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 20 #include "xfs.h" 21 #include "xfs_fs.h" 22 #include "xfs_shared.h" 23 #include "xfs_format.h" 24 #include "xfs_log_format.h" 25 #include "xfs_trans_resv.h" 26 #include "xfs_sb.h" 27 #include "xfs_mount.h" 28 #include "xfs_defer.h" 29 #include "xfs_inode.h" 30 #include "xfs_da_format.h" 31 #include "xfs_da_btree.h" 32 #include "xfs_dir2.h" 33 #include "xfs_attr_sf.h" 34 #include "xfs_attr.h" 35 #include "xfs_trans_space.h" 36 #include "xfs_trans.h" 37 #include "xfs_buf_item.h" 38 #include "xfs_inode_item.h" 39 #include "xfs_ialloc.h" 40 #include "xfs_bmap.h" 41 #include "xfs_bmap_util.h" 42 #include "xfs_error.h" 43 #include "xfs_quota.h" 44 #include "xfs_filestream.h" 45 #include "xfs_cksum.h" 46 #include "xfs_trace.h" 47 #include "xfs_icache.h" 48 #include "xfs_symlink.h" 49 #include "xfs_trans_priv.h" 50 #include "xfs_log.h" 51 #include "xfs_bmap_btree.h" 52 #include "xfs_reflink.h" 53 #include "xfs_dir2_priv.h" 54 55 kmem_zone_t *xfs_inode_zone; 56 57 /* 58 * Used in xfs_itruncate_extents(). This is the maximum number of extents 59 * freed from a file in a single transaction. 60 */ 61 #define XFS_ITRUNC_MAX_EXTENTS 2 62 63 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); 64 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); 65 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); 66 67 /* 68 * helper function to extract extent size hint from inode 69 */ 70 xfs_extlen_t 71 xfs_get_extsz_hint( 72 struct xfs_inode *ip) 73 { 74 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 75 return ip->i_d.di_extsize; 76 if (XFS_IS_REALTIME_INODE(ip)) 77 return ip->i_mount->m_sb.sb_rextsize; 78 return 0; 79 } 80 81 /* 82 * Helper function to extract CoW extent size hint from inode. 83 * Between the extent size hint and the CoW extent size hint, we 84 * return the greater of the two. If the value is zero (automatic), 85 * use the default size. 86 */ 87 xfs_extlen_t 88 xfs_get_cowextsz_hint( 89 struct xfs_inode *ip) 90 { 91 xfs_extlen_t a, b; 92 93 a = 0; 94 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 95 a = ip->i_d.di_cowextsize; 96 b = xfs_get_extsz_hint(ip); 97 98 a = max(a, b); 99 if (a == 0) 100 return XFS_DEFAULT_COWEXTSZ_HINT; 101 return a; 102 } 103 104 /* 105 * These two are wrapper routines around the xfs_ilock() routine used to 106 * centralize some grungy code. They are used in places that wish to lock the 107 * inode solely for reading the extents. The reason these places can't just 108 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to 109 * bringing in of the extents from disk for a file in b-tree format. If the 110 * inode is in b-tree format, then we need to lock the inode exclusively until 111 * the extents are read in. Locking it exclusively all the time would limit 112 * our parallelism unnecessarily, though. What we do instead is check to see 113 * if the extents have been read in yet, and only lock the inode exclusively 114 * if they have not. 115 * 116 * The functions return a value which should be given to the corresponding 117 * xfs_iunlock() call. 118 */ 119 uint 120 xfs_ilock_data_map_shared( 121 struct xfs_inode *ip) 122 { 123 uint lock_mode = XFS_ILOCK_SHARED; 124 125 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 126 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) 127 lock_mode = XFS_ILOCK_EXCL; 128 xfs_ilock(ip, lock_mode); 129 return lock_mode; 130 } 131 132 uint 133 xfs_ilock_attr_map_shared( 134 struct xfs_inode *ip) 135 { 136 uint lock_mode = XFS_ILOCK_SHARED; 137 138 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && 139 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) 140 lock_mode = XFS_ILOCK_EXCL; 141 xfs_ilock(ip, lock_mode); 142 return lock_mode; 143 } 144 145 /* 146 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 147 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows 148 * various combinations of the locks to be obtained. 149 * 150 * The 3 locks should always be ordered so that the IO lock is obtained first, 151 * the mmap lock second and the ilock last in order to prevent deadlock. 152 * 153 * Basic locking order: 154 * 155 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock 156 * 157 * mmap_sem locking order: 158 * 159 * i_rwsem -> page lock -> mmap_sem 160 * mmap_sem -> i_mmap_lock -> page_lock 161 * 162 * The difference in mmap_sem locking order mean that we cannot hold the 163 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can 164 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem 165 * in get_user_pages() to map the user pages into the kernel address space for 166 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because 167 * page faults already hold the mmap_sem. 168 * 169 * Hence to serialise fully against both syscall and mmap based IO, we need to 170 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both 171 * taken in places where we need to invalidate the page cache in a race 172 * free manner (e.g. truncate, hole punch and other extent manipulation 173 * functions). 174 */ 175 void 176 xfs_ilock( 177 xfs_inode_t *ip, 178 uint lock_flags) 179 { 180 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 181 182 /* 183 * You can't set both SHARED and EXCL for the same lock, 184 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 185 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 186 */ 187 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 188 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 189 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 190 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 191 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 192 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 193 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 194 195 if (lock_flags & XFS_IOLOCK_EXCL) { 196 down_write_nested(&VFS_I(ip)->i_rwsem, 197 XFS_IOLOCK_DEP(lock_flags)); 198 } else if (lock_flags & XFS_IOLOCK_SHARED) { 199 down_read_nested(&VFS_I(ip)->i_rwsem, 200 XFS_IOLOCK_DEP(lock_flags)); 201 } 202 203 if (lock_flags & XFS_MMAPLOCK_EXCL) 204 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 205 else if (lock_flags & XFS_MMAPLOCK_SHARED) 206 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 207 208 if (lock_flags & XFS_ILOCK_EXCL) 209 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 210 else if (lock_flags & XFS_ILOCK_SHARED) 211 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 212 } 213 214 /* 215 * This is just like xfs_ilock(), except that the caller 216 * is guaranteed not to sleep. It returns 1 if it gets 217 * the requested locks and 0 otherwise. If the IO lock is 218 * obtained but the inode lock cannot be, then the IO lock 219 * is dropped before returning. 220 * 221 * ip -- the inode being locked 222 * lock_flags -- this parameter indicates the inode's locks to be 223 * to be locked. See the comment for xfs_ilock() for a list 224 * of valid values. 225 */ 226 int 227 xfs_ilock_nowait( 228 xfs_inode_t *ip, 229 uint lock_flags) 230 { 231 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 232 233 /* 234 * You can't set both SHARED and EXCL for the same lock, 235 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 236 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 237 */ 238 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 239 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 240 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 241 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 242 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 243 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 244 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 245 246 if (lock_flags & XFS_IOLOCK_EXCL) { 247 if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) 248 goto out; 249 } else if (lock_flags & XFS_IOLOCK_SHARED) { 250 if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) 251 goto out; 252 } 253 254 if (lock_flags & XFS_MMAPLOCK_EXCL) { 255 if (!mrtryupdate(&ip->i_mmaplock)) 256 goto out_undo_iolock; 257 } else if (lock_flags & XFS_MMAPLOCK_SHARED) { 258 if (!mrtryaccess(&ip->i_mmaplock)) 259 goto out_undo_iolock; 260 } 261 262 if (lock_flags & XFS_ILOCK_EXCL) { 263 if (!mrtryupdate(&ip->i_lock)) 264 goto out_undo_mmaplock; 265 } else if (lock_flags & XFS_ILOCK_SHARED) { 266 if (!mrtryaccess(&ip->i_lock)) 267 goto out_undo_mmaplock; 268 } 269 return 1; 270 271 out_undo_mmaplock: 272 if (lock_flags & XFS_MMAPLOCK_EXCL) 273 mrunlock_excl(&ip->i_mmaplock); 274 else if (lock_flags & XFS_MMAPLOCK_SHARED) 275 mrunlock_shared(&ip->i_mmaplock); 276 out_undo_iolock: 277 if (lock_flags & XFS_IOLOCK_EXCL) 278 up_write(&VFS_I(ip)->i_rwsem); 279 else if (lock_flags & XFS_IOLOCK_SHARED) 280 up_read(&VFS_I(ip)->i_rwsem); 281 out: 282 return 0; 283 } 284 285 /* 286 * xfs_iunlock() is used to drop the inode locks acquired with 287 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 288 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 289 * that we know which locks to drop. 290 * 291 * ip -- the inode being unlocked 292 * lock_flags -- this parameter indicates the inode's locks to be 293 * to be unlocked. See the comment for xfs_ilock() for a list 294 * of valid values for this parameter. 295 * 296 */ 297 void 298 xfs_iunlock( 299 xfs_inode_t *ip, 300 uint lock_flags) 301 { 302 /* 303 * You can't set both SHARED and EXCL for the same lock, 304 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 305 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 306 */ 307 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 308 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 309 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 310 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 311 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 312 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 313 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 314 ASSERT(lock_flags != 0); 315 316 if (lock_flags & XFS_IOLOCK_EXCL) 317 up_write(&VFS_I(ip)->i_rwsem); 318 else if (lock_flags & XFS_IOLOCK_SHARED) 319 up_read(&VFS_I(ip)->i_rwsem); 320 321 if (lock_flags & XFS_MMAPLOCK_EXCL) 322 mrunlock_excl(&ip->i_mmaplock); 323 else if (lock_flags & XFS_MMAPLOCK_SHARED) 324 mrunlock_shared(&ip->i_mmaplock); 325 326 if (lock_flags & XFS_ILOCK_EXCL) 327 mrunlock_excl(&ip->i_lock); 328 else if (lock_flags & XFS_ILOCK_SHARED) 329 mrunlock_shared(&ip->i_lock); 330 331 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 332 } 333 334 /* 335 * give up write locks. the i/o lock cannot be held nested 336 * if it is being demoted. 337 */ 338 void 339 xfs_ilock_demote( 340 xfs_inode_t *ip, 341 uint lock_flags) 342 { 343 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); 344 ASSERT((lock_flags & 345 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 346 347 if (lock_flags & XFS_ILOCK_EXCL) 348 mrdemote(&ip->i_lock); 349 if (lock_flags & XFS_MMAPLOCK_EXCL) 350 mrdemote(&ip->i_mmaplock); 351 if (lock_flags & XFS_IOLOCK_EXCL) 352 downgrade_write(&VFS_I(ip)->i_rwsem); 353 354 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 355 } 356 357 #if defined(DEBUG) || defined(XFS_WARN) 358 int 359 xfs_isilocked( 360 xfs_inode_t *ip, 361 uint lock_flags) 362 { 363 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 364 if (!(lock_flags & XFS_ILOCK_SHARED)) 365 return !!ip->i_lock.mr_writer; 366 return rwsem_is_locked(&ip->i_lock.mr_lock); 367 } 368 369 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { 370 if (!(lock_flags & XFS_MMAPLOCK_SHARED)) 371 return !!ip->i_mmaplock.mr_writer; 372 return rwsem_is_locked(&ip->i_mmaplock.mr_lock); 373 } 374 375 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 376 if (!(lock_flags & XFS_IOLOCK_SHARED)) 377 return !debug_locks || 378 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); 379 return rwsem_is_locked(&VFS_I(ip)->i_rwsem); 380 } 381 382 ASSERT(0); 383 return 0; 384 } 385 #endif 386 387 #ifdef DEBUG 388 int xfs_locked_n; 389 int xfs_small_retries; 390 int xfs_middle_retries; 391 int xfs_lots_retries; 392 int xfs_lock_delays; 393 #endif 394 395 /* 396 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when 397 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined 398 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build 399 * errors and warnings. 400 */ 401 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) 402 static bool 403 xfs_lockdep_subclass_ok( 404 int subclass) 405 { 406 return subclass < MAX_LOCKDEP_SUBCLASSES; 407 } 408 #else 409 #define xfs_lockdep_subclass_ok(subclass) (true) 410 #endif 411 412 /* 413 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 414 * value. This can be called for any type of inode lock combination, including 415 * parent locking. Care must be taken to ensure we don't overrun the subclass 416 * storage fields in the class mask we build. 417 */ 418 static inline int 419 xfs_lock_inumorder(int lock_mode, int subclass) 420 { 421 int class = 0; 422 423 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | 424 XFS_ILOCK_RTSUM))); 425 ASSERT(xfs_lockdep_subclass_ok(subclass)); 426 427 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 428 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); 429 class += subclass << XFS_IOLOCK_SHIFT; 430 } 431 432 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 433 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); 434 class += subclass << XFS_MMAPLOCK_SHIFT; 435 } 436 437 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { 438 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); 439 class += subclass << XFS_ILOCK_SHIFT; 440 } 441 442 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; 443 } 444 445 /* 446 * The following routine will lock n inodes in exclusive mode. We assume the 447 * caller calls us with the inodes in i_ino order. 