Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Olaf Weber | 3e57ecf | 2006-06-09 14:48:12 +1000 | [diff] [blame] | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 3 | * All Rights Reserved. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * published by the Free Software Foundation. |
| 8 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
Nathan Scott | 7b71876 | 2005-11-02 14:58:39 +1100 | [diff] [blame] | 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | */ |
Robert P. J. Day | 40ebd81 | 2007-11-23 16:30:51 +1100 | [diff] [blame] | 18 | #include <linux/log2.h> |
| 19 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include "xfs.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 21 | #include "xfs_fs.h" |
Dave Chinner | 6ca1c90 | 2013-08-12 20:49:26 +1000 | [diff] [blame] | 22 | #include "xfs_format.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include "xfs_log.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 24 | #include "xfs_inum.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include "xfs_trans.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 26 | #include "xfs_trans_space.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include "xfs_trans_priv.h" |
| 28 | #include "xfs_sb.h" |
| 29 | #include "xfs_ag.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include "xfs_mount.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 31 | #include "xfs_da_btree.h" |
| 32 | #include "xfs_dir2_format.h" |
| 33 | #include "xfs_dir2.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include "xfs_bmap_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 35 | #include "xfs_alloc_btree.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include "xfs_ialloc_btree.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 37 | #include "xfs_attr_sf.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 38 | #include "xfs_attr.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 39 | #include "xfs_dinode.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | #include "xfs_inode.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #include "xfs_buf_item.h" |
Nathan Scott | a844f45 | 2005-11-02 14:38:42 +1100 | [diff] [blame] | 42 | #include "xfs_inode_item.h" |
| 43 | #include "xfs_btree.h" |
| 44 | #include "xfs_alloc.h" |
| 45 | #include "xfs_ialloc.h" |
| 46 | #include "xfs_bmap.h" |
Dave Chinner | 6898811 | 2013-08-12 20:49:42 +1000 | [diff] [blame] | 47 | #include "xfs_bmap_util.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | #include "xfs_error.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | #include "xfs_quota.h" |
David Chinner | 2a82b8b | 2007-07-11 11:09:12 +1000 | [diff] [blame] | 50 | #include "xfs_filestream.h" |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 51 | #include "xfs_cksum.h" |
Christoph Hellwig | 0b1b213 | 2009-12-14 23:14:59 +0000 | [diff] [blame] | 52 | #include "xfs_trace.h" |
Dave Chinner | 33479e0 | 2012-10-08 21:56:11 +1100 | [diff] [blame] | 53 | #include "xfs_icache.h" |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 54 | #include "xfs_symlink.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | kmem_zone_t *xfs_inode_zone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
| 58 | /* |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 59 | * Used in xfs_itruncate_extents(). This is the maximum number of extents |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | * freed from a file in a single transaction. |
| 61 | */ |
| 62 | #define XFS_ITRUNC_MAX_EXTENTS 2 |
| 63 | |
| 64 | STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | |
Dave Chinner | 2a0ec1d | 2012-04-23 15:59:02 +1000 | [diff] [blame] | 66 | /* |
| 67 | * helper function to extract extent size hint from inode |
| 68 | */ |
| 69 | xfs_extlen_t |
| 70 | xfs_get_extsz_hint( |
| 71 | struct xfs_inode *ip) |
| 72 | { |
| 73 | if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) |
| 74 | return ip->i_d.di_extsize; |
| 75 | if (XFS_IS_REALTIME_INODE(ip)) |
| 76 | return ip->i_mount->m_sb.sb_rextsize; |
| 77 | return 0; |
| 78 | } |
| 79 | |
Dave Chinner | fa96aca | 2012-10-08 21:56:10 +1100 | [diff] [blame] | 80 | /* |
| 81 | * This is a wrapper routine around the xfs_ilock() routine used to centralize |
| 82 | * some grungy code. It is used in places that wish to lock the inode solely |
| 83 | * for reading the extents. The reason these places can't just call |
| 84 | * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the |
| 85 | * extents from disk for a file in b-tree format. If the inode is in b-tree |
| 86 | * format, then we need to lock the inode exclusively until the extents are read |
| 87 | * in. Locking it exclusively all the time would limit our parallelism |
| 88 | * unnecessarily, though. What we do instead is check to see if the extents |
| 89 | * have been read in yet, and only lock the inode exclusively if they have not. |
| 90 | * |
| 91 | * The function returns a value which should be given to the corresponding |
| 92 | * xfs_iunlock_map_shared(). This value is the mode in which the lock was |
| 93 | * actually taken. |
| 94 | */ |
| 95 | uint |
| 96 | xfs_ilock_map_shared( |
| 97 | xfs_inode_t *ip) |
| 98 | { |
| 99 | uint lock_mode; |
| 100 | |
| 101 | if ((ip->i_d.di_format == XFS_DINODE_FMT_BTREE) && |
| 102 | ((ip->i_df.if_flags & XFS_IFEXTENTS) == 0)) { |
| 103 | lock_mode = XFS_ILOCK_EXCL; |
| 104 | } else { |
| 105 | lock_mode = XFS_ILOCK_SHARED; |
| 106 | } |
| 107 | |
| 108 | xfs_ilock(ip, lock_mode); |
| 109 | |
| 110 | return lock_mode; |
| 111 | } |
| 112 | |
| 113 | /* |
| 114 | * This is simply the unlock routine to go with xfs_ilock_map_shared(). |
| 115 | * All it does is call xfs_iunlock() with the given lock_mode. |
| 116 | */ |
| 117 | void |
| 118 | xfs_iunlock_map_shared( |
| 119 | xfs_inode_t *ip, |
| 120 | unsigned int lock_mode) |
| 121 | { |
| 122 | xfs_iunlock(ip, lock_mode); |
| 123 | } |
| 124 | |
| 125 | /* |
| 126 | * The xfs inode contains 2 locks: a multi-reader lock called the |
| 127 | * i_iolock and a multi-reader lock called the i_lock. This routine |
| 128 | * allows either or both of the locks to be obtained. |
| 129 | * |
| 130 | * The 2 locks should always be ordered so that the IO lock is |
| 131 | * obtained first in order to prevent deadlock. |
| 132 | * |
| 133 | * ip -- the inode being locked |
| 134 | * lock_flags -- this parameter indicates the inode's locks |
| 135 | * to be locked. It can be: |
| 136 | * XFS_IOLOCK_SHARED, |
| 137 | * XFS_IOLOCK_EXCL, |
| 138 | * XFS_ILOCK_SHARED, |
| 139 | * XFS_ILOCK_EXCL, |
| 140 | * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, |
| 141 | * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, |
| 142 | * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, |
| 143 | * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL |
| 144 | */ |
| 145 | void |
| 146 | xfs_ilock( |
| 147 | xfs_inode_t *ip, |
| 148 | uint lock_flags) |
| 149 | { |
| 150 | trace_xfs_ilock(ip, lock_flags, _RET_IP_); |
| 151 | |
| 152 | /* |
| 153 | * You can't set both SHARED and EXCL for the same lock, |
| 154 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 155 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 156 | */ |
| 157 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 158 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 159 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 160 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 161 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
| 162 | |
| 163 | if (lock_flags & XFS_IOLOCK_EXCL) |
| 164 | mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
| 165 | else if (lock_flags & XFS_IOLOCK_SHARED) |
| 166 | mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); |
| 167 | |
| 168 | if (lock_flags & XFS_ILOCK_EXCL) |
| 169 | mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
| 170 | else if (lock_flags & XFS_ILOCK_SHARED) |
| 171 | mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); |
| 172 | } |
| 173 | |
| 174 | /* |
| 175 | * This is just like xfs_ilock(), except that the caller |
| 176 | * is guaranteed not to sleep. It returns 1 if it gets |
| 177 | * the requested locks and 0 otherwise. If the IO lock is |
| 178 | * obtained but the inode lock cannot be, then the IO lock |
| 179 | * is dropped before returning. |
| 180 | * |
| 181 | * ip -- the inode being locked |
| 182 | * lock_flags -- this parameter indicates the inode's locks to be |
| 183 | * to be locked. See the comment for xfs_ilock() for a list |
| 184 | * of valid values. |
| 185 | */ |
| 186 | int |
| 187 | xfs_ilock_nowait( |
| 188 | xfs_inode_t *ip, |
| 189 | uint lock_flags) |
| 190 | { |
| 191 | trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); |
| 192 | |
| 193 | /* |
| 194 | * You can't set both SHARED and EXCL for the same lock, |
| 195 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 196 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 197 | */ |
| 198 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 199 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 200 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 201 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 202 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
| 203 | |
| 204 | if (lock_flags & XFS_IOLOCK_EXCL) { |
| 205 | if (!mrtryupdate(&ip->i_iolock)) |
| 206 | goto out; |
| 207 | } else if (lock_flags & XFS_IOLOCK_SHARED) { |
| 208 | if (!mrtryaccess(&ip->i_iolock)) |
| 209 | goto out; |
| 210 | } |
| 211 | if (lock_flags & XFS_ILOCK_EXCL) { |
| 212 | if (!mrtryupdate(&ip->i_lock)) |
| 213 | goto out_undo_iolock; |
| 214 | } else if (lock_flags & XFS_ILOCK_SHARED) { |
| 215 | if (!mrtryaccess(&ip->i_lock)) |
| 216 | goto out_undo_iolock; |
| 217 | } |
| 218 | return 1; |
| 219 | |
| 220 | out_undo_iolock: |
| 221 | if (lock_flags & XFS_IOLOCK_EXCL) |
| 222 | mrunlock_excl(&ip->i_iolock); |
| 223 | else if (lock_flags & XFS_IOLOCK_SHARED) |
| 224 | mrunlock_shared(&ip->i_iolock); |
| 225 | out: |
| 226 | return 0; |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * xfs_iunlock() is used to drop the inode locks acquired with |
| 231 | * xfs_ilock() and xfs_ilock_nowait(). The caller must pass |
| 232 | * in the flags given to xfs_ilock() or xfs_ilock_nowait() so |
| 233 | * that we know which locks to drop. |
| 234 | * |
| 235 | * ip -- the inode being unlocked |
| 236 | * lock_flags -- this parameter indicates the inode's locks to be |
| 237 | * to be unlocked. See the comment for xfs_ilock() for a list |
| 238 | * of valid values for this parameter. |
| 239 | * |
| 240 | */ |
| 241 | void |
| 242 | xfs_iunlock( |
| 243 | xfs_inode_t *ip, |
| 244 | uint lock_flags) |
| 245 | { |
| 246 | /* |
| 247 | * You can't set both SHARED and EXCL for the same lock, |
| 248 | * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, |
| 249 | * and XFS_ILOCK_EXCL are valid values to set in lock_flags. |
| 250 | */ |
| 251 | ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != |
| 252 | (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); |
| 253 | ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != |
| 254 | (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
| 255 | ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); |
| 256 | ASSERT(lock_flags != 0); |
| 257 | |
| 258 | if (lock_flags & XFS_IOLOCK_EXCL) |
| 259 | mrunlock_excl(&ip->i_iolock); |
| 260 | else if (lock_flags & XFS_IOLOCK_SHARED) |
| 261 | mrunlock_shared(&ip->i_iolock); |
| 262 | |
| 263 | if (lock_flags & XFS_ILOCK_EXCL) |
| 264 | mrunlock_excl(&ip->i_lock); |
| 265 | else if (lock_flags & XFS_ILOCK_SHARED) |
| 266 | mrunlock_shared(&ip->i_lock); |
| 267 | |
| 268 | trace_xfs_iunlock(ip, lock_flags, _RET_IP_); |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * give up write locks. the i/o lock cannot be held nested |
| 273 | * if it is being demoted. |
| 274 | */ |
| 275 | void |
| 276 | xfs_ilock_demote( |
| 277 | xfs_inode_t *ip, |
| 278 | uint lock_flags) |
| 279 | { |
| 280 | ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); |
| 281 | ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); |
| 282 | |
| 283 | if (lock_flags & XFS_ILOCK_EXCL) |
| 284 | mrdemote(&ip->i_lock); |
| 285 | if (lock_flags & XFS_IOLOCK_EXCL) |
| 286 | mrdemote(&ip->i_iolock); |
| 287 | |
| 288 | trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); |
| 289 | } |
| 290 | |
Dave Chinner | 742ae1e | 2013-04-30 21:39:34 +1000 | [diff] [blame] | 291 | #if defined(DEBUG) || defined(XFS_WARN) |
Dave Chinner | fa96aca | 2012-10-08 21:56:10 +1100 | [diff] [blame] | 292 | int |
| 293 | xfs_isilocked( |
| 294 | xfs_inode_t *ip, |
| 295 | uint lock_flags) |
| 296 | { |
| 297 | if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { |
| 298 | if (!(lock_flags & XFS_ILOCK_SHARED)) |
| 299 | return !!ip->i_lock.mr_writer; |
| 300 | return rwsem_is_locked(&ip->i_lock.mr_lock); |
| 301 | } |
| 302 | |
| 303 | if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { |
| 304 | if (!(lock_flags & XFS_IOLOCK_SHARED)) |
| 305 | return !!ip->i_iolock.mr_writer; |
| 306 | return rwsem_is_locked(&ip->i_iolock.mr_lock); |
| 307 | } |
| 308 | |
| 309 | ASSERT(0); |
| 310 | return 0; |
| 311 | } |
| 312 | #endif |
| 313 | |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 314 | #ifdef DEBUG |
| 315 | int xfs_locked_n; |
| 316 | int xfs_small_retries; |
| 317 | int xfs_middle_retries; |
| 318 | int xfs_lots_retries; |
| 319 | int xfs_lock_delays; |
| 320 | #endif |
| 321 | |
| 322 | /* |
| 323 | * Bump the subclass so xfs_lock_inodes() acquires each lock with |
| 324 | * a different value |
| 325 | */ |
| 326 | static inline int |
| 327 | xfs_lock_inumorder(int lock_mode, int subclass) |
| 328 | { |
| 329 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
| 330 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; |
| 331 | if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) |
| 332 | lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; |
| 333 | |
| 334 | return lock_mode; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * The following routine will lock n inodes in exclusive mode. |
| 339 | * We assume the caller calls us with the inodes in i_ino order. |
| 340 | * |
| 341 | * We need to detect deadlock where an inode that we lock |
| 342 | * is in the AIL and we start waiting for another inode that is locked |
| 343 | * by a thread in a long running transaction (such as truncate). This can |
| 344 | * result in deadlock since the long running trans might need to wait |
| 345 | * for the inode we just locked in order to push the tail and free space |
| 346 | * in the log. |
| 347 | */ |
| 348 | void |
| 349 | xfs_lock_inodes( |
| 350 | xfs_inode_t **ips, |
| 351 | int inodes, |
| 352 | uint lock_mode) |
| 353 | { |
| 354 | int attempts = 0, i, j, try_lock; |
| 355 | xfs_log_item_t *lp; |
| 356 | |
| 357 | ASSERT(ips && (inodes >= 2)); /* we need at least two */ |
| 358 | |
| 359 | try_lock = 0; |
| 360 | i = 0; |
| 361 | |
| 362 | again: |
| 363 | for (; i < inodes; i++) { |
| 364 | ASSERT(ips[i]); |
| 365 | |
| 366 | if (i && (ips[i] == ips[i-1])) /* Already locked */ |
| 367 | continue; |
| 368 | |
| 369 | /* |
| 370 | * If try_lock is not set yet, make sure all locked inodes |
| 371 | * are not in the AIL. |
| 372 | * If any are, set try_lock to be used later. |
| 373 | */ |
| 374 | |
| 375 | if (!try_lock) { |
| 376 | for (j = (i - 1); j >= 0 && !try_lock; j--) { |
| 377 | lp = (xfs_log_item_t *)ips[j]->i_itemp; |
| 378 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { |
| 379 | try_lock++; |
| 380 | } |
| 381 | } |
| 382 | } |
| 383 | |
| 384 | /* |
| 385 | * If any of the previous locks we have locked is in the AIL, |
| 386 | * we must TRY to get the second and subsequent locks. If |
| 387 | * we can't get any, we must release all we have |
| 388 | * and try again. |
| 389 | */ |
| 390 | |
| 391 | if (try_lock) { |
| 392 | /* try_lock must be 0 if i is 0. */ |
| 393 | /* |
| 394 | * try_lock means we have an inode locked |
| 395 | * that is in the AIL. |
| 396 | */ |
| 397 | ASSERT(i != 0); |
| 398 | if (!xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) { |
| 399 | attempts++; |
| 400 | |
| 401 | /* |
| 402 | * Unlock all previous guys and try again. |
| 403 | * xfs_iunlock will try to push the tail |
| 404 | * if the inode is in the AIL. |
| 405 | */ |
| 406 | |
| 407 | for(j = i - 1; j >= 0; j--) { |
| 408 | |
| 409 | /* |
| 410 | * Check to see if we've already |
| 411 | * unlocked this one. |
| 412 | * Not the first one going back, |
| 413 | * and the inode ptr is the same. |
| 414 | */ |
| 415 | if ((j != (i - 1)) && ips[j] == |
| 416 | ips[j+1]) |
| 417 | continue; |
| 418 | |
| 419 | xfs_iunlock(ips[j], lock_mode); |
| 420 | } |
| 421 | |
| 422 | if ((attempts % 5) == 0) { |
| 423 | delay(1); /* Don't just spin the CPU */ |
| 424 | #ifdef DEBUG |
| 425 | xfs_lock_delays++; |
| 426 | #endif |
| 427 | } |
| 428 | i = 0; |
| 429 | try_lock = 0; |
| 430 | goto again; |
| 431 | } |
| 432 | } else { |
| 433 | xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); |
| 434 | } |
| 435 | } |
| 436 | |
| 437 | #ifdef DEBUG |
| 438 | if (attempts) { |
| 439 | if (attempts < 5) xfs_small_retries++; |
| 440 | else if (attempts < 100) xfs_middle_retries++; |
| 441 | else xfs_lots_retries++; |
| 442 | } else { |
| 443 | xfs_locked_n++; |
| 444 | } |
| 445 | #endif |
| 446 | } |
| 447 | |
| 448 | /* |
| 449 | * xfs_lock_two_inodes() can only be used to lock one type of lock |
| 450 | * at a time - the iolock or the ilock, but not both at once. If |
| 451 | * we lock both at once, lockdep will report false positives saying |
| 452 | * we have violated locking orders. |
| 453 | */ |
| 454 | void |
| 455 | xfs_lock_two_inodes( |
| 456 | xfs_inode_t *ip0, |
| 457 | xfs_inode_t *ip1, |
| 458 | uint lock_mode) |
| 459 | { |
| 460 | xfs_inode_t *temp; |
| 461 | int attempts = 0; |
| 462 | xfs_log_item_t *lp; |
| 463 | |
| 464 | if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) |
| 465 | ASSERT((lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) == 0); |
| 466 | ASSERT(ip0->i_ino != ip1->i_ino); |
| 467 | |
| 468 | if (ip0->i_ino > ip1->i_ino) { |
| 469 | temp = ip0; |
| 470 | ip0 = ip1; |
| 471 | ip1 = temp; |
| 472 | } |
| 473 | |
| 474 | again: |
| 475 | xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); |
| 476 | |
| 477 | /* |
| 478 | * If the first lock we have locked is in the AIL, we must TRY to get |
| 479 | * the second lock. If we can't get it, we must release the first one |
| 480 | * and try again. |
| 481 | */ |
| 482 | lp = (xfs_log_item_t *)ip0->i_itemp; |
