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_errortag.h" 43 #include "xfs_error.h" 44 #include "xfs_quota.h" 45 #include "xfs_filestream.h" 46 #include "xfs_cksum.h" 47 #include "xfs_trace.h" 48 #include "xfs_icache.h" 49 #include "xfs_symlink.h" 50 #include "xfs_trans_priv.h" 51 #include "xfs_log.h" 52 #include "xfs_bmap_btree.h" 53 #include "xfs_reflink.h" 54 #include "xfs_dir2_priv.h" 55 56 kmem_zone_t *xfs_inode_zone; 57 58 /* 59 * Used in xfs_itruncate_extents(). This is the maximum number of extents 60 * freed from a file in a single transaction. 61 */ 62 #define XFS_ITRUNC_MAX_EXTENTS 2 63 64 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); 65 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); 66 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); 67 68 /* 69 * helper function to extract extent size hint from inode 70 */ 71 xfs_extlen_t 72 xfs_get_extsz_hint( 73 struct xfs_inode *ip) 74 { 75 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 76 return ip->i_d.di_extsize; 77 if (XFS_IS_REALTIME_INODE(ip)) 78 return ip->i_mount->m_sb.sb_rextsize; 79 return 0; 80 } 81 82 /* 83 * Helper function to extract CoW extent size hint from inode. 84 * Between the extent size hint and the CoW extent size hint, we 85 * return the greater of the two. If the value is zero (automatic), 86 * use the default size. 87 */ 88 xfs_extlen_t 89 xfs_get_cowextsz_hint( 90 struct xfs_inode *ip) 91 { 92 xfs_extlen_t a, b; 93 94 a = 0; 95 if (ip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 96 a = ip->i_d.di_cowextsize; 97 b = xfs_get_extsz_hint(ip); 98 99 a = max(a, b); 100 if (a == 0) 101 return XFS_DEFAULT_COWEXTSZ_HINT; 102 return a; 103 } 104 105 /* 106 * These two are wrapper routines around the xfs_ilock() routine used to 107 * centralize some grungy code. They are used in places that wish to lock the 108 * inode solely for reading the extents. The reason these places can't just 109 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to 110 * bringing in of the extents from disk for a file in b-tree format. If the 111 * inode is in b-tree format, then we need to lock the inode exclusively until 112 * the extents are read in. Locking it exclusively all the time would limit 113 * our parallelism unnecessarily, though. What we do instead is check to see 114 * if the extents have been read in yet, and only lock the inode exclusively 115 * if they have not. 116 * 117 * The functions return a value which should be given to the corresponding 118 * xfs_iunlock() call. 119 */ 120 uint 121 xfs_ilock_data_map_shared( 122 struct xfs_inode *ip) 123 { 124 uint lock_mode = XFS_ILOCK_SHARED; 125 126 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 127 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) 128 lock_mode = XFS_ILOCK_EXCL; 129 xfs_ilock(ip, lock_mode); 130 return lock_mode; 131 } 132 133 uint 134 xfs_ilock_attr_map_shared( 135 struct xfs_inode *ip) 136 { 137 uint lock_mode = XFS_ILOCK_SHARED; 138 139 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && 140 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) 141 lock_mode = XFS_ILOCK_EXCL; 142 xfs_ilock(ip, lock_mode); 143 return lock_mode; 144 } 145 146 /* 147 * In addition to i_rwsem in the VFS inode, the xfs inode contains 2 148 * multi-reader locks: i_mmap_lock and the i_lock. This routine allows 149 * various combinations of the locks to be obtained. 150 * 151 * The 3 locks should always be ordered so that the IO lock is obtained first, 152 * the mmap lock second and the ilock last in order to prevent deadlock. 153 * 154 * Basic locking order: 155 * 156 * i_rwsem -> i_mmap_lock -> page_lock -> i_ilock 157 * 158 * mmap_sem locking order: 159 * 160 * i_rwsem -> page lock -> mmap_sem 161 * mmap_sem -> i_mmap_lock -> page_lock 162 * 163 * The difference in mmap_sem locking order mean that we cannot hold the 164 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can 165 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem 166 * in get_user_pages() to map the user pages into the kernel address space for 167 * direct IO. Similarly the i_rwsem cannot be taken inside a page fault because 168 * page faults already hold the mmap_sem. 169 * 170 * Hence to serialise fully against both syscall and mmap based IO, we need to 171 * take both the i_rwsem and the i_mmap_lock. These locks should *only* be both 172 * taken in places where we need to invalidate the page cache in a race 173 * free manner (e.g. truncate, hole punch and other extent manipulation 174 * functions). 175 */ 176 void 177 xfs_ilock( 178 xfs_inode_t *ip, 179 uint lock_flags) 180 { 181 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 182 183 /* 184 * You can't set both SHARED and EXCL for the same lock, 185 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 186 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 187 */ 188 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 189 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 190 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 191 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 192 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 193 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 194 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 195 196 if (lock_flags & XFS_IOLOCK_EXCL) { 197 down_write_nested(&VFS_I(ip)->i_rwsem, 198 XFS_IOLOCK_DEP(lock_flags)); 199 } else if (lock_flags & XFS_IOLOCK_SHARED) { 200 down_read_nested(&VFS_I(ip)->i_rwsem, 201 XFS_IOLOCK_DEP(lock_flags)); 202 } 203 204 if (lock_flags & XFS_MMAPLOCK_EXCL) 205 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 206 else if (lock_flags & XFS_MMAPLOCK_SHARED) 207 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 208 209 if (lock_flags & XFS_ILOCK_EXCL) 210 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 211 else if (lock_flags & XFS_ILOCK_SHARED) 212 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 213 } 214 215 /* 216 * This is just like xfs_ilock(), except that the caller 217 * is guaranteed not to sleep. It returns 1 if it gets 218 * the requested locks and 0 otherwise. If the IO lock is 219 * obtained but the inode lock cannot be, then the IO lock 220 * is dropped before returning. 221 * 222 * ip -- the inode being locked 223 * lock_flags -- this parameter indicates the inode's locks to be 224 * to be locked. See the comment for xfs_ilock() for a list 225 * of valid values. 226 */ 227 int 228 xfs_ilock_nowait( 229 xfs_inode_t *ip, 230 uint lock_flags) 231 { 232 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 233 234 /* 235 * You can't set both SHARED and EXCL for the same lock, 236 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 237 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 238 */ 239 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 240 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 241 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 242 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 243 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 244 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 245 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 246 247 if (lock_flags & XFS_IOLOCK_EXCL) { 248 if (!down_write_trylock(&VFS_I(ip)->i_rwsem)) 249 goto out; 250 } else if (lock_flags & XFS_IOLOCK_SHARED) { 251 if (!down_read_trylock(&VFS_I(ip)->i_rwsem)) 252 goto out; 253 } 254 255 if (lock_flags & XFS_MMAPLOCK_EXCL) { 256 if (!mrtryupdate(&ip->i_mmaplock)) 257 goto out_undo_iolock; 258 } else if (lock_flags & XFS_MMAPLOCK_SHARED) { 259 if (!mrtryaccess(&ip->i_mmaplock)) 260 goto out_undo_iolock; 261 } 262 263 if (lock_flags & XFS_ILOCK_EXCL) { 264 if (!mrtryupdate(&ip->i_lock)) 265 goto out_undo_mmaplock; 266 } else if (lock_flags & XFS_ILOCK_SHARED) { 267 if (!mrtryaccess(&ip->i_lock)) 268 goto out_undo_mmaplock; 269 } 270 return 1; 271 272 out_undo_mmaplock: 273 if (lock_flags & XFS_MMAPLOCK_EXCL) 274 mrunlock_excl(&ip->i_mmaplock); 275 else if (lock_flags & XFS_MMAPLOCK_SHARED) 276 mrunlock_shared(&ip->i_mmaplock); 277 out_undo_iolock: 278 if (lock_flags & XFS_IOLOCK_EXCL) 279 up_write(&VFS_I(ip)->i_rwsem); 280 else if (lock_flags & XFS_IOLOCK_SHARED) 281 up_read(&VFS_I(ip)->i_rwsem); 282 out: 283 return 0; 284 } 285 286 /* 287 * xfs_iunlock() is used to drop the inode locks acquired with 288 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 289 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 290 * that we know which locks to drop. 291 * 292 * ip -- the inode being unlocked 293 * lock_flags -- this parameter indicates the inode's locks to be 294 * to be unlocked. See the comment for xfs_ilock() for a list 295 * of valid values for this parameter. 296 * 297 */ 298 void 299 xfs_iunlock( 300 xfs_inode_t *ip, 301 uint lock_flags) 302 { 303 /* 304 * You can't set both SHARED and EXCL for the same lock, 305 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 306 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 307 */ 308 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 309 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 310 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 311 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 312 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 313 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 314 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 315 ASSERT(lock_flags != 0); 316 317 if (lock_flags & XFS_IOLOCK_EXCL) 318 up_write(&VFS_I(ip)->i_rwsem); 319 else if (lock_flags & XFS_IOLOCK_SHARED) 320 up_read(&VFS_I(ip)->i_rwsem); 321 322 if (lock_flags & XFS_MMAPLOCK_EXCL) 323 mrunlock_excl(&ip->i_mmaplock); 324 else if (lock_flags & XFS_MMAPLOCK_SHARED) 325 mrunlock_shared(&ip->i_mmaplock); 326 327 if (lock_flags & XFS_ILOCK_EXCL) 328 mrunlock_excl(&ip->i_lock); 329 else if (lock_flags & XFS_ILOCK_SHARED) 330 mrunlock_shared(&ip->i_lock); 331 332 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 333 } 334 335 /* 336 * give up write locks. the i/o lock cannot be held nested 337 * if it is being demoted. 338 */ 339 void 340 xfs_ilock_demote( 341 xfs_inode_t *ip, 342 uint lock_flags) 343 { 344 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); 345 ASSERT((lock_flags & 346 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 347 348 if (lock_flags & XFS_ILOCK_EXCL) 349 mrdemote(&ip->i_lock); 350 if (lock_flags & XFS_MMAPLOCK_EXCL) 351 mrdemote(&ip->i_mmaplock); 352 if (lock_flags & XFS_IOLOCK_EXCL) 353 downgrade_write(&VFS_I(ip)->i_rwsem); 354 355 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 356 } 357 358 #if defined(DEBUG) || defined(XFS_WARN) 359 int 360 xfs_isilocked( 361 xfs_inode_t *ip, 362 uint lock_flags) 363 { 364 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 365 if (!(lock_flags & XFS_ILOCK_SHARED)) 366 return !!ip->i_lock.mr_writer; 367 return rwsem_is_locked(&ip->i_lock.mr_lock); 368 } 369 370 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { 371 if (!(lock_flags & XFS_MMAPLOCK_SHARED)) 372 return !!ip->i_mmaplock.mr_writer; 373 return rwsem_is_locked(&ip->i_mmaplock.mr_lock); 374 } 375 376 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 377 if (!(lock_flags & XFS_IOLOCK_SHARED)) 378 return !debug_locks || 379 lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0); 380 return rwsem_is_locked(&VFS_I(ip)->i_rwsem); 381 } 382 383 ASSERT(0); 384 return 0; 385 } 386 #endif 387 388 /* 389 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when 390 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined 391 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build 392 * errors and warnings. 393 */ 394 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) 395 static bool 396 xfs_lockdep_subclass_ok( 397 int subclass) 398 { 399 return subclass < MAX_LOCKDEP_SUBCLASSES; 400 } 401 #else 402 #define xfs_lockdep_subclass_ok(subclass) (true) 403 #endif 404 405 /* 406 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 407 * value. This can be called for any type of inode lock combination, including 408 * parent locking. Care must be taken to ensure we don't overrun the subclass 409 * storage fields in the class mask we build. 410 */ 411 static inline int 412 xfs_lock_inumorder(int lock_mode, int subclass) 413 { 414 int class = 0; 415 416 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | 417 XFS_ILOCK_RTSUM))); 418 ASSERT(xfs_lockdep_subclass_ok(subclass)); 419 420 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 421 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); 422 class += subclass << XFS_IOLOCK_SHIFT; 423 } 424 425 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 426 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); 427 class += subclass << XFS_MMAPLOCK_SHIFT; 428 } 429 430 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { 431 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); 432 class += subclass << XFS_ILOCK_SHIFT; 433 } 434 435 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; 436 } 437 438 /* 439 * The following routine will lock n inodes in exclusive mode. We assume the 440 * caller calls us with the inodes in i_ino order. 441 * 442 * We need to detect deadlock where an inode that we lock is in the AIL and we 443 * start waiting for another inode that is locked by a thread in a long running 444 * transaction (such as truncate). This can result in deadlock since the long 445 * running trans might need to wait for the inode we just locked in order to 446 * push the tail and free space in the log. 447 * 448 * xfs_lock_inodes() can only be used to lock one type of lock at a time - 449 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 450 * lock more than one at a time, lockdep will report false positives saying we 451 * have violated locking orders. 