448 * 449 * We need to detect deadlock where an inode that we lock is in the AIL and we 450 * start waiting for another inode that is locked by a thread in a long running 451 * transaction (such as truncate). This can result in deadlock since the long 452 * running trans might need to wait for the inode we just locked in order to 453 * push the tail and free space in the log. 454 * 455 * xfs_lock_inodes() can only be used to lock one type of lock at a time - 456 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 457 * lock more than one at a time, lockdep will report false positives saying we 458 * have violated locking orders. 459 */ 460 static void 461 xfs_lock_inodes( 462 xfs_inode_t **ips, 463 int inodes, 464 uint lock_mode) 465 { 466 int attempts = 0, i, j, try_lock; 467 xfs_log_item_t *lp; 468 469 /* 470 * Currently supports between 2 and 5 inodes with exclusive locking. We 471 * support an arbitrary depth of locking here, but absolute limits on 472 * inodes depend on the the type of locking and the limits placed by 473 * lockdep annotations in xfs_lock_inumorder. These are all checked by 474 * the asserts. 475 */ 476 ASSERT(ips && inodes >= 2 && inodes <= 5); 477 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | 478 XFS_ILOCK_EXCL)); 479 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | 480 XFS_ILOCK_SHARED))); 481 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || 482 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); 483 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || 484 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); 485 486 if (lock_mode & XFS_IOLOCK_EXCL) { 487 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); 488 } else if (lock_mode & XFS_MMAPLOCK_EXCL) 489 ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); 490 491 try_lock = 0; 492 i = 0; 493 again: 494 for (; i < inodes; i++) { 495 ASSERT(ips[i]); 496 497 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 498 continue; 499 500 /* 501 * If try_lock is not set yet, make sure all locked inodes are 502 * not in the AIL. If any are, set try_lock to be used later. 503 */ 504 if (!try_lock) { 505 for (j = (i - 1); j >= 0 && !try_lock; j--) { 506 lp = (xfs_log_item_t *)ips[j]->i_itemp; 507 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) 508 try_lock++; 509 } 510 } 511 512 /* 513 * If any of the previous locks we have locked is in the AIL, 514 * we must TRY to get the second and subsequent locks. If 515 * we can't get any, we must release all we have 516 * and try again. 517 */ 518 if (!try_lock) { 519 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 520 continue; 521 } 522 523 /* try_lock means we have an inode locked that is in the AIL. */ 524 ASSERT(i != 0); 525 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 526 continue; 527 528 /* 529 * Unlock all previous guys and try again. xfs_iunlock will try 530 * to push the tail if the inode is in the AIL. 531 */ 532 attempts++; 533 for (j = i - 1; j >= 0; j--) { 534 /* 535 * Check to see if we've already unlocked this one. Not 536 * the first one going back, and the inode ptr is the 537 * same. 538 */ 539 if (j != (i - 1) && ips[j] == ips[j + 1]) 540 continue; 541 542 xfs_iunlock(ips[j], lock_mode); 543 } 544 545 if ((attempts % 5) == 0) { 546 delay(1); /* Don't just spin the CPU */ 547 #ifdef DEBUG 548 xfs_lock_delays++; 549 #endif 550 } 551 i = 0; 552 try_lock = 0; 553 goto again; 554 } 555 556 #ifdef DEBUG 557 if (attempts) { 558 if (attempts < 5) xfs_small_retries++; 559 else if (attempts < 100) xfs_middle_retries++; 560 else xfs_lots_retries++; 561 } else { 562 xfs_locked_n++; 563 } 564 #endif 565 } 566 567 /* 568 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 569 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 570 * lock more than one at a time, lockdep will report false positives saying we 571 * have violated locking orders. 572 */ 573 void 574 xfs_lock_two_inodes( 575 xfs_inode_t *ip0, 576 xfs_inode_t *ip1, 577 uint lock_mode) 578 { 579 xfs_inode_t *temp; 580 int attempts = 0; 581 xfs_log_item_t *lp; 582 583 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); 584 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) 585 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 586 587 ASSERT(ip0->i_ino != ip1->i_ino); 588 589 if (ip0->i_ino > ip1->i_ino) { 590 temp = ip0; 591 ip0 = ip1; 592 ip1 = temp; 593 } 594 595 again: 596 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); 597 598 /* 599 * If the first lock we have locked is in the AIL, we must TRY to get 600 * the second lock. If we can't get it, we must release the first one 601 * and try again. 602 */ 603 lp = (xfs_log_item_t *)ip0->i_itemp; 604 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { 605 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { 606 xfs_iunlock(ip0, lock_mode); 607 if ((++attempts % 5) == 0) 608 delay(1); /* Don't just spin the CPU */ 609 goto again; 610 } 611 } else { 612 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); 613 } 614 } 615 616 617 void 618 __xfs_iflock( 619 struct xfs_inode *ip) 620 { 621 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 622 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 623 624 do { 625 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 626 if (xfs_isiflocked(ip)) 627 io_schedule(); 628 } while (!xfs_iflock_nowait(ip)); 629 630 finish_wait(wq, &wait.wq_entry); 631 } 632 633 STATIC uint 634 _xfs_dic2xflags( 635 uint16_t di_flags, 636 uint64_t di_flags2, 637 bool has_attr) 638 { 639 uint flags = 0; 640 641 if (di_flags & XFS_DIFLAG_ANY) { 642 if (di_flags & XFS_DIFLAG_REALTIME) 643 flags |= FS_XFLAG_REALTIME; 644 if (di_flags & XFS_DIFLAG_PREALLOC) 645 flags |= FS_XFLAG_PREALLOC; 646 if (di_flags & XFS_DIFLAG_IMMUTABLE) 647 flags |= FS_XFLAG_IMMUTABLE; 648 if (di_flags & XFS_DIFLAG_APPEND) 649 flags |= FS_XFLAG_APPEND; 650 if (di_flags & XFS_DIFLAG_SYNC) 651 flags |= FS_XFLAG_SYNC; 652 if (di_flags & XFS_DIFLAG_NOATIME) 653 flags |= FS_XFLAG_NOATIME; 654 if (di_flags & XFS_DIFLAG_NODUMP) 655 flags |= FS_XFLAG_NODUMP; 656 if (di_flags & XFS_DIFLAG_RTINHERIT) 657 flags |= FS_XFLAG_RTINHERIT; 658 if (di_flags & XFS_DIFLAG_PROJINHERIT) 659 flags |= FS_XFLAG_PROJINHERIT; 660 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 661 flags |= FS_XFLAG_NOSYMLINKS; 662 if (di_flags & XFS_DIFLAG_EXTSIZE) 663 flags |= FS_XFLAG_EXTSIZE; 664 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 665 flags |= FS_XFLAG_EXTSZINHERIT; 666 if (di_flags & XFS_DIFLAG_NODEFRAG) 667 flags |= FS_XFLAG_NODEFRAG; 668 if (di_flags & XFS_DIFLAG_FILESTREAM) 669 flags |= FS_XFLAG_FILESTREAM; 670 } 671 672 if (di_flags2 & XFS_DIFLAG2_ANY) { 673 if (di_flags2 & XFS_DIFLAG2_DAX) 674 flags |= FS_XFLAG_DAX; 675 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 676 flags |= FS_XFLAG_COWEXTSIZE; 677 } 678 679 if (has_attr) 680 flags |= FS_XFLAG_HASATTR; 681 682 return flags; 683 } 684 685 uint 686 xfs_ip2xflags( 687 struct xfs_inode *ip) 688 { 689 struct xfs_icdinode *dic = &ip->i_d; 690 691 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); 692 } 693 694 /* 695 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 696 * is allowed, otherwise it has to be an exact match. If a CI match is found, 697 * ci_name->name will point to a the actual name (caller must free) or 698 * will be set to NULL if an exact match is found. 699 */ 700 int 701 xfs_lookup( 702 xfs_inode_t *dp, 703 struct xfs_name *name, 704 xfs_inode_t **ipp, 705 struct xfs_name *ci_name) 706 { 707 xfs_ino_t inum; 708 int error; 709 710 trace_xfs_lookup(dp, name); 711 712 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 713 return -EIO; 714 715 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 716 if (error) 717 goto out_unlock; 718 719 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 720 if (error) 721 goto out_free_name; 722 723 return 0; 724 725 out_free_name: 726 if (ci_name) 727 kmem_free(ci_name->name); 728 out_unlock: 729 *ipp = NULL; 730 return error; 731 } 732 733 /* 734 * Allocate an inode on disk and return a copy of its in-core version. 735 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 736 * appropriately within the inode. The uid and gid for the inode are 737 * set according to the contents of the given cred structure. 738 * 739 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 740 * has a free inode available, call xfs_iget() to obtain the in-core 741 * version of the allocated inode. Finally, fill in the inode and 742 * log its initial contents. In this case, ialloc_context would be 743 * set to NULL. 744 * 745 * If xfs_dialloc() does not have an available inode, it will replenish 746 * its supply by doing an allocation. Since we can only do one 747 * allocation within a transaction without deadlocks, we must commit 748 * the current transaction before returning the inode itself. 749 * In this case, therefore, we will set ialloc_context and return. 750 * The caller should then commit the current transaction, start a new 751 * transaction, and call xfs_ialloc() again to actually get the inode. 752 * 753 * To ensure that some other process does not grab the inode that 754 * was allocated during the first call to xfs_ialloc(), this routine 755 * also returns the [locked] bp pointing to the head of the freelist 756 * as ialloc_context. The caller should hold this buffer across 757 * the commit and pass it back into this routine on the second call. 758 * 759 * If we are allocating quota inodes, we do not have a parent inode 760 * to attach to or associate with (i.e. pip == NULL) because they 761 * are not linked into the directory structure - they are attached 762 * directly to the superblock - and so have no parent. 763 */ 764 static int 765 xfs_ialloc( 766 xfs_trans_t *tp, 767 xfs_inode_t *pip, 768 umode_t mode, 769 xfs_nlink_t nlink, 770 xfs_dev_t rdev, 771 prid_t prid, 772 int okalloc, 773 xfs_buf_t **ialloc_context, 774 xfs_inode_t **ipp) 775 { 776 struct xfs_mount *mp = tp->t_mountp; 777 xfs_ino_t ino; 778 xfs_inode_t *ip; 779 uint flags; 780 int error; 781 struct timespec tv; 782 struct inode *inode; 783 784 /* 785 * Call the space management code to pick 786 * the on-disk inode to be allocated. 787 */ 788 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 789 ialloc_context, &ino); 790 if (error) 791 return error; 792 if (*ialloc_context || ino == NULLFSINO) { 793 *ipp = NULL; 794 return 0; 795 } 796 ASSERT(*ialloc_context == NULL); 797 798 /* 799 * Get the in-core inode with the lock held exclusively. 800 * This is because we're setting fields here we need 801 * to prevent others from looking at until we're done. 802 */ 803 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 804 XFS_ILOCK_EXCL, &ip); 805 if (error) 806 return error; 807 ASSERT(ip != NULL); 808 inode = VFS_I(ip); 809 810 /* 811 * We always convert v1 inodes to v2 now - we only support filesystems 812 * with >= v2 inode capability, so there is no reason for ever leaving 813 * an inode in v1 format. 814 */ 815 if (ip->i_d.di_version == 1) 816 ip->i_d.di_version = 2; 817 818 inode->i_mode = mode; 819 set_nlink(inode, nlink); 820 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 821 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 822 xfs_set_projid(ip, prid); 823 824 if (pip && XFS_INHERIT_GID(pip)) { 825 ip->i_d.di_gid = pip->i_d.di_gid; 826 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) 827 inode->i_mode |= S_ISGID; 828 } 829 830 /* 831 * If the group ID of the new file does not match the effective group 832 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 833 * (and only if the irix_sgid_inherit compatibility variable is set). 