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 xfs_buf_t **ialloc_context, 753 xfs_inode_t **ipp) 754 { 755 struct xfs_mount *mp = tp->t_mountp; 756 xfs_ino_t ino; 757 xfs_inode_t *ip; 758 uint flags; 759 int error; 760 struct timespec tv; 761 struct inode *inode; 762 763 /* 764 * Call the space management code to pick 765 * the on-disk inode to be allocated. 766 */ 767 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, 768 ialloc_context, &ino); 769 if (error) 770 return error; 771 if (*ialloc_context || ino == NULLFSINO) { 772 *ipp = NULL; 773 return 0; 774 } 775 ASSERT(*ialloc_context == NULL); 776 777 /* 778 * Get the in-core inode with the lock held exclusively. 779 * This is because we're setting fields here we need 780 * to prevent others from looking at until we're done. 781 */ 782 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 783 XFS_ILOCK_EXCL, &ip); 784 if (error) 785 return error; 786 ASSERT(ip != NULL); 787 inode = VFS_I(ip); 788 789 /* 790 * We always convert v1 inodes to v2 now - we only support filesystems 791 * with >= v2 inode capability, so there is no reason for ever leaving 792 * an inode in v1 format. 793 */ 794 if (ip->i_d.di_version == 1) 795 ip->i_d.di_version = 2; 796 797 inode->i_mode = mode; 798 set_nlink(inode, nlink); 799 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 800 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 801 inode->i_rdev = rdev; 802 xfs_set_projid(ip, prid); 803 804 if (pip && XFS_INHERIT_GID(pip)) { 805 ip->i_d.di_gid = pip->i_d.di_gid; 806 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) 807 inode->i_mode |= S_ISGID; 808 } 809 810 /* 811 * If the group ID of the new file does not match the effective group 812 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 813 * (and only if the irix_sgid_inherit compatibility variable is set). 814 */ 815 if ((irix_sgid_inherit) && 816 (inode->i_mode & S_ISGID) && 817 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) 818 inode->i_mode &= ~S_ISGID; 819 820 ip->i_d.di_size = 0; 821 ip->i_d.di_nextents = 0; 822 ASSERT(ip->i_d.di_nblocks == 0); 823 824 tv = current_time(inode); 825 inode->i_mtime = tv; 826 inode->i_atime = tv; 827 inode->i_ctime = tv; 828 829 ip->i_d.di_extsize = 0; 830 ip->i_d.di_dmevmask = 0; 831 ip->i_d.di_dmstate = 0; 832 ip->i_d.di_flags = 0; 833 834 if (ip->i_d.di_version == 3) { 835 inode->i_version = 1; 836 ip->i_d.di_flags2 = 0; 837 ip->i_d.di_cowextsize = 0; 838 ip->i_d.di_crtime.t_sec = (int32_t)tv.tv_sec; 839 ip->i_d.di_crtime.t_nsec = (int32_t)tv.tv_nsec; 840 } 841 842 843 flags = XFS_ILOG_CORE; 844 switch (mode & S_IFMT) { 845 case S_IFIFO: 846 case S_IFCHR: 847 case S_IFBLK: 848 case S_IFSOCK: 849 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 850 ip->i_df.if_flags = 0; 851 flags |= XFS_ILOG_DEV; 852 break; 853 case S_IFREG: 854 case S_IFDIR: 855 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 856 uint di_flags = 0; 857 858 if (S_ISDIR(mode)) { 859 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 860 di_flags |= XFS_DIFLAG_RTINHERIT; 861 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 862 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 863 ip->i_d.di_extsize = pip->i_d.di_extsize; 864 } 865 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 866 di_flags |= XFS_DIFLAG_PROJINHERIT; 867 } else if (S_ISREG(mode)) { 868 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 869 di_flags |= XFS_DIFLAG_REALTIME; 870 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 871 di_flags |= XFS_DIFLAG_EXTSIZE; 872 ip->i_d.di_extsize = pip->i_d.di_extsize; 873 } 874 } 875 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 876 xfs_inherit_noatime) 877 di_flags |= XFS_DIFLAG_NOATIME; 878 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 879 xfs_inherit_nodump) 880 di_flags |= XFS_DIFLAG_NODUMP; 881 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 882 xfs_inherit_sync) 883 di_flags |= XFS_DIFLAG_SYNC; 884 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 885 xfs_inherit_nosymlinks) 886 di_flags |= XFS_DIFLAG_NOSYMLINKS; 887 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 888 xfs_inherit_nodefrag) 889 di_flags |= XFS_DIFLAG_NODEFRAG; 890 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 891 di_flags |= XFS_DIFLAG_FILESTREAM; 892 893 ip->i_d.di_flags |= di_flags; 894 } 895 if (pip && 896 (pip->i_d.di_flags2 & XFS_DIFLAG2_ANY) && 897 pip->i_d.di_version == 3 && 898 ip->i_d.di_version == 3) { 899 uint64_t di_flags2 = 0; 900 901 if (pip->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) { 902 di_flags2 |= XFS_DIFLAG2_COWEXTSIZE; 903 ip->i_d.di_cowextsize = pip->i_d.di_cowextsize; 904 } 905 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 906 di_flags2 |= XFS_DIFLAG2_DAX; 907 908 ip->i_d.di_flags2 |= di_flags2; 909 } 910 /* FALLTHROUGH */ 911 case S_IFLNK: 912 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 913 ip->i_df.if_flags = XFS_IFEXTENTS; 914 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 915 ip->i_df.if_u1.if_root = NULL; 916 break; 917 default: 918 ASSERT(0); 919 } 920 /* 921 * Attribute fork settings for new inode. 922 */ 923 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 924 ip->i_d.di_anextents = 0; 925 926 /* 927 * Log the new values stuffed into the inode. 928 */ 929 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 930 xfs_trans_log_inode(tp, ip, flags); 931 932 /* now that we have an i_mode we can setup the inode structure */ 933 xfs_setup_inode(ip); 934 935 *ipp = ip; 936 return 0; 937 } 938 939 /* 940 * Allocates a new inode from disk and return a pointer to the 941 * incore copy. This routine will internally commit the current 942 * transaction and allocate a new one if the Space Manager needed 943 * to do an allocation to replenish the inode free-list. 944 * 945 * This routine is designed to be called from xfs_create and 946 * xfs_create_dir. 947 * 948 */ 949 int 950 xfs_dir_ialloc( 951 xfs_trans_t **tpp, /* input: current transaction; 952 output: may be a new transaction. */ 953 xfs_inode_t *dp, /* directory within whose allocate 954 the inode. */ 955 umode_t mode, 956 xfs_nlink_t nlink, 957 dev_t rdev, 958 prid_t prid, /* project id */ 959 xfs_inode_t **ipp, /* pointer to inode; it will be 960 locked. */ 961 int *committed) 962 963 { 964 xfs_trans_t *tp; 965 xfs_inode_t *ip; 966 xfs_buf_t *ialloc_context = NULL; 967 int code; 968 void *dqinfo; 969 uint tflags; 970 971 tp = *tpp; 972 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 973 974 /* 975 * xfs_ialloc will return a pointer to an incore inode if 976 * the Space Manager has an available inode on the free 977 * list. Otherwise, it will do an allocation and replenish 978 * the freelist. Since we can only do one allocation per 979 * transaction without deadlocks, we will need to commit the 980 * current transaction and start a new one. We will then 981 * need to call xfs_ialloc again to get the inode. 982 * 983 * If xfs_ialloc did an allocation to replenish the freelist, 984 * it returns the bp containing the head of the freelist as 985 * ialloc_context. We will hold a lock on it across the 986 * transaction commit so that no other process can steal 987 * the inode(s) that we've just allocated. 988 */ 989 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, &ialloc_context, 990 &ip); 991 992 /* 993 * Return an error if we were unable to allocate a new inode. 994 * This should only happen if we run out of space on disk or 995 * encounter a disk error. 996 */ 997 if (code) { 998 *ipp = NULL; 999 return code; 1000 } 1001 if (!ialloc_context && !ip) { 1002 *ipp = NULL; 1003 return -ENOSPC; 1004 } 1005 1006 /* 1007 * If the AGI buffer is non-NULL, then we were unable to get an 1008 * inode in one operation. We need to commit the current 1009 * transaction and call xfs_ialloc() again. It is guaranteed 1010 * to succeed the second time. 1011 */ 1012 if (ialloc_context) { 1013 /* 1014 * Normally, xfs_trans_commit releases all the locks. 1015 * We call bhold to hang on to the ialloc_context across 1016 * the commit. Holding this buffer prevents any other 1017 * processes from doing any allocations in this 1018 * allocation group. 1019 */ 1020 xfs_trans_bhold(tp, ialloc_context); 1021 1022 /* 1023 * We want the quota changes to be associated with the next 1024 * transaction, NOT this one. So, detach the dqinfo from this 1025 * and attach it to the next transaction. 1026 */ 1027 dqinfo = NULL; 1028 tflags = 0; 1029 if (tp->t_dqinfo) { 1030 dqinfo = (void *)tp->t_dqinfo; 1031 tp->t_dqinfo = NULL; 1032 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1033 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1034 } 1035 1036 code = xfs_trans_roll(&tp); 1037 if (committed != NULL) 1038 *committed = 1; 1039 1040 /* 1041 * Re-attach the quota info that we detached from prev trx. 1042 */ 1043 if (dqinfo) { 1044 tp->t_dqinfo = dqinfo; 1045 tp->t_flags |= tflags; 1046 } 1047 1048 if (code) { 1049 xfs_buf_relse(ialloc_context); 1050 *tpp = tp; 1051 *ipp = NULL; 1052 return code; 1053 } 1054 xfs_trans_bjoin(tp, ialloc_context); 1055 1056 /* 1057 * Call ialloc again. Since we've locked out all 1058 * other allocations in this allocation group, 1059 * this call should always succeed. 1060 */ 1061 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1062 &ialloc_context, &ip); 1063 1064 /* 1065 * If we get an error at this point, return to the caller 1066 * so that the current transaction can be aborted. 1067 */ 1068 if (code) { 1069 *tpp = tp; 1070 *ipp = NULL; 1071 return code; 1072 } 1073 ASSERT(!ialloc_context && ip); 1074 1075 } else { 1076 if (committed != NULL) 1077 *committed = 0; 1078 } 1079 1080 *ipp = ip; 1081 *tpp = tp; 1082 1083 return 0; 1084 } 1085 1086 /* 1087 * Decrement the link count on an inode & log the change. If this causes the 1088 * link count to go to zero, move the inode to AGI unlinked list so that it can 1089 * be freed when the last active reference goes away via xfs_inactive(). 1090 */ 1091 static int /* error */ 1092 xfs_droplink( 1093 xfs_trans_t *tp, 1094 xfs_inode_t *ip) 1095 { 1096 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1097 1098 drop_nlink(VFS_I(ip)); 1099 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1100 1101 if (VFS_I(ip)->i_nlink) 1102 return 0; 1103 1104 return xfs_iunlink(tp, ip); 1105 } 1106 1107 /* 1108 * Increment the link count on an inode & log the change. 