and enospc problems. 4218f46b5a66SChristoph Hellwig */ 4219f46b5a66SChristoph Hellwig long btrfs_ioctl_trans_end(struct file *file) 4220f46b5a66SChristoph Hellwig { 4221496ad9aaSAl Viro struct inode *inode = file_inode(file); 4222f46b5a66SChristoph Hellwig struct btrfs_root *root = BTRFS_I(inode)->root; 4223f46b5a66SChristoph Hellwig struct btrfs_trans_handle *trans; 4224f46b5a66SChristoph Hellwig 4225f46b5a66SChristoph Hellwig trans = file->private_data; 42261ab86aedSSage Weil if (!trans) 42271ab86aedSSage Weil return -EINVAL; 4228b214107eSChristoph Hellwig file->private_data = NULL; 42299ca9ee09SSage Weil 42301ab86aedSSage Weil btrfs_end_transaction(trans, root); 42311ab86aedSSage Weil 4232a4abeea4SJosef Bacik atomic_dec(&root->fs_info->open_ioctl_trans); 42339ca9ee09SSage Weil 42342a79f17eSAl Viro mnt_drop_write_file(file); 42351ab86aedSSage Weil return 0; 4236f46b5a66SChristoph Hellwig } 4237f46b5a66SChristoph Hellwig 42389a8c28beSMiao Xie static noinline long btrfs_ioctl_start_sync(struct btrfs_root *root, 42399a8c28beSMiao Xie void __user *argp) 424046204592SSage Weil { 424146204592SSage Weil struct btrfs_trans_handle *trans; 424246204592SSage Weil u64 transid; 4243db5b493aSTsutomu Itoh int ret; 424446204592SSage Weil 4245d4edf39bSMiao Xie trans = btrfs_attach_transaction_barrier(root); 4246ff7c1d33SMiao Xie if (IS_ERR(trans)) { 4247ff7c1d33SMiao Xie if (PTR_ERR(trans) != -ENOENT) 424898d5dc13STsutomu Itoh return PTR_ERR(trans); 4249ff7c1d33SMiao Xie 4250ff7c1d33SMiao Xie /* No running transaction, don't bother */ 4251ff7c1d33SMiao Xie transid = root->fs_info->last_trans_committed; 4252ff7c1d33SMiao Xie goto out; 4253ff7c1d33SMiao Xie } 425446204592SSage Weil transid = trans->transid; 4255db5b493aSTsutomu Itoh ret = btrfs_commit_transaction_async(trans, root, 0); 42568b2b2d3cSTsutomu Itoh if (ret) { 42578b2b2d3cSTsutomu Itoh btrfs_end_transaction(trans, root); 4258db5b493aSTsutomu Itoh return ret; 42598b2b2d3cSTsutomu Itoh } 4260ff7c1d33SMiao Xie out: 426146204592SSage Weil if (argp) 426246204592SSage Weil if (copy_to_user(argp, &transid, sizeof(transid))) 426346204592SSage Weil return -EFAULT; 426446204592SSage Weil return 0; 426546204592SSage Weil } 426646204592SSage Weil 42679a8c28beSMiao Xie static noinline long btrfs_ioctl_wait_sync(struct btrfs_root *root, 42689a8c28beSMiao Xie void __user *argp) 426946204592SSage Weil { 427046204592SSage Weil u64 transid; 427146204592SSage Weil 427246204592SSage Weil if (argp) { 427346204592SSage Weil if (copy_from_user(&transid, argp, sizeof(transid))) 427446204592SSage Weil return -EFAULT; 427546204592SSage Weil } else { 427646204592SSage Weil transid = 0; /* current trans */ 427746204592SSage Weil } 427846204592SSage Weil return btrfs_wait_for_commit(root, transid); 427946204592SSage Weil } 428046204592SSage Weil 4281b8e95489SMiao Xie static long btrfs_ioctl_scrub(struct file *file, void __user *arg) 4282475f6387SJan Schmidt { 4283496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4284475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 4285b8e95489SMiao Xie int ret; 4286475f6387SJan Schmidt 4287475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4288475f6387SJan Schmidt return -EPERM; 4289475f6387SJan Schmidt 4290475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 4291475f6387SJan Schmidt if (IS_ERR(sa)) 4292475f6387SJan Schmidt return PTR_ERR(sa); 4293475f6387SJan Schmidt 4294b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) { 4295b8e95489SMiao Xie ret = mnt_want_write_file(file); 4296b8e95489SMiao Xie if (ret) 4297b8e95489SMiao Xie goto out; 4298b8e95489SMiao Xie } 4299b8e95489SMiao Xie 4300aa1b8cd4SStefan Behrens ret = btrfs_scrub_dev(root->fs_info, sa->devid, sa->start, sa->end, 430163a212abSStefan Behrens &sa->progress, sa->flags & BTRFS_SCRUB_READONLY, 430263a212abSStefan Behrens 0); 4303475f6387SJan Schmidt 4304475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 4305475f6387SJan Schmidt ret = -EFAULT; 4306475f6387SJan Schmidt 4307b8e95489SMiao Xie if (!(sa->flags & BTRFS_SCRUB_READONLY)) 4308b8e95489SMiao Xie mnt_drop_write_file(file); 4309b8e95489SMiao Xie out: 4310475f6387SJan Schmidt kfree(sa); 4311475f6387SJan Schmidt return ret; 4312475f6387SJan Schmidt } 4313475f6387SJan Schmidt 4314475f6387SJan Schmidt static long btrfs_ioctl_scrub_cancel(struct btrfs_root *root, void __user *arg) 4315475f6387SJan Schmidt { 4316475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4317475f6387SJan Schmidt return -EPERM; 4318475f6387SJan Schmidt 4319aa1b8cd4SStefan Behrens return btrfs_scrub_cancel(root->fs_info); 4320475f6387SJan Schmidt } 4321475f6387SJan Schmidt 4322475f6387SJan Schmidt static long btrfs_ioctl_scrub_progress(struct btrfs_root *root, 4323475f6387SJan Schmidt void __user *arg) 4324475f6387SJan Schmidt { 4325475f6387SJan Schmidt struct btrfs_ioctl_scrub_args *sa; 4326475f6387SJan Schmidt int ret; 4327475f6387SJan Schmidt 4328475f6387SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4329475f6387SJan Schmidt return -EPERM; 4330475f6387SJan Schmidt 4331475f6387SJan Schmidt sa = memdup_user(arg, sizeof(*sa)); 4332475f6387SJan Schmidt if (IS_ERR(sa)) 4333475f6387SJan Schmidt return PTR_ERR(sa); 4334475f6387SJan Schmidt 4335475f6387SJan Schmidt ret = btrfs_scrub_progress(root, sa->devid, &sa->progress); 4336475f6387SJan Schmidt 4337475f6387SJan Schmidt if (copy_to_user(arg, sa, sizeof(*sa))) 4338475f6387SJan Schmidt ret = -EFAULT; 4339475f6387SJan Schmidt 4340475f6387SJan Schmidt kfree(sa); 4341475f6387SJan Schmidt return ret; 4342475f6387SJan Schmidt } 4343475f6387SJan Schmidt 4344c11d2c23SStefan Behrens static long btrfs_ioctl_get_dev_stats(struct btrfs_root *root, 4345b27f7c0cSDavid Sterba void __user *arg) 4346c11d2c23SStefan Behrens { 4347c11d2c23SStefan Behrens struct btrfs_ioctl_get_dev_stats *sa; 4348c11d2c23SStefan Behrens int ret; 4349c11d2c23SStefan Behrens 4350c11d2c23SStefan Behrens sa = memdup_user(arg, sizeof(*sa)); 4351c11d2c23SStefan Behrens if (IS_ERR(sa)) 4352c11d2c23SStefan Behrens return PTR_ERR(sa); 4353c11d2c23SStefan Behrens 4354b27f7c0cSDavid Sterba if ((sa->flags & BTRFS_DEV_STATS_RESET) && !capable(CAP_SYS_ADMIN)) { 4355b27f7c0cSDavid Sterba kfree(sa); 4356b27f7c0cSDavid Sterba return -EPERM; 4357b27f7c0cSDavid Sterba } 4358b27f7c0cSDavid Sterba 4359b27f7c0cSDavid Sterba ret = btrfs_get_dev_stats(root, sa); 4360c11d2c23SStefan Behrens 4361c11d2c23SStefan Behrens if (copy_to_user(arg, sa, sizeof(*sa))) 4362c11d2c23SStefan Behrens ret = -EFAULT; 4363c11d2c23SStefan Behrens 4364c11d2c23SStefan Behrens kfree(sa); 4365c11d2c23SStefan Behrens return ret; 4366c11d2c23SStefan Behrens } 4367c11d2c23SStefan Behrens 43683f6bcfbdSStefan Behrens static long btrfs_ioctl_dev_replace(struct btrfs_root *root, void __user *arg) 43693f6bcfbdSStefan Behrens { 43703f6bcfbdSStefan Behrens struct btrfs_ioctl_dev_replace_args *p; 43713f6bcfbdSStefan Behrens int ret; 43723f6bcfbdSStefan Behrens 43733f6bcfbdSStefan Behrens if (!capable(CAP_SYS_ADMIN)) 43743f6bcfbdSStefan Behrens return -EPERM; 43753f6bcfbdSStefan Behrens 43763f6bcfbdSStefan Behrens p = memdup_user(arg, sizeof(*p)); 43773f6bcfbdSStefan Behrens if (IS_ERR(p)) 43783f6bcfbdSStefan Behrens return PTR_ERR(p); 43793f6bcfbdSStefan Behrens 43803f6bcfbdSStefan Behrens switch (p->cmd) { 43813f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_START: 4382adfa97cbSIlya Dryomov if (root->fs_info->sb->s_flags & MS_RDONLY) { 4383adfa97cbSIlya Dryomov ret = -EROFS; 4384adfa97cbSIlya Dryomov goto out; 4385adfa97cbSIlya Dryomov } 43863f6bcfbdSStefan Behrens if (atomic_xchg( 43873f6bcfbdSStefan Behrens &root->fs_info->mutually_exclusive_operation_running, 43883f6bcfbdSStefan Behrens 1)) { 4389e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 43903f6bcfbdSStefan Behrens } else { 43913f6bcfbdSStefan Behrens ret = btrfs_dev_replace_start(root, p); 43923f6bcfbdSStefan Behrens atomic_set( 43933f6bcfbdSStefan Behrens &root->fs_info->mutually_exclusive_operation_running, 43943f6bcfbdSStefan Behrens 0); 43953f6bcfbdSStefan Behrens } 43963f6bcfbdSStefan Behrens break; 43973f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_STATUS: 43983f6bcfbdSStefan Behrens btrfs_dev_replace_status(root->fs_info, p); 43993f6bcfbdSStefan Behrens ret = 0; 44003f6bcfbdSStefan Behrens break; 