834 */ 835 if ((irix_sgid_inherit) && 836 (inode->i_mode & S_ISGID) && 837 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) 838 inode->i_mode &= ~S_ISGID; 839 840 ip->i_d.di_size = 0; 841 ip->i_d.di_nextents = 0; 842 ASSERT(ip->i_d.di_nblocks == 0); 843 844 tv = current_time(inode); 845 inode->i_mtime = tv; 846 inode->i_atime = tv; 847 inode->i_ctime = tv; 848 849 ip->i_d.di_extsize = 0; 850 ip->i_d.di_dmevmask = 0; 851 ip->i_d.di_dmstate = 0; 852 ip->i_d.di_flags = 0; 853 854 if (ip->i_d.di_version == 3) { 855 inode->i_version = 1; 856 ip->i_d.di_flags2 = 0; 857 ip->i_d.di_cowextsize = 0; 858 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; 859 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; 860 } 861 862 863 flags = XFS_ILOG_CORE; 864 switch (mode & S_IFMT) { 865 case S_IFIFO: 866 case S_IFCHR: 867 case S_IFBLK: 868 case S_IFSOCK: 869 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 870 ip->i_df.if_u2.if_rdev = rdev; 871 ip->i_df.if_flags = 0; 872 flags |= XFS_ILOG_DEV; 873 break; 874 case S_IFREG: 875 case S_IFDIR: 876 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 877 uint di_flags = 0; 878 879 if (S_ISDIR(mode)) { 880 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 881 di_flags |= XFS_DIFLAG_RTINHERIT; 882 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 883 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 884 ip->i_d.di_extsize = pip->i_d.di_extsize; 885 } 886 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 887 di_flags |= XFS_DIFLAG_PROJINHERIT; 888 } else if (S_ISREG(mode)) { 889 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 890 di_flags |= XFS_DIFLAG_REALTIME; 891 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 892 di_flags |= XFS_DIFLAG_EXTSIZE; 893 ip->i_d.di_extsize = pip->i_d.di_extsize; 894 } 895 } 896 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 897 xfs_inherit_noatime) 898 di_flags |= XFS_DIFLAG_NOATIME; 899 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 900 xfs_inherit_nodump) 901 di_flags |= XFS_DIFLAG_NODUMP; 902 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 903 xfs_inherit_sync) 904 di_flags |= XFS_DIFLAG_SYNC; 905 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 906 xfs_inherit_nosymlinks) 907 di_flags |= XFS_DIFLAG_NOSYMLINKS; 908 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 909 xfs_inherit_nodefrag) 910 di_flags |= XFS_DIFLAG_NODEFRAG; 911 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 912 di_flags |= XFS_DIFLAG_FILESTREAM; 913 914 ip->i_d.di_flags |= di_flags; 915 } 916 if (pip && 917 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && 918 pip->i_d.di_version == 3 && 919 ip->i_d.di_version == 3) { 920 uint64_t di_flags2 = 0; 921 922 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { 923 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; 924 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; 925 } 926 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 927 di_flags2 |= XFS_DIFLAG2_DAX; 928 929 ip->i_d.di_flags2 |= di_flags2; 930 } 931 /* FALLTHROUGH */ 932 case S_IFLNK: 933 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 934 ip->i_df.if_flags = XFS_IFEXTENTS; 935 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 936 ip->i_df.if_u1.if_extents = NULL; 937 break; 938 default: 939 ASSERT(0); 940 } 941 /* 942 * Attribute fork settings for new inode. 943 */ 944 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 945 ip->i_d.di_anextents = 0; 946 947 /* 948 * Log the new values stuffed into the inode. 949 */ 950 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 951 xfs_trans_log_inode(tp, ip, flags); 952 953 /* now that we have an i_mode we can setup the inode structure */ 954 xfs_setup_inode(ip); 955 956 *ipp = ip; 957 return 0; 958 } 959 960 /* 961 * Allocates a new inode from disk and return a pointer to the 962 * incore copy. This routine will internally commit the current 963 * transaction and allocate a new one if the Space Manager needed 964 * to do an allocation to replenish the inode free-list. 965 * 966 * This routine is designed to be called from xfs_create and 967 * xfs_create_dir. 968 * 969 */ 970 int 971 xfs_dir_ialloc( 972 xfs_trans_t **tpp, /* input: current transaction; 973 output: may be a new transaction. */ 974 xfs_inode_t *dp, /* directory within whose allocate 975 the inode. */ 976 umode_t mode, 977 xfs_nlink_t nlink, 978 xfs_dev_t rdev, 979 prid_t prid, /* project id */ 980 int okalloc, /* ok to allocate new space */ 981 xfs_inode_t **ipp, /* pointer to inode; it will be 982 locked. */ 983 int *committed) 984 985 { 986 xfs_trans_t *tp; 987 xfs_inode_t *ip; 988 xfs_buf_t *ialloc_context = NULL; 989 int code; 990 void *dqinfo; 991 uint tflags; 992 993 tp = *tpp; 994 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 995 996 /* 997 * xfs_ialloc will return a pointer to an incore inode if 998 * the Space Manager has an available inode on the free 999 * list. Otherwise, it will do an allocation and replenish 1000 * the freelist. Since we can only do one allocation per 1001 * transaction without deadlocks, we will need to commit the 1002 * current transaction and start a new one. We will then 1003 * need to call xfs_ialloc again to get the inode. 1004 * 1005 * If xfs_ialloc did an allocation to replenish the freelist, 1006 * it returns the bp containing the head of the freelist as 1007 * ialloc_context. We will hold a lock on it across the 1008 * transaction commit so that no other process can steal 1009 * the inode(s) that we've just allocated. 1010 */ 1011 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 1012 &ialloc_context, &ip); 1013 1014 /* 1015 * Return an error if we were unable to allocate a new inode. 1016 * This should only happen if we run out of space on disk or 1017 * encounter a disk error. 1018 */ 1019 if (code) { 1020 *ipp = NULL; 1021 return code; 1022 } 1023 if (!ialloc_context && !ip) { 1024 *ipp = NULL; 1025 return -ENOSPC; 1026 } 1027 1028 /* 1029 * If the AGI buffer is non-NULL, then we were unable to get an 1030 * inode in one operation. We need to commit the current 1031 * transaction and call xfs_ialloc() again. It is guaranteed 1032 * to succeed the second time. 1033 */ 1034 if (ialloc_context) { 1035 /* 1036 * Normally, xfs_trans_commit releases all the locks. 1037 * We call bhold to hang on to the ialloc_context across 1038 * the commit. Holding this buffer prevents any other 1039 * processes from doing any allocations in this 1040 * allocation group. 1041 */ 1042 xfs_trans_bhold(tp, ialloc_context); 1043 1044 /* 1045 * We want the quota changes to be associated with the next 1046 * transaction, NOT this one. So, detach the dqinfo from this 1047 * and attach it to the next transaction. 1048 */ 1049 dqinfo = NULL; 1050 tflags = 0; 1051 if (tp->t_dqinfo) { 1052 dqinfo = (void *)tp->t_dqinfo; 1053 tp->t_dqinfo = NULL; 1054 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1055 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1056 } 1057 1058 code = xfs_trans_roll(&tp, NULL); 1059 if (committed != NULL) 1060 *committed = 1; 1061 1062 /* 1063 * Re-attach the quota info that we detached from prev trx. 1064 */ 1065 if (dqinfo) { 1066 tp->t_dqinfo = dqinfo; 1067 tp->t_flags |= tflags; 1068 } 1069 1070 if (code) { 1071 xfs_buf_relse(ialloc_context); 1072 *tpp = tp; 1073 *ipp = NULL; 1074 return code; 1075 } 1076 xfs_trans_bjoin(tp, ialloc_context); 1077 1078 /* 1079 * Call ialloc again. Since we've locked out all 1080 * other allocations in this allocation group, 1081 * this call should always succeed. 1082 */ 1083 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1084 okalloc, &ialloc_context, &ip); 1085 1086 /* 1087 * If we get an error at this point, return to the caller 1088 * so that the current transaction can be aborted. 1089 */ 1090 if (code) { 1091 *tpp = tp; 1092 *ipp = NULL; 1093 return code; 1094 } 1095 ASSERT(!ialloc_context && ip); 1096 1097 } else { 1098 if (committed != NULL) 1099 *committed = 0; 1100 } 1101 1102 *ipp = ip; 1103 *tpp = tp; 1104 1105 return 0; 1106 } 1107 1108 /* 1109 * Decrement the link count on an inode & log the change. If this causes the 1110 * link count to go to zero, move the inode to AGI unlinked list so that it can 1111 * be freed when the last active reference goes away via xfs_inactive(). 1112 */ 1113 static int /* error */ 1114 xfs_droplink( 1115 xfs_trans_t *tp, 1116 xfs_inode_t *ip) 1117 { 1118 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1119 1120 drop_nlink(VFS_I(ip)); 1121 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1122 1123 if (VFS_I(ip)->i_nlink) 1124 return 0; 1125 1126 return xfs_iunlink(tp, ip); 1127 } 1128 1129 /* 1130 * Increment the link count on an inode & log the change. 1131 */ 1132 static int 1133 xfs_bumplink( 1134 xfs_trans_t *tp, 1135 xfs_inode_t *ip) 1136 { 1137 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1138 1139 ASSERT(ip->i_d.di_version > 1); 1140 inc_nlink(VFS_I(ip)); 1141 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1142 return 0; 1143 } 1144 1145 int 1146 xfs_create( 1147 xfs_inode_t *dp, 1148 struct xfs_name *name, 1149 umode_t mode, 1150 xfs_dev_t rdev, 1151 xfs_inode_t **ipp) 1152 { 1153 int is_dir = S_ISDIR(mode); 1154 struct xfs_mount *mp = dp->i_mount; 1155 struct xfs_inode *ip = NULL; 1156 struct xfs_trans *tp = NULL; 1157 int error; 1158 struct xfs_defer_ops dfops; 1159 xfs_fsblock_t first_block; 1160 bool unlock_dp_on_error = false; 1161 prid_t prid; 1162 struct xfs_dquot *udqp = NULL; 1163 struct xfs_dquot *gdqp = NULL; 1164 struct xfs_dquot *pdqp = NULL; 1165 struct xfs_trans_res *tres; 1166 uint resblks; 1167 1168 trace_xfs_create(dp, name); 1169 1170 if (XFS_FORCED_SHUTDOWN(mp)) 1171 return -EIO; 1172 1173 prid = xfs_get_initial_prid(dp); 1174 1175 /* 1176 * Make sure that we have allocated dquot(s) on disk. 1177 */ 1178 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1179 xfs_kgid_to_gid(current_fsgid()), prid, 1180 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1181 &udqp, &gdqp, &pdqp); 1182 if (error) 1183 return error; 1184 1185 if (is_dir) { 1186 rdev = 0; 1187 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1188 tres = &M_RES(mp)->tr_mkdir; 1189 } else { 1190 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1191 tres = &M_RES(mp)->tr_create; 1192 } 1193 1194 /* 1195 * Initially assume that the file does not exist and 1196 * reserve the resources for that case. If that is not 1197 * the case we'll drop the one we have and get a more 1198 * appropriate transaction later. 1199 */ 1200 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1201 if (error == -ENOSPC) { 1202 /* flush outstanding delalloc blocks and retry */ 1203 xfs_flush_inodes(mp); 1204 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1205 } 1206 if (error == -ENOSPC) { 1207 /* No space at all so try a "no-allocation" reservation */ 1208 resblks = 0; 1209 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1210 } 1211 if (error) 1212 goto out_release_inode; 1213 1214 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1215 unlock_dp_on_error = true; 1216 1217 xfs_defer_init(&dfops, &first_block); 1218 1219 /* 1220 * Reserve disk quota and the inode. 1221 */ 1222 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1223 pdqp, resblks, 1, 0); 1224 if (error) 1225 goto out_trans_cancel; 1226 1227 if (!resblks) { 1228 error = xfs_dir_canenter(tp, dp, name); 1229 if (error) 1230 goto out_trans_cancel; 1231 } 1232 1233 /* 1234 * A newly created regular or special file just has one directory 1235 * entry pointing to them, but a directory also the "." entry 1236 * pointing to itself. 1237 */ 1238 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1239 prid, resblks > 0, &ip, NULL); 1240 if (error) 1241 goto out_trans_cancel; 1242 1243 /* 1244 * Now we join the directory inode to the transaction. We do not do it 1245 * earlier because xfs_dir_ialloc might commit the previous transaction 1246 * (and release all the locks). An error from here on will result in 1247 * the transaction cancel unlocking dp so don't do it explicitly in the 1248 * error path. 1249 */ 1250 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1251 unlock_dp_on_error = false; 1252 1253 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1254 &first_block, &dfops, resblks ? 1255 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1256 if (error) { 1257 ASSERT(error != -ENOSPC); 1258 goto out_trans_cancel; 1259 } 1260 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1261 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1262 1263 if (is_dir) { 1264 error = xfs_dir_init(tp, ip, dp); 1265 if (error) 1266 goto out_bmap_cancel; 1267 1268 error = xfs_bumplink(tp, dp); 1269 if (error) 1270 goto out_bmap_cancel; 1271 } 1272 1273 /* 1274 * If this is a synchronous mount, make sure that the 1275 * create transaction goes to disk before returning to 1276 * the user. 1277 */ 1278 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1279 xfs_trans_set_sync(tp); 1280 1281 /* 1282 * Attach the dquot(s) to the inodes and modify them incore. 1283 * These ids of the inode couldn't have changed since the new 1284 * inode has been locked ever since it was created. 1285 */ 1286 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1287 1288 error = xfs_defer_finish(&tp, &dfops, NULL); 1289 if (error) 1290 goto out_bmap_cancel; 1291 1292 error = xfs_trans_commit(tp); 1293 if (error) 1294 goto out_release_inode; 1295 1296 xfs_qm_dqrele(udqp); 1297 xfs_qm_dqrele(gdqp); 1298 xfs_qm_dqrele(pdqp); 1299 1300 *ipp = ip; 1301 return 0; 1302 1303 out_bmap_cancel: 1304 xfs_defer_cancel(&dfops); 1305 out_trans_cancel: 1306 xfs_trans_cancel(tp); 1307 out_release_inode: 1308 /* 1309 * Wait until after the current transaction is aborted to finish the 1310 * setup of the inode and release the inode. This prevents recursive 1311 * transactions and deadlocks from xfs_inactive. 1312 */ 1313 if (ip) { 1314 xfs_finish_inode_setup(ip); 1315 IRELE(ip); 1316 } 1317 1318 xfs_qm_dqrele(udqp); 1319 xfs_qm_dqrele(gdqp); 1320 xfs_qm_dqrele(pdqp); 1321 1322 if (unlock_dp_on_error) 1323 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1324 return error; 1325 } 1326 1327 int 1328 xfs_create_tmpfile( 1329 struct xfs_inode *dp, 1330 struct dentry *dentry, 1331 umode_t mode, 1332 struct xfs_inode **ipp) 1333 { 1334 struct xfs_mount *mp = dp->i_mount; 1335 struct xfs_inode *ip = NULL; 1336 struct xfs_trans *tp = NULL; 1337 int error; 1338 prid_t prid; 1339 struct xfs_dquot *udqp = NULL; 1340 struct xfs_dquot *gdqp = NULL; 1341 struct xfs_dquot *pdqp = NULL; 1342 struct xfs_trans_res *tres; 1343 uint resblks; 1344 1345 if (XFS_FORCED_SHUTDOWN(mp)) 1346 return -EIO; 1347 1348 prid = xfs_get_initial_prid(dp); 1349 1350 /* 1351 * Make sure that we have allocated dquot(s) on disk. 1352 */ 1353 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1354 xfs_kgid_to_gid(current_fsgid()), prid, 1355 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1356 &udqp, &gdqp, &pdqp); 1357 if (error) 1358 return error; 1359 1360 resblks = XFS_IALLOC_SPACE_RES(mp); 1361 tres = &M_RES(mp)->tr_create_tmpfile; 1362 1363 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1364 if (error == -ENOSPC) { 1365 /* No space at all so try a "no-allocation" reservation */ 1366 resblks = 0; 1367 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1368 } 1369 if (error) 1370 goto out_release_inode; 1371 1372 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1373 pdqp, resblks, 1, 0); 1374 if (error) 1375 goto out_trans_cancel; 1376 1377 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1378 prid, resblks > 0, &ip, NULL); 1379 if (error) 1380 goto out_trans_cancel; 1381 1382 if (mp->m_flags & XFS_MOUNT_WSYNC) 1383 xfs_trans_set_sync(tp); 1384 1385 /* 1386 * Attach the dquot(s) to the inodes and modify them incore. 