452 */ 453 static void 454 xfs_lock_inodes( 455 xfs_inode_t **ips, 456 int inodes, 457 uint lock_mode) 458 { 459 int attempts = 0, i, j, try_lock; 460 xfs_log_item_t *lp; 461 462 /* 463 * Currently supports between 2 and 5 inodes with exclusive locking. We 464 * support an arbitrary depth of locking here, but absolute limits on 465 * inodes depend on the the type of locking and the limits placed by 466 * lockdep annotations in xfs_lock_inumorder. These are all checked by 467 * the asserts. 468 */ 469 ASSERT(ips && inodes >= 2 && inodes <= 5); 470 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | 471 XFS_ILOCK_EXCL)); 472 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | 473 XFS_ILOCK_SHARED))); 474 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || 475 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); 476 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || 477 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); 478 479 if (lock_mode & XFS_IOLOCK_EXCL) { 480 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); 481 } else if (lock_mode & XFS_MMAPLOCK_EXCL) 482 ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); 483 484 try_lock = 0; 485 i = 0; 486 again: 487 for (; i < inodes; i++) { 488 ASSERT(ips[i]); 489 490 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 491 continue; 492 493 /* 494 * If try_lock is not set yet, make sure all locked inodes are 495 * not in the AIL. If any are, set try_lock to be used later. 496 */ 497 if (!try_lock) { 498 for (j = (i - 1); j >= 0 && !try_lock; j--) { 499 lp = (xfs_log_item_t *)ips[j]->i_itemp; 500 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) 501 try_lock++; 502 } 503 } 504 505 /* 506 * If any of the previous locks we have locked is in the AIL, 507 * we must TRY to get the second and subsequent locks. If 508 * we can't get any, we must release all we have 509 * and try again. 510 */ 511 if (!try_lock) { 512 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 513 continue; 514 } 515 516 /* try_lock means we have an inode locked that is in the AIL. */ 517 ASSERT(i != 0); 518 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 519 continue; 520 521 /* 522 * Unlock all previous guys and try again. xfs_iunlock will try 523 * to push the tail if the inode is in the AIL. 524 */ 525 attempts++; 526 for (j = i - 1; j >= 0; j--) { 527 /* 528 * Check to see if we've already unlocked this one. Not 529 * the first one going back, and the inode ptr is the 530 * same. 531 */ 532 if (j != (i - 1) && ips[j] == ips[j + 1]) 533 continue; 534 535 xfs_iunlock(ips[j], lock_mode); 536 } 537 538 if ((attempts % 5) == 0) { 539 delay(1); /* Don't just spin the CPU */ 540 } 541 i = 0; 542 try_lock = 0; 543 goto again; 544 } 545 } 546 547 /* 548 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 549 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 550 * lock more than one at a time, lockdep will report false positives saying we 551 * have violated locking orders. 552 */ 553 void 554 xfs_lock_two_inodes( 555 xfs_inode_t *ip0, 556 xfs_inode_t *ip1, 557 uint lock_mode) 558 { 559 xfs_inode_t *temp; 560 int attempts = 0; 561 xfs_log_item_t *lp; 562 563 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL))); 564 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) 565 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 566 567 ASSERT(ip0->i_ino != ip1->i_ino); 568 569 if (ip0->i_ino > ip1->i_ino) { 570 temp = ip0; 571 ip0 = ip1; 572 ip1 = temp; 573 } 574 575 again: 576 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); 577 578 /* 579 * If the first lock we have locked is in the AIL, we must TRY to get 580 * the second lock. If we can't get it, we must release the first one 581 * and try again. 582 */ 583 lp = (xfs_log_item_t *)ip0->i_itemp; 584 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { 585 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { 586 xfs_iunlock(ip0, lock_mode); 587 if ((++attempts % 5) == 0) 588 delay(1); /* Don't just spin the CPU */ 589 goto again; 590 } 591 } else { 592 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); 593 } 594 } 595 596 597 void 598 __xfs_iflock( 599 struct xfs_inode *ip) 600 { 601 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 602 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 603 604 do { 605 prepare_to_wait_exclusive(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 606 if (xfs_isiflocked(ip)) 607 io_schedule(); 608 } while (!xfs_iflock_nowait(ip)); 609 610 finish_wait(wq, &wait.wq_entry); 611 } 612 613 STATIC uint 614 _xfs_dic2xflags( 615 uint16_t di_flags, 616 uint64_t di_flags2, 617 bool has_attr) 618 { 619 uint flags = 0; 620 621 if (di_flags & XFS_DIFLAG_ANY) { 622 if (di_flags & XFS_DIFLAG_REALTIME) 623 flags |= FS_XFLAG_REALTIME; 624 if (di_flags & XFS_DIFLAG_PREALLOC) 625 flags |= FS_XFLAG_PREALLOC; 626 if (di_flags & XFS_DIFLAG_IMMUTABLE) 627 flags |= FS_XFLAG_IMMUTABLE; 628 if (di_flags & XFS_DIFLAG_APPEND) 629 flags |= FS_XFLAG_APPEND; 630 if (di_flags & XFS_DIFLAG_SYNC) 631 flags |= FS_XFLAG_SYNC; 632 if (di_flags & XFS_DIFLAG_NOATIME) 633 flags |= FS_XFLAG_NOATIME; 634 if (di_flags & XFS_DIFLAG_NODUMP) 635 flags |= FS_XFLAG_NODUMP; 636 if (di_flags & XFS_DIFLAG_RTINHERIT) 637 flags |= FS_XFLAG_RTINHERIT; 638 if (di_flags & XFS_DIFLAG_PROJINHERIT) 639 flags |= FS_XFLAG_PROJINHERIT; 640 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 641 flags |= FS_XFLAG_NOSYMLINKS; 642 if (di_flags & XFS_DIFLAG_EXTSIZE) 643 flags |= FS_XFLAG_EXTSIZE; 644 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 645 flags |= FS_XFLAG_EXTSZINHERIT; 646 if (di_flags & XFS_DIFLAG_NODEFRAG) 647 flags |= FS_XFLAG_NODEFRAG; 648 if (di_flags & XFS_DIFLAG_FILESTREAM) 649 flags |= FS_XFLAG_FILESTREAM; 650 } 651 652 if (di_flags2 & XFS_DIFLAG2_ANY) { 653 if (di_flags2 & XFS_DIFLAG2_DAX) 654 flags |= FS_XFLAG_DAX; 655 if (di_flags2 & XFS_DIFLAG2_COWEXTSIZE) 656 flags |= FS_XFLAG_COWEXTSIZE; 657 } 658 659 if (has_attr) 660 flags |= FS_XFLAG_HASATTR; 661 662 return flags; 663 } 664 665 uint 666 xfs_ip2xflags( 667 struct xfs_inode *ip) 668 { 669 struct xfs_icdinode *dic = &ip->i_d; 670 671 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); 672 } 673 674 /* 675 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 676 * is allowed, otherwise it has to be an exact match. If a CI match is found, 677 * ci_name->name will point to a the actual name (caller must free) or 678 * will be set to NULL if an exact match is found. 679 */ 680 int 681 xfs_lookup( 682 xfs_inode_t *dp, 683 struct xfs_name *name, 684 xfs_inode_t **ipp, 685 struct xfs_name *ci_name) 686 { 687 xfs_ino_t inum; 688 int error; 689 690 trace_xfs_lookup(dp, name); 691 692 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 693 return -EIO; 694 695 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 696 if (error) 697 goto out_unlock; 698 699 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 700 if (error) 701 goto out_free_name; 702 703 return 0; 704 705 out_free_name: 706 if (ci_name) 707 kmem_free(ci_name->name); 708 out_unlock: 709 *ipp = NULL; 710 return error; 711 } 712 713 /* 714 * Allocate an inode on disk and return a copy of its in-core version. 715 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 716 * appropriately within the inode. The uid and gid for the inode are 717 * set according to the contents of the given cred structure. 718 * 719 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 720 * has a free inode available, call xfs_iget() to obtain the in-core 721 * version of the allocated inode. Finally, fill in the inode and 722 * log its initial contents. In this case, ialloc_context would be 723 * set to NULL. 724 * 725 * If xfs_dialloc() does not have an available inode, it will replenish 726 * its supply by doing an allocation. Since we can only do one 727 * allocation within a transaction without deadlocks, we must commit 728 * the current transaction before returning the inode itself. 729 * In this case, therefore, we will set ialloc_context and return. 730 * The caller should then commit the current transaction, start a new 731 * transaction, and call xfs_ialloc() again to actually get the inode. 732 * 733 * To ensure that some other process does not grab the inode that 734 * was allocated during the first call to xfs_ialloc(), this routine 735 * also returns the [locked] bp pointing to the head of the freelist 736 * as ialloc_context. The caller should hold this buffer across 737 * the commit and pass it back into this routine on the second call. 738 * 739 * If we are allocating quota inodes, we do not have a parent inode 740 * to attach to or associate with (i.e. pip == NULL) because they 741 * are not linked into the directory structure - they are attached 742 * directly to the superblock - and so have no parent. 743 */ 744 static int 745 xfs_ialloc( 746 xfs_trans_t *tp, 747 xfs_inode_t *pip, 748 umode_t mode, 749 xfs_nlink_t nlink, 750 dev_t rdev, 751 prid_t prid, 752 int okalloc, 753 xfs_buf_t **ialloc_context, 754 xfs_inode_t **ipp) 755 { 756 struct xfs_mount *mp = tp->t_mountp; 757 xfs_ino_t ino; 758 xfs_inode_t *ip; 759 uint flags; 760 int error; 761 struct timespec tv; 762 struct inode *inode; 763 764 /* 765 * Call the space management code to pick 766 * the on-disk inode to be allocated. 767 */ 768 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 769 ialloc_context, &ino); 770 if (error) 771 return error; 772 if (*ialloc_context || ino == NULLFSINO) { 773 *ipp = NULL; 774 return 0; 775 } 776 ASSERT(*ialloc_context == NULL); 777 778 /* 779 * Get the in-core inode with the lock held exclusively. 780 * This is because we're setting fields here we need 781 * to prevent others from looking at until we're done. 782 */ 783 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 784 XFS_ILOCK_EXCL, &ip); 785 if (error) 786 return error; 787 ASSERT(ip != NULL); 788 inode = VFS_I(ip); 789 790 /* 791 * We always convert v1 inodes to v2 now - we only support filesystems 792 * with >= v2 inode capability, so there is no reason for ever leaving 793 * an inode in v1 format. 794 */ 795 if (ip->i_d.di_version == 1) 796 ip->i_d.di_version = 2; 797 798 inode->i_mode = mode; 799 set_nlink(inode, nlink); 800 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 801 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 802 inode->i_rdev = rdev; 803 xfs_set_projid(ip, prid); 804 805 if (pip && XFS_INHERIT_GID(pip)) { 806 ip->i_d.di_gid = pip->i_d.di_gid; 807 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) 808 inode->i_mode |= S_ISGID; 809 } 810 811 /* 812 * If the group ID of the new file does not match the effective group 813 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 814 * (and only if the irix_sgid_inherit compatibility variable is set). 815 */ 816 if ((irix_sgid_inherit) && 817 (inode->i_mode & S_ISGID) && 818 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) 819 inode->i_mode &= ~S_ISGID; 820 821 ip->i_d.di_size = 0; 822 ip->i_d.di_nextents = 0; 823 ASSERT(ip->i_d.di_nblocks == 0); 824 825 tv = current_time(inode); 826 inode->i_mtime = tv; 827 inode->i_atime = tv; 828 inode->i_ctime = tv; 829 830 ip->i_d.di_extsize = 0; 831 ip->i_d.di_dmevmask = 0; 832 ip->i_d.di_dmstate = 0; 833 ip->i_d.di_flags = 0; 834 835 if (ip->i_d.di_version == 3) { 836 inode->i_version = 1; 837 ip->i_d.di_flags2 = 0; 838 ip->i_d.di_cowextsize = 0; 839 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; 840 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; 841 } 842 843 844 flags = XFS_ILOG_CORE; 845 switch (mode & S_IFMT) { 846 case S_IFIFO: 847 case S_IFCHR: 848 case S_IFBLK: 849 case S_IFSOCK: 850 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 851 ip->i_df.if_flags = 0; 852 flags |= XFS_ILOG_DEV; 853 break; 854 case S_IFREG: 855 case S_IFDIR: 856 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 857 uint di_flags = 0; 858 859 if (S_ISDIR(mode)) { 860 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 861 di_flags |= XFS_DIFLAG_RTINHERIT; 862 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 863 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 864 ip->i_d.di_extsize = pip->i_d.di_extsize; 865 } 866 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 867 di_flags |= XFS_DIFLAG_PROJINHERIT; 868 } else if (S_ISREG(mode)) { 869 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 870 di_flags |= XFS_DIFLAG_REALTIME; 871 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 872 di_flags |= XFS_DIFLAG_EXTSIZE; 873 ip->i_d.di_extsize = pip->i_d.di_extsize; 874 } 875 } 876 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 877 xfs_inherit_noatime) 878 di_flags |= XFS_DIFLAG_NOATIME; 879 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 880 xfs_inherit_nodump) 881 di_flags |= XFS_DIFLAG_NODUMP; 882 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 883 xfs_inherit_sync) 884 di_flags |= XFS_DIFLAG_SYNC; 885 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 886 xfs_inherit_nosymlinks) 887 di_flags |= XFS_DIFLAG_NOSYMLINKS; 888 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 889 xfs_inherit_nodefrag) 890 di_flags |= XFS_DIFLAG_NODEFRAG; 891 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 892 di_flags |= XFS_DIFLAG_FILESTREAM; 893 894 ip->i_d.di_flags |= di_flags; 895 } 896 if (pip && 897 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && 898 pip->i_d.di_version == 3 && 899 ip->i_d.di_version == 3) { 900 uint64_t di_flags2 = 0; 901 902 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { 903 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; 904 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; 905 } 906 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 907 di_flags2 |= XFS_DIFLAG2_DAX; 908 909 ip->i_d.di_flags2 |= di_flags2; 910 } 911 /* FALLTHROUGH */ 912 case S_IFLNK: 913 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 914 ip->i_df.if_flags = XFS_IFEXTENTS; 915 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 916 ip->i_df.if_u1.if_root = NULL; 917 break; 918 default: 919 ASSERT(0); 920 } 921 /* 922 * Attribute fork settings for new inode. 