44013f6bcfbdSStefan Behrens case BTRFS_IOCTL_DEV_REPLACE_CMD_CANCEL: 44023f6bcfbdSStefan Behrens ret = btrfs_dev_replace_cancel(root->fs_info, p); 44033f6bcfbdSStefan Behrens break; 44043f6bcfbdSStefan Behrens default: 44053f6bcfbdSStefan Behrens ret = -EINVAL; 44063f6bcfbdSStefan Behrens break; 44073f6bcfbdSStefan Behrens } 44083f6bcfbdSStefan Behrens 44093f6bcfbdSStefan Behrens if (copy_to_user(arg, p, sizeof(*p))) 44103f6bcfbdSStefan Behrens ret = -EFAULT; 4411adfa97cbSIlya Dryomov out: 44123f6bcfbdSStefan Behrens kfree(p); 44133f6bcfbdSStefan Behrens return ret; 44143f6bcfbdSStefan Behrens } 44153f6bcfbdSStefan Behrens 4416d7728c96SJan Schmidt static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg) 4417d7728c96SJan Schmidt { 4418d7728c96SJan Schmidt int ret = 0; 4419d7728c96SJan Schmidt int i; 4420740c3d22SChris Mason u64 rel_ptr; 4421d7728c96SJan Schmidt int size; 4422806468f8SChris Mason struct btrfs_ioctl_ino_path_args *ipa = NULL; 4423d7728c96SJan Schmidt struct inode_fs_paths *ipath = NULL; 4424d7728c96SJan Schmidt struct btrfs_path *path; 4425d7728c96SJan Schmidt 442682b22ac8SKusanagi Kouichi if (!capable(CAP_DAC_READ_SEARCH)) 4427d7728c96SJan Schmidt return -EPERM; 4428d7728c96SJan Schmidt 4429d7728c96SJan Schmidt path = btrfs_alloc_path(); 4430d7728c96SJan Schmidt if (!path) { 4431d7728c96SJan Schmidt ret = -ENOMEM; 4432d7728c96SJan Schmidt goto out; 4433d7728c96SJan Schmidt } 4434d7728c96SJan Schmidt 4435d7728c96SJan Schmidt ipa = memdup_user(arg, sizeof(*ipa)); 4436d7728c96SJan Schmidt if (IS_ERR(ipa)) { 4437d7728c96SJan Schmidt ret = PTR_ERR(ipa); 4438d7728c96SJan Schmidt ipa = NULL; 4439d7728c96SJan Schmidt goto out; 4440d7728c96SJan Schmidt } 4441d7728c96SJan Schmidt 4442d7728c96SJan Schmidt size = min_t(u32, ipa->size, 4096); 4443d7728c96SJan Schmidt ipath = init_ipath(size, root, path); 4444d7728c96SJan Schmidt if (IS_ERR(ipath)) { 4445d7728c96SJan Schmidt ret = PTR_ERR(ipath); 4446d7728c96SJan Schmidt ipath = NULL; 4447d7728c96SJan Schmidt goto out; 4448d7728c96SJan Schmidt } 4449d7728c96SJan Schmidt 4450d7728c96SJan Schmidt ret = paths_from_inode(ipa->inum, ipath); 4451d7728c96SJan Schmidt if (ret < 0) 4452d7728c96SJan Schmidt goto out; 4453d7728c96SJan Schmidt 4454d7728c96SJan Schmidt for (i = 0; i < ipath->fspath->elem_cnt; ++i) { 4455745c4d8eSJeff Mahoney rel_ptr = ipath->fspath->val[i] - 4456745c4d8eSJeff Mahoney (u64)(unsigned long)ipath->fspath->val; 4457740c3d22SChris Mason ipath->fspath->val[i] = rel_ptr; 4458d7728c96SJan Schmidt } 4459d7728c96SJan Schmidt 4460745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)ipa->fspath, 4461745c4d8eSJeff Mahoney (void *)(unsigned long)ipath->fspath, size); 4462d7728c96SJan Schmidt if (ret) { 4463d7728c96SJan Schmidt ret = -EFAULT; 4464d7728c96SJan Schmidt goto out; 4465d7728c96SJan Schmidt } 4466d7728c96SJan Schmidt 4467d7728c96SJan Schmidt out: 4468d7728c96SJan Schmidt btrfs_free_path(path); 4469d7728c96SJan Schmidt free_ipath(ipath); 4470d7728c96SJan Schmidt kfree(ipa); 4471d7728c96SJan Schmidt 4472d7728c96SJan Schmidt return ret; 4473d7728c96SJan Schmidt } 4474d7728c96SJan Schmidt 4475d7728c96SJan Schmidt static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx) 4476d7728c96SJan Schmidt { 4477d7728c96SJan Schmidt struct btrfs_data_container *inodes = ctx; 4478d7728c96SJan Schmidt const size_t c = 3 * sizeof(u64); 4479d7728c96SJan Schmidt 4480d7728c96SJan Schmidt if (inodes->bytes_left >= c) { 4481d7728c96SJan Schmidt inodes->bytes_left -= c; 4482d7728c96SJan Schmidt inodes->val[inodes->elem_cnt] = inum; 4483d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 1] = offset; 4484d7728c96SJan Schmidt inodes->val[inodes->elem_cnt + 2] = root; 4485d7728c96SJan Schmidt inodes->elem_cnt += 3; 4486d7728c96SJan Schmidt } else { 4487d7728c96SJan Schmidt inodes->bytes_missing += c - inodes->bytes_left; 4488d7728c96SJan Schmidt inodes->bytes_left = 0; 4489d7728c96SJan Schmidt inodes->elem_missed += 3; 4490d7728c96SJan Schmidt } 4491d7728c96SJan Schmidt 4492d7728c96SJan Schmidt return 0; 4493d7728c96SJan Schmidt } 4494d7728c96SJan Schmidt 4495d7728c96SJan Schmidt static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root, 4496d7728c96SJan Schmidt void __user *arg) 4497d7728c96SJan Schmidt { 4498d7728c96SJan Schmidt int ret = 0; 4499d7728c96SJan Schmidt int size; 4500d7728c96SJan Schmidt struct btrfs_ioctl_logical_ino_args *loi; 4501d7728c96SJan Schmidt struct btrfs_data_container *inodes = NULL; 4502d7728c96SJan Schmidt struct btrfs_path *path = NULL; 4503d7728c96SJan Schmidt 4504d7728c96SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 4505d7728c96SJan Schmidt return -EPERM; 4506d7728c96SJan Schmidt 4507d7728c96SJan Schmidt loi = memdup_user(arg, sizeof(*loi)); 4508d7728c96SJan Schmidt if (IS_ERR(loi)) { 4509d7728c96SJan Schmidt ret = PTR_ERR(loi); 4510d7728c96SJan Schmidt loi = NULL; 4511d7728c96SJan Schmidt goto out; 4512d7728c96SJan Schmidt } 4513d7728c96SJan Schmidt 4514d7728c96SJan Schmidt path = btrfs_alloc_path(); 4515d7728c96SJan Schmidt if (!path) { 4516d7728c96SJan Schmidt ret = -ENOMEM; 4517d7728c96SJan Schmidt goto out; 4518d7728c96SJan Schmidt } 4519d7728c96SJan Schmidt 4520ee22184bSByongho Lee size = min_t(u32, loi->size, SZ_64K); 4521d7728c96SJan Schmidt inodes = init_data_container(size); 4522d7728c96SJan Schmidt if (IS_ERR(inodes)) { 4523d7728c96SJan Schmidt ret = PTR_ERR(inodes); 4524d7728c96SJan Schmidt inodes = NULL; 4525d7728c96SJan Schmidt goto out; 4526d7728c96SJan Schmidt } 4527d7728c96SJan Schmidt 4528df031f07SLiu Bo ret = iterate_inodes_from_logical(loi->logical, root->fs_info, path, 4529df031f07SLiu Bo build_ino_list, inodes); 4530df031f07SLiu Bo if (ret == -EINVAL) 4531d7728c96SJan Schmidt ret = -ENOENT; 4532d7728c96SJan Schmidt if (ret < 0) 4533d7728c96SJan Schmidt goto out; 4534d7728c96SJan Schmidt 4535745c4d8eSJeff Mahoney ret = copy_to_user((void *)(unsigned long)loi->inodes, 4536745c4d8eSJeff Mahoney (void *)(unsigned long)inodes, size); 4537d7728c96SJan Schmidt if (ret) 4538d7728c96SJan Schmidt ret = -EFAULT; 4539d7728c96SJan Schmidt 4540d7728c96SJan Schmidt out: 4541d7728c96SJan Schmidt btrfs_free_path(path); 4542425d17a2SLiu Bo vfree(inodes); 4543d7728c96SJan Schmidt kfree(loi); 4544d7728c96SJan Schmidt 4545d7728c96SJan Schmidt return ret; 4546d7728c96SJan Schmidt } 4547d7728c96SJan Schmidt 454819a39dceSIlya Dryomov void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, 4549c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs) 4550c9e9f97bSIlya Dryomov { 4551c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl = fs_info->balance_ctl; 4552c9e9f97bSIlya Dryomov 4553c9e9f97bSIlya Dryomov bargs->flags = bctl->flags; 4554c9e9f97bSIlya Dryomov 4555837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_running)) 4556837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_RUNNING; 4557837d5b6eSIlya Dryomov if (atomic_read(&fs_info->balance_pause_req)) 4558837d5b6eSIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_PAUSE_REQ; 4559a7e99c69SIlya Dryomov if (atomic_read(&fs_info->balance_cancel_req)) 4560a7e99c69SIlya Dryomov bargs->state |= BTRFS_BALANCE_STATE_CANCEL_REQ; 4561837d5b6eSIlya Dryomov 4562c9e9f97bSIlya Dryomov memcpy(&bargs->data, &bctl->data, sizeof(bargs->data)); 4563c9e9f97bSIlya Dryomov memcpy(&bargs->meta, &bctl->meta, sizeof(bargs->meta)); 4564c9e9f97bSIlya Dryomov memcpy(&bargs->sys, &bctl->sys, sizeof(bargs->sys)); 456519a39dceSIlya Dryomov 456619a39dceSIlya Dryomov if (lock) { 456719a39dceSIlya Dryomov spin_lock(&fs_info->balance_lock); 456819a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 456919a39dceSIlya Dryomov spin_unlock(&fs_info->balance_lock); 457019a39dceSIlya Dryomov } else { 457119a39dceSIlya Dryomov memcpy(&bargs->stat, &bctl->stat, sizeof(bargs->stat)); 457219a39dceSIlya Dryomov } 4573c9e9f97bSIlya Dryomov } 4574c9e9f97bSIlya Dryomov 45759ba1f6e4SLiu Bo static long btrfs_ioctl_balance(struct file *file, void __user *arg) 4576c9e9f97bSIlya Dryomov { 4577496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 4578c9e9f97bSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 4579c9e9f97bSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 4580c9e9f97bSIlya Dryomov struct btrfs_balance_control *bctl; 4581ed0fb78fSIlya Dryomov bool need_unlock; /* for