1387 * These ids of the inode couldn't have changed since the new 1388 * inode has been locked ever since it was created. 1389 */ 1390 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1391 1392 error = xfs_iunlink(tp, ip); 1393 if (error) 1394 goto out_trans_cancel; 1395 1396 error = xfs_trans_commit(tp); 1397 if (error) 1398 goto out_release_inode; 1399 1400 xfs_qm_dqrele(udqp); 1401 xfs_qm_dqrele(gdqp); 1402 xfs_qm_dqrele(pdqp); 1403 1404 *ipp = ip; 1405 return 0; 1406 1407 out_trans_cancel: 1408 xfs_trans_cancel(tp); 1409 out_release_inode: 1410 /* 1411 * Wait until after the current transaction is aborted to finish the 1412 * setup of the inode and release the inode. This prevents recursive 1413 * transactions and deadlocks from xfs_inactive. 1414 */ 1415 if (ip) { 1416 xfs_finish_inode_setup(ip); 1417 IRELE(ip); 1418 } 1419 1420 xfs_qm_dqrele(udqp); 1421 xfs_qm_dqrele(gdqp); 1422 xfs_qm_dqrele(pdqp); 1423 1424 return error; 1425 } 1426 1427 int 1428 xfs_link( 1429 xfs_inode_t *tdp, 1430 xfs_inode_t *sip, 1431 struct xfs_name *target_name) 1432 { 1433 xfs_mount_t *mp = tdp->i_mount; 1434 xfs_trans_t *tp; 1435 int error; 1436 struct xfs_defer_ops dfops; 1437 xfs_fsblock_t first_block; 1438 int resblks; 1439 1440 trace_xfs_link(tdp, target_name); 1441 1442 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1443 1444 if (XFS_FORCED_SHUTDOWN(mp)) 1445 return -EIO; 1446 1447 error = xfs_qm_dqattach(sip, 0); 1448 if (error) 1449 goto std_return; 1450 1451 error = xfs_qm_dqattach(tdp, 0); 1452 if (error) 1453 goto std_return; 1454 1455 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1456 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1457 if (error == -ENOSPC) { 1458 resblks = 0; 1459 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1460 } 1461 if (error) 1462 goto std_return; 1463 1464 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1465 1466 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1467 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1468 1469 /* 1470 * If we are using project inheritance, we only allow hard link 1471 * creation in our tree when the project IDs are the same; else 1472 * the tree quota mechanism could be circumvented. 1473 */ 1474 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1475 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1476 error = -EXDEV; 1477 goto error_return; 1478 } 1479 1480 if (!resblks) { 1481 error = xfs_dir_canenter(tp, tdp, target_name); 1482 if (error) 1483 goto error_return; 1484 } 1485 1486 xfs_defer_init(&dfops, &first_block); 1487 1488 /* 1489 * Handle initial link state of O_TMPFILE inode 1490 */ 1491 if (VFS_I(sip)->i_nlink == 0) { 1492 error = xfs_iunlink_remove(tp, sip); 1493 if (error) 1494 goto error_return; 1495 } 1496 1497 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1498 &first_block, &dfops, resblks); 1499 if (error) 1500 goto error_return; 1501 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1502 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1503 1504 error = xfs_bumplink(tp, sip); 1505 if (error) 1506 goto error_return; 1507 1508 /* 1509 * If this is a synchronous mount, make sure that the 1510 * link transaction goes to disk before returning to 1511 * the user. 1512 */ 1513 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1514 xfs_trans_set_sync(tp); 1515 1516 error = xfs_defer_finish(&tp, &dfops, NULL); 1517 if (error) { 1518 xfs_defer_cancel(&dfops); 1519 goto error_return; 1520 } 1521 1522 return xfs_trans_commit(tp); 1523 1524 error_return: 1525 xfs_trans_cancel(tp); 1526 std_return: 1527 return error; 1528 } 1529 1530 /* 1531 * Free up the underlying blocks past new_size. The new size must be smaller 1532 * than the current size. This routine can be used both for the attribute and 1533 * data fork, and does not modify the inode size, which is left to the caller. 1534 * 1535 * The transaction passed to this routine must have made a permanent log 1536 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1537 * given transaction and start new ones, so make sure everything involved in 1538 * the transaction is tidy before calling here. Some transaction will be 1539 * returned to the caller to be committed. The incoming transaction must 1540 * already include the inode, and both inode locks must be held exclusively. 1541 * The inode must also be "held" within the transaction. On return the inode 1542 * will be "held" within the returned transaction. This routine does NOT 1543 * require any disk space to be reserved for it within the transaction. 1544 * 1545 * If we get an error, we must return with the inode locked and linked into the 1546 * current transaction. This keeps things simple for the higher level code, 1547 * because it always knows that the inode is locked and held in the transaction 1548 * that returns to it whether errors occur or not. We don't mark the inode 1549 * dirty on error so that transactions can be easily aborted if possible. 1550 */ 1551 int 1552 xfs_itruncate_extents( 1553 struct xfs_trans **tpp, 1554 struct xfs_inode *ip, 1555 int whichfork, 1556 xfs_fsize_t new_size) 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 /* 1580 * Since it is possible for space to become allocated beyond 1581 * the end of the file (in a crash where the space is allocated 1582 * but the inode size is not yet updated), simply remove any 1583 * blocks which show up between the new EOF and the maximum 1584 * possible file size. If the first block to be removed is 1585 * beyond the maximum file size (ie it is the same as last_block), 1586 * then there is nothing to do. 1587 */ 1588 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1589 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1590 if (first_unmap_block == last_block) 1591 return 0; 1592 1593 ASSERT(first_unmap_block < last_block); 1594 unmap_len = last_block - first_unmap_block + 1; 1595 while (!done) { 1596 xfs_defer_init(&dfops, &first_block); 1597 error = xfs_bunmapi(tp, ip, 1598 first_unmap_block, unmap_len, 1599 xfs_bmapi_aflag(whichfork), 1600 XFS_ITRUNC_MAX_EXTENTS, 1601 &first_block, &dfops, 1602 &done); 1603 if (error) 1604 goto out_bmap_cancel; 1605 1606 /* 1607 * Duplicate the transaction that has the permanent 1608 * reservation and commit the old transaction. 1609 */ 1610 error = xfs_defer_finish(&tp, &dfops, ip); 1611 if (error) 1612 goto out_bmap_cancel; 1613 1614 error = xfs_trans_roll(&tp, ip); 1615 if (error) 1616 goto out; 1617 } 1618 1619 /* Remove all pending CoW reservations. */ 1620 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, 1621 last_block, true); 1622 if (error) 1623 goto out; 1624 1625 /* 1626 * Clear the reflink flag if we truncated everything. 1627 */ 1628 if (ip->i_d.di_nblocks == 0 && xfs_is_reflink_inode(ip)) { 1629 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 1630 xfs_inode_clear_cowblocks_tag(ip); 1631 } 1632 1633 /* 1634 * Always re-log the inode so that our permanent transaction can keep 1635 * on rolling it forward in the log. 1636 */ 1637 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1638 1639 trace_xfs_itruncate_extents_end(ip, new_size); 1640 1641 out: 1642 *tpp = tp; 1643 return error; 1644 out_bmap_cancel: 1645 /* 1646 * If the bunmapi call encounters an error, return to the caller where 1647 * the transaction can be properly aborted. We just need to make sure 1648 * we're not holding any resources that we were not when we came in. 1649 */ 1650 xfs_defer_cancel(&dfops); 1651 goto out; 1652 } 1653 1654 int 1655 xfs_release( 1656 xfs_inode_t *ip) 1657 { 1658 xfs_mount_t *mp = ip->i_mount; 1659 int error; 1660 1661 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1662 return 0; 1663 1664 /* If this is a read-only mount, don't do this (would generate I/O) */ 1665 if (mp->m_flags & XFS_MOUNT_RDONLY) 1666 return 0; 1667 1668 if (!XFS_FORCED_SHUTDOWN(mp)) { 1669 int truncated; 1670 1671 /* 1672 * If we previously truncated this file and removed old data 1673 * in the process, we want to initiate "early" writeout on 1674 * the last close. This is an attempt to combat the notorious 1675 * NULL files problem which is particularly noticeable from a 1676 * truncate down, buffered (re-)write (delalloc), followed by 1677 * a crash. What we are effectively doing here is 1678 * significantly reducing the time window where we'd otherwise 1679 * be exposed to that problem. 1680 */ 1681 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1682 if (truncated) { 1683 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1684 if (ip->i_delayed_blks > 0) { 1685 error = filemap_flush(VFS_I(ip)->i_mapping); 1686 if (error) 1687 return error; 1688 } 1689 } 1690 } 1691 1692 if (VFS_I(ip)->i_nlink == 0) 1693 return 0; 1694 1695 if (xfs_can_free_eofblocks(ip, false)) { 1696 1697 /* 1698 * Check if the inode is being opened, written and closed 1699 * frequently and we have delayed allocation blocks outstanding 1700 * (e.g. streaming writes from the NFS server), truncating the 1701 * blocks past EOF will cause fragmentation to occur. 1702 * 1703 * In this case don't do the truncation, but we have to be 1704 * careful how we detect this case. Blocks beyond EOF show up as 1705 * i_delayed_blks even when the inode is clean, so we need to 1706 * truncate them away first before checking for a dirty release. 1707 * Hence on the first dirty close we will still remove the 1708 * speculative allocation, but after that we will leave it in 1709 * place. 1710 */ 1711 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1712 return 0; 1713 /* 1714 * If we can't get the iolock just skip truncating the blocks 1715 * past EOF because we could deadlock with the mmap_sem 1716 * otherwise. We'll get another chance to drop them once the 1717 * last reference to the inode is dropped, so we'll never leak 1718 * blocks permanently. 1719 */ 1720 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { 1721 error = xfs_free_eofblocks(ip); 1722 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 1723 if (error) 1724 return error; 1725 } 1726 1727 /* delalloc blocks after truncation means it really is dirty */ 1728 if (ip->i_delayed_blks) 1729 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1730 } 1731 return 0; 1732 } 1733 1734 /* 1735 * xfs_inactive_truncate 1736 * 1737 * Called to perform a truncate when an inode becomes unlinked. 1738 */ 1739 STATIC int 1740 xfs_inactive_truncate( 1741 struct xfs_inode *ip) 1742 { 1743 struct xfs_mount *mp = ip->i_mount; 1744 struct xfs_trans *tp; 1745 int error; 1746 1747 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1748 if (error) { 1749 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1750 return error; 1751 } 1752 1753 xfs_ilock(ip, XFS_ILOCK_EXCL); 1754 xfs_trans_ijoin(tp, ip, 0); 1755 1756 /* 1757 * Log the inode size first to prevent stale data exposure in the event 1758 * of a system crash before the truncate completes. See the related 1759 * comment in xfs_vn_setattr_size() for details. 1760 */ 1761 ip->i_d.di_size = 0; 1762 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1763 1764 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1765 if (error) 1766 goto error_trans_cancel; 1767 1768 ASSERT(ip->i_d.di_nextents == 0); 1769 1770 error = xfs_trans_commit(tp); 1771 if (error) 1772 goto error_unlock; 1773 1774 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1775 return 0; 1776 1777 error_trans_cancel: 1778 xfs_trans_cancel(tp); 1779 error_unlock: 1780 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1781 return error; 1782 } 1783 1784 /* 1785 * xfs_inactive_ifree() 1786 * 1787 * Perform the inode free when an inode is unlinked. 1788 */ 1789 STATIC int 1790 xfs_inactive_ifree( 1791 struct xfs_inode *ip) 1792 { 1793 struct xfs_defer_ops dfops; 1794 xfs_fsblock_t first_block; 1795 struct xfs_mount *mp = ip->i_mount; 1796 struct xfs_trans *tp; 1797 int error; 1798 1799 /* 1800 * We try to use a per-AG reservation for any block needed by the finobt 1801 * tree, but as the finobt feature predates the per-AG reservation 1802 * support a degraded file system might not have enough space for the 1803 * reservation at mount time. In that case try to dip into the reserved 1804 * pool and pray. 1805 * 1806 * Send a warning if the reservation does happen to fail, as the inode 1807 * now remains allocated and sits on the unlinked list until the fs is 1808 * repaired. 