1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 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 7 * published by the Free Software Foundation. 8 * 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. 13 * 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 17 */ 18 #include <linux/log2.h> 19 #include <linux/iversion.h> 20 21 #include "xfs.h" 22 #include "xfs_fs.h" 23 #include "xfs_shared.h" 24 #include "xfs_format.h" 25 #include "xfs_log_format.h" 26 #include "xfs_trans_resv.h" 27 #include "xfs_sb.h" 28 #include "xfs_mount.h" 29 #include "xfs_defer.h" 30 #include "xfs_inode.h" 31 #include "xfs_da_format.h" 32 #include "xfs_da_btree.h" 33 #include "xfs_dir2.h" 34 #include "xfs_attr_sf.h" 35 #include "xfs_attr.h" 36 #include "xfs_trans_space.h" 37 #include "xfs_trans.h" 38 #include "xfs_buf_item.h" 39 #include "xfs_inode_item.h" 40 #include "xfs_ialloc.h" 41 #include "xfs_bmap.h" 42 #include "xfs_bmap_util.h" 43 #include "xfs_errortag.h" 44 #include "xfs_error.h" 45 #include "xfs_quota.h" 46 #include "xfs_filestream.h" 47 #include "xfs_cksum.h" 48 #include "xfs_trace.h" 49 #include "xfs_icache.h" 50 #include "xfs_symlink.h" 51 #include "xfs_trans_priv.h" 52 #include "xfs_log.h" 53 #include "xfs_bmap_btree.h" 54 #include "xfs_reflink.h" 55 #include "xfs_dir2_priv.h" 56 57 kmem_zone_t *xfs_inode_zone; 58 59 /* 60 * Used in xfs_itruncate_extents(). This is the maximum number of extents 61 * freed from a file in a single transaction. 62 */ 63 #define XFS_ITRUNC_MAX_EXTENTS 2 64 65 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); 66 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); 67 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); 68 69 /* 70 * helper function to extract extent size hint from inode 71 */ 72 xfs_extlen_t 73 xfs_get_extsz_hint( 74 struct xfs_inode *ip) 75 { 76 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 77 return ip->i_d.di_extsize; 78 if (XFS_IS_REALTIME_INODE(ip)) 79 return ip->i_mount->m_sb.sb_rextsize; 80 return 0; 81 } 82 83 /* 84 * Helper function to extract CoW extent size hint from inode. 85 * Between the extent size hint and the CoW extent size hint, we 86 * return the greater of the two. If the value is zero (automatic), 87 * use the default size. 88 */ 89 xfs_extlen_t 90 xfs_get_cowextsz_hint( 91 struct xfs_inode *ip) 92 { 93 xfs_extlen_t a, b; 94 95 a = 0; 96 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 97 a = ip->i_d.di_cowextsize; 98 b = xfs_get_extsz_hint(ip); 99 100 a = max(a, b); 101 if (a == 0) 102 return XFS_DEFAULT_COWEXTSZ_HINT; 103 return a; 104 } 105 106 /* 107 * These two are wrapper routines around the xfs_ilock() routine used to 108 * centralize some grungy code. They are used in places that wish to lock the 109 * inode solely for reading the extents. The reason these places can't just 110 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to 111 * bringing in of the extents from disk for a file in b-tree format. If the 112 * inode is in b-tree format, then we need to lock the inode exclusively until 113 * the extents are read in. Locking it exclusively all the time would limit 114 * our parallelism unnecessarily, though. What we do instead is check to see 115 * if the extents have been read in yet, and only lock the inode exclusively 116 * if they have not. 117 * 118 * The functions return a value which should be given to the corresponding 119 * xfs_iunlock() call. 120 */ 121 uint 122 xfs_ilock_data_map_shared( 123 struct xfs_inode *ip) 124 { 125 uint lock_mode = XFS_ILOCK_SHARED; 126 127 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 128 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) 129 lock_mode = XFS_ILOCK_EXCL; 130 xfs_ilock(ip, lock_mode); 131 return lock_mode; 132 } 133 134 uint 135 xfs_ilock_attr_map_shared( 136 struct xfs_inode *ip) 137 { 138 uint lock_mode = XFS_ILOCK_SHARED; 139 140 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && 141 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) 142 lock_mode = XFS_ILOCK_EXCL; 143 xfs_ilock(ip, lock_mode); 144 return lock_mode; 145 } 146 147 /* 148 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 149 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows 150 * various combinations of the locks to be obtained. 151 * 152 * The 3 locks should always be ordered so that the IO lock is obtained first, 153 * the mmap lock second and the ilock last in order to prevent deadlock. 154 * 155 * Basic locking order: 156 * 157 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock 158 * 159 * mmap_sem locking order: 160 * 161 * i_rwsem -> page lock -> mmap_sem 162 * mmap_sem -> i_mmap_lock -> page_lock 163 * 164 * The difference in mmap_sem locking order mean that we cannot hold the 165 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can 166 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem 167 * in get_user_pages() to map the user pages into the kernel address space for 168 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because 169 * page faults already hold the mmap_sem. 170 * 171 * Hence to serialise fully against both syscall and mmap based IO, we need to 172 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both 173 * taken in places where we need to invalidate the page cache in a race 174 * free manner (e.g. truncate, hole punch and other extent manipulation 175 * functions). 176 */ 177 void 178 xfs_ilock( 179 xfs_inode_t *ip, 180 uint lock_flags) 181 { 182 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 183 184 /* 185 * You can't set both SHARED and EXCL for the same lock, 186 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 187 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 188 */ 189 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 190 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 191 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 192 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 193 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 194 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 195 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 196 197 if (lock_flags & XFS_IOLOCK_EXCL) { 198 down_write_nested(&VFS_I(ip)->i_rwsem, 199 XFS_IOLOCK_DEP(lock_flags)); 200 } else if (lock_flags & XFS_IOLOCK_SHARED) { 201 down_read_nested(&VFS_I(ip)->i_rwsem, 202 XFS_IOLOCK_DEP(lock_flags)); 203 } 204 205 if (lock_flags & XFS_MMAPLOCK_EXCL) 206 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 207 else if (lock_flags & XFS_MMAPLOCK_SHARED) 208 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 209 210 if (lock_flags & XFS_ILOCK_EXCL) 211 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 212 else if (lock_flags & XFS_ILOCK_SHARED) 213 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 214 } 215 216 /* 217 * This is just like xfs_ilock(), except that the caller 218 * is guaranteed not to sleep. It returns 1 if it gets 219 * the requested locks and 0 otherwise. If the IO lock is 220 * obtained but the inode lock cannot be, then the IO lock 221 * is dropped before returning. 222 * 223 * ip -- the inode being locked 224 * lock_flags -- this parameter indicates the inode's locks to be 225 * to be locked. See the comment for xfs_ilock() for a list 226 * of valid values. 227 */ 228 int 229 xfs_ilock_nowait( 230 xfs_inode_t *ip, 231 uint lock_flags) 232 { 233 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 234 235 /* 236 * You can't set both SHARED and EXCL for the same lock, 237 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 238 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 239 */ 240 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 241 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 242 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 243 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 244 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 245 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 246 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 247 248 if (lock_flags & XFS_IOLOCK_EXCL) { 249 if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) 250 goto out; 251 } else if (lock_flags & XFS_IOLOCK_SHARED) { 252 if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) 253 goto out; 254 } 255 256 if (lock_flags & XFS_MMAPLOCK_EXCL) { 257 if (!mrtryupdate(&ip->i_mmaplock)) 258 goto out_undo_iolock; 259 } else if (lock_flags & XFS_MMAPLOCK_SHARED) { 260 if (!mrtryaccess(&ip->i_mmaplock)) 261 goto out_undo_iolock; 262 } 263 264 if (lock_flags & XFS_ILOCK_EXCL) { 265 if (!mrtryupdate(&ip->i_lock)) 266 goto out_undo_mmaplock; 267 } else if (lock_flags & XFS_ILOCK_SHARED) { 268 if (!mrtryaccess(&ip->i_lock)) 269 goto out_undo_mmaplock; 270 } 271 return 1; 272 273 out_undo_mmaplock: 274 if (lock_flags & XFS_MMAPLOCK_EXCL) 275 mrunlock_excl(&ip->i_mmaplock); 276 else if (lock_flags & XFS_MMAPLOCK_SHARED) 277 mrunlock_shared(&ip->i_mmaplock); 278 out_undo_iolock: 279 if (lock_flags & XFS_IOLOCK_EXCL) 280 up_write(&VFS_I(ip)->i_rwsem); 281 else if (lock_flags & XFS_IOLOCK_SHARED) 282 up_read(&VFS_I(ip)->i_rwsem); 283 out: 284 return 0; 285 } 286 287 /* 288 * xfs_iunlock() is used to drop the inode locks acquired with 289 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 290 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 291 * that we know which locks to drop. 292 * 293 * ip -- the inode being unlocked 294 * lock_flags -- this parameter indicates the inode's locks to be 295 * to be unlocked. See the comment for xfs_ilock() for a list 296 * of valid values for this parameter. 297 * 298 */ 299 void 300 xfs_iunlock( 301 xfs_inode_t *ip, 302 uint lock_flags) 303 { 304 /* 305 * You can't set both SHARED and EXCL for the same lock, 306 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 307 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 308 */ 309 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 310 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 311 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 312 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 313 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 314 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 315 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 316 ASSERT(lock_flags != 0); 317 318 if (lock_flags & XFS_IOLOCK_EXCL) 319 up_write(&VFS_I(ip)->i_rwsem); 320 else if (lock_flags & XFS_IOLOCK_SHARED) 321 up_read(&VFS_I(ip)->i_rwsem); 322 323 if (lock_flags & XFS_MMAPLOCK_EXCL) 324 mrunlock_excl(&ip->i_mmaplock); 325 else if (lock_flags & XFS_MMAPLOCK_SHARED) 326 mrunlock_shared(&ip->i_mmaplock); 327 328 if (lock_flags & XFS_ILOCK_EXCL) 329 mrunlock_excl(&ip->i_lock); 330 else if (lock_flags & XFS_ILOCK_SHARED) 331 mrunlock_shared(&ip->i_lock); 332 333 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 334 } 335 336 /* 337 * give up write locks. the i/o lock cannot be held nested 338 * if it is being demoted. 339 */ 340 void 341 xfs_ilock_demote( 342 xfs_inode_t *ip, 343 uint lock_flags) 344 { 345 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); 346 ASSERT((lock_flags & 347 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 348 349 if (lock_flags & XFS_ILOCK_EXCL) 350 mrdemote(&ip->i_lock); 351 if (lock_flags & XFS_MMAPLOCK_EXCL) 352 mrdemote(&ip->i_mmaplock); 353 if (lock_flags & XFS_IOLOCK_EXCL) 354 downgrade_write(&VFS_I(ip)->i_rwsem); 355 356 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 357 } 358 359 #if defined(DEBUG) || defined(XFS_WARN) 360 int 361 xfs_isilocked( 362 xfs_inode_t *ip, 363 uint lock_flags) 364 { 365 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 366 if (!(lock_flags & XFS_ILOCK_SHARED)) 367 return !!ip->i_lock.mr_writer; 368 return rwsem_is_locked(&ip->i_lock.mr_lock); 369 } 370 371 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { 372 if (!(lock_flags & XFS_MMAPLOCK_SHARED)) 373 return !!ip->i_mmaplock.mr_writer; 374 return rwsem_is_locked(&ip->i_mmaplock.mr_lock); 375 } 376 377 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 378 if (!(lock_flags & XFS_IOLOCK_SHARED)) 379 return !debug_locks || 380 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); 381 return rwsem_is_locked(&VFS_I(ip)->i_rwsem); 382 } 383 384 ASSERT(0); 385 return 0; 386 } 387 #endif 388 389 /* 390 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when 391 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined 392 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build 393 * errors and warnings. 394 */ 395 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) 396 static bool 397 xfs_lockdep_subclass_ok( 398 int subclass) 399 { 400 return subclass < MAX_LOCKDEP_SUBCLASSES; 401 } 402 #else 403 #define xfs_lockdep_subclass_ok(subclass) (true) 404 #endif 405 406 /* 407 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 408 * value. This can be called for any type of inode lock combination, including 409 * parent locking. Care must be taken to ensure we don't overrun the subclass 410 * storage fields in the class mask we build. 411 */ 412 static inline int 413 xfs_lock_inumorder(int lock_mode, int subclass) 414 { 415 int class = 0; 416 417 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | 418 XFS_ILOCK_RTSUM))); 419 ASSERT(xfs_lockdep_subclass_ok(subclass)); 420 421 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 422 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); 423 class += subclass << XFS_IOLOCK_SHIFT; 424 } 425 426 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 427 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); 428 class += subclass << XFS_MMAPLOCK_SHIFT; 429 } 430 431 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { 432 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); 433 class += subclass << XFS_ILOCK_SHIFT; 434 } 435 436 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; 437 } 438 439 /* 440 * The following routine will lock n inodes in exclusive mode. We assume the 441 * caller calls us with the inodes in i_ino order. 