mut. excl. ops lock */ 4582c9e9f97bSIlya Dryomov int ret; 4583c9e9f97bSIlya Dryomov 4584c9e9f97bSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4585c9e9f97bSIlya Dryomov return -EPERM; 4586c9e9f97bSIlya Dryomov 4587e54bfa31SLiu Bo ret = mnt_want_write_file(file); 45889ba1f6e4SLiu Bo if (ret) 45899ba1f6e4SLiu Bo return ret; 45909ba1f6e4SLiu Bo 4591ed0fb78fSIlya Dryomov again: 4592ed0fb78fSIlya Dryomov if (!atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)) { 4593c9e9f97bSIlya Dryomov mutex_lock(&fs_info->volume_mutex); 4594c9e9f97bSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4595ed0fb78fSIlya Dryomov need_unlock = true; 4596ed0fb78fSIlya Dryomov goto locked; 4597ed0fb78fSIlya Dryomov } 4598ed0fb78fSIlya Dryomov 4599ed0fb78fSIlya Dryomov /* 4600ed0fb78fSIlya Dryomov * mut. excl. ops lock is locked. Three possibilites: 4601ed0fb78fSIlya Dryomov * (1) some other op is running 4602ed0fb78fSIlya Dryomov * (2) balance is running 4603ed0fb78fSIlya Dryomov * (3) balance is paused -- special case (think resume) 4604ed0fb78fSIlya Dryomov */ 4605ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4606ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4607ed0fb78fSIlya Dryomov /* this is either (2) or (3) */ 4608ed0fb78fSIlya Dryomov if (!atomic_read(&fs_info->balance_running)) { 4609ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4610ed0fb78fSIlya Dryomov if (!mutex_trylock(&fs_info->volume_mutex)) 4611ed0fb78fSIlya Dryomov goto again; 4612ed0fb78fSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 4613ed0fb78fSIlya Dryomov 4614ed0fb78fSIlya Dryomov if (fs_info->balance_ctl && 4615ed0fb78fSIlya Dryomov !atomic_read(&fs_info->balance_running)) { 4616ed0fb78fSIlya Dryomov /* this is (3) */ 4617ed0fb78fSIlya Dryomov need_unlock = false; 4618ed0fb78fSIlya Dryomov goto locked; 4619ed0fb78fSIlya Dryomov } 4620ed0fb78fSIlya Dryomov 4621ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4622ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->volume_mutex); 4623ed0fb78fSIlya Dryomov goto again; 4624ed0fb78fSIlya Dryomov } else { 4625ed0fb78fSIlya Dryomov /* this is (2) */ 4626ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4627ed0fb78fSIlya Dryomov ret = -EINPROGRESS; 4628ed0fb78fSIlya Dryomov goto out; 4629ed0fb78fSIlya Dryomov } 4630ed0fb78fSIlya Dryomov } else { 4631ed0fb78fSIlya Dryomov /* this is (1) */ 4632ed0fb78fSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4633e57138b3SAnand Jain ret = BTRFS_ERROR_DEV_EXCL_RUN_IN_PROGRESS; 4634ed0fb78fSIlya Dryomov goto out; 4635ed0fb78fSIlya Dryomov } 4636ed0fb78fSIlya Dryomov 4637ed0fb78fSIlya Dryomov locked: 4638ed0fb78fSIlya Dryomov BUG_ON(!atomic_read(&fs_info->mutually_exclusive_operation_running)); 4639c9e9f97bSIlya Dryomov 4640c9e9f97bSIlya Dryomov if (arg) { 4641c9e9f97bSIlya Dryomov bargs = memdup_user(arg, sizeof(*bargs)); 4642c9e9f97bSIlya Dryomov if (IS_ERR(bargs)) { 4643c9e9f97bSIlya Dryomov ret = PTR_ERR(bargs); 4644ed0fb78fSIlya Dryomov goto out_unlock; 4645c9e9f97bSIlya Dryomov } 4646de322263SIlya Dryomov 4647de322263SIlya Dryomov if (bargs->flags & BTRFS_BALANCE_RESUME) { 4648de322263SIlya Dryomov if (!fs_info->balance_ctl) { 4649de322263SIlya Dryomov ret = -ENOTCONN; 4650de322263SIlya Dryomov goto out_bargs; 4651de322263SIlya Dryomov } 4652de322263SIlya Dryomov 4653de322263SIlya Dryomov bctl = fs_info->balance_ctl; 4654de322263SIlya Dryomov spin_lock(&fs_info->balance_lock); 4655de322263SIlya Dryomov bctl->flags |= BTRFS_BALANCE_RESUME; 4656de322263SIlya Dryomov spin_unlock(&fs_info->balance_lock); 4657de322263SIlya Dryomov 4658de322263SIlya Dryomov goto do_balance; 4659de322263SIlya Dryomov } 4660c9e9f97bSIlya Dryomov } else { 4661c9e9f97bSIlya Dryomov bargs = NULL; 4662c9e9f97bSIlya Dryomov } 4663c9e9f97bSIlya Dryomov 4664ed0fb78fSIlya Dryomov if (fs_info->balance_ctl) { 4665837d5b6eSIlya Dryomov ret = -EINPROGRESS; 4666837d5b6eSIlya Dryomov goto out_bargs; 4667837d5b6eSIlya Dryomov } 4668837d5b6eSIlya Dryomov 46698d2db785SDavid Sterba bctl = kzalloc(sizeof(*bctl), GFP_KERNEL); 4670c9e9f97bSIlya Dryomov if (!bctl) { 4671c9e9f97bSIlya Dryomov ret = -ENOMEM; 4672c9e9f97bSIlya Dryomov goto out_bargs; 4673c9e9f97bSIlya Dryomov } 4674c9e9f97bSIlya Dryomov 4675c9e9f97bSIlya Dryomov bctl->fs_info = fs_info; 4676c9e9f97bSIlya Dryomov if (arg) { 4677c9e9f97bSIlya Dryomov memcpy(&bctl->data, &bargs->data, sizeof(bctl->data)); 4678c9e9f97bSIlya Dryomov memcpy(&bctl->meta, &bargs->meta, sizeof(bctl->meta)); 4679c9e9f97bSIlya Dryomov memcpy(&bctl->sys, &bargs->sys, sizeof(bctl->sys)); 4680c9e9f97bSIlya Dryomov 4681c9e9f97bSIlya Dryomov bctl->flags = bargs->flags; 4682f43ffb60SIlya Dryomov } else { 4683f43ffb60SIlya Dryomov /* balance everything - no filters */ 4684f43ffb60SIlya Dryomov bctl->flags |= BTRFS_BALANCE_TYPE_MASK; 4685c9e9f97bSIlya Dryomov } 4686c9e9f97bSIlya Dryomov 46878eb93459SDavid Sterba if (bctl->flags & ~(BTRFS_BALANCE_ARGS_MASK | BTRFS_BALANCE_TYPE_MASK)) { 46888eb93459SDavid Sterba ret = -EINVAL; 46890f89abf5SChristian Engelmayer goto out_bctl; 46908eb93459SDavid Sterba } 46918eb93459SDavid Sterba 4692de322263SIlya Dryomov do_balance: 4693c9e9f97bSIlya Dryomov /* 4694ed0fb78fSIlya Dryomov * Ownership of bctl and mutually_exclusive_operation_running 4695ed0fb78fSIlya Dryomov * goes to to btrfs_balance. bctl is freed in __cancel_balance, 4696ed0fb78fSIlya Dryomov * or, if restriper was paused all the way until unmount, in 4697ed0fb78fSIlya Dryomov * free_fs_info. mutually_exclusive_operation_running is 4698ed0fb78fSIlya Dryomov * cleared in __cancel_balance. 4699c9e9f97bSIlya Dryomov */ 4700ed0fb78fSIlya Dryomov need_unlock = false; 4701ed0fb78fSIlya Dryomov 4702ed0fb78fSIlya Dryomov ret = btrfs_balance(bctl, bargs); 47030f89abf5SChristian Engelmayer bctl = NULL; 4704ed0fb78fSIlya Dryomov 4705c9e9f97bSIlya Dryomov if (arg) { 4706c9e9f97bSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 4707c9e9f97bSIlya Dryomov ret = -EFAULT; 4708c9e9f97bSIlya Dryomov } 4709c9e9f97bSIlya Dryomov 47100f89abf5SChristian Engelmayer out_bctl: 47110f89abf5SChristian Engelmayer kfree(bctl); 4712c9e9f97bSIlya Dryomov out_bargs: 4713c9e9f97bSIlya Dryomov kfree(bargs); 4714ed0fb78fSIlya Dryomov out_unlock: 4715c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 4716c9e9f97bSIlya Dryomov mutex_unlock(&fs_info->volume_mutex); 4717ed0fb78fSIlya Dryomov if (need_unlock) 4718ed0fb78fSIlya Dryomov atomic_set(&fs_info->mutually_exclusive_operation_running, 0); 4719ed0fb78fSIlya Dryomov out: 4720e54bfa31SLiu Bo mnt_drop_write_file(file); 4721c9e9f97bSIlya Dryomov return ret; 4722c9e9f97bSIlya Dryomov } 4723c9e9f97bSIlya Dryomov 4724837d5b6eSIlya Dryomov static long btrfs_ioctl_balance_ctl(struct btrfs_root *root, int cmd) 4725837d5b6eSIlya Dryomov { 4726837d5b6eSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 4727837d5b6eSIlya Dryomov return -EPERM; 4728837d5b6eSIlya Dryomov 4729837d5b6eSIlya Dryomov switch (cmd) { 4730837d5b6eSIlya Dryomov case BTRFS_BALANCE_CTL_PAUSE: 4731837d5b6eSIlya Dryomov return btrfs_pause_balance(root->fs_info); 4732a7e99c69SIlya Dryomov case BTRFS_BALANCE_CTL_CANCEL: 4733a7e99c69SIlya Dryomov return btrfs_cancel_balance(root->fs_info); 4734837d5b6eSIlya Dryomov } 4735837d5b6eSIlya Dryomov 4736837d5b6eSIlya Dryomov return -EINVAL; 4737837d5b6eSIlya Dryomov } 4738837d5b6eSIlya Dryomov 473919a39dceSIlya Dryomov static long btrfs_ioctl_balance_progress(struct btrfs_root *root, 474019a39dceSIlya Dryomov void __user *arg) 474119a39dceSIlya Dryomov { 474219a39dceSIlya Dryomov struct btrfs_fs_info *fs_info = root->fs_info; 474319a39dceSIlya Dryomov struct btrfs_ioctl_balance_args *bargs; 474419a39dceSIlya Dryomov int ret = 0; 474519a39dceSIlya Dryomov 474619a39dceSIlya Dryomov if (!capable(CAP_SYS_ADMIN)) 474719a39dceSIlya Dryomov return -EPERM; 474819a39dceSIlya Dryomov 474919a39dceSIlya Dryomov mutex_lock(&fs_info->balance_mutex); 475019a39dceSIlya Dryomov if (!fs_info->balance_ctl) { 475119a39dceSIlya Dryomov ret = -ENOTCONN; 475219a39dceSIlya Dryomov goto out; 475319a39dceSIlya Dryomov } 475419a39dceSIlya Dryomov 47558d2db785SDavid Sterba bargs = kzalloc(sizeof(*bargs), GFP_KERNEL); 475619a39dceSIlya Dryomov if (!bargs) { 475719a39dceSIlya Dryomov ret = -ENOMEM; 475819a39dceSIlya Dryomov goto out; 