1109 */ 1110 static int 1111 xfs_bumplink( 1112 xfs_trans_t *tp, 1113 xfs_inode_t *ip) 1114 { 1115 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1116 1117 ASSERT(ip->i_d.di_version > 1); 1118 inc_nlink(VFS_I(ip)); 1119 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1120 return 0; 1121 } 1122 1123 int 1124 xfs_create( 1125 xfs_inode_t *dp, 1126 struct xfs_name *name, 1127 umode_t mode, 1128 dev_t rdev, 1129 xfs_inode_t **ipp) 1130 { 1131 int is_dir = S_ISDIR(mode); 1132 struct xfs_mount *mp = dp->i_mount; 1133 struct xfs_inode *ip = NULL; 1134 struct xfs_trans *tp = NULL; 1135 int error; 1136 struct xfs_defer_ops dfops; 1137 xfs_fsblock_t first_block; 1138 bool unlock_dp_on_error = false; 1139 prid_t prid; 1140 struct xfs_dquot *udqp = NULL; 1141 struct xfs_dquot *gdqp = NULL; 1142 struct xfs_dquot *pdqp = NULL; 1143 struct xfs_trans_res *tres; 1144 uint resblks; 1145 1146 trace_xfs_create(dp, name); 1147 1148 if (XFS_FORCED_SHUTDOWN(mp)) 1149 return -EIO; 1150 1151 prid = xfs_get_initial_prid(dp); 1152 1153 /* 1154 * Make sure that we have allocated dquot(s) on disk. 1155 */ 1156 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1157 xfs_kgid_to_gid(current_fsgid()), prid, 1158 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1159 &udqp, &gdqp, &pdqp); 1160 if (error) 1161 return error; 1162 1163 if (is_dir) { 1164 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1165 tres = &M_RES(mp)->tr_mkdir; 1166 } else { 1167 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1168 tres = &M_RES(mp)->tr_create; 1169 } 1170 1171 /* 1172 * Initially assume that the file does not exist and 1173 * reserve the resources for that case. If that is not 1174 * the case we'll drop the one we have and get a more 1175 * appropriate transaction later. 1176 */ 1177 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1178 if (error == -ENOSPC) { 1179 /* flush outstanding delalloc blocks and retry */ 1180 xfs_flush_inodes(mp); 1181 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1182 } 1183 if (error) 1184 goto out_release_inode; 1185 1186 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1187 unlock_dp_on_error = true; 1188 1189 xfs_defer_init(&dfops, &first_block); 1190 1191 /* 1192 * Reserve disk quota and the inode. 1193 */ 1194 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1195 pdqp, resblks, 1, 0); 1196 if (error) 1197 goto out_trans_cancel; 1198 1199 /* 1200 * A newly created regular or special file just has one directory 1201 * entry pointing to them, but a directory also the "." entry 1202 * pointing to itself. 1203 */ 1204 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, prid, &ip, 1205 NULL); 1206 if (error) 1207 goto out_trans_cancel; 1208 1209 /* 1210 * Now we join the directory inode to the transaction. We do not do it 1211 * earlier because xfs_dir_ialloc might commit the previous transaction 1212 * (and release all the locks). An error from here on will result in 1213 * the transaction cancel unlocking dp so don't do it explicitly in the 1214 * error path. 1215 */ 1216 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1217 unlock_dp_on_error = false; 1218 1219 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1220 &first_block, &dfops, resblks ? 1221 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1222 if (error) { 1223 ASSERT(error != -ENOSPC); 1224 goto out_trans_cancel; 1225 } 1226 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1227 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1228 1229 if (is_dir) { 1230 error = xfs_dir_init(tp, ip, dp); 1231 if (error) 1232 goto out_bmap_cancel; 1233 1234 error = xfs_bumplink(tp, dp); 1235 if (error) 1236 goto out_bmap_cancel; 1237 } 1238 1239 /* 1240 * If this is a synchronous mount, make sure that the 1241 * create transaction goes to disk before returning to 1242 * the user. 1243 */ 1244 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1245 xfs_trans_set_sync(tp); 1246 1247 /* 1248 * Attach the dquot(s) to the inodes and modify them incore. 1249 * These ids of the inode couldn't have changed since the new 1250 * inode has been locked ever since it was created. 1251 */ 1252 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1253 1254 error = xfs_defer_finish(&tp, &dfops); 1255 if (error) 1256 goto out_bmap_cancel; 1257 1258 error = xfs_trans_commit(tp); 1259 if (error) 1260 goto out_release_inode; 1261 1262 xfs_qm_dqrele(udqp); 1263 xfs_qm_dqrele(gdqp); 1264 xfs_qm_dqrele(pdqp); 1265 1266 *ipp = ip; 1267 return 0; 1268 1269 out_bmap_cancel: 1270 xfs_defer_cancel(&dfops); 1271 out_trans_cancel: 1272 xfs_trans_cancel(tp); 1273 out_release_inode: 1274 /* 1275 * Wait until after the current transaction is aborted to finish the 1276 * setup of the inode and release the inode. This prevents recursive 1277 * transactions and deadlocks from xfs_inactive. 1278 */ 1279 if (ip) { 1280 xfs_finish_inode_setup(ip); 1281 IRELE(ip); 1282 } 1283 1284 xfs_qm_dqrele(udqp); 1285 xfs_qm_dqrele(gdqp); 1286 xfs_qm_dqrele(pdqp); 1287 1288 if (unlock_dp_on_error) 1289 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1290 return error; 1291 } 1292 1293 int 1294 xfs_create_tmpfile( 1295 struct xfs_inode *dp, 1296 struct dentry *dentry, 1297 umode_t mode, 1298 struct xfs_inode **ipp) 1299 { 1300 struct xfs_mount *mp = dp->i_mount; 1301 struct xfs_inode *ip = NULL; 1302 struct xfs_trans *tp = NULL; 1303 int error; 1304 prid_t prid; 1305 struct xfs_dquot *udqp = NULL; 1306 struct xfs_dquot *gdqp = NULL; 1307 struct xfs_dquot *pdqp = NULL; 1308 struct xfs_trans_res *tres; 1309 uint resblks; 1310 1311 if (XFS_FORCED_SHUTDOWN(mp)) 1312 return -EIO; 1313 1314 prid = xfs_get_initial_prid(dp); 1315 1316 /* 1317 * Make sure that we have allocated dquot(s) on disk. 1318 */ 1319 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1320 xfs_kgid_to_gid(current_fsgid()), prid, 1321 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1322 &udqp, &gdqp, &pdqp); 1323 if (error) 1324 return error; 1325 1326 resblks = XFS_IALLOC_SPACE_RES(mp); 1327 tres = &M_RES(mp)->tr_create_tmpfile; 1328 1329 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1330 if (error) 1331 goto out_release_inode; 1332 1333 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1334 pdqp, resblks, 1, 0); 1335 if (error) 1336 goto out_trans_cancel; 1337 1338 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, prid, &ip, NULL); 1339 if (error) 1340 goto out_trans_cancel; 1341 1342 if (mp->m_flags & XFS_MOUNT_WSYNC) 1343 xfs_trans_set_sync(tp); 1344 1345 /* 1346 * Attach the dquot(s) to the inodes and modify them incore. 1347 * These ids of the inode couldn't have changed since the new 1348 * inode has been locked ever since it was created. 1349 */ 1350 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1351 1352 error = xfs_iunlink(tp, ip); 1353 if (error) 1354 goto out_trans_cancel; 1355 1356 error = xfs_trans_commit(tp); 1357 if (error) 1358 goto out_release_inode; 1359 1360 xfs_qm_dqrele(udqp); 1361 xfs_qm_dqrele(gdqp); 1362 xfs_qm_dqrele(pdqp); 1363 1364 *ipp = ip; 1365 return 0; 1366 1367 out_trans_cancel: 1368 xfs_trans_cancel(tp); 1369 out_release_inode: 1370 /* 1371 * Wait until after the current transaction is aborted to finish the 1372 * setup of the inode and release the inode. This prevents recursive 1373 * transactions and deadlocks from xfs_inactive. 1374 */ 1375 if (ip) { 1376 xfs_finish_inode_setup(ip); 1377 IRELE(ip); 1378 } 1379 1380 xfs_qm_dqrele(udqp); 1381 xfs_qm_dqrele(gdqp); 1382 xfs_qm_dqrele(pdqp); 1383 1384 return error; 1385 } 1386 1387 int 1388 xfs_link( 1389 xfs_inode_t *tdp, 1390 xfs_inode_t *sip, 1391 struct xfs_name *target_name) 1392 { 1393 xfs_mount_t *mp = tdp->i_mount; 1394 xfs_trans_t *tp; 1395 int error; 1396 struct xfs_defer_ops dfops; 1397 xfs_fsblock_t first_block; 1398 int resblks; 1399 1400 trace_xfs_link(tdp, target_name); 1401 1402 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1403 1404 if (XFS_FORCED_SHUTDOWN(mp)) 1405 return -EIO; 1406 1407 error = xfs_qm_dqattach(sip, 0); 1408 if (error) 1409 goto std_return; 1410 1411 error = xfs_qm_dqattach(tdp, 0); 1412 if (error) 1413 goto std_return; 1414 1415 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1416 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1417 if (error == -ENOSPC) { 1418 resblks = 0; 1419 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1420 } 1421 if (error) 1422 goto std_return; 1423 1424 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1425 1426 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1427 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1428 1429 /* 1430 * If we are using project inheritance, we only allow hard link 1431 * creation in our tree when the project IDs are the same; else 1432 * the tree quota mechanism could be circumvented. 1433 */ 1434 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1435 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1436 error = -EXDEV; 1437 goto error_return; 1438 } 1439 1440 if (!resblks) { 1441 error = xfs_dir_canenter(tp, tdp, target_name); 1442 if (error) 1443 goto error_return; 1444 } 1445 1446 xfs_defer_init(&dfops, &first_block); 1447 1448 /* 1449 * Handle initial link state of O_TMPFILE inode 1450 */ 1451 if (VFS_I(sip)->i_nlink == 0) { 1452 error = xfs_iunlink_remove(tp, sip); 1453 if (error) 1454 goto error_return; 1455 } 1456 1457 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1458 &first_block, &dfops, resblks); 1459 if (error) 1460 goto error_return; 1461 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1462 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1463 1464 error = xfs_bumplink(tp, sip); 1465 if (error) 1466 goto error_return; 1467 1468 /* 1469 * If this is a synchronous mount, make sure that the 1470 * link transaction goes to disk before returning to 1471 * the user. 1472 */ 1473 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1474 xfs_trans_set_sync(tp); 1475 1476 error = xfs_defer_finish(&tp, &dfops); 1477 if (error) { 1478 xfs_defer_cancel(&dfops); 1479 goto error_return; 1480 } 1481 1482 return xfs_trans_commit(tp); 1483 1484 error_return: 1485 xfs_trans_cancel(tp); 1486 std_return: 1487 return error; 1488 } 1489 1490 /* Clear the reflink flag and the cowblocks tag if possible. */ 1491 static void 1492 xfs_itruncate_clear_reflink_flags( 1493 struct xfs_inode *ip) 1494 { 1495 struct xfs_ifork *dfork; 1496 struct xfs_ifork *cfork; 1497 1498 if (!xfs_is_reflink_inode(ip)) 1499 return; 1500 dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 1501 cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK); 1502 if (dfork->if_bytes == 0 && cfork->if_bytes == 0) 1503 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 1504 if (cfork->if_bytes == 0) 1505 xfs_inode_clear_cowblocks_tag(ip); 1506 } 1507 1508 /* 1509 * Free up the underlying blocks past new_size. The new size must be smaller 1510 * than the current size. This routine can be used both for the attribute and 1511 * data fork, and does not modify the inode size, which is left to the caller. 1512 * 1513 * The transaction passed to this routine must have made a permanent log 1514 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1515 * given transaction and start new ones, so make sure everything involved in 1516 * the transaction is tidy before calling here. Some transaction will be 1517 * returned to the caller to be committed. The incoming transaction must 1518 * already include the inode, and both inode locks must be held exclusively. 