923 */ 924 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 925 ip->i_d.di_anextents = 0; 926 927 /* 928 * Log the new values stuffed into the inode. 929 */ 930 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 931 xfs_trans_log_inode(tp, ip, flags); 932 933 /* now that we have an i_mode we can setup the inode structure */ 934 xfs_setup_inode(ip); 935 936 *ipp = ip; 937 return 0; 938 } 939 940 /* 941 * Allocates a new inode from disk and return a pointer to the 942 * incore copy. This routine will internally commit the current 943 * transaction and allocate a new one if the Space Manager needed 944 * to do an allocation to replenish the inode free-list. 945 * 946 * This routine is designed to be called from xfs_create and 947 * xfs_create_dir. 948 * 949 */ 950 int 951 xfs_dir_ialloc( 952 xfs_trans_t **tpp, /* input: current transaction; 953 output: may be a new transaction. */ 954 xfs_inode_t *dp, /* directory within whose allocate 955 the inode. */ 956 umode_t mode, 957 xfs_nlink_t nlink, 958 dev_t rdev, 959 prid_t prid, /* project id */ 960 int okalloc, /* ok to allocate new space */ 961 xfs_inode_t **ipp, /* pointer to inode; it will be 962 locked. */ 963 int *committed) 964 965 { 966 xfs_trans_t *tp; 967 xfs_inode_t *ip; 968 xfs_buf_t *ialloc_context = NULL; 969 int code; 970 void *dqinfo; 971 uint tflags; 972 973 tp = *tpp; 974 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 975 976 /* 977 * xfs_ialloc will return a pointer to an incore inode if 978 * the Space Manager has an available inode on the free 979 * list. Otherwise, it will do an allocation and replenish 980 * the freelist. Since we can only do one allocation per 981 * transaction without deadlocks, we will need to commit the 982 * current transaction and start a new one. We will then 983 * need to call xfs_ialloc again to get the inode. 984 * 985 * If xfs_ialloc did an allocation to replenish the freelist, 986 * it returns the bp containing the head of the freelist as 987 * ialloc_context. We will hold a lock on it across the 988 * transaction commit so that no other process can steal 989 * the inode(s) that we've just allocated. 990 */ 991 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 992 &ialloc_context, &ip); 993 994 /* 995 * Return an error if we were unable to allocate a new inode. 996 * This should only happen if we run out of space on disk or 997 * encounter a disk error. 998 */ 999 if (code) { 1000 *ipp = NULL; 1001 return code; 1002 } 1003 if (!ialloc_context && !ip) { 1004 *ipp = NULL; 1005 return -ENOSPC; 1006 } 1007 1008 /* 1009 * If the AGI buffer is non-NULL, then we were unable to get an 1010 * inode in one operation. We need to commit the current 1011 * transaction and call xfs_ialloc() again. It is guaranteed 1012 * to succeed the second time. 1013 */ 1014 if (ialloc_context) { 1015 /* 1016 * Normally, xfs_trans_commit releases all the locks. 1017 * We call bhold to hang on to the ialloc_context across 1018 * the commit. Holding this buffer prevents any other 1019 * processes from doing any allocations in this 1020 * allocation group. 1021 */ 1022 xfs_trans_bhold(tp, ialloc_context); 1023 1024 /* 1025 * We want the quota changes to be associated with the next 1026 * transaction, NOT this one. So, detach the dqinfo from this 1027 * and attach it to the next transaction. 1028 */ 1029 dqinfo = NULL; 1030 tflags = 0; 1031 if (tp->t_dqinfo) { 1032 dqinfo = (void *)tp->t_dqinfo; 1033 tp->t_dqinfo = NULL; 1034 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1035 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1036 } 1037 1038 code = xfs_trans_roll(&tp); 1039 if (committed != NULL) 1040 *committed = 1; 1041 1042 /* 1043 * Re-attach the quota info that we detached from prev trx. 1044 */ 1045 if (dqinfo) { 1046 tp->t_dqinfo = dqinfo; 1047 tp->t_flags |= tflags; 1048 } 1049 1050 if (code) { 1051 xfs_buf_relse(ialloc_context); 1052 *tpp = tp; 1053 *ipp = NULL; 1054 return code; 1055 } 1056 xfs_trans_bjoin(tp, ialloc_context); 1057 1058 /* 1059 * Call ialloc again. Since we've locked out all 1060 * other allocations in this allocation group, 1061 * this call should always succeed. 1062 */ 1063 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1064 okalloc, &ialloc_context, &ip); 1065 1066 /* 1067 * If we get an error at this point, return to the caller 1068 * so that the current transaction can be aborted. 1069 */ 1070 if (code) { 1071 *tpp = tp; 1072 *ipp = NULL; 1073 return code; 1074 } 1075 ASSERT(!ialloc_context && ip); 1076 1077 } else { 1078 if (committed != NULL) 1079 *committed = 0; 1080 } 1081 1082 *ipp = ip; 1083 *tpp = tp; 1084 1085 return 0; 1086 } 1087 1088 /* 1089 * Decrement the link count on an inode & log the change. If this causes the 1090 * link count to go to zero, move the inode to AGI unlinked list so that it can 1091 * be freed when the last active reference goes away via xfs_inactive(). 1092 */ 1093 static int /* error */ 1094 xfs_droplink( 1095 xfs_trans_t *tp, 1096 xfs_inode_t *ip) 1097 { 1098 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1099 1100 drop_nlink(VFS_I(ip)); 1101 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1102 1103 if (VFS_I(ip)->i_nlink) 1104 return 0; 1105 1106 return xfs_iunlink(tp, ip); 1107 } 1108 1109 /* 1110 * Increment the link count on an inode & log the change. 1111 */ 1112 static int 1113 xfs_bumplink( 1114 xfs_trans_t *tp, 1115 xfs_inode_t *ip) 1116 { 1117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1118 1119 ASSERT(ip->i_d.di_version > 1); 1120 inc_nlink(VFS_I(ip)); 1121 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1122 return 0; 1123 } 1124 1125 int 1126 xfs_create( 1127 xfs_inode_t *dp, 1128 struct xfs_name *name, 1129 umode_t mode, 1130 dev_t rdev, 1131 xfs_inode_t **ipp) 1132 { 1133 int is_dir = S_ISDIR(mode); 1134 struct xfs_mount *mp = dp->i_mount; 1135 struct xfs_inode *ip = NULL; 1136 struct xfs_trans *tp = NULL; 1137 int error; 1138 struct xfs_defer_ops dfops; 1139 xfs_fsblock_t first_block; 1140 bool unlock_dp_on_error = false; 1141 prid_t prid; 1142 struct xfs_dquot *udqp = NULL; 1143 struct xfs_dquot *gdqp = NULL; 1144 struct xfs_dquot *pdqp = NULL; 1145 struct xfs_trans_res *tres; 1146 uint resblks; 1147 1148 trace_xfs_create(dp, name); 1149 1150 if (XFS_FORCED_SHUTDOWN(mp)) 1151 return -EIO; 1152 1153 prid = xfs_get_initial_prid(dp); 1154 1155 /* 1156 * Make sure that we have allocated dquot(s) on disk. 1157 */ 1158 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1159 xfs_kgid_to_gid(current_fsgid()), prid, 1160 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1161 &udqp, &gdqp, &pdqp); 1162 if (error) 1163 return error; 1164 1165 if (is_dir) { 1166 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1167 tres = &M_RES(mp)->tr_mkdir; 1168 } else { 1169 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1170 tres = &M_RES(mp)->tr_create; 1171 } 1172 1173 /* 1174 * Initially assume that the file does not exist and 1175 * reserve the resources for that case. If that is not 1176 * the case we'll drop the one we have and get a more 1177 * appropriate transaction later. 1178 */ 1179 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1180 if (error == -ENOSPC) { 1181 /* flush outstanding delalloc blocks and retry */ 1182 xfs_flush_inodes(mp); 1183 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1184 } 1185 if (error == -ENOSPC) { 1186 /* No space at all so try a "no-allocation" reservation */ 1187 resblks = 0; 1188 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1189 } 1190 if (error) 1191 goto out_release_inode; 1192 1193 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1194 unlock_dp_on_error = true; 1195 1196 xfs_defer_init(&dfops, &first_block); 1197 1198 /* 1199 * Reserve disk quota and the inode. 1200 */ 1201 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1202 pdqp, resblks, 1, 0); 1203 if (error) 1204 goto out_trans_cancel; 1205 1206 if (!resblks) { 1207 error = xfs_dir_canenter(tp, dp, name); 1208 if (error) 1209 goto out_trans_cancel; 1210 } 1211 1212 /* 1213 * A newly created regular or special file just has one directory 1214 * entry pointing to them, but a directory also the "." entry 1215 * pointing to itself. 1216 */ 1217 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1218 prid, resblks > 0, &ip, NULL); 1219 if (error) 1220 goto out_trans_cancel; 1221 1222 /* 1223 * Now we join the directory inode to the transaction. We do not do it 1224 * earlier because xfs_dir_ialloc might commit the previous transaction 1225 * (and release all the locks). An error from here on will result in 1226 * the transaction cancel unlocking dp so don't do it explicitly in the 1227 * error path. 1228 */ 1229 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1230 unlock_dp_on_error = false; 1231 1232 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1233 &first_block, &dfops, resblks ? 1234 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1235 if (error) { 1236 ASSERT(error != -ENOSPC); 1237 goto out_trans_cancel; 1238 } 1239 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1240 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1241 1242 if (is_dir) { 1243 error = xfs_dir_init(tp, ip, dp); 1244 if (error) 1245 goto out_bmap_cancel; 1246 1247 error = xfs_bumplink(tp, dp); 1248 if (error) 1249 goto out_bmap_cancel; 1250 } 1251 1252 /* 1253 * If this is a synchronous mount, make sure that the 1254 * create transaction goes to disk before returning to 1255 * the user. 1256 */ 1257 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1258 xfs_trans_set_sync(tp); 1259 1260 /* 1261 * Attach the dquot(s) to the inodes and modify them incore. 1262 * These ids of the inode couldn't have changed since the new 1263 * inode has been locked ever since it was created. 1264 */ 1265 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1266 1267 error = xfs_defer_finish(&tp, &dfops); 1268 if (error) 1269 goto out_bmap_cancel; 1270 1271 error = xfs_trans_commit(tp); 1272 if (error) 1273 goto out_release_inode; 1274 1275 xfs_qm_dqrele(udqp); 1276 xfs_qm_dqrele(gdqp); 1277 xfs_qm_dqrele(pdqp); 1278 1279 *ipp = ip; 1280 return 0; 1281 1282 out_bmap_cancel: 1283 xfs_defer_cancel(&dfops); 1284 out_trans_cancel: 1285 xfs_trans_cancel(tp); 1286 out_release_inode: 1287 /* 1288 * Wait until after the current transaction is aborted to finish the 1289 * setup of the inode and release the inode. This prevents recursive 1290 * transactions and deadlocks from xfs_inactive. 1291 */ 1292 if (ip) { 1293 xfs_finish_inode_setup(ip); 1294 IRELE(ip); 1295 } 1296 1297 xfs_qm_dqrele(udqp); 1298 xfs_qm_dqrele(gdqp); 1299 xfs_qm_dqrele(pdqp); 1300 1301 if (unlock_dp_on_error) 1302 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1303 return error; 1304 } 1305 1306 int 1307 xfs_create_tmpfile( 1308 struct xfs_inode *dp, 1309 struct dentry *dentry, 1310 umode_t mode, 1311 struct xfs_inode **ipp) 1312 { 1313 struct xfs_mount *mp = dp->i_mount; 1314 struct xfs_inode *ip = NULL; 1315 struct xfs_trans *tp = NULL; 1316 int error; 1317 prid_t prid; 1318 struct xfs_dquot *udqp = NULL; 1319 struct xfs_dquot *gdqp = NULL; 1320 struct xfs_dquot *pdqp = NULL; 1321 struct xfs_trans_res *tres; 1322 uint resblks; 1323 1324 if (XFS_FORCED_SHUTDOWN(mp)) 1325 return -EIO; 1326 1327 prid = xfs_get_initial_prid(dp); 1328 1329 /* 1330 * Make sure that we have allocated dquot(s) on disk. 1331 */ 1332 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1333 xfs_kgid_to_gid(current_fsgid()), prid, 1334 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1335 &udqp, &gdqp, &pdqp); 1336 if (error) 1337 return error; 1338 1339 resblks = XFS_IALLOC_SPACE_RES(mp); 1340 tres = &M_RES(mp)->tr_create_tmpfile; 1341 1342 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1343 if (error == -ENOSPC) { 1344 /* No space at all so try a "no-allocation" reservation */ 1345 resblks = 0; 1346 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1347 } 1348 if (error) 1349 goto out_release_inode; 1350 1351 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1352 pdqp, resblks, 1, 0); 1353 if (error) 1354 goto out_trans_cancel; 1355 1356 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1357 prid, resblks > 0, &ip, NULL); 1358 if (error) 1359 goto out_trans_cancel; 1360 1361 if (mp->m_flags & XFS_MOUNT_WSYNC) 1362 xfs_trans_set_sync(tp); 1363 1364 /* 1365 * Attach the dquot(s) to the inodes and modify them incore. 1366 * These ids of the inode couldn't have changed since the new 1367 * inode has been locked ever since it was created. 1368 */ 1369 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1370 1371 error = xfs_iunlink(tp, ip); 1372 if (error) 1373 goto out_trans_cancel; 1374 1375 error = xfs_trans_commit(tp); 1376 if (error) 1377 goto out_release_inode; 1378 1379 xfs_qm_dqrele(udqp); 1380 xfs_qm_dqrele(gdqp); 1381 xfs_qm_dqrele(pdqp); 1382 1383 *ipp = ip; 1384 return 0; 1385 1386 out_trans_cancel: 1387 xfs_trans_cancel(tp); 1388 out_release_inode: 1389 /* 1390 * Wait until after the current transaction is aborted to finish the 1391 * setup of the inode and release the inode. This prevents recursive 1392 * transactions and deadlocks from xfs_inactive. 