475919a39dceSIlya Dryomov } 476019a39dceSIlya Dryomov 476119a39dceSIlya Dryomov update_ioctl_balance_args(fs_info, 1, bargs); 476219a39dceSIlya Dryomov 476319a39dceSIlya Dryomov if (copy_to_user(arg, bargs, sizeof(*bargs))) 476419a39dceSIlya Dryomov ret = -EFAULT; 476519a39dceSIlya Dryomov 476619a39dceSIlya Dryomov kfree(bargs); 476719a39dceSIlya Dryomov out: 476819a39dceSIlya Dryomov mutex_unlock(&fs_info->balance_mutex); 476919a39dceSIlya Dryomov return ret; 477019a39dceSIlya Dryomov } 477119a39dceSIlya Dryomov 4772905b0ddaSMiao Xie static long btrfs_ioctl_quota_ctl(struct file *file, void __user *arg) 47735d13a37bSArne Jansen { 4774496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 47755d13a37bSArne Jansen struct btrfs_ioctl_quota_ctl_args *sa; 47765d13a37bSArne Jansen struct btrfs_trans_handle *trans = NULL; 47775d13a37bSArne Jansen int ret; 47785d13a37bSArne Jansen int err; 47795d13a37bSArne Jansen 47805d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 47815d13a37bSArne Jansen return -EPERM; 47825d13a37bSArne Jansen 4783905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4784905b0ddaSMiao Xie if (ret) 4785905b0ddaSMiao Xie return ret; 47865d13a37bSArne Jansen 47875d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4788905b0ddaSMiao Xie if (IS_ERR(sa)) { 4789905b0ddaSMiao Xie ret = PTR_ERR(sa); 4790905b0ddaSMiao Xie goto drop_write; 4791905b0ddaSMiao Xie } 47925d13a37bSArne Jansen 47937708f029SWang Shilong down_write(&root->fs_info->subvol_sem); 479492f183aaSWang Shilong trans = btrfs_start_transaction(root->fs_info->tree_root, 2); 47955d13a37bSArne Jansen if (IS_ERR(trans)) { 47965d13a37bSArne Jansen ret = PTR_ERR(trans); 47975d13a37bSArne Jansen goto out; 47985d13a37bSArne Jansen } 47995d13a37bSArne Jansen 48005d13a37bSArne Jansen switch (sa->cmd) { 48015d13a37bSArne Jansen case BTRFS_QUOTA_CTL_ENABLE: 48025d13a37bSArne Jansen ret = btrfs_quota_enable(trans, root->fs_info); 48035d13a37bSArne Jansen break; 48045d13a37bSArne Jansen case BTRFS_QUOTA_CTL_DISABLE: 48055d13a37bSArne Jansen ret = btrfs_quota_disable(trans, root->fs_info); 48065d13a37bSArne Jansen break; 48075d13a37bSArne Jansen default: 48085d13a37bSArne Jansen ret = -EINVAL; 48095d13a37bSArne Jansen break; 48105d13a37bSArne Jansen } 48115d13a37bSArne Jansen 481292f183aaSWang Shilong err = btrfs_commit_transaction(trans, root->fs_info->tree_root); 48135d13a37bSArne Jansen if (err && !ret) 48145d13a37bSArne Jansen ret = err; 48155d13a37bSArne Jansen out: 48165d13a37bSArne Jansen kfree(sa); 48177708f029SWang Shilong up_write(&root->fs_info->subvol_sem); 4818905b0ddaSMiao Xie drop_write: 4819905b0ddaSMiao Xie mnt_drop_write_file(file); 48205d13a37bSArne Jansen return ret; 48215d13a37bSArne Jansen } 48225d13a37bSArne Jansen 4823905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_assign(struct file *file, void __user *arg) 48245d13a37bSArne Jansen { 4825496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 48265d13a37bSArne Jansen struct btrfs_ioctl_qgroup_assign_args *sa; 48275d13a37bSArne Jansen struct btrfs_trans_handle *trans; 48285d13a37bSArne Jansen int ret; 48295d13a37bSArne Jansen int err; 48305d13a37bSArne Jansen 48315d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 48325d13a37bSArne Jansen return -EPERM; 48335d13a37bSArne Jansen 4834905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4835905b0ddaSMiao Xie if (ret) 4836905b0ddaSMiao Xie return ret; 48375d13a37bSArne Jansen 48385d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4839905b0ddaSMiao Xie if (IS_ERR(sa)) { 4840905b0ddaSMiao Xie ret = PTR_ERR(sa); 4841905b0ddaSMiao Xie goto drop_write; 4842905b0ddaSMiao Xie } 48435d13a37bSArne Jansen 48445d13a37bSArne Jansen trans = btrfs_join_transaction(root); 48455d13a37bSArne Jansen if (IS_ERR(trans)) { 48465d13a37bSArne Jansen ret = PTR_ERR(trans); 48475d13a37bSArne Jansen goto out; 48485d13a37bSArne Jansen } 48495d13a37bSArne Jansen 48505d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 48515d13a37bSArne Jansen if (sa->assign) { 48525d13a37bSArne Jansen ret = btrfs_add_qgroup_relation(trans, root->fs_info, 48535d13a37bSArne Jansen sa->src, sa->dst); 48545d13a37bSArne Jansen } else { 48555d13a37bSArne Jansen ret = btrfs_del_qgroup_relation(trans, root->fs_info, 48565d13a37bSArne Jansen sa->src, sa->dst); 48575d13a37bSArne Jansen } 48585d13a37bSArne Jansen 4859e082f563SQu Wenruo /* update qgroup status and info */ 4860e082f563SQu Wenruo err = btrfs_run_qgroups(trans, root->fs_info); 4861e082f563SQu Wenruo if (err < 0) 4862a4553fefSAnand Jain btrfs_std_error(root->fs_info, ret, 4863e082f563SQu Wenruo "failed to update qgroup status and info\n"); 48645d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 48655d13a37bSArne Jansen if (err && !ret) 48665d13a37bSArne Jansen ret = err; 48675d13a37bSArne Jansen 48685d13a37bSArne Jansen out: 48695d13a37bSArne Jansen kfree(sa); 4870905b0ddaSMiao Xie drop_write: 4871905b0ddaSMiao Xie mnt_drop_write_file(file); 48725d13a37bSArne Jansen return ret; 48735d13a37bSArne Jansen } 48745d13a37bSArne Jansen 4875905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_create(struct file *file, void __user *arg) 48765d13a37bSArne Jansen { 4877496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 48785d13a37bSArne Jansen struct btrfs_ioctl_qgroup_create_args *sa; 48795d13a37bSArne Jansen struct btrfs_trans_handle *trans; 48805d13a37bSArne Jansen int ret; 48815d13a37bSArne Jansen int err; 48825d13a37bSArne Jansen 48835d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 48845d13a37bSArne Jansen return -EPERM; 48855d13a37bSArne Jansen 4886905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4887905b0ddaSMiao Xie if (ret) 4888905b0ddaSMiao Xie return ret; 48895d13a37bSArne Jansen 48905d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4891905b0ddaSMiao Xie if (IS_ERR(sa)) { 4892905b0ddaSMiao Xie ret = PTR_ERR(sa); 4893905b0ddaSMiao Xie goto drop_write; 4894905b0ddaSMiao Xie } 48955d13a37bSArne Jansen 4896d86e56cfSMiao Xie if (!sa->qgroupid) { 4897d86e56cfSMiao Xie ret = -EINVAL; 4898d86e56cfSMiao Xie goto out; 4899d86e56cfSMiao Xie } 4900d86e56cfSMiao Xie 49015d13a37bSArne Jansen trans = btrfs_join_transaction(root); 49025d13a37bSArne Jansen if (IS_ERR(trans)) { 49035d13a37bSArne Jansen ret = PTR_ERR(trans); 49045d13a37bSArne Jansen goto out; 49055d13a37bSArne Jansen } 49065d13a37bSArne Jansen 49075d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 49085d13a37bSArne Jansen if (sa->create) { 49094087cf24SDongsheng Yang ret = btrfs_create_qgroup(trans, root->fs_info, sa->qgroupid); 49105d13a37bSArne Jansen } else { 49115d13a37bSArne Jansen ret = btrfs_remove_qgroup(trans, root->fs_info, sa->qgroupid); 49125d13a37bSArne Jansen } 49135d13a37bSArne Jansen 49145d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 49155d13a37bSArne Jansen if (err && !ret) 49165d13a37bSArne Jansen ret = err; 49175d13a37bSArne Jansen 49185d13a37bSArne Jansen out: 49195d13a37bSArne Jansen kfree(sa); 4920905b0ddaSMiao Xie drop_write: 4921905b0ddaSMiao Xie mnt_drop_write_file(file); 49225d13a37bSArne Jansen return ret; 49235d13a37bSArne Jansen } 49245d13a37bSArne Jansen 4925905b0ddaSMiao Xie static long btrfs_ioctl_qgroup_limit(struct file *file, void __user *arg) 49265d13a37bSArne Jansen { 4927496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 49285d13a37bSArne Jansen struct btrfs_ioctl_qgroup_limit_args *sa; 49295d13a37bSArne Jansen struct btrfs_trans_handle *trans; 49305d13a37bSArne Jansen int ret; 49315d13a37bSArne Jansen int err; 49325d13a37bSArne Jansen u64 qgroupid; 49335d13a37bSArne Jansen 49345d13a37bSArne Jansen if (!capable(CAP_SYS_ADMIN)) 49355d13a37bSArne Jansen return -EPERM; 49365d13a37bSArne Jansen 4937905b0ddaSMiao Xie ret = mnt_want_write_file(file); 4938905b0ddaSMiao Xie if (ret) 4939905b0ddaSMiao Xie return ret; 49405d13a37bSArne Jansen 49415d13a37bSArne Jansen sa = memdup_user(arg, sizeof(*sa)); 4942905b0ddaSMiao Xie if (IS_ERR(sa)) { 4943905b0ddaSMiao Xie ret = PTR_ERR(sa); 4944905b0ddaSMiao Xie goto drop_write; 4945905b0ddaSMiao Xie } 49465d13a37bSArne Jansen 49475d13a37bSArne Jansen trans = btrfs_join_transaction(root); 49485d13a37bSArne Jansen if (IS_ERR(trans)) { 49495d13a37bSArne Jansen ret = PTR_ERR(trans); 49505d13a37bSArne Jansen goto out; 49515d13a37bSArne Jansen } 49525d13a37bSArne Jansen 49535d13a37bSArne Jansen qgroupid = sa->qgroupid; 49545d13a37bSArne Jansen if (!qgroupid) { 49555d13a37bSArne Jansen /* take the current subvol as qgroup */ 49565d13a37bSArne Jansen qgroupid = root->root_key.objectid; 49575d13a37bSArne Jansen } 49585d13a37bSArne Jansen 49595d13a37bSArne Jansen /* FIXME: check if the IDs really exist */ 49605d13a37bSArne Jansen ret = btrfs_limit_qgroup(trans, root->fs_info, qgroupid, &sa->lim); 49615d13a37bSArne Jansen 49625d13a37bSArne Jansen err = btrfs_end_transaction(trans, root); 49635d13a37bSArne Jansen if (err && !ret) 49645d13a37bSArne Jansen ret = err; 49655d13a37bSArne Jansen 49665d13a37bSArne Jansen out: 49675d13a37bSArne Jansen kfree(sa); 4968905b0ddaSMiao Xie drop_write: 4969905b0ddaSMiao Xie mnt_drop_write_file(file); 49705d13a37bSArne Jansen return ret; 49715d13a37bSArne Jansen } 49725d13a37bSArne Jansen 49732f232036SJan Schmidt static long btrfs_ioctl_quota_rescan(struct file *file, void __user *arg) 49742f232036SJan Schmidt { 49756d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 49762f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 49772f232036SJan Schmidt int ret; 49782f232036SJan Schmidt 49792f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 49802f232036SJan Schmidt return -EPERM; 49812f232036SJan Schmidt 49822f232036SJan Schmidt ret = mnt_want_write_file(file); 49832f232036SJan Schmidt if (ret) 49842f232036SJan Schmidt return ret; 49852f232036SJan Schmidt 49862f232036SJan Schmidt qsa = memdup_user(arg, sizeof(*qsa)); 49872f232036SJan Schmidt if (IS_ERR(qsa)) { 49882f232036SJan Schmidt ret = PTR_ERR(qsa); 49892f232036SJan Schmidt goto drop_write; 49902f232036SJan Schmidt } 49912f232036SJan Schmidt 49922f232036SJan Schmidt if (qsa->flags) { 49932f232036SJan Schmidt ret = -EINVAL; 49942f232036SJan Schmidt goto out; 49952f232036SJan Schmidt } 49962f232036SJan Schmidt 49972f232036SJan Schmidt ret = btrfs_qgroup_rescan(root->fs_info); 49982f232036SJan Schmidt 49992f232036SJan Schmidt out: 50002f232036SJan Schmidt kfree(qsa); 50012f232036SJan Schmidt drop_write: 50022f232036SJan Schmidt mnt_drop_write_file(file); 50032f232036SJan Schmidt return ret; 50042f232036SJan Schmidt } 50052f232036SJan Schmidt 50062f232036SJan Schmidt static long btrfs_ioctl_quota_rescan_status(struct file *file, void __user *arg) 50072f232036SJan Schmidt { 50086d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 50092f232036SJan Schmidt struct btrfs_ioctl_quota_rescan_args *qsa; 50102f232036SJan Schmidt int ret = 0; 50112f232036SJan Schmidt 50122f232036SJan Schmidt if (!capable(CAP_SYS_ADMIN)) 50132f232036SJan Schmidt return -EPERM; 50142f232036SJan Schmidt 50158d2db785SDavid Sterba qsa = kzalloc(sizeof(*qsa), GFP_KERNEL); 50162f232036SJan Schmidt if (!qsa) 50172f232036SJan Schmidt return -ENOMEM; 50182f232036SJan Schmidt 50192f232036SJan Schmidt if (root->fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) { 50202f232036SJan Schmidt qsa->flags = 1; 50212f232036SJan Schmidt qsa->progress = root->fs_info->qgroup_rescan_progress.objectid; 50222f232036SJan Schmidt } 50232f232036SJan Schmidt 50242f232036SJan Schmidt if (copy_to_user(arg, qsa, sizeof(*qsa))) 50252f232036SJan Schmidt ret = -EFAULT; 50262f232036SJan Schmidt 50272f232036SJan Schmidt kfree(qsa); 50282f232036SJan Schmidt return ret; 50292f232036SJan Schmidt } 50302f232036SJan Schmidt 503157254b6eSJan Schmidt static long btrfs_ioctl_quota_rescan_wait(struct file *file, void __user *arg) 503257254b6eSJan Schmidt { 503354563d41SAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 503457254b6eSJan Schmidt 503557254b6eSJan Schmidt if (!capable(CAP_SYS_ADMIN)) 503657254b6eSJan Schmidt return -EPERM; 503757254b6eSJan Schmidt 503857254b6eSJan Schmidt return btrfs_qgroup_wait_for_completion(root->fs_info); 503957254b6eSJan Schmidt } 504057254b6eSJan Schmidt 5041abccd00fSHugo Mills static long _btrfs_ioctl_set_received_subvol(struct file *file, 5042abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa) 50438ea05e3aSAlexander Block { 5044496ad9aaSAl Viro struct inode *inode = file_inode(file); 50458ea05e3aSAlexander Block struct btrfs_root *root = BTRFS_I(inode)->root; 50468ea05e3aSAlexander Block struct btrfs_root_item *root_item = &root->root_item; 50478ea05e3aSAlexander Block struct btrfs_trans_handle *trans; 504804b285f3SDeepa Dinamani struct timespec ct = current_fs_time(inode->i_sb); 50498ea05e3aSAlexander Block int ret = 0; 5050dd5f9615SStefan Behrens int received_uuid_changed; 50518ea05e3aSAlexander Block 5052bd60ea0fSDavid Sterba if (!inode_owner_or_capable(inode)) 5053bd60ea0fSDavid Sterba return -EPERM; 5054bd60ea0fSDavid Sterba 50558ea05e3aSAlexander Block ret = mnt_want_write_file(file); 50568ea05e3aSAlexander Block if (ret < 0) 50578ea05e3aSAlexander Block return ret; 50588ea05e3aSAlexander Block 50598ea05e3aSAlexander Block down_write(&root->fs_info->subvol_sem); 50608ea05e3aSAlexander Block 50618ea05e3aSAlexander Block if (btrfs_ino(inode) != BTRFS_FIRST_FREE_OBJECTID) { 50628ea05e3aSAlexander Block ret = -EINVAL; 50638ea05e3aSAlexander Block goto out; 50648ea05e3aSAlexander Block } 50658ea05e3aSAlexander Block 50668ea05e3aSAlexander Block if (btrfs_root_readonly(root)) { 50678ea05e3aSAlexander Block ret = -EROFS; 50688ea05e3aSAlexander Block goto out; 50698ea05e3aSAlexander Block } 50708ea05e3aSAlexander Block 5071dd5f9615SStefan Behrens /* 5072dd5f9615SStefan Behrens * 1 - root item 5073dd5f9615SStefan Behrens * 2 - uuid items (received uuid + subvol uuid) 5074dd5f9615SStefan Behrens */ 5075dd5f9615SStefan Behrens trans = btrfs_start_transaction(root, 3); 50768ea05e3aSAlexander Block if (IS_ERR(trans)) { 50778ea05e3aSAlexander Block ret = PTR_ERR(trans); 50788ea05e3aSAlexander Block trans = NULL; 50798ea05e3aSAlexander Block goto out; 50808ea05e3aSAlexander Block } 50818ea05e3aSAlexander Block 50828ea05e3aSAlexander Block sa->rtransid = trans->transid; 50838ea05e3aSAlexander Block sa->rtime.sec = ct.tv_sec; 50848ea05e3aSAlexander Block sa->rtime.nsec = ct.tv_nsec; 50858ea05e3aSAlexander Block 5086dd5f9615SStefan Behrens received_uuid_changed = memcmp(root_item->received_uuid, sa->uuid, 5087dd5f9615SStefan Behrens BTRFS_UUID_SIZE); 5088dd5f9615SStefan Behrens if (received_uuid_changed && 5089dd5f9615SStefan Behrens !btrfs_is_empty_uuid(root_item->received_uuid)) 5090dd5f9615SStefan Behrens btrfs_uuid_tree_rem(trans, root->fs_info->uuid_root, 5091dd5f9615SStefan Behrens root_item->received_uuid, 5092dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 5093dd5f9615SStefan Behrens root->root_key.objectid); 50948ea05e3aSAlexander Block memcpy(root_item->received_uuid, sa->uuid, BTRFS_UUID_SIZE); 50958ea05e3aSAlexander Block btrfs_set_root_stransid(root_item, sa->stransid); 50968ea05e3aSAlexander Block btrfs_set_root_rtransid(root_item, sa->rtransid); 50973cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->stime, sa->stime.sec); 50983cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->stime, sa->stime.nsec); 50993cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&root_item->rtime, sa->rtime.sec); 51003cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&root_item->rtime, sa->rtime.nsec); 51018ea05e3aSAlexander Block 51028ea05e3aSAlexander Block ret = btrfs_update_root(trans, root->fs_info->tree_root, 51038ea05e3aSAlexander Block &root->root_key, &root->root_item); 51048ea05e3aSAlexander Block if (ret < 0) { 51058ea05e3aSAlexander Block btrfs_end_transaction(trans, root); 51068ea05e3aSAlexander Block goto out; 5107dd5f9615SStefan Behrens } 5108dd5f9615SStefan Behrens if (received_uuid_changed && !btrfs_is_empty_uuid(sa->uuid)) { 5109dd5f9615SStefan Behrens ret = btrfs_uuid_tree_add(trans, root->fs_info->uuid_root, 5110dd5f9615SStefan Behrens sa->uuid, 5111dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 5112dd5f9615SStefan Behrens root->root_key.objectid); 5113dd5f9615SStefan Behrens if (ret < 0 && ret != -EEXIST) { 5114dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 5115dd5f9615SStefan Behrens goto out; 5116dd5f9615SStefan Behrens } 5117dd5f9615SStefan Behrens } 51188ea05e3aSAlexander Block ret = btrfs_commit_transaction(trans, root); 5119dd5f9615SStefan Behrens if (ret < 0) { 5120dd5f9615SStefan Behrens btrfs_abort_transaction(trans, root, ret); 