1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18 #include <linux/log2.h> 19 20 #include "xfs.h" 21 #include "xfs_fs.h" 22 #include "xfs_shared.h" 23 #include "xfs_format.h" 24 #include "xfs_log_format.h" 25 #include "xfs_trans_resv.h" 26 #include "xfs_sb.h" 27 #include "xfs_mount.h" 28 #include "xfs_defer.h" 29 #include "xfs_inode.h" 30 #include "xfs_da_format.h" 31 #include "xfs_da_btree.h" 32 #include "xfs_dir2.h" 33 #include "xfs_attr_sf.h" 34 #include "xfs_attr.h" 35 #include "xfs_trans_space.h" 36 #include "xfs_trans.h" 37 #include "xfs_buf_item.h" 38 #include "xfs_inode_item.h" 39 #include "xfs_ialloc.h" 40 #include "xfs_bmap.h" 41 #include "xfs_bmap_util.h" 42 #include "xfs_error.h" 43 #include "xfs_quota.h" 44 #include "xfs_filestream.h" 45 #include "xfs_cksum.h" 46 #include "xfs_trace.h" 47 #include "xfs_icache.h" 48 #include "xfs_symlink.h" 49 #include "xfs_trans_priv.h" 50 #include "xfs_log.h" 51 #include "xfs_bmap_btree.h" 52 53 kmem_zone_t *xfs_inode_zone; 54 55 /* 56 * Used in xfs_itruncate_extents(). This is the maximum number of extents 57 * freed from a file in a single transaction. 58 */ 59 #define XFS_ITRUNC_MAX_EXTENTS 2 60 61 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); 62 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); 63 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); 64 65 /* 66 * helper function to extract extent size hint from inode 67 */ 68 xfs_extlen_t 69 xfs_get_extsz_hint( 70 struct xfs_inode *ip) 71 { 72 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 73 return ip->i_d.di_extsize; 74 if (XFS_IS_REALTIME_INODE(ip)) 75 return ip->i_mount->m_sb.sb_rextsize; 76 return 0; 77 } 78 79 /* 80 * These two are wrapper routines around the xfs_ilock() routine used to 81 * centralize some grungy code. They are used in places that wish to lock the 82 * inode solely for reading the extents. The reason these places can't just 83 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to 84 * bringing in of the extents from disk for a file in b-tree format. If the 85 * inode is in b-tree format, then we need to lock the inode exclusively until 86 * the extents are read in. Locking it exclusively all the time would limit 87 * our parallelism unnecessarily, though. What we do instead is check to see 88 * if the extents have been read in yet, and only lock the inode exclusively 89 * if they have not. 90 * 91 * The functions return a value which should be given to the corresponding 92 * xfs_iunlock() call. 93 */ 94 uint 95 xfs_ilock_data_map_shared( 96 struct xfs_inode *ip) 97 { 98 uint lock_mode = XFS_ILOCK_SHARED; 99 100 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 101 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) 102 lock_mode = XFS_ILOCK_EXCL; 103 xfs_ilock(ip, lock_mode); 104 return lock_mode; 105 } 106 107 uint 108 xfs_ilock_attr_map_shared( 109 struct xfs_inode *ip) 110 { 111 uint lock_mode = XFS_ILOCK_SHARED; 112 113 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && 114 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) 115 lock_mode = XFS_ILOCK_EXCL; 116 xfs_ilock(ip, lock_mode); 117 return lock_mode; 118 } 119 120 /* 121 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and 122 * the i_lock. This routine allows various combinations of the locks to be 123 * obtained. 124 * 125 * The 3 locks should always be ordered so that the IO lock is obtained first, 126 * the mmap lock second and the ilock last in order to prevent deadlock. 127 * 128 * Basic locking order: 129 * 130 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock 131 * 132 * mmap_sem locking order: 133 * 134 * i_iolock -> page lock -> mmap_sem 135 * mmap_sem -> i_mmap_lock -> page_lock 136 * 137 * The difference in mmap_sem locking order mean that we cannot hold the 138 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can 139 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem 140 * in get_user_pages() to map the user pages into the kernel address space for 141 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because 142 * page faults already hold the mmap_sem. 143 * 144 * Hence to serialise fully against both syscall and mmap based IO, we need to 145 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both 146 * taken in places where we need to invalidate the page cache in a race 147 * free manner (e.g. truncate, hole punch and other extent manipulation 148 * functions). 149 */ 150 void 151 xfs_ilock( 152 xfs_inode_t *ip, 153 uint lock_flags) 154 { 155 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 156 157 /* 158 * You can't set both SHARED and EXCL for the same lock, 159 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 160 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 161 */ 162 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 163 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 164 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 165 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 166 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 167 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 168 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 169 170 if (lock_flags & XFS_IOLOCK_EXCL) 171 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 172 else if (lock_flags & XFS_IOLOCK_SHARED) 173 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 174 175 if (lock_flags & XFS_MMAPLOCK_EXCL) 176 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 177 else if (lock_flags & XFS_MMAPLOCK_SHARED) 178 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 179 180 if (lock_flags & XFS_ILOCK_EXCL) 181 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 182 else if (lock_flags & XFS_ILOCK_SHARED) 183 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 184 } 185 186 /* 187 * This is just like xfs_ilock(), except that the caller 188 * is guaranteed not to sleep. It returns 1 if it gets 189 * the requested locks and 0 otherwise. If the IO lock is 190 * obtained but the inode lock cannot be, then the IO lock 191 * is dropped before returning. 192 * 193 * ip -- the inode being locked 194 * lock_flags -- this parameter indicates the inode's locks to be 195 * to be locked. See the comment for xfs_ilock() for a list 196 * of valid values. 197 */ 198 int 199 xfs_ilock_nowait( 200 xfs_inode_t *ip, 201 uint lock_flags) 202 { 203 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 204 205 /* 206 * You can't set both SHARED and EXCL for the same lock, 207 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 208 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 209 */ 210 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 211 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 212 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 213 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 214 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 215 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 216 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 217 218 if (lock_flags & XFS_IOLOCK_EXCL) { 219 if (!mrtryupdate(&ip->i_iolock)) 220 goto out; 221 } else if (lock_flags & XFS_IOLOCK_SHARED) { 222 if (!mrtryaccess(&ip->i_iolock)) 223 goto out; 224 } 225 226 if (lock_flags & XFS_MMAPLOCK_EXCL) { 227 if (!mrtryupdate(&ip->i_mmaplock)) 228 goto out_undo_iolock; 229 } else if (lock_flags & XFS_MMAPLOCK_SHARED) { 230 if (!mrtryaccess(&ip->i_mmaplock)) 231 goto out_undo_iolock; 232 } 233 234 if (lock_flags & XFS_ILOCK_EXCL) { 235 if (!mrtryupdate(&ip->i_lock)) 236 goto out_undo_mmaplock; 237 } else if (lock_flags & XFS_ILOCK_SHARED) { 238 if (!mrtryaccess(&ip->i_lock)) 239 goto out_undo_mmaplock; 240 } 241 return 1; 242 243 out_undo_mmaplock: 244 if (lock_flags & XFS_MMAPLOCK_EXCL) 245 mrunlock_excl(&ip->i_mmaplock); 246 else if (lock_flags & XFS_MMAPLOCK_SHARED) 247 mrunlock_shared(&ip->i_mmaplock); 248 out_undo_iolock: 249 if (lock_flags & XFS_IOLOCK_EXCL) 250 mrunlock_excl(&ip->i_iolock); 251 else if (lock_flags & XFS_IOLOCK_SHARED) 252 mrunlock_shared(&ip->i_iolock); 253 out: 254 return 0; 255 } 256 257 /* 258 * xfs_iunlock() is used to drop the inode locks acquired with 259 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 260 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 261 * that we know which locks to drop. 262 * 263 * ip -- the inode being unlocked 264 * lock_flags -- this parameter indicates the inode's locks to be 265 * to be unlocked. See the comment for xfs_ilock() for a list 266 * of valid values for this parameter. 267 * 268 */ 269 void 270 xfs_iunlock( 271 xfs_inode_t *ip, 272 uint lock_flags) 273 { 274 /* 275 * You can't set both SHARED and EXCL for the same lock, 276 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 277 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 278 */ 279 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 280 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 281 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 282 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 283 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 284 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 285 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 286 ASSERT(lock_flags != 0); 287 288 if (lock_flags & XFS_IOLOCK_EXCL) 289 mrunlock_excl(&ip->i_iolock); 290 else if (lock_flags & XFS_IOLOCK_SHARED) 291 mrunlock_shared(&ip->i_iolock); 292 293 if (lock_flags & XFS_MMAPLOCK_EXCL) 294 mrunlock_excl(&ip->i_mmaplock); 295 else if (lock_flags & XFS_MMAPLOCK_SHARED) 296 mrunlock_shared(&ip->i_mmaplock); 297 298 if (lock_flags & XFS_ILOCK_EXCL) 299 mrunlock_excl(&ip->i_lock); 300 else if (lock_flags & XFS_ILOCK_SHARED) 301 mrunlock_shared(&ip->i_lock); 302 303 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 304 } 305 306 /* 307 * give up write locks. the i/o lock cannot be held nested 308 * if it is being demoted. 