1393 */ 1394 if (ip) { 1395 xfs_finish_inode_setup(ip); 1396 IRELE(ip); 1397 } 1398 1399 xfs_qm_dqrele(udqp); 1400 xfs_qm_dqrele(gdqp); 1401 xfs_qm_dqrele(pdqp); 1402 1403 return error; 1404 } 1405 1406 int 1407 xfs_link( 1408 xfs_inode_t *tdp, 1409 xfs_inode_t *sip, 1410 struct xfs_name *target_name) 1411 { 1412 xfs_mount_t *mp = tdp->i_mount; 1413 xfs_trans_t *tp; 1414 int error; 1415 struct xfs_defer_ops dfops; 1416 xfs_fsblock_t first_block; 1417 int resblks; 1418 1419 trace_xfs_link(tdp, target_name); 1420 1421 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1422 1423 if (XFS_FORCED_SHUTDOWN(mp)) 1424 return -EIO; 1425 1426 error = xfs_qm_dqattach(sip, 0); 1427 if (error) 1428 goto std_return; 1429 1430 error = xfs_qm_dqattach(tdp, 0); 1431 if (error) 1432 goto std_return; 1433 1434 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1435 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1436 if (error == -ENOSPC) { 1437 resblks = 0; 1438 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1439 } 1440 if (error) 1441 goto std_return; 1442 1443 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1444 1445 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1446 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1447 1448 /* 1449 * If we are using project inheritance, we only allow hard link 1450 * creation in our tree when the project IDs are the same; else 1451 * the tree quota mechanism could be circumvented. 1452 */ 1453 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1454 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1455 error = -EXDEV; 1456 goto error_return; 1457 } 1458 1459 if (!resblks) { 1460 error = xfs_dir_canenter(tp, tdp, target_name); 1461 if (error) 1462 goto error_return; 1463 } 1464 1465 xfs_defer_init(&dfops, &first_block); 1466 1467 /* 1468 * Handle initial link state of O_TMPFILE inode 1469 */ 1470 if (VFS_I(sip)->i_nlink == 0) { 1471 error = xfs_iunlink_remove(tp, sip); 1472 if (error) 1473 goto error_return; 1474 } 1475 1476 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1477 &first_block, &dfops, resblks); 1478 if (error) 1479 goto error_return; 1480 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1481 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1482 1483 error = xfs_bumplink(tp, sip); 1484 if (error) 1485 goto error_return; 1486 1487 /* 1488 * If this is a synchronous mount, make sure that the 1489 * link transaction goes to disk before returning to 1490 * the user. 1491 */ 1492 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1493 xfs_trans_set_sync(tp); 1494 1495 error = xfs_defer_finish(&tp, &dfops); 1496 if (error) { 1497 xfs_defer_cancel(&dfops); 1498 goto error_return; 1499 } 1500 1501 return xfs_trans_commit(tp); 1502 1503 error_return: 1504 xfs_trans_cancel(tp); 1505 std_return: 1506 return error; 1507 } 1508 1509 /* 1510 * Free up the underlying blocks past new_size. The new size must be smaller 1511 * than the current size. This routine can be used both for the attribute and 1512 * data fork, and does not modify the inode size, which is left to the caller. 1513 * 1514 * The transaction passed to this routine must have made a permanent log 1515 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1516 * given transaction and start new ones, so make sure everything involved in 1517 * the transaction is tidy before calling here. Some transaction will be 1518 * returned to the caller to be committed. The incoming transaction must 1519 * already include the inode, and both inode locks must be held exclusively. 1520 * The inode must also be "held" within the transaction. On return the inode 1521 * will be "held" within the returned transaction. This routine does NOT 1522 * require any disk space to be reserved for it within the transaction. 1523 * 1524 * If we get an error, we must return with the inode locked and linked into the 1525 * current transaction. This keeps things simple for the higher level code, 1526 * because it always knows that the inode is locked and held in the transaction 1527 * that returns to it whether errors occur or not. We don't mark the inode 1528 * dirty on error so that transactions can be easily aborted if possible. 1529 */ 1530 int 1531 xfs_itruncate_extents( 1532 struct xfs_trans **tpp, 1533 struct xfs_inode *ip, 1534 int whichfork, 1535 xfs_fsize_t new_size) 1536 { 1537 struct xfs_mount *mp = ip->i_mount; 1538 struct xfs_trans *tp = *tpp; 1539 struct xfs_defer_ops dfops; 1540 xfs_fsblock_t first_block; 1541 xfs_fileoff_t first_unmap_block; 1542 xfs_fileoff_t last_block; 1543 xfs_filblks_t unmap_len; 1544 int error = 0; 1545 int done = 0; 1546 1547 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1548 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1549 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1550 ASSERT(new_size <= XFS_ISIZE(ip)); 1551 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1552 ASSERT(ip->i_itemp != NULL); 1553 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1554 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1555 1556 trace_xfs_itruncate_extents_start(ip, new_size); 1557 1558 /* 1559 * Since it is possible for space to become allocated beyond 1560 * the end of the file (in a crash where the space is allocated 1561 * but the inode size is not yet updated), simply remove any 1562 * blocks which show up between the new EOF and the maximum 1563 * possible file size. If the first block to be removed is 1564 * beyond the maximum file size (ie it is the same as last_block), 1565 * then there is nothing to do. 1566 */ 1567 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1568 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1569 if (first_unmap_block == last_block) 1570 return 0; 1571 1572 ASSERT(first_unmap_block < last_block); 1573 unmap_len = last_block - first_unmap_block + 1; 1574 while (!done) { 1575 xfs_defer_init(&dfops, &first_block); 1576 error = xfs_bunmapi(tp, ip, 1577 first_unmap_block, unmap_len, 1578 xfs_bmapi_aflag(whichfork), 1579 XFS_ITRUNC_MAX_EXTENTS, 1580 &first_block, &dfops, 1581 &done); 1582 if (error) 1583 goto out_bmap_cancel; 1584 1585 /* 1586 * Duplicate the transaction that has the permanent 1587 * reservation and commit the old transaction. 1588 */ 1589 xfs_defer_ijoin(&dfops, ip); 1590 error = xfs_defer_finish(&tp, &dfops); 1591 if (error) 1592 goto out_bmap_cancel; 1593 1594 error = xfs_trans_roll_inode(&tp, ip); 1595 if (error) 1596 goto out; 1597 } 1598 1599 /* Remove all pending CoW reservations. */ 1600 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, 1601 last_block, true); 1602 if (error) 1603 goto out; 1604 1605 /* 1606 * Clear the reflink flag if there are no data fork blocks and 1607 * there are no extents staged in the cow fork. 1608 */ 1609 if (xfs_is_reflink_inode(ip) && ip->i_cnextents == 0) { 1610 if (ip->i_d.di_nblocks == 0) 1611 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 1612 xfs_inode_clear_cowblocks_tag(ip); 1613 } 1614 1615 /* 1616 * Always re-log the inode so that our permanent transaction can keep 1617 * on rolling it forward in the log. 1618 */ 1619 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1620 1621 trace_xfs_itruncate_extents_end(ip, new_size); 1622 1623 out: 1624 *tpp = tp; 1625 return error; 1626 out_bmap_cancel: 1627 /* 1628 * If the bunmapi call encounters an error, return to the caller where 1629 * the transaction can be properly aborted. We just need to make sure 1630 * we're not holding any resources that we were not when we came in. 1631 */ 1632 xfs_defer_cancel(&dfops); 1633 goto out; 1634 } 1635 1636 int 1637 xfs_release( 1638 xfs_inode_t *ip) 1639 { 1640 xfs_mount_t *mp = ip->i_mount; 1641 int error; 1642 1643 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1644 return 0; 1645 1646 /* If this is a read-only mount, don't do this (would generate I/O) */ 1647 if (mp->m_flags & XFS_MOUNT_RDONLY) 1648 return 0; 1649 1650 if (!XFS_FORCED_SHUTDOWN(mp)) { 1651 int truncated; 1652 1653 /* 1654 * If we previously truncated this file and removed old data 1655 * in the process, we want to initiate "early" writeout on 1656 * the last close. This is an attempt to combat the notorious 1657 * NULL files problem which is particularly noticeable from a 1658 * truncate down, buffered (re-)write (delalloc), followed by 1659 * a crash. What we are effectively doing here is 1660 * significantly reducing the time window where we'd otherwise 1661 * be exposed to that problem. 1662 */ 1663 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1664 if (truncated) { 1665 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1666 if (ip->i_delayed_blks > 0) { 1667 error = filemap_flush(VFS_I(ip)->i_mapping); 1668 if (error) 1669 return error; 1670 } 1671 } 1672 } 1673 1674 if (VFS_I(ip)->i_nlink == 0) 1675 return 0; 1676 1677 if (xfs_can_free_eofblocks(ip, false)) { 1678 1679 /* 1680 * Check if the inode is being opened, written and closed 1681 * frequently and we have delayed allocation blocks outstanding 1682 * (e.g. streaming writes from the NFS server), truncating the 1683 * blocks past EOF will cause fragmentation to occur. 1684 * 1685 * In this case don't do the truncation, but we have to be 1686 * careful how we detect this case. Blocks beyond EOF show up as 1687 * i_delayed_blks even when the inode is clean, so we need to 1688 * truncate them away first before checking for a dirty release. 1689 * Hence on the first dirty close we will still remove the 1690 * speculative allocation, but after that we will leave it in 1691 * place. 1692 */ 1693 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1694 return 0; 1695 /* 1696 * If we can't get the iolock just skip truncating the blocks 1697 * past EOF because we could deadlock with the mmap_sem 1698 * otherwise. We'll get another chance to drop them once the 1699 * last reference to the inode is dropped, so we'll never leak 1700 * blocks permanently. 1701 */ 1702 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { 1703 error = xfs_free_eofblocks(ip); 1704 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 1705 if (error) 1706 return error; 1707 } 1708 1709 /* delalloc blocks after truncation means it really is dirty */ 1710 if (ip->i_delayed_blks) 1711 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1712 } 1713 return 0; 1714 } 1715 1716 /* 1717 * xfs_inactive_truncate 1718 * 1719 * Called to perform a truncate when an inode becomes unlinked. 1720 */ 1721 STATIC int 1722 xfs_inactive_truncate( 1723 struct xfs_inode *ip) 1724 { 1725 struct xfs_mount *mp = ip->i_mount; 1726 struct xfs_trans *tp; 1727 int error; 1728 1729 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1730 if (error) { 1731 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1732 return error; 1733 } 1734 1735 xfs_ilock(ip, XFS_ILOCK_EXCL); 1736 xfs_trans_ijoin(tp, ip, 0); 1737 1738 /* 1739 * Log the inode size first to prevent stale data exposure in the event 1740 * of a system crash before the truncate completes. See the related 1741 * comment in xfs_vn_setattr_size() for details. 1742 */ 1743 ip->i_d.di_size = 0; 1744 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1745 1746 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1747 if (error) 1748 goto error_trans_cancel; 1749 1750 ASSERT(ip->i_d.di_nextents == 0); 1751 1752 error = xfs_trans_commit(tp); 1753 if (error) 1754 goto error_unlock; 1755 1756 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1757 return 0; 1758 1759 error_trans_cancel: 1760 xfs_trans_cancel(tp); 1761 error_unlock: 1762 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1763 return error; 1764 } 1765 1766 /* 1767 * xfs_inactive_ifree() 1768 * 1769 * Perform the inode free when an inode is unlinked. 1770 */ 1771 STATIC int 1772 xfs_inactive_ifree( 1773 struct xfs_inode *ip) 1774 { 1775 struct xfs_defer_ops dfops; 1776 xfs_fsblock_t first_block; 1777 struct xfs_mount *mp = ip->i_mount; 1778 struct xfs_trans *tp; 1779 int error; 1780 1781 /* 1782 * We try to use a per-AG reservation for any block needed by the finobt 1783 * tree, but as the finobt feature predates the per-AG reservation 1784 * support a degraded file system might not have enough space for the 1785 * reservation at mount time. In that case try to dip into the reserved 1786 * pool and pray. 1787 * 1788 * Send a warning if the reservation does happen to fail, as the inode 1789 * now remains allocated and sits on the unlinked list until the fs is 1790 * repaired. 1791 */ 1792 if (unlikely(mp->m_inotbt_nores)) { 1793 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1794 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 1795 &tp); 1796 } else { 1797 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); 1798 } 1799 if (error) { 1800 if (error == -ENOSPC) { 1801 xfs_warn_ratelimited(mp, 1802 "Failed to remove inode(s) from unlinked list. " 1803 "Please free space, unmount and run xfs_repair."); 1804 } else { 1805 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1806 } 1807 return error; 1808 } 1809 1810 xfs_ilock(ip, XFS_ILOCK_EXCL); 1811 xfs_trans_ijoin(tp, ip, 0); 1812 1813 xfs_defer_init(&dfops, &first_block); 1814 error = xfs_ifree(tp, ip, &dfops); 1815 if (error) { 1816 /* 1817 * If we fail to free the inode, shut down. The cancel 1818 * might do that, we need to make sure. Otherwise the 1819 * inode might be lost for a long time or forever. 1820 */ 1821 if (!XFS_FORCED_SHUTDOWN(mp)) { 1822 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1823 __func__, error); 1824 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1825 } 1826 xfs_trans_cancel(tp); 1827 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1828 return error; 1829 } 1830 1831 /* 1832 * Credit the quota account(s). The inode is gone. 1833 */ 1834 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1835 1836 /* 1837 * Just ignore errors at this point. There is nothing we can do except 1838 * to try to keep going. Make sure it's not a silent error. 