| 483 | if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { |
| 484 | if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { |
| 485 | xfs_iunlock(ip0, lock_mode); |
| 486 | if ((++attempts % 5) == 0) |
| 487 | delay(1); /* Don't just spin the CPU */ |
| 488 | goto again; |
| 489 | } |
| 490 | } else { |
| 491 | xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | |
Dave Chinner | fa96aca | 2012-10-08 21:56:10 +1100 | [diff] [blame] | 496 | void |
| 497 | __xfs_iflock( |
| 498 | struct xfs_inode *ip) |
| 499 | { |
| 500 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); |
| 501 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); |
| 502 | |
| 503 | do { |
| 504 | prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
| 505 | if (xfs_isiflocked(ip)) |
| 506 | io_schedule(); |
| 507 | } while (!xfs_iflock_nowait(ip)); |
| 508 | |
| 509 | finish_wait(wq, &wait.wait); |
| 510 | } |
| 511 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | STATIC uint |
| 513 | _xfs_dic2xflags( |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | __uint16_t di_flags) |
| 515 | { |
| 516 | uint flags = 0; |
| 517 | |
| 518 | if (di_flags & XFS_DIFLAG_ANY) { |
| 519 | if (di_flags & XFS_DIFLAG_REALTIME) |
| 520 | flags |= XFS_XFLAG_REALTIME; |
| 521 | if (di_flags & XFS_DIFLAG_PREALLOC) |
| 522 | flags |= XFS_XFLAG_PREALLOC; |
| 523 | if (di_flags & XFS_DIFLAG_IMMUTABLE) |
| 524 | flags |= XFS_XFLAG_IMMUTABLE; |
| 525 | if (di_flags & XFS_DIFLAG_APPEND) |
| 526 | flags |= XFS_XFLAG_APPEND; |
| 527 | if (di_flags & XFS_DIFLAG_SYNC) |
| 528 | flags |= XFS_XFLAG_SYNC; |
| 529 | if (di_flags & XFS_DIFLAG_NOATIME) |
| 530 | flags |= XFS_XFLAG_NOATIME; |
| 531 | if (di_flags & XFS_DIFLAG_NODUMP) |
| 532 | flags |= XFS_XFLAG_NODUMP; |
| 533 | if (di_flags & XFS_DIFLAG_RTINHERIT) |
| 534 | flags |= XFS_XFLAG_RTINHERIT; |
| 535 | if (di_flags & XFS_DIFLAG_PROJINHERIT) |
| 536 | flags |= XFS_XFLAG_PROJINHERIT; |
| 537 | if (di_flags & XFS_DIFLAG_NOSYMLINKS) |
| 538 | flags |= XFS_XFLAG_NOSYMLINKS; |
Nathan Scott | dd9f438 | 2006-01-11 15:28:28 +1100 | [diff] [blame] | 539 | if (di_flags & XFS_DIFLAG_EXTSIZE) |
| 540 | flags |= XFS_XFLAG_EXTSIZE; |
| 541 | if (di_flags & XFS_DIFLAG_EXTSZINHERIT) |
| 542 | flags |= XFS_XFLAG_EXTSZINHERIT; |
Barry Naujok | d3446ea | 2006-06-09 14:54:19 +1000 | [diff] [blame] | 543 | if (di_flags & XFS_DIFLAG_NODEFRAG) |
| 544 | flags |= XFS_XFLAG_NODEFRAG; |
David Chinner | 2a82b8b | 2007-07-11 11:09:12 +1000 | [diff] [blame] | 545 | if (di_flags & XFS_DIFLAG_FILESTREAM) |
| 546 | flags |= XFS_XFLAG_FILESTREAM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 547 | } |
| 548 | |
| 549 | return flags; |
| 550 | } |
| 551 | |
| 552 | uint |
| 553 | xfs_ip2xflags( |
| 554 | xfs_inode_t *ip) |
| 555 | { |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 556 | xfs_icdinode_t *dic = &ip->i_d; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 557 | |
Nathan Scott | a916e2b | 2006-06-09 17:12:17 +1000 | [diff] [blame] | 558 | return _xfs_dic2xflags(dic->di_flags) | |
Christoph Hellwig | 45ba598 | 2007-12-07 14:07:20 +1100 | [diff] [blame] | 559 | (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | } |
| 561 | |
| 562 | uint |
| 563 | xfs_dic2xflags( |
Christoph Hellwig | 45ba598 | 2007-12-07 14:07:20 +1100 | [diff] [blame] | 564 | xfs_dinode_t *dip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 565 | { |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 566 | return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | |
Christoph Hellwig | 45ba598 | 2007-12-07 14:07:20 +1100 | [diff] [blame] | 567 | (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | } |
| 569 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 570 | /* |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 571 | * Lookups up an inode from "name". If ci_name is not NULL, then a CI match |
| 572 | * is allowed, otherwise it has to be an exact match. If a CI match is found, |
| 573 | * ci_name->name will point to a the actual name (caller must free) or |
| 574 | * will be set to NULL if an exact match is found. |
| 575 | */ |
| 576 | int |
| 577 | xfs_lookup( |
| 578 | xfs_inode_t *dp, |
| 579 | struct xfs_name *name, |
| 580 | xfs_inode_t **ipp, |
| 581 | struct xfs_name *ci_name) |
| 582 | { |
| 583 | xfs_ino_t inum; |
| 584 | int error; |
| 585 | uint lock_mode; |
| 586 | |
| 587 | trace_xfs_lookup(dp, name); |
| 588 | |
| 589 | if (XFS_FORCED_SHUTDOWN(dp->i_mount)) |
| 590 | return XFS_ERROR(EIO); |
| 591 | |
| 592 | lock_mode = xfs_ilock_map_shared(dp); |
| 593 | error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); |
| 594 | xfs_iunlock_map_shared(dp, lock_mode); |
| 595 | |
| 596 | if (error) |
| 597 | goto out; |
| 598 | |
| 599 | error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); |
| 600 | if (error) |
| 601 | goto out_free_name; |
| 602 | |
| 603 | return 0; |
| 604 | |
| 605 | out_free_name: |
| 606 | if (ci_name) |
| 607 | kmem_free(ci_name->name); |
| 608 | out: |
| 609 | *ipp = NULL; |
| 610 | return error; |
| 611 | } |
| 612 | |
| 613 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | * Allocate an inode on disk and return a copy of its in-core version. |
| 615 | * The in-core inode is locked exclusively. Set mode, nlink, and rdev |
| 616 | * appropriately within the inode. The uid and gid for the inode are |
| 617 | * set according to the contents of the given cred structure. |
| 618 | * |
| 619 | * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() |
Carlos Maiolino | cd856db | 2012-10-20 11:08:19 -0300 | [diff] [blame] | 620 | * has a free inode available, call xfs_iget() to obtain the in-core |
| 621 | * version of the allocated inode. Finally, fill in the inode and |
| 622 | * log its initial contents. In this case, ialloc_context would be |
| 623 | * set to NULL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | * |
Carlos Maiolino | cd856db | 2012-10-20 11:08:19 -0300 | [diff] [blame] | 625 | * If xfs_dialloc() does not have an available inode, it will replenish |
| 626 | * its supply by doing an allocation. Since we can only do one |
| 627 | * allocation within a transaction without deadlocks, we must commit |
| 628 | * the current transaction before returning the inode itself. |
| 629 | * In this case, therefore, we will set ialloc_context and return. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | * The caller should then commit the current transaction, start a new |
| 631 | * transaction, and call xfs_ialloc() again to actually get the inode. |
| 632 | * |
| 633 | * To ensure that some other process does not grab the inode that |
| 634 | * was allocated during the first call to xfs_ialloc(), this routine |
| 635 | * also returns the [locked] bp pointing to the head of the freelist |
| 636 | * as ialloc_context. The caller should hold this buffer across |
| 637 | * the commit and pass it back into this routine on the second call. |
David Chinner | b11f94d | 2007-07-11 11:09:33 +1000 | [diff] [blame] | 638 | * |
| 639 | * If we are allocating quota inodes, we do not have a parent inode |
| 640 | * to attach to or associate with (i.e. pip == NULL) because they |
| 641 | * are not linked into the directory structure - they are attached |
| 642 | * directly to the superblock - and so have no parent. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | */ |
| 644 | int |
| 645 | xfs_ialloc( |
| 646 | xfs_trans_t *tp, |
| 647 | xfs_inode_t *pip, |
Al Viro | 576b1d6 | 2011-07-26 02:50:15 -0400 | [diff] [blame] | 648 | umode_t mode, |
Nathan Scott | 31b084a | 2005-05-05 13:25:00 -0700 | [diff] [blame] | 649 | xfs_nlink_t nlink, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 650 | xfs_dev_t rdev, |
Arkadiusz Mi?kiewicz | 6743099 | 2010-09-26 06:10:18 +0000 | [diff] [blame] | 651 | prid_t prid, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | int okalloc, |
| 653 | xfs_buf_t **ialloc_context, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 654 | xfs_inode_t **ipp) |
| 655 | { |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 656 | struct xfs_mount *mp = tp->t_mountp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 657 | xfs_ino_t ino; |
| 658 | xfs_inode_t *ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 659 | uint flags; |
| 660 | int error; |
Christoph Hellwig | dff35fd | 2008-08-13 16:44:15 +1000 | [diff] [blame] | 661 | timespec_t tv; |
David Chinner | bf90424 | 2008-10-30 17:36:14 +1100 | [diff] [blame] | 662 | int filestreams = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | |
| 664 | /* |
| 665 | * Call the space management code to pick |
| 666 | * the on-disk inode to be allocated. |
| 667 | */ |
David Chinner | b11f94d | 2007-07-11 11:09:33 +1000 | [diff] [blame] | 668 | error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, |
Christoph Hellwig | 0835890 | 2012-07-04 10:54:47 -0400 | [diff] [blame] | 669 | ialloc_context, &ino); |
David Chinner | bf90424 | 2008-10-30 17:36:14 +1100 | [diff] [blame] | 670 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 671 | return error; |
Christoph Hellwig | 0835890 | 2012-07-04 10:54:47 -0400 | [diff] [blame] | 672 | if (*ialloc_context || ino == NULLFSINO) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | *ipp = NULL; |
| 674 | return 0; |
| 675 | } |
| 676 | ASSERT(*ialloc_context == NULL); |
| 677 | |
| 678 | /* |
| 679 | * Get the in-core inode with the lock held exclusively. |
| 680 | * This is because we're setting fields here we need |
| 681 | * to prevent others from looking at until we're done. |
| 682 | */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 683 | error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, |
Christoph Hellwig | ec3ba85 | 2011-02-13 13:26:42 +0000 | [diff] [blame] | 684 | XFS_ILOCK_EXCL, &ip); |
David Chinner | bf90424 | 2008-10-30 17:36:14 +1100 | [diff] [blame] | 685 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 687 | ASSERT(ip != NULL); |
| 688 | |
Al Viro | 576b1d6 | 2011-07-26 02:50:15 -0400 | [diff] [blame] | 689 | ip->i_d.di_mode = mode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | ip->i_d.di_onlink = 0; |
| 691 | ip->i_d.di_nlink = nlink; |
| 692 | ASSERT(ip->i_d.di_nlink == nlink); |
Dwight Engen | 7aab1b2 | 2013-08-15 14:08:01 -0400 | [diff] [blame] | 693 | ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); |
| 694 | ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); |
Arkadiusz Mi?kiewicz | 6743099 | 2010-09-26 06:10:18 +0000 | [diff] [blame] | 695 | xfs_set_projid(ip, prid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
| 697 | |
| 698 | /* |
| 699 | * If the superblock version is up to where we support new format |
| 700 | * inodes and this is currently an old format inode, then change |
| 701 | * the inode version number now. This way we only do the conversion |
| 702 | * here rather than here and in the flush/logging code. |
| 703 | */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 704 | if (xfs_sb_version_hasnlink(&mp->m_sb) && |
Christoph Hellwig | 51ce16d | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 705 | ip->i_d.di_version == 1) { |
| 706 | ip->i_d.di_version = 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 707 | /* |
| 708 | * We've already zeroed the old link count, the projid field, |
| 709 | * and the pad field. |
| 710 | */ |
| 711 | } |
| 712 | |
| 713 | /* |
| 714 | * Project ids won't be stored on disk if we are using a version 1 inode. |
| 715 | */ |
Christoph Hellwig | 51ce16d | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 716 | if ((prid != 0) && (ip->i_d.di_version == 1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 717 | xfs_bump_ino_vers2(tp, ip); |
| 718 | |
Christoph Hellwig | bd186aa | 2007-08-30 17:21:12 +1000 | [diff] [blame] | 719 | if (pip && XFS_INHERIT_GID(pip)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 720 | ip->i_d.di_gid = pip->i_d.di_gid; |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 721 | if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 722 | ip->i_d.di_mode |= S_ISGID; |
| 723 | } |
| 724 | } |
| 725 | |
| 726 | /* |
| 727 | * If the group ID of the new file does not match the effective group |
| 728 | * ID or one of the supplementary group IDs, the S_ISGID bit is cleared |
| 729 | * (and only if the irix_sgid_inherit compatibility variable is set). |
| 730 | */ |
| 731 | if ((irix_sgid_inherit) && |
| 732 | (ip->i_d.di_mode & S_ISGID) && |
Dwight Engen | 7aab1b2 | 2013-08-15 14:08:01 -0400 | [diff] [blame] | 733 | (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | ip->i_d.di_mode &= ~S_ISGID; |
| 735 | } |
| 736 | |
| 737 | ip->i_d.di_size = 0; |
| 738 | ip->i_d.di_nextents = 0; |
| 739 | ASSERT(ip->i_d.di_nblocks == 0); |
Christoph Hellwig | dff35fd | 2008-08-13 16:44:15 +1000 | [diff] [blame] | 740 | |
| 741 | nanotime(&tv); |
| 742 | ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; |
| 743 | ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; |
| 744 | ip->i_d.di_atime = ip->i_d.di_mtime; |
| 745 | ip->i_d.di_ctime = ip->i_d.di_mtime; |
| 746 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | /* |
| 748 | * di_gen will have been taken care of in xfs_iread. |
| 749 | */ |
| 750 | ip->i_d.di_extsize = 0; |
| 751 | ip->i_d.di_dmevmask = 0; |
| 752 | ip->i_d.di_dmstate = 0; |
| 753 | ip->i_d.di_flags = 0; |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 754 | |
| 755 | if (ip->i_d.di_version == 3) { |
| 756 | ASSERT(ip->i_d.di_ino == ino); |
| 757 | ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); |
| 758 | ip->i_d.di_crc = 0; |
| 759 | ip->i_d.di_changecount = 1; |
| 760 | ip->i_d.di_lsn = 0; |
| 761 | ip->i_d.di_flags2 = 0; |
| 762 | memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); |
| 763 | ip->i_d.di_crtime = ip->i_d.di_mtime; |
| 764 | } |
| 765 | |
| 766 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 767 | flags = XFS_ILOG_CORE; |
| 768 | switch (mode & S_IFMT) { |
| 769 | case S_IFIFO: |
| 770 | case S_IFCHR: |
| 771 | case S_IFBLK: |
| 772 | case S_IFSOCK: |
| 773 | ip->i_d.di_format = XFS_DINODE_FMT_DEV; |
| 774 | ip->i_df.if_u2.if_rdev = rdev; |
| 775 | ip->i_df.if_flags = 0; |
| 776 | flags |= XFS_ILOG_DEV; |
| 777 | break; |
| 778 | case S_IFREG: |
David Chinner | bf90424 | 2008-10-30 17:36:14 +1100 | [diff] [blame] | 779 | /* |
| 780 | * we can't set up filestreams until after the VFS inode |
| 781 | * is set up properly. |
| 782 | */ |
| 783 | if (pip && xfs_inode_is_filestream(pip)) |
| 784 | filestreams = 1; |
David Chinner | 2a82b8b | 2007-07-11 11:09:12 +1000 | [diff] [blame] | 785 | /* fall through */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | case S_IFDIR: |
David Chinner | b11f94d | 2007-07-11 11:09:33 +1000 | [diff] [blame] | 787 | if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 788 | uint di_flags = 0; |
| 789 | |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 790 | if (S_ISDIR(mode)) { |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 791 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
| 792 | di_flags |= XFS_DIFLAG_RTINHERIT; |
Nathan Scott | dd9f438 | 2006-01-11 15:28:28 +1100 | [diff] [blame] | 793 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
| 794 | di_flags |= XFS_DIFLAG_EXTSZINHERIT; |
| 795 | ip->i_d.di_extsize = pip->i_d.di_extsize; |
| 796 | } |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 797 | } else if (S_ISREG(mode)) { |
Christoph Hellwig | 613d704 | 2007-10-11 17:44:08 +1000 | [diff] [blame] | 798 | if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 799 | di_flags |= XFS_DIFLAG_REALTIME; |
Nathan Scott | dd9f438 | 2006-01-11 15:28:28 +1100 | [diff] [blame] | 800 | if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { |
| 801 | di_flags |= XFS_DIFLAG_EXTSIZE; |
| 802 | ip->i_d.di_extsize = pip->i_d.di_extsize; |
| 803 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | } |
| 805 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && |
| 806 | xfs_inherit_noatime) |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 807 | di_flags |= XFS_DIFLAG_NOATIME; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && |
| 809 | xfs_inherit_nodump) |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 810 | di_flags |= XFS_DIFLAG_NODUMP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 811 | if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && |
| 812 | xfs_inherit_sync) |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 813 | di_flags |= XFS_DIFLAG_SYNC; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && |
| 815 | xfs_inherit_nosymlinks) |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 816 | di_flags |= XFS_DIFLAG_NOSYMLINKS; |
| 817 | if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
| 818 | di_flags |= XFS_DIFLAG_PROJINHERIT; |
Barry Naujok | d3446ea | 2006-06-09 14:54:19 +1000 | [diff] [blame] | 819 | if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && |
| 820 | xfs_inherit_nodefrag) |
| 821 | di_flags |= XFS_DIFLAG_NODEFRAG; |
David Chinner | 2a82b8b | 2007-07-11 11:09:12 +1000 | [diff] [blame] | 822 | if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) |
| 823 | di_flags |= XFS_DIFLAG_FILESTREAM; |
Nathan Scott | 365ca83 | 2005-06-21 15:39:12 +1000 | [diff] [blame] | 824 | ip->i_d.di_flags |= di_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | } |
| 826 | /* FALLTHROUGH */ |
| 827 | case S_IFLNK: |
| 828 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
| 829 | ip->i_df.if_flags = XFS_IFEXTENTS; |
| 830 | ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; |
| 831 | ip->i_df.if_u1.if_extents = NULL; |
| 832 | break; |
| 833 | default: |
| 834 | ASSERT(0); |
| 835 | } |
| 836 | /* |
| 837 | * Attribute fork settings for new inode. |
| 838 | */ |
| 839 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; |
| 840 | ip->i_d.di_anextents = 0; |
| 841 | |
| 842 | /* |
| 843 | * Log the new values stuffed into the inode. |
| 844 | */ |
Christoph Hellwig | ddc3415 | 2011-09-19 15:00:54 +0000 | [diff] [blame] | 845 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | xfs_trans_log_inode(tp, ip, flags); |
| 847 | |
Nathan Scott | b83bd13 | 2006-06-09 16:48:30 +1000 | [diff] [blame] | 848 | /* now that we have an i_mode we can setup inode ops and unlock */ |
Christoph Hellwig | 41be8be | 2008-08-13 16:23:13 +1000 | [diff] [blame] | 849 | xfs_setup_inode(ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | |
David Chinner | bf90424 | 2008-10-30 17:36:14 +1100 | [diff] [blame] | 851 | /* now we have set up the vfs inode we can associate the filestream */ |
| 852 | if (filestreams) { |
| 853 | error = xfs_filestream_associate(pip, ip); |