309 */ 310 void 311 xfs_ilock_demote( 312 xfs_inode_t *ip, 313 uint lock_flags) 314 { 315 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); 316 ASSERT((lock_flags & 317 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 318 319 if (lock_flags & XFS_ILOCK_EXCL) 320 mrdemote(&ip->i_lock); 321 if (lock_flags & XFS_MMAPLOCK_EXCL) 322 mrdemote(&ip->i_mmaplock); 323 if (lock_flags & XFS_IOLOCK_EXCL) 324 mrdemote(&ip->i_iolock); 325 326 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 327 } 328 329 #if defined(DEBUG) || defined(XFS_WARN) 330 int 331 xfs_isilocked( 332 xfs_inode_t *ip, 333 uint lock_flags) 334 { 335 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 336 if (!(lock_flags & XFS_ILOCK_SHARED)) 337 return !!ip->i_lock.mr_writer; 338 return rwsem_is_locked(&ip->i_lock.mr_lock); 339 } 340 341 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { 342 if (!(lock_flags & XFS_MMAPLOCK_SHARED)) 343 return !!ip->i_mmaplock.mr_writer; 344 return rwsem_is_locked(&ip->i_mmaplock.mr_lock); 345 } 346 347 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 348 if (!(lock_flags & XFS_IOLOCK_SHARED)) 349 return !!ip->i_iolock.mr_writer; 350 return rwsem_is_locked(&ip->i_iolock.mr_lock); 351 } 352 353 ASSERT(0); 354 return 0; 355 } 356 #endif 357 358 #ifdef DEBUG 359 int xfs_locked_n; 360 int xfs_small_retries; 361 int xfs_middle_retries; 362 int xfs_lots_retries; 363 int xfs_lock_delays; 364 #endif 365 366 /* 367 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when 368 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined 369 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build 370 * errors and warnings. 371 */ 372 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) 373 static bool 374 xfs_lockdep_subclass_ok( 375 int subclass) 376 { 377 return subclass < MAX_LOCKDEP_SUBCLASSES; 378 } 379 #else 380 #define xfs_lockdep_subclass_ok(subclass) (true) 381 #endif 382 383 /* 384 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 385 * value. This can be called for any type of inode lock combination, including 386 * parent locking. Care must be taken to ensure we don't overrun the subclass 387 * storage fields in the class mask we build. 388 */ 389 static inline int 390 xfs_lock_inumorder(int lock_mode, int subclass) 391 { 392 int class = 0; 393 394 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | 395 XFS_ILOCK_RTSUM))); 396 ASSERT(xfs_lockdep_subclass_ok(subclass)); 397 398 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 399 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); 400 ASSERT(xfs_lockdep_subclass_ok(subclass + 401 XFS_IOLOCK_PARENT_VAL)); 402 class += subclass << XFS_IOLOCK_SHIFT; 403 if (lock_mode & XFS_IOLOCK_PARENT) 404 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT; 405 } 406 407 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 408 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); 409 class += subclass << XFS_MMAPLOCK_SHIFT; 410 } 411 412 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { 413 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); 414 class += subclass << XFS_ILOCK_SHIFT; 415 } 416 417 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; 418 } 419 420 /* 421 * The following routine will lock n inodes in exclusive mode. We assume the 422 * caller calls us with the inodes in i_ino order. 423 * 424 * We need to detect deadlock where an inode that we lock is in the AIL and we 425 * start waiting for another inode that is locked by a thread in a long running 426 * transaction (such as truncate). This can result in deadlock since the long 427 * running trans might need to wait for the inode we just locked in order to 428 * push the tail and free space in the log. 429 * 430 * xfs_lock_inodes() can only be used to lock one type of lock at a time - 431 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 432 * lock more than one at a time, lockdep will report false positives saying we 433 * have violated locking orders. 434 */ 435 static void 436 xfs_lock_inodes( 437 xfs_inode_t **ips, 438 int inodes, 439 uint lock_mode) 440 { 441 int attempts = 0, i, j, try_lock; 442 xfs_log_item_t *lp; 443 444 /* 445 * Currently supports between 2 and 5 inodes with exclusive locking. We 446 * support an arbitrary depth of locking here, but absolute limits on 447 * inodes depend on the the type of locking and the limits placed by 448 * lockdep annotations in xfs_lock_inumorder. These are all checked by 449 * the asserts. 450 */ 451 ASSERT(ips && inodes >= 2 && inodes <= 5); 452 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | 453 XFS_ILOCK_EXCL)); 454 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | 455 XFS_ILOCK_SHARED))); 456 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) || 457 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1); 458 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || 459 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); 460 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || 461 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); 462 463 if (lock_mode & XFS_IOLOCK_EXCL) { 464 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); 465 } else if (lock_mode & XFS_MMAPLOCK_EXCL) 466 ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); 467 468 try_lock = 0; 469 i = 0; 470 again: 471 for (; i < inodes; i++) { 472 ASSERT(ips[i]); 473 474 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 475 continue; 476 477 /* 478 * If try_lock is not set yet, make sure all locked inodes are 479 * not in the AIL. If any are, set try_lock to be used later. 480 */ 481 if (!try_lock) { 482 for (j = (i - 1); j >= 0 && !try_lock; j--) { 483 lp = (xfs_log_item_t *)ips[j]->i_itemp; 484 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) 485 try_lock++; 486 } 487 } 488 489 /* 490 * If any of the previous locks we have locked is in the AIL, 491 * we must TRY to get the second and subsequent locks. If 492 * we can't get any, we must release all we have 493 * and try again. 494 */ 495 if (!try_lock) { 496 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 497 continue; 498 } 499 500 /* try_lock means we have an inode locked that is in the AIL. */ 501 ASSERT(i != 0); 502 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 503 continue; 504 505 /* 506 * Unlock all previous guys and try again. xfs_iunlock will try 507 * to push the tail if the inode is in the AIL. 508 */ 509 attempts++; 510 for (j = i - 1; j >= 0; j--) { 511 /* 512 * Check to see if we've already unlocked this one. Not 513 * the first one going back, and the inode ptr is the 514 * same. 515 */ 516 if (j != (i - 1) && ips[j] == ips[j + 1]) 517 continue; 518 519 xfs_iunlock(ips[j], lock_mode); 520 } 521 522 if ((attempts % 5) == 0) { 523 delay(1); /* Don't just spin the CPU */ 524 #ifdef DEBUG 525 xfs_lock_delays++; 526 #endif 527 } 528 i = 0; 529 try_lock = 0; 530 goto again; 531 } 532 533 #ifdef DEBUG 534 if (attempts) { 535 if (attempts < 5) xfs_small_retries++; 536 else if (attempts < 100) xfs_middle_retries++; 537 else xfs_lots_retries++; 538 } else { 539 xfs_locked_n++; 540 } 541 #endif 542 } 543 544 /* 545 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 546 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 547 * lock more than one at a time, lockdep will report false positives saying we 548 * have violated locking orders. 549 */ 550 void 551 xfs_lock_two_inodes( 552 xfs_inode_t *ip0, 553 xfs_inode_t *ip1, 554 uint lock_mode) 555 { 556 xfs_inode_t *temp; 557 int attempts = 0; 558 xfs_log_item_t *lp; 559 560 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 561 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); 562 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 563 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) 564 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 565 566 ASSERT(ip0->i_ino != ip1->i_ino); 567 568 if (ip0->i_ino > ip1->i_ino) { 569 temp = ip0; 570 ip0 = ip1; 571 ip1 = temp; 572 } 573 574 again: 575 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); 576 577 /* 578 * If the first lock we have locked is in the AIL, we must TRY to get 579 * the second lock. If we can't get it, we must release the first one 580 * and try again. 581 */ 582 lp = (xfs_log_item_t *)ip0->i_itemp; 583 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { 584 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { 585 xfs_iunlock(ip0, lock_mode); 586 if ((++attempts % 5) == 0) 587 delay(1); /* Don't just spin the CPU */ 588 goto again; 589 } 590 } else { 591 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); 592 } 593 } 594 595 596 void 597 __xfs_iflock( 598 struct xfs_inode *ip) 599 { 600 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 601 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 602 603 do { 604 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 605 if (xfs_isiflocked(ip)) 606 io_schedule(); 607 } while (!xfs_iflock_nowait(ip)); 608 609 finish_wait(wq, &wait.wait); 610 } 611 612 STATIC uint 613 _xfs_dic2xflags( 614 __uint16_t di_flags, 615 uint64_t di_flags2, 616 bool has_attr) 617 { 618 uint flags = 0; 619 620 if (di_flags & XFS_DIFLAG_ANY) { 621 if (di_flags & XFS_DIFLAG_REALTIME) 622 flags |= FS_XFLAG_REALTIME; 623 if (di_flags & XFS_DIFLAG_PREALLOC) 624 flags |= FS_XFLAG_PREALLOC; 625 if (di_flags & XFS_DIFLAG_IMMUTABLE) 626 flags |= FS_XFLAG_IMMUTABLE; 627 if (di_flags & XFS_DIFLAG_APPEND) 628 flags |= FS_XFLAG_APPEND; 629 if (di_flags & XFS_DIFLAG_SYNC) 630 flags |= FS_XFLAG_SYNC; 631 if (di_flags & XFS_DIFLAG_NOATIME) 632 flags |= FS_XFLAG_NOATIME; 633 if (di_flags & XFS_DIFLAG_NODUMP) 634 flags |= FS_XFLAG_NODUMP; 635 if (di_flags & XFS_DIFLAG_RTINHERIT) 636 flags |= FS_XFLAG_RTINHERIT; 637 if (di_flags & XFS_DIFLAG_PROJINHERIT) 638 flags |= FS_XFLAG_PROJINHERIT; 639 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 640 flags |= FS_XFLAG_NOSYMLINKS; 641 if (di_flags & XFS_DIFLAG_EXTSIZE) 642 flags |= FS_XFLAG_EXTSIZE; 643 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 644 flags |= FS_XFLAG_EXTSZINHERIT; 645 if (di_flags & XFS_DIFLAG_NODEFRAG) 646 flags |= FS_XFLAG_NODEFRAG; 647 if (di_flags & XFS_DIFLAG_FILESTREAM) 648 flags |= FS_XFLAG_FILESTREAM; 649 } 650 651 if (di_flags2 & XFS_DIFLAG2_ANY) { 652 if (di_flags2 & XFS_DIFLAG2_DAX) 653 flags |= FS_XFLAG_DAX; 654 } 655 656 if (has_attr) 657 flags |= FS_XFLAG_HASATTR; 658 659 return flags; 660 } 661 662 uint 663 xfs_ip2xflags( 664 struct xfs_inode *ip) 665 { 666 struct xfs_icdinode *dic = &ip->i_d; 667 668 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); 669 } 670 671 /* 672 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 673 * is allowed, otherwise it has to be an exact match. If a CI match is found, 674 * ci_name->name will point to a the actual name (caller must free) or 675 * will be set to NULL if an exact match is found. 