1519 * The inode must also be "held" within the transaction. On return the inode 1520 * will be "held" within the returned transaction. This routine does NOT 1521 * require any disk space to be reserved for it within the transaction. 1522 * 1523 * If we get an error, we must return with the inode locked and linked into the 1524 * current transaction. This keeps things simple for the higher level code, 1525 * because it always knows that the inode is locked and held in the transaction 1526 * that returns to it whether errors occur or not. We don't mark the inode 1527 * dirty on error so that transactions can be easily aborted if possible. 1528 */ 1529 int 1530 xfs_itruncate_extents( 1531 struct xfs_trans **tpp, 1532 struct xfs_inode *ip, 1533 int whichfork, 1534 xfs_fsize_t new_size) 1535 { 1536 struct xfs_mount *mp = ip->i_mount; 1537 struct xfs_trans *tp = *tpp; 1538 struct xfs_defer_ops dfops; 1539 xfs_fsblock_t first_block; 1540 xfs_fileoff_t first_unmap_block; 1541 xfs_fileoff_t last_block; 1542 xfs_filblks_t unmap_len; 1543 int error = 0; 1544 int done = 0; 1545 1546 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1547 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1548 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1549 ASSERT(new_size <= XFS_ISIZE(ip)); 1550 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1551 ASSERT(ip->i_itemp != NULL); 1552 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1553 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1554 1555 trace_xfs_itruncate_extents_start(ip, new_size); 1556 1557 /* 1558 * Since it is possible for space to become allocated beyond 1559 * the end of the file (in a crash where the space is allocated 1560 * but the inode size is not yet updated), simply remove any 1561 * blocks which show up between the new EOF and the maximum 1562 * possible file size. If the first block to be removed is 1563 * beyond the maximum file size (ie it is the same as last_block), 1564 * then there is nothing to do. 1565 */ 1566 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1567 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1568 if (first_unmap_block == last_block) 1569 return 0; 1570 1571 ASSERT(first_unmap_block < last_block); 1572 unmap_len = last_block - first_unmap_block + 1; 1573 while (!done) { 1574 xfs_defer_init(&dfops, &first_block); 1575 error = xfs_bunmapi(tp, ip, 1576 first_unmap_block, unmap_len, 1577 xfs_bmapi_aflag(whichfork), 1578 XFS_ITRUNC_MAX_EXTENTS, 1579 &first_block, &dfops, 1580 &done); 1581 if (error) 1582 goto out_bmap_cancel; 1583 1584 /* 1585 * Duplicate the transaction that has the permanent 1586 * reservation and commit the old transaction. 1587 */ 1588 xfs_defer_ijoin(&dfops, ip); 1589 error = xfs_defer_finish(&tp, &dfops); 1590 if (error) 1591 goto out_bmap_cancel; 1592 1593 error = xfs_trans_roll_inode(&tp, ip); 1594 if (error) 1595 goto out; 1596 } 1597 1598 /* Remove all pending CoW reservations. */ 1599 error = xfs_reflink_cancel_cow_blocks(ip, &tp, first_unmap_block, 1600 last_block, true); 1601 if (error) 1602 goto out; 1603 1604 xfs_itruncate_clear_reflink_flags(ip); 1605 1606 /* 1607 * Always re-log the inode so that our permanent transaction can keep 1608 * on rolling it forward in the log. 1609 */ 1610 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1611 1612 trace_xfs_itruncate_extents_end(ip, new_size); 1613 1614 out: 1615 *tpp = tp; 1616 return error; 1617 out_bmap_cancel: 1618 /* 1619 * If the bunmapi call encounters an error, return to the caller where 1620 * the transaction can be properly aborted. We just need to make sure 1621 * we're not holding any resources that we were not when we came in. 1622 */ 1623 xfs_defer_cancel(&dfops); 1624 goto out; 1625 } 1626 1627 int 1628 xfs_release( 1629 xfs_inode_t *ip) 1630 { 1631 xfs_mount_t *mp = ip->i_mount; 1632 int error; 1633 1634 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1635 return 0; 1636 1637 /* If this is a read-only mount, don't do this (would generate I/O) */ 1638 if (mp->m_flags & XFS_MOUNT_RDONLY) 1639 return 0; 1640 1641 if (!XFS_FORCED_SHUTDOWN(mp)) { 1642 int truncated; 1643 1644 /* 1645 * If we previously truncated this file and removed old data 1646 * in the process, we want to initiate "early" writeout on 1647 * the last close. This is an attempt to combat the notorious 1648 * NULL files problem which is particularly noticeable from a 1649 * truncate down, buffered (re-)write (delalloc), followed by 1650 * a crash. What we are effectively doing here is 1651 * significantly reducing the time window where we'd otherwise 1652 * be exposed to that problem. 1653 */ 1654 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1655 if (truncated) { 1656 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1657 if (ip->i_delayed_blks > 0) { 1658 error = filemap_flush(VFS_I(ip)->i_mapping); 1659 if (error) 1660 return error; 1661 } 1662 } 1663 } 1664 1665 if (VFS_I(ip)->i_nlink == 0) 1666 return 0; 1667 1668 if (xfs_can_free_eofblocks(ip, false)) { 1669 1670 /* 1671 * Check if the inode is being opened, written and closed 1672 * frequently and we have delayed allocation blocks outstanding 1673 * (e.g. streaming writes from the NFS server), truncating the 1674 * blocks past EOF will cause fragmentation to occur. 1675 * 1676 * In this case don't do the truncation, but we have to be 1677 * careful how we detect this case. Blocks beyond EOF show up as 1678 * i_delayed_blks even when the inode is clean, so we need to 1679 * truncate them away first before checking for a dirty release. 1680 * Hence on the first dirty close we will still remove the 1681 * speculative allocation, but after that we will leave it in 1682 * place. 1683 */ 1684 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1685 return 0; 1686 /* 1687 * If we can't get the iolock just skip truncating the blocks 1688 * past EOF because we could deadlock with the mmap_sem 1689 * otherwise. We'll get another chance to drop them once the 1690 * last reference to the inode is dropped, so we'll never leak 1691 * blocks permanently. 1692 */ 1693 if (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) { 1694 error = xfs_free_eofblocks(ip); 1695 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 1696 if (error) 1697 return error; 1698 } 1699 1700 /* delalloc blocks after truncation means it really is dirty */ 1701 if (ip->i_delayed_blks) 1702 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1703 } 1704 return 0; 1705 } 1706 1707 /* 1708 * xfs_inactive_truncate 1709 * 1710 * Called to perform a truncate when an inode becomes unlinked. 1711 */ 1712 STATIC int 1713 xfs_inactive_truncate( 1714 struct xfs_inode *ip) 1715 { 1716 struct xfs_mount *mp = ip->i_mount; 1717 struct xfs_trans *tp; 1718 int error; 1719 1720 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1721 if (error) { 1722 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1723 return error; 1724 } 1725 1726 xfs_ilock(ip, XFS_ILOCK_EXCL); 1727 xfs_trans_ijoin(tp, ip, 0); 1728 1729 /* 1730 * Log the inode size first to prevent stale data exposure in the event 1731 * of a system crash before the truncate completes. See the related 1732 * comment in xfs_vn_setattr_size() for details. 1733 */ 1734 ip->i_d.di_size = 0; 1735 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1736 1737 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1738 if (error) 1739 goto error_trans_cancel; 1740 1741 ASSERT(ip->i_d.di_nextents == 0); 1742 1743 error = xfs_trans_commit(tp); 1744 if (error) 1745 goto error_unlock; 1746 1747 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1748 return 0; 1749 1750 error_trans_cancel: 1751 xfs_trans_cancel(tp); 1752 error_unlock: 1753 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1754 return error; 1755 } 1756 1757 /* 1758 * xfs_inactive_ifree() 1759 * 1760 * Perform the inode free when an inode is unlinked. 1761 */ 1762 STATIC int 1763 xfs_inactive_ifree( 1764 struct xfs_inode *ip) 1765 { 1766 struct xfs_defer_ops dfops; 1767 xfs_fsblock_t first_block; 1768 struct xfs_mount *mp = ip->i_mount; 1769 struct xfs_trans *tp; 1770 int error; 1771 1772 /* 1773 * We try to use a per-AG reservation for any block needed by the finobt 1774 * tree, but as the finobt feature predates the per-AG reservation 1775 * support a degraded file system might not have enough space for the 1776 * reservation at mount time. In that case try to dip into the reserved 1777 * pool and pray. 1778 * 1779 * Send a warning if the reservation does happen to fail, as the inode 1780 * now remains allocated and sits on the unlinked list until the fs is 1781 * repaired. 1782 */ 1783 if (unlikely(mp->m_inotbt_nores)) { 1784 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1785 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, 1786 &tp); 1787 } else { 1788 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp); 1789 } 1790 if (error) { 1791 if (error == -ENOSPC) { 1792 xfs_warn_ratelimited(mp, 1793 "Failed to remove inode(s) from unlinked list. " 1794 "Please free space, unmount and run xfs_repair."); 1795 } else { 1796 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1797 } 1798 return error; 1799 } 1800 1801 xfs_ilock(ip, XFS_ILOCK_EXCL); 1802 xfs_trans_ijoin(tp, ip, 0); 1803 1804 xfs_defer_init(&dfops, &first_block); 1805 error = xfs_ifree(tp, ip, &dfops); 1806 if (error) { 1807 /* 1808 * If we fail to free the inode, shut down. The cancel 1809 * might do that, we need to make sure. Otherwise the 1810 * inode might be lost for a long time or forever. 1811 */ 1812 if (!XFS_FORCED_SHUTDOWN(mp)) { 1813 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1814 __func__, error); 1815 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1816 } 1817 xfs_trans_cancel(tp); 1818 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1819 return error; 1820 } 1821 1822 /* 1823 * Credit the quota account(s). The inode is gone. 1824 */ 1825 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1826 1827 /* 1828 * Just ignore errors at this point. There is nothing we can do except 1829 * to try to keep going. Make sure it's not a silent error. 1830 */ 1831 error = xfs_defer_finish(&tp, &dfops); 1832 if (error) { 1833 xfs_notice(mp, "%s: xfs_defer_finish returned error %d", 1834 __func__, error); 1835 xfs_defer_cancel(&dfops); 1836 } 1837 error = xfs_trans_commit(tp); 1838 if (error) 1839 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1840 __func__, error); 1841 1842 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1843 return 0; 1844 } 1845 1846 /* 1847 * xfs_inactive 1848 * 1849 * This is called when the vnode reference count for the vnode 1850 * goes to zero. If the file has been unlinked, then it must 1851 * now be truncated. Also, we clear all of the read-ahead state 1852 * kept for the inode here since the file is now closed. 