1839 */ 1840 error = xfs_defer_finish(&tp, &dfops); 1841 if (error) { 1842 xfs_notice(mp, "%s: xfs_defer_finish returned error %d", 1843 __func__, error); 1844 xfs_defer_cancel(&dfops); 1845 } 1846 error = xfs_trans_commit(tp); 1847 if (error) 1848 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1849 __func__, error); 1850 1851 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1852 return 0; 1853 } 1854 1855 /* 1856 * xfs_inactive 1857 * 1858 * This is called when the vnode reference count for the vnode 1859 * goes to zero. If the file has been unlinked, then it must 1860 * now be truncated. Also, we clear all of the read-ahead state 1861 * kept for the inode here since the file is now closed. 1862 */ 1863 void 1864 xfs_inactive( 1865 xfs_inode_t *ip) 1866 { 1867 struct xfs_mount *mp; 1868 int error; 1869 int truncate = 0; 1870 1871 /* 1872 * If the inode is already free, then there can be nothing 1873 * to clean up here. 1874 */ 1875 if (VFS_I(ip)->i_mode == 0) { 1876 ASSERT(ip->i_df.if_real_bytes == 0); 1877 ASSERT(ip->i_df.if_broot_bytes == 0); 1878 return; 1879 } 1880 1881 mp = ip->i_mount; 1882 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); 1883 1884 /* If this is a read-only mount, don't do this (would generate I/O) */ 1885 if (mp->m_flags & XFS_MOUNT_RDONLY) 1886 return; 1887 1888 if (VFS_I(ip)->i_nlink != 0) { 1889 /* 1890 * force is true because we are evicting an inode from the 1891 * cache. Post-eof blocks must be freed, lest we end up with 1892 * broken free space accounting. 1893 * 1894 * Note: don't bother with iolock here since lockdep complains 1895 * about acquiring it in reclaim context. We have the only 1896 * reference to the inode at this point anyways. 1897 */ 1898 if (xfs_can_free_eofblocks(ip, true)) 1899 xfs_free_eofblocks(ip); 1900 1901 return; 1902 } 1903 1904 if (S_ISREG(VFS_I(ip)->i_mode) && 1905 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1906 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1907 truncate = 1; 1908 1909 error = xfs_qm_dqattach(ip, 0); 1910 if (error) 1911 return; 1912 1913 if (S_ISLNK(VFS_I(ip)->i_mode)) 1914 error = xfs_inactive_symlink(ip); 1915 else if (truncate) 1916 error = xfs_inactive_truncate(ip); 1917 if (error) 1918 return; 1919 1920 /* 1921 * If there are attributes associated with the file then blow them away 1922 * now. The code calls a routine that recursively deconstructs the 1923 * attribute fork. If also blows away the in-core attribute fork. 1924 */ 1925 if (XFS_IFORK_Q(ip)) { 1926 error = xfs_attr_inactive(ip); 1927 if (error) 1928 return; 1929 } 1930 1931 ASSERT(!ip->i_afp); 1932 ASSERT(ip->i_d.di_anextents == 0); 1933 ASSERT(ip->i_d.di_forkoff == 0); 1934 1935 /* 1936 * Free the inode. 1937 */ 1938 error = xfs_inactive_ifree(ip); 1939 if (error) 1940 return; 1941 1942 /* 1943 * Release the dquots held by inode, if any. 1944 */ 1945 xfs_qm_dqdetach(ip); 1946 } 1947 1948 /* 1949 * This is called when the inode's link count goes to 0 or we are creating a 1950 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1951 * set to true as the link count is dropped to zero by the VFS after we've 1952 * created the file successfully, so we have to add it to the unlinked list 1953 * while the link count is non-zero. 1954 * 1955 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1956 * list when the inode is freed. 1957 */ 1958 STATIC int 1959 xfs_iunlink( 1960 struct xfs_trans *tp, 1961 struct xfs_inode *ip) 1962 { 1963 xfs_mount_t *mp = tp->t_mountp; 1964 xfs_agi_t *agi; 1965 xfs_dinode_t *dip; 1966 xfs_buf_t *agibp; 1967 xfs_buf_t *ibp; 1968 xfs_agino_t agino; 1969 short bucket_index; 1970 int offset; 1971 int error; 1972 1973 ASSERT(VFS_I(ip)->i_mode != 0); 1974 1975 /* 1976 * Get the agi buffer first. It ensures lock ordering 1977 * on the list. 1978 */ 1979 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1980 if (error) 1981 return error; 1982 agi = XFS_BUF_TO_AGI(agibp); 1983 1984 /* 1985 * Get the index into the agi hash table for the 1986 * list this inode will go on. 1987 */ 1988 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1989 ASSERT(agino != 0); 1990 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1991 ASSERT(agi->agi_unlinked[bucket_index]); 1992 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1993 1994 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1995 /* 1996 * There is already another inode in the bucket we need 1997 * to add ourselves to. Add us at the front of the list. 1998 * Here we put the head pointer into our next pointer, 1999 * and then we fall through to point the head at us. 2000 */ 2001 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2002 0, 0); 2003 if (error) 2004 return error; 2005 2006 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 2007 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 2008 offset = ip->i_imap.im_boffset + 2009 offsetof(xfs_dinode_t, di_next_unlinked); 2010 2011 /* need to recalc the inode CRC if appropriate */ 2012 xfs_dinode_calc_crc(mp, dip); 2013 2014 xfs_trans_inode_buf(tp, ibp); 2015 xfs_trans_log_buf(tp, ibp, offset, 2016 (offset + sizeof(xfs_agino_t) - 1)); 2017 xfs_inobp_check(mp, ibp); 2018 } 2019 2020 /* 2021 * Point the bucket head pointer at the inode being inserted. 2022 */ 2023 ASSERT(agino != 0); 2024 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 2025 offset = offsetof(xfs_agi_t, agi_unlinked) + 2026 (sizeof(xfs_agino_t) * bucket_index); 2027 xfs_trans_log_buf(tp, agibp, offset, 2028 (offset + sizeof(xfs_agino_t) - 1)); 2029 return 0; 2030 } 2031 2032 /* 2033 * Pull the on-disk inode from the AGI unlinked list. 2034 */ 2035 STATIC int 2036 xfs_iunlink_remove( 2037 xfs_trans_t *tp, 2038 xfs_inode_t *ip) 2039 { 2040 xfs_ino_t next_ino; 2041 xfs_mount_t *mp; 2042 xfs_agi_t *agi; 2043 xfs_dinode_t *dip; 2044 xfs_buf_t *agibp; 2045 xfs_buf_t *ibp; 2046 xfs_agnumber_t agno; 2047 xfs_agino_t agino; 2048 xfs_agino_t next_agino; 2049 xfs_buf_t *last_ibp; 2050 xfs_dinode_t *last_dip = NULL; 2051 short bucket_index; 2052 int offset, last_offset = 0; 2053 int error; 2054 2055 mp = tp->t_mountp; 2056 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2057 2058 /* 2059 * Get the agi buffer first. It ensures lock ordering 2060 * on the list. 2061 */ 2062 error = xfs_read_agi(mp, tp, agno, &agibp); 2063 if (error) 2064 return error; 2065 2066 agi = XFS_BUF_TO_AGI(agibp); 2067 2068 /* 2069 * Get the index into the agi hash table for the 2070 * list this inode will go on. 2071 */ 2072 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2073 ASSERT(agino != 0); 2074 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2075 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2076 ASSERT(agi->agi_unlinked[bucket_index]); 2077 2078 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2079 /* 2080 * We're at the head of the list. Get the inode's on-disk 2081 * buffer to see if there is anyone after us on the list. 2082 * Only modify our next pointer if it is not already NULLAGINO. 2083 * This saves us the overhead of dealing with the buffer when 2084 * there is no need to change it. 2085 */ 2086 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2087 0, 0); 2088 if (error) { 2089 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2090 __func__, error); 2091 return error; 2092 } 2093 next_agino = be32_to_cpu(dip->di_next_unlinked); 2094 ASSERT(next_agino != 0); 2095 if (next_agino != NULLAGINO) { 2096 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2097 offset = ip->i_imap.im_boffset + 2098 offsetof(xfs_dinode_t, di_next_unlinked); 2099 2100 /* need to recalc the inode CRC if appropriate */ 2101 xfs_dinode_calc_crc(mp, dip); 2102 2103 xfs_trans_inode_buf(tp, ibp); 2104 xfs_trans_log_buf(tp, ibp, offset, 2105 (offset + sizeof(xfs_agino_t) - 1)); 2106 xfs_inobp_check(mp, ibp); 2107 } else { 2108 xfs_trans_brelse(tp, ibp); 2109 } 2110 /* 2111 * Point the bucket head pointer at the next inode. 2112 */ 2113 ASSERT(next_agino != 0); 2114 ASSERT(next_agino != agino); 2115 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2116 offset = offsetof(xfs_agi_t, agi_unlinked) + 2117 (sizeof(xfs_agino_t) * bucket_index); 2118 xfs_trans_log_buf(tp, agibp, offset, 2119 (offset + sizeof(xfs_agino_t) - 1)); 2120 } else { 2121 /* 2122 * We need to search the list for the inode being freed. 2123 */ 2124 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2125 last_ibp = NULL; 2126 while (next_agino != agino) { 2127 struct xfs_imap imap; 2128 2129 if (last_ibp) 2130 xfs_trans_brelse(tp, last_ibp); 2131 2132 imap.im_blkno = 0; 2133 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2134 2135 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2136 if (error) { 2137 xfs_warn(mp, 2138 "%s: xfs_imap returned error %d.", 2139 __func__, error); 2140 return error; 2141 } 2142 2143 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2144 &last_ibp, 0, 0); 2145 if (error) { 2146 xfs_warn(mp, 2147 "%s: xfs_imap_to_bp returned error %d.", 2148 __func__, error); 2149 return error; 2150 } 2151 2152 last_offset = imap.im_boffset; 2153 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2154 ASSERT(next_agino != NULLAGINO); 2155 ASSERT(next_agino != 0); 2156 } 2157 2158 /* 2159 * Now last_ibp points to the buffer previous to us on the 2160 * unlinked list. Pull us from the list. 2161 */ 2162 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2163 0, 0); 2164 if (error) { 2165 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2166 __func__, error); 2167 return error; 2168 } 2169 next_agino = be32_to_cpu(dip->di_next_unlinked); 2170 ASSERT(next_agino != 0); 2171 ASSERT(next_agino != agino); 2172 if (next_agino != NULLAGINO) { 2173 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2174 offset = ip->i_imap.im_boffset + 2175 offsetof(xfs_dinode_t, di_next_unlinked); 2176 2177 /* need to recalc the inode CRC if appropriate */ 2178 xfs_dinode_calc_crc(mp, dip); 2179 2180 xfs_trans_inode_buf(tp, ibp); 2181 xfs_trans_log_buf(tp, ibp, offset, 2182 (offset + sizeof(xfs_agino_t) - 1)); 2183 xfs_inobp_check(mp, ibp); 2184 } else { 2185 xfs_trans_brelse(tp, ibp); 2186 } 2187 /* 2188 * Point the previous inode on the list to the next inode. 2189 */ 2190 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2191 ASSERT(next_agino != 0); 2192 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2193 2194 /* need to recalc the inode CRC if appropriate */ 2195 xfs_dinode_calc_crc(mp, last_dip); 2196 2197 xfs_trans_inode_buf(tp, last_ibp); 2198 xfs_trans_log_buf(tp, last_ibp, offset, 2199 (offset + sizeof(xfs_agino_t) - 1)); 2200 xfs_inobp_check(mp, last_ibp); 2201 } 2202 return 0; 2203 } 2204 2205 /* 2206 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2207 * inodes that are in memory - they all must be marked stale and attached to 2208 * the cluster buffer. 2209 */ 2210 STATIC int 2211 xfs_ifree_cluster( 2212 xfs_inode_t *free_ip, 2213 xfs_trans_t *tp, 2214 struct xfs_icluster *xic) 2215 { 2216 xfs_mount_t *mp = free_ip->i_mount; 2217 int blks_per_cluster; 2218 int inodes_per_cluster; 2219 int nbufs; 2220 int i, j; 2221 int ioffset; 2222 xfs_daddr_t blkno; 2223 xfs_buf_t *bp; 2224 xfs_inode_t *ip; 2225 xfs_inode_log_item_t *iip; 2226 xfs_log_item_t *lip; 2227 struct xfs_perag *pag; 2228 xfs_ino_t inum; 2229 2230 inum = xic->first_ino; 2231 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2232 blks_per_cluster = xfs_icluster_size_fsb(mp); 2233 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2234 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2235 2236 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2237 /* 2238 * The allocation bitmap tells us which inodes of the chunk were 2239 * physically allocated. Skip the cluster if an inode falls into 2240 * a sparse region. 2241 */ 2242 ioffset = inum - xic->first_ino; 2243 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2244 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2245 continue; 2246 } 2247 2248 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2249 XFS_INO_TO_AGBNO(mp, inum)); 2250 2251 /* 2252 * We obtain and lock the backing buffer first in the process 2253 * here, as we have to ensure that any dirty inode that we 2254 * can't get the flush lock on is attached to the buffer. 2255 * If we scan the in-memory inodes first, then buffer IO can 2256 * complete before we get a lock on it, and hence we may fail 2257 * to mark all the active inodes on the buffer stale. 2258 */ 2259 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2260 mp->m_bsize * blks_per_cluster, 2261 XBF_UNMAPPED); 2262 2263 if (!bp) 2264 return -ENOMEM; 2265 2266 /* 2267 * This buffer may not have been correctly initialised as we 2268 * didn't read it from disk. That's not important because we are 2269 * only using to mark the buffer as stale in the log, and to 2270 * attach stale cached inodes on it. That means it will never be 2271 * dispatched for IO. If it is, we want to know about it, and we 2272 * want it to fail. We can acheive this by adding a write 2273 * verifier to the buffer. 