51218ea05e3aSAlexander Block goto out; 51228ea05e3aSAlexander Block } 51238ea05e3aSAlexander Block 5124abccd00fSHugo Mills out: 5125abccd00fSHugo Mills up_write(&root->fs_info->subvol_sem); 5126abccd00fSHugo Mills mnt_drop_write_file(file); 5127abccd00fSHugo Mills return ret; 5128abccd00fSHugo Mills } 5129abccd00fSHugo Mills 5130abccd00fSHugo Mills #ifdef CONFIG_64BIT 5131abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol_32(struct file *file, 5132abccd00fSHugo Mills void __user *arg) 5133abccd00fSHugo Mills { 5134abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args_32 *args32 = NULL; 5135abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *args64 = NULL; 5136abccd00fSHugo Mills int ret = 0; 5137abccd00fSHugo Mills 5138abccd00fSHugo Mills args32 = memdup_user(arg, sizeof(*args32)); 5139abccd00fSHugo Mills if (IS_ERR(args32)) { 5140abccd00fSHugo Mills ret = PTR_ERR(args32); 5141abccd00fSHugo Mills args32 = NULL; 5142abccd00fSHugo Mills goto out; 5143abccd00fSHugo Mills } 5144abccd00fSHugo Mills 51458d2db785SDavid Sterba args64 = kmalloc(sizeof(*args64), GFP_KERNEL); 514684dbeb87SDan Carpenter if (!args64) { 514784dbeb87SDan Carpenter ret = -ENOMEM; 5148abccd00fSHugo Mills goto out; 5149abccd00fSHugo Mills } 5150abccd00fSHugo Mills 5151abccd00fSHugo Mills memcpy(args64->uuid, args32->uuid, BTRFS_UUID_SIZE); 5152abccd00fSHugo Mills args64->stransid = args32->stransid; 5153abccd00fSHugo Mills args64->rtransid = args32->rtransid; 5154abccd00fSHugo Mills args64->stime.sec = args32->stime.sec; 5155abccd00fSHugo Mills args64->stime.nsec = args32->stime.nsec; 5156abccd00fSHugo Mills args64->rtime.sec = args32->rtime.sec; 5157abccd00fSHugo Mills args64->rtime.nsec = args32->rtime.nsec; 5158abccd00fSHugo Mills args64->flags = args32->flags; 5159abccd00fSHugo Mills 5160abccd00fSHugo Mills ret = _btrfs_ioctl_set_received_subvol(file, args64); 5161abccd00fSHugo Mills if (ret) 5162abccd00fSHugo Mills goto out; 5163abccd00fSHugo Mills 5164abccd00fSHugo Mills memcpy(args32->uuid, args64->uuid, BTRFS_UUID_SIZE); 5165abccd00fSHugo Mills args32->stransid = args64->stransid; 5166abccd00fSHugo Mills args32->rtransid = args64->rtransid; 5167abccd00fSHugo Mills args32->stime.sec = args64->stime.sec; 5168abccd00fSHugo Mills args32->stime.nsec = args64->stime.nsec; 5169abccd00fSHugo Mills args32->rtime.sec = args64->rtime.sec; 5170abccd00fSHugo Mills args32->rtime.nsec = args64->rtime.nsec; 5171abccd00fSHugo Mills args32->flags = args64->flags; 5172abccd00fSHugo Mills 5173abccd00fSHugo Mills ret = copy_to_user(arg, args32, sizeof(*args32)); 5174abccd00fSHugo Mills if (ret) 5175abccd00fSHugo Mills ret = -EFAULT; 5176abccd00fSHugo Mills 5177abccd00fSHugo Mills out: 5178abccd00fSHugo Mills kfree(args32); 5179abccd00fSHugo Mills kfree(args64); 5180abccd00fSHugo Mills return ret; 5181abccd00fSHugo Mills } 5182abccd00fSHugo Mills #endif 5183abccd00fSHugo Mills 5184abccd00fSHugo Mills static long btrfs_ioctl_set_received_subvol(struct file *file, 5185abccd00fSHugo Mills void __user *arg) 5186abccd00fSHugo Mills { 5187abccd00fSHugo Mills struct btrfs_ioctl_received_subvol_args *sa = NULL; 5188abccd00fSHugo Mills int ret = 0; 5189abccd00fSHugo Mills 5190abccd00fSHugo Mills sa = memdup_user(arg, sizeof(*sa)); 5191abccd00fSHugo Mills if (IS_ERR(sa)) { 5192abccd00fSHugo Mills ret = PTR_ERR(sa); 5193abccd00fSHugo Mills sa = NULL; 5194abccd00fSHugo Mills goto out; 5195abccd00fSHugo Mills } 5196abccd00fSHugo Mills 5197abccd00fSHugo Mills ret = _btrfs_ioctl_set_received_subvol(file, sa); 5198abccd00fSHugo Mills 5199abccd00fSHugo Mills if (ret) 5200abccd00fSHugo Mills goto out; 5201abccd00fSHugo Mills 52028ea05e3aSAlexander Block ret = copy_to_user(arg, sa, sizeof(*sa)); 52038ea05e3aSAlexander Block if (ret) 52048ea05e3aSAlexander Block ret = -EFAULT; 52058ea05e3aSAlexander Block 52068ea05e3aSAlexander Block out: 52078ea05e3aSAlexander Block kfree(sa); 52088ea05e3aSAlexander Block return ret; 52098ea05e3aSAlexander Block } 52108ea05e3aSAlexander Block 5211867ab667Sjeff.liu static int btrfs_ioctl_get_fslabel(struct file *file, void __user *arg) 5212867ab667Sjeff.liu { 52136d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 5214a1b83ac5SAnand Jain size_t len; 5215867ab667Sjeff.liu int ret; 5216a1b83ac5SAnand Jain char label[BTRFS_LABEL_SIZE]; 5217a1b83ac5SAnand Jain 5218a1b83ac5SAnand Jain spin_lock(&root->fs_info->super_lock); 5219a1b83ac5SAnand Jain memcpy(label, root->fs_info->super_copy->label, BTRFS_LABEL_SIZE); 5220a1b83ac5SAnand Jain spin_unlock(&root->fs_info->super_lock); 5221a1b83ac5SAnand Jain 5222a1b83ac5SAnand Jain len = strnlen(label, BTRFS_LABEL_SIZE); 5223867ab667Sjeff.liu 5224867ab667Sjeff.liu if (len == BTRFS_LABEL_SIZE) { 5225efe120a0SFrank Holton btrfs_warn(root->fs_info, 5226efe120a0SFrank Holton "label is too long, return the first %zu bytes", --len); 5227867ab667Sjeff.liu } 5228867ab667Sjeff.liu 5229867ab667Sjeff.liu ret = copy_to_user(arg, label, len); 5230867ab667Sjeff.liu 5231867ab667Sjeff.liu return ret ? -EFAULT : 0; 5232867ab667Sjeff.liu } 5233867ab667Sjeff.liu 5234a8bfd4abSjeff.liu static int btrfs_ioctl_set_fslabel(struct file *file, void __user *arg) 5235a8bfd4abSjeff.liu { 52366d0379ecSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 5237a8bfd4abSjeff.liu struct btrfs_super_block *super_block = root->fs_info->super_copy; 5238a8bfd4abSjeff.liu struct btrfs_trans_handle *trans; 5239a8bfd4abSjeff.liu char label[BTRFS_LABEL_SIZE]; 5240a8bfd4abSjeff.liu int ret; 5241a8bfd4abSjeff.liu 5242a8bfd4abSjeff.liu if (!capable(CAP_SYS_ADMIN)) 5243a8bfd4abSjeff.liu return -EPERM; 5244a8bfd4abSjeff.liu 5245a8bfd4abSjeff.liu if (copy_from_user(label, arg, sizeof(label))) 5246a8bfd4abSjeff.liu return -EFAULT; 5247a8bfd4abSjeff.liu 5248a8bfd4abSjeff.liu if (strnlen(label, BTRFS_LABEL_SIZE) == BTRFS_LABEL_SIZE) { 5249efe120a0SFrank Holton btrfs_err(root->fs_info, "unable to set label with more than %d bytes", 5250a8bfd4abSjeff.liu BTRFS_LABEL_SIZE - 1); 5251a8bfd4abSjeff.liu return -EINVAL; 5252a8bfd4abSjeff.liu } 5253a8bfd4abSjeff.liu 5254a8bfd4abSjeff.liu ret = mnt_want_write_file(file); 5255a8bfd4abSjeff.liu if (ret) 5256a8bfd4abSjeff.liu return ret; 5257a8bfd4abSjeff.liu 5258a8bfd4abSjeff.liu trans = btrfs_start_transaction(root, 0); 5259a8bfd4abSjeff.liu if (IS_ERR(trans)) { 5260a8bfd4abSjeff.liu ret = PTR_ERR(trans); 5261a8bfd4abSjeff.liu goto out_unlock; 5262a8bfd4abSjeff.liu } 5263a8bfd4abSjeff.liu 5264a1b83ac5SAnand Jain spin_lock(&root->fs_info->super_lock); 5265a8bfd4abSjeff.liu strcpy(super_block->label, label); 5266a1b83ac5SAnand Jain spin_unlock(&root->fs_info->super_lock); 5267d0270acaSJeff Mahoney ret = btrfs_commit_transaction(trans, root); 5268a8bfd4abSjeff.liu 5269a8bfd4abSjeff.liu out_unlock: 5270a8bfd4abSjeff.liu mnt_drop_write_file(file); 5271a8bfd4abSjeff.liu return ret; 5272a8bfd4abSjeff.liu } 5273a8bfd4abSjeff.liu 52742eaa055fSJeff Mahoney #define INIT_FEATURE_FLAGS(suffix) \ 52752eaa055fSJeff Mahoney { .compat_flags = BTRFS_FEATURE_COMPAT_##suffix, \ 52762eaa055fSJeff Mahoney .compat_ro_flags = BTRFS_FEATURE_COMPAT_RO_##suffix, \ 52772eaa055fSJeff Mahoney .incompat_flags = BTRFS_FEATURE_INCOMPAT_##suffix } 52782eaa055fSJeff Mahoney 5279d5131b65SDavid Sterba int btrfs_ioctl_get_supported_features(void __user *arg) 52802eaa055fSJeff Mahoney { 52814d4ab6d6SDavid Sterba static const struct btrfs_ioctl_feature_flags features[3] = { 52822eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SUPP), 52832eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_SET), 52842eaa055fSJeff Mahoney INIT_FEATURE_FLAGS(SAFE_CLEAR) 52852eaa055fSJeff Mahoney }; 52862eaa055fSJeff Mahoney 52872eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 52882eaa055fSJeff Mahoney return -EFAULT; 52892eaa055fSJeff Mahoney 52902eaa055fSJeff Mahoney return 0; 52912eaa055fSJeff Mahoney } 52922eaa055fSJeff Mahoney 52932eaa055fSJeff Mahoney static int btrfs_ioctl_get_features(struct file *file, void __user *arg) 52942eaa055fSJeff Mahoney { 52952eaa055fSJeff Mahoney struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 52962eaa055fSJeff Mahoney struct btrfs_super_block *super_block = root->fs_info->super_copy; 52972eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags features; 52982eaa055fSJeff Mahoney 52992eaa055fSJeff