442 * 443 * We need to detect deadlock where an inode that we lock is in the AIL and we 444 * start waiting for another inode that is locked by a thread in a long running 445 * transaction (such as truncate). This can result in deadlock since the long 446 * running trans might need to wait for the inode we just locked in order to 447 * push the tail and free space in the log. 448 * 449 * xfs_lock_inodes() can only be used to lock one type of lock at a time - 450 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 451 * lock more than one at a time, lockdep will report false positives saying we 452 * have violated locking orders. 453 */ 454 static void 455 xfs_lock_inodes( 456 xfs_inode_t **ips, 457 int inodes, 458 uint lock_mode) 459 { 460 int attempts = 0, i, j, try_lock; 461 xfs_log_item_t *lp; 462 463 /* 464 * Currently supports between 2 and 5 inodes with exclusive locking. We 465 * support an arbitrary depth of locking here, but absolute limits on 466 * inodes depend on the the type of locking and the limits placed by 467 * lockdep annotations in xfs_lock_inumorder. These are all checked by 468 * the asserts. 469 */ 470 ASSERT(ips && inodes >= 2 && inodes <= 5); 471 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | 472 XFS_ILOCK_EXCL)); 473 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | 474 XFS_ILOCK_SHARED))); 475 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || 476 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); 477 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || 478 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); 479 480 if (lock_mode & XFS_IOLOCK_EXCL) { 481 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); 482 } else if (lock_mode & XFS_MMAPLOCK_EXCL) 483 ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); 484 485 try_lock = 0; 486 i = 0; 487 again: 488 for (; i < inodes; i++) { 489 ASSERT(ips[i]); 490 491 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 492 continue; 493 494 /* 495 * If try_lock is not set yet, make sure all locked inodes are 496 * not in the AIL. If any are, set try_lock to be used later. 497 */ 498 if (!try_lock) { 499 for (j = (i - 1); j >= 0 && !try_lock; j--) { 500 lp = (xfs_log_item_t *)ips[j]->i_itemp; 501 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) 502 try_lock++; 503 } 504 } 505 506 /* 507 * If any of the previous locks we have locked is in the AIL, 508 * we must TRY to get the second and subsequent locks. If 509 * we can't get any, we must release all we have 510 * and try again. 511 */ 512 if (!try_lock) { 513 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 514 continue; 515 } 516 517 /* try_lock means we have an inode locked that is in the AIL. */ 518 ASSERT(i != 0); 519 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 520 continue; 521 522 /* 523 * Unlock all previous guys and try again. xfs_iunlock will try 524 * to push the tail if the inode is in the AIL. 525 */ 526 attempts++; 527 for (j = i - 1; j >= 0; j--) { 528 /* 529 * Check to see if we've already unlocked this one. Not 530 * the first one going back, and the inode ptr is the 531 * same. 532 */ 533 if (j != (i - 1) && ips[j] == ips[j + 1]) 534 continue; 535 536 xfs_iunlock(ips[j], lock_mode); 537 } 538 539 if ((attempts % 5) == 0) { 540 delay(1); /* Don't just spin the CPU */ 541 } 542 i = 0; 543 try_lock = 0; 544 goto again; 545 } 546 } 547 548 /* 549 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 550 * the mmaplock or the ilock, but not more than one type at a time. If we lock 551 * more than one at a time, lockdep will report false positives saying we have 552 * violated locking orders. The iolock must be double-locked separately since 553 * we use i_rwsem for that. We now support taking one lock EXCL and the other 554 * SHARED. 555 */ 556 void 557 xfs_lock_two_inodes( 558 struct xfs_inode *ip0, 559 uint ip0_mode, 560 struct xfs_inode *ip1, 561 uint ip1_mode) 562 { 563 struct xfs_inode *temp; 564 uint mode_temp; 565 int attempts = 0; 566 xfs_log_item_t *lp; 567 568 ASSERT(hweight32(ip0_mode) == 1); 569 ASSERT(hweight32(ip1_mode) == 1); 570 ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); 571 ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); 572 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || 573 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 574 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || 575 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 576 ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || 577 !(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 578 ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) || 579 !(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 580 581 ASSERT(ip0->i_ino != ip1->i_ino); 582 583 if (ip0->i_ino > ip1->i_ino) { 584 temp = ip0; 585 ip0 = ip1; 586 ip1 = temp; 587 mode_temp = ip0_mode; 588 ip0_mode = ip1_mode; 589 ip1_mode = mode_temp; 590 } 591 592 again: 593 xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0)); 594 595 /* 596 * If the first lock we have locked is in the AIL, we must TRY to get 597 * the second lock. If we can't get it, we must release the first one 598 * and try again. 599 */ 600 lp = (xfs_log_item_t *)ip0->i_itemp; 601 if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) { 602 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) { 603 xfs_iunlock(ip0, ip0_mode); 604 if ((++attempts % 5) == 0) 605 delay(1); /* Don't just spin the CPU */ 606 goto again; 607 } 608 } else { 609 xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1)); 610 } 611 } 612 613 void 614 __xfs_iflock( 615 struct xfs_inode *ip) 616 { 617 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 618 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 619 620 do { 621 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 622 if (xfs_isiflocked(ip)) 623 io_schedule(); 624 } while (!xfs_iflock_nowait(ip)); 625 626 finish_wait(wq, &wait.wq_entry); 627 } 628 629 STATIC uint 630 _xfs_dic2xflags( 631 uint16_t di_flags, 632 uint64_t di_flags2, 633 bool has_attr) 634 { 635 uint flags = 0; 636 637 if (di_flags & XFS_DIFLAG_ANY) { 638 if (di_flags & XFS_DIFLAG_REALTIME) 639 flags |= FS_XFLAG_REALTIME; 640 if (di_flags & XFS_DIFLAG_PREALLOC) 641 flags |= FS_XFLAG_PREALLOC; 642 if (di_flags & XFS_DIFLAG_IMMUTABLE) 643 flags |= FS_XFLAG_IMMUTABLE; 644 if (di_flags & XFS_DIFLAG_APPEND) 645 flags |= FS_XFLAG_APPEND; 646 if (di_flags & XFS_DIFLAG_SYNC) 647 flags |= FS_XFLAG_SYNC; 648 if (di_flags & XFS_DIFLAG_NOATIME) 649 flags |= FS_XFLAG_NOATIME; 650 if (di_flags & XFS_DIFLAG_NODUMP) 651 flags |= FS_XFLAG_NODUMP; 652 if (di_flags & XFS_DIFLAG_RTINHERIT) 653 flags |= FS_XFLAG_RTINHERIT; 654 if (di_flags & XFS_DIFLAG_PROJINHERIT) 655 flags |= FS_XFLAG_PROJINHERIT; 656 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 657 flags |= FS_XFLAG_NOSYMLINKS; 658 if (di_flags & XFS_DIFLAG_EXTSIZE) 659 flags |= FS_XFLAG_EXTSIZE; 660 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 661 flags |= FS_XFLAG_EXTSZINHERIT; 662 if (di_flags & XFS_DIFLAG_NODEFRAG) 663 flags |= FS_XFLAG_NODEFRAG; 664 if (di_flags & XFS_DIFLAG_FILESTREAM) 665 flags |= FS_XFLAG_FILESTREAM; 666 } 667 668 if (di_flags2 & XFS_DIFLAG2_ANY) { 669 if (di_flags2 & XFS_DIFLAG2_DAX) 670 flags |= FS_XFLAG_DAX; 671 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 672 flags |= FS_XFLAG_COWEXTSIZE; 673 } 674 675 if (has_attr) 676 flags |= FS_XFLAG_HASATTR; 677 678 return flags; 679 } 680 681 uint 682 xfs_ip2xflags( 683 struct xfs_inode *ip) 684 { 685 struct xfs_icdinode *dic = &ip->i_d; 686 687 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); 688 } 689 690 /* 691 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 692 * is allowed, otherwise it has to be an exact match. If a CI match is found, 693 * ci_name->name will point to a the actual name (caller must free) or 694 * will be set to NULL if an exact match is found. 695 */ 696 int 697 xfs_lookup( 698 xfs_inode_t *dp, 699 struct xfs_name *name, 700 xfs_inode_t **ipp, 701 struct xfs_name *ci_name) 702 { 703 xfs_ino_t inum; 704 int error; 705 706 trace_xfs_lookup(dp, name); 707 708 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 709 return -EIO; 710 711 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 712 if (error) 713 goto out_unlock; 714 715 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 716 if (error) 717 goto out_free_name; 718 719 return 0; 720 721 out_free_name: 722 if (ci_name) 723 kmem_free(ci_name->name); 724 out_unlock: 725 *ipp = NULL; 726 return error; 727 } 728 729 /* 730 * Allocate an inode on disk and return a copy of its in-core version. 731 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 732 * appropriately within the inode. The uid and gid for the inode are 733 * set according to the contents of the given cred structure. 734 * 735 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 736 * has a free inode available, call xfs_iget() to obtain the in-core 737 * version of the allocated inode. Finally, fill in the inode and 738 * log its initial contents. In this case, ialloc_context would be 739 * set to NULL. 740 * 741 * If xfs_dialloc() does not have an available inode, it will replenish 742 * its supply by doing an allocation. Since we can only do one 743 * allocation within a transaction without deadlocks, we must commit 744 * the current transaction before returning the inode itself. 745 * In this case, therefore, we will set ialloc_context and return. 746 * The caller should then commit the current transaction, start a new 747 * transaction, and call xfs_ialloc() again to actually get the inode. 748 * 749 * To ensure that some other process does not grab the inode that 750 * was allocated during the first call to xfs_ialloc(), this routine 751 * also returns the [locked] bp pointing to the head of the freelist 752 * as ialloc_context. The caller should hold this buffer across 753 * the commit and pass it back into this routine on the second call. 754 * 755 * If we are allocating quota inodes, we do not have a parent inode 756 * to attach to or associate with (i.e. pip == NULL) because they 757 * are not linked into the directory structure - they are attached 758 * directly to the superblock - and so have no parent. 759 */ 760 static int 761 xfs_ialloc( 762 xfs_trans_t *tp, 763 xfs_inode_t *pip, 764 umode_t mode, 765 xfs_nlink_t nlink, 766 dev_t rdev, 767 prid_t prid, 768 xfs_buf_t **ialloc_context, 769 xfs_inode_t **ipp) 770 { 771 struct xfs_mount *mp = tp->t_mountp; 772 xfs_ino_t ino; 773 xfs_inode_t *ip; 774 uint flags; 775 int error; 776 struct timespec tv; 777 struct inode *inode; 778 779 /* 780 * Call the space management code to pick 781 * the on-disk inode to be allocated. 782 */ 783 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, 784 ialloc_context, &ino); 785 if (error) 786 return error; 787 if (*ialloc_context || ino == NULLFSINO) { 788 *ipp = NULL; 789 return 0; 790 } 791 ASSERT(*ialloc_context == NULL); 792 793 /* 794 * Protect against obviously corrupt allocation btree records. Later 795 * xfs_iget checks will catch re-allocation of other active in-memory 796 * and on-disk inodes. If we don't catch reallocating the parent inode 797 * here we will deadlock in xfs_iget() so we have to do these checks 798 * first. 799 */ 800 if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) { 801 xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino); 802 return -EFSCORRUPTED; 803 } 804 805 /* 806 * Get the in-core inode with the lock held exclusively. 807 * This is because we're setting fields here we need 808 * to prevent others from looking at until we're done. 809 */ 810 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 811 XFS_ILOCK_EXCL, &ip); 812 if (error) 813 return error; 814 ASSERT(ip != NULL); 815 inode = VFS_I(ip); 816 817 /* 818 * We always convert v1 inodes to v2 now - we only support filesystems 819 * with >= v2 inode capability, so there is no reason for ever leaving 820 * an inode in v1 format. 821 */ 822 if (ip->i_d.di_version == 1) 823 ip->i_d.di_version = 2; 824 825 inode->i_mode = mode; 826 set_nlink(inode, nlink); 827 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 828 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 829 inode->i_rdev = rdev; 830 xfs_set_projid(ip, prid); 831 832 if (pip && XFS_INHERIT_GID(pip)) { 833 ip->i_d.di_gid = pip->i_d.di_gid; 834 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) 835 inode->i_mode |= S_ISGID; 836 } 837 838 /* 839 * If the group ID of the new file does not match the effective group 840 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 841 * (and only if the irix_sgid_inherit compatibility variable is set). 842 */ 843 if ((irix_sgid_inherit) && 844 (inode->i_mode & S_ISGID) && 845 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) 846 inode->i_mode &= ~S_ISGID; 847 848 ip->i_d.di_size = 0; 849 ip->i_d.di_nextents = 0; 850 ASSERT(ip->i_d.di_nblocks == 0); 851 852 tv = current_time(inode); 853 inode->i_mtime = tv; 854 inode->i_atime = tv; 855 inode->i_ctime = tv; 856 857 ip->i_d.di_extsize = 0; 858 ip->i_d.di_dmevmask = 0; 859 ip->i_d.di_dmstate = 0; 860 ip->i_d.di_flags = 0; 861 862 if (ip->i_d.di_version == 3) { 863 inode_set_iversion(inode, 1); 864 ip->i_d.di_flags2 = 0; 865 ip->i_d.di_cowextsize = 0; 866 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; 867 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; 868 } 869 870 871 flags = XFS_ILOG_CORE; 872 switch (mode & S_IFMT) { 873 case S_IFIFO: 874 case S_IFCHR: 875 case S_IFBLK: 876 case S_IFSOCK: 877 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 878 ip->i_df.if_flags = 0; 879 flags |= XFS_ILOG_DEV; 880 break; 881 case S_IFREG: 882 case S_IFDIR: 883 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 884 uint di_flags = 0; 885 886 if (S_ISDIR(mode)) { 887 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 888 di_flags |= XFS_DIFLAG_RTINHERIT; 889 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 890 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 891 ip->i_d.di_extsize = pip->i_d.di_extsize; 892 } 893 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 894 di_flags |= XFS_DIFLAG_PROJINHERIT; 895 } else if (S_ISREG(mode)) { 896 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 897 di_flags |= XFS_DIFLAG_REALTIME; 898 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 899 di_flags |= XFS_DIFLAG_EXTSIZE; 900 ip->i_d.di_extsize = pip->i_d.di_extsize; 901 } 902 } 903 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 904 xfs_inherit_noatime) 905 di_flags |= XFS_DIFLAG_NOATIME; 906 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 907 xfs_inherit_nodump) 908 di_flags |= XFS_DIFLAG_NODUMP; 909 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 910 xfs_inherit_sync) 911 di_flags |= XFS_DIFLAG_SYNC; 912 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 913 xfs_inherit_nosymlinks) 914 di_flags |= XFS_DIFLAG_NOSYMLINKS; 915 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 916 xfs_inherit_nodefrag) 917 di_flags |= XFS_DIFLAG_NODEFRAG; 918 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 919 di_flags |= XFS_DIFLAG_FILESTREAM; 920 921 ip->i_d.di_flags |= di_flags; 922 } 923 if (pip && 924 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && 925 pip->i_d.di_version == 3 && 926 ip->i_d.di_version == 3) { 927 uint64_t di_flags2 = 0; 928 929 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { 930 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; 931 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; 932 } 933 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 934 di_flags2 |= XFS_DIFLAG2_DAX; 935 936 ip->i_d.di_flags2 |= di_flags2; 937 } 938 /* FALLTHROUGH */ 939 case S_IFLNK: 940 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 941 ip->i_df.if_flags = XFS_IFEXTENTS; 942 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 943 ip->i_df.if_u1.if_root = NULL; 944 break; 945 default: 946 ASSERT(0); 947 } 948 /* 949 * Attribute fork settings for new inode. 