1809 */ 1810 if (unlikely(mp->m_inotbt_nores)) { 1811 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1812 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 1813 &tp); 1814 } else { 1815 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); 1816 } 1817 if (error) { 1818 if (error == -ENOSPC) { 1819 xfs_warn_ratelimited(mp, 1820 "Failed to remove inode(s) from unlinked list. " 1821 "Please free space, unmount and run xfs_repair."); 1822 } else { 1823 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1824 } 1825 return error; 1826 } 1827 1828 xfs_ilock(ip, XFS_ILOCK_EXCL); 1829 xfs_trans_ijoin(tp, ip, 0); 1830 1831 xfs_defer_init(&dfops, &first_block); 1832 error = xfs_ifree(tp, ip, &dfops); 1833 if (error) { 1834 /* 1835 * If we fail to free the inode, shut down. The cancel 1836 * might do that, we need to make sure. Otherwise the 1837 * inode might be lost for a long time or forever. 1838 */ 1839 if (!XFS_FORCED_SHUTDOWN(mp)) { 1840 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1841 __func__, error); 1842 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1843 } 1844 xfs_trans_cancel(tp); 1845 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1846 return error; 1847 } 1848 1849 /* 1850 * Credit the quota account(s). The inode is gone. 1851 */ 1852 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1853 1854 /* 1855 * Just ignore errors at this point. There is nothing we can do except 1856 * to try to keep going. Make sure it's not a silent error. 1857 */ 1858 error = xfs_defer_finish(&tp, &dfops, NULL); 1859 if (error) { 1860 xfs_notice(mp, "%s: xfs_defer_finish returned error %d", 1861 __func__, error); 1862 xfs_defer_cancel(&dfops); 1863 } 1864 error = xfs_trans_commit(tp); 1865 if (error) 1866 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1867 __func__, error); 1868 1869 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1870 return 0; 1871 } 1872 1873 /* 1874 * xfs_inactive 1875 * 1876 * This is called when the vnode reference count for the vnode 1877 * goes to zero. If the file has been unlinked, then it must 1878 * now be truncated. Also, we clear all of the read-ahead state 1879 * kept for the inode here since the file is now closed. 1880 */ 1881 void 1882 xfs_inactive( 1883 xfs_inode_t *ip) 1884 { 1885 struct xfs_mount *mp; 1886 int error; 1887 int truncate = 0; 1888 1889 /* 1890 * If the inode is already free, then there can be nothing 1891 * to clean up here. 1892 */ 1893 if (VFS_I(ip)->i_mode == 0) { 1894 ASSERT(ip->i_df.if_real_bytes == 0); 1895 ASSERT(ip->i_df.if_broot_bytes == 0); 1896 return; 1897 } 1898 1899 mp = ip->i_mount; 1900 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); 1901 1902 /* If this is a read-only mount, don't do this (would generate I/O) */ 1903 if (mp->m_flags & XFS_MOUNT_RDONLY) 1904 return; 1905 1906 if (VFS_I(ip)->i_nlink != 0) { 1907 /* 1908 * force is true because we are evicting an inode from the 1909 * cache. Post-eof blocks must be freed, lest we end up with 1910 * broken free space accounting. 1911 * 1912 * Note: don't bother with iolock here since lockdep complains 1913 * about acquiring it in reclaim context. We have the only 1914 * reference to the inode at this point anyways. 1915 */ 1916 if (xfs_can_free_eofblocks(ip, true)) 1917 xfs_free_eofblocks(ip); 1918 1919 return; 1920 } 1921 1922 if (S_ISREG(VFS_I(ip)->i_mode) && 1923 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1924 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1925 truncate = 1; 1926 1927 error = xfs_qm_dqattach(ip, 0); 1928 if (error) 1929 return; 1930 1931 if (S_ISLNK(VFS_I(ip)->i_mode)) 1932 error = xfs_inactive_symlink(ip); 1933 else if (truncate) 1934 error = xfs_inactive_truncate(ip); 1935 if (error) 1936 return; 1937 1938 /* 1939 * If there are attributes associated with the file then blow them away 1940 * now. The code calls a routine that recursively deconstructs the 1941 * attribute fork. If also blows away the in-core attribute fork. 1942 */ 1943 if (XFS_IFORK_Q(ip)) { 1944 error = xfs_attr_inactive(ip); 1945 if (error) 1946 return; 1947 } 1948 1949 ASSERT(!ip->i_afp); 1950 ASSERT(ip->i_d.di_anextents == 0); 1951 ASSERT(ip->i_d.di_forkoff == 0); 1952 1953 /* 1954 * Free the inode. 1955 */ 1956 error = xfs_inactive_ifree(ip); 1957 if (error) 1958 return; 1959 1960 /* 1961 * Release the dquots held by inode, if any. 1962 */ 1963 xfs_qm_dqdetach(ip); 1964 } 1965 1966 /* 1967 * This is called when the inode's link count goes to 0 or we are creating a 1968 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1969 * set to true as the link count is dropped to zero by the VFS after we've 1970 * created the file successfully, so we have to add it to the unlinked list 1971 * while the link count is non-zero. 1972 * 1973 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1974 * list when the inode is freed. 1975 */ 1976 STATIC int 1977 xfs_iunlink( 1978 struct xfs_trans *tp, 1979 struct xfs_inode *ip) 1980 { 1981 xfs_mount_t *mp = tp->t_mountp; 1982 xfs_agi_t *agi; 1983 xfs_dinode_t *dip; 1984 xfs_buf_t *agibp; 1985 xfs_buf_t *ibp; 1986 xfs_agino_t agino; 1987 short bucket_index; 1988 int offset; 1989 int error; 1990 1991 ASSERT(VFS_I(ip)->i_mode != 0); 1992 1993 /* 1994 * Get the agi buffer first. It ensures lock ordering 1995 * on the list. 1996 */ 1997 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1998 if (error) 1999 return error; 2000 agi = XFS_BUF_TO_AGI(agibp); 2001 2002 /* 2003 * Get the index into the agi hash table for the 2004 * list this inode will go on. 2005 */ 2006 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2007 ASSERT(agino != 0); 2008 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2009 ASSERT(agi->agi_unlinked[bucket_index]); 2010 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 2011 2012 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 2013 /* 2014 * There is already another inode in the bucket we need 2015 * to add ourselves to. Add us at the front of the list. 2016 * Here we put the head pointer into our next pointer, 2017 * and then we fall through to point the head at us. 2018 */ 2019 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2020 0, 0); 2021 if (error) 2022 return error; 2023 2024 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 2025 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 2026 offset = ip->i_imap.im_boffset + 2027 offsetof(xfs_dinode_t, di_next_unlinked); 2028 2029 /* need to recalc the inode CRC if appropriate */ 2030 xfs_dinode_calc_crc(mp, dip); 2031 2032 xfs_trans_inode_buf(tp, ibp); 2033 xfs_trans_log_buf(tp, ibp, offset, 2034 (offset + sizeof(xfs_agino_t) - 1)); 2035 xfs_inobp_check(mp, ibp); 2036 } 2037 2038 /* 2039 * Point the bucket head pointer at the inode being inserted. 2040 */ 2041 ASSERT(agino != 0); 2042 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 2043 offset = offsetof(xfs_agi_t, agi_unlinked) + 2044 (sizeof(xfs_agino_t) * bucket_index); 2045 xfs_trans_log_buf(tp, agibp, offset, 2046 (offset + sizeof(xfs_agino_t) - 1)); 2047 return 0; 2048 } 2049 2050 /* 2051 * Pull the on-disk inode from the AGI unlinked list. 2052 */ 2053 STATIC int 2054 xfs_iunlink_remove( 2055 xfs_trans_t *tp, 2056 xfs_inode_t *ip) 2057 { 2058 xfs_ino_t next_ino; 2059 xfs_mount_t *mp; 2060 xfs_agi_t *agi; 2061 xfs_dinode_t *dip; 2062 xfs_buf_t *agibp; 2063 xfs_buf_t *ibp; 2064 xfs_agnumber_t agno; 2065 xfs_agino_t agino; 2066 xfs_agino_t next_agino; 2067 xfs_buf_t *last_ibp; 2068 xfs_dinode_t *last_dip = NULL; 2069 short bucket_index; 2070 int offset, last_offset = 0; 2071 int error; 2072 2073 mp = tp->t_mountp; 2074 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2075 2076 /* 2077 * Get the agi buffer first. It ensures lock ordering 2078 * on the list. 2079 */ 2080 error = xfs_read_agi(mp, tp, agno, &agibp); 2081 if (error) 2082 return error; 2083 2084 agi = XFS_BUF_TO_AGI(agibp); 2085 2086 /* 2087 * Get the index into the agi hash table for the 2088 * list this inode will go on. 2089 */ 2090 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2091 ASSERT(agino != 0); 2092 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2093 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2094 ASSERT(agi->agi_unlinked[bucket_index]); 2095 2096 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2097 /* 2098 * We're at the head of the list. Get the inode's on-disk 2099 * buffer to see if there is anyone after us on the list. 2100 * Only modify our next pointer if it is not already NULLAGINO. 2101 * This saves us the overhead of dealing with the buffer when 2102 * there is no need to change it. 2103 */ 2104 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2105 0, 0); 2106 if (error) { 2107 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2108 __func__, error); 2109 return error; 2110 } 2111 next_agino = be32_to_cpu(dip->di_next_unlinked); 2112 ASSERT(next_agino != 0); 2113 if (next_agino != NULLAGINO) { 2114 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2115 offset = ip->i_imap.im_boffset + 2116 offsetof(xfs_dinode_t, di_next_unlinked); 2117 2118 /* need to recalc the inode CRC if appropriate */ 2119 xfs_dinode_calc_crc(mp, dip); 2120 2121 xfs_trans_inode_buf(tp, ibp); 2122 xfs_trans_log_buf(tp, ibp, offset, 2123 (offset + sizeof(xfs_agino_t) - 1)); 2124 xfs_inobp_check(mp, ibp); 2125 } else { 2126 xfs_trans_brelse(tp, ibp); 2127 } 2128 /* 2129 * Point the bucket head pointer at the next inode. 2130 */ 2131 ASSERT(next_agino != 0); 2132 ASSERT(next_agino != agino); 2133 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2134 offset = offsetof(xfs_agi_t, agi_unlinked) + 2135 (sizeof(xfs_agino_t) * bucket_index); 2136 xfs_trans_log_buf(tp, agibp, offset, 2137 (offset + sizeof(xfs_agino_t) - 1)); 2138 } else { 2139 /* 2140 * We need to search the list for the inode being freed. 2141 */ 2142 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2143 last_ibp = NULL; 2144 while (next_agino != agino) { 2145 struct xfs_imap imap; 2146 2147 if (last_ibp) 2148 xfs_trans_brelse(tp, last_ibp); 2149 2150 imap.im_blkno = 0; 2151 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2152 2153 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2154 if (error) { 2155 xfs_warn(mp, 2156 "%s: xfs_imap returned error %d.", 2157 __func__, error); 2158 return error; 2159 } 2160 2161 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2162 &last_ibp, 0, 0); 2163 if (error) { 2164 xfs_warn(mp, 2165 "%s: xfs_imap_to_bp returned error %d.", 2166 __func__, error); 2167 return error; 2168 } 2169 2170 last_offset = imap.im_boffset; 2171 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2172 ASSERT(next_agino != NULLAGINO); 2173 ASSERT(next_agino != 0); 2174 } 2175 2176 /* 2177 * Now last_ibp points to the buffer previous to us on the 2178 * unlinked list. Pull us from the list. 2179 */ 2180 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2181 0, 0); 2182 if (error) { 2183 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2184 __func__, error); 2185 return error; 2186 } 2187 next_agino = be32_to_cpu(dip->di_next_unlinked); 2188 ASSERT(next_agino != 0); 2189 ASSERT(next_agino != agino); 2190 if (next_agino != NULLAGINO) { 2191 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2192 offset = ip->i_imap.im_boffset + 2193 offsetof(xfs_dinode_t, di_next_unlinked); 2194 2195 /* need to recalc the inode CRC if appropriate */ 2196 xfs_dinode_calc_crc(mp, dip); 2197 2198 xfs_trans_inode_buf(tp, ibp); 2199 xfs_trans_log_buf(tp, ibp, offset, 2200 (offset + sizeof(xfs_agino_t) - 1)); 2201 xfs_inobp_check(mp, ibp); 2202 } else { 2203 xfs_trans_brelse(tp, ibp); 2204 } 2205 /* 2206 * Point the previous inode on the list to the next inode. 2207 */ 2208 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2209 ASSERT(next_agino != 0); 2210 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2211 2212 /* need to recalc the inode CRC if appropriate */ 2213 xfs_dinode_calc_crc(mp, last_dip); 2214 2215 xfs_trans_inode_buf(tp, last_ibp); 2216 xfs_trans_log_buf(tp, last_ibp, offset, 2217 (offset + sizeof(xfs_agino_t) - 1)); 2218 xfs_inobp_check(mp, last_ibp); 2219 } 2220 return 0; 2221 } 2222 2223 /* 2224 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2225 * inodes that are in memory - they all must be marked stale and attached to 2226 * the cluster buffer. 2227 */ 2228 STATIC int 2229 xfs_ifree_cluster( 2230 xfs_inode_t *free_ip, 2231 xfs_trans_t *tp, 2232 struct xfs_icluster *xic) 2233 { 2234 xfs_mount_t *mp = free_ip->i_mount; 2235 int blks_per_cluster; 2236 int inodes_per_cluster; 2237 int nbufs; 2238 int i, j; 2239 int ioffset; 2240 xfs_daddr_t blkno; 2241 xfs_buf_t *bp; 2242 xfs_inode_t *ip; 2243 xfs_inode_log_item_t *iip; 2244 xfs_log_item_t *lip; 2245 struct xfs_perag *pag; 2246 xfs_ino_t inum; 2247 2248 inum = xic->first_ino; 2249 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2250 blks_per_cluster = xfs_icluster_size_fsb(mp); 2251 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2252 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2253 2254 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2255 /* 2256 * The allocation bitmap tells us which inodes of the chunk were 2257 * physically allocated. Skip the cluster if an inode falls into 2258 * a sparse region. 2259 */ 2260 ioffset = inum - xic->first_ino; 2261 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2262 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2263 continue; 2264 } 2265 2266 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2267 XFS_INO_TO_AGBNO(mp, inum)); 2268 2269 /* 2270 * We obtain and lock the backing buffer first in the process 2271 * here, as we have to ensure that any dirty inode that we 2272 * can't get the flush lock on is attached to the buffer. 