Mahoney features.compat_flags = btrfs_super_compat_flags(super_block); 53002eaa055fSJeff Mahoney features.compat_ro_flags = btrfs_super_compat_ro_flags(super_block); 53012eaa055fSJeff Mahoney features.incompat_flags = btrfs_super_incompat_flags(super_block); 53022eaa055fSJeff Mahoney 53032eaa055fSJeff Mahoney if (copy_to_user(arg, &features, sizeof(features))) 53042eaa055fSJeff Mahoney return -EFAULT; 53052eaa055fSJeff Mahoney 53062eaa055fSJeff Mahoney return 0; 53072eaa055fSJeff Mahoney } 53082eaa055fSJeff Mahoney 53093b02a68aSJeff Mahoney static int check_feature_bits(struct btrfs_root *root, 53103b02a68aSJeff Mahoney enum btrfs_feature_set set, 53112eaa055fSJeff Mahoney u64 change_mask, u64 flags, u64 supported_flags, 53122eaa055fSJeff Mahoney u64 safe_set, u64 safe_clear) 53132eaa055fSJeff Mahoney { 53143b02a68aSJeff Mahoney const char *type = btrfs_feature_set_names[set]; 53153b02a68aSJeff Mahoney char *names; 53162eaa055fSJeff Mahoney u64 disallowed, unsupported; 53172eaa055fSJeff Mahoney u64 set_mask = flags & change_mask; 53182eaa055fSJeff Mahoney u64 clear_mask = ~flags & change_mask; 53192eaa055fSJeff Mahoney 53202eaa055fSJeff Mahoney unsupported = set_mask & ~supported_flags; 53212eaa055fSJeff Mahoney if (unsupported) { 53223b02a68aSJeff Mahoney names = btrfs_printable_features(set, unsupported); 53233b02a68aSJeff Mahoney if (names) { 53243b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 53253b02a68aSJeff Mahoney "this kernel does not support the %s feature bit%s", 53263b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 53273b02a68aSJeff Mahoney kfree(names); 53283b02a68aSJeff Mahoney } else 53292eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 53302eaa055fSJeff Mahoney "this kernel does not support %s bits 0x%llx", 53312eaa055fSJeff Mahoney type, unsupported); 53322eaa055fSJeff Mahoney return -EOPNOTSUPP; 53332eaa055fSJeff Mahoney } 53342eaa055fSJeff Mahoney 53352eaa055fSJeff Mahoney disallowed = set_mask & ~safe_set; 53362eaa055fSJeff Mahoney if (disallowed) { 53373b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 53383b02a68aSJeff Mahoney if (names) { 53393b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 53403b02a68aSJeff Mahoney "can't set the %s feature bit%s while mounted", 53413b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 53423b02a68aSJeff Mahoney kfree(names); 53433b02a68aSJeff Mahoney } else 53442eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 53452eaa055fSJeff Mahoney "can't set %s bits 0x%llx while mounted", 53462eaa055fSJeff Mahoney type, disallowed); 53472eaa055fSJeff Mahoney return -EPERM; 53482eaa055fSJeff Mahoney } 53492eaa055fSJeff Mahoney 53502eaa055fSJeff Mahoney disallowed = clear_mask & ~safe_clear; 53512eaa055fSJeff Mahoney if (disallowed) { 53523b02a68aSJeff Mahoney names = btrfs_printable_features(set, disallowed); 53533b02a68aSJeff Mahoney if (names) { 53543b02a68aSJeff Mahoney btrfs_warn(root->fs_info, 53553b02a68aSJeff Mahoney "can't clear the %s feature bit%s while mounted", 53563b02a68aSJeff Mahoney names, strchr(names, ',') ? "s" : ""); 53573b02a68aSJeff Mahoney kfree(names); 53583b02a68aSJeff Mahoney } else 53592eaa055fSJeff Mahoney btrfs_warn(root->fs_info, 53602eaa055fSJeff Mahoney "can't clear %s bits 0x%llx while mounted", 53612eaa055fSJeff Mahoney type, disallowed); 53622eaa055fSJeff Mahoney return -EPERM; 53632eaa055fSJeff Mahoney } 53642eaa055fSJeff Mahoney 53652eaa055fSJeff Mahoney return 0; 53662eaa055fSJeff Mahoney } 53672eaa055fSJeff Mahoney 53682eaa055fSJeff Mahoney #define check_feature(root, change_mask, flags, mask_base) \ 53693b02a68aSJeff Mahoney check_feature_bits(root, FEAT_##mask_base, change_mask, flags, \ 53702eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SUPP, \ 53712eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_SET, \ 53722eaa055fSJeff Mahoney BTRFS_FEATURE_ ## mask_base ## _SAFE_CLEAR) 53732eaa055fSJeff Mahoney 53742eaa055fSJeff Mahoney static int btrfs_ioctl_set_features(struct file *file, void __user *arg) 53752eaa055fSJeff Mahoney { 53762eaa055fSJeff Mahoney struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 53772eaa055fSJeff Mahoney struct btrfs_super_block *super_block = root->fs_info->super_copy; 53782eaa055fSJeff Mahoney struct btrfs_ioctl_feature_flags flags[2]; 53792eaa055fSJeff Mahoney struct btrfs_trans_handle *trans; 53802eaa055fSJeff Mahoney u64 newflags; 53812eaa055fSJeff Mahoney int ret; 53822eaa055fSJeff Mahoney 53832eaa055fSJeff Mahoney if (!capable(CAP_SYS_ADMIN)) 53842eaa055fSJeff Mahoney return -EPERM; 53852eaa055fSJeff Mahoney 53862eaa055fSJeff Mahoney if (copy_from_user(flags, arg, sizeof(flags))) 53872eaa055fSJeff Mahoney return -EFAULT; 53882eaa055fSJeff Mahoney 53892eaa055fSJeff Mahoney /* Nothing to do */ 53902eaa055fSJeff Mahoney if (!flags[0].compat_flags && !flags[0].compat_ro_flags && 53912eaa055fSJeff Mahoney !flags[0].incompat_flags) 53922eaa055fSJeff Mahoney return 0; 53932eaa055fSJeff Mahoney 53942eaa055fSJeff Mahoney ret = check_feature(root, flags[0].compat_flags, 53952eaa055fSJeff Mahoney flags[1].compat_flags, COMPAT); 53962eaa055fSJeff Mahoney if (ret) 53972eaa055fSJeff Mahoney return ret; 53982eaa055fSJeff Mahoney 53992eaa055fSJeff Mahoney ret = check_feature(root, flags[0].compat_ro_flags, 54002eaa055fSJeff Mahoney flags[1].compat_ro_flags, COMPAT_RO); 54012eaa055fSJeff Mahoney if (ret) 54022eaa055fSJeff Mahoney return ret; 54032eaa055fSJeff Mahoney 54042eaa055fSJeff Mahoney ret = check_feature(root, flags[0].incompat_flags, 54052eaa055fSJeff Mahoney flags[1].incompat_flags, INCOMPAT); 54062eaa055fSJeff Mahoney if (ret) 54072eaa055fSJeff Mahoney return ret; 54082eaa055fSJeff Mahoney 54097ab19625SDavid Sterba ret = mnt_want_write_file(file); 54107ab19625SDavid Sterba if (ret) 54117ab19625SDavid Sterba return ret; 54127ab19625SDavid Sterba 54138051aa1aSDavid Sterba trans = btrfs_start_transaction(root, 0); 54147ab19625SDavid Sterba if (IS_ERR(trans)) { 54157ab19625SDavid Sterba ret = PTR_ERR(trans); 54167ab19625SDavid Sterba goto out_drop_write; 54177ab19625SDavid Sterba } 54182eaa055fSJeff Mahoney 54192eaa055fSJeff Mahoney spin_lock(&root->fs_info->super_lock); 54202eaa055fSJeff Mahoney newflags = btrfs_super_compat_flags(super_block); 54212eaa055fSJeff Mahoney newflags |= flags[0].compat_flags & flags[1].compat_flags; 54222eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_flags & ~flags[1].compat_flags); 54232eaa055fSJeff Mahoney btrfs_set_super_compat_flags(super_block, newflags); 54242eaa055fSJeff Mahoney 54252eaa055fSJeff Mahoney newflags = btrfs_super_compat_ro_flags(super_block); 54262eaa055fSJeff Mahoney newflags |= flags[0].compat_ro_flags & flags[1].compat_ro_flags; 54272eaa055fSJeff Mahoney newflags &= ~(flags[0].compat_ro_flags & ~flags[1].compat_ro_flags); 54282eaa055fSJeff Mahoney btrfs_set_super_compat_ro_flags(super_block, newflags); 54292eaa055fSJeff Mahoney 54302eaa055fSJeff Mahoney newflags = btrfs_super_incompat_flags(super_block); 54312eaa055fSJeff Mahoney newflags |= flags[0].incompat_flags & flags[1].incompat_flags; 54322eaa055fSJeff Mahoney newflags &= ~(flags[0].incompat_flags & ~flags[1].incompat_flags); 54332eaa055fSJeff Mahoney btrfs_set_super_incompat_flags(super_block, newflags); 54342eaa055fSJeff Mahoney spin_unlock(&root->fs_info->super_lock); 54352eaa055fSJeff Mahoney 54367ab19625SDavid Sterba ret = btrfs_commit_transaction(trans, root); 54377ab19625SDavid Sterba out_drop_write: 54387ab19625SDavid Sterba mnt_drop_write_file(file); 54397ab19625SDavid Sterba 54407ab19625SDavid Sterba return ret; 54412eaa055fSJeff Mahoney } 54422eaa055fSJeff Mahoney 5443f46b5a66SChristoph Hellwig long btrfs_ioctl(struct file *file, unsigned int 5444f46b5a66SChristoph Hellwig cmd, unsigned long arg) 5445f46b5a66SChristoph Hellwig { 5446496ad9aaSAl Viro struct btrfs_root *root = BTRFS_I(file_inode(file))->root; 54474bcabaa3SChristoph Hellwig void __user *argp = (void __user *)arg; 5448f46b5a66SChristoph Hellwig 5449f46b5a66SChristoph Hellwig switch (cmd) { 54506cbff00fSChristoph Hellwig case FS_IOC_GETFLAGS: 54516cbff00fSChristoph Hellwig return btrfs_ioctl_getflags(file, argp); 54526cbff00fSChristoph Hellwig case FS_IOC_SETFLAGS: 54536cbff00fSChristoph Hellwig return btrfs_ioctl_setflags(file, argp); 54546cbff00fSChristoph