676 */ 677 int 678 xfs_lookup( 679 xfs_inode_t *dp, 680 struct xfs_name *name, 681 xfs_inode_t **ipp, 682 struct xfs_name *ci_name) 683 { 684 xfs_ino_t inum; 685 int error; 686 687 trace_xfs_lookup(dp, name); 688 689 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 690 return -EIO; 691 692 xfs_ilock(dp, XFS_IOLOCK_SHARED); 693 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 694 if (error) 695 goto out_unlock; 696 697 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 698 if (error) 699 goto out_free_name; 700 701 xfs_iunlock(dp, XFS_IOLOCK_SHARED); 702 return 0; 703 704 out_free_name: 705 if (ci_name) 706 kmem_free(ci_name->name); 707 out_unlock: 708 xfs_iunlock(dp, XFS_IOLOCK_SHARED); 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 xfs_dev_t rdev, 751 prid_t prid, 752 int okalloc, 753 xfs_buf_t **ialloc_context, 754 xfs_inode_t **ipp) 755 { 756 struct xfs_mount *mp = tp->t_mountp; 757 xfs_ino_t ino; 758 xfs_inode_t *ip; 759 uint flags; 760 int error; 761 struct timespec tv; 762 struct inode *inode; 763 764 /* 765 * Call the space management code to pick 766 * the on-disk inode to be allocated. 767 */ 768 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 769 ialloc_context, &ino); 770 if (error) 771 return error; 772 if (*ialloc_context || ino == NULLFSINO) { 773 *ipp = NULL; 774 return 0; 775 } 776 ASSERT(*ialloc_context == NULL); 777 778 /* 779 * Get the in-core inode with the lock held exclusively. 780 * This is because we're setting fields here we need 781 * to prevent others from looking at until we're done. 782 */ 783 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 784 XFS_ILOCK_EXCL, &ip); 785 if (error) 786 return error; 787 ASSERT(ip != NULL); 788 inode = VFS_I(ip); 789 790 /* 791 * We always convert v1 inodes to v2 now - we only support filesystems 792 * with >= v2 inode capability, so there is no reason for ever leaving 793 * an inode in v1 format. 794 */ 795 if (ip->i_d.di_version == 1) 796 ip->i_d.di_version = 2; 797 798 inode->i_mode = mode; 799 set_nlink(inode, nlink); 800 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 801 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 802 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_fs_time(mp->m_super); 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_crtime.t_sec = (__int32_t)tv.tv_sec; 838 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec; 839 } 840 841 842 flags = XFS_ILOG_CORE; 843 switch (mode & S_IFMT) { 844 case S_IFIFO: 845 case S_IFCHR: 846 case S_IFBLK: 847 case S_IFSOCK: 848 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 849 ip->i_df.if_u2.if_rdev = rdev; 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 uint64_t di_flags2 = 0; 857 uint di_flags = 0; 858 859 if (S_ISDIR(mode)) { 860 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 861 di_flags |= XFS_DIFLAG_RTINHERIT; 862 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 863 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 864 ip->i_d.di_extsize = pip->i_d.di_extsize; 865 } 866 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 867 di_flags |= XFS_DIFLAG_PROJINHERIT; 868 } else if (S_ISREG(mode)) { 869 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 870 di_flags |= XFS_DIFLAG_REALTIME; 871 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 872 di_flags |= XFS_DIFLAG_EXTSIZE; 873 ip->i_d.di_extsize = pip->i_d.di_extsize; 874 } 875 } 876 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 877 xfs_inherit_noatime) 878 di_flags |= XFS_DIFLAG_NOATIME; 879 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 880 xfs_inherit_nodump) 881 di_flags |= XFS_DIFLAG_NODUMP; 882 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 883 xfs_inherit_sync) 884 di_flags |= XFS_DIFLAG_SYNC; 885 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 886 xfs_inherit_nosymlinks) 887 di_flags |= XFS_DIFLAG_NOSYMLINKS; 888 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 889 xfs_inherit_nodefrag) 890 di_flags |= XFS_DIFLAG_NODEFRAG; 891 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 892 di_flags |= XFS_DIFLAG_FILESTREAM; 893 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 894 di_flags2 |= XFS_DIFLAG2_DAX; 895 896 ip->i_d.di_flags |= di_flags; 897 ip->i_d.di_flags2 |= di_flags2; 898 } 899 /* FALLTHROUGH */ 900 case S_IFLNK: 901 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 902 ip->i_df.if_flags = XFS_IFEXTENTS; 903 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 904 ip->i_df.if_u1.if_extents = NULL; 905 break; 906 default: 907 ASSERT(0); 908 } 909 /* 910 * Attribute fork settings for new inode. 911 */ 912 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 913 ip->i_d.di_anextents = 0; 914 915 /* 916 * Log the new values stuffed into the inode. 917 */ 918 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 919 xfs_trans_log_inode(tp, ip, flags); 920 921 /* now that we have an i_mode we can setup the inode structure */ 922 xfs_setup_inode(ip); 923 924 *ipp = ip; 925 return 0; 926 } 927 928 /* 929 * Allocates a new inode from disk and return a pointer to the 930 * incore copy. This routine will internally commit the current 931 * transaction and allocate a new one if the Space Manager needed 932 * to do an allocation to replenish the inode free-list. 933 * 934 * This routine is designed to be called from xfs_create and 935 * xfs_create_dir. 936 * 937 */ 938 int 939 xfs_dir_ialloc( 940 xfs_trans_t **tpp, /* input: current transaction; 941 output: may be a new transaction. */ 942 xfs_inode_t *dp, /* directory within whose allocate 943 the inode. */ 944 umode_t mode, 945 xfs_nlink_t nlink, 946 xfs_dev_t rdev, 947 prid_t prid, /* project id */ 948 int okalloc, /* ok to allocate new space */ 949 xfs_inode_t **ipp, /* pointer to inode; it will be 950 locked. */ 951 int *committed) 952 953 { 954 xfs_trans_t *tp; 955 xfs_inode_t *ip; 956 xfs_buf_t *ialloc_context = NULL; 957 int code; 958 void *dqinfo; 959 uint tflags; 960 961 tp = *tpp; 962 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 963 964 /* 965 * xfs_ialloc will return a pointer to an incore inode if 966 * the Space Manager has an available inode on the free 967 * list. Otherwise, it will do an allocation and replenish 968 * the freelist. Since we can only do one allocation per 969 * transaction without deadlocks, we will need to commit the 970 * current transaction and start a new one. We will then 971 * need to call xfs_ialloc again to get the inode. 972 * 973 * If xfs_ialloc did an allocation to replenish the freelist, 974 * it returns the bp containing the head of the freelist as 975 * ialloc_context. We will hold a lock on it across the 976 * transaction commit so that no other process can steal 977 * the inode(s) that we've just allocated. 978 */ 979 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 980 &ialloc_context, &ip); 981 982 /* 983 * Return an error if we were unable to allocate a new inode. 984 * This should only happen if we run out of space on disk or 985 * encounter a disk error. 986 */ 987 if (code) { 988 *ipp = NULL; 989 return code; 990 } 991 if (!ialloc_context && !ip) { 992 *ipp = NULL; 993 return -ENOSPC; 994 } 995 996 /* 997 * If the AGI buffer is non-NULL, then we were unable to get an 998 * inode in one operation. We need to commit the current 999 * transaction and call xfs_ialloc() again. It is guaranteed 1000 * to succeed the second time. 1001 */ 1002 if (ialloc_context) { 1003 /* 1004 * Normally, xfs_trans_commit releases all the locks. 1005 * We call bhold to hang on to the ialloc_context across 1006 * the commit. Holding this buffer prevents any other 1007 * processes from doing any allocations in this 1008 * allocation group. 1009 */ 1010 xfs_trans_bhold(tp, ialloc_context); 1011 1012 /* 1013 * We want the quota changes to be associated with the next 1014 * transaction, NOT this one. So, detach the dqinfo from this 1015 * and attach it to the next transaction. 1016 */ 1017 dqinfo = NULL; 1018 tflags = 0; 1019 if (tp->t_dqinfo) { 1020 dqinfo = (void *)tp->t_dqinfo; 1021 tp->t_dqinfo = NULL; 1022 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1023 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1024 } 1025 1026 code = xfs_trans_roll(&tp, NULL); 1027 if (committed != NULL) 1028 *committed = 1; 1029 1030 /* 1031 * Re-attach the quota info that we detached from prev trx. 1032 */ 1033 if (dqinfo) { 1034 tp->t_dqinfo = dqinfo; 1035 tp->t_flags |= tflags; 1036 } 1037 1038 if (code) { 1039 xfs_buf_relse(ialloc_context); 1040 *tpp = tp; 1041 *ipp = NULL; 1042 return code; 1043 } 1044 xfs_trans_bjoin(tp, ialloc_context); 1045 1046 /* 1047 * Call ialloc again. Since we've locked out all 1048 * other allocations in this allocation group, 1049 * this call should always succeed. 1050 */ 1051 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1052 okalloc, &ialloc_context, &ip); 1053 1054 /* 1055 * If we get an error at this point, return to the caller 1056 * so that the current transaction can be aborted. 