2273 * If we scan the in-memory inodes first, then buffer IO can 2274 * complete before we get a lock on it, and hence we may fail 2275 * to mark all the active inodes on the buffer stale. 2276 */ 2277 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2278 mp->m_bsize * blks_per_cluster, 2279 XBF_UNMAPPED); 2280 2281 if (!bp) 2282 return -ENOMEM; 2283 2284 /* 2285 * This buffer may not have been correctly initialised as we 2286 * didn't read it from disk. That's not important because we are 2287 * only using to mark the buffer as stale in the log, and to 2288 * attach stale cached inodes on it. That means it will never be 2289 * dispatched for IO. If it is, we want to know about it, and we 2290 * want it to fail. We can acheive this by adding a write 2291 * verifier to the buffer. 2292 */ 2293 bp->b_ops = &xfs_inode_buf_ops; 2294 2295 /* 2296 * Walk the inodes already attached to the buffer and mark them 2297 * stale. These will all have the flush locks held, so an 2298 * in-memory inode walk can't lock them. By marking them all 2299 * stale first, we will not attempt to lock them in the loop 2300 * below as the XFS_ISTALE flag will be set. 2301 */ 2302 lip = bp->b_fspriv; 2303 while (lip) { 2304 if (lip->li_type == XFS_LI_INODE) { 2305 iip = (xfs_inode_log_item_t *)lip; 2306 ASSERT(iip->ili_logged == 1); 2307 lip->li_cb = xfs_istale_done; 2308 xfs_trans_ail_copy_lsn(mp->m_ail, 2309 &iip->ili_flush_lsn, 2310 &iip->ili_item.li_lsn); 2311 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2312 } 2313 lip = lip->li_bio_list; 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 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2364 rcu_read_unlock(); 2365 delay(1); 2366 goto retry; 2367 } 2368 rcu_read_unlock(); 2369 2370 xfs_iflock(ip); 2371 xfs_iflags_set(ip, XFS_ISTALE); 2372 2373 /* 2374 * we don't need to attach clean inodes or those only 2375 * with unlogged changes (which we throw away, anyway). 2376 */ 2377 iip = ip->i_itemp; 2378 if (!iip || xfs_inode_clean(ip)) { 2379 ASSERT(ip != free_ip); 2380 xfs_ifunlock(ip); 2381 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2382 continue; 2383 } 2384 2385 iip->ili_last_fields = iip->ili_fields; 2386 iip->ili_fields = 0; 2387 iip->ili_fsync_fields = 0; 2388 iip->ili_logged = 1; 2389 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2390 &iip->ili_item.li_lsn); 2391 2392 xfs_buf_attach_iodone(bp, xfs_istale_done, 2393 &iip->ili_item); 2394 2395 if (ip != free_ip) 2396 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2397 } 2398 2399 xfs_trans_stale_inode_buf(tp, bp); 2400 xfs_trans_binval(tp, bp); 2401 } 2402 2403 xfs_perag_put(pag); 2404 return 0; 2405 } 2406 2407 /* 2408 * This is called to return an inode to the inode free list. 2409 * The inode should already be truncated to 0 length and have 2410 * no pages associated with it. This routine also assumes that 2411 * the inode is already a part of the transaction. 2412 * 2413 * The on-disk copy of the inode will have been added to the list 2414 * of unlinked inodes in the AGI. We need to remove the inode from 2415 * that list atomically with respect to freeing it here. 2416 */ 2417 int 2418 xfs_ifree( 2419 xfs_trans_t *tp, 2420 xfs_inode_t *ip, 2421 struct xfs_defer_ops *dfops) 2422 { 2423 int error; 2424 struct xfs_icluster xic = { 0 }; 2425 2426 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2427 ASSERT(VFS_I(ip)->i_nlink == 0); 2428 ASSERT(ip->i_d.di_nextents == 0); 2429 ASSERT(ip->i_d.di_anextents == 0); 2430 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2431 ASSERT(ip->i_d.di_nblocks == 0); 2432 2433 /* 2434 * Pull the on-disk inode from the AGI unlinked list. 2435 */ 2436 error = xfs_iunlink_remove(tp, ip); 2437 if (error) 2438 return error; 2439 2440 error = xfs_difree(tp, ip->i_ino, dfops, &xic); 2441 if (error) 2442 return error; 2443 2444 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2445 ip->i_d.di_flags = 0; 2446 ip->i_d.di_dmevmask = 0; 2447 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2448 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2449 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2450 /* 2451 * Bump the generation count so no one will be confused 2452 * by reincarnations of this inode. 2453 */ 2454 VFS_I(ip)->i_generation++; 2455 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2456 2457 if (xic.deleted) 2458 error = xfs_ifree_cluster(ip, tp, &xic); 2459 2460 return error; 2461 } 2462 2463 /* 2464 * This is called to unpin an inode. The caller must have the inode locked 2465 * in at least shared mode so that the buffer cannot be subsequently pinned 2466 * once someone is waiting for it to be unpinned. 2467 */ 2468 static void 2469 xfs_iunpin( 2470 struct xfs_inode *ip) 2471 { 2472 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2473 2474 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2475 2476 /* Give the log a push to start the unpinning I/O */ 2477 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2478 2479 } 2480 2481 static void 2482 __xfs_iunpin_wait( 2483 struct xfs_inode *ip) 2484 { 2485 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2486 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2487 2488 xfs_iunpin(ip); 2489 2490 do { 2491 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 2492 if (xfs_ipincount(ip)) 2493 io_schedule(); 2494 } while (xfs_ipincount(ip)); 2495 finish_wait(wq, &wait.wq_entry); 2496 } 2497 2498 void 2499 xfs_iunpin_wait( 2500 struct xfs_inode *ip) 2501 { 2502 if (xfs_ipincount(ip)) 2503 __xfs_iunpin_wait(ip); 2504 } 2505 2506 /* 2507 * Removing an inode from the namespace involves removing the directory entry 2508 * and dropping the link count on the inode. Removing the directory entry can 2509 * result in locking an AGF (directory blocks were freed) and removing a link 2510 * count can result in placing the inode on an unlinked list which results in 2511 * locking an AGI. 2512 * 2513 * The big problem here is that we have an ordering constraint on AGF and AGI 2514 * locking - inode allocation locks the AGI, then can allocate a new extent for 2515 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2516 * removes the inode from the unlinked list, requiring that we lock the AGI 2517 * first, and then freeing the inode can result in an inode chunk being freed 2518 * and hence freeing disk space requiring that we lock an AGF. 2519 * 2520 * Hence the ordering that is imposed by other parts of the code is AGI before 2521 * AGF. This means we cannot remove the directory entry before we drop the inode 2522 * reference count and put it on the unlinked list as this results in a lock 2523 * order of AGF then AGI, and this can deadlock against inode allocation and 2524 * freeing. Therefore we must drop the link counts before we remove the 2525 * directory entry. 2526 * 2527 * This is still safe from a transactional point of view - it is not until we 2528 * get to xfs_defer_finish() that we have the possibility of multiple 2529 * transactions in this operation. Hence as long as we remove the directory 2530 * entry and drop the link count in the first transaction of the remove 2531 * operation, there are no transactional constraints on the ordering here. 2532 */ 2533 int 2534 xfs_remove( 2535 xfs_inode_t *dp, 2536 struct xfs_name *name, 2537 xfs_inode_t *ip) 2538 { 2539 xfs_mount_t *mp = dp->i_mount; 2540 xfs_trans_t *tp = NULL; 2541 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2542 int error = 0; 2543 struct xfs_defer_ops dfops; 2544 xfs_fsblock_t first_block; 2545 uint resblks; 2546 2547 trace_xfs_remove(dp, name); 2548 2549 if (XFS_FORCED_SHUTDOWN(mp)) 2550 return -EIO; 2551 2552 error = xfs_qm_dqattach(dp, 0); 2553 if (error) 2554 goto std_return; 2555 2556 error = xfs_qm_dqattach(ip, 0); 2557 if (error) 2558 goto std_return; 2559 2560 /* 2561 * We try to get the real space reservation first, 2562 * allowing for directory btree deletion(s) implying 2563 * possible bmap insert(s). If we can't get the space 2564 * reservation then we use 0 instead, and avoid the bmap 2565 * btree insert(s) in the directory code by, if the bmap 2566 * insert tries to happen, instead trimming the LAST 2567 * block from the directory. 2568 */ 2569 resblks = XFS_REMOVE_SPACE_RES(mp); 2570 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2571 if (error == -ENOSPC) { 2572 resblks = 0; 2573 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2574 &tp); 2575 } 2576 if (error) { 2577 ASSERT(error != -ENOSPC); 2578 goto std_return; 2579 } 2580 2581 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2582 2583 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2584 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2585 2586 /* 2587 * If we're removing a directory perform some additional validation. 2588 */ 2589 if (is_dir) { 2590 ASSERT(VFS_I(ip)->i_nlink >= 2); 2591 if (VFS_I(ip)->i_nlink != 2) { 2592 error = -ENOTEMPTY; 2593 goto out_trans_cancel; 2594 } 2595 if (!xfs_dir_isempty(ip)) { 2596 error = -ENOTEMPTY; 2597 goto out_trans_cancel; 2598 } 2599 2600 /* Drop the link from ip's "..". */ 2601 error = xfs_droplink(tp, dp); 2602 if (error) 2603 goto out_trans_cancel; 2604 2605 /* Drop the "." link from ip to self. */ 2606 error = xfs_droplink(tp, ip); 2607 if (error) 2608 goto out_trans_cancel; 2609 } else { 2610 /* 2611 * When removing a non-directory we need to log the parent 2612 * inode here. For a directory this is done implicitly 2613 * by the xfs_droplink call for the ".." entry. 2614 */ 2615 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2616 } 2617 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2618 2619 /* Drop the link from dp to ip. */ 2620 error = xfs_droplink(tp, ip); 2621 if (error) 2622 goto out_trans_cancel; 2623 2624 xfs_defer_init(&dfops, &first_block); 2625 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2626 &first_block, &dfops, resblks); 2627 if (error) { 2628 ASSERT(error != -ENOENT); 2629 goto out_bmap_cancel; 2630 } 2631 2632 /* 2633 * If this is a synchronous mount, make sure that the 2634 * remove transaction goes to disk before returning to 2635 * the user. 2636 */ 2637 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2638 xfs_trans_set_sync(tp); 2639 2640 error = xfs_defer_finish(&tp, &dfops, NULL); 2641 if (error) 2642 goto out_bmap_cancel; 2643 2644 error = xfs_trans_commit(tp); 2645 if (error) 2646 goto std_return; 2647 2648 if (is_dir && xfs_inode_is_filestream(ip)) 2649 xfs_filestream_deassociate(ip); 2650 2651 return 0; 2652 2653 out_bmap_cancel: 2654 xfs_defer_cancel(&dfops); 2655 out_trans_cancel: 2656 xfs_trans_cancel(tp); 2657 std_return: 2658 return error; 2659 } 2660 2661 /* 2662 * Enter all inodes for a rename transaction into a sorted array. 2663 */ 2664 #define __XFS_SORT_INODES 5 2665 STATIC void 2666 xfs_sort_for_rename( 2667 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2668 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2669 struct xfs_inode *ip1, /* in: inode of old entry */ 2670 struct xfs_inode *ip2, /* in: inode of new entry */ 2671 struct xfs_inode *wip, /* in: whiteout inode */ 2672 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2673 int *num_inodes) /* in/out: inodes in array */ 2674 { 2675 int i, j; 2676 2677 ASSERT(*num_inodes == __XFS_SORT_INODES); 2678 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2679 2680 /* 2681 * i_tab contains a list of pointers to inodes. We initialize 2682 * the table here & we'll sort it. We will then use it to 2683 * order the acquisition of the inode locks. 2684 * 2685 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2686 */ 2687 i = 0; 2688 i_tab[i++] = dp1; 2689 i_tab[i++] = dp2; 2690 i_tab[i++] = ip1; 2691 if (ip2) 2692 i_tab[i++] = ip2; 2693 if (wip) 2694 i_tab[i++] = wip; 2695 *num_inodes = i; 2696 2697 /* 2698 * Sort the elements via bubble sort. (Remember, there are at 2699 * most 5 elements to sort, so this is adequate.) 2700 */ 2701 for (i = 0; i < *num_inodes; i++) { 2702 for (j = 1; j < *num_inodes; j++) { 2703 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2704 struct xfs_inode *temp = i_tab[j]; 2705 i_tab[j] = i_tab[j-1]; 2706 i_tab[j-1] = temp; 2707 } 2708 } 2709 } 2710 } 2711 2712 static int 2713 xfs_finish_rename( 2714 struct xfs_trans *tp, 2715 struct xfs_defer_ops *dfops) 2716 { 2717 int error; 2718 2719 /* 2720 * If this is a synchronous mount, make sure that the rename transaction 2721 * goes to disk before returning to the user. 2722 */ 2723 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2724 xfs_trans_set_sync(tp); 2725 2726 error = xfs_defer_finish(&tp, dfops, NULL); 2727 if (error) { 2728 xfs_defer_cancel(dfops); 2729 xfs_trans_cancel(tp); 2730 return error; 2731 } 2732 2733 return xfs_trans_commit(tp); 2734 } 2735 2736 /* 2737 * xfs_cross_rename() 2738 * 2739 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2740 */ 2741 STATIC int 2742 xfs_cross_rename( 2743 struct xfs_trans *tp, 2744 struct xfs_inode *dp1, 2745 struct xfs_name *name1, 2746 struct xfs_inode *ip1, 2747 struct xfs_inode *dp2, 2748 struct xfs_name *name2, 2749 struct xfs_inode *ip2, 2750 struct xfs_defer_ops *dfops, 2751 xfs_fsblock_t *first_block, 2752 int spaceres) 2753 { 2754 int error = 0; 2755 int ip1_flags = 0; 2756 int ip2_flags = 0; 2757 int dp2_flags = 0; 2758 2759 /* Swap inode number for dirent in first parent */ 2760 error = xfs_dir_replace(tp, dp1, name1, 2761 ip2->i_ino, 2762 first_block, dfops, spaceres); 2763 if (error) 2764 goto out_trans_abort; 2765 2766 /* Swap inode number for dirent in second parent */ 2767 error = xfs_dir_replace(tp, dp2, name2, 2768 ip1->i_ino, 2769 first_block, dfops, spaceres); 2770 if (error) 2771 goto out_trans_abort; 2772 2773 /* 2774 * If we're renaming one or more directories across different parents, 2775 * update the respective ".." entries (and link counts) to match the new 2776 * parents. 