1853 */ 1854 void 1855 xfs_inactive( 1856 xfs_inode_t *ip) 1857 { 1858 struct xfs_mount *mp; 1859 int error; 1860 int truncate = 0; 1861 1862 /* 1863 * If the inode is already free, then there can be nothing 1864 * to clean up here. 1865 */ 1866 if (VFS_I(ip)->i_mode == 0) { 1867 ASSERT(ip->i_df.if_real_bytes == 0); 1868 ASSERT(ip->i_df.if_broot_bytes == 0); 1869 return; 1870 } 1871 1872 mp = ip->i_mount; 1873 ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY)); 1874 1875 /* If this is a read-only mount, don't do this (would generate I/O) */ 1876 if (mp->m_flags & XFS_MOUNT_RDONLY) 1877 return; 1878 1879 if (VFS_I(ip)->i_nlink != 0) { 1880 /* 1881 * force is true because we are evicting an inode from the 1882 * cache. Post-eof blocks must be freed, lest we end up with 1883 * broken free space accounting. 1884 * 1885 * Note: don't bother with iolock here since lockdep complains 1886 * about acquiring it in reclaim context. We have the only 1887 * reference to the inode at this point anyways. 1888 */ 1889 if (xfs_can_free_eofblocks(ip, true)) 1890 xfs_free_eofblocks(ip); 1891 1892 return; 1893 } 1894 1895 if (S_ISREG(VFS_I(ip)->i_mode) && 1896 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1897 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1898 truncate = 1; 1899 1900 error = xfs_qm_dqattach(ip, 0); 1901 if (error) 1902 return; 1903 1904 if (S_ISLNK(VFS_I(ip)->i_mode)) 1905 error = xfs_inactive_symlink(ip); 1906 else if (truncate) 1907 error = xfs_inactive_truncate(ip); 1908 if (error) 1909 return; 1910 1911 /* 1912 * If there are attributes associated with the file then blow them away 1913 * now. The code calls a routine that recursively deconstructs the 1914 * attribute fork. If also blows away the in-core attribute fork. 1915 */ 1916 if (XFS_IFORK_Q(ip)) { 1917 error = xfs_attr_inactive(ip); 1918 if (error) 1919 return; 1920 } 1921 1922 ASSERT(!ip->i_afp); 1923 ASSERT(ip->i_d.di_anextents == 0); 1924 ASSERT(ip->i_d.di_forkoff == 0); 1925 1926 /* 1927 * Free the inode. 1928 */ 1929 error = xfs_inactive_ifree(ip); 1930 if (error) 1931 return; 1932 1933 /* 1934 * Release the dquots held by inode, if any. 1935 */ 1936 xfs_qm_dqdetach(ip); 1937 } 1938 1939 /* 1940 * This is called when the inode's link count goes to 0 or we are creating a 1941 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1942 * set to true as the link count is dropped to zero by the VFS after we've 1943 * created the file successfully, so we have to add it to the unlinked list 1944 * while the link count is non-zero. 1945 * 1946 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1947 * list when the inode is freed. 1948 */ 1949 STATIC int 1950 xfs_iunlink( 1951 struct xfs_trans *tp, 1952 struct xfs_inode *ip) 1953 { 1954 xfs_mount_t *mp = tp->t_mountp; 1955 xfs_agi_t *agi; 1956 xfs_dinode_t *dip; 1957 xfs_buf_t *agibp; 1958 xfs_buf_t *ibp; 1959 xfs_agino_t agino; 1960 short bucket_index; 1961 int offset; 1962 int error; 1963 1964 ASSERT(VFS_I(ip)->i_mode != 0); 1965 1966 /* 1967 * Get the agi buffer first. It ensures lock ordering 1968 * on the list. 1969 */ 1970 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1971 if (error) 1972 return error; 1973 agi = XFS_BUF_TO_AGI(agibp); 1974 1975 /* 1976 * Get the index into the agi hash table for the 1977 * list this inode will go on. 1978 */ 1979 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1980 ASSERT(agino != 0); 1981 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1982 ASSERT(agi->agi_unlinked[bucket_index]); 1983 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1984 1985 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1986 /* 1987 * There is already another inode in the bucket we need 1988 * to add ourselves to. Add us at the front of the list. 1989 * Here we put the head pointer into our next pointer, 1990 * and then we fall through to point the head at us. 1991 */ 1992 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1993 0, 0); 1994 if (error) 1995 return error; 1996 1997 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1998 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1999 offset = ip->i_imap.im_boffset + 2000 offsetof(xfs_dinode_t, di_next_unlinked); 2001 2002 /* need to recalc the inode CRC if appropriate */ 2003 xfs_dinode_calc_crc(mp, dip); 2004 2005 xfs_trans_inode_buf(tp, ibp); 2006 xfs_trans_log_buf(tp, ibp, offset, 2007 (offset + sizeof(xfs_agino_t) - 1)); 2008 xfs_inobp_check(mp, ibp); 2009 } 2010 2011 /* 2012 * Point the bucket head pointer at the inode being inserted. 2013 */ 2014 ASSERT(agino != 0); 2015 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 2016 offset = offsetof(xfs_agi_t, agi_unlinked) + 2017 (sizeof(xfs_agino_t) * bucket_index); 2018 xfs_trans_log_buf(tp, agibp, offset, 2019 (offset + sizeof(xfs_agino_t) - 1)); 2020 return 0; 2021 } 2022 2023 /* 2024 * Pull the on-disk inode from the AGI unlinked list. 2025 */ 2026 STATIC int 2027 xfs_iunlink_remove( 2028 xfs_trans_t *tp, 2029 xfs_inode_t *ip) 2030 { 2031 xfs_ino_t next_ino; 2032 xfs_mount_t *mp; 2033 xfs_agi_t *agi; 2034 xfs_dinode_t *dip; 2035 xfs_buf_t *agibp; 2036 xfs_buf_t *ibp; 2037 xfs_agnumber_t agno; 2038 xfs_agino_t agino; 2039 xfs_agino_t next_agino; 2040 xfs_buf_t *last_ibp; 2041 xfs_dinode_t *last_dip = NULL; 2042 short bucket_index; 2043 int offset, last_offset = 0; 2044 int error; 2045 2046 mp = tp->t_mountp; 2047 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2048 2049 /* 2050 * Get the agi buffer first. It ensures lock ordering 2051 * on the list. 2052 */ 2053 error = xfs_read_agi(mp, tp, agno, &agibp); 2054 if (error) 2055 return error; 2056 2057 agi = XFS_BUF_TO_AGI(agibp); 2058 2059 /* 2060 * Get the index into the agi hash table for the 2061 * list this inode will go on. 2062 */ 2063 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2064 ASSERT(agino != 0); 2065 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2066 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2067 ASSERT(agi->agi_unlinked[bucket_index]); 2068 2069 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2070 /* 2071 * We're at the head of the list. Get the inode's on-disk 2072 * buffer to see if there is anyone after us on the list. 2073 * Only modify our next pointer if it is not already NULLAGINO. 2074 * This saves us the overhead of dealing with the buffer when 2075 * there is no need to change it. 2076 */ 2077 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2078 0, 0); 2079 if (error) { 2080 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2081 __func__, error); 2082 return error; 2083 } 2084 next_agino = be32_to_cpu(dip->di_next_unlinked); 2085 ASSERT(next_agino != 0); 2086 if (next_agino != NULLAGINO) { 2087 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2088 offset = ip->i_imap.im_boffset + 2089 offsetof(xfs_dinode_t, di_next_unlinked); 2090 2091 /* need to recalc the inode CRC if appropriate */ 2092 xfs_dinode_calc_crc(mp, dip); 2093 2094 xfs_trans_inode_buf(tp, ibp); 2095 xfs_trans_log_buf(tp, ibp, offset, 2096 (offset + sizeof(xfs_agino_t) - 1)); 2097 xfs_inobp_check(mp, ibp); 2098 } else { 2099 xfs_trans_brelse(tp, ibp); 2100 } 2101 /* 2102 * Point the bucket head pointer at the next inode. 2103 */ 2104 ASSERT(next_agino != 0); 2105 ASSERT(next_agino != agino); 2106 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2107 offset = offsetof(xfs_agi_t, agi_unlinked) + 2108 (sizeof(xfs_agino_t) * bucket_index); 2109 xfs_trans_log_buf(tp, agibp, offset, 2110 (offset + sizeof(xfs_agino_t) - 1)); 2111 } else { 2112 /* 2113 * We need to search the list for the inode being freed. 2114 */ 2115 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2116 last_ibp = NULL; 2117 while (next_agino != agino) { 2118 struct xfs_imap imap; 2119 2120 if (last_ibp) 2121 xfs_trans_brelse(tp, last_ibp); 2122 2123 imap.im_blkno = 0; 2124 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2125 2126 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2127 if (error) { 2128 xfs_warn(mp, 2129 "%s: xfs_imap returned error %d.", 2130 __func__, error); 2131 return error; 2132 } 2133 2134 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2135 &last_ibp, 0, 0); 2136 if (error) { 2137 xfs_warn(mp, 2138 "%s: xfs_imap_to_bp returned error %d.", 2139 __func__, error); 2140 return error; 2141 } 2142 2143 last_offset = imap.im_boffset; 2144 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2145 ASSERT(next_agino != NULLAGINO); 2146 ASSERT(next_agino != 0); 2147 } 2148 2149 /* 2150 * Now last_ibp points to the buffer previous to us on the 2151 * unlinked list. Pull us from the list. 2152 */ 2153 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2154 0, 0); 2155 if (error) { 2156 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2157 __func__, error); 2158 return error; 2159 } 2160 next_agino = be32_to_cpu(dip->di_next_unlinked); 2161 ASSERT(next_agino != 0); 2162 ASSERT(next_agino != agino); 2163 if (next_agino != NULLAGINO) { 2164 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2165 offset = ip->i_imap.im_boffset + 2166 offsetof(xfs_dinode_t, di_next_unlinked); 2167 2168 /* need to recalc the inode CRC if appropriate */ 2169 xfs_dinode_calc_crc(mp, dip); 2170 2171 xfs_trans_inode_buf(tp, ibp); 2172 xfs_trans_log_buf(tp, ibp, offset, 2173 (offset + sizeof(xfs_agino_t) - 1)); 2174 xfs_inobp_check(mp, ibp); 2175 } else { 2176 xfs_trans_brelse(tp, ibp); 2177 } 2178 /* 2179 * Point the previous inode on the list to the next inode. 2180 */ 2181 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2182 ASSERT(next_agino != 0); 2183 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2184 2185 /* need to recalc the inode CRC if appropriate */ 2186 xfs_dinode_calc_crc(mp, last_dip); 2187 2188 xfs_trans_inode_buf(tp, last_ibp); 2189 xfs_trans_log_buf(tp, last_ibp, offset, 2190 (offset + sizeof(xfs_agino_t) - 1)); 2191 xfs_inobp_check(mp, last_ibp); 2192 } 2193 return 0; 2194 } 2195 2196 /* 2197 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2198 * inodes that are in memory - they all must be marked stale and attached to 2199 * the cluster buffer. 2200 */ 2201 STATIC int 2202 xfs_ifree_cluster( 2203 xfs_inode_t *free_ip, 2204 xfs_trans_t *tp, 2205 struct xfs_icluster *xic) 2206 { 2207 xfs_mount_t *mp = free_ip->i_mount; 2208 int blks_per_cluster; 2209 int inodes_per_cluster; 2210 int nbufs; 2211 int i, j; 2212 int ioffset; 2213 xfs_daddr_t blkno; 2214 xfs_buf_t *bp; 2215 xfs_inode_t *ip; 2216 xfs_inode_log_item_t *iip; 2217 xfs_log_item_t *lip; 2218 struct xfs_perag *pag; 2219 xfs_ino_t inum; 2220 2221 inum = xic->first_ino; 2222 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2223 blks_per_cluster = xfs_icluster_size_fsb(mp); 2224 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2225 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2226 2227 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2228 /* 2229 * The allocation bitmap tells us which inodes of the chunk were 2230 * physically allocated. Skip the cluster if an inode falls into 2231 * a sparse region. 