950 */ 951 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 952 ip->i_d.di_anextents = 0; 953 954 /* 955 * Log the new values stuffed into the inode. 956 */ 957 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 958 xfs_trans_log_inode(tp, ip, flags); 959 960 /* now that we have an i_mode we can setup the inode structure */ 961 xfs_setup_inode(ip); 962 963 *ipp = ip; 964 return 0; 965 } 966 967 /* 968 * Allocates a new inode from disk and return a pointer to the 969 * incore copy. This routine will internally commit the current 970 * transaction and allocate a new one if the Space Manager needed 971 * to do an allocation to replenish the inode free-list. 972 * 973 * This routine is designed to be called from xfs_create and 974 * xfs_create_dir. 975 * 976 */ 977 int 978 xfs_dir_ialloc( 979 xfs_trans_t **tpp, /* input: current transaction; 980 output: may be a new transaction. */ 981 xfs_inode_t *dp, /* directory within whose allocate 982 the inode. */ 983 umode_t mode, 984 xfs_nlink_t nlink, 985 dev_t rdev, 986 prid_t prid, /* project id */ 987 xfs_inode_t **ipp) /* pointer to inode; it will be 988 locked. */ 989 { 990 xfs_trans_t *tp; 991 xfs_inode_t *ip; 992 xfs_buf_t *ialloc_context = NULL; 993 int code; 994 void *dqinfo; 995 uint tflags; 996 997 tp = *tpp; 998 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 999 1000 /* 1001 * xfs_ialloc will return a pointer to an incore inode if 1002 * the Space Manager has an available inode on the free 1003 * list. Otherwise, it will do an allocation and replenish 1004 * the freelist. Since we can only do one allocation per 1005 * transaction without deadlocks, we will need to commit the 1006 * current transaction and start a new one. We will then 1007 * need to call xfs_ialloc again to get the inode. 1008 * 1009 * If xfs_ialloc did an allocation to replenish the freelist, 1010 * it returns the bp containing the head of the freelist as 1011 * ialloc_context. We will hold a lock on it across the 1012 * transaction commit so that no other process can steal 1013 * the inode(s) that we've just allocated. 1014 */ 1015 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context, 1016 &ip); 1017 1018 /* 1019 * Return an error if we were unable to allocate a new inode. 1020 * This should only happen if we run out of space on disk or 1021 * encounter a disk error. 1022 */ 1023 if (code) { 1024 *ipp = NULL; 1025 return code; 1026 } 1027 if (!ialloc_context && !ip) { 1028 *ipp = NULL; 1029 return -ENOSPC; 1030 } 1031 1032 /* 1033 * If the AGI buffer is non-NULL, then we were unable to get an 1034 * inode in one operation. We need to commit the current 1035 * transaction and call xfs_ialloc() again. It is guaranteed 1036 * to succeed the second time. 1037 */ 1038 if (ialloc_context) { 1039 /* 1040 * Normally, xfs_trans_commit releases all the locks. 1041 * We call bhold to hang on to the ialloc_context across 1042 * the commit. Holding this buffer prevents any other 1043 * processes from doing any allocations in this 1044 * allocation group. 1045 */ 1046 xfs_trans_bhold(tp, ialloc_context); 1047 1048 /* 1049 * We want the quota changes to be associated with the next 1050 * transaction, NOT this one. So, detach the dqinfo from this 1051 * and attach it to the next transaction. 1052 */ 1053 dqinfo = NULL; 1054 tflags = 0; 1055 if (tp->t_dqinfo) { 1056 dqinfo = (void *)tp->t_dqinfo; 1057 tp->t_dqinfo = NULL; 1058 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1059 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1060 } 1061 1062 code = xfs_trans_roll(&tp); 1063 1064 /* 1065 * Re-attach the quota info that we detached from prev trx. 1066 */ 1067 if (dqinfo) { 1068 tp->t_dqinfo = dqinfo; 1069 tp->t_flags |= tflags; 1070 } 1071 1072 if (code) { 1073 xfs_buf_relse(ialloc_context); 1074 *tpp = tp; 1075 *ipp = NULL; 1076 return code; 1077 } 1078 xfs_trans_bjoin(tp, ialloc_context); 1079 1080 /* 1081 * Call ialloc again. Since we've locked out all 1082 * other allocations in this allocation group, 1083 * this call should always succeed. 1084 */ 1085 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1086 &ialloc_context, &ip); 1087 1088 /* 1089 * If we get an error at this point, return to the caller 1090 * so that the current transaction can be aborted. 1091 */ 1092 if (code) { 1093 *tpp = tp; 1094 *ipp = NULL; 1095 return code; 1096 } 1097 ASSERT(!ialloc_context && ip); 1098 1099 } 1100 1101 *ipp = ip; 1102 *tpp = tp; 1103 1104 return 0; 1105 } 1106 1107 /* 1108 * Decrement the link count on an inode & log the change. If this causes the 1109 * link count to go to zero, move the inode to AGI unlinked list so that it can 1110 * be freed when the last active reference goes away via xfs_inactive(). 1111 */ 1112 static int /* error */ 1113 xfs_droplink( 1114 xfs_trans_t *tp, 1115 xfs_inode_t *ip) 1116 { 1117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1118 1119 drop_nlink(VFS_I(ip)); 1120 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1121 1122 if (VFS_I(ip)->i_nlink) 1123 return 0; 1124 1125 return xfs_iunlink(tp, ip); 1126 } 1127 1128 /* 1129 * Increment the link count on an inode & log the change. 1130 */ 1131 static int 1132 xfs_bumplink( 1133 xfs_trans_t *tp, 1134 xfs_inode_t *ip) 1135 { 1136 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1137 1138 ASSERT(ip->i_d.di_version > 1); 1139 inc_nlink(VFS_I(ip)); 1140 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1141 return 0; 1142 } 1143 1144 int 1145 xfs_create( 1146 xfs_inode_t *dp, 1147 struct xfs_name *name, 1148 umode_t mode, 1149 dev_t rdev, 1150 xfs_inode_t **ipp) 1151 { 1152 int is_dir = S_ISDIR(mode); 1153 struct xfs_mount *mp = dp->i_mount; 1154 struct xfs_inode *ip = NULL; 1155 struct xfs_trans *tp = NULL; 1156 int error; 1157 struct xfs_defer_ops dfops; 1158 xfs_fsblock_t first_block; 1159 bool unlock_dp_on_error = false; 1160 prid_t prid; 1161 struct xfs_dquot *udqp = NULL; 1162 struct xfs_dquot *gdqp = NULL; 1163 struct xfs_dquot *pdqp = NULL; 1164 struct xfs_trans_res *tres; 1165 uint resblks; 1166 1167 trace_xfs_create(dp, name); 1168 1169 if (XFS_FORCED_SHUTDOWN(mp)) 1170 return -EIO; 1171 1172 prid = xfs_get_initial_prid(dp); 1173 1174 /* 1175 * Make sure that we have allocated dquot(s) on disk. 1176 */ 1177 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1178 xfs_kgid_to_gid(current_fsgid()), prid, 1179 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1180 &udqp, &gdqp, &pdqp); 1181 if (error) 1182 return error; 1183 1184 if (is_dir) { 1185 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1186 tres = &M_RES(mp)->tr_mkdir; 1187 } else { 1188 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1189 tres = &M_RES(mp)->tr_create; 1190 } 1191 1192 /* 1193 * Initially assume that the file does not exist and 1194 * reserve the resources for that case. If that is not 1195 * the case we'll drop the one we have and get a more 1196 * appropriate transaction later. 1197 */ 1198 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1199 if (error == -ENOSPC) { 1200 /* flush outstanding delalloc blocks and retry */ 1201 xfs_flush_inodes(mp); 1202 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1203 } 1204 if (error) 1205 goto out_release_inode; 1206 1207 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1208 unlock_dp_on_error = true; 1209 1210 xfs_defer_init(&dfops, &first_block); 1211 tp->t_agfl_dfops = &dfops; 1212 1213 /* 1214 * Reserve disk quota and the inode. 1215 */ 1216 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1217 pdqp, resblks, 1, 0); 1218 if (error) 1219 goto out_trans_cancel; 1220 1221 /* 1222 * A newly created regular or special file just has one directory 1223 * entry pointing to them, but a directory also the "." entry 1224 * pointing to itself. 1225 */ 1226 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip); 1227 if (error) 1228 goto out_trans_cancel; 1229 1230 /* 1231 * Now we join the directory inode to the transaction. We do not do it 1232 * earlier because xfs_dir_ialloc might commit the previous transaction 1233 * (and release all the locks). An error from here on will result in 1234 * the transaction cancel unlocking dp so don't do it explicitly in the 1235 * error path. 1236 */ 1237 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1238 unlock_dp_on_error = false; 1239 1240 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1241 &first_block, &dfops, resblks ? 1242 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1243 if (error) { 1244 ASSERT(error != -ENOSPC); 1245 goto out_trans_cancel; 1246 } 1247 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1248 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1249 1250 if (is_dir) { 1251 error = xfs_dir_init(tp, ip, dp); 1252 if (error) 1253 goto out_bmap_cancel; 1254 1255 error = xfs_bumplink(tp, dp); 1256 if (error) 1257 goto out_bmap_cancel; 1258 } 1259 1260 /* 1261 * If this is a synchronous mount, make sure that the 1262 * create transaction goes to disk before returning to 1263 * the user. 1264 */ 1265 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1266 xfs_trans_set_sync(tp); 1267 1268 /* 1269 * Attach the dquot(s) to the inodes and modify them incore. 1270 * These ids of the inode couldn't have changed since the new 1271 * inode has been locked ever since it was created. 1272 */ 1273 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1274 1275 error = xfs_defer_finish(&tp, &dfops); 1276 if (error) 1277 goto out_bmap_cancel; 1278 1279 error = xfs_trans_commit(tp); 1280 if (error) 1281 goto out_release_inode; 1282 1283 xfs_qm_dqrele(udqp); 1284 xfs_qm_dqrele(gdqp); 1285 xfs_qm_dqrele(pdqp); 1286 1287 *ipp = ip; 1288 return 0; 1289 1290 out_bmap_cancel: 1291 xfs_defer_cancel(&dfops); 1292 out_trans_cancel: 1293 xfs_trans_cancel(tp); 1294 out_release_inode: 1295 /* 1296 * Wait until after the current transaction is aborted to finish the 1297 * setup of the inode and release the inode. This prevents recursive 1298 * transactions and deadlocks from xfs_inactive. 1299 */ 1300 if (ip) { 1301 xfs_finish_inode_setup(ip); 1302 IRELE(ip); 1303 } 1304 1305 xfs_qm_dqrele(udqp); 1306 xfs_qm_dqrele(gdqp); 1307 xfs_qm_dqrele(pdqp); 1308 1309 if (unlock_dp_on_error) 1310 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1311 return error; 1312 } 1313 1314 int 1315 xfs_create_tmpfile( 1316 struct xfs_inode *dp, 1317 umode_t mode, 1318 struct xfs_inode **ipp) 1319 { 1320 struct xfs_mount *mp = dp->i_mount; 1321 struct xfs_inode *ip = NULL; 1322 struct xfs_trans *tp = NULL; 1323 int error; 1324 prid_t prid; 1325 struct xfs_dquot *udqp = NULL; 1326 struct xfs_dquot *gdqp = NULL; 1327 struct xfs_dquot *pdqp = NULL; 1328 struct xfs_trans_res *tres; 1329 uint resblks; 1330 1331 if (XFS_FORCED_SHUTDOWN(mp)) 1332 return -EIO; 1333 1334 prid = xfs_get_initial_prid(dp); 1335 1336 /* 1337 * Make sure that we have allocated dquot(s) on disk. 1338 */ 1339 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1340 xfs_kgid_to_gid(current_fsgid()), prid, 1341 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1342 &udqp, &gdqp, &pdqp); 1343 if (error) 1344 return error; 1345 1346 resblks = XFS_IALLOC_SPACE_RES(mp); 1347 tres = &M_RES(mp)->tr_create_tmpfile; 1348 1349 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1350 if (error) 1351 goto out_release_inode; 1352 1353 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1354 pdqp, resblks, 1, 0); 1355 if (error) 1356 goto out_trans_cancel; 1357 1358 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip); 1359 if (error) 1360 goto out_trans_cancel; 1361 1362 if (mp->m_flags & XFS_MOUNT_WSYNC) 1363 xfs_trans_set_sync(tp); 1364 1365 /* 1366 * Attach the dquot(s) to the inodes and modify them incore. 1367 * These ids of the inode couldn't have changed since the new 1368 * inode has been locked ever since it was created. 1369 */ 1370 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1371 1372 error = xfs_iunlink(tp, ip); 1373 if (error) 1374 goto out_trans_cancel; 1375 1376 error = xfs_trans_commit(tp); 1377 if (error) 1378 goto out_release_inode; 1379 1380 xfs_qm_dqrele(udqp); 1381 xfs_qm_dqrele(gdqp); 1382 xfs_qm_dqrele(pdqp); 1383 1384 *ipp = ip; 1385 return 0; 1386 1387 out_trans_cancel: 1388 xfs_trans_cancel(tp); 1389 out_release_inode: 1390 /* 1391 * Wait until after the current transaction is aborted to finish the 1392 * setup of the inode and release the inode. This prevents recursive 1393 * transactions and deadlocks from xfs_inactive. 