2274 */ 2275 bp->b_ops = &xfs_inode_buf_ops; 2276 2277 /* 2278 * Walk the inodes already attached to the buffer and mark them 2279 * stale. These will all have the flush locks held, so an 2280 * in-memory inode walk can't lock them. By marking them all 2281 * stale first, we will not attempt to lock them in the loop 2282 * below as the XFS_ISTALE flag will be set. 2283 */ 2284 lip = bp->b_fspriv; 2285 while (lip) { 2286 if (lip->li_type == XFS_LI_INODE) { 2287 iip = (xfs_inode_log_item_t *)lip; 2288 ASSERT(iip->ili_logged == 1); 2289 lip->li_cb = xfs_istale_done; 2290 xfs_trans_ail_copy_lsn(mp->m_ail, 2291 &iip->ili_flush_lsn, 2292 &iip->ili_item.li_lsn); 2293 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2294 } 2295 lip = lip->li_bio_list; 2296 } 2297 2298 2299 /* 2300 * For each inode in memory attempt to add it to the inode 2301 * buffer and set it up for being staled on buffer IO 2302 * completion. This is safe as we've locked out tail pushing 2303 * and flushing by locking the buffer. 2304 * 2305 * We have already marked every inode that was part of a 2306 * transaction stale above, which means there is no point in 2307 * even trying to lock them. 2308 */ 2309 for (i = 0; i < inodes_per_cluster; i++) { 2310 retry: 2311 rcu_read_lock(); 2312 ip = radix_tree_lookup(&pag->pag_ici_root, 2313 XFS_INO_TO_AGINO(mp, (inum + i))); 2314 2315 /* Inode not in memory, nothing to do */ 2316 if (!ip) { 2317 rcu_read_unlock(); 2318 continue; 2319 } 2320 2321 /* 2322 * because this is an RCU protected lookup, we could 2323 * find a recently freed or even reallocated inode 2324 * during the lookup. We need to check under the 2325 * i_flags_lock for a valid inode here. Skip it if it 2326 * is not valid, the wrong inode or stale. 2327 */ 2328 spin_lock(&ip->i_flags_lock); 2329 if (ip->i_ino != inum + i || 2330 __xfs_iflags_test(ip, XFS_ISTALE)) { 2331 spin_unlock(&ip->i_flags_lock); 2332 rcu_read_unlock(); 2333 continue; 2334 } 2335 spin_unlock(&ip->i_flags_lock); 2336 2337 /* 2338 * Don't try to lock/unlock the current inode, but we 2339 * _cannot_ skip the other inodes that we did not find 2340 * in the list attached to the buffer and are not 2341 * already marked stale. If we can't lock it, back off 2342 * and retry. 2343 */ 2344 if (ip != free_ip) { 2345 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2346 rcu_read_unlock(); 2347 delay(1); 2348 goto retry; 2349 } 2350 2351 /* 2352 * Check the inode number again in case we're 2353 * racing with freeing in xfs_reclaim_inode(). 2354 * See the comments in that function for more 2355 * information as to why the initial check is 2356 * not sufficient. 2357 */ 2358 if (ip->i_ino != inum + i) { 2359 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2360 rcu_read_unlock(); 2361 continue; 2362 } 2363 } 2364 rcu_read_unlock(); 2365 2366 xfs_iflock(ip); 2367 xfs_iflags_set(ip, XFS_ISTALE); 2368 2369 /* 2370 * we don't need to attach clean inodes or those only 2371 * with unlogged changes (which we throw away, anyway). 2372 */ 2373 iip = ip->i_itemp; 2374 if (!iip || xfs_inode_clean(ip)) { 2375 ASSERT(ip != free_ip); 2376 xfs_ifunlock(ip); 2377 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2378 continue; 2379 } 2380 2381 iip->ili_last_fields = iip->ili_fields; 2382 iip->ili_fields = 0; 2383 iip->ili_fsync_fields = 0; 2384 iip->ili_logged = 1; 2385 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2386 &iip->ili_item.li_lsn); 2387 2388 xfs_buf_attach_iodone(bp, xfs_istale_done, 2389 &iip->ili_item); 2390 2391 if (ip != free_ip) 2392 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2393 } 2394 2395 xfs_trans_stale_inode_buf(tp, bp); 2396 xfs_trans_binval(tp, bp); 2397 } 2398 2399 xfs_perag_put(pag); 2400 return 0; 2401 } 2402 2403 /* 2404 * This is called to return an inode to the inode free list. 2405 * The inode should already be truncated to 0 length and have 2406 * no pages associated with it. This routine also assumes that 2407 * the inode is already a part of the transaction. 2408 * 2409 * The on-disk copy of the inode will have been added to the list 2410 * of unlinked inodes in the AGI. We need to remove the inode from 2411 * that list atomically with respect to freeing it here. 2412 */ 2413 int 2414 xfs_ifree( 2415 xfs_trans_t *tp, 2416 xfs_inode_t *ip, 2417 struct xfs_defer_ops *dfops) 2418 { 2419 int error; 2420 struct xfs_icluster xic = { 0 }; 2421 2422 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2423 ASSERT(VFS_I(ip)->i_nlink == 0); 2424 ASSERT(ip->i_d.di_nextents == 0); 2425 ASSERT(ip->i_d.di_anextents == 0); 2426 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2427 ASSERT(ip->i_d.di_nblocks == 0); 2428 2429 /* 2430 * Pull the on-disk inode from the AGI unlinked list. 2431 */ 2432 error = xfs_iunlink_remove(tp, ip); 2433 if (error) 2434 return error; 2435 2436 error = xfs_difree(tp, ip->i_ino, dfops, &xic); 2437 if (error) 2438 return error; 2439 2440 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2441 ip->i_d.di_flags = 0; 2442 ip->i_d.di_dmevmask = 0; 2443 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2444 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2445 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2446 /* 2447 * Bump the generation count so no one will be confused 2448 * by reincarnations of this inode. 2449 */ 2450 VFS_I(ip)->i_generation++; 2451 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2452 2453 if (xic.deleted) 2454 error = xfs_ifree_cluster(ip, tp, &xic); 2455 2456 return error; 2457 } 2458 2459 /* 2460 * This is called to unpin an inode. The caller must have the inode locked 2461 * in at least shared mode so that the buffer cannot be subsequently pinned 2462 * once someone is waiting for it to be unpinned. 2463 */ 2464 static void 2465 xfs_iunpin( 2466 struct xfs_inode *ip) 2467 { 2468 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2469 2470 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2471 2472 /* Give the log a push to start the unpinning I/O */ 2473 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2474 2475 } 2476 2477 static void 2478 __xfs_iunpin_wait( 2479 struct xfs_inode *ip) 2480 { 2481 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2482 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2483 2484 xfs_iunpin(ip); 2485 2486 do { 2487 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 2488 if (xfs_ipincount(ip)) 2489 io_schedule(); 2490 } while (xfs_ipincount(ip)); 2491 finish_wait(wq, &wait.wq_entry); 2492 } 2493 2494 void 2495 xfs_iunpin_wait( 2496 struct xfs_inode *ip) 2497 { 2498 if (xfs_ipincount(ip)) 2499 __xfs_iunpin_wait(ip); 2500 } 2501 2502 /* 2503 * Removing an inode from the namespace involves removing the directory entry 2504 * and dropping the link count on the inode. Removing the directory entry can 2505 * result in locking an AGF (directory blocks were freed) and removing a link 2506 * count can result in placing the inode on an unlinked list which results in 2507 * locking an AGI. 2508 * 2509 * The big problem here is that we have an ordering constraint on AGF and AGI 2510 * locking - inode allocation locks the AGI, then can allocate a new extent for 2511 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2512 * removes the inode from the unlinked list, requiring that we lock the AGI 2513 * first, and then freeing the inode can result in an inode chunk being freed 2514 * and hence freeing disk space requiring that we lock an AGF. 2515 * 2516 * Hence the ordering that is imposed by other parts of the code is AGI before 2517 * AGF. This means we cannot remove the directory entry before we drop the inode 2518 * reference count and put it on the unlinked list as this results in a lock 2519 * order of AGF then AGI, and this can deadlock against inode allocation and 2520 * freeing. Therefore we must drop the link counts before we remove the 2521 * directory entry. 2522 * 2523 * This is still safe from a transactional point of view - it is not until we 2524 * get to xfs_defer_finish() that we have the possibility of multiple 2525 * transactions in this operation. Hence as long as we remove the directory 2526 * entry and drop the link count in the first transaction of the remove 2527 * operation, there are no transactional constraints on the ordering here. 2528 */ 2529 int 2530 xfs_remove( 2531 xfs_inode_t *dp, 2532 struct xfs_name *name, 2533 xfs_inode_t *ip) 2534 { 2535 xfs_mount_t *mp = dp->i_mount; 2536 xfs_trans_t *tp = NULL; 2537 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2538 int error = 0; 2539 struct xfs_defer_ops dfops; 2540 xfs_fsblock_t first_block; 2541 uint resblks; 2542 2543 trace_xfs_remove(dp, name); 2544 2545 if (XFS_FORCED_SHUTDOWN(mp)) 2546 return -EIO; 2547 2548 error = xfs_qm_dqattach(dp, 0); 2549 if (error) 2550 goto std_return; 2551 2552 error = xfs_qm_dqattach(ip, 0); 2553 if (error) 2554 goto std_return; 2555 2556 /* 2557 * We try to get the real space reservation first, 2558 * allowing for directory btree deletion(s) implying 2559 * possible bmap insert(s). If we can't get the space 2560 * reservation then we use 0 instead, and avoid the bmap 2561 * btree insert(s) in the directory code by, if the bmap 2562 * insert tries to happen, instead trimming the LAST 2563 * block from the directory. 2564 */ 2565 resblks = XFS_REMOVE_SPACE_RES(mp); 2566 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2567 if (error == -ENOSPC) { 2568 resblks = 0; 2569 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2570 &tp); 2571 } 2572 if (error) { 2573 ASSERT(error != -ENOSPC); 2574 goto std_return; 2575 } 2576 2577 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2578 2579 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2580 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2581 2582 /* 2583 * If we're removing a directory perform some additional validation. 2584 */ 2585 if (is_dir) { 2586 ASSERT(VFS_I(ip)->i_nlink >= 2); 2587 if (VFS_I(ip)->i_nlink != 2) { 2588 error = -ENOTEMPTY; 2589 goto out_trans_cancel; 2590 } 2591 if (!xfs_dir_isempty(ip)) { 2592 error = -ENOTEMPTY; 2593 goto out_trans_cancel; 2594 } 2595 2596 /* Drop the link from ip's "..". */ 2597 error = xfs_droplink(tp, dp); 2598 if (error) 2599 goto out_trans_cancel; 2600 2601 /* Drop the "." link from ip to self. */ 2602 error = xfs_droplink(tp, ip); 2603 if (error) 2604 goto out_trans_cancel; 2605 } else { 2606 /* 2607 * When removing a non-directory we need to log the parent 2608 * inode here. For a directory this is done implicitly 2609 * by the xfs_droplink call for the ".." entry. 2610 */ 2611 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2612 } 2613 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2614 2615 /* Drop the link from dp to ip. */ 2616 error = xfs_droplink(tp, ip); 2617 if (error) 2618 goto out_trans_cancel; 2619 2620 xfs_defer_init(&dfops, &first_block); 2621 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2622 &first_block, &dfops, resblks); 2623 if (error) { 2624 ASSERT(error != -ENOENT); 2625 goto out_bmap_cancel; 2626 } 2627 2628 /* 2629 * If this is a synchronous mount, make sure that the 2630 * remove transaction goes to disk before returning to 2631 * the user. 2632 */ 2633 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2634 xfs_trans_set_sync(tp); 2635 2636 error = xfs_defer_finish(&tp, &dfops); 2637 if (error) 2638 goto out_bmap_cancel; 2639 2640 error = xfs_trans_commit(tp); 2641 if (error) 2642 goto std_return; 2643 2644 if (is_dir && xfs_inode_is_filestream(ip)) 2645 xfs_filestream_deassociate(ip); 2646 2647 return 0; 2648 2649 out_bmap_cancel: 2650 xfs_defer_cancel(&dfops); 2651 out_trans_cancel: 2652 xfs_trans_cancel(tp); 2653 std_return: 2654 return error; 2655 } 2656 2657 /* 2658 * Enter all inodes for a rename transaction into a sorted array. 2659 */ 2660 #define __XFS_SORT_INODES 5 2661 STATIC void 2662 xfs_sort_for_rename( 2663 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2664 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2665 struct xfs_inode *ip1, /* in: inode of old entry */ 2666 struct xfs_inode *ip2, /* in: inode of new entry */ 2667 struct xfs_inode *wip, /* in: whiteout inode */ 2668 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2669 int *num_inodes) /* in/out: inodes in array */ 2670 { 2671 int i, j; 2672 2673 ASSERT(*num_inodes == __XFS_SORT_INODES); 2674 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2675 2676 /* 2677 * i_tab contains a list of pointers to inodes. We initialize 2678 * the table here & we'll sort it. We will then use it to 2679 * order the acquisition of the inode locks. 2680 * 2681 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2682 */ 2683 i = 0; 2684 i_tab[i++] = dp1; 2685 i_tab[i++] = dp2; 2686 i_tab[i++] = ip1; 2687 if (ip2) 2688 i_tab[i++] = ip2; 2689 if (wip) 2690 i_tab[i++] = wip; 2691 *num_inodes = i; 2692 2693 /* 2694 * Sort the elements via bubble sort. (Remember, there are at 2695 * most 5 elements to sort, so this is adequate.) 2696 */ 2697 for (i = 0; i < *num_inodes; i++) { 2698 for (j = 1; j < *num_inodes; j++) { 2699 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2700 struct xfs_inode *temp = i_tab[j]; 2701 i_tab[j] = i_tab[j-1]; 2702 i_tab[j-1] = temp; 2703 } 2704 } 2705 } 2706 } 2707 2708 static int 2709 xfs_finish_rename( 2710 struct xfs_trans *tp, 2711 struct xfs_defer_ops *dfops) 2712 { 2713 int error; 2714 2715 /* 2716 * If this is a synchronous mount, make sure that the rename transaction 2717 * goes to disk before returning to the user. 2718 */ 2719 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2720 xfs_trans_set_sync(tp); 2721 2722 error = xfs_defer_finish(&tp, dfops); 2723 if (error) { 2724 xfs_defer_cancel(dfops); 2725 xfs_trans_cancel(tp); 2726 return error; 2727 } 2728 2729 return xfs_trans_commit(tp); 2730 } 2731 2732 /* 2733 * xfs_cross_rename() 2734 * 2735 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2736 */ 2737 STATIC int 2738 xfs_cross_rename( 2739 struct xfs_trans *tp, 2740 struct xfs_inode *dp1, 2741 struct xfs_name *name1, 2742 struct xfs_inode *ip1, 2743 struct xfs_inode *dp2, 2744 struct xfs_name *name2, 2745 struct xfs_inode *ip2, 2746 struct xfs_defer_ops *dfops, 2747 xfs_fsblock_t *first_block, 2748 int spaceres) 2749 { 2750 int error = 0; 2751 int ip1_flags = 0; 2752 int ip2_flags = 0; 2753 int dp2_flags = 0; 2754 2755 /* Swap inode number for dirent in first parent */ 2756 error = xfs_dir_replace(tp, dp1, name1, 2757 ip2->i_ino, 2758 first_block, dfops, spaceres); 2759 if (error) 2760 goto out_trans_abort; 2761 2762 /* Swap inode number for dirent in second parent */ 2763 error = xfs_dir_replace(tp, dp2, name2, 2764 ip1->i_ino, 2765 first_block, dfops, spaceres); 2766 if (error) 2767 goto out_trans_abort; 2768 2769 /* 2770 * If we're renaming one or more directories across different parents, 2771 * update the respective ".." entries (and link counts) to match the new 2772 * parents. 