1057 */ 1058 if (code) { 1059 *tpp = tp; 1060 *ipp = NULL; 1061 return code; 1062 } 1063 ASSERT(!ialloc_context && ip); 1064 1065 } else { 1066 if (committed != NULL) 1067 *committed = 0; 1068 } 1069 1070 *ipp = ip; 1071 *tpp = tp; 1072 1073 return 0; 1074 } 1075 1076 /* 1077 * Decrement the link count on an inode & log the change. If this causes the 1078 * link count to go to zero, move the inode to AGI unlinked list so that it can 1079 * be freed when the last active reference goes away via xfs_inactive(). 1080 */ 1081 static int /* error */ 1082 xfs_droplink( 1083 xfs_trans_t *tp, 1084 xfs_inode_t *ip) 1085 { 1086 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1087 1088 drop_nlink(VFS_I(ip)); 1089 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1090 1091 if (VFS_I(ip)->i_nlink) 1092 return 0; 1093 1094 return xfs_iunlink(tp, ip); 1095 } 1096 1097 /* 1098 * Increment the link count on an inode & log the change. 1099 */ 1100 static int 1101 xfs_bumplink( 1102 xfs_trans_t *tp, 1103 xfs_inode_t *ip) 1104 { 1105 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1106 1107 ASSERT(ip->i_d.di_version > 1); 1108 inc_nlink(VFS_I(ip)); 1109 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1110 return 0; 1111 } 1112 1113 int 1114 xfs_create( 1115 xfs_inode_t *dp, 1116 struct xfs_name *name, 1117 umode_t mode, 1118 xfs_dev_t rdev, 1119 xfs_inode_t **ipp) 1120 { 1121 int is_dir = S_ISDIR(mode); 1122 struct xfs_mount *mp = dp->i_mount; 1123 struct xfs_inode *ip = NULL; 1124 struct xfs_trans *tp = NULL; 1125 int error; 1126 struct xfs_defer_ops dfops; 1127 xfs_fsblock_t first_block; 1128 bool unlock_dp_on_error = false; 1129 prid_t prid; 1130 struct xfs_dquot *udqp = NULL; 1131 struct xfs_dquot *gdqp = NULL; 1132 struct xfs_dquot *pdqp = NULL; 1133 struct xfs_trans_res *tres; 1134 uint resblks; 1135 1136 trace_xfs_create(dp, name); 1137 1138 if (XFS_FORCED_SHUTDOWN(mp)) 1139 return -EIO; 1140 1141 prid = xfs_get_initial_prid(dp); 1142 1143 /* 1144 * Make sure that we have allocated dquot(s) on disk. 1145 */ 1146 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1147 xfs_kgid_to_gid(current_fsgid()), prid, 1148 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1149 &udqp, &gdqp, &pdqp); 1150 if (error) 1151 return error; 1152 1153 if (is_dir) { 1154 rdev = 0; 1155 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1156 tres = &M_RES(mp)->tr_mkdir; 1157 } else { 1158 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1159 tres = &M_RES(mp)->tr_create; 1160 } 1161 1162 /* 1163 * Initially assume that the file does not exist and 1164 * reserve the resources for that case. If that is not 1165 * the case we'll drop the one we have and get a more 1166 * appropriate transaction later. 1167 */ 1168 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1169 if (error == -ENOSPC) { 1170 /* flush outstanding delalloc blocks and retry */ 1171 xfs_flush_inodes(mp); 1172 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1173 } 1174 if (error == -ENOSPC) { 1175 /* No space at all so try a "no-allocation" reservation */ 1176 resblks = 0; 1177 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1178 } 1179 if (error) 1180 goto out_release_inode; 1181 1182 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | 1183 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT); 1184 unlock_dp_on_error = true; 1185 1186 xfs_defer_init(&dfops, &first_block); 1187 1188 /* 1189 * Reserve disk quota and the inode. 1190 */ 1191 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1192 pdqp, resblks, 1, 0); 1193 if (error) 1194 goto out_trans_cancel; 1195 1196 if (!resblks) { 1197 error = xfs_dir_canenter(tp, dp, name); 1198 if (error) 1199 goto out_trans_cancel; 1200 } 1201 1202 /* 1203 * A newly created regular or special file just has one directory 1204 * entry pointing to them, but a directory also the "." entry 1205 * pointing to itself. 1206 */ 1207 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1208 prid, resblks > 0, &ip, NULL); 1209 if (error) 1210 goto out_trans_cancel; 1211 1212 /* 1213 * Now we join the directory inode to the transaction. We do not do it 1214 * earlier because xfs_dir_ialloc might commit the previous transaction 1215 * (and release all the locks). An error from here on will result in 1216 * the transaction cancel unlocking dp so don't do it explicitly in the 1217 * error path. 1218 */ 1219 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1220 unlock_dp_on_error = false; 1221 1222 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1223 &first_block, &dfops, resblks ? 1224 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1225 if (error) { 1226 ASSERT(error != -ENOSPC); 1227 goto out_trans_cancel; 1228 } 1229 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1230 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1231 1232 if (is_dir) { 1233 error = xfs_dir_init(tp, ip, dp); 1234 if (error) 1235 goto out_bmap_cancel; 1236 1237 error = xfs_bumplink(tp, dp); 1238 if (error) 1239 goto out_bmap_cancel; 1240 } 1241 1242 /* 1243 * If this is a synchronous mount, make sure that the 1244 * create transaction goes to disk before returning to 1245 * the user. 1246 */ 1247 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1248 xfs_trans_set_sync(tp); 1249 1250 /* 1251 * Attach the dquot(s) to the inodes and modify them incore. 1252 * These ids of the inode couldn't have changed since the new 1253 * inode has been locked ever since it was created. 1254 */ 1255 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1256 1257 error = xfs_defer_finish(&tp, &dfops, NULL); 1258 if (error) 1259 goto out_bmap_cancel; 1260 1261 error = xfs_trans_commit(tp); 1262 if (error) 1263 goto out_release_inode; 1264 1265 xfs_qm_dqrele(udqp); 1266 xfs_qm_dqrele(gdqp); 1267 xfs_qm_dqrele(pdqp); 1268 1269 *ipp = ip; 1270 return 0; 1271 1272 out_bmap_cancel: 1273 xfs_defer_cancel(&dfops); 1274 out_trans_cancel: 1275 xfs_trans_cancel(tp); 1276 out_release_inode: 1277 /* 1278 * Wait until after the current transaction is aborted to finish the 1279 * setup of the inode and release the inode. This prevents recursive 1280 * transactions and deadlocks from xfs_inactive. 1281 */ 1282 if (ip) { 1283 xfs_finish_inode_setup(ip); 1284 IRELE(ip); 1285 } 1286 1287 xfs_qm_dqrele(udqp); 1288 xfs_qm_dqrele(gdqp); 1289 xfs_qm_dqrele(pdqp); 1290 1291 if (unlock_dp_on_error) 1292 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1293 return error; 1294 } 1295 1296 int 1297 xfs_create_tmpfile( 1298 struct xfs_inode *dp, 1299 struct dentry *dentry, 1300 umode_t mode, 1301 struct xfs_inode **ipp) 1302 { 1303 struct xfs_mount *mp = dp->i_mount; 1304 struct xfs_inode *ip = NULL; 1305 struct xfs_trans *tp = NULL; 1306 int error; 1307 prid_t prid; 1308 struct xfs_dquot *udqp = NULL; 1309 struct xfs_dquot *gdqp = NULL; 1310 struct xfs_dquot *pdqp = NULL; 1311 struct xfs_trans_res *tres; 1312 uint resblks; 1313 1314 if (XFS_FORCED_SHUTDOWN(mp)) 1315 return -EIO; 1316 1317 prid = xfs_get_initial_prid(dp); 1318 1319 /* 1320 * Make sure that we have allocated dquot(s) on disk. 1321 */ 1322 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1323 xfs_kgid_to_gid(current_fsgid()), prid, 1324 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1325 &udqp, &gdqp, &pdqp); 1326 if (error) 1327 return error; 1328 1329 resblks = XFS_IALLOC_SPACE_RES(mp); 1330 tres = &M_RES(mp)->tr_create_tmpfile; 1331 1332 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1333 if (error == -ENOSPC) { 1334 /* No space at all so try a "no-allocation" reservation */ 1335 resblks = 0; 1336 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1337 } 1338 if (error) 1339 goto out_release_inode; 1340 1341 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1342 pdqp, resblks, 1, 0); 1343 if (error) 1344 goto out_trans_cancel; 1345 1346 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1347 prid, resblks > 0, &ip, NULL); 1348 if (error) 1349 goto out_trans_cancel; 1350 1351 if (mp->m_flags & XFS_MOUNT_WSYNC) 1352 xfs_trans_set_sync(tp); 1353 1354 /* 1355 * Attach the dquot(s) to the inodes and modify them incore. 1356 * These ids of the inode couldn't have changed since the new 1357 * inode has been locked ever since it was created. 1358 */ 1359 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1360 1361 error = xfs_iunlink(tp, ip); 1362 if (error) 1363 goto out_trans_cancel; 1364 1365 error = xfs_trans_commit(tp); 1366 if (error) 1367 goto out_release_inode; 1368 1369 xfs_qm_dqrele(udqp); 1370 xfs_qm_dqrele(gdqp); 1371 xfs_qm_dqrele(pdqp); 1372 1373 *ipp = ip; 1374 return 0; 1375 1376 out_trans_cancel: 1377 xfs_trans_cancel(tp); 1378 out_release_inode: 1379 /* 1380 * Wait until after the current transaction is aborted to finish the 1381 * setup of the inode and release the inode. This prevents recursive 1382 * transactions and deadlocks from xfs_inactive. 1383 */ 1384 if (ip) { 1385 xfs_finish_inode_setup(ip); 1386 IRELE(ip); 1387 } 1388 1389 xfs_qm_dqrele(udqp); 1390 xfs_qm_dqrele(gdqp); 1391 xfs_qm_dqrele(pdqp); 1392 1393 return error; 1394 } 1395 1396 int 1397 xfs_link( 1398 xfs_inode_t *tdp, 1399 xfs_inode_t *sip, 1400 struct xfs_name *target_name) 1401 { 1402 xfs_mount_t *mp = tdp->i_mount; 1403 xfs_trans_t *tp; 1404 int error; 1405 struct xfs_defer_ops dfops; 1406 xfs_fsblock_t first_block; 1407 int resblks; 1408 1409 trace_xfs_link(tdp, target_name); 1410 1411 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1412 1413 if (XFS_FORCED_SHUTDOWN(mp)) 1414 return -EIO; 1415 1416 error = xfs_qm_dqattach(sip, 0); 1417 if (error) 1418 goto std_return; 1419 1420 error = xfs_qm_dqattach(tdp, 0); 1421 if (error) 1422 goto std_return; 1423 1424 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1425 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1426 if (error == -ENOSPC) { 1427 resblks = 0; 1428 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1429 } 1430 if (error) 1431 goto std_return; 1432 1433 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 1434 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1435 1436 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1437 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1438 1439 /* 1440 * If we are using project inheritance, we only allow hard link 1441 * creation in our tree when the project IDs are the same; else 1442 * the tree quota mechanism could be circumvented. 