Hellwig case FS_IOC_GETVERSION: 54556cbff00fSChristoph Hellwig return btrfs_ioctl_getversion(file, argp); 5456f7039b1dSLi Dongyang case FITRIM: 5457f7039b1dSLi Dongyang return btrfs_ioctl_fitrim(file, argp); 5458f46b5a66SChristoph Hellwig case BTRFS_IOC_SNAP_CREATE: 5459fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 0); 5460fdfb1e4fSLi Zefan case BTRFS_IOC_SNAP_CREATE_V2: 5461fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create_v2(file, argp, 0); 54623de4586cSChris Mason case BTRFS_IOC_SUBVOL_CREATE: 5463fa0d2b9bSLi Zefan return btrfs_ioctl_snap_create(file, argp, 1); 54646f72c7e2SArne Jansen case BTRFS_IOC_SUBVOL_CREATE_V2: 54656f72c7e2SArne Jansen return btrfs_ioctl_snap_create_v2(file, argp, 1); 546676dda93cSYan, Zheng case BTRFS_IOC_SNAP_DESTROY: 546776dda93cSYan, Zheng return btrfs_ioctl_snap_destroy(file, argp); 54680caa102dSLi Zefan case BTRFS_IOC_SUBVOL_GETFLAGS: 54690caa102dSLi Zefan return btrfs_ioctl_subvol_getflags(file, argp); 54700caa102dSLi Zefan case BTRFS_IOC_SUBVOL_SETFLAGS: 54710caa102dSLi Zefan return btrfs_ioctl_subvol_setflags(file, argp); 54726ef5ed0dSJosef Bacik case BTRFS_IOC_DEFAULT_SUBVOL: 54736ef5ed0dSJosef Bacik return btrfs_ioctl_default_subvol(file, argp); 5474f46b5a66SChristoph Hellwig case BTRFS_IOC_DEFRAG: 54751e701a32SChris Mason return btrfs_ioctl_defrag(file, NULL); 54761e701a32SChris Mason case BTRFS_IOC_DEFRAG_RANGE: 54771e701a32SChris Mason return btrfs_ioctl_defrag(file, argp); 5478f46b5a66SChristoph Hellwig case BTRFS_IOC_RESIZE: 5479198605a8SMiao Xie return btrfs_ioctl_resize(file, argp); 5480f46b5a66SChristoph Hellwig case BTRFS_IOC_ADD_DEV: 54814bcabaa3SChristoph Hellwig return btrfs_ioctl_add_dev(root, argp); 5482f46b5a66SChristoph Hellwig case BTRFS_IOC_RM_DEV: 5483da24927bSMiao Xie return btrfs_ioctl_rm_dev(file, argp); 5484475f6387SJan Schmidt case BTRFS_IOC_FS_INFO: 5485475f6387SJan Schmidt return btrfs_ioctl_fs_info(root, argp); 5486475f6387SJan Schmidt case BTRFS_IOC_DEV_INFO: 5487475f6387SJan Schmidt return btrfs_ioctl_dev_info(root, argp); 5488f46b5a66SChristoph Hellwig case BTRFS_IOC_BALANCE: 54899ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, NULL); 5490f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_START: 5491f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_start(file); 5492f46b5a66SChristoph Hellwig case BTRFS_IOC_TRANS_END: 5493f46b5a66SChristoph Hellwig return btrfs_ioctl_trans_end(file); 5494ac8e9819SChris Mason case BTRFS_IOC_TREE_SEARCH: 5495ac8e9819SChris Mason return btrfs_ioctl_tree_search(file, argp); 5496cc68a8a5SGerhard Heift case BTRFS_IOC_TREE_SEARCH_V2: 5497cc68a8a5SGerhard Heift return btrfs_ioctl_tree_search_v2(file, argp); 5498ac8e9819SChris Mason case BTRFS_IOC_INO_LOOKUP: 5499ac8e9819SChris Mason return btrfs_ioctl_ino_lookup(file, argp); 5500d7728c96SJan Schmidt case BTRFS_IOC_INO_PATHS: 5501d7728c96SJan Schmidt return btrfs_ioctl_ino_to_path(root, argp); 5502d7728c96SJan Schmidt case BTRFS_IOC_LOGICAL_INO: 5503d7728c96SJan Schmidt return btrfs_ioctl_logical_to_ino(root, argp); 55041406e432SJosef Bacik case BTRFS_IOC_SPACE_INFO: 55051406e432SJosef Bacik return btrfs_ioctl_space_info(root, argp); 55069b199859SFilipe David Borba Manana case BTRFS_IOC_SYNC: { 55079b199859SFilipe David Borba Manana int ret; 55089b199859SFilipe David Borba Manana 55096c255e67SMiao Xie ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1); 55109b199859SFilipe David Borba Manana if (ret) 55119b199859SFilipe David Borba Manana return ret; 5512ddb52f4fSAl Viro ret = btrfs_sync_fs(file_inode(file)->i_sb, 1); 55132fad4e83SDavid Sterba /* 55142fad4e83SDavid Sterba * The transaction thread may want to do more work, 55152fad4e83SDavid Sterba * namely it pokes the cleaner ktread that will start 55162fad4e83SDavid Sterba * processing uncleaned subvols. 55172fad4e83SDavid Sterba */ 55182fad4e83SDavid Sterba wake_up_process(root->fs_info->transaction_kthread); 55199b199859SFilipe David Borba Manana return ret; 55209b199859SFilipe David Borba Manana } 552146204592SSage Weil case BTRFS_IOC_START_SYNC: 55229a8c28beSMiao Xie return btrfs_ioctl_start_sync(root, argp); 552346204592SSage Weil case BTRFS_IOC_WAIT_SYNC: 55249a8c28beSMiao Xie return btrfs_ioctl_wait_sync(root, argp); 5525475f6387SJan Schmidt case BTRFS_IOC_SCRUB: 5526b8e95489SMiao Xie return btrfs_ioctl_scrub(file, argp); 5527475f6387SJan Schmidt case BTRFS_IOC_SCRUB_CANCEL: 5528475f6387SJan Schmidt return btrfs_ioctl_scrub_cancel(root, argp); 5529475f6387SJan Schmidt case BTRFS_IOC_SCRUB_PROGRESS: 5530475f6387SJan Schmidt return btrfs_ioctl_scrub_progress(root, argp); 5531c9e9f97bSIlya Dryomov case BTRFS_IOC_BALANCE_V2: 55329ba1f6e4SLiu Bo return btrfs_ioctl_balance(file, argp); 5533837d5b6eSIlya Dryomov case BTRFS_IOC_BALANCE_CTL: 5534837d5b6eSIlya Dryomov return btrfs_ioctl_balance_ctl(root, arg); 553519a39dceSIlya Dryomov case BTRFS_IOC_BALANCE_PROGRESS: 553619a39dceSIlya Dryomov return btrfs_ioctl_balance_progress(root, argp); 55378ea05e3aSAlexander Block case BTRFS_IOC_SET_RECEIVED_SUBVOL: 55388ea05e3aSAlexander Block return btrfs_ioctl_set_received_subvol(file, argp); 5539abccd00fSHugo Mills #ifdef CONFIG_64BIT 5540abccd00fSHugo Mills case BTRFS_IOC_SET_RECEIVED_SUBVOL_32: 5541abccd00fSHugo Mills return btrfs_ioctl_set_received_subvol_32(file, argp); 5542abccd00fSHugo Mills #endif 554331db9f7cSAlexander Block case BTRFS_IOC_SEND: 554431db9f7cSAlexander Block return btrfs_ioctl_send(file, argp); 5545c11d2c23SStefan Behrens case BTRFS_IOC_GET_DEV_STATS: 5546b27f7c0cSDavid Sterba return btrfs_ioctl_get_dev_stats(root, argp); 55475d13a37bSArne Jansen case BTRFS_IOC_QUOTA_CTL: 5548905b0ddaSMiao Xie return btrfs_ioctl_quota_ctl(file, argp); 55495d13a37bSArne Jansen case BTRFS_IOC_QGROUP_ASSIGN: 5550905b0ddaSMiao Xie return btrfs_ioctl_qgroup_assign(file, argp); 55515d13a37bSArne Jansen case BTRFS_IOC_QGROUP_CREATE: 5552905b0ddaSMiao Xie return btrfs_ioctl_qgroup_create(file, argp); 55535d13a37bSArne Jansen case BTRFS_IOC_QGROUP_LIMIT: 5554905b0ddaSMiao Xie return btrfs_ioctl_qgroup_limit(file, argp); 55552f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN: 55562f232036SJan Schmidt return btrfs_ioctl_quota_rescan(file, argp); 55572f232036SJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_STATUS: 55582f232036SJan Schmidt return btrfs_ioctl_quota_rescan_status(file, argp); 555957254b6eSJan Schmidt case BTRFS_IOC_QUOTA_RESCAN_WAIT: 556057254b6eSJan Schmidt return btrfs_ioctl_quota_rescan_wait(file, argp); 55613f6bcfbdSStefan Behrens case BTRFS_IOC_DEV_REPLACE: 55623f6bcfbdSStefan Behrens return btrfs_ioctl_dev_replace(root, argp); 5563867ab667Sjeff.liu case BTRFS_IOC_GET_FSLABEL: 5564867ab667Sjeff.liu return btrfs_ioctl_get_fslabel(file, argp); 5565a8bfd4abSjeff.liu case BTRFS_IOC_SET_FSLABEL: 5566a8bfd4abSjeff.liu return btrfs_ioctl_set_fslabel(file, argp); 55672eaa055fSJeff Mahoney case BTRFS_IOC_GET_SUPPORTED_FEATURES: 5568d5131b65SDavid Sterba return btrfs_ioctl_get_supported_features(argp); 55692eaa055fSJeff Mahoney case BTRFS_IOC_GET_FEATURES: 55702eaa055fSJeff Mahoney return btrfs_ioctl_get_features(file, argp); 55712eaa055fSJeff Mahoney case BTRFS_IOC_SET_FEATURES: 55722eaa055fSJeff Mahoney return btrfs_ioctl_set_features(file, argp); 5573f46b5a66SChristoph Hellwig } 5574f46b5a66SChristoph Hellwig 5575f46b5a66SChristoph Hellwig return -ENOTTY; 5576f46b5a66SChristoph Hellwig } 55774c63c245SLuke Dashjr 55784c63c245SLuke Dashjr #ifdef CONFIG_COMPAT 55794c63c245SLuke Dashjr long btrfs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg) 55804c63c245SLuke Dashjr { 55814c63c245SLuke Dashjr switch (cmd) { 55824c63c245SLuke Dashjr case FS_IOC32_GETFLAGS: 55834c63c245SLuke Dashjr cmd = FS_IOC_GETFLAGS; 55844c63c245SLuke Dashjr break; 55854c63c245SLuke Dashjr case FS_IOC32_SETFLAGS: 55864c63c245SLuke Dashjr cmd = FS_IOC_SETFLAGS; 55874c63c245SLuke Dashjr break; 55884c63c245SLuke Dashjr case FS_IOC32_GETVERSION: 55894c63c245SLuke Dashjr cmd = FS_IOC_GETVERSION; 55904c63c245SLuke Dashjr break; 55914c63c245SLuke Dashjr default: 55924c63c245SLuke Dashjr return -ENOIOCTLCMD; 55934c63c245SLuke Dashjr } 55944c63c245SLuke Dashjr 55954c63c245SLuke Dashjr return btrfs_ioctl(file, cmd, (unsigned long) compat_ptr(arg)); 55964c63c245SLuke Dashjr } 55974c63c245SLuke Dashjr #endif 5598