| 854 | if (error < 0) |
| 855 | return -error; |
| 856 | if (!error) |
| 857 | xfs_iflags_set(ip, XFS_IFILESTREAM); |
| 858 | } |
| 859 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | *ipp = ip; |
| 861 | return 0; |
| 862 | } |
| 863 | |
Dave Chinner | e546cb7 | 2013-08-12 20:49:47 +1000 | [diff] [blame] | 864 | /* |
| 865 | * Allocates a new inode from disk and return a pointer to the |
| 866 | * incore copy. This routine will internally commit the current |
| 867 | * transaction and allocate a new one if the Space Manager needed |
| 868 | * to do an allocation to replenish the inode free-list. |
| 869 | * |
| 870 | * This routine is designed to be called from xfs_create and |
| 871 | * xfs_create_dir. |
| 872 | * |
| 873 | */ |
| 874 | int |
| 875 | xfs_dir_ialloc( |
| 876 | xfs_trans_t **tpp, /* input: current transaction; |
| 877 | output: may be a new transaction. */ |
| 878 | xfs_inode_t *dp, /* directory within whose allocate |
| 879 | the inode. */ |
| 880 | umode_t mode, |
| 881 | xfs_nlink_t nlink, |
| 882 | xfs_dev_t rdev, |
| 883 | prid_t prid, /* project id */ |
| 884 | int okalloc, /* ok to allocate new space */ |
| 885 | xfs_inode_t **ipp, /* pointer to inode; it will be |
| 886 | locked. */ |
| 887 | int *committed) |
| 888 | |
| 889 | { |
| 890 | xfs_trans_t *tp; |
| 891 | xfs_trans_t *ntp; |
| 892 | xfs_inode_t *ip; |
| 893 | xfs_buf_t *ialloc_context = NULL; |
| 894 | int code; |
Dave Chinner | e546cb7 | 2013-08-12 20:49:47 +1000 | [diff] [blame] | 895 | void *dqinfo; |
| 896 | uint tflags; |
| 897 | |
| 898 | tp = *tpp; |
| 899 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
| 900 | |
| 901 | /* |
| 902 | * xfs_ialloc will return a pointer to an incore inode if |
| 903 | * the Space Manager has an available inode on the free |
| 904 | * list. Otherwise, it will do an allocation and replenish |
| 905 | * the freelist. Since we can only do one allocation per |
| 906 | * transaction without deadlocks, we will need to commit the |
| 907 | * current transaction and start a new one. We will then |
| 908 | * need to call xfs_ialloc again to get the inode. |
| 909 | * |
| 910 | * If xfs_ialloc did an allocation to replenish the freelist, |
| 911 | * it returns the bp containing the head of the freelist as |
| 912 | * ialloc_context. We will hold a lock on it across the |
| 913 | * transaction commit so that no other process can steal |
| 914 | * the inode(s) that we've just allocated. |
| 915 | */ |
| 916 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, |
| 917 | &ialloc_context, &ip); |
| 918 | |
| 919 | /* |
| 920 | * Return an error if we were unable to allocate a new inode. |
| 921 | * This should only happen if we run out of space on disk or |
| 922 | * encounter a disk error. |
| 923 | */ |
| 924 | if (code) { |
| 925 | *ipp = NULL; |
| 926 | return code; |
| 927 | } |
| 928 | if (!ialloc_context && !ip) { |
| 929 | *ipp = NULL; |
| 930 | return XFS_ERROR(ENOSPC); |
| 931 | } |
| 932 | |
| 933 | /* |
| 934 | * If the AGI buffer is non-NULL, then we were unable to get an |
| 935 | * inode in one operation. We need to commit the current |
| 936 | * transaction and call xfs_ialloc() again. It is guaranteed |
| 937 | * to succeed the second time. |
| 938 | */ |
| 939 | if (ialloc_context) { |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 940 | struct xfs_trans_res tres; |
| 941 | |
Dave Chinner | e546cb7 | 2013-08-12 20:49:47 +1000 | [diff] [blame] | 942 | /* |
| 943 | * Normally, xfs_trans_commit releases all the locks. |
| 944 | * We call bhold to hang on to the ialloc_context across |
| 945 | * the commit. Holding this buffer prevents any other |
| 946 | * processes from doing any allocations in this |
| 947 | * allocation group. |
| 948 | */ |
| 949 | xfs_trans_bhold(tp, ialloc_context); |
| 950 | /* |
| 951 | * Save the log reservation so we can use |
| 952 | * them in the next transaction. |
| 953 | */ |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 954 | tres.tr_logres = xfs_trans_get_log_res(tp); |
| 955 | tres.tr_logcount = xfs_trans_get_log_count(tp); |
Dave Chinner | e546cb7 | 2013-08-12 20:49:47 +1000 | [diff] [blame] | 956 | |
| 957 | /* |
| 958 | * We want the quota changes to be associated with the next |
| 959 | * transaction, NOT this one. So, detach the dqinfo from this |
| 960 | * and attach it to the next transaction. |
| 961 | */ |
| 962 | dqinfo = NULL; |
| 963 | tflags = 0; |
| 964 | if (tp->t_dqinfo) { |
| 965 | dqinfo = (void *)tp->t_dqinfo; |
| 966 | tp->t_dqinfo = NULL; |
| 967 | tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; |
| 968 | tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); |
| 969 | } |
| 970 | |
| 971 | ntp = xfs_trans_dup(tp); |
| 972 | code = xfs_trans_commit(tp, 0); |
| 973 | tp = ntp; |
| 974 | if (committed != NULL) { |
| 975 | *committed = 1; |
| 976 | } |
| 977 | /* |
| 978 | * If we get an error during the commit processing, |
| 979 | * release the buffer that is still held and return |
| 980 | * to the caller. |
| 981 | */ |
| 982 | if (code) { |
| 983 | xfs_buf_relse(ialloc_context); |
| 984 | if (dqinfo) { |
| 985 | tp->t_dqinfo = dqinfo; |
| 986 | xfs_trans_free_dqinfo(tp); |
| 987 | } |
| 988 | *tpp = ntp; |
| 989 | *ipp = NULL; |
| 990 | return code; |
| 991 | } |
| 992 | |
| 993 | /* |
| 994 | * transaction commit worked ok so we can drop the extra ticket |
| 995 | * reference that we gained in xfs_trans_dup() |
| 996 | */ |
| 997 | xfs_log_ticket_put(tp->t_ticket); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 998 | tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; |
| 999 | code = xfs_trans_reserve(tp, &tres, 0, 0); |
| 1000 | |
Dave Chinner | e546cb7 | 2013-08-12 20:49:47 +1000 | [diff] [blame] | 1001 | /* |
| 1002 | * Re-attach the quota info that we detached from prev trx. |
| 1003 | */ |
| 1004 | if (dqinfo) { |
| 1005 | tp->t_dqinfo = dqinfo; |
| 1006 | tp->t_flags |= tflags; |
| 1007 | } |
| 1008 | |
| 1009 | if (code) { |
| 1010 | xfs_buf_relse(ialloc_context); |
| 1011 | *tpp = ntp; |
| 1012 | *ipp = NULL; |
| 1013 | return code; |
| 1014 | } |
| 1015 | xfs_trans_bjoin(tp, ialloc_context); |
| 1016 | |
| 1017 | /* |
| 1018 | * Call ialloc again. Since we've locked out all |
| 1019 | * other allocations in this allocation group, |
| 1020 | * this call should always succeed. |
| 1021 | */ |
| 1022 | code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, |
| 1023 | okalloc, &ialloc_context, &ip); |
| 1024 | |
| 1025 | /* |
| 1026 | * If we get an error at this point, return to the caller |
| 1027 | * so that the current transaction can be aborted. |
| 1028 | */ |
| 1029 | if (code) { |
| 1030 | *tpp = tp; |
| 1031 | *ipp = NULL; |
| 1032 | return code; |
| 1033 | } |
| 1034 | ASSERT(!ialloc_context && ip); |
| 1035 | |
| 1036 | } else { |
| 1037 | if (committed != NULL) |
| 1038 | *committed = 0; |
| 1039 | } |
| 1040 | |
| 1041 | *ipp = ip; |
| 1042 | *tpp = tp; |
| 1043 | |
| 1044 | return 0; |
| 1045 | } |
| 1046 | |
| 1047 | /* |
| 1048 | * Decrement the link count on an inode & log the change. |
| 1049 | * If this causes the link count to go to zero, initiate the |
| 1050 | * logging activity required to truncate a file. |
| 1051 | */ |
| 1052 | int /* error */ |
| 1053 | xfs_droplink( |
| 1054 | xfs_trans_t *tp, |
| 1055 | xfs_inode_t *ip) |
| 1056 | { |
| 1057 | int error; |
| 1058 | |
| 1059 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
| 1060 | |
| 1061 | ASSERT (ip->i_d.di_nlink > 0); |
| 1062 | ip->i_d.di_nlink--; |
| 1063 | drop_nlink(VFS_I(ip)); |
| 1064 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 1065 | |
| 1066 | error = 0; |
| 1067 | if (ip->i_d.di_nlink == 0) { |
| 1068 | /* |
| 1069 | * We're dropping the last link to this file. |
| 1070 | * Move the on-disk inode to the AGI unlinked list. |
| 1071 | * From xfs_inactive() we will pull the inode from |
| 1072 | * the list and free it. |
| 1073 | */ |
| 1074 | error = xfs_iunlink(tp, ip); |
| 1075 | } |
| 1076 | return error; |
| 1077 | } |
| 1078 | |
| 1079 | /* |
| 1080 | * This gets called when the inode's version needs to be changed from 1 to 2. |
| 1081 | * Currently this happens when the nlink field overflows the old 16-bit value |
| 1082 | * or when chproj is called to change the project for the first time. |
| 1083 | * As a side effect the superblock version will also get rev'd |
| 1084 | * to contain the NLINK bit. |
| 1085 | */ |
| 1086 | void |
| 1087 | xfs_bump_ino_vers2( |
| 1088 | xfs_trans_t *tp, |
| 1089 | xfs_inode_t *ip) |
| 1090 | { |
| 1091 | xfs_mount_t *mp; |
| 1092 | |
| 1093 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
| 1094 | ASSERT(ip->i_d.di_version == 1); |
| 1095 | |
| 1096 | ip->i_d.di_version = 2; |
| 1097 | ip->i_d.di_onlink = 0; |
| 1098 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
| 1099 | mp = tp->t_mountp; |
| 1100 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
| 1101 | spin_lock(&mp->m_sb_lock); |
| 1102 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
| 1103 | xfs_sb_version_addnlink(&mp->m_sb); |
| 1104 | spin_unlock(&mp->m_sb_lock); |
| 1105 | xfs_mod_sb(tp, XFS_SB_VERSIONNUM); |
| 1106 | } else { |
| 1107 | spin_unlock(&mp->m_sb_lock); |
| 1108 | } |
| 1109 | } |
| 1110 | /* Caller must log the inode */ |
| 1111 | } |
| 1112 | |
| 1113 | /* |
| 1114 | * Increment the link count on an inode & log the change. |
| 1115 | */ |
| 1116 | int |
| 1117 | xfs_bumplink( |
| 1118 | xfs_trans_t *tp, |
| 1119 | xfs_inode_t *ip) |
| 1120 | { |
| 1121 | xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); |
| 1122 | |
| 1123 | ASSERT(ip->i_d.di_nlink > 0); |
| 1124 | ip->i_d.di_nlink++; |
| 1125 | inc_nlink(VFS_I(ip)); |
| 1126 | if ((ip->i_d.di_version == 1) && |
| 1127 | (ip->i_d.di_nlink > XFS_MAXLINK_1)) { |
| 1128 | /* |
| 1129 | * The inode has increased its number of links beyond |
| 1130 | * what can fit in an old format inode. It now needs |
| 1131 | * to be converted to a version 2 inode with a 32 bit |
| 1132 | * link count. If this is the first inode in the file |
| 1133 | * system to do this, then we need to bump the superblock |
| 1134 | * version number as well. |
| 1135 | */ |
| 1136 | xfs_bump_ino_vers2(tp, ip); |
| 1137 | } |
| 1138 | |
| 1139 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 1140 | return 0; |
| 1141 | } |
| 1142 | |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1143 | int |
| 1144 | xfs_create( |
| 1145 | xfs_inode_t *dp, |
| 1146 | struct xfs_name *name, |
| 1147 | umode_t mode, |
| 1148 | xfs_dev_t rdev, |
| 1149 | xfs_inode_t **ipp) |
| 1150 | { |
| 1151 | int is_dir = S_ISDIR(mode); |
| 1152 | struct xfs_mount *mp = dp->i_mount; |
| 1153 | struct xfs_inode *ip = NULL; |
| 1154 | struct xfs_trans *tp = NULL; |
| 1155 | int error; |
| 1156 | xfs_bmap_free_t free_list; |
| 1157 | xfs_fsblock_t first_block; |
| 1158 | bool unlock_dp_on_error = false; |
| 1159 | uint cancel_flags; |
| 1160 | int committed; |
| 1161 | prid_t prid; |
| 1162 | struct xfs_dquot *udqp = NULL; |
| 1163 | struct xfs_dquot *gdqp = NULL; |
| 1164 | struct xfs_dquot *pdqp = NULL; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1165 | struct xfs_trans_res tres; |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1166 | uint resblks; |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1167 | |
| 1168 | trace_xfs_create(dp, name); |
| 1169 | |
| 1170 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 1171 | return XFS_ERROR(EIO); |
| 1172 | |
| 1173 | if (dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) |
| 1174 | prid = xfs_get_projid(dp); |
| 1175 | else |
| 1176 | prid = XFS_PROJID_DEFAULT; |
| 1177 | |
| 1178 | /* |
| 1179 | * Make sure that we have allocated dquot(s) on disk. |
| 1180 | */ |
Dwight Engen | 7aab1b2 | 2013-08-15 14:08:01 -0400 | [diff] [blame] | 1181 | error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), |
| 1182 | xfs_kgid_to_gid(current_fsgid()), prid, |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1183 | XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, |
| 1184 | &udqp, &gdqp, &pdqp); |
| 1185 | if (error) |
| 1186 | return error; |
| 1187 | |
| 1188 | if (is_dir) { |
| 1189 | rdev = 0; |
| 1190 | resblks = XFS_MKDIR_SPACE_RES(mp, name->len); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1191 | tres.tr_logres = M_RES(mp)->tr_mkdir.tr_logres; |
| 1192 | tres.tr_logcount = XFS_MKDIR_LOG_COUNT; |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1193 | tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); |
| 1194 | } else { |
| 1195 | resblks = XFS_CREATE_SPACE_RES(mp, name->len); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1196 | tres.tr_logres = M_RES(mp)->tr_create.tr_logres; |
| 1197 | tres.tr_logcount = XFS_CREATE_LOG_COUNT; |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1198 | tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); |
| 1199 | } |
| 1200 | |
| 1201 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
| 1202 | |
| 1203 | /* |
| 1204 | * Initially assume that the file does not exist and |
| 1205 | * reserve the resources for that case. If that is not |
| 1206 | * the case we'll drop the one we have and get a more |
| 1207 | * appropriate transaction later. |
| 1208 | */ |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1209 | tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; |
| 1210 | error = xfs_trans_reserve(tp, &tres, resblks, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1211 | if (error == ENOSPC) { |
| 1212 | /* flush outstanding delalloc blocks and retry */ |
| 1213 | xfs_flush_inodes(mp); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1214 | error = xfs_trans_reserve(tp, &tres, resblks, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1215 | } |
| 1216 | if (error == ENOSPC) { |
| 1217 | /* No space at all so try a "no-allocation" reservation */ |
| 1218 | resblks = 0; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1219 | error = xfs_trans_reserve(tp, &tres, 0, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1220 | } |
| 1221 | if (error) { |
| 1222 | cancel_flags = 0; |
| 1223 | goto out_trans_cancel; |
| 1224 | } |
| 1225 | |
| 1226 | xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
| 1227 | unlock_dp_on_error = true; |
| 1228 | |
| 1229 | xfs_bmap_init(&free_list, &first_block); |
| 1230 | |
| 1231 | /* |
| 1232 | * Reserve disk quota and the inode. |
| 1233 | */ |
| 1234 | error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, |
| 1235 | pdqp, resblks, 1, 0); |
| 1236 | if (error) |
| 1237 | goto out_trans_cancel; |
| 1238 | |
| 1239 | error = xfs_dir_canenter(tp, dp, name, resblks); |
| 1240 | if (error) |
| 1241 | goto out_trans_cancel; |
| 1242 | |
| 1243 | /* |
| 1244 | * A newly created regular or special file just has one directory |
| 1245 | * entry pointing to them, but a directory also the "." entry |
| 1246 | * pointing to itself. |
| 1247 | */ |
| 1248 | error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, |
| 1249 | prid, resblks > 0, &ip, &committed); |
| 1250 | if (error) { |
| 1251 | if (error == ENOSPC) |
| 1252 | goto out_trans_cancel; |
| 1253 | goto out_trans_abort; |
| 1254 | } |
| 1255 | |
| 1256 | /* |
| 1257 | * Now we join the directory inode to the transaction. We do not do it |
| 1258 | * earlier because xfs_dir_ialloc might commit the previous transaction |
| 1259 | * (and release all the locks). An error from here on will result in |
| 1260 | * the transaction cancel unlocking dp so don't do it explicitly in the |
| 1261 | * error path. |
| 1262 | */ |
| 1263 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
| 1264 | unlock_dp_on_error = false; |
| 1265 | |
| 1266 | error = xfs_dir_createname(tp, dp, name, ip->i_ino, |
| 1267 | &first_block, &free_list, resblks ? |
| 1268 | resblks - XFS_IALLOC_SPACE_RES(mp) : 0); |
| 1269 | if (error) { |
| 1270 | ASSERT(error != ENOSPC); |
| 1271 | goto out_trans_abort; |
| 1272 | } |
| 1273 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 1274 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
| 1275 | |
| 1276 | if (is_dir) { |
| 1277 | error = xfs_dir_init(tp, ip, dp); |
| 1278 | if (error) |
| 1279 | goto out_bmap_cancel; |
| 1280 | |
| 1281 | error = xfs_bumplink(tp, dp); |
| 1282 | if (error) |
| 1283 | goto out_bmap_cancel; |
| 1284 | } |
| 1285 | |
| 1286 | /* |
| 1287 | * If this is a synchronous mount, make sure that the |
| 1288 | * create transaction goes to disk before returning to |
| 1289 | * the user. |
| 1290 | */ |
| 1291 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
| 1292 | xfs_trans_set_sync(tp); |
| 1293 | |
| 1294 | /* |
| 1295 | * Attach the dquot(s) to the inodes and modify them incore. |
| 1296 | * These ids of the inode couldn't have changed since the new |
| 1297 | * inode has been locked ever since it was created. |
| 1298 | */ |
| 1299 | xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); |
| 1300 | |
| 1301 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
| 1302 | if (error) |
| 1303 | goto out_bmap_cancel; |
| 1304 | |
| 1305 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 1306 | if (error) |
| 1307 | goto out_release_inode; |
| 1308 | |
| 1309 | xfs_qm_dqrele(udqp); |
| 1310 | xfs_qm_dqrele(gdqp); |
| 1311 | xfs_qm_dqrele(pdqp); |
| 1312 | |
| 1313 | *ipp = ip; |
| 1314 | return 0; |
| 1315 | |
| 1316 | out_bmap_cancel: |
| 1317 | xfs_bmap_cancel(&free_list); |
| 1318 | out_trans_abort: |
| 1319 | cancel_flags |= XFS_TRANS_ABORT; |
| 1320 | out_trans_cancel: |
| 1321 | xfs_trans_cancel(tp, cancel_flags); |
| 1322 | out_release_inode: |
| 1323 | /* |
| 1324 | * Wait until after the current transaction is aborted to |
| 1325 | * release the inode. This prevents recursive transactions |
| 1326 | * and deadlocks from xfs_inactive. |
| 1327 | */ |
| 1328 | if (ip) |
| 1329 | IRELE(ip); |
| 1330 | |
| 1331 | xfs_qm_dqrele(udqp); |
| 1332 | xfs_qm_dqrele(gdqp); |
| 1333 | xfs_qm_dqrele(pdqp); |
| 1334 | |
| 1335 | if (unlock_dp_on_error) |
| 1336 | xfs_iunlock(dp, XFS_ILOCK_EXCL); |
| 1337 | return error; |
| 1338 | } |
| 1339 | |
| 1340 | int |
| 1341 | xfs_link( |
| 1342 | xfs_inode_t *tdp, |
| 1343 | xfs_inode_t *sip, |
| 1344 | struct xfs_name *target_name) |
| 1345 | { |
| 1346 | xfs_mount_t *mp = tdp->i_mount; |
| 1347 | xfs_trans_t *tp; |
| 1348 | int error; |
| 1349 | xfs_bmap_free_t free_list; |
| 1350 | xfs_fsblock_t first_block; |
| 1351 | int cancel_flags; |
| 1352 | int committed; |
| 1353 | int resblks; |
| 1354 | |
| 1355 | trace_xfs_link(tdp, target_name); |
| 1356 | |
| 1357 | ASSERT(!S_ISDIR(sip->i_d.di_mode)); |
| 1358 | |