2232 */ 2233 ioffset = inum - xic->first_ino; 2234 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2235 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2236 continue; 2237 } 2238 2239 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2240 XFS_INO_TO_AGBNO(mp, inum)); 2241 2242 /* 2243 * We obtain and lock the backing buffer first in the process 2244 * here, as we have to ensure that any dirty inode that we 2245 * can't get the flush lock on is attached to the buffer. 2246 * If we scan the in-memory inodes first, then buffer IO can 2247 * complete before we get a lock on it, and hence we may fail 2248 * to mark all the active inodes on the buffer stale. 2249 */ 2250 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2251 mp->m_bsize * blks_per_cluster, 2252 XBF_UNMAPPED); 2253 2254 if (!bp) 2255 return -ENOMEM; 2256 2257 /* 2258 * This buffer may not have been correctly initialised as we 2259 * didn't read it from disk. That's not important because we are 2260 * only using to mark the buffer as stale in the log, and to 2261 * attach stale cached inodes on it. That means it will never be 2262 * dispatched for IO. If it is, we want to know about it, and we 2263 * want it to fail. We can acheive this by adding a write 2264 * verifier to the buffer. 2265 */ 2266 bp->b_ops = &xfs_inode_buf_ops; 2267 2268 /* 2269 * Walk the inodes already attached to the buffer and mark them 2270 * stale. These will all have the flush locks held, so an 2271 * in-memory inode walk can't lock them. By marking them all 2272 * stale first, we will not attempt to lock them in the loop 2273 * below as the XFS_ISTALE flag will be set. 2274 */ 2275 lip = bp->b_fspriv; 2276 while (lip) { 2277 if (lip->li_type == XFS_LI_INODE) { 2278 iip = (xfs_inode_log_item_t *)lip; 2279 ASSERT(iip->ili_logged == 1); 2280 lip->li_cb = xfs_istale_done; 2281 xfs_trans_ail_copy_lsn(mp->m_ail, 2282 &iip->ili_flush_lsn, 2283 &iip->ili_item.li_lsn); 2284 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2285 } 2286 lip = lip->li_bio_list; 2287 } 2288 2289 2290 /* 2291 * For each inode in memory attempt to add it to the inode 2292 * buffer and set it up for being staled on buffer IO 2293 * completion. This is safe as we've locked out tail pushing 2294 * and flushing by locking the buffer. 2295 * 2296 * We have already marked every inode that was part of a 2297 * transaction stale above, which means there is no point in 2298 * even trying to lock them. 2299 */ 2300 for (i = 0; i < inodes_per_cluster; i++) { 2301 retry: 2302 rcu_read_lock(); 2303 ip = radix_tree_lookup(&pag->pag_ici_root, 2304 XFS_INO_TO_AGINO(mp, (inum + i))); 2305 2306 /* Inode not in memory, nothing to do */ 2307 if (!ip) { 2308 rcu_read_unlock(); 2309 continue; 2310 } 2311 2312 /* 2313 * because this is an RCU protected lookup, we could 2314 * find a recently freed or even reallocated inode 2315 * during the lookup. We need to check under the 2316 * i_flags_lock for a valid inode here. Skip it if it 2317 * is not valid, the wrong inode or stale. 2318 */ 2319 spin_lock(&ip->i_flags_lock); 2320 if (ip->i_ino != inum + i || 2321 __xfs_iflags_test(ip, XFS_ISTALE)) { 2322 spin_unlock(&ip->i_flags_lock); 2323 rcu_read_unlock(); 2324 continue; 2325 } 2326 spin_unlock(&ip->i_flags_lock); 2327 2328 /* 2329 * Don't try to lock/unlock the current inode, but we 2330 * _cannot_ skip the other inodes that we did not find 2331 * in the list attached to the buffer and are not 2332 * already marked stale. If we can't lock it, back off 2333 * and retry. 2334 */ 2335 if (ip != free_ip) { 2336 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2337 rcu_read_unlock(); 2338 delay(1); 2339 goto retry; 2340 } 2341 2342 /* 2343 * Check the inode number again in case we're 2344 * racing with freeing in xfs_reclaim_inode(). 2345 * See the comments in that function for more 2346 * information as to why the initial check is 2347 * not sufficient. 2348 */ 2349 if (ip->i_ino != inum + i) { 2350 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2351 rcu_read_unlock(); 2352 continue; 2353 } 2354 } 2355 rcu_read_unlock(); 2356 2357 xfs_iflock(ip); 2358 xfs_iflags_set(ip, XFS_ISTALE); 2359 2360 /* 2361 * we don't need to attach clean inodes or those only 2362 * with unlogged changes (which we throw away, anyway). 2363 */ 2364 iip = ip->i_itemp; 2365 if (!iip || xfs_inode_clean(ip)) { 2366 ASSERT(ip != free_ip); 2367 xfs_ifunlock(ip); 2368 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2369 continue; 2370 } 2371 2372 iip->ili_last_fields = iip->ili_fields; 2373 iip->ili_fields = 0; 2374 iip->ili_fsync_fields = 0; 2375 iip->ili_logged = 1; 2376 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2377 &iip->ili_item.li_lsn); 2378 2379 xfs_buf_attach_iodone(bp, xfs_istale_done, 2380 &iip->ili_item); 2381 2382 if (ip != free_ip) 2383 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2384 } 2385 2386 xfs_trans_stale_inode_buf(tp, bp); 2387 xfs_trans_binval(tp, bp); 2388 } 2389 2390 xfs_perag_put(pag); 2391 return 0; 2392 } 2393 2394 /* 2395 * Free any local-format buffers sitting around before we reset to 2396 * extents format. 2397 */ 2398 static inline void 2399 xfs_ifree_local_data( 2400 struct xfs_inode *ip, 2401 int whichfork) 2402 { 2403 struct xfs_ifork *ifp; 2404 2405 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL) 2406 return; 2407 2408 ifp = XFS_IFORK_PTR(ip, whichfork); 2409 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork); 2410 } 2411 2412 /* 2413 * This is called to return an inode to the inode free list. 2414 * The inode should already be truncated to 0 length and have 2415 * no pages associated with it. This routine also assumes that 2416 * the inode is already a part of the transaction. 2417 * 2418 * The on-disk copy of the inode will have been added to the list 2419 * of unlinked inodes in the AGI. We need to remove the inode from 2420 * that list atomically with respect to freeing it here. 2421 */ 2422 int 2423 xfs_ifree( 2424 xfs_trans_t *tp, 2425 xfs_inode_t *ip, 2426 struct xfs_defer_ops *dfops) 2427 { 2428 int error; 2429 struct xfs_icluster xic = { 0 }; 2430 2431 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2432 ASSERT(VFS_I(ip)->i_nlink == 0); 2433 ASSERT(ip->i_d.di_nextents == 0); 2434 ASSERT(ip->i_d.di_anextents == 0); 2435 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2436 ASSERT(ip->i_d.di_nblocks == 0); 2437 2438 /* 2439 * Pull the on-disk inode from the AGI unlinked list. 2440 */ 2441 error = xfs_iunlink_remove(tp, ip); 2442 if (error) 2443 return error; 2444 2445 error = xfs_difree(tp, ip->i_ino, dfops, &xic); 2446 if (error) 2447 return error; 2448 2449 xfs_ifree_local_data(ip, XFS_DATA_FORK); 2450 xfs_ifree_local_data(ip, XFS_ATTR_FORK); 2451 2452 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2453 ip->i_d.di_flags = 0; 2454 ip->i_d.di_dmevmask = 0; 2455 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2456 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2457 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2458 /* 2459 * Bump the generation count so no one will be confused 2460 * by reincarnations of this inode. 2461 */ 2462 VFS_I(ip)->i_generation++; 2463 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2464 2465 if (xic.deleted) 2466 error = xfs_ifree_cluster(ip, tp, &xic); 2467 2468 return error; 2469 } 2470 2471 /* 2472 * This is called to unpin an inode. The caller must have the inode locked 2473 * in at least shared mode so that the buffer cannot be subsequently pinned 2474 * once someone is waiting for it to be unpinned. 2475 */ 2476 static void 2477 xfs_iunpin( 2478 struct xfs_inode *ip) 2479 { 2480 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2481 2482 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2483 2484 /* Give the log a push to start the unpinning I/O */ 2485 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2486 2487 } 2488 2489 static void 2490 __xfs_iunpin_wait( 2491 struct xfs_inode *ip) 2492 { 2493 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2494 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2495 2496 xfs_iunpin(ip); 2497 2498 do { 2499 prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); 2500 if (xfs_ipincount(ip)) 2501 io_schedule(); 2502 } while (xfs_ipincount(ip)); 2503 finish_wait(wq, &wait.wq_entry); 2504 } 2505 2506 void 2507 xfs_iunpin_wait( 2508 struct xfs_inode *ip) 2509 { 2510 if (xfs_ipincount(ip)) 2511 __xfs_iunpin_wait(ip); 2512 } 2513 2514 /* 2515 * Removing an inode from the namespace involves removing the directory entry 2516 * and dropping the link count on the inode. Removing the directory entry can 2517 * result in locking an AGF (directory blocks were freed) and removing a link 2518 * count can result in placing the inode on an unlinked list which results in 2519 * locking an AGI. 2520 * 2521 * The big problem here is that we have an ordering constraint on AGF and AGI 2522 * locking - inode allocation locks the AGI, then can allocate a new extent for 2523 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2524 * removes the inode from the unlinked list, requiring that we lock the AGI 2525 * first, and then freeing the inode can result in an inode chunk being freed 2526 * and hence freeing disk space requiring that we lock an AGF. 2527 * 2528 * Hence the ordering that is imposed by other parts of the code is AGI before 2529 * AGF. This means we cannot remove the directory entry before we drop the inode 2530 * reference count and put it on the unlinked list as this results in a lock 2531 * order of AGF then AGI, and this can deadlock against inode allocation and 2532 * freeing. Therefore we must drop the link counts before we remove the 2533 * directory entry. 2534 * 2535 * This is still safe from a transactional point of view - it is not until we 2536 * get to xfs_defer_finish() that we have the possibility of multiple 2537 * transactions in this operation. Hence as long as we remove the directory 2538 * entry and drop the link count in the first transaction of the remove 2539 * operation, there are no transactional constraints on the ordering here. 2540 */ 2541 int 2542 xfs_remove( 2543 xfs_inode_t *dp, 2544 struct xfs_name *name, 2545 xfs_inode_t *ip) 2546 { 2547 xfs_mount_t *mp = dp->i_mount; 2548 xfs_trans_t *tp = NULL; 2549 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2550 int error = 0; 2551 struct xfs_defer_ops dfops; 2552 xfs_fsblock_t first_block; 2553 uint resblks; 2554 2555 trace_xfs_remove(dp, name); 2556 2557 if (XFS_FORCED_SHUTDOWN(mp)) 2558 return -EIO; 2559 2560 error = xfs_qm_dqattach(dp, 0); 2561 if (error) 2562 goto std_return; 2563 2564 error = xfs_qm_dqattach(ip, 0); 2565 if (error) 2566 goto std_return; 2567 2568 /* 2569 * We try to get the real space reservation first, 2570 * allowing for directory btree deletion(s) implying 2571 * possible bmap insert(s). If we can't get the space 2572 * reservation then we use 0 instead, and avoid the bmap 2573 * btree insert(s) in the directory code by, if the bmap 2574 * insert tries to happen, instead trimming the LAST 2575 * block from the directory. 