1394 */ 1395 if (ip) { 1396 xfs_finish_inode_setup(ip); 1397 IRELE(ip); 1398 } 1399 1400 xfs_qm_dqrele(udqp); 1401 xfs_qm_dqrele(gdqp); 1402 xfs_qm_dqrele(pdqp); 1403 1404 return error; 1405 } 1406 1407 int 1408 xfs_link( 1409 xfs_inode_t *tdp, 1410 xfs_inode_t *sip, 1411 struct xfs_name *target_name) 1412 { 1413 xfs_mount_t *mp = tdp->i_mount; 1414 xfs_trans_t *tp; 1415 int error; 1416 struct xfs_defer_ops dfops; 1417 xfs_fsblock_t first_block; 1418 int resblks; 1419 1420 trace_xfs_link(tdp, target_name); 1421 1422 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1423 1424 if (XFS_FORCED_SHUTDOWN(mp)) 1425 return -EIO; 1426 1427 error = xfs_qm_dqattach(sip); 1428 if (error) 1429 goto std_return; 1430 1431 error = xfs_qm_dqattach(tdp); 1432 if (error) 1433 goto std_return; 1434 1435 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1436 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1437 if (error == -ENOSPC) { 1438 resblks = 0; 1439 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1440 } 1441 if (error) 1442 goto std_return; 1443 1444 xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL); 1445 1446 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1447 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1448 1449 /* 1450 * If we are using project inheritance, we only allow hard link 1451 * creation in our tree when the project IDs are the same; else 1452 * the tree quota mechanism could be circumvented. 1453 */ 1454 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1455 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1456 error = -EXDEV; 1457 goto error_return; 1458 } 1459 1460 if (!resblks) { 1461 error = xfs_dir_canenter(tp, tdp, target_name); 1462 if (error) 1463 goto error_return; 1464 } 1465 1466 xfs_defer_init(&dfops, &first_block); 1467 tp->t_agfl_dfops = &dfops; 1468 1469 /* 1470 * Handle initial link state of O_TMPFILE inode 1471 */ 1472 if (VFS_I(sip)->i_nlink == 0) { 1473 error = xfs_iunlink_remove(tp, sip); 1474 if (error) 1475 goto error_return; 1476 } 1477 1478 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1479 &first_block, &dfops, resblks); 1480 if (error) 1481 goto error_return; 1482 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1483 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1484 1485 error = xfs_bumplink(tp, sip); 1486 if (error) 1487 goto error_return; 1488 1489 /* 1490 * If this is a synchronous mount, make sure that the 1491 * link transaction goes to disk before returning to 1492 * the user. 1493 */ 1494 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1495 xfs_trans_set_sync(tp); 1496 1497 error = xfs_defer_finish(&tp, &dfops); 1498 if (error) { 1499 xfs_defer_cancel(&dfops); 1500 goto error_return; 1501 } 1502 1503 return xfs_trans_commit(tp); 1504 1505 error_return: 1506 xfs_trans_cancel(tp); 1507 std_return: 1508 return error; 1509 } 1510 1511 /* Clear the reflink flag and the cowblocks tag if possible. */ 1512 static void 1513 xfs_itruncate_clear_reflink_flags( 1514 struct xfs_inode *ip) 1515 { 1516 struct xfs_ifork *dfork; 1517 struct xfs_ifork *cfork; 1518 1519 if (!xfs_is_reflink_inode(ip)) 1520 return; 1521 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 1522 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK); 1523 if (dfork->if_bytes == 0 && cfork->if_bytes == 0) 1524 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 1525 if (cfork->if_bytes == 0) 1526 xfs_inode_clear_cowblocks_tag(ip); 1527 } 1528 1529 /* 1530 * Free up the underlying blocks past new_size. The new size must be smaller 1531 * than the current size. This routine can be used both for the attribute and 1532 * data fork, and does not modify the inode size, which is left to the caller. 1533 * 1534 * The transaction passed to this routine must have made a permanent log 1535 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1536 * given transaction and start new ones, so make sure everything involved in 1537 * the transaction is tidy before calling here. Some transaction will be 1538 * returned to the caller to be committed. The incoming transaction must 1539 * already include the inode, and both inode locks must be held exclusively. 1540 * The inode must also be "held" within the transaction. On return the inode 1541 * will be "held" within the returned transaction. This routine does NOT 1542 * require any disk space to be reserved for it within the transaction. 1543 * 1544 * If we get an error, we must return with the inode locked and linked into the 1545 * current transaction. This keeps things simple for the higher level code, 1546 * because it always knows that the inode is locked and held in the transaction 1547 * that returns to it whether errors occur or not. We don't mark the inode 1548 * dirty on error so that transactions can be easily aborted if possible. 1549 */ 1550 int 1551 xfs_itruncate_extents_flags( 1552 struct xfs_trans **tpp, 1553 struct xfs_inode *ip, 1554 int whichfork, 1555 xfs_fsize_t new_size, 1556 int flags) 1557 { 1558 struct xfs_mount *mp = ip->i_mount; 1559 struct xfs_trans *tp = *tpp; 1560 struct xfs_defer_ops dfops; 1561 xfs_fsblock_t first_block; 1562 xfs_fileoff_t first_unmap_block; 1563 xfs_fileoff_t last_block; 1564 xfs_filblks_t unmap_len; 1565 int error = 0; 1566 int done = 0; 1567 1568 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1569 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1570 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1571 ASSERT(new_size <= XFS_ISIZE(ip)); 1572 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1573 ASSERT(ip->i_itemp != NULL); 1574 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1575 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1576 1577 trace_xfs_itruncate_extents_start(ip, new_size); 1578 1579 flags |= xfs_bmapi_aflag(whichfork); 1580 1581 /* 1582 * Since it is possible for space to become allocated beyond 1583 * the end of the file (in a crash where the space is allocated 1584 * but the inode size is not yet updated), simply remove any 1585 * blocks which show up between the new EOF and the maximum 1586 * possible file size. If the first block to be removed is 1587 * beyond the maximum file size (ie it is the same as last_block), 1588 * then there is nothing to do. 1589 */ 1590 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1591 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1592 if (first_unmap_block == last_block) 1593 return 0; 1594 1595 ASSERT(first_unmap_block < last_block); 1596 unmap_len = last_block - first_unmap_block + 1; 1597 while (!done) { 1598 xfs_defer_init(&dfops, &first_block); 1599 error = xfs_bunmapi(tp, ip, first_unmap_block, unmap_len, flags, 1600 XFS_ITRUNC_MAX_EXTENTS, &first_block, 1601 &dfops, &done); 1602 if (error) 1603 goto out_bmap_cancel; 1604 1605 /* 1606 * Duplicate the transaction that has the permanent 1607 * reservation and commit the old transaction. 1608 */ 1609 xfs_defer_ijoin(&dfops, ip); 1610 error = xfs_defer_finish(&tp, &dfops); 1611 if (error) 1612 goto out_bmap_cancel; 1613 1614 error = xfs_trans_roll_inode(&tp, ip); 1615 if (error) 1616 goto out; 1617 } 1618 1619 if (whichfork == XFS_DATA_FORK) { 1620 /* Remove all pending CoW reservations. */ 1621 error = xfs_reflink_cancel_cow_blocks(ip, &tp, 1622 first_unmap_block, last_block, true); 1623 if (error) 1624 goto out; 1625 1626 xfs_itruncate_clear_reflink_flags(ip); 1627 } 1628 1629 /* 1630 * Always re-log the inode so that our permanent transaction can keep 1631 * on rolling it forward in the log. 1632 */ 1633 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1634 1635 trace_xfs_itruncate_extents_end(ip, new_size); 1636 1637 out: 1638 *tpp = tp; 1639 return error; 1640 out_bmap_cancel: 1641 /* 1642 * If the bunmapi call encounters an error, return to the caller where 1643 * the transaction can be properly aborted. We just need to make sure 1644 * we're not holding any resources that we were not when we came in. 1645 */ 1646 xfs_defer_cancel(&dfops); 1647 goto out; 1648 } 1649 1650 int 1651 xfs_release( 1652 xfs_inode_t *ip) 1653 { 1654 xfs_mount_t *mp = ip->i_mount; 1655 int error; 1656 1657 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1658 return 0; 1659 1660 /* If this is a read-only mount, don't do this (would generate I/O) */ 1661 if (mp->m_flags & XFS_MOUNT_RDONLY) 1662 return 0; 1663 1664 if (!XFS_FORCED_SHUTDOWN(mp)) { 1665 int truncated; 1666 1667 /* 1668 * If we previously truncated this file and removed old data 1669 * in the process, we want to initiate "early" writeout on 1670 * the last close. This is an attempt to combat the notorious 1671 * NULL files problem which is particularly noticeable from a 1672 * truncate down, buffered (re-)write (delalloc), followed by 1673 * a crash. What we are effectively doing here is 1674 * significantly reducing the time window where we'd otherwise 1675 * be exposed to that problem. 1676 */ 1677 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1678 if (truncated) { 1679 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1680 if (ip->i_delayed_blks > 0) { 1681 error = filemap_flush(VFS_I(ip)->i_mapping); 1682 if (error) 1683 return error; 1684 } 1685 } 1686 } 1687 1688 if (VFS_I(ip)->i_nlink == 0) 1689 return 0; 1690 1691 if (xfs_can_free_eofblocks(ip, false)) { 1692 1693 /* 1694 * Check if the inode is being opened, written and closed 1695 * frequently and we have delayed allocation blocks outstanding 1696 * (e.g. streaming writes from the NFS server), truncating the 1697 * blocks past EOF will cause fragmentation to occur. 1698 * 1699 * In this case don't do the truncation, but we have to be 1700 * careful how we detect this case. Blocks beyond EOF show up as 1701 * i_delayed_blks even when the inode is clean, so we need to 1702 * truncate them away first before checking for a dirty release. 1703 * Hence on the first dirty close we will still remove the 1704 * speculative allocation, but after that we will leave it in 1705 * place. 1706 */ 1707 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1708 return 0; 1709 /* 1710 * If we can't get the iolock just skip truncating the blocks 1711 * past EOF because we could deadlock with the mmap_sem 1712 * otherwise. We'll get another chance to drop them once the 1713 * last reference to the inode is dropped, so we'll never leak 1714 * blocks permanently. 1715 */ 1716 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { 1717 error = xfs_free_eofblocks(ip); 1718 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 1719 if (error) 1720 return error; 1721 } 1722 1723 /* delalloc blocks after truncation means it really is dirty */ 1724 if (ip->i_delayed_blks) 1725 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1726 } 1727 return 0; 1728 } 1729 1730 /* 1731 * xfs_inactive_truncate 1732 * 1733 * Called to perform a truncate when an inode becomes unlinked. 1734 */ 1735 STATIC int 1736 xfs_inactive_truncate( 1737 struct xfs_inode *ip) 1738 { 1739 struct xfs_mount *mp = ip->i_mount; 1740 struct xfs_trans *tp; 1741 int error; 1742 1743 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1744 if (error) { 1745 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1746 return error; 1747 } 1748 1749 xfs_ilock(ip, XFS_ILOCK_EXCL); 1750 xfs_trans_ijoin(tp, ip, 0); 1751 1752 /* 1753 * Log the inode size first to prevent stale data exposure in the event 1754 * of a system crash before the truncate completes. See the related 1755 * comment in xfs_vn_setattr_size() for details. 1756 */ 1757 ip->i_d.di_size = 0; 1758 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1759 1760 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1761 if (error) 1762 goto error_trans_cancel; 1763 1764 ASSERT(ip->i_d.di_nextents == 0); 1765 1766 error = xfs_trans_commit(tp); 1767 if (error) 1768 goto error_unlock; 1769 1770 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1771 return 0; 1772 1773 error_trans_cancel: 1774 xfs_trans_cancel(tp); 1775 error_unlock: 1776 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1777 return error; 1778 } 1779 1780 /* 1781 * xfs_inactive_ifree() 1782 * 1783 * Perform the inode free when an inode is unlinked. 1784 */ 1785 STATIC int 1786 xfs_inactive_ifree( 1787 struct xfs_inode *ip) 1788 { 1789 struct xfs_defer_ops dfops; 1790 xfs_fsblock_t first_block; 1791 struct xfs_mount *mp = ip->i_mount; 1792 struct xfs_trans *tp; 1793 int error; 1794 1795 /* 1796 * We try to use a per-AG reservation for any block needed by the finobt 1797 * tree, but as the finobt feature predates the per-AG reservation 1798 * support a degraded file system might not have enough space for the 1799 * reservation at mount time. In that case try to dip into the reserved 1800 * pool and pray. 1801 * 1802 * Send a warning if the reservation does happen to fail, as the inode 1803 * now remains allocated and sits on the unlinked list until the fs is 1804 * repaired. 1805 */ 1806 if (unlikely(mp->m_inotbt_nores)) { 1807 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1808 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 1809 &tp); 1810 } else { 1811 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); 1812 } 1813 if (error) { 1814 if (error == -ENOSPC) { 1815 xfs_warn_ratelimited(mp, 1816 "Failed to remove inode(s) from unlinked list. " 1817 "Please free space, unmount and run xfs_repair."); 1818 } else { 1819 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1820 } 1821 return error; 1822 } 1823 1824 xfs_ilock(ip, XFS_ILOCK_EXCL); 1825 xfs_trans_ijoin(tp, ip, 0); 1826 1827 xfs_defer_init(&dfops, &first_block); 1828 tp->t_agfl_dfops = &dfops; 1829 error = xfs_ifree(tp, ip, &dfops); 1830 if (error) { 1831 /* 1832 * If we fail to free the inode, shut down. The cancel 1833 * might do that, we need to make sure. Otherwise the 1834 * inode might be lost for a long time or forever. 1835 */ 1836 if (!XFS_FORCED_SHUTDOWN(mp)) { 1837 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1838 __func__, error); 1839 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1840 } 1841 xfs_trans_cancel(tp); 1842 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1843 return error; 1844 } 1845 1846 /* 1847 * Credit the quota account(s). The inode is gone. 1848 */ 1849 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1850 1851 /* 1852 * Just ignore errors at this point. There is nothing we can do except 1853 * to try to keep going. Make sure it's not a silent error. 