2777 */ 2778 if (dp1 != dp2) { 2779 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2780 2781 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2782 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2783 dp1->i_ino, first_block, 2784 dfops, spaceres); 2785 if (error) 2786 goto out_trans_abort; 2787 2788 /* transfer ip2 ".." reference to dp1 */ 2789 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2790 error = xfs_droplink(tp, dp2); 2791 if (error) 2792 goto out_trans_abort; 2793 error = xfs_bumplink(tp, dp1); 2794 if (error) 2795 goto out_trans_abort; 2796 } 2797 2798 /* 2799 * Although ip1 isn't changed here, userspace needs 2800 * to be warned about the change, so that applications 2801 * relying on it (like backup ones), will properly 2802 * notify the change 2803 */ 2804 ip1_flags |= XFS_ICHGTIME_CHG; 2805 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2806 } 2807 2808 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2809 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2810 dp2->i_ino, first_block, 2811 dfops, spaceres); 2812 if (error) 2813 goto out_trans_abort; 2814 2815 /* transfer ip1 ".." reference to dp2 */ 2816 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2817 error = xfs_droplink(tp, dp1); 2818 if (error) 2819 goto out_trans_abort; 2820 error = xfs_bumplink(tp, dp2); 2821 if (error) 2822 goto out_trans_abort; 2823 } 2824 2825 /* 2826 * Although ip2 isn't changed here, userspace needs 2827 * to be warned about the change, so that applications 2828 * relying on it (like backup ones), will properly 2829 * notify the change 2830 */ 2831 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2832 ip2_flags |= XFS_ICHGTIME_CHG; 2833 } 2834 } 2835 2836 if (ip1_flags) { 2837 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2838 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2839 } 2840 if (ip2_flags) { 2841 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2842 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2843 } 2844 if (dp2_flags) { 2845 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2846 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2847 } 2848 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2849 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2850 return xfs_finish_rename(tp, dfops); 2851 2852 out_trans_abort: 2853 xfs_defer_cancel(dfops); 2854 xfs_trans_cancel(tp); 2855 return error; 2856 } 2857 2858 /* 2859 * xfs_rename_alloc_whiteout() 2860 * 2861 * Return a referenced, unlinked, unlocked inode that that can be used as a 2862 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2863 * crash between allocating the inode and linking it into the rename transaction 2864 * recovery will free the inode and we won't leak it. 2865 */ 2866 static int 2867 xfs_rename_alloc_whiteout( 2868 struct xfs_inode *dp, 2869 struct xfs_inode **wip) 2870 { 2871 struct xfs_inode *tmpfile; 2872 int error; 2873 2874 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2875 if (error) 2876 return error; 2877 2878 /* 2879 * Prepare the tmpfile inode as if it were created through the VFS. 2880 * Otherwise, the link increment paths will complain about nlink 0->1. 2881 * Drop the link count as done by d_tmpfile(), complete the inode setup 2882 * and flag it as linkable. 2883 */ 2884 drop_nlink(VFS_I(tmpfile)); 2885 xfs_setup_iops(tmpfile); 2886 xfs_finish_inode_setup(tmpfile); 2887 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2888 2889 *wip = tmpfile; 2890 return 0; 2891 } 2892 2893 /* 2894 * xfs_rename 2895 */ 2896 int 2897 xfs_rename( 2898 struct xfs_inode *src_dp, 2899 struct xfs_name *src_name, 2900 struct xfs_inode *src_ip, 2901 struct xfs_inode *target_dp, 2902 struct xfs_name *target_name, 2903 struct xfs_inode *target_ip, 2904 unsigned int flags) 2905 { 2906 struct xfs_mount *mp = src_dp->i_mount; 2907 struct xfs_trans *tp; 2908 struct xfs_defer_ops dfops; 2909 xfs_fsblock_t first_block; 2910 struct xfs_inode *wip = NULL; /* whiteout inode */ 2911 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2912 int num_inodes = __XFS_SORT_INODES; 2913 bool new_parent = (src_dp != target_dp); 2914 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2915 int spaceres; 2916 int error; 2917 2918 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2919 2920 if ((flags & RENAME_EXCHANGE) && !target_ip) 2921 return -EINVAL; 2922 2923 /* 2924 * If we are doing a whiteout operation, allocate the whiteout inode 2925 * we will be placing at the target and ensure the type is set 2926 * appropriately. 2927 */ 2928 if (flags & RENAME_WHITEOUT) { 2929 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2930 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2931 if (error) 2932 return error; 2933 2934 /* setup target dirent info as whiteout */ 2935 src_name->type = XFS_DIR3_FT_CHRDEV; 2936 } 2937 2938 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2939 inodes, &num_inodes); 2940 2941 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2942 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2943 if (error == -ENOSPC) { 2944 spaceres = 0; 2945 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2946 &tp); 2947 } 2948 if (error) 2949 goto out_release_wip; 2950 2951 /* 2952 * Attach the dquots to the inodes 2953 */ 2954 error = xfs_qm_vop_rename_dqattach(inodes); 2955 if (error) 2956 goto out_trans_cancel; 2957 2958 /* 2959 * Lock all the participating inodes. Depending upon whether 2960 * the target_name exists in the target directory, and 2961 * whether the target directory is the same as the source 2962 * directory, we can lock from 2 to 4 inodes. 2963 */ 2964 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2965 2966 /* 2967 * Join all the inodes to the transaction. From this point on, 2968 * we can rely on either trans_commit or trans_cancel to unlock 2969 * them. 2970 */ 2971 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 2972 if (new_parent) 2973 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 2974 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2975 if (target_ip) 2976 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2977 if (wip) 2978 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2979 2980 /* 2981 * If we are using project inheritance, we only allow renames 2982 * into our tree when the project IDs are the same; else the 2983 * tree quota mechanism would be circumvented. 2984 */ 2985 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2986 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2987 error = -EXDEV; 2988 goto out_trans_cancel; 2989 } 2990 2991 xfs_defer_init(&dfops, &first_block); 2992 2993 /* RENAME_EXCHANGE is unique from here on. */ 2994 if (flags & RENAME_EXCHANGE) 2995 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2996 target_dp, target_name, target_ip, 2997 &dfops, &first_block, spaceres); 2998 2999 /* 3000 * Set up the target. 3001 */ 3002 if (target_ip == NULL) { 3003 /* 3004 * If there's no space reservation, check the entry will 3005 * fit before actually inserting it. 3006 */ 3007 if (!spaceres) { 3008 error = xfs_dir_canenter(tp, target_dp, target_name); 3009 if (error) 3010 goto out_trans_cancel; 3011 } 3012 /* 3013 * If target does not exist and the rename crosses 3014 * directories, adjust the target directory link count 3015 * to account for the ".." reference from the new entry. 3016 */ 3017 error = xfs_dir_createname(tp, target_dp, target_name, 3018 src_ip->i_ino, &first_block, 3019 &dfops, spaceres); 3020 if (error) 3021 goto out_bmap_cancel; 3022 3023 xfs_trans_ichgtime(tp, target_dp, 3024 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3025 3026 if (new_parent && src_is_directory) { 3027 error = xfs_bumplink(tp, target_dp); 3028 if (error) 3029 goto out_bmap_cancel; 3030 } 3031 } else { /* target_ip != NULL */ 3032 /* 3033 * If target exists and it's a directory, check that both 3034 * target and source are directories and that target can be 3035 * destroyed, or that neither is a directory. 3036 */ 3037 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 3038 /* 3039 * Make sure target dir is empty. 3040 */ 3041 if (!(xfs_dir_isempty(target_ip)) || 3042 (VFS_I(target_ip)->i_nlink > 2)) { 3043 error = -EEXIST; 3044 goto out_trans_cancel; 3045 } 3046 } 3047 3048 /* 3049 * Link the source inode under the target name. 3050 * If the source inode is a directory and we are moving 3051 * it across directories, its ".." entry will be 3052 * inconsistent until we replace that down below. 3053 * 3054 * In case there is already an entry with the same 3055 * name at the destination directory, remove it first. 3056 */ 3057 error = xfs_dir_replace(tp, target_dp, target_name, 3058 src_ip->i_ino, 3059 &first_block, &dfops, spaceres); 3060 if (error) 3061 goto out_bmap_cancel; 3062 3063 xfs_trans_ichgtime(tp, target_dp, 3064 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3065 3066 /* 3067 * Decrement the link count on the target since the target 3068 * dir no longer points to it. 3069 */ 3070 error = xfs_droplink(tp, target_ip); 3071 if (error) 3072 goto out_bmap_cancel; 3073 3074 if (src_is_directory) { 3075 /* 3076 * Drop the link from the old "." entry. 3077 */ 3078 error = xfs_droplink(tp, target_ip); 3079 if (error) 3080 goto out_bmap_cancel; 3081 } 3082 } /* target_ip != NULL */ 3083 3084 /* 3085 * Remove the source. 3086 */ 3087 if (new_parent && src_is_directory) { 3088 /* 3089 * Rewrite the ".." entry to point to the new 3090 * directory. 3091 */ 3092 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3093 target_dp->i_ino, 3094 &first_block, &dfops, spaceres); 3095 ASSERT(error != -EEXIST); 3096 if (error) 3097 goto out_bmap_cancel; 3098 } 3099 3100 /* 3101 * We always want to hit the ctime on the source inode. 3102 * 3103 * This isn't strictly required by the standards since the source 3104 * inode isn't really being changed, but old unix file systems did 3105 * it and some incremental backup programs won't work without it. 3106 */ 3107 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3108 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3109 3110 /* 3111 * Adjust the link count on src_dp. This is necessary when 3112 * renaming a directory, either within one parent when 3113 * the target existed, or across two parent directories. 3114 */ 3115 if (src_is_directory && (new_parent || target_ip != NULL)) { 3116 3117 /* 3118 * Decrement link count on src_directory since the 3119 * entry that's moved no longer points to it. 3120 */ 3121 error = xfs_droplink(tp, src_dp); 3122 if (error) 3123 goto out_bmap_cancel; 3124 } 3125 3126 /* 3127 * For whiteouts, we only need to update the source dirent with the 3128 * inode number of the whiteout inode rather than removing it 3129 * altogether. 3130 */ 3131 if (wip) { 3132 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3133 &first_block, &dfops, spaceres); 3134 } else 3135 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3136 &first_block, &dfops, spaceres); 3137 if (error) 3138 goto out_bmap_cancel; 3139 3140 /* 3141 * For whiteouts, we need to bump the link count on the whiteout inode. 3142 * This means that failures all the way up to this point leave the inode 3143 * on the unlinked list and so cleanup is a simple matter of dropping 3144 * the remaining reference to it. If we fail here after bumping the link 3145 * count, we're shutting down the filesystem so we'll never see the 3146 * intermediate state on disk. 3147 */ 3148 if (wip) { 3149 ASSERT(VFS_I(wip)->i_nlink == 0); 3150 error = xfs_bumplink(tp, wip); 3151 if (error) 3152 goto out_bmap_cancel; 3153 error = xfs_iunlink_remove(tp, wip); 3154 if (error) 3155 goto out_bmap_cancel; 3156 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3157 3158 /* 3159 * Now we have a real link, clear the "I'm a tmpfile" state 3160 * flag from the inode so it doesn't accidentally get misused in 3161 * future. 3162 */ 3163 VFS_I(wip)->i_state &= ~I_LINKABLE; 3164 } 3165 3166 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3167 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3168 if (new_parent) 3169 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3170 3171 error = xfs_finish_rename(tp, &dfops); 3172 if (wip) 3173 IRELE(wip); 3174 return error; 3175 3176 out_bmap_cancel: 3177 xfs_defer_cancel(&dfops); 3178 out_trans_cancel: 3179 xfs_trans_cancel(tp); 3180 out_release_wip: 3181 if (wip) 3182 IRELE(wip); 3183 return error; 3184 } 3185 3186 STATIC int 3187 xfs_iflush_cluster( 3188 struct xfs_inode *ip, 3189 struct xfs_buf *bp) 3190 { 3191 struct xfs_mount *mp = ip->i_mount; 3192 struct xfs_perag *pag; 3193 unsigned long first_index, mask; 3194 unsigned long inodes_per_cluster; 3195 int cilist_size; 3196 struct xfs_inode **cilist; 3197 struct xfs_inode *cip; 3198 int nr_found; 3199 int clcount = 0; 3200 int bufwasdelwri; 3201 int i; 3202 3203 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3204 3205 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3206 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3207 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3208 if (!cilist) 3209 goto out_put; 3210 3211 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3212 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3213 rcu_read_lock(); 3214 /* really need a gang lookup range call here */ 3215 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3216 first_index, inodes_per_cluster); 3217 if (nr_found == 0) 3218 goto out_free; 3219 3220 for (i = 0; i < nr_found; i++) { 3221 cip = cilist[i]; 3222 if (cip == ip) 3223 continue; 3224 3225 /* 3226 * because this is an RCU protected lookup, we could find a 3227 * recently freed or even reallocated inode during the lookup. 