| 1359 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 1360 | return XFS_ERROR(EIO); |
| 1361 | |
| 1362 | error = xfs_qm_dqattach(sip, 0); |
| 1363 | if (error) |
| 1364 | goto std_return; |
| 1365 | |
| 1366 | error = xfs_qm_dqattach(tdp, 0); |
| 1367 | if (error) |
| 1368 | goto std_return; |
| 1369 | |
| 1370 | tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); |
| 1371 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
| 1372 | resblks = XFS_LINK_SPACE_RES(mp, target_name->len); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1373 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1374 | if (error == ENOSPC) { |
| 1375 | resblks = 0; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1376 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1377 | } |
| 1378 | if (error) { |
| 1379 | cancel_flags = 0; |
| 1380 | goto error_return; |
| 1381 | } |
| 1382 | |
| 1383 | xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); |
| 1384 | |
| 1385 | xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); |
| 1386 | xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); |
| 1387 | |
| 1388 | /* |
| 1389 | * If we are using project inheritance, we only allow hard link |
| 1390 | * creation in our tree when the project IDs are the same; else |
| 1391 | * the tree quota mechanism could be circumvented. |
| 1392 | */ |
| 1393 | if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && |
| 1394 | (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { |
| 1395 | error = XFS_ERROR(EXDEV); |
| 1396 | goto error_return; |
| 1397 | } |
| 1398 | |
| 1399 | error = xfs_dir_canenter(tp, tdp, target_name, resblks); |
| 1400 | if (error) |
| 1401 | goto error_return; |
| 1402 | |
| 1403 | xfs_bmap_init(&free_list, &first_block); |
| 1404 | |
| 1405 | error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, |
| 1406 | &first_block, &free_list, resblks); |
| 1407 | if (error) |
| 1408 | goto abort_return; |
| 1409 | xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 1410 | xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); |
| 1411 | |
| 1412 | error = xfs_bumplink(tp, sip); |
| 1413 | if (error) |
| 1414 | goto abort_return; |
| 1415 | |
| 1416 | /* |
| 1417 | * If this is a synchronous mount, make sure that the |
| 1418 | * link transaction goes to disk before returning to |
| 1419 | * the user. |
| 1420 | */ |
| 1421 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { |
| 1422 | xfs_trans_set_sync(tp); |
| 1423 | } |
| 1424 | |
| 1425 | error = xfs_bmap_finish (&tp, &free_list, &committed); |
| 1426 | if (error) { |
| 1427 | xfs_bmap_cancel(&free_list); |
| 1428 | goto abort_return; |
| 1429 | } |
| 1430 | |
| 1431 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 1432 | |
| 1433 | abort_return: |
| 1434 | cancel_flags |= XFS_TRANS_ABORT; |
| 1435 | error_return: |
| 1436 | xfs_trans_cancel(tp, cancel_flags); |
| 1437 | std_return: |
| 1438 | return error; |
| 1439 | } |
| 1440 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1441 | /* |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1442 | * Free up the underlying blocks past new_size. The new size must be smaller |
| 1443 | * than the current size. This routine can be used both for the attribute and |
| 1444 | * data fork, and does not modify the inode size, which is left to the caller. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1445 | * |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1446 | * The transaction passed to this routine must have made a permanent log |
| 1447 | * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the |
| 1448 | * given transaction and start new ones, so make sure everything involved in |
| 1449 | * the transaction is tidy before calling here. Some transaction will be |
| 1450 | * returned to the caller to be committed. The incoming transaction must |
| 1451 | * already include the inode, and both inode locks must be held exclusively. |
| 1452 | * The inode must also be "held" within the transaction. On return the inode |
| 1453 | * will be "held" within the returned transaction. This routine does NOT |
| 1454 | * require any disk space to be reserved for it within the transaction. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | * |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1456 | * If we get an error, we must return with the inode locked and linked into the |
| 1457 | * current transaction. This keeps things simple for the higher level code, |
| 1458 | * because it always knows that the inode is locked and held in the transaction |
| 1459 | * that returns to it whether errors occur or not. We don't mark the inode |
| 1460 | * dirty on error so that transactions can be easily aborted if possible. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1461 | */ |
| 1462 | int |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1463 | xfs_itruncate_extents( |
| 1464 | struct xfs_trans **tpp, |
| 1465 | struct xfs_inode *ip, |
| 1466 | int whichfork, |
| 1467 | xfs_fsize_t new_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1468 | { |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1469 | struct xfs_mount *mp = ip->i_mount; |
| 1470 | struct xfs_trans *tp = *tpp; |
| 1471 | struct xfs_trans *ntp; |
| 1472 | xfs_bmap_free_t free_list; |
| 1473 | xfs_fsblock_t first_block; |
| 1474 | xfs_fileoff_t first_unmap_block; |
| 1475 | xfs_fileoff_t last_block; |
| 1476 | xfs_filblks_t unmap_len; |
| 1477 | int committed; |
| 1478 | int error = 0; |
| 1479 | int done = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1480 | |
Christoph Hellwig | 0b56185 | 2012-07-04 11:13:31 -0400 | [diff] [blame] | 1481 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
| 1482 | ASSERT(!atomic_read(&VFS_I(ip)->i_count) || |
| 1483 | xfs_isilocked(ip, XFS_IOLOCK_EXCL)); |
Christoph Hellwig | ce7ae151 | 2011-12-18 20:00:11 +0000 | [diff] [blame] | 1484 | ASSERT(new_size <= XFS_ISIZE(ip)); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1485 | ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | ASSERT(ip->i_itemp != NULL); |
Christoph Hellwig | 898621d | 2010-06-24 11:36:58 +1000 | [diff] [blame] | 1487 | ASSERT(ip->i_itemp->ili_lock_flags == 0); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1488 | ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1489 | |
Christoph Hellwig | 673e8e5 | 2011-12-18 20:00:04 +0000 | [diff] [blame] | 1490 | trace_xfs_itruncate_extents_start(ip, new_size); |
| 1491 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | /* |
| 1493 | * Since it is possible for space to become allocated beyond |
| 1494 | * the end of the file (in a crash where the space is allocated |
| 1495 | * but the inode size is not yet updated), simply remove any |
| 1496 | * blocks which show up between the new EOF and the maximum |
| 1497 | * possible file size. If the first block to be removed is |
| 1498 | * beyond the maximum file size (ie it is the same as last_block), |
| 1499 | * then there is nothing to do. |
| 1500 | */ |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1501 | first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); |
Dave Chinner | 3297238 | 2012-06-08 15:44:54 +1000 | [diff] [blame] | 1502 | last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1503 | if (first_unmap_block == last_block) |
| 1504 | return 0; |
| 1505 | |
| 1506 | ASSERT(first_unmap_block < last_block); |
| 1507 | unmap_len = last_block - first_unmap_block + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | while (!done) { |
Eric Sandeen | 9d87c31 | 2009-01-14 23:22:07 -0600 | [diff] [blame] | 1509 | xfs_bmap_init(&free_list, &first_block); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1510 | error = xfs_bunmapi(tp, ip, |
Olaf Weber | 3e57ecf | 2006-06-09 14:48:12 +1000 | [diff] [blame] | 1511 | first_unmap_block, unmap_len, |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1512 | xfs_bmapi_aflag(whichfork), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1513 | XFS_ITRUNC_MAX_EXTENTS, |
Olaf Weber | 3e57ecf | 2006-06-09 14:48:12 +1000 | [diff] [blame] | 1514 | &first_block, &free_list, |
Christoph Hellwig | b4e9181 | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 1515 | &done); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1516 | if (error) |
| 1517 | goto out_bmap_cancel; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | |
| 1519 | /* |
| 1520 | * Duplicate the transaction that has the permanent |
| 1521 | * reservation and commit the old transaction. |
| 1522 | */ |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1523 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
Christoph Hellwig | 898621d | 2010-06-24 11:36:58 +1000 | [diff] [blame] | 1524 | if (committed) |
Christoph Hellwig | ddc3415 | 2011-09-19 15:00:54 +0000 | [diff] [blame] | 1525 | xfs_trans_ijoin(tp, ip, 0); |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1526 | if (error) |
| 1527 | goto out_bmap_cancel; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | |
| 1529 | if (committed) { |
| 1530 | /* |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1531 | * Mark the inode dirty so it will be logged and |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1532 | * moved forward in the log as part of every commit. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1533 | */ |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1534 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1535 | } |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1536 | |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1537 | ntp = xfs_trans_dup(tp); |
| 1538 | error = xfs_trans_commit(tp, 0); |
| 1539 | tp = ntp; |
David Chinner | e5720ee | 2008-04-10 12:21:18 +1000 | [diff] [blame] | 1540 | |
Christoph Hellwig | ddc3415 | 2011-09-19 15:00:54 +0000 | [diff] [blame] | 1541 | xfs_trans_ijoin(tp, ip, 0); |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1542 | |
Dave Chinner | cc09c0d | 2008-11-17 17:37:10 +1100 | [diff] [blame] | 1543 | if (error) |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1544 | goto out; |
| 1545 | |
Dave Chinner | cc09c0d | 2008-11-17 17:37:10 +1100 | [diff] [blame] | 1546 | /* |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1547 | * Transaction commit worked ok so we can drop the extra ticket |
Dave Chinner | cc09c0d | 2008-11-17 17:37:10 +1100 | [diff] [blame] | 1548 | * reference that we gained in xfs_trans_dup() |
| 1549 | */ |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1550 | xfs_log_ticket_put(tp->t_ticket); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1551 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); |
David Chinner | f648505 | 2008-04-17 16:50:04 +1000 | [diff] [blame] | 1552 | if (error) |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1553 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1554 | } |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1555 | |
Christoph Hellwig | 673e8e5 | 2011-12-18 20:00:04 +0000 | [diff] [blame] | 1556 | /* |
| 1557 | * Always re-log the inode so that our permanent transaction can keep |
| 1558 | * on rolling it forward in the log. |
| 1559 | */ |
| 1560 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 1561 | |
| 1562 | trace_xfs_itruncate_extents_end(ip, new_size); |
| 1563 | |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1564 | out: |
| 1565 | *tpp = tp; |
| 1566 | return error; |
| 1567 | out_bmap_cancel: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1568 | /* |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1569 | * If the bunmapi call encounters an error, return to the caller where |
| 1570 | * the transaction can be properly aborted. We just need to make sure |
| 1571 | * we're not holding any resources that we were not when we came in. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1572 | */ |
Christoph Hellwig | 8f04c47 | 2011-07-08 14:34:34 +0200 | [diff] [blame] | 1573 | xfs_bmap_cancel(&free_list); |
| 1574 | goto out; |
| 1575 | } |
| 1576 | |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1577 | int |
| 1578 | xfs_release( |
| 1579 | xfs_inode_t *ip) |
| 1580 | { |
| 1581 | xfs_mount_t *mp = ip->i_mount; |
| 1582 | int error; |
| 1583 | |
| 1584 | if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) |
| 1585 | return 0; |
| 1586 | |
| 1587 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
| 1588 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
| 1589 | return 0; |
| 1590 | |
| 1591 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
| 1592 | int truncated; |
| 1593 | |
| 1594 | /* |
| 1595 | * If we are using filestreams, and we have an unlinked |
| 1596 | * file that we are processing the last close on, then nothing |
| 1597 | * will be able to reopen and write to this file. Purge this |
| 1598 | * inode from the filestreams cache so that it doesn't delay |
| 1599 | * teardown of the inode. |
| 1600 | */ |
| 1601 | if ((ip->i_d.di_nlink == 0) && xfs_inode_is_filestream(ip)) |
| 1602 | xfs_filestream_deassociate(ip); |
| 1603 | |
| 1604 | /* |
| 1605 | * If we previously truncated this file and removed old data |
| 1606 | * in the process, we want to initiate "early" writeout on |
| 1607 | * the last close. This is an attempt to combat the notorious |
| 1608 | * NULL files problem which is particularly noticeable from a |
| 1609 | * truncate down, buffered (re-)write (delalloc), followed by |
| 1610 | * a crash. What we are effectively doing here is |
| 1611 | * significantly reducing the time window where we'd otherwise |
| 1612 | * be exposed to that problem. |
| 1613 | */ |
| 1614 | truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); |
| 1615 | if (truncated) { |
| 1616 | xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); |
| 1617 | if (VN_DIRTY(VFS_I(ip)) && ip->i_delayed_blks > 0) { |
| 1618 | error = -filemap_flush(VFS_I(ip)->i_mapping); |
| 1619 | if (error) |
| 1620 | return error; |
| 1621 | } |
| 1622 | } |
| 1623 | } |
| 1624 | |
| 1625 | if (ip->i_d.di_nlink == 0) |
| 1626 | return 0; |
| 1627 | |
| 1628 | if (xfs_can_free_eofblocks(ip, false)) { |
| 1629 | |
| 1630 | /* |
| 1631 | * If we can't get the iolock just skip truncating the blocks |
| 1632 | * past EOF because we could deadlock with the mmap_sem |
| 1633 | * otherwise. We'll get another chance to drop them once the |
| 1634 | * last reference to the inode is dropped, so we'll never leak |
| 1635 | * blocks permanently. |
| 1636 | * |
| 1637 | * Further, check if the inode is being opened, written and |
| 1638 | * closed frequently and we have delayed allocation blocks |
| 1639 | * outstanding (e.g. streaming writes from the NFS server), |
| 1640 | * truncating the blocks past EOF will cause fragmentation to |
| 1641 | * occur. |
| 1642 | * |
| 1643 | * In this case don't do the truncation, either, but we have to |
| 1644 | * be careful how we detect this case. Blocks beyond EOF show |
| 1645 | * up as i_delayed_blks even when the inode is clean, so we |
| 1646 | * need to truncate them away first before checking for a dirty |
| 1647 | * release. Hence on the first dirty close we will still remove |
| 1648 | * the speculative allocation, but after that we will leave it |
| 1649 | * in place. |
| 1650 | */ |
| 1651 | if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) |
| 1652 | return 0; |
| 1653 | |
| 1654 | error = xfs_free_eofblocks(mp, ip, true); |
| 1655 | if (error && error != EAGAIN) |
| 1656 | return error; |
| 1657 | |
| 1658 | /* delalloc blocks after truncation means it really is dirty */ |
| 1659 | if (ip->i_delayed_blks) |
| 1660 | xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); |
| 1661 | } |
| 1662 | return 0; |
| 1663 | } |
| 1664 | |
| 1665 | /* |
| 1666 | * xfs_inactive |
| 1667 | * |
| 1668 | * This is called when the vnode reference count for the vnode |
| 1669 | * goes to zero. If the file has been unlinked, then it must |
| 1670 | * now be truncated. Also, we clear all of the read-ahead state |
| 1671 | * kept for the inode here since the file is now closed. |
| 1672 | */ |
| 1673 | int |
| 1674 | xfs_inactive( |
| 1675 | xfs_inode_t *ip) |
| 1676 | { |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1677 | xfs_bmap_free_t free_list; |
| 1678 | xfs_fsblock_t first_block; |
| 1679 | int committed; |
| 1680 | struct xfs_trans *tp; |
| 1681 | struct xfs_mount *mp; |
| 1682 | struct xfs_trans_res *resp; |
| 1683 | int error; |
| 1684 | int truncate = 0; |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1685 | |
| 1686 | /* |
| 1687 | * If the inode is already free, then there can be nothing |
| 1688 | * to clean up here. |
| 1689 | */ |
Ben Myers | d948709 | 2013-09-10 18:11:22 -0500 | [diff] [blame] | 1690 | if (ip->i_d.di_mode == 0) { |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1691 | ASSERT(ip->i_df.if_real_bytes == 0); |
| 1692 | ASSERT(ip->i_df.if_broot_bytes == 0); |
| 1693 | return VN_INACTIVE_CACHE; |
| 1694 | } |
| 1695 | |
| 1696 | mp = ip->i_mount; |
| 1697 | |
| 1698 | error = 0; |
| 1699 | |
| 1700 | /* If this is a read-only mount, don't do this (would generate I/O) */ |
| 1701 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
| 1702 | goto out; |
| 1703 | |
| 1704 | if (ip->i_d.di_nlink != 0) { |
| 1705 | /* |
| 1706 | * force is true because we are evicting an inode from the |
| 1707 | * cache. Post-eof blocks must be freed, lest we end up with |
| 1708 | * broken free space accounting. |
| 1709 | */ |
| 1710 | if (xfs_can_free_eofblocks(ip, true)) { |
| 1711 | error = xfs_free_eofblocks(mp, ip, false); |
| 1712 | if (error) |
| 1713 | return VN_INACTIVE_CACHE; |
| 1714 | } |
| 1715 | goto out; |
| 1716 | } |
| 1717 | |
| 1718 | if (S_ISREG(ip->i_d.di_mode) && |
| 1719 | (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || |
| 1720 | ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) |
| 1721 | truncate = 1; |
| 1722 | |
| 1723 | error = xfs_qm_dqattach(ip, 0); |
| 1724 | if (error) |
| 1725 | return VN_INACTIVE_CACHE; |
| 1726 | |
Brian Foster | 36b21dd | 2013-09-20 11:06:09 -0400 | [diff] [blame^] | 1727 | if (S_ISLNK(ip->i_d.di_mode)) { |
| 1728 | error = xfs_inactive_symlink(ip); |
| 1729 | if (error) |
| 1730 | goto out; |
| 1731 | } |
| 1732 | |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1733 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); |
Brian Foster | 36b21dd | 2013-09-20 11:06:09 -0400 | [diff] [blame^] | 1734 | resp = truncate ? &M_RES(mp)->tr_itruncate : &M_RES(mp)->tr_ifree; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1735 | |
| 1736 | error = xfs_trans_reserve(tp, resp, 0, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1737 | if (error) { |