2576 */ 2577 resblks = XFS_REMOVE_SPACE_RES(mp); 2578 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2579 if (error == -ENOSPC) { 2580 resblks = 0; 2581 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2582 &tp); 2583 } 2584 if (error) { 2585 ASSERT(error != -ENOSPC); 2586 goto std_return; 2587 } 2588 2589 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2590 2591 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2592 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2593 2594 /* 2595 * If we're removing a directory perform some additional validation. 2596 */ 2597 if (is_dir) { 2598 ASSERT(VFS_I(ip)->i_nlink >= 2); 2599 if (VFS_I(ip)->i_nlink != 2) { 2600 error = -ENOTEMPTY; 2601 goto out_trans_cancel; 2602 } 2603 if (!xfs_dir_isempty(ip)) { 2604 error = -ENOTEMPTY; 2605 goto out_trans_cancel; 2606 } 2607 2608 /* Drop the link from ip's "..". */ 2609 error = xfs_droplink(tp, dp); 2610 if (error) 2611 goto out_trans_cancel; 2612 2613 /* Drop the "." link from ip to self. */ 2614 error = xfs_droplink(tp, ip); 2615 if (error) 2616 goto out_trans_cancel; 2617 } else { 2618 /* 2619 * When removing a non-directory we need to log the parent 2620 * inode here. For a directory this is done implicitly 2621 * by the xfs_droplink call for the ".." entry. 2622 */ 2623 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2624 } 2625 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2626 2627 /* Drop the link from dp to ip. */ 2628 error = xfs_droplink(tp, ip); 2629 if (error) 2630 goto out_trans_cancel; 2631 2632 xfs_defer_init(&dfops, &first_block); 2633 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2634 &first_block, &dfops, resblks); 2635 if (error) { 2636 ASSERT(error != -ENOENT); 2637 goto out_bmap_cancel; 2638 } 2639 2640 /* 2641 * If this is a synchronous mount, make sure that the 2642 * remove transaction goes to disk before returning to 2643 * the user. 2644 */ 2645 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2646 xfs_trans_set_sync(tp); 2647 2648 error = xfs_defer_finish(&tp, &dfops); 2649 if (error) 2650 goto out_bmap_cancel; 2651 2652 error = xfs_trans_commit(tp); 2653 if (error) 2654 goto std_return; 2655 2656 if (is_dir && xfs_inode_is_filestream(ip)) 2657 xfs_filestream_deassociate(ip); 2658 2659 return 0; 2660 2661 out_bmap_cancel: 2662 xfs_defer_cancel(&dfops); 2663 out_trans_cancel: 2664 xfs_trans_cancel(tp); 2665 std_return: 2666 return error; 2667 } 2668 2669 /* 2670 * Enter all inodes for a rename transaction into a sorted array. 2671 */ 2672 #define __XFS_SORT_INODES 5 2673 STATIC void 2674 xfs_sort_for_rename( 2675 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2676 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2677 struct xfs_inode *ip1, /* in: inode of old entry */ 2678 struct xfs_inode *ip2, /* in: inode of new entry */ 2679 struct xfs_inode *wip, /* in: whiteout inode */ 2680 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2681 int *num_inodes) /* in/out: inodes in array */ 2682 { 2683 int i, j; 2684 2685 ASSERT(*num_inodes == __XFS_SORT_INODES); 2686 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2687 2688 /* 2689 * i_tab contains a list of pointers to inodes. We initialize 2690 * the table here & we'll sort it. We will then use it to 2691 * order the acquisition of the inode locks. 2692 * 2693 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2694 */ 2695 i = 0; 2696 i_tab[i++] = dp1; 2697 i_tab[i++] = dp2; 2698 i_tab[i++] = ip1; 2699 if (ip2) 2700 i_tab[i++] = ip2; 2701 if (wip) 2702 i_tab[i++] = wip; 2703 *num_inodes = i; 2704 2705 /* 2706 * Sort the elements via bubble sort. (Remember, there are at 2707 * most 5 elements to sort, so this is adequate.) 2708 */ 2709 for (i = 0; i < *num_inodes; i++) { 2710 for (j = 1; j < *num_inodes; j++) { 2711 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2712 struct xfs_inode *temp = i_tab[j]; 2713 i_tab[j] = i_tab[j-1]; 2714 i_tab[j-1] = temp; 2715 } 2716 } 2717 } 2718 } 2719 2720 static int 2721 xfs_finish_rename( 2722 struct xfs_trans *tp, 2723 struct xfs_defer_ops *dfops) 2724 { 2725 int error; 2726 2727 /* 2728 * If this is a synchronous mount, make sure that the rename transaction 2729 * goes to disk before returning to the user. 2730 */ 2731 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2732 xfs_trans_set_sync(tp); 2733 2734 error = xfs_defer_finish(&tp, dfops); 2735 if (error) { 2736 xfs_defer_cancel(dfops); 2737 xfs_trans_cancel(tp); 2738 return error; 2739 } 2740 2741 return xfs_trans_commit(tp); 2742 } 2743 2744 /* 2745 * xfs_cross_rename() 2746 * 2747 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2748 */ 2749 STATIC int 2750 xfs_cross_rename( 2751 struct xfs_trans *tp, 2752 struct xfs_inode *dp1, 2753 struct xfs_name *name1, 2754 struct xfs_inode *ip1, 2755 struct xfs_inode *dp2, 2756 struct xfs_name *name2, 2757 struct xfs_inode *ip2, 2758 struct xfs_defer_ops *dfops, 2759 xfs_fsblock_t *first_block, 2760 int spaceres) 2761 { 2762 int error = 0; 2763 int ip1_flags = 0; 2764 int ip2_flags = 0; 2765 int dp2_flags = 0; 2766 2767 /* Swap inode number for dirent in first parent */ 2768 error = xfs_dir_replace(tp, dp1, name1, 2769 ip2->i_ino, 2770 first_block, dfops, spaceres); 2771 if (error) 2772 goto out_trans_abort; 2773 2774 /* Swap inode number for dirent in second parent */ 2775 error = xfs_dir_replace(tp, dp2, name2, 2776 ip1->i_ino, 2777 first_block, dfops, spaceres); 2778 if (error) 2779 goto out_trans_abort; 2780 2781 /* 2782 * If we're renaming one or more directories across different parents, 2783 * update the respective ".." entries (and link counts) to match the new 2784 * parents. 2785 */ 2786 if (dp1 != dp2) { 2787 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2788 2789 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2790 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2791 dp1->i_ino, first_block, 2792 dfops, spaceres); 2793 if (error) 2794 goto out_trans_abort; 2795 2796 /* transfer ip2 ".." reference to dp1 */ 2797 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2798 error = xfs_droplink(tp, dp2); 2799 if (error) 2800 goto out_trans_abort; 2801 error = xfs_bumplink(tp, dp1); 2802 if (error) 2803 goto out_trans_abort; 2804 } 2805 2806 /* 2807 * Although ip1 isn't changed here, userspace needs 2808 * to be warned about the change, so that applications 2809 * relying on it (like backup ones), will properly 2810 * notify the change 2811 */ 2812 ip1_flags |= XFS_ICHGTIME_CHG; 2813 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2814 } 2815 2816 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2817 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2818 dp2->i_ino, first_block, 2819 dfops, spaceres); 2820 if (error) 2821 goto out_trans_abort; 2822 2823 /* transfer ip1 ".." reference to dp2 */ 2824 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2825 error = xfs_droplink(tp, dp1); 2826 if (error) 2827 goto out_trans_abort; 2828 error = xfs_bumplink(tp, dp2); 2829 if (error) 2830 goto out_trans_abort; 2831 } 2832 2833 /* 2834 * Although ip2 isn't changed here, userspace needs 2835 * to be warned about the change, so that applications 2836 * relying on it (like backup ones), will properly 2837 * notify the change 2838 */ 2839 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2840 ip2_flags |= XFS_ICHGTIME_CHG; 2841 } 2842 } 2843 2844 if (ip1_flags) { 2845 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2846 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2847 } 2848 if (ip2_flags) { 2849 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2850 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2851 } 2852 if (dp2_flags) { 2853 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2854 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2855 } 2856 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2857 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2858 return xfs_finish_rename(tp, dfops); 2859 2860 out_trans_abort: 2861 xfs_defer_cancel(dfops); 2862 xfs_trans_cancel(tp); 2863 return error; 2864 } 2865 2866 /* 2867 * xfs_rename_alloc_whiteout() 2868 * 2869 * Return a referenced, unlinked, unlocked inode that that can be used as a 2870 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2871 * crash between allocating the inode and linking it into the rename transaction 2872 * recovery will free the inode and we won't leak it. 2873 */ 2874 static int 2875 xfs_rename_alloc_whiteout( 2876 struct xfs_inode *dp, 2877 struct xfs_inode **wip) 2878 { 2879 struct xfs_inode *tmpfile; 2880 int error; 2881 2882 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2883 if (error) 2884 return error; 2885 2886 /* 2887 * Prepare the tmpfile inode as if it were created through the VFS. 2888 * Otherwise, the link increment paths will complain about nlink 0->1. 2889 * Drop the link count as done by d_tmpfile(), complete the inode setup 2890 * and flag it as linkable. 2891 */ 2892 drop_nlink(VFS_I(tmpfile)); 2893 xfs_setup_iops(tmpfile); 2894 xfs_finish_inode_setup(tmpfile); 2895 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2896 2897 *wip = tmpfile; 2898 return 0; 2899 } 2900 2901 /* 2902 * xfs_rename 2903 */ 2904 int 2905 xfs_rename( 2906 struct xfs_inode *src_dp, 2907 struct xfs_name *src_name, 2908 struct xfs_inode *src_ip, 2909 struct xfs_inode *target_dp, 2910 struct xfs_name *target_name, 2911 struct xfs_inode *target_ip, 2912 unsigned int flags) 2913 { 2914 struct xfs_mount *mp = src_dp->i_mount; 2915 struct xfs_trans *tp; 2916 struct xfs_defer_ops dfops; 2917 xfs_fsblock_t first_block; 2918 struct xfs_inode *wip = NULL; /* whiteout inode */ 2919 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2920 int num_inodes = __XFS_SORT_INODES; 2921 bool new_parent = (src_dp != target_dp); 2922 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2923 int spaceres; 2924 int error; 2925 2926 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2927 2928 if ((flags & RENAME_EXCHANGE) && !target_ip) 2929 return -EINVAL; 2930 2931 /* 2932 * If we are doing a whiteout operation, allocate the whiteout inode 2933 * we will be placing at the target and ensure the type is set 2934 * appropriately. 