1854 */ 1855 error = xfs_defer_finish(&tp, &dfops); 1856 if (error) { 1857 xfs_notice(mp, "%s: xfs_defer_finish returned error %d", 1858 __func__, error); 1859 xfs_defer_cancel(&dfops); 1860 } 1861 error = xfs_trans_commit(tp); 1862 if (error) 1863 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1864 __func__, error); 1865 1866 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1867 return 0; 1868 } 1869 1870 /* 1871 * xfs_inactive 1872 * 1873 * This is called when the vnode reference count for the vnode 1874 * goes to zero. If the file has been unlinked, then it must 1875 * now be truncated. Also, we clear all of the read-ahead state 1876 * kept for the inode here since the file is now closed. 1877 */ 1878 void 1879 xfs_inactive( 1880 xfs_inode_t *ip) 1881 { 1882 struct xfs_mount *mp; 1883 struct xfs_ifork *cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK); 1884 int error; 1885 int truncate = 0; 1886 1887 /* 1888 * If the inode is already free, then there can be nothing 1889 * to clean up here. 1890 */ 1891 if (VFS_I(ip)->i_mode == 0) { 1892 ASSERT(ip->i_df.if_real_bytes == 0); 1893 ASSERT(ip->i_df.if_broot_bytes == 0); 1894 return; 1895 } 1896 1897 mp = ip->i_mount; 1898 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); 1899 1900 /* If this is a read-only mount, don't do this (would generate I/O) */ 1901 if (mp->m_flags & XFS_MOUNT_RDONLY) 1902 return; 1903 1904 /* Try to clean out the cow blocks if there are any. */ 1905 if (xfs_is_reflink_inode(ip) && cow_ifp->if_bytes > 0) 1906 xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true); 1907 1908 if (VFS_I(ip)->i_nlink != 0) { 1909 /* 1910 * force is true because we are evicting an inode from the 1911 * cache. Post-eof blocks must be freed, lest we end up with 1912 * broken free space accounting. 1913 * 1914 * Note: don't bother with iolock here since lockdep complains 1915 * about acquiring it in reclaim context. We have the only 1916 * reference to the inode at this point anyways. 1917 */ 1918 if (xfs_can_free_eofblocks(ip, true)) 1919 xfs_free_eofblocks(ip); 1920 1921 return; 1922 } 1923 1924 if (S_ISREG(VFS_I(ip)->i_mode) && 1925 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1926 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1927 truncate = 1; 1928 1929 error = xfs_qm_dqattach(ip); 1930 if (error) 1931 return; 1932 1933 if (S_ISLNK(VFS_I(ip)->i_mode)) 1934 error = xfs_inactive_symlink(ip); 1935 else if (truncate) 1936 error = xfs_inactive_truncate(ip); 1937 if (error) 1938 return; 1939 1940 /* 1941 * If there are attributes associated with the file then blow them away 1942 * now. The code calls a routine that recursively deconstructs the 1943 * attribute fork. If also blows away the in-core attribute fork. 1944 */ 1945 if (XFS_IFORK_Q(ip)) { 1946 error = xfs_attr_inactive(ip); 1947 if (error) 1948 return; 1949 } 1950 1951 ASSERT(!ip->i_afp); 1952 ASSERT(ip->i_d.di_anextents == 0); 1953 ASSERT(ip->i_d.di_forkoff == 0); 1954 1955 /* 1956 * Free the inode. 1957 */ 1958 error = xfs_inactive_ifree(ip); 1959 if (error) 1960 return; 1961 1962 /* 1963 * Release the dquots held by inode, if any. 1964 */ 1965 xfs_qm_dqdetach(ip); 1966 } 1967 1968 /* 1969 * This is called when the inode's link count goes to 0 or we are creating a 1970 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1971 * set to true as the link count is dropped to zero by the VFS after we've 1972 * created the file successfully, so we have to add it to the unlinked list 1973 * while the link count is non-zero. 1974 * 1975 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1976 * list when the inode is freed. 1977 */ 1978 STATIC int 1979 xfs_iunlink( 1980 struct xfs_trans *tp, 1981 struct xfs_inode *ip) 1982 { 1983 xfs_mount_t *mp = tp->t_mountp; 1984 xfs_agi_t *agi; 1985 xfs_dinode_t *dip; 1986 xfs_buf_t *agibp; 1987 xfs_buf_t *ibp; 1988 xfs_agino_t agino; 1989 short bucket_index; 1990 int offset; 1991 int error; 1992 1993 ASSERT(VFS_I(ip)->i_mode != 0); 1994 1995 /* 1996 * Get the agi buffer first. It ensures lock ordering 1997 * on the list. 1998 */ 1999 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 2000 if (error) 2001 return error; 2002 agi = XFS_BUF_TO_AGI(agibp); 2003 2004 /* 2005 * Get the index into the agi hash table for the 2006 * list this inode will go on. 2007 */ 2008 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2009 ASSERT(agino != 0); 2010 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2011 ASSERT(agi->agi_unlinked[bucket_index]); 2012 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 2013 2014 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 2015 /* 2016 * There is already another inode in the bucket we need 2017 * to add ourselves to. Add us at the front of the list. 2018 * Here we put the head pointer into our next pointer, 2019 * and then we fall through to point the head at us. 2020 */ 2021 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2022 0, 0); 2023 if (error) 2024 return error; 2025 2026 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 2027 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 2028 offset = ip->i_imap.im_boffset + 2029 offsetof(xfs_dinode_t, di_next_unlinked); 2030 2031 /* need to recalc the inode CRC if appropriate */ 2032 xfs_dinode_calc_crc(mp, dip); 2033 2034 xfs_trans_inode_buf(tp, ibp); 2035 xfs_trans_log_buf(tp, ibp, offset, 2036 (offset + sizeof(xfs_agino_t) - 1)); 2037 xfs_inobp_check(mp, ibp); 2038 } 2039 2040 /* 2041 * Point the bucket head pointer at the inode being inserted. 2042 */ 2043 ASSERT(agino != 0); 2044 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 2045 offset = offsetof(xfs_agi_t, agi_unlinked) + 2046 (sizeof(xfs_agino_t) * bucket_index); 2047 xfs_trans_log_buf(tp, agibp, offset, 2048 (offset + sizeof(xfs_agino_t) - 1)); 2049 return 0; 2050 } 2051 2052 /* 2053 * Pull the on-disk inode from the AGI unlinked list. 2054 */ 2055 STATIC int 2056 xfs_iunlink_remove( 2057 xfs_trans_t *tp, 2058 xfs_inode_t *ip) 2059 { 2060 xfs_ino_t next_ino; 2061 xfs_mount_t *mp; 2062 xfs_agi_t *agi; 2063 xfs_dinode_t *dip; 2064 xfs_buf_t *agibp; 2065 xfs_buf_t *ibp; 2066 xfs_agnumber_t agno; 2067 xfs_agino_t agino; 2068 xfs_agino_t next_agino; 2069 xfs_buf_t *last_ibp; 2070 xfs_dinode_t *last_dip = NULL; 2071 short bucket_index; 2072 int offset, last_offset = 0; 2073 int error; 2074 2075 mp = tp->t_mountp; 2076 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2077 2078 /* 2079 * Get the agi buffer first. It ensures lock ordering 2080 * on the list. 2081 */ 2082 error = xfs_read_agi(mp, tp, agno, &agibp); 2083 if (error) 2084 return error; 2085 2086 agi = XFS_BUF_TO_AGI(agibp); 2087 2088 /* 2089 * Get the index into the agi hash table for the 2090 * list this inode will go on. 2091 */ 2092 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2093 ASSERT(agino != 0); 2094 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2095 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2096 ASSERT(agi->agi_unlinked[bucket_index]); 2097 2098 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2099 /* 2100 * We're at the head of the list. Get the inode's on-disk 2101 * buffer to see if there is anyone after us on the list. 2102 * Only modify our next pointer if it is not already NULLAGINO. 2103 * This saves us the overhead of dealing with the buffer when 2104 * there is no need to change it. 2105 */ 2106 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2107 0, 0); 2108 if (error) { 2109 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2110 __func__, error); 2111 return error; 2112 } 2113 next_agino = be32_to_cpu(dip->di_next_unlinked); 2114 ASSERT(next_agino != 0); 2115 if (next_agino != NULLAGINO) { 2116 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2117 offset = ip->i_imap.im_boffset + 2118 offsetof(xfs_dinode_t, di_next_unlinked); 2119 2120 /* need to recalc the inode CRC if appropriate */ 2121 xfs_dinode_calc_crc(mp, dip); 2122 2123 xfs_trans_inode_buf(tp, ibp); 2124 xfs_trans_log_buf(tp, ibp, offset, 2125 (offset + sizeof(xfs_agino_t) - 1)); 2126 xfs_inobp_check(mp, ibp); 2127 } else { 2128 xfs_trans_brelse(tp, ibp); 2129 } 2130 /* 2131 * Point the bucket head pointer at the next inode. 2132 */ 2133 ASSERT(next_agino != 0); 2134 ASSERT(next_agino != agino); 2135 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2136 offset = offsetof(xfs_agi_t, agi_unlinked) + 2137 (sizeof(xfs_agino_t) * bucket_index); 2138 xfs_trans_log_buf(tp, agibp, offset, 2139 (offset + sizeof(xfs_agino_t) - 1)); 2140 } else { 2141 /* 2142 * We need to search the list for the inode being freed. 2143 */ 2144 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2145 last_ibp = NULL; 2146 while (next_agino != agino) { 2147 struct xfs_imap imap; 2148 2149 if (last_ibp) 2150 xfs_trans_brelse(tp, last_ibp); 2151 2152 imap.im_blkno = 0; 2153 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2154 2155 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2156 if (error) { 2157 xfs_warn(mp, 2158 "%s: xfs_imap returned error %d.", 2159 __func__, error); 2160 return error; 2161 } 2162 2163 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2164 &last_ibp, 0, 0); 2165 if (error) { 2166 xfs_warn(mp, 2167 "%s: xfs_imap_to_bp returned error %d.", 2168 __func__, error); 2169 return error; 2170 } 2171 2172 last_offset = imap.im_boffset; 2173 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2174 ASSERT(next_agino != NULLAGINO); 2175 ASSERT(next_agino != 0); 2176 } 2177 2178 /* 2179 * Now last_ibp points to the buffer previous to us on the 2180 * unlinked list. Pull us from the list. 2181 */ 2182 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2183 0, 0); 2184 if (error) { 2185 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2186 __func__, error); 2187 return error; 2188 } 2189 next_agino = be32_to_cpu(dip->di_next_unlinked); 2190 ASSERT(next_agino != 0); 2191 ASSERT(next_agino != agino); 2192 if (next_agino != NULLAGINO) { 2193 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2194 offset = ip->i_imap.im_boffset + 2195 offsetof(xfs_dinode_t, di_next_unlinked); 2196 2197 /* need to recalc the inode CRC if appropriate */ 2198 xfs_dinode_calc_crc(mp, dip); 2199 2200 xfs_trans_inode_buf(tp, ibp); 2201 xfs_trans_log_buf(tp, ibp, offset, 2202 (offset + sizeof(xfs_agino_t) - 1)); 2203 xfs_inobp_check(mp, ibp); 2204 } else { 2205 xfs_trans_brelse(tp, ibp); 2206 } 2207 /* 2208 * Point the previous inode on the list to the next inode. 2209 */ 2210 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2211 ASSERT(next_agino != 0); 2212 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2213 2214 /* need to recalc the inode CRC if appropriate */ 2215 xfs_dinode_calc_crc(mp, last_dip); 2216 2217 xfs_trans_inode_buf(tp, last_ibp); 2218 xfs_trans_log_buf(tp, last_ibp, offset, 2219 (offset + sizeof(xfs_agino_t) - 1)); 2220 xfs_inobp_check(mp, last_ibp); 2221 } 2222 return 0; 2223 } 2224 2225 /* 2226 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2227 * inodes that are in memory - they all must be marked stale and attached to 2228 * the cluster buffer. 2229 */ 2230 STATIC int 2231 xfs_ifree_cluster( 2232 xfs_inode_t *free_ip, 2233 xfs_trans_t *tp, 2234 struct xfs_icluster *xic) 2235 { 2236 xfs_mount_t *mp = free_ip->i_mount; 2237 int blks_per_cluster; 2238 int inodes_per_cluster; 2239 int nbufs; 2240 int i, j; 2241 int ioffset; 2242 xfs_daddr_t blkno; 2243 xfs_buf_t *bp; 2244 xfs_inode_t *ip; 2245 xfs_inode_log_item_t *iip; 2246 struct xfs_log_item *lip; 2247 struct xfs_perag *pag; 2248 xfs_ino_t inum; 2249 2250 inum = xic->first_ino; 2251 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2252 blks_per_cluster = xfs_icluster_size_fsb(mp); 2253 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2254 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2255 2256 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2257 /* 2258 * The allocation bitmap tells us which inodes of the chunk were 2259 * physically allocated. Skip the cluster if an inode falls into 2260 * a sparse region. 2261 */ 2262 ioffset = inum - xic->first_ino; 2263 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2264 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2265 continue; 2266 } 2267 2268 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2269 XFS_INO_TO_AGBNO(mp, inum)); 2270 2271 /* 2272 * We obtain and lock the backing buffer first in the process 2273 * here, as we have to ensure that any dirty inode that we 2274 * can't get the flush lock on is attached to the buffer. 2275 * If we scan the in-memory inodes first, then buffer IO can 2276 * complete before we get a lock on it, and hence we may fail 2277 * to mark all the active inodes on the buffer stale. 2278 */ 2279 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2280 mp->m_bsize * blks_per_cluster, 2281 XBF_UNMAPPED); 2282 2283 if (!bp) 2284 return -ENOMEM; 2285 2286 /* 2287 * This buffer may not have been correctly initialised as we 2288 * didn't read it from disk. That's not important because we are 2289 * only using to mark the buffer as stale in the log, and to 2290 * attach stale cached inodes on it. That means it will never be 2291 * dispatched for IO. If it is, we want to know about it, and we 2292 * want it to fail. We can acheive this by adding a write 2293 * verifier to the buffer. 2294 */ 2295 bp->b_ops = &xfs_inode_buf_ops; 2296 2297 /* 2298 * Walk the inodes already attached to the buffer and mark them 2299 * stale. These will all have the flush locks held, so an 2300 * in-memory inode walk can't lock them. By marking them all 2301 * stale first, we will not attempt to lock them in the loop 2302 * below as the XFS_ISTALE flag will be set. 2303 */ 2304 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) { 2305 if (lip->li_type == XFS_LI_INODE) { 2306 iip = (xfs_inode_log_item_t *)lip; 2307 ASSERT(iip->ili_logged == 1); 2308 lip->li_cb = xfs_istale_done; 2309 xfs_trans_ail_copy_lsn(mp->m_ail, 2310 &iip->ili_flush_lsn, 2311 &iip->ili_item.li_lsn); 2312 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2313 } 2314 } 2315 2316 2317 /* 2318 * For each inode in memory attempt to add it to the inode 2319 * buffer and set it up for being staled on buffer IO 2320 * completion. This is safe as we've locked out tail pushing 2321 * and flushing by locking the buffer. 