| 1738 | ASSERT(XFS_FORCED_SHUTDOWN(mp)); |
| 1739 | xfs_trans_cancel(tp, 0); |
| 1740 | return VN_INACTIVE_CACHE; |
| 1741 | } |
| 1742 | |
| 1743 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 1744 | xfs_trans_ijoin(tp, ip, 0); |
| 1745 | |
Brian Foster | 36b21dd | 2013-09-20 11:06:09 -0400 | [diff] [blame^] | 1746 | if (truncate) { |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1747 | ip->i_d.di_size = 0; |
| 1748 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 1749 | |
| 1750 | error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); |
| 1751 | if (error) |
| 1752 | goto out_cancel; |
| 1753 | |
| 1754 | ASSERT(ip->i_d.di_nextents == 0); |
| 1755 | } |
| 1756 | |
| 1757 | /* |
| 1758 | * If there are attributes associated with the file then blow them away |
| 1759 | * now. The code calls a routine that recursively deconstructs the |
| 1760 | * attribute fork. We need to just commit the current transaction |
| 1761 | * because we can't use it for xfs_attr_inactive(). |
| 1762 | */ |
| 1763 | if (ip->i_d.di_anextents > 0) { |
| 1764 | ASSERT(ip->i_d.di_forkoff != 0); |
| 1765 | |
| 1766 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 1767 | if (error) |
| 1768 | goto out_unlock; |
| 1769 | |
| 1770 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 1771 | |
| 1772 | error = xfs_attr_inactive(ip); |
| 1773 | if (error) |
| 1774 | goto out; |
| 1775 | |
| 1776 | tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 1777 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 0, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 1778 | if (error) { |
| 1779 | xfs_trans_cancel(tp, 0); |
| 1780 | goto out; |
| 1781 | } |
| 1782 | |
| 1783 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
| 1784 | xfs_trans_ijoin(tp, ip, 0); |
| 1785 | } |
| 1786 | |
| 1787 | if (ip->i_afp) |
| 1788 | xfs_idestroy_fork(ip, XFS_ATTR_FORK); |
| 1789 | |
| 1790 | ASSERT(ip->i_d.di_anextents == 0); |
| 1791 | |
| 1792 | /* |
| 1793 | * Free the inode. |
| 1794 | */ |
| 1795 | xfs_bmap_init(&free_list, &first_block); |
| 1796 | error = xfs_ifree(tp, ip, &free_list); |
| 1797 | if (error) { |
| 1798 | /* |
| 1799 | * If we fail to free the inode, shut down. The cancel |
| 1800 | * might do that, we need to make sure. Otherwise the |
| 1801 | * inode might be lost for a long time or forever. |
| 1802 | */ |
| 1803 | if (!XFS_FORCED_SHUTDOWN(mp)) { |
| 1804 | xfs_notice(mp, "%s: xfs_ifree returned error %d", |
| 1805 | __func__, error); |
| 1806 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
| 1807 | } |
| 1808 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); |
| 1809 | } else { |
| 1810 | /* |
| 1811 | * Credit the quota account(s). The inode is gone. |
| 1812 | */ |
| 1813 | xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); |
| 1814 | |
| 1815 | /* |
| 1816 | * Just ignore errors at this point. There is nothing we can |
| 1817 | * do except to try to keep going. Make sure it's not a silent |
| 1818 | * error. |
| 1819 | */ |
| 1820 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
| 1821 | if (error) |
| 1822 | xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", |
| 1823 | __func__, error); |
| 1824 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 1825 | if (error) |
| 1826 | xfs_notice(mp, "%s: xfs_trans_commit returned error %d", |
| 1827 | __func__, error); |
| 1828 | } |
| 1829 | |
| 1830 | /* |
| 1831 | * Release the dquots held by inode, if any. |
| 1832 | */ |
| 1833 | xfs_qm_dqdetach(ip); |
| 1834 | out_unlock: |
| 1835 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 1836 | out: |
| 1837 | return VN_INACTIVE_CACHE; |
| 1838 | out_cancel: |
| 1839 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
| 1840 | goto out_unlock; |
| 1841 | } |
| 1842 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1843 | /* |
| 1844 | * This is called when the inode's link count goes to 0. |
| 1845 | * We place the on-disk inode on a list in the AGI. It |
| 1846 | * will be pulled from this list when the inode is freed. |
| 1847 | */ |
| 1848 | int |
| 1849 | xfs_iunlink( |
| 1850 | xfs_trans_t *tp, |
| 1851 | xfs_inode_t *ip) |
| 1852 | { |
| 1853 | xfs_mount_t *mp; |
| 1854 | xfs_agi_t *agi; |
| 1855 | xfs_dinode_t *dip; |
| 1856 | xfs_buf_t *agibp; |
| 1857 | xfs_buf_t *ibp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1858 | xfs_agino_t agino; |
| 1859 | short bucket_index; |
| 1860 | int offset; |
| 1861 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1862 | |
| 1863 | ASSERT(ip->i_d.di_nlink == 0); |
| 1864 | ASSERT(ip->i_d.di_mode != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1865 | |
| 1866 | mp = tp->t_mountp; |
| 1867 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1868 | /* |
| 1869 | * Get the agi buffer first. It ensures lock ordering |
| 1870 | * on the list. |
| 1871 | */ |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 1872 | error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); |
Vlad Apostolov | 859d718 | 2007-10-11 17:44:18 +1000 | [diff] [blame] | 1873 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1874 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1875 | agi = XFS_BUF_TO_AGI(agibp); |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 1876 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | /* |
| 1878 | * Get the index into the agi hash table for the |
| 1879 | * list this inode will go on. |
| 1880 | */ |
| 1881 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
| 1882 | ASSERT(agino != 0); |
| 1883 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; |
| 1884 | ASSERT(agi->agi_unlinked[bucket_index]); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 1885 | ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1886 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 1887 | if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1888 | /* |
| 1889 | * There is already another inode in the bucket we need |
| 1890 | * to add ourselves to. Add us at the front of the list. |
| 1891 | * Here we put the head pointer into our next pointer, |
| 1892 | * and then we fall through to point the head at us. |
| 1893 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 1894 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
| 1895 | 0, 0); |
Vlad Apostolov | c319b58 | 2007-11-23 16:27:51 +1100 | [diff] [blame] | 1896 | if (error) |
| 1897 | return error; |
| 1898 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 1899 | ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; |
Christoph Hellwig | 92bfc6e | 2008-11-28 14:23:41 +1100 | [diff] [blame] | 1901 | offset = ip->i_imap.im_boffset + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1902 | offsetof(xfs_dinode_t, di_next_unlinked); |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 1903 | |
| 1904 | /* need to recalc the inode CRC if appropriate */ |
| 1905 | xfs_dinode_calc_crc(mp, dip); |
| 1906 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1907 | xfs_trans_inode_buf(tp, ibp); |
| 1908 | xfs_trans_log_buf(tp, ibp, offset, |
| 1909 | (offset + sizeof(xfs_agino_t) - 1)); |
| 1910 | xfs_inobp_check(mp, ibp); |
| 1911 | } |
| 1912 | |
| 1913 | /* |
| 1914 | * Point the bucket head pointer at the inode being inserted. |
| 1915 | */ |
| 1916 | ASSERT(agino != 0); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 1917 | agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1918 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
| 1919 | (sizeof(xfs_agino_t) * bucket_index); |
| 1920 | xfs_trans_log_buf(tp, agibp, offset, |
| 1921 | (offset + sizeof(xfs_agino_t) - 1)); |
| 1922 | return 0; |
| 1923 | } |
| 1924 | |
| 1925 | /* |
| 1926 | * Pull the on-disk inode from the AGI unlinked list. |
| 1927 | */ |
| 1928 | STATIC int |
| 1929 | xfs_iunlink_remove( |
| 1930 | xfs_trans_t *tp, |
| 1931 | xfs_inode_t *ip) |
| 1932 | { |
| 1933 | xfs_ino_t next_ino; |
| 1934 | xfs_mount_t *mp; |
| 1935 | xfs_agi_t *agi; |
| 1936 | xfs_dinode_t *dip; |
| 1937 | xfs_buf_t *agibp; |
| 1938 | xfs_buf_t *ibp; |
| 1939 | xfs_agnumber_t agno; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1940 | xfs_agino_t agino; |
| 1941 | xfs_agino_t next_agino; |
| 1942 | xfs_buf_t *last_ibp; |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1943 | xfs_dinode_t *last_dip = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1944 | short bucket_index; |
Nathan Scott | 6fdf8cc | 2006-06-28 10:13:52 +1000 | [diff] [blame] | 1945 | int offset, last_offset = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1947 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1948 | mp = tp->t_mountp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1949 | agno = XFS_INO_TO_AGNO(mp, ip->i_ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1950 | |
| 1951 | /* |
| 1952 | * Get the agi buffer first. It ensures lock ordering |
| 1953 | * on the list. |
| 1954 | */ |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 1955 | error = xfs_read_agi(mp, tp, agno, &agibp); |
| 1956 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1957 | return error; |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 1958 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1959 | agi = XFS_BUF_TO_AGI(agibp); |
Christoph Hellwig | 5e1be0f | 2008-11-28 14:23:37 +1100 | [diff] [blame] | 1960 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1961 | /* |
| 1962 | * Get the index into the agi hash table for the |
| 1963 | * list this inode will go on. |
| 1964 | */ |
| 1965 | agino = XFS_INO_TO_AGINO(mp, ip->i_ino); |
| 1966 | ASSERT(agino != 0); |
| 1967 | bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 1968 | ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1969 | ASSERT(agi->agi_unlinked[bucket_index]); |
| 1970 | |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 1971 | if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1972 | /* |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 1973 | * We're at the head of the list. Get the inode's on-disk |
| 1974 | * buffer to see if there is anyone after us on the list. |
| 1975 | * Only modify our next pointer if it is not already NULLAGINO. |
| 1976 | * This saves us the overhead of dealing with the buffer when |
| 1977 | * there is no need to change it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1978 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 1979 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
| 1980 | 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1981 | if (error) { |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 1982 | xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 1983 | __func__, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1984 | return error; |
| 1985 | } |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 1986 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | ASSERT(next_agino != 0); |
| 1988 | if (next_agino != NULLAGINO) { |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 1989 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
Christoph Hellwig | 92bfc6e | 2008-11-28 14:23:41 +1100 | [diff] [blame] | 1990 | offset = ip->i_imap.im_boffset + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1991 | offsetof(xfs_dinode_t, di_next_unlinked); |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 1992 | |
| 1993 | /* need to recalc the inode CRC if appropriate */ |
| 1994 | xfs_dinode_calc_crc(mp, dip); |
| 1995 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | xfs_trans_inode_buf(tp, ibp); |
| 1997 | xfs_trans_log_buf(tp, ibp, offset, |
| 1998 | (offset + sizeof(xfs_agino_t) - 1)); |
| 1999 | xfs_inobp_check(mp, ibp); |
| 2000 | } else { |
| 2001 | xfs_trans_brelse(tp, ibp); |
| 2002 | } |
| 2003 | /* |
| 2004 | * Point the bucket head pointer at the next inode. |
| 2005 | */ |
| 2006 | ASSERT(next_agino != 0); |
| 2007 | ASSERT(next_agino != agino); |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 2008 | agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2009 | offset = offsetof(xfs_agi_t, agi_unlinked) + |
| 2010 | (sizeof(xfs_agino_t) * bucket_index); |
| 2011 | xfs_trans_log_buf(tp, agibp, offset, |
| 2012 | (offset + sizeof(xfs_agino_t) - 1)); |
| 2013 | } else { |
| 2014 | /* |
| 2015 | * We need to search the list for the inode being freed. |
| 2016 | */ |
Christoph Hellwig | 16259e7 | 2005-11-02 15:11:25 +1100 | [diff] [blame] | 2017 | next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2018 | last_ibp = NULL; |
| 2019 | while (next_agino != agino) { |
Christoph Hellwig | 129dbc9 | 2012-07-03 12:21:51 -0400 | [diff] [blame] | 2020 | struct xfs_imap imap; |
| 2021 | |
| 2022 | if (last_ibp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2023 | xfs_trans_brelse(tp, last_ibp); |
Christoph Hellwig | 129dbc9 | 2012-07-03 12:21:51 -0400 | [diff] [blame] | 2024 | |
| 2025 | imap.im_blkno = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2026 | next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); |
Christoph Hellwig | 129dbc9 | 2012-07-03 12:21:51 -0400 | [diff] [blame] | 2027 | |
| 2028 | error = xfs_imap(mp, tp, next_ino, &imap, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2029 | if (error) { |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 2030 | xfs_warn(mp, |
Christoph Hellwig | 129dbc9 | 2012-07-03 12:21:51 -0400 | [diff] [blame] | 2031 | "%s: xfs_imap returned error %d.", |
| 2032 | __func__, error); |
| 2033 | return error; |
| 2034 | } |
| 2035 | |
| 2036 | error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, |
| 2037 | &last_ibp, 0, 0); |
| 2038 | if (error) { |
| 2039 | xfs_warn(mp, |
| 2040 | "%s: xfs_imap_to_bp returned error %d.", |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 2041 | __func__, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | return error; |
| 2043 | } |
Christoph Hellwig | 129dbc9 | 2012-07-03 12:21:51 -0400 | [diff] [blame] | 2044 | |
| 2045 | last_offset = imap.im_boffset; |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2046 | next_agino = be32_to_cpu(last_dip->di_next_unlinked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2047 | ASSERT(next_agino != NULLAGINO); |
| 2048 | ASSERT(next_agino != 0); |
| 2049 | } |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 2050 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2051 | /* |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 2052 | * Now last_ibp points to the buffer previous to us on the |
| 2053 | * unlinked list. Pull us from the list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 2055 | error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, |
| 2056 | 0, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2057 | if (error) { |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 2058 | xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", |
Dave Chinner | 0b932cc | 2011-03-07 10:08:35 +1100 | [diff] [blame] | 2059 | __func__, error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2060 | return error; |
| 2061 | } |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2062 | next_agino = be32_to_cpu(dip->di_next_unlinked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2063 | ASSERT(next_agino != 0); |
| 2064 | ASSERT(next_agino != agino); |
| 2065 | if (next_agino != NULLAGINO) { |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2066 | dip->di_next_unlinked = cpu_to_be32(NULLAGINO); |
Christoph Hellwig | 92bfc6e | 2008-11-28 14:23:41 +1100 | [diff] [blame] | 2067 | offset = ip->i_imap.im_boffset + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2068 | offsetof(xfs_dinode_t, di_next_unlinked); |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2069 | |
| 2070 | /* need to recalc the inode CRC if appropriate */ |
| 2071 | xfs_dinode_calc_crc(mp, dip); |
| 2072 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2073 | xfs_trans_inode_buf(tp, ibp); |
| 2074 | xfs_trans_log_buf(tp, ibp, offset, |
| 2075 | (offset + sizeof(xfs_agino_t) - 1)); |
| 2076 | xfs_inobp_check(mp, ibp); |
| 2077 | } else { |
| 2078 | xfs_trans_brelse(tp, ibp); |
| 2079 | } |
| 2080 | /* |
| 2081 | * Point the previous inode on the list to the next inode. |
| 2082 | */ |
Christoph Hellwig | 347d1c0 | 2007-08-28 13:57:51 +1000 | [diff] [blame] | 2083 | last_dip->di_next_unlinked = cpu_to_be32(next_agino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2084 | ASSERT(next_agino != 0); |
| 2085 | offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); |
Dave Chinner | 0a32c26 | 2013-06-05 12:09:08 +1000 | [diff] [blame] | 2086 | |
| 2087 | /* need to recalc the inode CRC if appropriate */ |
| 2088 | xfs_dinode_calc_crc(mp, last_dip); |
| 2089 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | xfs_trans_inode_buf(tp, last_ibp); |
| 2091 | xfs_trans_log_buf(tp, last_ibp, offset, |
| 2092 | (offset + sizeof(xfs_agino_t) - 1)); |
| 2093 | xfs_inobp_check(mp, last_ibp); |
| 2094 | } |
| 2095 | return 0; |
| 2096 | } |
| 2097 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2098 | /* |
Zhi Yong Wu | 0b8182d | 2013-08-12 03:14:59 +0000 | [diff] [blame] | 2099 | * A big issue when freeing the inode cluster is that we _cannot_ skip any |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2100 | * inodes that are in memory - they all must be marked stale and attached to |
| 2101 | * the cluster buffer. |
| 2102 | */ |
Chandra Seetharaman | 2a30f36d | 2011-09-20 13:56:55 +0000 | [diff] [blame] | 2103 | STATIC int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2104 | xfs_ifree_cluster( |
| 2105 | xfs_inode_t *free_ip, |
| 2106 | xfs_trans_t *tp, |
| 2107 | xfs_ino_t inum) |
| 2108 | { |
| 2109 | xfs_mount_t *mp = free_ip->i_mount; |
| 2110 | int blks_per_cluster; |
| 2111 | int nbufs; |
| 2112 | int ninodes; |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2113 | int i, j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | xfs_daddr_t blkno; |
| 2115 | xfs_buf_t *bp; |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2116 | xfs_inode_t *ip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2117 | xfs_inode_log_item_t *iip; |
| 2118 | xfs_log_item_t *lip; |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 2119 | struct xfs_perag *pag; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2120 | |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 2121 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2122 | if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { |
| 2123 | blks_per_cluster = 1; |
| 2124 | ninodes = mp->m_sb.sb_inopblock; |
| 2125 | nbufs = XFS_IALLOC_BLOCKS(mp); |
| 2126 | } else { |
| 2127 | blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / |
| 2128 | mp->m_sb.sb_blocksize; |