2935 */ 2936 if (flags & RENAME_WHITEOUT) { 2937 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2938 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2939 if (error) 2940 return error; 2941 2942 /* setup target dirent info as whiteout */ 2943 src_name->type = XFS_DIR3_FT_CHRDEV; 2944 } 2945 2946 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2947 inodes, &num_inodes); 2948 2949 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2950 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2951 if (error == -ENOSPC) { 2952 spaceres = 0; 2953 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2954 &tp); 2955 } 2956 if (error) 2957 goto out_release_wip; 2958 2959 /* 2960 * Attach the dquots to the inodes 2961 */ 2962 error = xfs_qm_vop_rename_dqattach(inodes); 2963 if (error) 2964 goto out_trans_cancel; 2965 2966 /* 2967 * Lock all the participating inodes. Depending upon whether 2968 * the target_name exists in the target directory, and 2969 * whether the target directory is the same as the source 2970 * directory, we can lock from 2 to 4 inodes. 2971 */ 2972 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2973 2974 /* 2975 * Join all the inodes to the transaction. From this point on, 2976 * we can rely on either trans_commit or trans_cancel to unlock 2977 * them. 2978 */ 2979 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 2980 if (new_parent) 2981 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 2982 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2983 if (target_ip) 2984 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2985 if (wip) 2986 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2987 2988 /* 2989 * If we are using project inheritance, we only allow renames 2990 * into our tree when the project IDs are the same; else the 2991 * tree quota mechanism would be circumvented. 2992 */ 2993 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2994 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2995 error = -EXDEV; 2996 goto out_trans_cancel; 2997 } 2998 2999 xfs_defer_init(&dfops, &first_block); 3000 3001 /* RENAME_EXCHANGE is unique from here on. */ 3002 if (flags & RENAME_EXCHANGE) 3003 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 3004 target_dp, target_name, target_ip, 3005 &dfops, &first_block, spaceres); 3006 3007 /* 3008 * Set up the target. 3009 */ 3010 if (target_ip == NULL) { 3011 /* 3012 * If there's no space reservation, check the entry will 3013 * fit before actually inserting it. 3014 */ 3015 if (!spaceres) { 3016 error = xfs_dir_canenter(tp, target_dp, target_name); 3017 if (error) 3018 goto out_trans_cancel; 3019 } 3020 /* 3021 * If target does not exist and the rename crosses 3022 * directories, adjust the target directory link count 3023 * to account for the ".." reference from the new entry. 3024 */ 3025 error = xfs_dir_createname(tp, target_dp, target_name, 3026 src_ip->i_ino, &first_block, 3027 &dfops, spaceres); 3028 if (error) 3029 goto out_bmap_cancel; 3030 3031 xfs_trans_ichgtime(tp, target_dp, 3032 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3033 3034 if (new_parent && src_is_directory) { 3035 error = xfs_bumplink(tp, target_dp); 3036 if (error) 3037 goto out_bmap_cancel; 3038 } 3039 } else { /* target_ip != NULL */ 3040 /* 3041 * If target exists and it's a directory, check that both 3042 * target and source are directories and that target can be 3043 * destroyed, or that neither is a directory. 3044 */ 3045 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 3046 /* 3047 * Make sure target dir is empty. 3048 */ 3049 if (!(xfs_dir_isempty(target_ip)) || 3050 (VFS_I(target_ip)->i_nlink > 2)) { 3051 error = -EEXIST; 3052 goto out_trans_cancel; 3053 } 3054 } 3055 3056 /* 3057 * Link the source inode under the target name. 3058 * If the source inode is a directory and we are moving 3059 * it across directories, its ".." entry will be 3060 * inconsistent until we replace that down below. 3061 * 3062 * In case there is already an entry with the same 3063 * name at the destination directory, remove it first. 3064 */ 3065 error = xfs_dir_replace(tp, target_dp, target_name, 3066 src_ip->i_ino, 3067 &first_block, &dfops, spaceres); 3068 if (error) 3069 goto out_bmap_cancel; 3070 3071 xfs_trans_ichgtime(tp, target_dp, 3072 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3073 3074 /* 3075 * Decrement the link count on the target since the target 3076 * dir no longer points to it. 3077 */ 3078 error = xfs_droplink(tp, target_ip); 3079 if (error) 3080 goto out_bmap_cancel; 3081 3082 if (src_is_directory) { 3083 /* 3084 * Drop the link from the old "." entry. 3085 */ 3086 error = xfs_droplink(tp, target_ip); 3087 if (error) 3088 goto out_bmap_cancel; 3089 } 3090 } /* target_ip != NULL */ 3091 3092 /* 3093 * Remove the source. 3094 */ 3095 if (new_parent && src_is_directory) { 3096 /* 3097 * Rewrite the ".." entry to point to the new 3098 * directory. 3099 */ 3100 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3101 target_dp->i_ino, 3102 &first_block, &dfops, spaceres); 3103 ASSERT(error != -EEXIST); 3104 if (error) 3105 goto out_bmap_cancel; 3106 } 3107 3108 /* 3109 * We always want to hit the ctime on the source inode. 3110 * 3111 * This isn't strictly required by the standards since the source 3112 * inode isn't really being changed, but old unix file systems did 3113 * it and some incremental backup programs won't work without it. 3114 */ 3115 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3116 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3117 3118 /* 3119 * Adjust the link count on src_dp. This is necessary when 3120 * renaming a directory, either within one parent when 3121 * the target existed, or across two parent directories. 3122 */ 3123 if (src_is_directory && (new_parent || target_ip != NULL)) { 3124 3125 /* 3126 * Decrement link count on src_directory since the 3127 * entry that's moved no longer points to it. 3128 */ 3129 error = xfs_droplink(tp, src_dp); 3130 if (error) 3131 goto out_bmap_cancel; 3132 } 3133 3134 /* 3135 * For whiteouts, we only need to update the source dirent with the 3136 * inode number of the whiteout inode rather than removing it 3137 * altogether. 3138 */ 3139 if (wip) { 3140 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3141 &first_block, &dfops, spaceres); 3142 } else 3143 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3144 &first_block, &dfops, spaceres); 3145 if (error) 3146 goto out_bmap_cancel; 3147 3148 /* 3149 * For whiteouts, we need to bump the link count on the whiteout inode. 3150 * This means that failures all the way up to this point leave the inode 3151 * on the unlinked list and so cleanup is a simple matter of dropping 3152 * the remaining reference to it. If we fail here after bumping the link 3153 * count, we're shutting down the filesystem so we'll never see the 3154 * intermediate state on disk. 3155 */ 3156 if (wip) { 3157 ASSERT(VFS_I(wip)->i_nlink == 0); 3158 error = xfs_bumplink(tp, wip); 3159 if (error) 3160 goto out_bmap_cancel; 3161 error = xfs_iunlink_remove(tp, wip); 3162 if (error) 3163 goto out_bmap_cancel; 3164 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3165 3166 /* 3167 * Now we have a real link, clear the "I'm a tmpfile" state 3168 * flag from the inode so it doesn't accidentally get misused in 3169 * future. 3170 */ 3171 VFS_I(wip)->i_state &= ~I_LINKABLE; 3172 } 3173 3174 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3175 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3176 if (new_parent) 3177 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3178 3179 error = xfs_finish_rename(tp, &dfops); 3180 if (wip) 3181 IRELE(wip); 3182 return error; 3183 3184 out_bmap_cancel: 3185 xfs_defer_cancel(&dfops); 3186 out_trans_cancel: 3187 xfs_trans_cancel(tp); 3188 out_release_wip: 3189 if (wip) 3190 IRELE(wip); 3191 return error; 3192 } 3193 3194 STATIC int 3195 xfs_iflush_cluster( 3196 struct xfs_inode *ip, 3197 struct xfs_buf *bp) 3198 { 3199 struct xfs_mount *mp = ip->i_mount; 3200 struct xfs_perag *pag; 3201 unsigned long first_index, mask; 3202 unsigned long inodes_per_cluster; 3203 int cilist_size; 3204 struct xfs_inode **cilist; 3205 struct xfs_inode *cip; 3206 int nr_found; 3207 int clcount = 0; 3208 int bufwasdelwri; 3209 int i; 3210 3211 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3212 3213 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3214 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3215 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3216 if (!cilist) 3217 goto out_put; 3218 3219 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3220 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3221 rcu_read_lock(); 3222 /* really need a gang lookup range call here */ 3223 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3224 first_index, inodes_per_cluster); 3225 if (nr_found == 0) 3226 goto out_free; 3227 3228 for (i = 0; i < nr_found; i++) { 3229 cip = cilist[i]; 3230 if (cip == ip) 3231 continue; 3232 3233 /* 3234 * because this is an RCU protected lookup, we could find a 3235 * recently freed or even reallocated inode during the lookup. 3236 * We need to check under the i_flags_lock for a valid inode 3237 * here. Skip it if it is not valid or the wrong inode. 3238 */ 3239 spin_lock(&cip->i_flags_lock); 3240 if (!cip->i_ino || 3241 __xfs_iflags_test(cip, XFS_ISTALE)) { 3242 spin_unlock(&cip->i_flags_lock); 3243 continue; 3244 } 3245 3246 /* 3247 * Once we fall off the end of the cluster, no point checking 3248 * any more inodes in the list because they will also all be 3249 * outside the cluster. 3250 */ 3251 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3252 spin_unlock(&cip->i_flags_lock); 3253 break; 3254 } 3255 spin_unlock(&cip->i_flags_lock); 3256 3257 /* 3258 * Do an un-protected check to see if the inode is dirty and 3259 * is a candidate for flushing. These checks will be repeated 3260 * later after the appropriate locks are acquired. 3261 */ 3262 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3263 continue; 3264 3265 /* 3266 * Try to get locks. If any are unavailable or it is pinned, 3267 * then this inode cannot be flushed and is skipped. 3268 */ 3269 3270 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3271 continue; 3272 if (!xfs_iflock_nowait(cip)) { 3273 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3274 continue; 3275 } 3276 if (xfs_ipincount(cip)) { 3277 xfs_ifunlock(cip); 3278 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3279 continue; 3280 } 3281 3282 3283 /* 3284 * Check the inode number again, just to be certain we are not 3285 * racing with freeing in xfs_reclaim_inode(). See the comments 3286 * in that function for more information as to why the initial 3287 * check is not sufficient. 3288 */ 3289 if (!cip->i_ino) { 3290 xfs_ifunlock(cip); 3291 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3292 continue; 3293 } 3294 3295 /* 3296 * arriving here means that this inode can be flushed. First 3297 * re-check that it's dirty before flushing. 