2322 * 2323 * We have already marked every inode that was part of a 2324 * transaction stale above, which means there is no point in 2325 * even trying to lock them. 2326 */ 2327 for (i = 0; i < inodes_per_cluster; i++) { 2328 retry: 2329 rcu_read_lock(); 2330 ip = radix_tree_lookup(&pag->pag_ici_root, 2331 XFS_INO_TO_AGINO(mp, (inum + i))); 2332 2333 /* Inode not in memory, nothing to do */ 2334 if (!ip) { 2335 rcu_read_unlock(); 2336 continue; 2337 } 2338 2339 /* 2340 * because this is an RCU protected lookup, we could 2341 * find a recently freed or even reallocated inode 2342 * during the lookup. We need to check under the 2343 * i_flags_lock for a valid inode here. Skip it if it 2344 * is not valid, the wrong inode or stale. 2345 */ 2346 spin_lock(&ip->i_flags_lock); 2347 if (ip->i_ino != inum + i || 2348 __xfs_iflags_test(ip, XFS_ISTALE)) { 2349 spin_unlock(&ip->i_flags_lock); 2350 rcu_read_unlock(); 2351 continue; 2352 } 2353 spin_unlock(&ip->i_flags_lock); 2354 2355 /* 2356 * Don't try to lock/unlock the current inode, but we 2357 * _cannot_ skip the other inodes that we did not find 2358 * in the list attached to the buffer and are not 2359 * already marked stale. If we can't lock it, back off 2360 * and retry. 2361 */ 2362 if (ip != free_ip) { 2363 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2364 rcu_read_unlock(); 2365 delay(1); 2366 goto retry; 2367 } 2368 2369 /* 2370 * Check the inode number again in case we're 2371 * racing with freeing in xfs_reclaim_inode(). 2372 * See the comments in that function for more 2373 * information as to why the initial check is 2374 * not sufficient. 2375 */ 2376 if (ip->i_ino != inum + i) { 2377 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2378 rcu_read_unlock(); 2379 continue; 2380 } 2381 } 2382 rcu_read_unlock(); 2383 2384 xfs_iflock(ip); 2385 xfs_iflags_set(ip, XFS_ISTALE); 2386 2387 /* 2388 * we don't need to attach clean inodes or those only 2389 * with unlogged changes (which we throw away, anyway). 2390 */ 2391 iip = ip->i_itemp; 2392 if (!iip || xfs_inode_clean(ip)) { 2393 ASSERT(ip != free_ip); 2394 xfs_ifunlock(ip); 2395 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2396 continue; 2397 } 2398 2399 iip->ili_last_fields = iip->ili_fields; 2400 iip->ili_fields = 0; 2401 iip->ili_fsync_fields = 0; 2402 iip->ili_logged = 1; 2403 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2404 &iip->ili_item.li_lsn); 2405 2406 xfs_buf_attach_iodone(bp, xfs_istale_done, 2407 &iip->ili_item); 2408 2409 if (ip != free_ip) 2410 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2411 } 2412 2413 xfs_trans_stale_inode_buf(tp, bp); 2414 xfs_trans_binval(tp, bp); 2415 } 2416 2417 xfs_perag_put(pag); 2418 return 0; 2419 } 2420 2421 /* 2422 * Free any local-format buffers sitting around before we reset to 2423 * extents format. 2424 */ 2425 static inline void 2426 xfs_ifree_local_data( 2427 struct xfs_inode *ip, 2428 int whichfork) 2429 { 2430 struct xfs_ifork *ifp; 2431 2432 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL) 2433 return; 2434 2435 ifp = XFS_IFORK_PTR(ip, whichfork); 2436 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); 2437 } 2438 2439 /* 2440 * This is called to return an inode to the inode free list. 2441 * The inode should already be truncated to 0 length and have 2442 * no pages associated with it. This routine also assumes that 2443 * the inode is already a part of the transaction. 2444 * 2445 * The on-disk copy of the inode will have been added to the list 2446 * of unlinked inodes in the AGI. We need to remove the inode from 2447 * that list atomically with respect to freeing it here. 2448 */ 2449 int 2450 xfs_ifree( 2451 xfs_trans_t *tp, 2452 xfs_inode_t *ip, 2453 struct xfs_defer_ops *dfops) 2454 { 2455 int error; 2456 struct xfs_icluster xic = { 0 }; 2457 2458 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2459 ASSERT(VFS_I(ip)->i_nlink == 0); 2460 ASSERT(ip->i_d.di_nextents == 0); 2461 ASSERT(ip->i_d.di_anextents == 0); 2462 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2463 ASSERT(ip->i_d.di_nblocks == 0); 2464 2465 /* 2466 * Pull the on-disk inode from the AGI unlinked list. 2467 */ 2468 error = xfs_iunlink_remove(tp, ip); 2469 if (error) 2470 return error; 2471 2472 error = xfs_difree(tp, ip->i_ino, dfops, &xic); 2473 if (error) 2474 return error; 2475 2476 xfs_ifree_local_data(ip, XFS_DATA_FORK); 2477 xfs_ifree_local_data(ip, XFS_ATTR_FORK); 2478 2479 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2480 ip->i_d.di_flags = 0; 2481 ip->i_d.di_flags2 = 0; 2482 ip->i_d.di_dmevmask = 0; 2483 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2484 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2485 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2486 2487 /* Don't attempt to replay owner changes for a deleted inode */ 2488 ip->i_itemp->ili_fields &= ~(XFS_ILOG_AOWNER|XFS_ILOG_DOWNER); 2489 2490 /* 2491 * Bump the generation count so no one will be confused 2492 * by reincarnations of this inode. 2493 */ 2494 VFS_I(ip)->i_generation++; 2495 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2496 2497 if (xic.deleted) 2498 error = xfs_ifree_cluster(ip, tp, &xic); 2499 2500 return error; 2501 } 2502 2503 /* 2504 * This is called to unpin an inode. The caller must have the inode locked 2505 * in at least shared mode so that the buffer cannot be subsequently pinned 2506 * once someone is waiting for it to be unpinned. 2507 */ 2508 static void 2509 xfs_iunpin( 2510 struct xfs_inode *ip) 2511 { 2512 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2513 2514 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2515 2516 /* Give the log a push to start the unpinning I/O */ 2517 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0, NULL); 2518 2519 } 2520 2521 static void 2522 __xfs_iunpin_wait( 2523 struct xfs_inode *ip) 2524 { 2525 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2526 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2527 2528 xfs_iunpin(ip); 2529 2530 do { 2531 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 2532 if (xfs_ipincount(ip)) 2533 io_schedule(); 2534 } while (xfs_ipincount(ip)); 2535 finish_wait(wq, &wait.wq_entry); 2536 } 2537 2538 void 2539 xfs_iunpin_wait( 2540 struct xfs_inode *ip) 2541 { 2542 if (xfs_ipincount(ip)) 2543 __xfs_iunpin_wait(ip); 2544 } 2545 2546 /* 2547 * Removing an inode from the namespace involves removing the directory entry 2548 * and dropping the link count on the inode. Removing the directory entry can 2549 * result in locking an AGF (directory blocks were freed) and removing a link 2550 * count can result in placing the inode on an unlinked list which results in 2551 * locking an AGI. 2552 * 2553 * The big problem here is that we have an ordering constraint on AGF and AGI 2554 * locking - inode allocation locks the AGI, then can allocate a new extent for 2555 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2556 * removes the inode from the unlinked list, requiring that we lock the AGI 2557 * first, and then freeing the inode can result in an inode chunk being freed 2558 * and hence freeing disk space requiring that we lock an AGF. 2559 * 2560 * Hence the ordering that is imposed by other parts of the code is AGI before 2561 * AGF. This means we cannot remove the directory entry before we drop the inode 2562 * reference count and put it on the unlinked list as this results in a lock 2563 * order of AGF then AGI, and this can deadlock against inode allocation and 2564 * freeing. Therefore we must drop the link counts before we remove the 2565 * directory entry. 2566 * 2567 * This is still safe from a transactional point of view - it is not until we 2568 * get to xfs_defer_finish() that we have the possibility of multiple 2569 * transactions in this operation. Hence as long as we remove the directory 2570 * entry and drop the link count in the first transaction of the remove 2571 * operation, there are no transactional constraints on the ordering here. 2572 */ 2573 int 2574 xfs_remove( 2575 xfs_inode_t *dp, 2576 struct xfs_name *name, 2577 xfs_inode_t *ip) 2578 { 2579 xfs_mount_t *mp = dp->i_mount; 2580 xfs_trans_t *tp = NULL; 2581 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2582 int error = 0; 2583 struct xfs_defer_ops dfops; 2584 xfs_fsblock_t first_block; 2585 uint resblks; 2586 2587 trace_xfs_remove(dp, name); 2588 2589 if (XFS_FORCED_SHUTDOWN(mp)) 2590 return -EIO; 2591 2592 error = xfs_qm_dqattach(dp); 2593 if (error) 2594 goto std_return; 2595 2596 error = xfs_qm_dqattach(ip); 2597 if (error) 2598 goto std_return; 2599 2600 /* 2601 * We try to get the real space reservation first, 2602 * allowing for directory btree deletion(s) implying 2603 * possible bmap insert(s). If we can't get the space 2604 * reservation then we use 0 instead, and avoid the bmap 2605 * btree insert(s) in the directory code by, if the bmap 2606 * insert tries to happen, instead trimming the LAST 2607 * block from the directory. 2608 */ 2609 resblks = XFS_REMOVE_SPACE_RES(mp); 2610 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2611 if (error == -ENOSPC) { 2612 resblks = 0; 2613 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2614 &tp); 2615 } 2616 if (error) { 2617 ASSERT(error != -ENOSPC); 2618 goto std_return; 2619 } 2620 2621 xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL); 2622 2623 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2624 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2625 2626 /* 2627 * If we're removing a directory perform some additional validation. 2628 */ 2629 if (is_dir) { 2630 ASSERT(VFS_I(ip)->i_nlink >= 2); 2631 if (VFS_I(ip)->i_nlink != 2) { 2632 error = -ENOTEMPTY; 2633 goto out_trans_cancel; 2634 } 2635 if (!xfs_dir_isempty(ip)) { 2636 error = -ENOTEMPTY; 2637 goto out_trans_cancel; 2638 } 2639 2640 /* Drop the link from ip's "..". */ 2641 error = xfs_droplink(tp, dp); 2642 if (error) 2643 goto out_trans_cancel; 2644 2645 /* Drop the "." link from ip to self. */ 2646 error = xfs_droplink(tp, ip); 2647 if (error) 2648 goto out_trans_cancel; 2649 } else { 2650 /* 2651 * When removing a non-directory we need to log the parent 2652 * inode here. For a directory this is done implicitly 2653 * by the xfs_droplink call for the ".." entry. 2654 */ 2655 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2656 } 2657 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2658 2659 /* Drop the link from dp to ip. */ 2660 error = xfs_droplink(tp, ip); 2661 if (error) 2662 goto out_trans_cancel; 2663 2664 xfs_defer_init(&dfops, &first_block); 2665 tp->t_agfl_dfops = &dfops; 2666 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2667 &first_block, &dfops, resblks); 2668 if (error) { 2669 ASSERT(error != -ENOENT); 2670 goto out_bmap_cancel; 2671 } 2672 2673 /* 2674 * If this is a synchronous mount, make sure that the 2675 * remove transaction goes to disk before returning to 2676 * the user. 2677 */ 2678 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2679 xfs_trans_set_sync(tp); 2680 2681 error = xfs_defer_finish(&tp, &dfops); 2682 if (error) 2683 goto out_bmap_cancel; 2684 2685 error = xfs_trans_commit(tp); 2686 if (error) 2687 goto std_return; 2688 2689 if (is_dir && xfs_inode_is_filestream(ip)) 2690 xfs_filestream_deassociate(ip); 2691 2692 return 0; 2693 2694 out_bmap_cancel: 2695 xfs_defer_cancel(&dfops); 2696 out_trans_cancel: 2697 xfs_trans_cancel(tp); 2698 std_return: 2699 return error; 2700 } 2701 2702 /* 2703 * Enter all inodes for a rename transaction into a sorted array. 2704 */ 2705 #define __XFS_SORT_INODES 5 2706 STATIC void 2707 xfs_sort_for_rename( 2708 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2709 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2710 struct xfs_inode *ip1, /* in: inode of old entry */ 2711 struct xfs_inode *ip2, /* in: inode of new entry */ 2712 struct xfs_inode *wip, /* in: whiteout inode */ 2713 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2714 int *num_inodes) /* in/out: inodes in array */ 2715 { 2716 int i, j; 2717 2718 ASSERT(*num_inodes == __XFS_SORT_INODES); 2719 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2720 2721 /* 2722 * i_tab contains a list of pointers to inodes. We initialize 2723 * the table here & we'll sort it. We will then use it to 2724 * order the acquisition of the inode locks. 2725 * 2726 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2727 */ 2728 i = 0; 2729 i_tab[i++] = dp1; 2730 i_tab[i++] = dp2; 2731 i_tab[i++] = ip1; 2732 if (ip2) 2733 i_tab[i++] = ip2; 2734 if (wip) 2735 i_tab[i++] = wip; 2736 *num_inodes = i; 2737 2738 /* 2739 * Sort the elements via bubble sort. (Remember, there are at 2740 * most 5 elements to sort, so this is adequate.) 2741 */ 2742 for (i = 0; i < *num_inodes; i++) { 2743 for (j = 1; j < *num_inodes; j++) { 2744 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2745 struct xfs_inode *temp = i_tab[j]; 2746 i_tab[j] = i_tab[j-1]; 2747 i_tab[j-1] = temp; 2748 } 2749 } 2750 } 2751 } 2752 2753 static int 2754 xfs_finish_rename( 2755 struct xfs_trans *tp, 2756 struct xfs_defer_ops *dfops) 2757 { 2758 int error; 2759 2760 /* 2761 * If this is a synchronous mount, make sure that the rename transaction 2762 * goes to disk before returning to the user. 2763 */ 2764 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2765 xfs_trans_set_sync(tp); 2766 2767 error = xfs_defer_finish(&tp, dfops); 2768 if (error) { 2769 xfs_defer_cancel(dfops); 2770 xfs_trans_cancel(tp); 2771 return error; 2772 } 2773 2774 return xfs_trans_commit(tp); 2775 } 2776 2777 /* 2778 * xfs_cross_rename() 2779 * 2780 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2781 */ 2782 STATIC int 2783 xfs_cross_rename( 2784 struct xfs_trans *tp, 2785 struct xfs_inode *dp1, 2786 struct xfs_name *name1, 2787 struct xfs_inode *ip1, 2788 struct xfs_inode *dp2, 2789 struct xfs_name *name2, 2790 struct xfs_inode *ip2, 2791 struct xfs_defer_ops *dfops, 2792 xfs_fsblock_t *first_block, 2793 int spaceres) 2794 { 2795 int error = 0; 2796 int ip1_flags = 0; 2797 int ip2_flags = 0; 2798 int dp2_flags = 0; 2799 2800 /* Swap inode number for dirent in first parent */ 2801 error = xfs_dir_replace(tp, dp1, name1, 2802 ip2->i_ino, 2803 first_block, dfops, spaceres); 2804 if (error) 2805 goto out_trans_abort; 2806 2807 /* Swap inode number for dirent in second parent */ 2808 error = xfs_dir_replace(tp, dp2, name2, 2809 ip1->i_ino, 2810 first_block, dfops, spaceres); 2811 if (error) 2812 goto out_trans_abort; 2813 2814 /* 2815 * If we're renaming one or more directories across different parents, 2816 * update the respective ".." entries (and link counts) to match the new 2817 * parents. 