1443 */ 1444 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1445 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1446 error = -EXDEV; 1447 goto error_return; 1448 } 1449 1450 if (!resblks) { 1451 error = xfs_dir_canenter(tp, tdp, target_name); 1452 if (error) 1453 goto error_return; 1454 } 1455 1456 xfs_defer_init(&dfops, &first_block); 1457 1458 /* 1459 * Handle initial link state of O_TMPFILE inode 1460 */ 1461 if (VFS_I(sip)->i_nlink == 0) { 1462 error = xfs_iunlink_remove(tp, sip); 1463 if (error) 1464 goto error_return; 1465 } 1466 1467 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1468 &first_block, &dfops, resblks); 1469 if (error) 1470 goto error_return; 1471 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1472 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1473 1474 error = xfs_bumplink(tp, sip); 1475 if (error) 1476 goto error_return; 1477 1478 /* 1479 * If this is a synchronous mount, make sure that the 1480 * link transaction goes to disk before returning to 1481 * the user. 1482 */ 1483 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1484 xfs_trans_set_sync(tp); 1485 1486 error = xfs_defer_finish(&tp, &dfops, NULL); 1487 if (error) { 1488 xfs_defer_cancel(&dfops); 1489 goto error_return; 1490 } 1491 1492 return xfs_trans_commit(tp); 1493 1494 error_return: 1495 xfs_trans_cancel(tp); 1496 std_return: 1497 return error; 1498 } 1499 1500 /* 1501 * Free up the underlying blocks past new_size. The new size must be smaller 1502 * than the current size. This routine can be used both for the attribute and 1503 * data fork, and does not modify the inode size, which is left to the caller. 1504 * 1505 * The transaction passed to this routine must have made a permanent log 1506 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1507 * given transaction and start new ones, so make sure everything involved in 1508 * the transaction is tidy before calling here. Some transaction will be 1509 * returned to the caller to be committed. The incoming transaction must 1510 * already include the inode, and both inode locks must be held exclusively. 1511 * The inode must also be "held" within the transaction. On return the inode 1512 * will be "held" within the returned transaction. This routine does NOT 1513 * require any disk space to be reserved for it within the transaction. 1514 * 1515 * If we get an error, we must return with the inode locked and linked into the 1516 * current transaction. This keeps things simple for the higher level code, 1517 * because it always knows that the inode is locked and held in the transaction 1518 * that returns to it whether errors occur or not. We don't mark the inode 1519 * dirty on error so that transactions can be easily aborted if possible. 1520 */ 1521 int 1522 xfs_itruncate_extents( 1523 struct xfs_trans **tpp, 1524 struct xfs_inode *ip, 1525 int whichfork, 1526 xfs_fsize_t new_size) 1527 { 1528 struct xfs_mount *mp = ip->i_mount; 1529 struct xfs_trans *tp = *tpp; 1530 struct xfs_defer_ops dfops; 1531 xfs_fsblock_t first_block; 1532 xfs_fileoff_t first_unmap_block; 1533 xfs_fileoff_t last_block; 1534 xfs_filblks_t unmap_len; 1535 int error = 0; 1536 int done = 0; 1537 1538 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1539 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1540 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1541 ASSERT(new_size <= XFS_ISIZE(ip)); 1542 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1543 ASSERT(ip->i_itemp != NULL); 1544 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1545 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1546 1547 trace_xfs_itruncate_extents_start(ip, new_size); 1548 1549 /* 1550 * Since it is possible for space to become allocated beyond 1551 * the end of the file (in a crash where the space is allocated 1552 * but the inode size is not yet updated), simply remove any 1553 * blocks which show up between the new EOF and the maximum 1554 * possible file size. If the first block to be removed is 1555 * beyond the maximum file size (ie it is the same as last_block), 1556 * then there is nothing to do. 1557 */ 1558 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1559 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1560 if (first_unmap_block == last_block) 1561 return 0; 1562 1563 ASSERT(first_unmap_block < last_block); 1564 unmap_len = last_block - first_unmap_block + 1; 1565 while (!done) { 1566 xfs_defer_init(&dfops, &first_block); 1567 error = xfs_bunmapi(tp, ip, 1568 first_unmap_block, unmap_len, 1569 xfs_bmapi_aflag(whichfork), 1570 XFS_ITRUNC_MAX_EXTENTS, 1571 &first_block, &dfops, 1572 &done); 1573 if (error) 1574 goto out_bmap_cancel; 1575 1576 /* 1577 * Duplicate the transaction that has the permanent 1578 * reservation and commit the old transaction. 1579 */ 1580 error = xfs_defer_finish(&tp, &dfops, ip); 1581 if (error) 1582 goto out_bmap_cancel; 1583 1584 error = xfs_trans_roll(&tp, ip); 1585 if (error) 1586 goto out; 1587 } 1588 1589 /* 1590 * Always re-log the inode so that our permanent transaction can keep 1591 * on rolling it forward in the log. 1592 */ 1593 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1594 1595 trace_xfs_itruncate_extents_end(ip, new_size); 1596 1597 out: 1598 *tpp = tp; 1599 return error; 1600 out_bmap_cancel: 1601 /* 1602 * If the bunmapi call encounters an error, return to the caller where 1603 * the transaction can be properly aborted. We just need to make sure 1604 * we're not holding any resources that we were not when we came in. 1605 */ 1606 xfs_defer_cancel(&dfops); 1607 goto out; 1608 } 1609 1610 int 1611 xfs_release( 1612 xfs_inode_t *ip) 1613 { 1614 xfs_mount_t *mp = ip->i_mount; 1615 int error; 1616 1617 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1618 return 0; 1619 1620 /* If this is a read-only mount, don't do this (would generate I/O) */ 1621 if (mp->m_flags & XFS_MOUNT_RDONLY) 1622 return 0; 1623 1624 if (!XFS_FORCED_SHUTDOWN(mp)) { 1625 int truncated; 1626 1627 /* 1628 * If we previously truncated this file and removed old data 1629 * in the process, we want to initiate "early" writeout on 1630 * the last close. This is an attempt to combat the notorious 1631 * NULL files problem which is particularly noticeable from a 1632 * truncate down, buffered (re-)write (delalloc), followed by 1633 * a crash. What we are effectively doing here is 1634 * significantly reducing the time window where we'd otherwise 1635 * be exposed to that problem. 1636 */ 1637 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1638 if (truncated) { 1639 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1640 if (ip->i_delayed_blks > 0) { 1641 error = filemap_flush(VFS_I(ip)->i_mapping); 1642 if (error) 1643 return error; 1644 } 1645 } 1646 } 1647 1648 if (VFS_I(ip)->i_nlink == 0) 1649 return 0; 1650 1651 if (xfs_can_free_eofblocks(ip, false)) { 1652 1653 /* 1654 * If we can't get the iolock just skip truncating the blocks 1655 * past EOF because we could deadlock with the mmap_sem 1656 * otherwise. We'll get another chance to drop them once the 1657 * last reference to the inode is dropped, so we'll never leak 1658 * blocks permanently. 1659 * 1660 * Further, check if the inode is being opened, written and 1661 * closed frequently and we have delayed allocation blocks 1662 * outstanding (e.g. streaming writes from the NFS server), 1663 * truncating the blocks past EOF will cause fragmentation to 1664 * occur. 1665 * 1666 * In this case don't do the truncation, either, but we have to 1667 * be careful how we detect this case. Blocks beyond EOF show 1668 * up as i_delayed_blks even when the inode is clean, so we 1669 * need to truncate them away first before checking for a dirty 1670 * release. Hence on the first dirty close we will still remove 1671 * the speculative allocation, but after that we will leave it 1672 * in place. 1673 */ 1674 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1675 return 0; 1676 1677 error = xfs_free_eofblocks(mp, ip, true); 1678 if (error && error != -EAGAIN) 1679 return error; 1680 1681 /* delalloc blocks after truncation means it really is dirty */ 1682 if (ip->i_delayed_blks) 1683 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1684 } 1685 return 0; 1686 } 1687 1688 /* 1689 * xfs_inactive_truncate 1690 * 1691 * Called to perform a truncate when an inode becomes unlinked. 1692 */ 1693 STATIC int 1694 xfs_inactive_truncate( 1695 struct xfs_inode *ip) 1696 { 1697 struct xfs_mount *mp = ip->i_mount; 1698 struct xfs_trans *tp; 1699 int error; 1700 1701 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1702 if (error) { 1703 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1704 return error; 1705 } 1706 1707 xfs_ilock(ip, XFS_ILOCK_EXCL); 1708 xfs_trans_ijoin(tp, ip, 0); 1709 1710 /* 1711 * Log the inode size first to prevent stale data exposure in the event 1712 * of a system crash before the truncate completes. See the related 1713 * comment in xfs_setattr_size() for details. 