| 2129 | ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; |
| 2130 | nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; |
| 2131 | } |
| 2132 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2133 | for (j = 0; j < nbufs; j++, inum += ninodes) { |
| 2134 | blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), |
| 2135 | XFS_INO_TO_AGBNO(mp, inum)); |
| 2136 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2137 | /* |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2138 | * We obtain and lock the backing buffer first in the process |
| 2139 | * here, as we have to ensure that any dirty inode that we |
| 2140 | * can't get the flush lock on is attached to the buffer. |
| 2141 | * If we scan the in-memory inodes first, then buffer IO can |
| 2142 | * complete before we get a lock on it, and hence we may fail |
| 2143 | * to mark all the active inodes on the buffer stale. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2144 | */ |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2145 | bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, |
Dave Chinner | b6aff29 | 2012-11-02 11:38:42 +1100 | [diff] [blame] | 2146 | mp->m_bsize * blks_per_cluster, |
| 2147 | XBF_UNMAPPED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | |
Chandra Seetharaman | 2a30f36d | 2011-09-20 13:56:55 +0000 | [diff] [blame] | 2149 | if (!bp) |
| 2150 | return ENOMEM; |
Dave Chinner | b0f539d | 2012-11-14 17:53:49 +1100 | [diff] [blame] | 2151 | |
| 2152 | /* |
| 2153 | * This buffer may not have been correctly initialised as we |
| 2154 | * didn't read it from disk. That's not important because we are |
| 2155 | * only using to mark the buffer as stale in the log, and to |
| 2156 | * attach stale cached inodes on it. That means it will never be |
| 2157 | * dispatched for IO. If it is, we want to know about it, and we |
| 2158 | * want it to fail. We can acheive this by adding a write |
| 2159 | * verifier to the buffer. |
| 2160 | */ |
Dave Chinner | 1813dd6 | 2012-11-14 17:54:40 +1100 | [diff] [blame] | 2161 | bp->b_ops = &xfs_inode_buf_ops; |
Dave Chinner | b0f539d | 2012-11-14 17:53:49 +1100 | [diff] [blame] | 2162 | |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2163 | /* |
| 2164 | * Walk the inodes already attached to the buffer and mark them |
| 2165 | * stale. These will all have the flush locks held, so an |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2166 | * in-memory inode walk can't lock them. By marking them all |
| 2167 | * stale first, we will not attempt to lock them in the loop |
| 2168 | * below as the XFS_ISTALE flag will be set. |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2169 | */ |
Christoph Hellwig | adadbee | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2170 | lip = bp->b_fspriv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2171 | while (lip) { |
| 2172 | if (lip->li_type == XFS_LI_INODE) { |
| 2173 | iip = (xfs_inode_log_item_t *)lip; |
| 2174 | ASSERT(iip->ili_logged == 1); |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2175 | lip->li_cb = xfs_istale_done; |
David Chinner | 7b2e2a3 | 2008-10-30 17:39:12 +1100 | [diff] [blame] | 2176 | xfs_trans_ail_copy_lsn(mp->m_ail, |
| 2177 | &iip->ili_flush_lsn, |
| 2178 | &iip->ili_item.li_lsn); |
David Chinner | e5ffd2b | 2006-11-21 18:55:33 +1100 | [diff] [blame] | 2179 | xfs_iflags_set(iip->ili_inode, XFS_ISTALE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2180 | } |
| 2181 | lip = lip->li_bio_list; |
| 2182 | } |
| 2183 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2184 | |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2185 | /* |
| 2186 | * For each inode in memory attempt to add it to the inode |
| 2187 | * buffer and set it up for being staled on buffer IO |
| 2188 | * completion. This is safe as we've locked out tail pushing |
| 2189 | * and flushing by locking the buffer. |
| 2190 | * |
| 2191 | * We have already marked every inode that was part of a |
| 2192 | * transaction stale above, which means there is no point in |
| 2193 | * even trying to lock them. |
| 2194 | */ |
| 2195 | for (i = 0; i < ninodes; i++) { |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2196 | retry: |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2197 | rcu_read_lock(); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2198 | ip = radix_tree_lookup(&pag->pag_ici_root, |
| 2199 | XFS_INO_TO_AGINO(mp, (inum + i))); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2200 | |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2201 | /* Inode not in memory, nothing to do */ |
| 2202 | if (!ip) { |
| 2203 | rcu_read_unlock(); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2204 | continue; |
| 2205 | } |
| 2206 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2207 | /* |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2208 | * because this is an RCU protected lookup, we could |
| 2209 | * find a recently freed or even reallocated inode |
| 2210 | * during the lookup. We need to check under the |
| 2211 | * i_flags_lock for a valid inode here. Skip it if it |
| 2212 | * is not valid, the wrong inode or stale. |
| 2213 | */ |
| 2214 | spin_lock(&ip->i_flags_lock); |
| 2215 | if (ip->i_ino != inum + i || |
| 2216 | __xfs_iflags_test(ip, XFS_ISTALE)) { |
| 2217 | spin_unlock(&ip->i_flags_lock); |
| 2218 | rcu_read_unlock(); |
| 2219 | continue; |
| 2220 | } |
| 2221 | spin_unlock(&ip->i_flags_lock); |
| 2222 | |
| 2223 | /* |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2224 | * Don't try to lock/unlock the current inode, but we |
| 2225 | * _cannot_ skip the other inodes that we did not find |
| 2226 | * in the list attached to the buffer and are not |
| 2227 | * already marked stale. If we can't lock it, back off |
| 2228 | * and retry. |
| 2229 | */ |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2230 | if (ip != free_ip && |
| 2231 | !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2232 | rcu_read_unlock(); |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2233 | delay(1); |
| 2234 | goto retry; |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2235 | } |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2236 | rcu_read_unlock(); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2237 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2238 | xfs_iflock(ip); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2239 | xfs_iflags_set(ip, XFS_ISTALE); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2240 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2241 | /* |
| 2242 | * we don't need to attach clean inodes or those only |
| 2243 | * with unlogged changes (which we throw away, anyway). |
| 2244 | */ |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2245 | iip = ip->i_itemp; |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2246 | if (!iip || xfs_inode_clean(ip)) { |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2247 | ASSERT(ip != free_ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2248 | xfs_ifunlock(ip); |
| 2249 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
| 2250 | continue; |
| 2251 | } |
| 2252 | |
Christoph Hellwig | f5d8d5c | 2012-02-29 09:53:54 +0000 | [diff] [blame] | 2253 | iip->ili_last_fields = iip->ili_fields; |
| 2254 | iip->ili_fields = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2255 | iip->ili_logged = 1; |
David Chinner | 7b2e2a3 | 2008-10-30 17:39:12 +1100 | [diff] [blame] | 2256 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
| 2257 | &iip->ili_item.li_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2258 | |
Christoph Hellwig | ca30b2a | 2010-06-23 18:11:15 +1000 | [diff] [blame] | 2259 | xfs_buf_attach_iodone(bp, xfs_istale_done, |
| 2260 | &iip->ili_item); |
Dave Chinner | 5b257b4 | 2010-06-03 16:22:29 +1000 | [diff] [blame] | 2261 | |
| 2262 | if (ip != free_ip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2263 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2264 | } |
| 2265 | |
Dave Chinner | 5b3eed7 | 2010-08-24 11:42:41 +1000 | [diff] [blame] | 2266 | xfs_trans_stale_inode_buf(tp, bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2267 | xfs_trans_binval(tp, bp); |
| 2268 | } |
| 2269 | |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 2270 | xfs_perag_put(pag); |
Chandra Seetharaman | 2a30f36d | 2011-09-20 13:56:55 +0000 | [diff] [blame] | 2271 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2272 | } |
| 2273 | |
| 2274 | /* |
| 2275 | * This is called to return an inode to the inode free list. |
| 2276 | * The inode should already be truncated to 0 length and have |
| 2277 | * no pages associated with it. This routine also assumes that |
| 2278 | * the inode is already a part of the transaction. |
| 2279 | * |
| 2280 | * The on-disk copy of the inode will have been added to the list |
| 2281 | * of unlinked inodes in the AGI. We need to remove the inode from |
| 2282 | * that list atomically with respect to freeing it here. |
| 2283 | */ |
| 2284 | int |
| 2285 | xfs_ifree( |
| 2286 | xfs_trans_t *tp, |
| 2287 | xfs_inode_t *ip, |
| 2288 | xfs_bmap_free_t *flist) |
| 2289 | { |
| 2290 | int error; |
| 2291 | int delete; |
| 2292 | xfs_ino_t first_ino; |
| 2293 | |
Christoph Hellwig | 579aa9c | 2008-04-22 17:34:00 +1000 | [diff] [blame] | 2294 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2295 | ASSERT(ip->i_d.di_nlink == 0); |
| 2296 | ASSERT(ip->i_d.di_nextents == 0); |
| 2297 | ASSERT(ip->i_d.di_anextents == 0); |
Christoph Hellwig | ce7ae151 | 2011-12-18 20:00:11 +0000 | [diff] [blame] | 2298 | ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2299 | ASSERT(ip->i_d.di_nblocks == 0); |
| 2300 | |
| 2301 | /* |
| 2302 | * Pull the on-disk inode from the AGI unlinked list. |
| 2303 | */ |
| 2304 | error = xfs_iunlink_remove(tp, ip); |
Dave Chinner | 1baaed8 | 2013-06-27 16:04:50 +1000 | [diff] [blame] | 2305 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2306 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2307 | |
| 2308 | error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); |
Dave Chinner | 1baaed8 | 2013-06-27 16:04:50 +1000 | [diff] [blame] | 2309 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2310 | return error; |
Dave Chinner | 1baaed8 | 2013-06-27 16:04:50 +1000 | [diff] [blame] | 2311 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | ip->i_d.di_mode = 0; /* mark incore inode as free */ |
| 2313 | ip->i_d.di_flags = 0; |
| 2314 | ip->i_d.di_dmevmask = 0; |
| 2315 | ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2316 | ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; |
| 2317 | ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; |
| 2318 | /* |
| 2319 | * Bump the generation count so no one will be confused |
| 2320 | * by reincarnations of this inode. |
| 2321 | */ |
| 2322 | ip->i_d.di_gen++; |
| 2323 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); |
| 2324 | |
Dave Chinner | 1baaed8 | 2013-06-27 16:04:50 +1000 | [diff] [blame] | 2325 | if (delete) |
Chandra Seetharaman | 2a30f36d | 2011-09-20 13:56:55 +0000 | [diff] [blame] | 2326 | error = xfs_ifree_cluster(ip, tp, first_ino); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2327 | |
Chandra Seetharaman | 2a30f36d | 2011-09-20 13:56:55 +0000 | [diff] [blame] | 2328 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2329 | } |
| 2330 | |
| 2331 | /* |
Christoph Hellwig | 60ec6783 | 2010-02-17 19:43:56 +0000 | [diff] [blame] | 2332 | * This is called to unpin an inode. The caller must have the inode locked |
| 2333 | * in at least shared mode so that the buffer cannot be subsequently pinned |
| 2334 | * once someone is waiting for it to be unpinned. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2335 | */ |
Christoph Hellwig | 60ec6783 | 2010-02-17 19:43:56 +0000 | [diff] [blame] | 2336 | static void |
Christoph Hellwig | f392e63 | 2011-12-18 20:00:10 +0000 | [diff] [blame] | 2337 | xfs_iunpin( |
Christoph Hellwig | 60ec6783 | 2010-02-17 19:43:56 +0000 | [diff] [blame] | 2338 | struct xfs_inode *ip) |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 2339 | { |
Christoph Hellwig | 579aa9c | 2008-04-22 17:34:00 +1000 | [diff] [blame] | 2340 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 2341 | |
Dave Chinner | 4aaf15d | 2010-03-08 11:24:07 +1100 | [diff] [blame] | 2342 | trace_xfs_inode_unpin_nowait(ip, _RET_IP_); |
| 2343 | |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 2344 | /* Give the log a push to start the unpinning I/O */ |
Christoph Hellwig | 60ec6783 | 2010-02-17 19:43:56 +0000 | [diff] [blame] | 2345 | xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 2346 | |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 2347 | } |
| 2348 | |
Christoph Hellwig | f392e63 | 2011-12-18 20:00:10 +0000 | [diff] [blame] | 2349 | static void |
| 2350 | __xfs_iunpin_wait( |
| 2351 | struct xfs_inode *ip) |
| 2352 | { |
| 2353 | wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); |
| 2354 | DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); |
| 2355 | |
| 2356 | xfs_iunpin(ip); |
| 2357 | |
| 2358 | do { |
| 2359 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); |
| 2360 | if (xfs_ipincount(ip)) |
| 2361 | io_schedule(); |
| 2362 | } while (xfs_ipincount(ip)); |
| 2363 | finish_wait(wq, &wait.wait); |
| 2364 | } |
| 2365 | |
Dave Chinner | 777df5a | 2010-02-06 12:37:26 +1100 | [diff] [blame] | 2366 | void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2367 | xfs_iunpin_wait( |
Christoph Hellwig | 60ec6783 | 2010-02-17 19:43:56 +0000 | [diff] [blame] | 2368 | struct xfs_inode *ip) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2369 | { |
Christoph Hellwig | f392e63 | 2011-12-18 20:00:10 +0000 | [diff] [blame] | 2370 | if (xfs_ipincount(ip)) |
| 2371 | __xfs_iunpin_wait(ip); |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 2372 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2373 | |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 2374 | int |
| 2375 | xfs_remove( |
| 2376 | xfs_inode_t *dp, |
| 2377 | struct xfs_name *name, |
| 2378 | xfs_inode_t *ip) |
| 2379 | { |
| 2380 | xfs_mount_t *mp = dp->i_mount; |
| 2381 | xfs_trans_t *tp = NULL; |
| 2382 | int is_dir = S_ISDIR(ip->i_d.di_mode); |
| 2383 | int error = 0; |
| 2384 | xfs_bmap_free_t free_list; |
| 2385 | xfs_fsblock_t first_block; |
| 2386 | int cancel_flags; |
| 2387 | int committed; |
| 2388 | int link_zero; |
| 2389 | uint resblks; |
| 2390 | uint log_count; |
| 2391 | |
| 2392 | trace_xfs_remove(dp, name); |
| 2393 | |
| 2394 | if (XFS_FORCED_SHUTDOWN(mp)) |
| 2395 | return XFS_ERROR(EIO); |
| 2396 | |
| 2397 | error = xfs_qm_dqattach(dp, 0); |
| 2398 | if (error) |
| 2399 | goto std_return; |
| 2400 | |
| 2401 | error = xfs_qm_dqattach(ip, 0); |
| 2402 | if (error) |
| 2403 | goto std_return; |
| 2404 | |
| 2405 | if (is_dir) { |
| 2406 | tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); |
| 2407 | log_count = XFS_DEFAULT_LOG_COUNT; |
| 2408 | } else { |
| 2409 | tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); |
| 2410 | log_count = XFS_REMOVE_LOG_COUNT; |
| 2411 | } |
| 2412 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
| 2413 | |
| 2414 | /* |
| 2415 | * We try to get the real space reservation first, |
| 2416 | * allowing for directory btree deletion(s) implying |
| 2417 | * possible bmap insert(s). If we can't get the space |
| 2418 | * reservation then we use 0 instead, and avoid the bmap |
| 2419 | * btree insert(s) in the directory code by, if the bmap |
| 2420 | * insert tries to happen, instead trimming the LAST |
| 2421 | * block from the directory. |
| 2422 | */ |
| 2423 | resblks = XFS_REMOVE_SPACE_RES(mp); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 2424 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 2425 | if (error == ENOSPC) { |
| 2426 | resblks = 0; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 2427 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0); |
Dave Chinner | c24b5df | 2013-08-12 20:49:45 +1000 | [diff] [blame] | 2428 | } |
| 2429 | if (error) { |
| 2430 | ASSERT(error != ENOSPC); |
| 2431 | cancel_flags = 0; |
| 2432 | goto out_trans_cancel; |
| 2433 | } |
| 2434 | |
| 2435 | xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); |
| 2436 | |
| 2437 | xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); |
| 2438 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
| 2439 | |
| 2440 | /* |
| 2441 | * If we're removing a directory perform some additional validation. |
| 2442 | */ |
| 2443 | if (is_dir) { |
| 2444 | ASSERT(ip->i_d.di_nlink >= 2); |
| 2445 | if (ip->i_d.di_nlink != 2) { |
| 2446 | error = XFS_ERROR(ENOTEMPTY); |
| 2447 | goto out_trans_cancel; |
| 2448 | } |
| 2449 | if (!xfs_dir_isempty(ip)) { |
| 2450 | error = XFS_ERROR(ENOTEMPTY); |
| 2451 | goto out_trans_cancel; |
| 2452 | } |
| 2453 | } |
| 2454 | |
| 2455 | xfs_bmap_init(&free_list, &first_block); |
| 2456 | error = xfs_dir_removename(tp, dp, name, ip->i_ino, |
| 2457 | &first_block, &free_list, resblks); |
| 2458 | if (error) { |
| 2459 | ASSERT(error != ENOENT); |
| 2460 | goto out_bmap_cancel; |
| 2461 | } |
| 2462 | xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 2463 | |
| 2464 | if (is_dir) { |
| 2465 | /* |
| 2466 | * Drop the link from ip's "..". |
| 2467 | */ |
| 2468 | error = xfs_droplink(tp, dp); |
| 2469 | if (error) |
| 2470 | goto out_bmap_cancel; |
| 2471 | |
| 2472 | /* |
| 2473 | * Drop the "." link from ip to self. |
| 2474 | */ |
| 2475 | error = xfs_droplink(tp, ip); |
| 2476 | if (error) |
| 2477 | goto out_bmap_cancel; |
| 2478 | } else { |
| 2479 | /* |
| 2480 | * When removing a non-directory we need to log the parent |
| 2481 | * inode here. For a directory this is done implicitly |
| 2482 | * by the xfs_droplink call for the ".." entry. |
| 2483 | */ |
| 2484 | xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); |
| 2485 | } |
| 2486 | |
| 2487 | /* |
| 2488 | * Drop the link from dp to ip. |
| 2489 | */ |
| 2490 | error = xfs_droplink(tp, ip); |
| 2491 | if (error) |
| 2492 | goto out_bmap_cancel; |
| 2493 | |
| 2494 | /* |
| 2495 | * Determine if this is the last link while |
| 2496 | * we are in the transaction. |
| 2497 | */ |
| 2498 | link_zero = (ip->i_d.di_nlink == 0); |
| 2499 | |
| 2500 | /* |
| 2501 | * If this is a synchronous mount, make sure that the |
| 2502 | * remove transaction goes to disk before returning to |
| 2503 | * the user. |
| 2504 | */ |
| 2505 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) |
| 2506 | xfs_trans_set_sync(tp); |
| 2507 | |
| 2508 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