2818 */ 2819 if (dp1 != dp2) { 2820 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2821 2822 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2823 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2824 dp1->i_ino, first_block, 2825 dfops, spaceres); 2826 if (error) 2827 goto out_trans_abort; 2828 2829 /* transfer ip2 ".." reference to dp1 */ 2830 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2831 error = xfs_droplink(tp, dp2); 2832 if (error) 2833 goto out_trans_abort; 2834 error = xfs_bumplink(tp, dp1); 2835 if (error) 2836 goto out_trans_abort; 2837 } 2838 2839 /* 2840 * Although ip1 isn't changed here, userspace needs 2841 * to be warned about the change, so that applications 2842 * relying on it (like backup ones), will properly 2843 * notify the change 2844 */ 2845 ip1_flags |= XFS_ICHGTIME_CHG; 2846 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2847 } 2848 2849 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2850 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2851 dp2->i_ino, first_block, 2852 dfops, spaceres); 2853 if (error) 2854 goto out_trans_abort; 2855 2856 /* transfer ip1 ".." reference to dp2 */ 2857 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2858 error = xfs_droplink(tp, dp1); 2859 if (error) 2860 goto out_trans_abort; 2861 error = xfs_bumplink(tp, dp2); 2862 if (error) 2863 goto out_trans_abort; 2864 } 2865 2866 /* 2867 * Although ip2 isn't changed here, userspace needs 2868 * to be warned about the change, so that applications 2869 * relying on it (like backup ones), will properly 2870 * notify the change 2871 */ 2872 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2873 ip2_flags |= XFS_ICHGTIME_CHG; 2874 } 2875 } 2876 2877 if (ip1_flags) { 2878 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2879 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2880 } 2881 if (ip2_flags) { 2882 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2883 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2884 } 2885 if (dp2_flags) { 2886 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2887 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2888 } 2889 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2890 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2891 return xfs_finish_rename(tp, dfops); 2892 2893 out_trans_abort: 2894 xfs_defer_cancel(dfops); 2895 xfs_trans_cancel(tp); 2896 return error; 2897 } 2898 2899 /* 2900 * xfs_rename_alloc_whiteout() 2901 * 2902 * Return a referenced, unlinked, unlocked inode that that can be used as a 2903 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2904 * crash between allocating the inode and linking it into the rename transaction 2905 * recovery will free the inode and we won't leak it. 2906 */ 2907 static int 2908 xfs_rename_alloc_whiteout( 2909 struct xfs_inode *dp, 2910 struct xfs_inode **wip) 2911 { 2912 struct xfs_inode *tmpfile; 2913 int error; 2914 2915 error = xfs_create_tmpfile(dp, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2916 if (error) 2917 return error; 2918 2919 /* 2920 * Prepare the tmpfile inode as if it were created through the VFS. 2921 * Otherwise, the link increment paths will complain about nlink 0->1. 2922 * Drop the link count as done by d_tmpfile(), complete the inode setup 2923 * and flag it as linkable. 2924 */ 2925 drop_nlink(VFS_I(tmpfile)); 2926 xfs_setup_iops(tmpfile); 2927 xfs_finish_inode_setup(tmpfile); 2928 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2929 2930 *wip = tmpfile; 2931 return 0; 2932 } 2933 2934 /* 2935 * xfs_rename 2936 */ 2937 int 2938 xfs_rename( 2939 struct xfs_inode *src_dp, 2940 struct xfs_name *src_name, 2941 struct xfs_inode *src_ip, 2942 struct xfs_inode *target_dp, 2943 struct xfs_name *target_name, 2944 struct xfs_inode *target_ip, 2945 unsigned int flags) 2946 { 2947 struct xfs_mount *mp = src_dp->i_mount; 2948 struct xfs_trans *tp; 2949 struct xfs_defer_ops dfops; 2950 xfs_fsblock_t first_block; 2951 struct xfs_inode *wip = NULL; /* whiteout inode */ 2952 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2953 int num_inodes = __XFS_SORT_INODES; 2954 bool new_parent = (src_dp != target_dp); 2955 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2956 int spaceres; 2957 int error; 2958 2959 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2960 2961 if ((flags & RENAME_EXCHANGE) && !target_ip) 2962 return -EINVAL; 2963 2964 /* 2965 * If we are doing a whiteout operation, allocate the whiteout inode 2966 * we will be placing at the target and ensure the type is set 2967 * appropriately. 2968 */ 2969 if (flags & RENAME_WHITEOUT) { 2970 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2971 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2972 if (error) 2973 return error; 2974 2975 /* setup target dirent info as whiteout */ 2976 src_name->type = XFS_DIR3_FT_CHRDEV; 2977 } 2978 2979 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2980 inodes, &num_inodes); 2981 2982 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2983 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2984 if (error == -ENOSPC) { 2985 spaceres = 0; 2986 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2987 &tp); 2988 } 2989 if (error) 2990 goto out_release_wip; 2991 2992 /* 2993 * Attach the dquots to the inodes 2994 */ 2995 error = xfs_qm_vop_rename_dqattach(inodes); 2996 if (error) 2997 goto out_trans_cancel; 2998 2999 /* 3000 * Lock all the participating inodes. Depending upon whether 3001 * the target_name exists in the target directory, and 3002 * whether the target directory is the same as the source 3003 * directory, we can lock from 2 to 4 inodes. 3004 */ 3005 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 3006 3007 /* 3008 * Join all the inodes to the transaction. From this point on, 3009 * we can rely on either trans_commit or trans_cancel to unlock 3010 * them. 3011 */ 3012 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 3013 if (new_parent) 3014 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 3015 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 3016 if (target_ip) 3017 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 3018 if (wip) 3019 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 3020 3021 /* 3022 * If we are using project inheritance, we only allow renames 3023 * into our tree when the project IDs are the same; else the 3024 * tree quota mechanism would be circumvented. 3025 */ 3026 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 3027 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 3028 error = -EXDEV; 3029 goto out_trans_cancel; 3030 } 3031 3032 xfs_defer_init(&dfops, &first_block); 3033 tp->t_agfl_dfops = &dfops; 3034 3035 /* RENAME_EXCHANGE is unique from here on. */ 3036 if (flags & RENAME_EXCHANGE) 3037 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 3038 target_dp, target_name, target_ip, 3039 &dfops, &first_block, spaceres); 3040 3041 /* 3042 * Set up the target. 3043 */ 3044 if (target_ip == NULL) { 3045 /* 3046 * If there's no space reservation, check the entry will 3047 * fit before actually inserting it. 3048 */ 3049 if (!spaceres) { 3050 error = xfs_dir_canenter(tp, target_dp, target_name); 3051 if (error) 3052 goto out_trans_cancel; 3053 } 3054 /* 3055 * If target does not exist and the rename crosses 3056 * directories, adjust the target directory link count 3057 * to account for the ".." reference from the new entry. 3058 */ 3059 error = xfs_dir_createname(tp, target_dp, target_name, 3060 src_ip->i_ino, &first_block, 3061 &dfops, spaceres); 3062 if (error) 3063 goto out_bmap_cancel; 3064 3065 xfs_trans_ichgtime(tp, target_dp, 3066 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3067 3068 if (new_parent && src_is_directory) { 3069 error = xfs_bumplink(tp, target_dp); 3070 if (error) 3071 goto out_bmap_cancel; 3072 } 3073 } else { /* target_ip != NULL */ 3074 /* 3075 * If target exists and it's a directory, check that both 3076 * target and source are directories and that target can be 3077 * destroyed, or that neither is a directory. 3078 */ 3079 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 3080 /* 3081 * Make sure target dir is empty. 3082 */ 3083 if (!(xfs_dir_isempty(target_ip)) || 3084 (VFS_I(target_ip)->i_nlink > 2)) { 3085 error = -EEXIST; 3086 goto out_trans_cancel; 3087 } 3088 } 3089 3090 /* 3091 * Link the source inode under the target name. 3092 * If the source inode is a directory and we are moving 3093 * it across directories, its ".." entry will be 3094 * inconsistent until we replace that down below. 3095 * 3096 * In case there is already an entry with the same 3097 * name at the destination directory, remove it first. 3098 */ 3099 error = xfs_dir_replace(tp, target_dp, target_name, 3100 src_ip->i_ino, 3101 &first_block, &dfops, spaceres); 3102 if (error) 3103 goto out_bmap_cancel; 3104 3105 xfs_trans_ichgtime(tp, target_dp, 3106 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3107 3108 /* 3109 * Decrement the link count on the target since the target 3110 * dir no longer points to it. 3111 */ 3112 error = xfs_droplink(tp, target_ip); 3113 if (error) 3114 goto out_bmap_cancel; 3115 3116 if (src_is_directory) { 3117 /* 3118 * Drop the link from the old "." entry. 3119 */ 3120 error = xfs_droplink(tp, target_ip); 3121 if (error) 3122 goto out_bmap_cancel; 3123 } 3124 } /* target_ip != NULL */ 3125 3126 /* 3127 * Remove the source. 3128 */ 3129 if (new_parent && src_is_directory) { 3130 /* 3131 * Rewrite the ".." entry to point to the new 3132 * directory. 3133 */ 3134 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3135 target_dp->i_ino, 3136 &first_block, &dfops, spaceres); 3137 ASSERT(error != -EEXIST); 3138 if (error) 3139 goto out_bmap_cancel; 3140 } 3141 3142 /* 3143 * We always want to hit the ctime on the source inode. 3144 * 3145 * This isn't strictly required by the standards since the source 3146 * inode isn't really being changed, but old unix file systems did 3147 * it and some incremental backup programs won't work without it. 3148 */ 3149 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3150 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3151 3152 /* 3153 * Adjust the link count on src_dp. This is necessary when 3154 * renaming a directory, either within one parent when 3155 * the target existed, or across two parent directories. 3156 */ 3157 if (src_is_directory && (new_parent || target_ip != NULL)) { 3158 3159 /* 3160 * Decrement link count on src_directory since the 3161 * entry that's moved no longer points to it. 3162 */ 3163 error = xfs_droplink(tp, src_dp); 3164 if (error) 3165 goto out_bmap_cancel; 3166 } 3167 3168 /* 3169 * For whiteouts, we only need to update the source dirent with the 3170 * inode number of the whiteout inode rather than removing it 3171 * altogether. 3172 */ 3173 if (wip) { 3174 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3175 &first_block, &dfops, spaceres); 3176 } else 3177 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3178 &first_block, &dfops, spaceres); 3179 if (error) 3180 goto out_bmap_cancel; 3181 3182 /* 3183 * For whiteouts, we need to bump the link count on the whiteout inode. 3184 * This means that failures all the way up to this point leave the inode 3185 * on the unlinked list and so cleanup is a simple matter of dropping 3186 * the remaining reference to it. If we fail here after bumping the link 3187 * count, we're shutting down the filesystem so we'll never see the 3188 * intermediate state on disk. 3189 */ 3190 if (wip) { 3191 ASSERT(VFS_I(wip)->i_nlink == 0); 3192 error = xfs_bumplink(tp, wip); 3193 if (error) 3194 goto out_bmap_cancel; 3195 error = xfs_iunlink_remove(tp, wip); 3196 if (error) 3197 goto out_bmap_cancel; 3198 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3199 3200 /* 3201 * Now we have a real link, clear the "I'm a tmpfile" state 3202 * flag from the inode so it doesn't accidentally get misused in 3203 * future. 3204 */ 3205 VFS_I(wip)->i_state &= ~I_LINKABLE; 3206 } 3207 3208 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3209 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3210 if (new_parent) 3211 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3212 3213 error = xfs_finish_rename(tp, &dfops); 3214 if (wip) 3215 IRELE(wip); 3216 return error; 3217 3218 out_bmap_cancel: 3219 xfs_defer_cancel(&dfops); 3220 out_trans_cancel: 3221 xfs_trans_cancel(tp); 3222 out_release_wip: 3223 if (wip) 3224 IRELE(wip); 3225 return error; 3226 } 3227 3228 STATIC int 3229 xfs_iflush_cluster( 3230 struct xfs_inode *ip, 3231 struct xfs_buf *bp) 3232 { 3233 struct xfs_mount *mp = ip->i_mount; 3234 struct xfs_perag *pag; 3235 unsigned long first_index, mask; 3236 unsigned long inodes_per_cluster; 3237 int cilist_size; 3238 struct xfs_inode **cilist; 3239 struct xfs_inode *cip; 3240 int nr_found; 3241 int clcount = 0; 3242 int bufwasdelwri; 3243 int i; 3244 3245 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3246 3247 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3248 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3249 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3250 if (!cilist) 3251 goto out_put; 3252 3253 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3254 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3255 rcu_read_lock(); 3256 /* really need a gang lookup range call here */ 3257 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3258 first_index, inodes_per_cluster); 3259 if (nr_found == 0) 3260 goto out_free; 3261 3262 for (i = 0; i < nr_found; i++) { 3263 cip = cilist[i]; 3264 if (cip == ip) 3265 continue; 3266 3267 /* 3268 * because this is an RCU protected lookup, we could find a 3269 * recently freed or even reallocated inode during the lookup. 