1714 */ 1715 ip->i_d.di_size = 0; 1716 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1717 1718 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1719 if (error) 1720 goto error_trans_cancel; 1721 1722 ASSERT(ip->i_d.di_nextents == 0); 1723 1724 error = xfs_trans_commit(tp); 1725 if (error) 1726 goto error_unlock; 1727 1728 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1729 return 0; 1730 1731 error_trans_cancel: 1732 xfs_trans_cancel(tp); 1733 error_unlock: 1734 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1735 return error; 1736 } 1737 1738 /* 1739 * xfs_inactive_ifree() 1740 * 1741 * Perform the inode free when an inode is unlinked. 1742 */ 1743 STATIC int 1744 xfs_inactive_ifree( 1745 struct xfs_inode *ip) 1746 { 1747 struct xfs_defer_ops dfops; 1748 xfs_fsblock_t first_block; 1749 struct xfs_mount *mp = ip->i_mount; 1750 struct xfs_trans *tp; 1751 int error; 1752 1753 /* 1754 * The ifree transaction might need to allocate blocks for record 1755 * insertion to the finobt. We don't want to fail here at ENOSPC, so 1756 * allow ifree to dip into the reserved block pool if necessary. 1757 * 1758 * Freeing large sets of inodes generally means freeing inode chunks, 1759 * directory and file data blocks, so this should be relatively safe. 1760 * Only under severe circumstances should it be possible to free enough 1761 * inodes to exhaust the reserve block pool via finobt expansion while 1762 * at the same time not creating free space in the filesystem. 1763 * 1764 * Send a warning if the reservation does happen to fail, as the inode 1765 * now remains allocated and sits on the unlinked list until the fs is 1766 * repaired. 1767 */ 1768 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1769 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 1770 if (error) { 1771 if (error == -ENOSPC) { 1772 xfs_warn_ratelimited(mp, 1773 "Failed to remove inode(s) from unlinked list. " 1774 "Please free space, unmount and run xfs_repair."); 1775 } else { 1776 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1777 } 1778 return error; 1779 } 1780 1781 xfs_ilock(ip, XFS_ILOCK_EXCL); 1782 xfs_trans_ijoin(tp, ip, 0); 1783 1784 xfs_defer_init(&dfops, &first_block); 1785 error = xfs_ifree(tp, ip, &dfops); 1786 if (error) { 1787 /* 1788 * If we fail to free the inode, shut down. The cancel 1789 * might do that, we need to make sure. Otherwise the 1790 * inode might be lost for a long time or forever. 1791 */ 1792 if (!XFS_FORCED_SHUTDOWN(mp)) { 1793 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1794 __func__, error); 1795 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1796 } 1797 xfs_trans_cancel(tp); 1798 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1799 return error; 1800 } 1801 1802 /* 1803 * Credit the quota account(s). The inode is gone. 1804 */ 1805 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1806 1807 /* 1808 * Just ignore errors at this point. There is nothing we can do except 1809 * to try to keep going. Make sure it's not a silent error. 1810 */ 1811 error = xfs_defer_finish(&tp, &dfops, NULL); 1812 if (error) { 1813 xfs_notice(mp, "%s: xfs_defer_finish returned error %d", 1814 __func__, error); 1815 xfs_defer_cancel(&dfops); 1816 } 1817 error = xfs_trans_commit(tp); 1818 if (error) 1819 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1820 __func__, error); 1821 1822 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1823 return 0; 1824 } 1825 1826 /* 1827 * xfs_inactive 1828 * 1829 * This is called when the vnode reference count for the vnode 1830 * goes to zero. If the file has been unlinked, then it must 1831 * now be truncated. Also, we clear all of the read-ahead state 1832 * kept for the inode here since the file is now closed. 1833 */ 1834 void 1835 xfs_inactive( 1836 xfs_inode_t *ip) 1837 { 1838 struct xfs_mount *mp; 1839 int error; 1840 int truncate = 0; 1841 1842 /* 1843 * If the inode is already free, then there can be nothing 1844 * to clean up here. 1845 */ 1846 if (VFS_I(ip)->i_mode == 0) { 1847 ASSERT(ip->i_df.if_real_bytes == 0); 1848 ASSERT(ip->i_df.if_broot_bytes == 0); 1849 return; 1850 } 1851 1852 mp = ip->i_mount; 1853 1854 /* If this is a read-only mount, don't do this (would generate I/O) */ 1855 if (mp->m_flags & XFS_MOUNT_RDONLY) 1856 return; 1857 1858 if (VFS_I(ip)->i_nlink != 0) { 1859 /* 1860 * force is true because we are evicting an inode from the 1861 * cache. Post-eof blocks must be freed, lest we end up with 1862 * broken free space accounting. 1863 */ 1864 if (xfs_can_free_eofblocks(ip, true)) 1865 xfs_free_eofblocks(mp, ip, false); 1866 1867 return; 1868 } 1869 1870 if (S_ISREG(VFS_I(ip)->i_mode) && 1871 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1872 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1873 truncate = 1; 1874 1875 error = xfs_qm_dqattach(ip, 0); 1876 if (error) 1877 return; 1878 1879 if (S_ISLNK(VFS_I(ip)->i_mode)) 1880 error = xfs_inactive_symlink(ip); 1881 else if (truncate) 1882 error = xfs_inactive_truncate(ip); 1883 if (error) 1884 return; 1885 1886 /* 1887 * If there are attributes associated with the file then blow them away 1888 * now. The code calls a routine that recursively deconstructs the 1889 * attribute fork. If also blows away the in-core attribute fork. 1890 */ 1891 if (XFS_IFORK_Q(ip)) { 1892 error = xfs_attr_inactive(ip); 1893 if (error) 1894 return; 1895 } 1896 1897 ASSERT(!ip->i_afp); 1898 ASSERT(ip->i_d.di_anextents == 0); 1899 ASSERT(ip->i_d.di_forkoff == 0); 1900 1901 /* 1902 * Free the inode. 1903 */ 1904 error = xfs_inactive_ifree(ip); 1905 if (error) 1906 return; 1907 1908 /* 1909 * Release the dquots held by inode, if any. 1910 */ 1911 xfs_qm_dqdetach(ip); 1912 } 1913 1914 /* 1915 * This is called when the inode's link count goes to 0 or we are creating a 1916 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1917 * set to true as the link count is dropped to zero by the VFS after we've 1918 * created the file successfully, so we have to add it to the unlinked list 1919 * while the link count is non-zero. 1920 * 1921 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1922 * list when the inode is freed. 1923 */ 1924 STATIC int 1925 xfs_iunlink( 1926 struct xfs_trans *tp, 1927 struct xfs_inode *ip) 1928 { 1929 xfs_mount_t *mp = tp->t_mountp; 1930 xfs_agi_t *agi; 1931 xfs_dinode_t *dip; 1932 xfs_buf_t *agibp; 1933 xfs_buf_t *ibp; 1934 xfs_agino_t agino; 1935 short bucket_index; 1936 int offset; 1937 int error; 1938 1939 ASSERT(VFS_I(ip)->i_mode != 0); 1940 1941 /* 1942 * Get the agi buffer first. It ensures lock ordering 1943 * on the list. 1944 */ 1945 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1946 if (error) 1947 return error; 1948 agi = XFS_BUF_TO_AGI(agibp); 1949 1950 /* 1951 * Get the index into the agi hash table for the 1952 * list this inode will go on. 1953 */ 1954 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1955 ASSERT(agino != 0); 1956 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1957 ASSERT(agi->agi_unlinked[bucket_index]); 1958 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1959 1960 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1961 /* 1962 * There is already another inode in the bucket we need 1963 * to add ourselves to. Add us at the front of the list. 1964 * Here we put the head pointer into our next pointer, 1965 * and then we fall through to point the head at us. 1966 */ 1967 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1968 0, 0); 1969 if (error) 1970 return error; 1971 1972 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1973 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1974 offset = ip->i_imap.im_boffset + 1975 offsetof(xfs_dinode_t, di_next_unlinked); 1976 1977 /* need to recalc the inode CRC if appropriate */ 1978 xfs_dinode_calc_crc(mp, dip); 1979 1980 xfs_trans_inode_buf(tp, ibp); 1981 xfs_trans_log_buf(tp, ibp, offset, 1982 (offset + sizeof(xfs_agino_t) - 1)); 1983 xfs_inobp_check(mp, ibp); 1984 } 1985 1986 /* 1987 * Point the bucket head pointer at the inode being inserted. 1988 */ 1989 ASSERT(agino != 0); 1990 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1991 offset = offsetof(xfs_agi_t, agi_unlinked) + 1992 (sizeof(xfs_agino_t) * bucket_index); 1993 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 1994 xfs_trans_log_buf(tp, agibp, offset, 1995 (offset + sizeof(xfs_agino_t) - 1)); 1996 return 0; 1997 } 1998 1999 /* 2000 * Pull the on-disk inode from the AGI unlinked list. 2001 */ 2002 STATIC int 2003 xfs_iunlink_remove( 2004 xfs_trans_t *tp, 2005 xfs_inode_t *ip) 2006 { 2007 xfs_ino_t next_ino; 2008 xfs_mount_t *mp; 2009 xfs_agi_t *agi; 2010 xfs_dinode_t *dip; 2011 xfs_buf_t *agibp; 2012 xfs_buf_t *ibp; 2013 xfs_agnumber_t agno; 2014 xfs_agino_t agino; 2015 xfs_agino_t next_agino; 2016 xfs_buf_t *last_ibp; 2017 xfs_dinode_t *last_dip = NULL; 2018 short bucket_index; 2019 int offset, last_offset = 0; 2020 int error; 2021 2022 mp = tp->t_mountp; 2023 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2024 2025 /* 2026 * Get the agi buffer first. It ensures lock ordering 2027 * on the list. 2028 */ 2029 error = xfs_read_agi(mp, tp, agno, &agibp); 2030 if (error) 2031 return error; 2032 2033 agi = XFS_BUF_TO_AGI(agibp); 2034 2035 /* 2036 * Get the index into the agi hash table for the 2037 * list this inode will go on. 2038 */ 2039 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2040 ASSERT(agino != 0); 2041 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2042 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2043 ASSERT(agi->agi_unlinked[bucket_index]); 2044 2045 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2046 /* 2047 * We're at the head of the list. Get the inode's on-disk 2048 * buffer to see if there is anyone after us on the list. 2049 * Only modify our next pointer if it is not already NULLAGINO. 