| 2509 | if (error) |
| 2510 | goto out_bmap_cancel; |
| 2511 | |
| 2512 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 2513 | if (error) |
| 2514 | goto std_return; |
| 2515 | |
| 2516 | /* |
| 2517 | * If we are using filestreams, kill the stream association. |
| 2518 | * If the file is still open it may get a new one but that |
| 2519 | * will get killed on last close in xfs_close() so we don't |
| 2520 | * have to worry about that. |
| 2521 | */ |
| 2522 | if (!is_dir && link_zero && xfs_inode_is_filestream(ip)) |
| 2523 | xfs_filestream_deassociate(ip); |
| 2524 | |
| 2525 | return 0; |
| 2526 | |
| 2527 | out_bmap_cancel: |
| 2528 | xfs_bmap_cancel(&free_list); |
| 2529 | cancel_flags |= XFS_TRANS_ABORT; |
| 2530 | out_trans_cancel: |
| 2531 | xfs_trans_cancel(tp, cancel_flags); |
| 2532 | std_return: |
| 2533 | return error; |
| 2534 | } |
| 2535 | |
Dave Chinner | f6bba20 | 2013-08-12 20:49:46 +1000 | [diff] [blame] | 2536 | /* |
| 2537 | * Enter all inodes for a rename transaction into a sorted array. |
| 2538 | */ |
| 2539 | STATIC void |
| 2540 | xfs_sort_for_rename( |
| 2541 | xfs_inode_t *dp1, /* in: old (source) directory inode */ |
| 2542 | xfs_inode_t *dp2, /* in: new (target) directory inode */ |
| 2543 | xfs_inode_t *ip1, /* in: inode of old entry */ |
| 2544 | xfs_inode_t *ip2, /* in: inode of new entry, if it |
| 2545 | already exists, NULL otherwise. */ |
| 2546 | xfs_inode_t **i_tab,/* out: array of inode returned, sorted */ |
| 2547 | int *num_inodes) /* out: number of inodes in array */ |
| 2548 | { |
| 2549 | xfs_inode_t *temp; |
| 2550 | int i, j; |
| 2551 | |
| 2552 | /* |
| 2553 | * i_tab contains a list of pointers to inodes. We initialize |
| 2554 | * the table here & we'll sort it. We will then use it to |
| 2555 | * order the acquisition of the inode locks. |
| 2556 | * |
| 2557 | * Note that the table may contain duplicates. e.g., dp1 == dp2. |
| 2558 | */ |
| 2559 | i_tab[0] = dp1; |
| 2560 | i_tab[1] = dp2; |
| 2561 | i_tab[2] = ip1; |
| 2562 | if (ip2) { |
| 2563 | *num_inodes = 4; |
| 2564 | i_tab[3] = ip2; |
| 2565 | } else { |
| 2566 | *num_inodes = 3; |
| 2567 | i_tab[3] = NULL; |
| 2568 | } |
| 2569 | |
| 2570 | /* |
| 2571 | * Sort the elements via bubble sort. (Remember, there are at |
| 2572 | * most 4 elements to sort, so this is adequate.) |
| 2573 | */ |
| 2574 | for (i = 0; i < *num_inodes; i++) { |
| 2575 | for (j = 1; j < *num_inodes; j++) { |
| 2576 | if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { |
| 2577 | temp = i_tab[j]; |
| 2578 | i_tab[j] = i_tab[j-1]; |
| 2579 | i_tab[j-1] = temp; |
| 2580 | } |
| 2581 | } |
| 2582 | } |
| 2583 | } |
| 2584 | |
| 2585 | /* |
| 2586 | * xfs_rename |
| 2587 | */ |
| 2588 | int |
| 2589 | xfs_rename( |
| 2590 | xfs_inode_t *src_dp, |
| 2591 | struct xfs_name *src_name, |
| 2592 | xfs_inode_t *src_ip, |
| 2593 | xfs_inode_t *target_dp, |
| 2594 | struct xfs_name *target_name, |
| 2595 | xfs_inode_t *target_ip) |
| 2596 | { |
| 2597 | xfs_trans_t *tp = NULL; |
| 2598 | xfs_mount_t *mp = src_dp->i_mount; |
| 2599 | int new_parent; /* moving to a new dir */ |
| 2600 | int src_is_directory; /* src_name is a directory */ |
| 2601 | int error; |
| 2602 | xfs_bmap_free_t free_list; |
| 2603 | xfs_fsblock_t first_block; |
| 2604 | int cancel_flags; |
| 2605 | int committed; |
| 2606 | xfs_inode_t *inodes[4]; |
| 2607 | int spaceres; |
| 2608 | int num_inodes; |
| 2609 | |
| 2610 | trace_xfs_rename(src_dp, target_dp, src_name, target_name); |
| 2611 | |
| 2612 | new_parent = (src_dp != target_dp); |
| 2613 | src_is_directory = S_ISDIR(src_ip->i_d.di_mode); |
| 2614 | |
| 2615 | xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, |
| 2616 | inodes, &num_inodes); |
| 2617 | |
| 2618 | xfs_bmap_init(&free_list, &first_block); |
| 2619 | tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME); |
| 2620 | cancel_flags = XFS_TRANS_RELEASE_LOG_RES; |
| 2621 | spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 2622 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0); |
Dave Chinner | f6bba20 | 2013-08-12 20:49:46 +1000 | [diff] [blame] | 2623 | if (error == ENOSPC) { |
| 2624 | spaceres = 0; |
Jie Liu | 3d3c8b5 | 2013-08-12 20:49:59 +1000 | [diff] [blame] | 2625 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0); |
Dave Chinner | f6bba20 | 2013-08-12 20:49:46 +1000 | [diff] [blame] | 2626 | } |
| 2627 | if (error) { |
| 2628 | xfs_trans_cancel(tp, 0); |
| 2629 | goto std_return; |
| 2630 | } |
| 2631 | |
| 2632 | /* |
| 2633 | * Attach the dquots to the inodes |
| 2634 | */ |
| 2635 | error = xfs_qm_vop_rename_dqattach(inodes); |
| 2636 | if (error) { |
| 2637 | xfs_trans_cancel(tp, cancel_flags); |
| 2638 | goto std_return; |
| 2639 | } |
| 2640 | |
| 2641 | /* |
| 2642 | * Lock all the participating inodes. Depending upon whether |
| 2643 | * the target_name exists in the target directory, and |
| 2644 | * whether the target directory is the same as the source |
| 2645 | * directory, we can lock from 2 to 4 inodes. |
| 2646 | */ |
| 2647 | xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); |
| 2648 | |
| 2649 | /* |
| 2650 | * Join all the inodes to the transaction. From this point on, |
| 2651 | * we can rely on either trans_commit or trans_cancel to unlock |
| 2652 | * them. |
| 2653 | */ |
| 2654 | xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); |
| 2655 | if (new_parent) |
| 2656 | xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); |
| 2657 | xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); |
| 2658 | if (target_ip) |
| 2659 | xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); |
| 2660 | |
| 2661 | /* |
| 2662 | * If we are using project inheritance, we only allow renames |
| 2663 | * into our tree when the project IDs are the same; else the |
| 2664 | * tree quota mechanism would be circumvented. |
| 2665 | */ |
| 2666 | if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && |
| 2667 | (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { |
| 2668 | error = XFS_ERROR(EXDEV); |
| 2669 | goto error_return; |
| 2670 | } |
| 2671 | |
| 2672 | /* |
| 2673 | * Set up the target. |
| 2674 | */ |
| 2675 | if (target_ip == NULL) { |
| 2676 | /* |
| 2677 | * If there's no space reservation, check the entry will |
| 2678 | * fit before actually inserting it. |
| 2679 | */ |
| 2680 | error = xfs_dir_canenter(tp, target_dp, target_name, spaceres); |
| 2681 | if (error) |
| 2682 | goto error_return; |
| 2683 | /* |
| 2684 | * If target does not exist and the rename crosses |
| 2685 | * directories, adjust the target directory link count |
| 2686 | * to account for the ".." reference from the new entry. |
| 2687 | */ |
| 2688 | error = xfs_dir_createname(tp, target_dp, target_name, |
| 2689 | src_ip->i_ino, &first_block, |
| 2690 | &free_list, spaceres); |
| 2691 | if (error == ENOSPC) |
| 2692 | goto error_return; |
| 2693 | if (error) |
| 2694 | goto abort_return; |
| 2695 | |
| 2696 | xfs_trans_ichgtime(tp, target_dp, |
| 2697 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 2698 | |
| 2699 | if (new_parent && src_is_directory) { |
| 2700 | error = xfs_bumplink(tp, target_dp); |
| 2701 | if (error) |
| 2702 | goto abort_return; |
| 2703 | } |
| 2704 | } else { /* target_ip != NULL */ |
| 2705 | /* |
| 2706 | * If target exists and it's a directory, check that both |
| 2707 | * target and source are directories and that target can be |
| 2708 | * destroyed, or that neither is a directory. |
| 2709 | */ |
| 2710 | if (S_ISDIR(target_ip->i_d.di_mode)) { |
| 2711 | /* |
| 2712 | * Make sure target dir is empty. |
| 2713 | */ |
| 2714 | if (!(xfs_dir_isempty(target_ip)) || |
| 2715 | (target_ip->i_d.di_nlink > 2)) { |
| 2716 | error = XFS_ERROR(EEXIST); |
| 2717 | goto error_return; |
| 2718 | } |
| 2719 | } |
| 2720 | |
| 2721 | /* |
| 2722 | * Link the source inode under the target name. |
| 2723 | * If the source inode is a directory and we are moving |
| 2724 | * it across directories, its ".." entry will be |
| 2725 | * inconsistent until we replace that down below. |
| 2726 | * |
| 2727 | * In case there is already an entry with the same |
| 2728 | * name at the destination directory, remove it first. |
| 2729 | */ |
| 2730 | error = xfs_dir_replace(tp, target_dp, target_name, |
| 2731 | src_ip->i_ino, |
| 2732 | &first_block, &free_list, spaceres); |
| 2733 | if (error) |
| 2734 | goto abort_return; |
| 2735 | |
| 2736 | xfs_trans_ichgtime(tp, target_dp, |
| 2737 | XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 2738 | |
| 2739 | /* |
| 2740 | * Decrement the link count on the target since the target |
| 2741 | * dir no longer points to it. |
| 2742 | */ |
| 2743 | error = xfs_droplink(tp, target_ip); |
| 2744 | if (error) |
| 2745 | goto abort_return; |
| 2746 | |
| 2747 | if (src_is_directory) { |
| 2748 | /* |
| 2749 | * Drop the link from the old "." entry. |
| 2750 | */ |
| 2751 | error = xfs_droplink(tp, target_ip); |
| 2752 | if (error) |
| 2753 | goto abort_return; |
| 2754 | } |
| 2755 | } /* target_ip != NULL */ |
| 2756 | |
| 2757 | /* |
| 2758 | * Remove the source. |
| 2759 | */ |
| 2760 | if (new_parent && src_is_directory) { |
| 2761 | /* |
| 2762 | * Rewrite the ".." entry to point to the new |
| 2763 | * directory. |
| 2764 | */ |
| 2765 | error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, |
| 2766 | target_dp->i_ino, |
| 2767 | &first_block, &free_list, spaceres); |
| 2768 | ASSERT(error != EEXIST); |
| 2769 | if (error) |
| 2770 | goto abort_return; |
| 2771 | } |
| 2772 | |
| 2773 | /* |
| 2774 | * We always want to hit the ctime on the source inode. |
| 2775 | * |
| 2776 | * This isn't strictly required by the standards since the source |
| 2777 | * inode isn't really being changed, but old unix file systems did |
| 2778 | * it and some incremental backup programs won't work without it. |
| 2779 | */ |
| 2780 | xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); |
| 2781 | xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); |
| 2782 | |
| 2783 | /* |
| 2784 | * Adjust the link count on src_dp. This is necessary when |
| 2785 | * renaming a directory, either within one parent when |
| 2786 | * the target existed, or across two parent directories. |
| 2787 | */ |
| 2788 | if (src_is_directory && (new_parent || target_ip != NULL)) { |
| 2789 | |
| 2790 | /* |
| 2791 | * Decrement link count on src_directory since the |
| 2792 | * entry that's moved no longer points to it. |
| 2793 | */ |
| 2794 | error = xfs_droplink(tp, src_dp); |
| 2795 | if (error) |
| 2796 | goto abort_return; |
| 2797 | } |
| 2798 | |
| 2799 | error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, |
| 2800 | &first_block, &free_list, spaceres); |
| 2801 | if (error) |
| 2802 | goto abort_return; |
| 2803 | |
| 2804 | xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); |
| 2805 | xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); |
| 2806 | if (new_parent) |
| 2807 | xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); |
| 2808 | |
| 2809 | /* |
| 2810 | * If this is a synchronous mount, make sure that the |
| 2811 | * rename transaction goes to disk before returning to |
| 2812 | * the user. |
| 2813 | */ |
| 2814 | if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { |
| 2815 | xfs_trans_set_sync(tp); |
| 2816 | } |
| 2817 | |
| 2818 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
| 2819 | if (error) { |
| 2820 | xfs_bmap_cancel(&free_list); |
| 2821 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | |
| 2822 | XFS_TRANS_ABORT)); |
| 2823 | goto std_return; |
| 2824 | } |
| 2825 | |
| 2826 | /* |
| 2827 | * trans_commit will unlock src_ip, target_ip & decrement |
| 2828 | * the vnode references. |
| 2829 | */ |
| 2830 | return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
| 2831 | |
| 2832 | abort_return: |
| 2833 | cancel_flags |= XFS_TRANS_ABORT; |
| 2834 | error_return: |
| 2835 | xfs_bmap_cancel(&free_list); |
| 2836 | xfs_trans_cancel(tp, cancel_flags); |
| 2837 | std_return: |
| 2838 | return error; |
| 2839 | } |
| 2840 | |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2841 | STATIC int |
| 2842 | xfs_iflush_cluster( |
| 2843 | xfs_inode_t *ip, |
| 2844 | xfs_buf_t *bp) |
| 2845 | { |
| 2846 | xfs_mount_t *mp = ip->i_mount; |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 2847 | struct xfs_perag *pag; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2848 | unsigned long first_index, mask; |
David Chinner | c8f5f12 | 2008-05-20 11:30:15 +1000 | [diff] [blame] | 2849 | unsigned long inodes_per_cluster; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2850 | int ilist_size; |
| 2851 | xfs_inode_t **ilist; |
| 2852 | xfs_inode_t *iq; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2853 | int nr_found; |
| 2854 | int clcount = 0; |
| 2855 | int bufwasdelwri; |
| 2856 | int i; |
| 2857 | |
Dave Chinner | 5017e97 | 2010-01-11 11:47:40 +0000 | [diff] [blame] | 2858 | pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2859 | |
David Chinner | c8f5f12 | 2008-05-20 11:30:15 +1000 | [diff] [blame] | 2860 | inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; |
| 2861 | ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); |
David Chinner | 49383b0 | 2008-05-19 16:29:34 +1000 | [diff] [blame] | 2862 | ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2863 | if (!ilist) |
Dave Chinner | 44b56e0 | 2010-01-11 11:47:43 +0000 | [diff] [blame] | 2864 | goto out_put; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2865 | |
| 2866 | mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); |
| 2867 | first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2868 | rcu_read_lock(); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2869 | /* really need a gang lookup range call here */ |
| 2870 | nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, |
David Chinner | c8f5f12 | 2008-05-20 11:30:15 +1000 | [diff] [blame] | 2871 | first_index, inodes_per_cluster); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2872 | if (nr_found == 0) |
| 2873 | goto out_free; |
| 2874 | |
| 2875 | for (i = 0; i < nr_found; i++) { |
| 2876 | iq = ilist[i]; |
| 2877 | if (iq == ip) |
| 2878 | continue; |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2879 | |
| 2880 | /* |
| 2881 | * because this is an RCU protected lookup, we could find a |
| 2882 | * recently freed or even reallocated inode during the lookup. |
| 2883 | * We need to check under the i_flags_lock for a valid inode |
| 2884 | * here. Skip it if it is not valid or the wrong inode. |
| 2885 | */ |
| 2886 | spin_lock(&ip->i_flags_lock); |
| 2887 | if (!ip->i_ino || |
| 2888 | (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { |
| 2889 | spin_unlock(&ip->i_flags_lock); |
| 2890 | continue; |
| 2891 | } |
| 2892 | spin_unlock(&ip->i_flags_lock); |
| 2893 | |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2894 | /* |
| 2895 | * Do an un-protected check to see if the inode is dirty and |
| 2896 | * is a candidate for flushing. These checks will be repeated |
| 2897 | * later after the appropriate locks are acquired. |
| 2898 | */ |
David Chinner | 3354040 | 2008-03-06 13:43:59 +1100 | [diff] [blame] | 2899 | if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2900 | continue; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2901 | |
| 2902 | /* |
| 2903 | * Try to get locks. If any are unavailable or it is pinned, |
| 2904 | * then this inode cannot be flushed and is skipped. |
| 2905 | */ |
| 2906 | |
| 2907 | if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) |
| 2908 | continue; |
| 2909 | if (!xfs_iflock_nowait(iq)) { |
| 2910 | xfs_iunlock(iq, XFS_ILOCK_SHARED); |
| 2911 | continue; |
| 2912 | } |
| 2913 | if (xfs_ipincount(iq)) { |
| 2914 | xfs_ifunlock(iq); |
| 2915 | xfs_iunlock(iq, XFS_ILOCK_SHARED); |
| 2916 | continue; |
| 2917 | } |
| 2918 | |
| 2919 | /* |
| 2920 | * arriving here means that this inode can be flushed. First |
| 2921 | * re-check that it's dirty before flushing. |
| 2922 | */ |
David Chinner | 3354040 | 2008-03-06 13:43:59 +1100 | [diff] [blame] | 2923 | if (!xfs_inode_clean(iq)) { |
| 2924 | int error; |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2925 | error = xfs_iflush_int(iq, bp); |
| 2926 | if (error) { |
| 2927 | xfs_iunlock(iq, XFS_ILOCK_SHARED); |
| 2928 | goto cluster_corrupt_out; |
| 2929 | } |
| 2930 | clcount++; |
| 2931 | } else { |
| 2932 | xfs_ifunlock(iq); |
| 2933 | } |
| 2934 | xfs_iunlock(iq, XFS_ILOCK_SHARED); |
| 2935 | } |
| 2936 | |
| 2937 | if (clcount) { |
| 2938 | XFS_STATS_INC(xs_icluster_flushcnt); |
| 2939 | XFS_STATS_ADD(xs_icluster_flushinode, clcount); |
| 2940 | } |
| 2941 | |
| 2942 | out_free: |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2943 | rcu_read_unlock(); |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 2944 | kmem_free(ilist); |
Dave Chinner | 44b56e0 | 2010-01-11 11:47:43 +0000 | [diff] [blame] | 2945 | out_put: |
| 2946 | xfs_perag_put(pag); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2947 | return 0; |
| 2948 | |
| 2949 | |
| 2950 | cluster_corrupt_out: |
| 2951 | /* |
| 2952 | * Corruption detected in the clustering loop. Invalidate the |
| 2953 | * inode buffer and shut down the filesystem. |
| 2954 | */ |
Dave Chinner | 1a3e8f3 | 2010-12-17 17:29:43 +1100 | [diff] [blame] | 2955 | rcu_read_unlock(); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2956 | /* |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2957 | * Clean up the buffer. If it was delwri, just release it -- |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2958 | * brelse can handle it with no problems. If not, shut down the |
| 2959 | * filesystem before releasing the buffer. |
| 2960 | */ |
Christoph Hellwig | 43ff212 | 2012-04-23 15:58:39 +1000 | [diff] [blame] | 2961 | bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2962 | if (bufwasdelwri) |
| 2963 | xfs_buf_relse(bp); |
| 2964 | |
| 2965 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
| 2966 | |
| 2967 | if (!bufwasdelwri) { |
| 2968 | /* |
| 2969 | * Just like incore_relse: if we have b_iodone functions, |
| 2970 | * mark the buffer as an error and call them. Otherwise |
| 2971 | * mark it as stale and brelse. |
| 2972 | */ |
Christoph Hellwig | cb669ca | 2011-07-13 13:43:49 +0200 | [diff] [blame] | 2973 | if (bp->b_iodone) { |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2974 | XFS_BUF_UNDONE(bp); |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 2975 | xfs_buf_stale(bp); |
Chandra Seetharaman | 5a52c2a58 | 2011-07-22 23:39:51 +0000 | [diff] [blame] | 2976 | xfs_buf_ioerror(bp, EIO); |