2773 */ 2774 if (dp1 != dp2) { 2775 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2776 2777 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2778 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2779 dp1->i_ino, first_block, 2780 dfops, spaceres); 2781 if (error) 2782 goto out_trans_abort; 2783 2784 /* transfer ip2 ".." reference to dp1 */ 2785 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2786 error = xfs_droplink(tp, dp2); 2787 if (error) 2788 goto out_trans_abort; 2789 error = xfs_bumplink(tp, dp1); 2790 if (error) 2791 goto out_trans_abort; 2792 } 2793 2794 /* 2795 * Although ip1 isn't changed here, userspace needs 2796 * to be warned about the change, so that applications 2797 * relying on it (like backup ones), will properly 2798 * notify the change 2799 */ 2800 ip1_flags |= XFS_ICHGTIME_CHG; 2801 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2802 } 2803 2804 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2805 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2806 dp2->i_ino, first_block, 2807 dfops, spaceres); 2808 if (error) 2809 goto out_trans_abort; 2810 2811 /* transfer ip1 ".." reference to dp2 */ 2812 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2813 error = xfs_droplink(tp, dp1); 2814 if (error) 2815 goto out_trans_abort; 2816 error = xfs_bumplink(tp, dp2); 2817 if (error) 2818 goto out_trans_abort; 2819 } 2820 2821 /* 2822 * Although ip2 isn't changed here, userspace needs 2823 * to be warned about the change, so that applications 2824 * relying on it (like backup ones), will properly 2825 * notify the change 2826 */ 2827 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2828 ip2_flags |= XFS_ICHGTIME_CHG; 2829 } 2830 } 2831 2832 if (ip1_flags) { 2833 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2834 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2835 } 2836 if (ip2_flags) { 2837 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2838 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2839 } 2840 if (dp2_flags) { 2841 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2842 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2843 } 2844 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2845 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2846 return xfs_finish_rename(tp, dfops); 2847 2848 out_trans_abort: 2849 xfs_defer_cancel(dfops); 2850 xfs_trans_cancel(tp); 2851 return error; 2852 } 2853 2854 /* 2855 * xfs_rename_alloc_whiteout() 2856 * 2857 * Return a referenced, unlinked, unlocked inode that that can be used as a 2858 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2859 * crash between allocating the inode and linking it into the rename transaction 2860 * recovery will free the inode and we won't leak it. 2861 */ 2862 static int 2863 xfs_rename_alloc_whiteout( 2864 struct xfs_inode *dp, 2865 struct xfs_inode **wip) 2866 { 2867 struct xfs_inode *tmpfile; 2868 int error; 2869 2870 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2871 if (error) 2872 return error; 2873 2874 /* 2875 * Prepare the tmpfile inode as if it were created through the VFS. 2876 * Otherwise, the link increment paths will complain about nlink 0->1. 2877 * Drop the link count as done by d_tmpfile(), complete the inode setup 2878 * and flag it as linkable. 2879 */ 2880 drop_nlink(VFS_I(tmpfile)); 2881 xfs_setup_iops(tmpfile); 2882 xfs_finish_inode_setup(tmpfile); 2883 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2884 2885 *wip = tmpfile; 2886 return 0; 2887 } 2888 2889 /* 2890 * xfs_rename 2891 */ 2892 int 2893 xfs_rename( 2894 struct xfs_inode *src_dp, 2895 struct xfs_name *src_name, 2896 struct xfs_inode *src_ip, 2897 struct xfs_inode *target_dp, 2898 struct xfs_name *target_name, 2899 struct xfs_inode *target_ip, 2900 unsigned int flags) 2901 { 2902 struct xfs_mount *mp = src_dp->i_mount; 2903 struct xfs_trans *tp; 2904 struct xfs_defer_ops dfops; 2905 xfs_fsblock_t first_block; 2906 struct xfs_inode *wip = NULL; /* whiteout inode */ 2907 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2908 int num_inodes = __XFS_SORT_INODES; 2909 bool new_parent = (src_dp != target_dp); 2910 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2911 int spaceres; 2912 int error; 2913 2914 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2915 2916 if ((flags & RENAME_EXCHANGE) && !target_ip) 2917 return -EINVAL; 2918 2919 /* 2920 * If we are doing a whiteout operation, allocate the whiteout inode 2921 * we will be placing at the target and ensure the type is set 2922 * appropriately. 2923 */ 2924 if (flags & RENAME_WHITEOUT) { 2925 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2926 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2927 if (error) 2928 return error; 2929 2930 /* setup target dirent info as whiteout */ 2931 src_name->type = XFS_DIR3_FT_CHRDEV; 2932 } 2933 2934 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2935 inodes, &num_inodes); 2936 2937 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2938 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2939 if (error == -ENOSPC) { 2940 spaceres = 0; 2941 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2942 &tp); 2943 } 2944 if (error) 2945 goto out_release_wip; 2946 2947 /* 2948 * Attach the dquots to the inodes 2949 */ 2950 error = xfs_qm_vop_rename_dqattach(inodes); 2951 if (error) 2952 goto out_trans_cancel; 2953 2954 /* 2955 * Lock all the participating inodes. Depending upon whether 2956 * the target_name exists in the target directory, and 2957 * whether the target directory is the same as the source 2958 * directory, we can lock from 2 to 4 inodes. 2959 */ 2960 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2961 2962 /* 2963 * Join all the inodes to the transaction. From this point on, 2964 * we can rely on either trans_commit or trans_cancel to unlock 2965 * them. 2966 */ 2967 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 2968 if (new_parent) 2969 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 2970 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2971 if (target_ip) 2972 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2973 if (wip) 2974 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2975 2976 /* 2977 * If we are using project inheritance, we only allow renames 2978 * into our tree when the project IDs are the same; else the 2979 * tree quota mechanism would be circumvented. 2980 */ 2981 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2982 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2983 error = -EXDEV; 2984 goto out_trans_cancel; 2985 } 2986 2987 xfs_defer_init(&dfops, &first_block); 2988 2989 /* RENAME_EXCHANGE is unique from here on. */ 2990 if (flags & RENAME_EXCHANGE) 2991 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2992 target_dp, target_name, target_ip, 2993 &dfops, &first_block, spaceres); 2994 2995 /* 2996 * Set up the target. 2997 */ 2998 if (target_ip == NULL) { 2999 /* 3000 * If there's no space reservation, check the entry will 3001 * fit before actually inserting it. 3002 */ 3003 if (!spaceres) { 3004 error = xfs_dir_canenter(tp, target_dp, target_name); 3005 if (error) 3006 goto out_trans_cancel; 3007 } 3008 /* 3009 * If target does not exist and the rename crosses 3010 * directories, adjust the target directory link count 3011 * to account for the ".." reference from the new entry. 3012 */ 3013 error = xfs_dir_createname(tp, target_dp, target_name, 3014 src_ip->i_ino, &first_block, 3015 &dfops, spaceres); 3016 if (error) 3017 goto out_bmap_cancel; 3018 3019 xfs_trans_ichgtime(tp, target_dp, 3020 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3021 3022 if (new_parent && src_is_directory) { 3023 error = xfs_bumplink(tp, target_dp); 3024 if (error) 3025 goto out_bmap_cancel; 3026 } 3027 } else { /* target_ip != NULL */ 3028 /* 3029 * If target exists and it's a directory, check that both 3030 * target and source are directories and that target can be 3031 * destroyed, or that neither is a directory. 3032 */ 3033 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 3034 /* 3035 * Make sure target dir is empty. 3036 */ 3037 if (!(xfs_dir_isempty(target_ip)) || 3038 (VFS_I(target_ip)->i_nlink > 2)) { 3039 error = -EEXIST; 3040 goto out_trans_cancel; 3041 } 3042 } 3043 3044 /* 3045 * Link the source inode under the target name. 3046 * If the source inode is a directory and we are moving 3047 * it across directories, its ".." entry will be 3048 * inconsistent until we replace that down below. 3049 * 3050 * In case there is already an entry with the same 3051 * name at the destination directory, remove it first. 3052 */ 3053 error = xfs_dir_replace(tp, target_dp, target_name, 3054 src_ip->i_ino, 3055 &first_block, &dfops, spaceres); 3056 if (error) 3057 goto out_bmap_cancel; 3058 3059 xfs_trans_ichgtime(tp, target_dp, 3060 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3061 3062 /* 3063 * Decrement the link count on the target since the target 3064 * dir no longer points to it. 3065 */ 3066 error = xfs_droplink(tp, target_ip); 3067 if (error) 3068 goto out_bmap_cancel; 3069 3070 if (src_is_directory) { 3071 /* 3072 * Drop the link from the old "." entry. 3073 */ 3074 error = xfs_droplink(tp, target_ip); 3075 if (error) 3076 goto out_bmap_cancel; 3077 } 3078 } /* target_ip != NULL */ 3079 3080 /* 3081 * Remove the source. 3082 */ 3083 if (new_parent && src_is_directory) { 3084 /* 3085 * Rewrite the ".." entry to point to the new 3086 * directory. 3087 */ 3088 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3089 target_dp->i_ino, 3090 &first_block, &dfops, spaceres); 3091 ASSERT(error != -EEXIST); 3092 if (error) 3093 goto out_bmap_cancel; 3094 } 3095 3096 /* 3097 * We always want to hit the ctime on the source inode. 3098 * 3099 * This isn't strictly required by the standards since the source 3100 * inode isn't really being changed, but old unix file systems did 3101 * it and some incremental backup programs won't work without it. 3102 */ 3103 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3104 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3105 3106 /* 3107 * Adjust the link count on src_dp. This is necessary when 3108 * renaming a directory, either within one parent when 3109 * the target existed, or across two parent directories. 3110 */ 3111 if (src_is_directory && (new_parent || target_ip != NULL)) { 3112 3113 /* 3114 * Decrement link count on src_directory since the 3115 * entry that's moved no longer points to it. 3116 */ 3117 error = xfs_droplink(tp, src_dp); 3118 if (error) 3119 goto out_bmap_cancel; 3120 } 3121 3122 /* 3123 * For whiteouts, we only need to update the source dirent with the 3124 * inode number of the whiteout inode rather than removing it 3125 * altogether. 3126 */ 3127 if (wip) { 3128 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3129 &first_block, &dfops, spaceres); 3130 } else 3131 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3132 &first_block, &dfops, spaceres); 3133 if (error) 3134 goto out_bmap_cancel; 3135 3136 /* 3137 * For whiteouts, we need to bump the link count on the whiteout inode. 3138 * This means that failures all the way up to this point leave the inode 3139 * on the unlinked list and so cleanup is a simple matter of dropping 3140 * the remaining reference to it. If we fail here after bumping the link 3141 * count, we're shutting down the filesystem so we'll never see the 3142 * intermediate state on disk. 3143 */ 3144 if (wip) { 3145 ASSERT(VFS_I(wip)->i_nlink == 0); 3146 error = xfs_bumplink(tp, wip); 3147 if (error) 3148 goto out_bmap_cancel; 3149 error = xfs_iunlink_remove(tp, wip); 3150 if (error) 3151 goto out_bmap_cancel; 3152 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3153 3154 /* 3155 * Now we have a real link, clear the "I'm a tmpfile" state 3156 * flag from the inode so it doesn't accidentally get misused in 3157 * future. 3158 */ 3159 VFS_I(wip)->i_state &= ~I_LINKABLE; 3160 } 3161 3162 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3163 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3164 if (new_parent) 3165 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3166 3167 error = xfs_finish_rename(tp, &dfops); 3168 if (wip) 3169 IRELE(wip); 3170 return error; 3171 3172 out_bmap_cancel: 3173 xfs_defer_cancel(&dfops); 3174 out_trans_cancel: 3175 xfs_trans_cancel(tp); 3176 out_release_wip: 3177 if (wip) 3178 IRELE(wip); 3179 return error; 3180 } 3181 3182 STATIC int 3183 xfs_iflush_cluster( 3184 struct xfs_inode *ip, 3185 struct xfs_buf *bp) 3186 { 3187 struct xfs_mount *mp = ip->i_mount; 3188 struct xfs_perag *pag; 3189 unsigned long first_index, mask; 3190 unsigned long inodes_per_cluster; 3191 int cilist_size; 3192 struct xfs_inode **cilist; 3193 struct xfs_inode *cip; 3194 int nr_found; 3195 int clcount = 0; 3196 int bufwasdelwri; 3197 int i; 3198 3199 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3200 3201 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3202 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3203 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3204 if (!cilist) 3205 goto out_put; 3206 3207 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3208 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3209 rcu_read_lock(); 3210 /* really need a gang lookup range call here */ 3211 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3212 first_index, inodes_per_cluster); 3213 if (nr_found == 0) 3214 goto out_free; 3215 3216 for (i = 0; i < nr_found; i++) { 3217 cip = cilist[i]; 3218 if (cip == ip) 3219 continue; 3220 3221 /* 3222 * because this is an RCU protected lookup, we could find a 3223 * recently freed or even reallocated inode during the lookup. 