3298 */ 3299 if (!xfs_inode_clean(cip)) { 3300 int error; 3301 error = xfs_iflush_int(cip, bp); 3302 if (error) { 3303 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3304 goto cluster_corrupt_out; 3305 } 3306 clcount++; 3307 } else { 3308 xfs_ifunlock(cip); 3309 } 3310 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3311 } 3312 3313 if (clcount) { 3314 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3315 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3316 } 3317 3318 out_free: 3319 rcu_read_unlock(); 3320 kmem_free(cilist); 3321 out_put: 3322 xfs_perag_put(pag); 3323 return 0; 3324 3325 3326 cluster_corrupt_out: 3327 /* 3328 * Corruption detected in the clustering loop. Invalidate the 3329 * inode buffer and shut down the filesystem. 3330 */ 3331 rcu_read_unlock(); 3332 /* 3333 * Clean up the buffer. If it was delwri, just release it -- 3334 * brelse can handle it with no problems. If not, shut down the 3335 * filesystem before releasing the buffer. 3336 */ 3337 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3338 if (bufwasdelwri) 3339 xfs_buf_relse(bp); 3340 3341 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3342 3343 if (!bufwasdelwri) { 3344 /* 3345 * Just like incore_relse: if we have b_iodone functions, 3346 * mark the buffer as an error and call them. Otherwise 3347 * mark it as stale and brelse. 3348 */ 3349 if (bp->b_iodone) { 3350 bp->b_flags &= ~XBF_DONE; 3351 xfs_buf_stale(bp); 3352 xfs_buf_ioerror(bp, -EIO); 3353 xfs_buf_ioend(bp); 3354 } else { 3355 xfs_buf_stale(bp); 3356 xfs_buf_relse(bp); 3357 } 3358 } 3359 3360 /* 3361 * Unlocks the flush lock 3362 */ 3363 xfs_iflush_abort(cip, false); 3364 kmem_free(cilist); 3365 xfs_perag_put(pag); 3366 return -EFSCORRUPTED; 3367 } 3368 3369 /* 3370 * Flush dirty inode metadata into the backing buffer. 3371 * 3372 * The caller must have the inode lock and the inode flush lock held. The 3373 * inode lock will still be held upon return to the caller, and the inode 3374 * flush lock will be released after the inode has reached the disk. 3375 * 3376 * The caller must write out the buffer returned in *bpp and release it. 3377 */ 3378 int 3379 xfs_iflush( 3380 struct xfs_inode *ip, 3381 struct xfs_buf **bpp) 3382 { 3383 struct xfs_mount *mp = ip->i_mount; 3384 struct xfs_buf *bp = NULL; 3385 struct xfs_dinode *dip; 3386 int error; 3387 3388 XFS_STATS_INC(mp, xs_iflush_count); 3389 3390 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3391 ASSERT(xfs_isiflocked(ip)); 3392 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3393 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3394 3395 *bpp = NULL; 3396 3397 xfs_iunpin_wait(ip); 3398 3399 /* 3400 * For stale inodes we cannot rely on the backing buffer remaining 3401 * stale in cache for the remaining life of the stale inode and so 3402 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3403 * inodes below. We have to check this after ensuring the inode is 3404 * unpinned so that it is safe to reclaim the stale inode after the 3405 * flush call. 3406 */ 3407 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3408 xfs_ifunlock(ip); 3409 return 0; 3410 } 3411 3412 /* 3413 * This may have been unpinned because the filesystem is shutting 3414 * down forcibly. If that's the case we must not write this inode 3415 * to disk, because the log record didn't make it to disk. 3416 * 3417 * We also have to remove the log item from the AIL in this case, 3418 * as we wait for an empty AIL as part of the unmount process. 3419 */ 3420 if (XFS_FORCED_SHUTDOWN(mp)) { 3421 error = -EIO; 3422 goto abort_out; 3423 } 3424 3425 /* 3426 * Get the buffer containing the on-disk inode. We are doing a try-lock 3427 * operation here, so we may get an EAGAIN error. In that case, we 3428 * simply want to return with the inode still dirty. 3429 * 3430 * If we get any other error, we effectively have a corruption situation 3431 * and we cannot flush the inode, so we treat it the same as failing 3432 * xfs_iflush_int(). 3433 */ 3434 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3435 0); 3436 if (error == -EAGAIN) { 3437 xfs_ifunlock(ip); 3438 return error; 3439 } 3440 if (error) 3441 goto corrupt_out; 3442 3443 /* 3444 * First flush out the inode that xfs_iflush was called with. 3445 */ 3446 error = xfs_iflush_int(ip, bp); 3447 if (error) 3448 goto corrupt_out; 3449 3450 /* 3451 * If the buffer is pinned then push on the log now so we won't 3452 * get stuck waiting in the write for too long. 3453 */ 3454 if (xfs_buf_ispinned(bp)) 3455 xfs_log_force(mp, 0); 3456 3457 /* 3458 * inode clustering: 3459 * see if other inodes can be gathered into this write 3460 */ 3461 error = xfs_iflush_cluster(ip, bp); 3462 if (error) 3463 goto cluster_corrupt_out; 3464 3465 *bpp = bp; 3466 return 0; 3467 3468 corrupt_out: 3469 if (bp) 3470 xfs_buf_relse(bp); 3471 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3472 cluster_corrupt_out: 3473 error = -EFSCORRUPTED; 3474 abort_out: 3475 /* 3476 * Unlocks the flush lock 3477 */ 3478 xfs_iflush_abort(ip, false); 3479 return error; 3480 } 3481 3482 STATIC int 3483 xfs_iflush_int( 3484 struct xfs_inode *ip, 3485 struct xfs_buf *bp) 3486 { 3487 struct xfs_inode_log_item *iip = ip->i_itemp; 3488 struct xfs_dinode *dip; 3489 struct xfs_mount *mp = ip->i_mount; 3490 3491 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3492 ASSERT(xfs_isiflocked(ip)); 3493 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3494 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3495 ASSERT(iip != NULL && iip->ili_fields != 0); 3496 ASSERT(ip->i_d.di_version > 1); 3497 3498 /* set *dip = inode's place in the buffer */ 3499 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3500 3501 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3502 mp, XFS_ERRTAG_IFLUSH_1)) { 3503 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3504 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3505 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3506 goto corrupt_out; 3507 } 3508 if (S_ISREG(VFS_I(ip)->i_mode)) { 3509 if (XFS_TEST_ERROR( 3510 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3511 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3512 mp, XFS_ERRTAG_IFLUSH_3)) { 3513 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3514 "%s: Bad regular inode %Lu, ptr 0x%p", 3515 __func__, ip->i_ino, ip); 3516 goto corrupt_out; 3517 } 3518 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3519 if (XFS_TEST_ERROR( 3520 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3521 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3522 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3523 mp, XFS_ERRTAG_IFLUSH_4)) { 3524 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3525 "%s: Bad directory inode %Lu, ptr 0x%p", 3526 __func__, ip->i_ino, ip); 3527 goto corrupt_out; 3528 } 3529 } 3530 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3531 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) { 3532 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3533 "%s: detected corrupt incore inode %Lu, " 3534 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3535 __func__, ip->i_ino, 3536 ip->i_d.di_nextents + ip->i_d.di_anextents, 3537 ip->i_d.di_nblocks, ip); 3538 goto corrupt_out; 3539 } 3540 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3541 mp, XFS_ERRTAG_IFLUSH_6)) { 3542 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3543 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3544 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3545 goto corrupt_out; 3546 } 3547 3548 /* 3549 * Inode item log recovery for v2 inodes are dependent on the 3550 * di_flushiter count for correct sequencing. We bump the flush 3551 * iteration count so we can detect flushes which postdate a log record 3552 * during recovery. This is redundant as we now log every change and 3553 * hence this can't happen but we need to still do it to ensure 3554 * backwards compatibility with old kernels that predate logging all 3555 * inode changes. 3556 */ 3557 if (ip->i_d.di_version < 3) 3558 ip->i_d.di_flushiter++; 3559 3560 /* Check the inline directory data. */ 3561 if (S_ISDIR(VFS_I(ip)->i_mode) && 3562 ip->i_d.di_format == XFS_DINODE_FMT_LOCAL && 3563 xfs_dir2_sf_verify(ip)) 3564 goto corrupt_out; 3565 3566 /* 3567 * Copy the dirty parts of the inode into the on-disk inode. We always 3568 * copy out the core of the inode, because if the inode is dirty at all 3569 * the core must be. 3570 */ 3571 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3572 3573 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3574 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3575 ip->i_d.di_flushiter = 0; 3576 3577 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3578 if (XFS_IFORK_Q(ip)) 3579 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3580 xfs_inobp_check(mp, bp); 3581 3582 /* 3583 * We've recorded everything logged in the inode, so we'd like to clear 3584 * the ili_fields bits so we don't log and flush things unnecessarily. 3585 * However, we can't stop logging all this information until the data 3586 * we've copied into the disk buffer is written to disk. If we did we 3587 * might overwrite the copy of the inode in the log with all the data 3588 * after re-logging only part of it, and in the face of a crash we 3589 * wouldn't have all the data we need to recover. 3590 * 3591 * What we do is move the bits to the ili_last_fields field. When 3592 * logging the inode, these bits are moved back to the ili_fields field. 3593 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3594 * know that the information those bits represent is permanently on 3595 * disk. As long as the flush completes before the inode is logged 3596 * again, then both ili_fields and ili_last_fields will be cleared. 3597 * 3598 * We can play with the ili_fields bits here, because the inode lock 3599 * must be held exclusively in order to set bits there and the flush 3600 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3601 * done routine can tell whether or not to look in the AIL. Also, store 3602 * the current LSN of the inode so that we can tell whether the item has 3603 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3604 * need the AIL lock, because it is a 64 bit value that cannot be read 3605 * atomically. 3606 */ 3607 iip->ili_last_fields = iip->ili_fields; 3608 iip->ili_fields = 0; 3609 iip->ili_fsync_fields = 0; 3610 iip->ili_logged = 1; 3611 3612 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3613 &iip->ili_item.li_lsn); 3614 3615 /* 3616 * Attach the function xfs_iflush_done to the inode's 3617 * buffer. This will remove the inode from the AIL 3618 * and unlock the inode's flush lock when the inode is 3619 * completely written to disk. 3620 */ 3621 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3622 3623 /* generate the checksum. */ 3624 xfs_dinode_calc_crc(mp, dip); 3625 3626 ASSERT(bp->b_fspriv != NULL); 3627 ASSERT(bp->b_iodone != NULL); 3628 return 0; 3629 3630 corrupt_out: 3631 return -EFSCORRUPTED; 3632 } 3633