Christoph Hellwig | 1a1a3e9 | 2010-10-06 18:41:18 +0000 | [diff] [blame] | 2977 | xfs_buf_ioend(bp, 0); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2978 | } else { |
Christoph Hellwig | c867cb6 | 2011-10-10 16:52:46 +0000 | [diff] [blame] | 2979 | xfs_buf_stale(bp); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2980 | xfs_buf_relse(bp); |
| 2981 | } |
| 2982 | } |
| 2983 | |
| 2984 | /* |
| 2985 | * Unlocks the flush lock |
| 2986 | */ |
Dave Chinner | 04913fd | 2012-04-23 15:58:41 +1000 | [diff] [blame] | 2987 | xfs_iflush_abort(iq, false); |
Denys Vlasenko | f0e2d93 | 2008-05-19 16:31:57 +1000 | [diff] [blame] | 2988 | kmem_free(ilist); |
Dave Chinner | 44b56e0 | 2010-01-11 11:47:43 +0000 | [diff] [blame] | 2989 | xfs_perag_put(pag); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 2990 | return XFS_ERROR(EFSCORRUPTED); |
| 2991 | } |
| 2992 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2993 | /* |
Christoph Hellwig | 4c46819 | 2012-04-23 15:58:36 +1000 | [diff] [blame] | 2994 | * Flush dirty inode metadata into the backing buffer. |
| 2995 | * |
| 2996 | * The caller must have the inode lock and the inode flush lock held. The |
| 2997 | * inode lock will still be held upon return to the caller, and the inode |
| 2998 | * flush lock will be released after the inode has reached the disk. |
| 2999 | * |
| 3000 | * The caller must write out the buffer returned in *bpp and release it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | */ |
| 3002 | int |
| 3003 | xfs_iflush( |
Christoph Hellwig | 4c46819 | 2012-04-23 15:58:36 +1000 | [diff] [blame] | 3004 | struct xfs_inode *ip, |
| 3005 | struct xfs_buf **bpp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3006 | { |
Christoph Hellwig | 4c46819 | 2012-04-23 15:58:36 +1000 | [diff] [blame] | 3007 | struct xfs_mount *mp = ip->i_mount; |
| 3008 | struct xfs_buf *bp; |
| 3009 | struct xfs_dinode *dip; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | |
| 3012 | XFS_STATS_INC(xs_iflush_count); |
| 3013 | |
Christoph Hellwig | 579aa9c | 2008-04-22 17:34:00 +1000 | [diff] [blame] | 3014 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
Christoph Hellwig | 474fce0 | 2011-12-18 20:00:09 +0000 | [diff] [blame] | 3015 | ASSERT(xfs_isiflocked(ip)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3016 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
Christoph Hellwig | 8096b1e | 2011-12-18 20:00:07 +0000 | [diff] [blame] | 3017 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | |
Christoph Hellwig | 4c46819 | 2012-04-23 15:58:36 +1000 | [diff] [blame] | 3019 | *bpp = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3020 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3021 | xfs_iunpin_wait(ip); |
| 3022 | |
| 3023 | /* |
Dave Chinner | 4b6a468 | 2010-01-11 11:45:21 +0000 | [diff] [blame] | 3024 | * For stale inodes we cannot rely on the backing buffer remaining |
| 3025 | * stale in cache for the remaining life of the stale inode and so |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 3026 | * xfs_imap_to_bp() below may give us a buffer that no longer contains |
Dave Chinner | 4b6a468 | 2010-01-11 11:45:21 +0000 | [diff] [blame] | 3027 | * inodes below. We have to check this after ensuring the inode is |
| 3028 | * unpinned so that it is safe to reclaim the stale inode after the |
| 3029 | * flush call. |
| 3030 | */ |
| 3031 | if (xfs_iflags_test(ip, XFS_ISTALE)) { |
| 3032 | xfs_ifunlock(ip); |
| 3033 | return 0; |
| 3034 | } |
| 3035 | |
| 3036 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3037 | * This may have been unpinned because the filesystem is shutting |
| 3038 | * down forcibly. If that's the case we must not write this inode |
Christoph Hellwig | 32ce90a | 2012-04-23 15:58:32 +1000 | [diff] [blame] | 3039 | * to disk, because the log record didn't make it to disk. |
| 3040 | * |
| 3041 | * We also have to remove the log item from the AIL in this case, |
| 3042 | * as we wait for an empty AIL as part of the unmount process. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3043 | */ |
| 3044 | if (XFS_FORCED_SHUTDOWN(mp)) { |
Christoph Hellwig | 32ce90a | 2012-04-23 15:58:32 +1000 | [diff] [blame] | 3045 | error = XFS_ERROR(EIO); |
| 3046 | goto abort_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3047 | } |
| 3048 | |
| 3049 | /* |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 3050 | * Get the buffer containing the on-disk inode. |
| 3051 | */ |
Christoph Hellwig | 475ee41 | 2012-07-03 12:21:22 -0400 | [diff] [blame] | 3052 | error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, |
| 3053 | 0); |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 3054 | if (error || !bp) { |
| 3055 | xfs_ifunlock(ip); |
| 3056 | return error; |
| 3057 | } |
| 3058 | |
| 3059 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3060 | * First flush out the inode that xfs_iflush was called with. |
| 3061 | */ |
| 3062 | error = xfs_iflush_int(ip, bp); |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 3063 | if (error) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | goto corrupt_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3065 | |
| 3066 | /* |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 3067 | * If the buffer is pinned then push on the log now so we won't |
| 3068 | * get stuck waiting in the write for too long. |
| 3069 | */ |
Chandra Seetharaman | 811e64c | 2011-07-22 23:40:27 +0000 | [diff] [blame] | 3070 | if (xfs_buf_ispinned(bp)) |
Christoph Hellwig | a14a348 | 2010-01-19 09:56:46 +0000 | [diff] [blame] | 3071 | xfs_log_force(mp, 0); |
David Chinner | a3f74ff | 2008-03-06 13:43:42 +1100 | [diff] [blame] | 3072 | |
| 3073 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | * inode clustering: |
| 3075 | * see if other inodes can be gathered into this write |
| 3076 | */ |
David Chinner | bad5584 | 2008-03-06 13:43:49 +1100 | [diff] [blame] | 3077 | error = xfs_iflush_cluster(ip, bp); |
| 3078 | if (error) |
| 3079 | goto cluster_corrupt_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3080 | |
Christoph Hellwig | 4c46819 | 2012-04-23 15:58:36 +1000 | [diff] [blame] | 3081 | *bpp = bp; |
| 3082 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3083 | |
| 3084 | corrupt_out: |
| 3085 | xfs_buf_relse(bp); |
Nathan Scott | 7d04a33 | 2006-06-09 14:58:38 +1000 | [diff] [blame] | 3086 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3087 | cluster_corrupt_out: |
Christoph Hellwig | 32ce90a | 2012-04-23 15:58:32 +1000 | [diff] [blame] | 3088 | error = XFS_ERROR(EFSCORRUPTED); |
| 3089 | abort_out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3090 | /* |
| 3091 | * Unlocks the flush lock |
| 3092 | */ |
Dave Chinner | 04913fd | 2012-04-23 15:58:41 +1000 | [diff] [blame] | 3093 | xfs_iflush_abort(ip, false); |
Christoph Hellwig | 32ce90a | 2012-04-23 15:58:32 +1000 | [diff] [blame] | 3094 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3095 | } |
| 3096 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3097 | STATIC int |
| 3098 | xfs_iflush_int( |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3099 | struct xfs_inode *ip, |
| 3100 | struct xfs_buf *bp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3101 | { |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3102 | struct xfs_inode_log_item *iip = ip->i_itemp; |
| 3103 | struct xfs_dinode *dip; |
| 3104 | struct xfs_mount *mp = ip->i_mount; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3105 | |
Christoph Hellwig | 579aa9c | 2008-04-22 17:34:00 +1000 | [diff] [blame] | 3106 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); |
Christoph Hellwig | 474fce0 | 2011-12-18 20:00:09 +0000 | [diff] [blame] | 3107 | ASSERT(xfs_isiflocked(ip)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3108 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
Christoph Hellwig | 8096b1e | 2011-12-18 20:00:07 +0000 | [diff] [blame] | 3109 | ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3110 | ASSERT(iip != NULL && iip->ili_fields != 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3111 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3112 | /* set *dip = inode's place in the buffer */ |
Christoph Hellwig | 92bfc6e | 2008-11-28 14:23:41 +1100 | [diff] [blame] | 3113 | dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3114 | |
Christoph Hellwig | 69ef921 | 2011-07-08 14:36:05 +0200 | [diff] [blame] | 3115 | if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3116 | mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3117 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3118 | "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", |
| 3119 | __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3120 | goto corrupt_out; |
| 3121 | } |
| 3122 | if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, |
| 3123 | mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3124 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3125 | "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", |
| 3126 | __func__, ip->i_ino, ip, ip->i_d.di_magic); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3127 | goto corrupt_out; |
| 3128 | } |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 3129 | if (S_ISREG(ip->i_d.di_mode)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3130 | if (XFS_TEST_ERROR( |
| 3131 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3132 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), |
| 3133 | mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3134 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3135 | "%s: Bad regular inode %Lu, ptr 0x%p", |
| 3136 | __func__, ip->i_ino, ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3137 | goto corrupt_out; |
| 3138 | } |
Al Viro | abbede1 | 2011-07-26 02:31:30 -0400 | [diff] [blame] | 3139 | } else if (S_ISDIR(ip->i_d.di_mode)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3140 | if (XFS_TEST_ERROR( |
| 3141 | (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && |
| 3142 | (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && |
| 3143 | (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), |
| 3144 | mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3145 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3146 | "%s: Bad directory inode %Lu, ptr 0x%p", |
| 3147 | __func__, ip->i_ino, ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3148 | goto corrupt_out; |
| 3149 | } |
| 3150 | } |
| 3151 | if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > |
| 3152 | ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, |
| 3153 | XFS_RANDOM_IFLUSH_5)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3154 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3155 | "%s: detected corrupt incore inode %Lu, " |
| 3156 | "total extents = %d, nblocks = %Ld, ptr 0x%p", |
| 3157 | __func__, ip->i_ino, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3158 | ip->i_d.di_nextents + ip->i_d.di_anextents, |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3159 | ip->i_d.di_nblocks, ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3160 | goto corrupt_out; |
| 3161 | } |
| 3162 | if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, |
| 3163 | mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { |
Dave Chinner | 6a19d93 | 2011-03-07 10:02:35 +1100 | [diff] [blame] | 3164 | xfs_alert_tag(mp, XFS_PTAG_IFLUSH, |
| 3165 | "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", |
| 3166 | __func__, ip->i_ino, ip->i_d.di_forkoff, ip); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3167 | goto corrupt_out; |
| 3168 | } |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3169 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3170 | /* |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3171 | * Inode item log recovery for v1/v2 inodes are dependent on the |
| 3172 | * di_flushiter count for correct sequencing. We bump the flush |
| 3173 | * iteration count so we can detect flushes which postdate a log record |
| 3174 | * during recovery. This is redundant as we now log every change and |
| 3175 | * hence this can't happen but we need to still do it to ensure |
| 3176 | * backwards compatibility with old kernels that predate logging all |
| 3177 | * inode changes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3178 | */ |
Dave Chinner | e60896d | 2013-07-24 15:47:30 +1000 | [diff] [blame] | 3179 | if (ip->i_d.di_version < 3) |
| 3180 | ip->i_d.di_flushiter++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3181 | |
| 3182 | /* |
| 3183 | * Copy the dirty parts of the inode into the on-disk |
| 3184 | * inode. We always copy out the core of the inode, |
| 3185 | * because if the inode is dirty at all the core must |
| 3186 | * be. |
| 3187 | */ |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3188 | xfs_dinode_to_disk(dip, &ip->i_d); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3189 | |
| 3190 | /* Wrap, we never let the log put out DI_MAX_FLUSH */ |
| 3191 | if (ip->i_d.di_flushiter == DI_MAX_FLUSH) |
| 3192 | ip->i_d.di_flushiter = 0; |
| 3193 | |
| 3194 | /* |
| 3195 | * If this is really an old format inode and the superblock version |
| 3196 | * has not been updated to support only new format inodes, then |
| 3197 | * convert back to the old inode format. If the superblock version |
| 3198 | * has been updated, then make the conversion permanent. |
| 3199 | */ |
Christoph Hellwig | 51ce16d | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3200 | ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); |
| 3201 | if (ip->i_d.di_version == 1) { |
Eric Sandeen | 6211870 | 2008-03-06 13:44:28 +1100 | [diff] [blame] | 3202 | if (!xfs_sb_version_hasnlink(&mp->m_sb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3203 | /* |
| 3204 | * Convert it back. |
| 3205 | */ |
| 3206 | ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3207 | dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3208 | } else { |
| 3209 | /* |
| 3210 | * The superblock version has already been bumped, |
| 3211 | * so just make the conversion to the new inode |
| 3212 | * format permanent. |
| 3213 | */ |
Christoph Hellwig | 51ce16d | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3214 | ip->i_d.di_version = 2; |
| 3215 | dip->di_version = 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3216 | ip->i_d.di_onlink = 0; |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3217 | dip->di_onlink = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3218 | memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); |
Christoph Hellwig | 81591fe | 2008-11-28 14:23:39 +1100 | [diff] [blame] | 3219 | memset(&(dip->di_pad[0]), 0, |
| 3220 | sizeof(dip->di_pad)); |
Arkadiusz Mi?kiewicz | 6743099 | 2010-09-26 06:10:18 +0000 | [diff] [blame] | 3221 | ASSERT(xfs_get_projid(ip) == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3222 | } |
| 3223 | } |
| 3224 | |
David Chinner | e4ac967 | 2008-04-10 12:23:58 +1000 | [diff] [blame] | 3225 | xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); |
| 3226 | if (XFS_IFORK_Q(ip)) |
| 3227 | xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3228 | xfs_inobp_check(mp, bp); |
| 3229 | |
| 3230 | /* |
Christoph Hellwig | f5d8d5c | 2012-02-29 09:53:54 +0000 | [diff] [blame] | 3231 | * We've recorded everything logged in the inode, so we'd like to clear |
| 3232 | * the ili_fields bits so we don't log and flush things unnecessarily. |
| 3233 | * However, we can't stop logging all this information until the data |
| 3234 | * we've copied into the disk buffer is written to disk. If we did we |
| 3235 | * might overwrite the copy of the inode in the log with all the data |
| 3236 | * after re-logging only part of it, and in the face of a crash we |
| 3237 | * wouldn't have all the data we need to recover. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3238 | * |
Christoph Hellwig | f5d8d5c | 2012-02-29 09:53:54 +0000 | [diff] [blame] | 3239 | * What we do is move the bits to the ili_last_fields field. When |
| 3240 | * logging the inode, these bits are moved back to the ili_fields field. |
| 3241 | * In the xfs_iflush_done() routine we clear ili_last_fields, since we |
| 3242 | * know that the information those bits represent is permanently on |
| 3243 | * disk. As long as the flush completes before the inode is logged |
| 3244 | * again, then both ili_fields and ili_last_fields will be cleared. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | * |
Christoph Hellwig | f5d8d5c | 2012-02-29 09:53:54 +0000 | [diff] [blame] | 3246 | * We can play with the ili_fields bits here, because the inode lock |
| 3247 | * must be held exclusively in order to set bits there and the flush |
| 3248 | * lock protects the ili_last_fields bits. Set ili_logged so the flush |
| 3249 | * done routine can tell whether or not to look in the AIL. Also, store |
| 3250 | * the current LSN of the inode so that we can tell whether the item has |
| 3251 | * moved in the AIL from xfs_iflush_done(). In order to read the lsn we |
| 3252 | * need the AIL lock, because it is a 64 bit value that cannot be read |
| 3253 | * atomically. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3254 | */ |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3255 | iip->ili_last_fields = iip->ili_fields; |
| 3256 | iip->ili_fields = 0; |
| 3257 | iip->ili_logged = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3258 | |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3259 | xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, |
| 3260 | &iip->ili_item.li_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3261 | |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3262 | /* |
| 3263 | * Attach the function xfs_iflush_done to the inode's |
| 3264 | * buffer. This will remove the inode from the AIL |
| 3265 | * and unlock the inode's flush lock when the inode is |
| 3266 | * completely written to disk. |
| 3267 | */ |
| 3268 | xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3269 | |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3270 | /* update the lsn in the on disk inode if required */ |
| 3271 | if (ip->i_d.di_version == 3) |
| 3272 | dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3273 | |
Christoph Hellwig | 93848a9 | 2013-04-03 16:11:17 +1100 | [diff] [blame] | 3274 | /* generate the checksum. */ |
| 3275 | xfs_dinode_calc_crc(mp, dip); |
| 3276 | |
| 3277 | ASSERT(bp->b_fspriv != NULL); |
| 3278 | ASSERT(bp->b_iodone != NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3279 | return 0; |
| 3280 | |
| 3281 | corrupt_out: |
| 3282 | return XFS_ERROR(EFSCORRUPTED); |
| 3283 | } |