3224 * We need to check under the i_flags_lock for a valid inode 3225 * here. Skip it if it is not valid or the wrong inode. 3226 */ 3227 spin_lock(&cip->i_flags_lock); 3228 if (!cip->i_ino || 3229 __xfs_iflags_test(cip, XFS_ISTALE)) { 3230 spin_unlock(&cip->i_flags_lock); 3231 continue; 3232 } 3233 3234 /* 3235 * Once we fall off the end of the cluster, no point checking 3236 * any more inodes in the list because they will also all be 3237 * outside the cluster. 3238 */ 3239 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3240 spin_unlock(&cip->i_flags_lock); 3241 break; 3242 } 3243 spin_unlock(&cip->i_flags_lock); 3244 3245 /* 3246 * Do an un-protected check to see if the inode is dirty and 3247 * is a candidate for flushing. These checks will be repeated 3248 * later after the appropriate locks are acquired. 3249 */ 3250 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3251 continue; 3252 3253 /* 3254 * Try to get locks. If any are unavailable or it is pinned, 3255 * then this inode cannot be flushed and is skipped. 3256 */ 3257 3258 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3259 continue; 3260 if (!xfs_iflock_nowait(cip)) { 3261 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3262 continue; 3263 } 3264 if (xfs_ipincount(cip)) { 3265 xfs_ifunlock(cip); 3266 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3267 continue; 3268 } 3269 3270 3271 /* 3272 * Check the inode number again, just to be certain we are not 3273 * racing with freeing in xfs_reclaim_inode(). See the comments 3274 * in that function for more information as to why the initial 3275 * check is not sufficient. 3276 */ 3277 if (!cip->i_ino) { 3278 xfs_ifunlock(cip); 3279 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3280 continue; 3281 } 3282 3283 /* 3284 * arriving here means that this inode can be flushed. First 3285 * re-check that it's dirty before flushing. 3286 */ 3287 if (!xfs_inode_clean(cip)) { 3288 int error; 3289 error = xfs_iflush_int(cip, bp); 3290 if (error) { 3291 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3292 goto cluster_corrupt_out; 3293 } 3294 clcount++; 3295 } else { 3296 xfs_ifunlock(cip); 3297 } 3298 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3299 } 3300 3301 if (clcount) { 3302 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3303 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3304 } 3305 3306 out_free: 3307 rcu_read_unlock(); 3308 kmem_free(cilist); 3309 out_put: 3310 xfs_perag_put(pag); 3311 return 0; 3312 3313 3314 cluster_corrupt_out: 3315 /* 3316 * Corruption detected in the clustering loop. Invalidate the 3317 * inode buffer and shut down the filesystem. 3318 */ 3319 rcu_read_unlock(); 3320 /* 3321 * Clean up the buffer. If it was delwri, just release it -- 3322 * brelse can handle it with no problems. If not, shut down the 3323 * filesystem before releasing the buffer. 3324 */ 3325 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3326 if (bufwasdelwri) 3327 xfs_buf_relse(bp); 3328 3329 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3330 3331 if (!bufwasdelwri) { 3332 /* 3333 * Just like incore_relse: if we have b_iodone functions, 3334 * mark the buffer as an error and call them. Otherwise 3335 * mark it as stale and brelse. 3336 */ 3337 if (bp->b_iodone) { 3338 bp->b_flags &= ~XBF_DONE; 3339 xfs_buf_stale(bp); 3340 xfs_buf_ioerror(bp, -EIO); 3341 xfs_buf_ioend(bp); 3342 } else { 3343 xfs_buf_stale(bp); 3344 xfs_buf_relse(bp); 3345 } 3346 } 3347 3348 /* 3349 * Unlocks the flush lock 3350 */ 3351 xfs_iflush_abort(cip, false); 3352 kmem_free(cilist); 3353 xfs_perag_put(pag); 3354 return -EFSCORRUPTED; 3355 } 3356 3357 /* 3358 * Flush dirty inode metadata into the backing buffer. 3359 * 3360 * The caller must have the inode lock and the inode flush lock held. The 3361 * inode lock will still be held upon return to the caller, and the inode 3362 * flush lock will be released after the inode has reached the disk. 3363 * 3364 * The caller must write out the buffer returned in *bpp and release it. 3365 */ 3366 int 3367 xfs_iflush( 3368 struct xfs_inode *ip, 3369 struct xfs_buf **bpp) 3370 { 3371 struct xfs_mount *mp = ip->i_mount; 3372 struct xfs_buf *bp = NULL; 3373 struct xfs_dinode *dip; 3374 int error; 3375 3376 XFS_STATS_INC(mp, xs_iflush_count); 3377 3378 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3379 ASSERT(xfs_isiflocked(ip)); 3380 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3381 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3382 3383 *bpp = NULL; 3384 3385 xfs_iunpin_wait(ip); 3386 3387 /* 3388 * For stale inodes we cannot rely on the backing buffer remaining 3389 * stale in cache for the remaining life of the stale inode and so 3390 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3391 * inodes below. We have to check this after ensuring the inode is 3392 * unpinned so that it is safe to reclaim the stale inode after the 3393 * flush call. 3394 */ 3395 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3396 xfs_ifunlock(ip); 3397 return 0; 3398 } 3399 3400 /* 3401 * This may have been unpinned because the filesystem is shutting 3402 * down forcibly. If that's the case we must not write this inode 3403 * to disk, because the log record didn't make it to disk. 3404 * 3405 * We also have to remove the log item from the AIL in this case, 3406 * as we wait for an empty AIL as part of the unmount process. 3407 */ 3408 if (XFS_FORCED_SHUTDOWN(mp)) { 3409 error = -EIO; 3410 goto abort_out; 3411 } 3412 3413 /* 3414 * Get the buffer containing the on-disk inode. We are doing a try-lock 3415 * operation here, so we may get an EAGAIN error. In that case, we 3416 * simply want to return with the inode still dirty. 3417 * 3418 * If we get any other error, we effectively have a corruption situation 3419 * and we cannot flush the inode, so we treat it the same as failing 3420 * xfs_iflush_int(). 3421 */ 3422 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3423 0); 3424 if (error == -EAGAIN) { 3425 xfs_ifunlock(ip); 3426 return error; 3427 } 3428 if (error) 3429 goto corrupt_out; 3430 3431 /* 3432 * First flush out the inode that xfs_iflush was called with. 3433 */ 3434 error = xfs_iflush_int(ip, bp); 3435 if (error) 3436 goto corrupt_out; 3437 3438 /* 3439 * If the buffer is pinned then push on the log now so we won't 3440 * get stuck waiting in the write for too long. 3441 */ 3442 if (xfs_buf_ispinned(bp)) 3443 xfs_log_force(mp, 0); 3444 3445 /* 3446 * inode clustering: 3447 * see if other inodes can be gathered into this write 3448 */ 3449 error = xfs_iflush_cluster(ip, bp); 3450 if (error) 3451 goto cluster_corrupt_out; 3452 3453 *bpp = bp; 3454 return 0; 3455 3456 corrupt_out: 3457 if (bp) 3458 xfs_buf_relse(bp); 3459 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3460 cluster_corrupt_out: 3461 error = -EFSCORRUPTED; 3462 abort_out: 3463 /* 3464 * Unlocks the flush lock 3465 */ 3466 xfs_iflush_abort(ip, false); 3467 return error; 3468 } 3469 3470 STATIC int 3471 xfs_iflush_int( 3472 struct xfs_inode *ip, 3473 struct xfs_buf *bp) 3474 { 3475 struct xfs_inode_log_item *iip = ip->i_itemp; 3476 struct xfs_dinode *dip; 3477 struct xfs_mount *mp = ip->i_mount; 3478 3479 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3480 ASSERT(xfs_isiflocked(ip)); 3481 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3482 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3483 ASSERT(iip != NULL && iip->ili_fields != 0); 3484 ASSERT(ip->i_d.di_version > 1); 3485 3486 /* set *dip = inode's place in the buffer */ 3487 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3488 3489 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3490 mp, XFS_ERRTAG_IFLUSH_1)) { 3491 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3492 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3493 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3494 goto corrupt_out; 3495 } 3496 if (S_ISREG(VFS_I(ip)->i_mode)) { 3497 if (XFS_TEST_ERROR( 3498 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3499 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3500 mp, XFS_ERRTAG_IFLUSH_3)) { 3501 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3502 "%s: Bad regular inode %Lu, ptr 0x%p", 3503 __func__, ip->i_ino, ip); 3504 goto corrupt_out; 3505 } 3506 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3507 if (XFS_TEST_ERROR( 3508 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3509 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3510 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3511 mp, XFS_ERRTAG_IFLUSH_4)) { 3512 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3513 "%s: Bad directory inode %Lu, ptr 0x%p", 3514 __func__, ip->i_ino, ip); 3515 goto corrupt_out; 3516 } 3517 } 3518 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3519 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { 3520 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3521 "%s: detected corrupt incore inode %Lu, " 3522 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3523 __func__, ip->i_ino, 3524 ip->i_d.di_nextents + ip->i_d.di_anextents, 3525 ip->i_d.di_nblocks, ip); 3526 goto corrupt_out; 3527 } 3528 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3529 mp, XFS_ERRTAG_IFLUSH_6)) { 3530 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3531 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3532 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3533 goto corrupt_out; 3534 } 3535 3536 /* 3537 * Inode item log recovery for v2 inodes are dependent on the 3538 * di_flushiter count for correct sequencing. We bump the flush 3539 * iteration count so we can detect flushes which postdate a log record 3540 * during recovery. This is redundant as we now log every change and 3541 * hence this can't happen but we need to still do it to ensure 3542 * backwards compatibility with old kernels that predate logging all 3543 * inode changes. 3544 */ 3545 if (ip->i_d.di_version < 3) 3546 ip->i_d.di_flushiter++; 3547 3548 /* Check the inline directory data. */ 3549 if (S_ISDIR(VFS_I(ip)->i_mode) && 3550 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL && 3551 xfs_dir2_sf_verify(ip)) 3552 goto corrupt_out; 3553 3554 /* 3555 * Copy the dirty parts of the inode into the on-disk inode. We always 3556 * copy out the core of the inode, because if the inode is dirty at all 3557 * the core must be. 3558 */ 3559 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3560 3561 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3562 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3563 ip->i_d.di_flushiter = 0; 3564 3565 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3566 if (XFS_IFORK_Q(ip)) 3567 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3568 xfs_inobp_check(mp, bp); 3569 3570 /* 3571 * We've recorded everything logged in the inode, so we'd like to clear 3572 * the ili_fields bits so we don't log and flush things unnecessarily. 3573 * However, we can't stop logging all this information until the data 3574 * we've copied into the disk buffer is written to disk. If we did we 3575 * might overwrite the copy of the inode in the log with all the data 3576 * after re-logging only part of it, and in the face of a crash we 3577 * wouldn't have all the data we need to recover. 3578 * 3579 * What we do is move the bits to the ili_last_fields field. When 3580 * logging the inode, these bits are moved back to the ili_fields field. 3581 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3582 * know that the information those bits represent is permanently on 3583 * disk. As long as the flush completes before the inode is logged 3584 * again, then both ili_fields and ili_last_fields will be cleared. 3585 * 3586 * We can play with the ili_fields bits here, because the inode lock 3587 * must be held exclusively in order to set bits there and the flush 3588 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3589 * done routine can tell whether or not to look in the AIL. Also, store 3590 * the current LSN of the inode so that we can tell whether the item has 3591 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3592 * need the AIL lock, because it is a 64 bit value that cannot be read 3593 * atomically. 3594 */ 3595 iip->ili_last_fields = iip->ili_fields; 3596 iip->ili_fields = 0; 3597 iip->ili_fsync_fields = 0; 3598 iip->ili_logged = 1; 3599 3600 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3601 &iip->ili_item.li_lsn); 3602 3603 /* 3604 * Attach the function xfs_iflush_done to the inode's 3605 * buffer. This will remove the inode from the AIL 3606 * and unlock the inode's flush lock when the inode is 3607 * completely written to disk. 3608 */ 3609 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3610 3611 /* generate the checksum. */ 3612 xfs_dinode_calc_crc(mp, dip); 3613 3614 ASSERT(bp->b_fspriv != NULL); 3615 ASSERT(bp->b_iodone != NULL); 3616 return 0; 3617 3618 corrupt_out: 3619 return -EFSCORRUPTED; 3620 } 3621