3228 * We need to check under the i_flags_lock for a valid inode 3229 * here. Skip it if it is not valid or the wrong inode. 3230 */ 3231 spin_lock(&cip->i_flags_lock); 3232 if (!cip->i_ino || 3233 __xfs_iflags_test(cip, XFS_ISTALE)) { 3234 spin_unlock(&cip->i_flags_lock); 3235 continue; 3236 } 3237 3238 /* 3239 * Once we fall off the end of the cluster, no point checking 3240 * any more inodes in the list because they will also all be 3241 * outside the cluster. 3242 */ 3243 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3244 spin_unlock(&cip->i_flags_lock); 3245 break; 3246 } 3247 spin_unlock(&cip->i_flags_lock); 3248 3249 /* 3250 * Do an un-protected check to see if the inode is dirty and 3251 * is a candidate for flushing. These checks will be repeated 3252 * later after the appropriate locks are acquired. 3253 */ 3254 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3255 continue; 3256 3257 /* 3258 * Try to get locks. If any are unavailable or it is pinned, 3259 * then this inode cannot be flushed and is skipped. 3260 */ 3261 3262 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3263 continue; 3264 if (!xfs_iflock_nowait(cip)) { 3265 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3266 continue; 3267 } 3268 if (xfs_ipincount(cip)) { 3269 xfs_ifunlock(cip); 3270 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3271 continue; 3272 } 3273 3274 3275 /* 3276 * Check the inode number again, just to be certain we are not 3277 * racing with freeing in xfs_reclaim_inode(). See the comments 3278 * in that function for more information as to why the initial 3279 * check is not sufficient. 3280 */ 3281 if (!cip->i_ino) { 3282 xfs_ifunlock(cip); 3283 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3284 continue; 3285 } 3286 3287 /* 3288 * arriving here means that this inode can be flushed. First 3289 * re-check that it's dirty before flushing. 3290 */ 3291 if (!xfs_inode_clean(cip)) { 3292 int error; 3293 error = xfs_iflush_int(cip, bp); 3294 if (error) { 3295 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3296 goto cluster_corrupt_out; 3297 } 3298 clcount++; 3299 } else { 3300 xfs_ifunlock(cip); 3301 } 3302 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3303 } 3304 3305 if (clcount) { 3306 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3307 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3308 } 3309 3310 out_free: 3311 rcu_read_unlock(); 3312 kmem_free(cilist); 3313 out_put: 3314 xfs_perag_put(pag); 3315 return 0; 3316 3317 3318 cluster_corrupt_out: 3319 /* 3320 * Corruption detected in the clustering loop. Invalidate the 3321 * inode buffer and shut down the filesystem. 3322 */ 3323 rcu_read_unlock(); 3324 /* 3325 * Clean up the buffer. If it was delwri, just release it -- 3326 * brelse can handle it with no problems. If not, shut down the 3327 * filesystem before releasing the buffer. 3328 */ 3329 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3330 if (bufwasdelwri) 3331 xfs_buf_relse(bp); 3332 3333 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3334 3335 if (!bufwasdelwri) { 3336 /* 3337 * Just like incore_relse: if we have b_iodone functions, 3338 * mark the buffer as an error and call them. Otherwise 3339 * mark it as stale and brelse. 3340 */ 3341 if (bp->b_iodone) { 3342 bp->b_flags &= ~XBF_DONE; 3343 xfs_buf_stale(bp); 3344 xfs_buf_ioerror(bp, -EIO); 3345 xfs_buf_ioend(bp); 3346 } else { 3347 xfs_buf_stale(bp); 3348 xfs_buf_relse(bp); 3349 } 3350 } 3351 3352 /* 3353 * Unlocks the flush lock 3354 */ 3355 xfs_iflush_abort(cip, false); 3356 kmem_free(cilist); 3357 xfs_perag_put(pag); 3358 return -EFSCORRUPTED; 3359 } 3360 3361 /* 3362 * Flush dirty inode metadata into the backing buffer. 3363 * 3364 * The caller must have the inode lock and the inode flush lock held. The 3365 * inode lock will still be held upon return to the caller, and the inode 3366 * flush lock will be released after the inode has reached the disk. 3367 * 3368 * The caller must write out the buffer returned in *bpp and release it. 3369 */ 3370 int 3371 xfs_iflush( 3372 struct xfs_inode *ip, 3373 struct xfs_buf **bpp) 3374 { 3375 struct xfs_mount *mp = ip->i_mount; 3376 struct xfs_buf *bp = NULL; 3377 struct xfs_dinode *dip; 3378 int error; 3379 3380 XFS_STATS_INC(mp, xs_iflush_count); 3381 3382 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3383 ASSERT(xfs_isiflocked(ip)); 3384 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3385 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3386 3387 *bpp = NULL; 3388 3389 xfs_iunpin_wait(ip); 3390 3391 /* 3392 * For stale inodes we cannot rely on the backing buffer remaining 3393 * stale in cache for the remaining life of the stale inode and so 3394 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3395 * inodes below. We have to check this after ensuring the inode is 3396 * unpinned so that it is safe to reclaim the stale inode after the 3397 * flush call. 3398 */ 3399 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3400 xfs_ifunlock(ip); 3401 return 0; 3402 } 3403 3404 /* 3405 * This may have been unpinned because the filesystem is shutting 3406 * down forcibly. If that's the case we must not write this inode 3407 * to disk, because the log record didn't make it to disk. 3408 * 3409 * We also have to remove the log item from the AIL in this case, 3410 * as we wait for an empty AIL as part of the unmount process. 3411 */ 3412 if (XFS_FORCED_SHUTDOWN(mp)) { 3413 error = -EIO; 3414 goto abort_out; 3415 } 3416 3417 /* 3418 * Get the buffer containing the on-disk inode. We are doing a try-lock 3419 * operation here, so we may get an EAGAIN error. In that case, we 3420 * simply want to return with the inode still dirty. 3421 * 3422 * If we get any other error, we effectively have a corruption situation 3423 * and we cannot flush the inode, so we treat it the same as failing 3424 * xfs_iflush_int(). 3425 */ 3426 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3427 0); 3428 if (error == -EAGAIN) { 3429 xfs_ifunlock(ip); 3430 return error; 3431 } 3432 if (error) 3433 goto corrupt_out; 3434 3435 /* 3436 * First flush out the inode that xfs_iflush was called with. 3437 */ 3438 error = xfs_iflush_int(ip, bp); 3439 if (error) 3440 goto corrupt_out; 3441 3442 /* 3443 * If the buffer is pinned then push on the log now so we won't 3444 * get stuck waiting in the write for too long. 3445 */ 3446 if (xfs_buf_ispinned(bp)) 3447 xfs_log_force(mp, 0); 3448 3449 /* 3450 * inode clustering: 3451 * see if other inodes can be gathered into this write 3452 */ 3453 error = xfs_iflush_cluster(ip, bp); 3454 if (error) 3455 goto cluster_corrupt_out; 3456 3457 *bpp = bp; 3458 return 0; 3459 3460 corrupt_out: 3461 if (bp) 3462 xfs_buf_relse(bp); 3463 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3464 cluster_corrupt_out: 3465 error = -EFSCORRUPTED; 3466 abort_out: 3467 /* 3468 * Unlocks the flush lock 3469 */ 3470 xfs_iflush_abort(ip, false); 3471 return error; 3472 } 3473 3474 STATIC int 3475 xfs_iflush_int( 3476 struct xfs_inode *ip, 3477 struct xfs_buf *bp) 3478 { 3479 struct xfs_inode_log_item *iip = ip->i_itemp; 3480 struct xfs_dinode *dip; 3481 struct xfs_mount *mp = ip->i_mount; 3482 3483 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3484 ASSERT(xfs_isiflocked(ip)); 3485 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3486 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3487 ASSERT(iip != NULL && iip->ili_fields != 0); 3488 ASSERT(ip->i_d.di_version > 1); 3489 3490 /* set *dip = inode's place in the buffer */ 3491 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3492 3493 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3494 mp, XFS_ERRTAG_IFLUSH_1)) { 3495 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3496 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3497 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3498 goto corrupt_out; 3499 } 3500 if (S_ISREG(VFS_I(ip)->i_mode)) { 3501 if (XFS_TEST_ERROR( 3502 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3503 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3504 mp, XFS_ERRTAG_IFLUSH_3)) { 3505 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3506 "%s: Bad regular inode %Lu, ptr 0x%p", 3507 __func__, ip->i_ino, ip); 3508 goto corrupt_out; 3509 } 3510 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3511 if (XFS_TEST_ERROR( 3512 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3513 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3514 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3515 mp, XFS_ERRTAG_IFLUSH_4)) { 3516 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3517 "%s: Bad directory inode %Lu, ptr 0x%p", 3518 __func__, ip->i_ino, ip); 3519 goto corrupt_out; 3520 } 3521 } 3522 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3523 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { 3524 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3525 "%s: detected corrupt incore inode %Lu, " 3526 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3527 __func__, ip->i_ino, 3528 ip->i_d.di_nextents + ip->i_d.di_anextents, 3529 ip->i_d.di_nblocks, ip); 3530 goto corrupt_out; 3531 } 3532 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3533 mp, XFS_ERRTAG_IFLUSH_6)) { 3534 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3535 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3536 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3537 goto corrupt_out; 3538 } 3539 3540 /* 3541 * Inode item log recovery for v2 inodes are dependent on the 3542 * di_flushiter count for correct sequencing. We bump the flush 3543 * iteration count so we can detect flushes which postdate a log record 3544 * during recovery. This is redundant as we now log every change and 3545 * hence this can't happen but we need to still do it to ensure 3546 * backwards compatibility with old kernels that predate logging all 3547 * inode changes. 3548 */ 3549 if (ip->i_d.di_version < 3) 3550 ip->i_d.di_flushiter++; 3551 3552 /* Check the inline directory data. */ 3553 if (S_ISDIR(VFS_I(ip)->i_mode) && 3554 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL && 3555 xfs_dir2_sf_verify(ip)) 3556 goto corrupt_out; 3557 3558 /* 3559 * Copy the dirty parts of the inode into the on-disk inode. We always 3560 * copy out the core of the inode, because if the inode is dirty at all 3561 * the core must be. 3562 */ 3563 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3564 3565 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3566 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3567 ip->i_d.di_flushiter = 0; 3568 3569 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3570 if (XFS_IFORK_Q(ip)) 3571 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3572 xfs_inobp_check(mp, bp); 3573 3574 /* 3575 * We've recorded everything logged in the inode, so we'd like to clear 3576 * the ili_fields bits so we don't log and flush things unnecessarily. 3577 * However, we can't stop logging all this information until the data 3578 * we've copied into the disk buffer is written to disk. If we did we 3579 * might overwrite the copy of the inode in the log with all the data 3580 * after re-logging only part of it, and in the face of a crash we 3581 * wouldn't have all the data we need to recover. 3582 * 3583 * What we do is move the bits to the ili_last_fields field. When 3584 * logging the inode, these bits are moved back to the ili_fields field. 3585 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3586 * know that the information those bits represent is permanently on 3587 * disk. As long as the flush completes before the inode is logged 3588 * again, then both ili_fields and ili_last_fields will be cleared. 3589 * 3590 * We can play with the ili_fields bits here, because the inode lock 3591 * must be held exclusively in order to set bits there and the flush 3592 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3593 * done routine can tell whether or not to look in the AIL. Also, store 3594 * the current LSN of the inode so that we can tell whether the item has 3595 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3596 * need the AIL lock, because it is a 64 bit value that cannot be read 3597 * atomically. 3598 */ 3599 iip->ili_last_fields = iip->ili_fields; 3600 iip->ili_fields = 0; 3601 iip->ili_fsync_fields = 0; 3602 iip->ili_logged = 1; 3603 3604 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3605 &iip->ili_item.li_lsn); 3606 3607 /* 3608 * Attach the function xfs_iflush_done to the inode's 3609 * buffer. This will remove the inode from the AIL 3610 * and unlock the inode's flush lock when the inode is 3611 * completely written to disk. 3612 */ 3613 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3614 3615 /* generate the checksum. */ 3616 xfs_dinode_calc_crc(mp, dip); 3617 3618 ASSERT(bp->b_fspriv != NULL); 3619 ASSERT(bp->b_iodone != NULL); 3620 return 0; 3621 3622 corrupt_out: 3623 return -EFSCORRUPTED; 3624 } 3625