3270 * We need to check under the i_flags_lock for a valid inode 3271 * here. Skip it if it is not valid or the wrong inode. 3272 */ 3273 spin_lock(&cip->i_flags_lock); 3274 if (!cip->i_ino || 3275 __xfs_iflags_test(cip, XFS_ISTALE)) { 3276 spin_unlock(&cip->i_flags_lock); 3277 continue; 3278 } 3279 3280 /* 3281 * Once we fall off the end of the cluster, no point checking 3282 * any more inodes in the list because they will also all be 3283 * outside the cluster. 3284 */ 3285 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3286 spin_unlock(&cip->i_flags_lock); 3287 break; 3288 } 3289 spin_unlock(&cip->i_flags_lock); 3290 3291 /* 3292 * Do an un-protected check to see if the inode is dirty and 3293 * is a candidate for flushing. These checks will be repeated 3294 * later after the appropriate locks are acquired. 3295 */ 3296 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3297 continue; 3298 3299 /* 3300 * Try to get locks. If any are unavailable or it is pinned, 3301 * then this inode cannot be flushed and is skipped. 3302 */ 3303 3304 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3305 continue; 3306 if (!xfs_iflock_nowait(cip)) { 3307 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3308 continue; 3309 } 3310 if (xfs_ipincount(cip)) { 3311 xfs_ifunlock(cip); 3312 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3313 continue; 3314 } 3315 3316 3317 /* 3318 * Check the inode number again, just to be certain we are not 3319 * racing with freeing in xfs_reclaim_inode(). See the comments 3320 * in that function for more information as to why the initial 3321 * check is not sufficient. 3322 */ 3323 if (!cip->i_ino) { 3324 xfs_ifunlock(cip); 3325 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3326 continue; 3327 } 3328 3329 /* 3330 * arriving here means that this inode can be flushed. First 3331 * re-check that it's dirty before flushing. 3332 */ 3333 if (!xfs_inode_clean(cip)) { 3334 int error; 3335 error = xfs_iflush_int(cip, bp); 3336 if (error) { 3337 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3338 goto cluster_corrupt_out; 3339 } 3340 clcount++; 3341 } else { 3342 xfs_ifunlock(cip); 3343 } 3344 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3345 } 3346 3347 if (clcount) { 3348 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3349 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3350 } 3351 3352 out_free: 3353 rcu_read_unlock(); 3354 kmem_free(cilist); 3355 out_put: 3356 xfs_perag_put(pag); 3357 return 0; 3358 3359 3360 cluster_corrupt_out: 3361 /* 3362 * Corruption detected in the clustering loop. Invalidate the 3363 * inode buffer and shut down the filesystem. 3364 */ 3365 rcu_read_unlock(); 3366 /* 3367 * Clean up the buffer. If it was delwri, just release it -- 3368 * brelse can handle it with no problems. If not, shut down the 3369 * filesystem before releasing the buffer. 3370 */ 3371 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3372 if (bufwasdelwri) 3373 xfs_buf_relse(bp); 3374 3375 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3376 3377 if (!bufwasdelwri) { 3378 /* 3379 * Just like incore_relse: if we have b_iodone functions, 3380 * mark the buffer as an error and call them. Otherwise 3381 * mark it as stale and brelse. 3382 */ 3383 if (bp->b_iodone) { 3384 bp->b_flags &= ~XBF_DONE; 3385 xfs_buf_stale(bp); 3386 xfs_buf_ioerror(bp, -EIO); 3387 xfs_buf_ioend(bp); 3388 } else { 3389 xfs_buf_stale(bp); 3390 xfs_buf_relse(bp); 3391 } 3392 } 3393 3394 /* 3395 * Unlocks the flush lock 3396 */ 3397 xfs_iflush_abort(cip, false); 3398 kmem_free(cilist); 3399 xfs_perag_put(pag); 3400 return -EFSCORRUPTED; 3401 } 3402 3403 /* 3404 * Flush dirty inode metadata into the backing buffer. 3405 * 3406 * The caller must have the inode lock and the inode flush lock held. The 3407 * inode lock will still be held upon return to the caller, and the inode 3408 * flush lock will be released after the inode has reached the disk. 3409 * 3410 * The caller must write out the buffer returned in *bpp and release it. 3411 */ 3412 int 3413 xfs_iflush( 3414 struct xfs_inode *ip, 3415 struct xfs_buf **bpp) 3416 { 3417 struct xfs_mount *mp = ip->i_mount; 3418 struct xfs_buf *bp = NULL; 3419 struct xfs_dinode *dip; 3420 int error; 3421 3422 XFS_STATS_INC(mp, xs_iflush_count); 3423 3424 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3425 ASSERT(xfs_isiflocked(ip)); 3426 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3427 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3428 3429 *bpp = NULL; 3430 3431 xfs_iunpin_wait(ip); 3432 3433 /* 3434 * For stale inodes we cannot rely on the backing buffer remaining 3435 * stale in cache for the remaining life of the stale inode and so 3436 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3437 * inodes below. We have to check this after ensuring the inode is 3438 * unpinned so that it is safe to reclaim the stale inode after the 3439 * flush call. 3440 */ 3441 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3442 xfs_ifunlock(ip); 3443 return 0; 3444 } 3445 3446 /* 3447 * This may have been unpinned because the filesystem is shutting 3448 * down forcibly. If that's the case we must not write this inode 3449 * to disk, because the log record didn't make it to disk. 3450 * 3451 * We also have to remove the log item from the AIL in this case, 3452 * as we wait for an empty AIL as part of the unmount process. 3453 */ 3454 if (XFS_FORCED_SHUTDOWN(mp)) { 3455 error = -EIO; 3456 goto abort_out; 3457 } 3458 3459 /* 3460 * Get the buffer containing the on-disk inode. We are doing a try-lock 3461 * operation here, so we may get an EAGAIN error. In that case, we 3462 * simply want to return with the inode still dirty. 3463 * 3464 * If we get any other error, we effectively have a corruption situation 3465 * and we cannot flush the inode, so we treat it the same as failing 3466 * xfs_iflush_int(). 3467 */ 3468 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3469 0); 3470 if (error == -EAGAIN) { 3471 xfs_ifunlock(ip); 3472 return error; 3473 } 3474 if (error) 3475 goto corrupt_out; 3476 3477 /* 3478 * First flush out the inode that xfs_iflush was called with. 3479 */ 3480 error = xfs_iflush_int(ip, bp); 3481 if (error) 3482 goto corrupt_out; 3483 3484 /* 3485 * If the buffer is pinned then push on the log now so we won't 3486 * get stuck waiting in the write for too long. 3487 */ 3488 if (xfs_buf_ispinned(bp)) 3489 xfs_log_force(mp, 0); 3490 3491 /* 3492 * inode clustering: 3493 * see if other inodes can be gathered into this write 3494 */ 3495 error = xfs_iflush_cluster(ip, bp); 3496 if (error) 3497 goto cluster_corrupt_out; 3498 3499 *bpp = bp; 3500 return 0; 3501 3502 corrupt_out: 3503 if (bp) 3504 xfs_buf_relse(bp); 3505 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3506 cluster_corrupt_out: 3507 error = -EFSCORRUPTED; 3508 abort_out: 3509 /* 3510 * Unlocks the flush lock 3511 */ 3512 xfs_iflush_abort(ip, false); 3513 return error; 3514 } 3515 3516 /* 3517 * If there are inline format data / attr forks attached to this inode, 3518 * make sure they're not corrupt. 3519 */ 3520 bool 3521 xfs_inode_verify_forks( 3522 struct xfs_inode *ip) 3523 { 3524 struct xfs_ifork *ifp; 3525 xfs_failaddr_t fa; 3526 3527 fa = xfs_ifork_verify_data(ip, &xfs_default_ifork_ops); 3528 if (fa) { 3529 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 3530 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "data fork", 3531 ifp->if_u1.if_data, ifp->if_bytes, fa); 3532 return false; 3533 } 3534 3535 fa = xfs_ifork_verify_attr(ip, &xfs_default_ifork_ops); 3536 if (fa) { 3537 ifp = XFS_IFORK_PTR(ip, XFS_ATTR_FORK); 3538 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "attr fork", 3539 ifp ? ifp->if_u1.if_data : NULL, 3540 ifp ? ifp->if_bytes : 0, fa); 3541 return false; 3542 } 3543 return true; 3544 } 3545 3546 STATIC int 3547 xfs_iflush_int( 3548 struct xfs_inode *ip, 3549 struct xfs_buf *bp) 3550 { 3551 struct xfs_inode_log_item *iip = ip->i_itemp; 3552 struct xfs_dinode *dip; 3553 struct xfs_mount *mp = ip->i_mount; 3554 3555 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3556 ASSERT(xfs_isiflocked(ip)); 3557 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3558 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3559 ASSERT(iip != NULL && iip->ili_fields != 0); 3560 ASSERT(ip->i_d.di_version > 1); 3561 3562 /* set *dip = inode's place in the buffer */ 3563 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3564 3565 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3566 mp, XFS_ERRTAG_IFLUSH_1)) { 3567 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3568 "%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT, 3569 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3570 goto corrupt_out; 3571 } 3572 if (S_ISREG(VFS_I(ip)->i_mode)) { 3573 if (XFS_TEST_ERROR( 3574 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3575 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3576 mp, XFS_ERRTAG_IFLUSH_3)) { 3577 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3578 "%s: Bad regular inode %Lu, ptr "PTR_FMT, 3579 __func__, ip->i_ino, ip); 3580 goto corrupt_out; 3581 } 3582 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3583 if (XFS_TEST_ERROR( 3584 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3585 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3586 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3587 mp, XFS_ERRTAG_IFLUSH_4)) { 3588 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3589 "%s: Bad directory inode %Lu, ptr "PTR_FMT, 3590 __func__, ip->i_ino, ip); 3591 goto corrupt_out; 3592 } 3593 } 3594 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3595 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { 3596 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3597 "%s: detected corrupt incore inode %Lu, " 3598 "total extents = %d, nblocks = %Ld, ptr "PTR_FMT, 3599 __func__, ip->i_ino, 3600 ip->i_d.di_nextents + ip->i_d.di_anextents, 3601 ip->i_d.di_nblocks, ip); 3602 goto corrupt_out; 3603 } 3604 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3605 mp, XFS_ERRTAG_IFLUSH_6)) { 3606 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3607 "%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT, 3608 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3609 goto corrupt_out; 3610 } 3611 3612 /* 3613 * Inode item log recovery for v2 inodes are dependent on the 3614 * di_flushiter count for correct sequencing. We bump the flush 3615 * iteration count so we can detect flushes which postdate a log record 3616 * during recovery. This is redundant as we now log every change and 3617 * hence this can't happen but we need to still do it to ensure 3618 * backwards compatibility with old kernels that predate logging all 3619 * inode changes. 3620 */ 3621 if (ip->i_d.di_version < 3) 3622 ip->i_d.di_flushiter++; 3623 3624 /* Check the inline fork data before we write out. */ 3625 if (!xfs_inode_verify_forks(ip)) 3626 goto corrupt_out; 3627 3628 /* 3629 * Copy the dirty parts of the inode into the on-disk inode. We always 3630 * copy out the core of the inode, because if the inode is dirty at all 3631 * the core must be. 3632 */ 3633 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3634 3635 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3636 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3637 ip->i_d.di_flushiter = 0; 3638 3639 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3640 if (XFS_IFORK_Q(ip)) 3641 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3642 xfs_inobp_check(mp, bp); 3643 3644 /* 3645 * We've recorded everything logged in the inode, so we'd like to clear 3646 * the ili_fields bits so we don't log and flush things unnecessarily. 3647 * However, we can't stop logging all this information until the data 3648 * we've copied into the disk buffer is written to disk. If we did we 3649 * might overwrite the copy of the inode in the log with all the data 3650 * after re-logging only part of it, and in the face of a crash we 3651 * wouldn't have all the data we need to recover. 3652 * 3653 * What we do is move the bits to the ili_last_fields field. When 3654 * logging the inode, these bits are moved back to the ili_fields field. 3655 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3656 * know that the information those bits represent is permanently on 3657 * disk. As long as the flush completes before the inode is logged 3658 * again, then both ili_fields and ili_last_fields will be cleared. 3659 * 3660 * We can play with the ili_fields bits here, because the inode lock 3661 * must be held exclusively in order to set bits there and the flush 3662 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3663 * done routine can tell whether or not to look in the AIL. Also, store 3664 * the current LSN of the inode so that we can tell whether the item has 3665 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3666 * need the AIL lock, because it is a 64 bit value that cannot be read 3667 * atomically. 3668 */ 3669 iip->ili_last_fields = iip->ili_fields; 3670 iip->ili_fields = 0; 3671 iip->ili_fsync_fields = 0; 3672 iip->ili_logged = 1; 3673 3674 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3675 &iip->ili_item.li_lsn); 3676 3677 /* 3678 * Attach the function xfs_iflush_done to the inode's 3679 * buffer. This will remove the inode from the AIL 3680 * and unlock the inode's flush lock when the inode is 3681 * completely written to disk. 3682 */ 3683 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3684 3685 /* generate the checksum. */ 3686 xfs_dinode_calc_crc(mp, dip); 3687 3688 ASSERT(!list_empty(&bp->b_li_list)); 3689 ASSERT(bp->b_iodone != NULL); 3690 return 0; 3691 3692 corrupt_out: 3693 return -EFSCORRUPTED; 3694 } 3695