2050 * This saves us the overhead of dealing with the buffer when 2051 * there is no need to change it. 2052 */ 2053 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2054 0, 0); 2055 if (error) { 2056 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2057 __func__, error); 2058 return error; 2059 } 2060 next_agino = be32_to_cpu(dip->di_next_unlinked); 2061 ASSERT(next_agino != 0); 2062 if (next_agino != NULLAGINO) { 2063 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2064 offset = ip->i_imap.im_boffset + 2065 offsetof(xfs_dinode_t, di_next_unlinked); 2066 2067 /* need to recalc the inode CRC if appropriate */ 2068 xfs_dinode_calc_crc(mp, dip); 2069 2070 xfs_trans_inode_buf(tp, ibp); 2071 xfs_trans_log_buf(tp, ibp, offset, 2072 (offset + sizeof(xfs_agino_t) - 1)); 2073 xfs_inobp_check(mp, ibp); 2074 } else { 2075 xfs_trans_brelse(tp, ibp); 2076 } 2077 /* 2078 * Point the bucket head pointer at the next inode. 2079 */ 2080 ASSERT(next_agino != 0); 2081 ASSERT(next_agino != agino); 2082 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2083 offset = offsetof(xfs_agi_t, agi_unlinked) + 2084 (sizeof(xfs_agino_t) * bucket_index); 2085 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2086 xfs_trans_log_buf(tp, agibp, offset, 2087 (offset + sizeof(xfs_agino_t) - 1)); 2088 } else { 2089 /* 2090 * We need to search the list for the inode being freed. 2091 */ 2092 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2093 last_ibp = NULL; 2094 while (next_agino != agino) { 2095 struct xfs_imap imap; 2096 2097 if (last_ibp) 2098 xfs_trans_brelse(tp, last_ibp); 2099 2100 imap.im_blkno = 0; 2101 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2102 2103 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2104 if (error) { 2105 xfs_warn(mp, 2106 "%s: xfs_imap returned error %d.", 2107 __func__, error); 2108 return error; 2109 } 2110 2111 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2112 &last_ibp, 0, 0); 2113 if (error) { 2114 xfs_warn(mp, 2115 "%s: xfs_imap_to_bp returned error %d.", 2116 __func__, error); 2117 return error; 2118 } 2119 2120 last_offset = imap.im_boffset; 2121 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2122 ASSERT(next_agino != NULLAGINO); 2123 ASSERT(next_agino != 0); 2124 } 2125 2126 /* 2127 * Now last_ibp points to the buffer previous to us on the 2128 * unlinked list. Pull us from the list. 2129 */ 2130 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2131 0, 0); 2132 if (error) { 2133 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2134 __func__, error); 2135 return error; 2136 } 2137 next_agino = be32_to_cpu(dip->di_next_unlinked); 2138 ASSERT(next_agino != 0); 2139 ASSERT(next_agino != agino); 2140 if (next_agino != NULLAGINO) { 2141 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2142 offset = ip->i_imap.im_boffset + 2143 offsetof(xfs_dinode_t, di_next_unlinked); 2144 2145 /* need to recalc the inode CRC if appropriate */ 2146 xfs_dinode_calc_crc(mp, dip); 2147 2148 xfs_trans_inode_buf(tp, ibp); 2149 xfs_trans_log_buf(tp, ibp, offset, 2150 (offset + sizeof(xfs_agino_t) - 1)); 2151 xfs_inobp_check(mp, ibp); 2152 } else { 2153 xfs_trans_brelse(tp, ibp); 2154 } 2155 /* 2156 * Point the previous inode on the list to the next inode. 2157 */ 2158 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2159 ASSERT(next_agino != 0); 2160 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2161 2162 /* need to recalc the inode CRC if appropriate */ 2163 xfs_dinode_calc_crc(mp, last_dip); 2164 2165 xfs_trans_inode_buf(tp, last_ibp); 2166 xfs_trans_log_buf(tp, last_ibp, offset, 2167 (offset + sizeof(xfs_agino_t) - 1)); 2168 xfs_inobp_check(mp, last_ibp); 2169 } 2170 return 0; 2171 } 2172 2173 /* 2174 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2175 * inodes that are in memory - they all must be marked stale and attached to 2176 * the cluster buffer. 2177 */ 2178 STATIC int 2179 xfs_ifree_cluster( 2180 xfs_inode_t *free_ip, 2181 xfs_trans_t *tp, 2182 struct xfs_icluster *xic) 2183 { 2184 xfs_mount_t *mp = free_ip->i_mount; 2185 int blks_per_cluster; 2186 int inodes_per_cluster; 2187 int nbufs; 2188 int i, j; 2189 int ioffset; 2190 xfs_daddr_t blkno; 2191 xfs_buf_t *bp; 2192 xfs_inode_t *ip; 2193 xfs_inode_log_item_t *iip; 2194 xfs_log_item_t *lip; 2195 struct xfs_perag *pag; 2196 xfs_ino_t inum; 2197 2198 inum = xic->first_ino; 2199 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2200 blks_per_cluster = xfs_icluster_size_fsb(mp); 2201 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2202 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2203 2204 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2205 /* 2206 * The allocation bitmap tells us which inodes of the chunk were 2207 * physically allocated. Skip the cluster if an inode falls into 2208 * a sparse region. 2209 */ 2210 ioffset = inum - xic->first_ino; 2211 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2212 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2213 continue; 2214 } 2215 2216 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2217 XFS_INO_TO_AGBNO(mp, inum)); 2218 2219 /* 2220 * We obtain and lock the backing buffer first in the process 2221 * here, as we have to ensure that any dirty inode that we 2222 * can't get the flush lock on is attached to the buffer. 2223 * If we scan the in-memory inodes first, then buffer IO can 2224 * complete before we get a lock on it, and hence we may fail 2225 * to mark all the active inodes on the buffer stale. 2226 */ 2227 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2228 mp->m_bsize * blks_per_cluster, 2229 XBF_UNMAPPED); 2230 2231 if (!bp) 2232 return -ENOMEM; 2233 2234 /* 2235 * This buffer may not have been correctly initialised as we 2236 * didn't read it from disk. That's not important because we are 2237 * only using to mark the buffer as stale in the log, and to 2238 * attach stale cached inodes on it. That means it will never be 2239 * dispatched for IO. If it is, we want to know about it, and we 2240 * want it to fail. We can acheive this by adding a write 2241 * verifier to the buffer. 2242 */ 2243 bp->b_ops = &xfs_inode_buf_ops; 2244 2245 /* 2246 * Walk the inodes already attached to the buffer and mark them 2247 * stale. These will all have the flush locks held, so an 2248 * in-memory inode walk can't lock them. By marking them all 2249 * stale first, we will not attempt to lock them in the loop 2250 * below as the XFS_ISTALE flag will be set. 2251 */ 2252 lip = bp->b_fspriv; 2253 while (lip) { 2254 if (lip->li_type == XFS_LI_INODE) { 2255 iip = (xfs_inode_log_item_t *)lip; 2256 ASSERT(iip->ili_logged == 1); 2257 lip->li_cb = xfs_istale_done; 2258 xfs_trans_ail_copy_lsn(mp->m_ail, 2259 &iip->ili_flush_lsn, 2260 &iip->ili_item.li_lsn); 2261 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2262 } 2263 lip = lip->li_bio_list; 2264 } 2265 2266 2267 /* 2268 * For each inode in memory attempt to add it to the inode 2269 * buffer and set it up for being staled on buffer IO 2270 * completion. This is safe as we've locked out tail pushing 2271 * and flushing by locking the buffer. 2272 * 2273 * We have already marked every inode that was part of a 2274 * transaction stale above, which means there is no point in 2275 * even trying to lock them. 2276 */ 2277 for (i = 0; i < inodes_per_cluster; i++) { 2278 retry: 2279 rcu_read_lock(); 2280 ip = radix_tree_lookup(&pag->pag_ici_root, 2281 XFS_INO_TO_AGINO(mp, (inum + i))); 2282 2283 /* Inode not in memory, nothing to do */ 2284 if (!ip) { 2285 rcu_read_unlock(); 2286 continue; 2287 } 2288 2289 /* 2290 * because this is an RCU protected lookup, we could 2291 * find a recently freed or even reallocated inode 2292 * during the lookup. We need to check under the 2293 * i_flags_lock for a valid inode here. Skip it if it 2294 * is not valid, the wrong inode or stale. 2295 */ 2296 spin_lock(&ip->i_flags_lock); 2297 if (ip->i_ino != inum + i || 2298 __xfs_iflags_test(ip, XFS_ISTALE)) { 2299 spin_unlock(&ip->i_flags_lock); 2300 rcu_read_unlock(); 2301 continue; 2302 } 2303 spin_unlock(&ip->i_flags_lock); 2304 2305 /* 2306 * Don't try to lock/unlock the current inode, but we 2307 * _cannot_ skip the other inodes that we did not find 2308 * in the list attached to the buffer and are not 2309 * already marked stale. If we can't lock it, back off 2310 * and retry. 2311 */ 2312 if (ip != free_ip && 2313 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2314 rcu_read_unlock(); 2315 delay(1); 2316 goto retry; 2317 } 2318 rcu_read_unlock(); 2319 2320 xfs_iflock(ip); 2321 xfs_iflags_set(ip, XFS_ISTALE); 2322 2323 /* 2324 * we don't need to attach clean inodes or those only 2325 * with unlogged changes (which we throw away, anyway). 2326 */ 2327 iip = ip->i_itemp; 2328 if (!iip || xfs_inode_clean(ip)) { 2329 ASSERT(ip != free_ip); 2330 xfs_ifunlock(ip); 2331 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2332 continue; 2333 } 2334 2335 iip->ili_last_fields = iip->ili_fields; 2336 iip->ili_fields = 0; 2337 iip->ili_fsync_fields = 0; 2338 iip->ili_logged = 1; 2339 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2340 &iip->ili_item.li_lsn); 2341 2342 xfs_buf_attach_iodone(bp, xfs_istale_done, 2343 &iip->ili_item); 2344 2345 if (ip != free_ip) 2346 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2347 } 2348 2349 xfs_trans_stale_inode_buf(tp, bp); 2350 xfs_trans_binval(tp, bp); 2351 } 2352 2353 xfs_perag_put(pag); 2354 return 0; 2355 } 2356 2357 /* 2358 * This is called to return an inode to the inode free list. 2359 * The inode should already be truncated to 0 length and have 2360 * no pages associated with it. This routine also assumes that 2361 * the inode is already a part of the transaction. 2362 * 2363 * The on-disk copy of the inode will have been added to the list 2364 * of unlinked inodes in the AGI. We need to remove the inode from 2365 * that list atomically with respect to freeing it here. 2366 */ 2367 int 2368 xfs_ifree( 2369 xfs_trans_t *tp, 2370 xfs_inode_t *ip, 2371 struct xfs_defer_ops *dfops) 2372 { 2373 int error; 2374 struct xfs_icluster xic = { 0 }; 2375 2376 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2377 ASSERT(VFS_I(ip)->i_nlink == 0); 2378 ASSERT(ip->i_d.di_nextents == 0); 2379 ASSERT(ip->i_d.di_anextents == 0); 2380 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2381 ASSERT(ip->i_d.di_nblocks == 0); 2382 2383 /* 2384 * Pull the on-disk inode from the AGI unlinked list. 2385 */ 2386 error = xfs_iunlink_remove(tp, ip); 2387 if (error) 2388 return error; 2389 2390 error = xfs_difree(tp, ip->i_ino, dfops, &xic); 2391 if (error) 2392 return error; 2393 2394 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2395 ip->i_d.di_flags = 0; 2396 ip->i_d.di_dmevmask = 0; 2397 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2398 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2399 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2400 /* 2401 * Bump the generation count so no one will be confused 2402 * by reincarnations of this inode. 2403 */ 2404 VFS_I(ip)->i_generation++; 2405 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2406 2407 if (xic.deleted) 2408 error = xfs_ifree_cluster(ip, tp, &xic); 2409 2410 return error; 2411 } 2412 2413 /* 2414 * This is called to unpin an inode. The caller must have the inode locked 2415 * in at least shared mode so that the buffer cannot be subsequently pinned 2416 * once someone is waiting for it to be unpinned. 2417 */ 2418 static void 2419 xfs_iunpin( 2420 struct xfs_inode *ip) 2421 { 2422 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2423 2424 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2425 2426 /* Give the log a push to start the unpinning I/O */ 2427 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2428 2429 } 2430 2431 static void 2432 __xfs_iunpin_wait( 2433 struct xfs_inode *ip) 2434 { 2435 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2436 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2437 2438 xfs_iunpin(ip); 2439 2440 do { 2441 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2442 if (xfs_ipincount(ip)) 2443 io_schedule(); 2444 } while (xfs_ipincount(ip)); 2445 finish_wait(wq, &wait.wait); 2446 } 2447 2448 void 2449 xfs_iunpin_wait( 2450 struct xfs_inode *ip) 2451 { 2452 if (xfs_ipincount(ip)) 2453 __xfs_iunpin_wait(ip); 2454 } 2455 2456 /* 2457 * Removing an inode from the namespace involves removing the directory entry 2458 * and dropping the link count on the inode. Removing the directory entry can 2459 * result in locking an AGF (directory blocks were freed) and removing a link 2460 * count can result in placing the inode on an unlinked list which results in 2461 * locking an AGI. 2462 * 2463 * The big problem here is that we have an ordering constraint on AGF and AGI 2464 * locking - inode allocation locks the AGI, then can allocate a new extent for 2465 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2466 * removes the inode from the unlinked list, requiring that we lock the AGI 2467 * first, and then freeing the inode can result in an inode chunk being freed 2468 * and hence freeing disk space requiring that we lock an AGF. 2469 * 2470 * Hence the ordering that is imposed by other parts of the code is AGI before 2471 * AGF. This means we cannot remove the directory entry before we drop the inode 2472 * reference count and put it on the unlinked list as this results in a lock 2473 * order of AGF then AGI, and this can deadlock against inode allocation and 2474 * freeing. Therefore we must drop the link counts before we remove the 2475 * directory entry. 2476 * 2477 * This is still safe from a transactional point of view - it is not until we 2478 * get to xfs_defer_finish() that we have the possibility of multiple 2479 * transactions in this operation. Hence as long as we remove the directory 2480 * entry and drop the link count in the first transaction of the remove 2481 * operation, there are no transactional constraints on the ordering here. 2482 */ 2483 int 2484 xfs_remove( 2485 xfs_inode_t *dp, 2486 struct xfs_name *name, 2487 xfs_inode_t *ip) 2488 { 2489 xfs_mount_t *mp = dp->i_mount; 2490 xfs_trans_t *tp = NULL; 2491 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2492 int error = 0; 2493 struct xfs_defer_ops dfops; 2494 xfs_fsblock_t first_block; 2495 uint resblks; 2496 2497 trace_xfs_remove(dp, name); 2498 2499 if (XFS_FORCED_SHUTDOWN(mp)) 2500 return -EIO; 2501 2502 error = xfs_qm_dqattach(dp, 0); 2503 if (error) 2504 goto std_return; 2505 2506 error = xfs_qm_dqattach(ip, 0); 2507 if (error) 2508 goto std_return; 2509 2510 /* 2511 * We try to get the real space reservation first, 2512 * allowing for directory btree deletion(s) implying 2513 * possible bmap insert(s). If we can't get the space 2514 * reservation then we use 0 instead, and avoid the bmap 2515 * btree insert(s) in the directory code by, if the bmap 2516 * insert tries to happen, instead trimming the LAST 2517 * block from the directory. 2518 */ 2519 resblks = XFS_REMOVE_SPACE_RES(mp); 2520 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2521 if (error == -ENOSPC) { 2522 resblks = 0; 2523 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2524 &tp); 2525 } 2526 if (error) { 2527 ASSERT(error != -ENOSPC); 2528 goto std_return; 2529 } 2530 2531 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2532 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2533 2534 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2535 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2536 2537 /* 2538 * If we're removing a directory perform some additional validation. 2539 */ 2540 if (is_dir) { 2541 ASSERT(VFS_I(ip)->i_nlink >= 2); 2542 if (VFS_I(ip)->i_nlink != 2) { 2543 error = -ENOTEMPTY; 2544 goto out_trans_cancel; 2545 } 2546 if (!xfs_dir_isempty(ip)) { 2547 error = -ENOTEMPTY; 2548 goto out_trans_cancel; 2549 } 2550 2551 /* Drop the link from ip's "..". */ 2552 error = xfs_droplink(tp, dp); 2553 if (error) 2554 goto out_trans_cancel; 2555 2556 /* Drop the "." link from ip to self. */ 2557 error = xfs_droplink(tp, ip); 2558 if (error) 2559 goto out_trans_cancel; 2560 } else { 2561 /* 2562 * When removing a non-directory we need to log the parent 2563 * inode here. For a directory this is done implicitly 2564 * by the xfs_droplink call for the ".." entry. 2565 */ 2566 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2567 } 2568 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2569 2570 /* Drop the link from dp to ip. */ 2571 error = xfs_droplink(tp, ip); 2572 if (error) 2573 goto out_trans_cancel; 2574 2575 xfs_defer_init(&dfops, &first_block); 2576 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2577 &first_block, &dfops, resblks); 2578 if (error) { 2579 ASSERT(error != -ENOENT); 2580 goto out_bmap_cancel; 2581 } 2582 2583 /* 2584 * If this is a synchronous mount, make sure that the 2585 * remove transaction goes to disk before returning to 2586 * the user. 2587 */ 2588 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2589 xfs_trans_set_sync(tp); 2590 2591 error = xfs_defer_finish(&tp, &dfops, NULL); 2592 if (error) 2593 goto out_bmap_cancel; 2594 2595 error = xfs_trans_commit(tp); 2596 if (error) 2597 goto std_return; 2598 2599 if (is_dir && xfs_inode_is_filestream(ip)) 2600 xfs_filestream_deassociate(ip); 2601 2602 return 0; 2603 2604 out_bmap_cancel: 2605 xfs_defer_cancel(&dfops); 2606 out_trans_cancel: 2607 xfs_trans_cancel(tp); 2608 std_return: 2609 return error; 2610 } 2611 2612 /* 2613 * Enter all inodes for a rename transaction into a sorted array. 2614 */ 2615 #define __XFS_SORT_INODES 5 2616 STATIC void 2617 xfs_sort_for_rename( 2618 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2619 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2620 struct xfs_inode *ip1, /* in: inode of old entry */ 2621 struct xfs_inode *ip2, /* in: inode of new entry */ 2622 struct xfs_inode *wip, /* in: whiteout inode */ 2623 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2624 int *num_inodes) /* in/out: inodes in array */ 2625 { 2626 int i, j; 2627 2628 ASSERT(*num_inodes == __XFS_SORT_INODES); 2629 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2630 2631 /* 2632 * i_tab contains a list of pointers to inodes. We initialize 2633 * the table here & we'll sort it. We will then use it to 2634 * order the acquisition of the inode locks. 2635 * 2636 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2637 */ 2638 i = 0; 2639 i_tab[i++] = dp1; 2640 i_tab[i++] = dp2; 2641 i_tab[i++] = ip1; 2642 if (ip2) 2643 i_tab[i++] = ip2; 2644 if (wip) 2645 i_tab[i++] = wip; 2646 *num_inodes = i; 2647 2648 /* 2649 * Sort the elements via bubble sort. (Remember, there are at 2650 * most 5 elements to sort, so this is adequate.) 2651 */ 2652 for (i = 0; i < *num_inodes; i++) { 2653 for (j = 1; j < *num_inodes; j++) { 2654 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2655 struct xfs_inode *temp = i_tab[j]; 2656 i_tab[j] = i_tab[j-1]; 2657 i_tab[j-1] = temp; 2658 } 2659 } 2660 } 2661 } 2662 2663 static int 2664 xfs_finish_rename( 2665 struct xfs_trans *tp, 2666 struct xfs_defer_ops *dfops) 2667 { 2668 int error; 2669 2670 /* 2671 * If this is a synchronous mount, make sure that the rename transaction 2672 * goes to disk before returning to the user. 2673 */ 2674 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2675 xfs_trans_set_sync(tp); 2676 2677 error = xfs_defer_finish(&tp, dfops, NULL); 2678 if (error) { 2679 xfs_defer_cancel(dfops); 2680 xfs_trans_cancel(tp); 2681 return error; 2682 } 2683 2684 return xfs_trans_commit(tp); 2685 } 2686 2687 /* 2688 * xfs_cross_rename() 2689 * 2690 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2691 */ 2692 STATIC int 2693 xfs_cross_rename( 2694 struct xfs_trans *tp, 2695 struct xfs_inode *dp1, 2696 struct xfs_name *name1, 2697 struct xfs_inode *ip1, 2698 struct xfs_inode *dp2, 2699 struct xfs_name *name2, 2700 struct xfs_inode *ip2, 2701 struct xfs_defer_ops *dfops, 2702 xfs_fsblock_t *first_block, 2703 int spaceres) 2704 { 2705 int error = 0; 2706 int ip1_flags = 0; 2707 int ip2_flags = 0; 2708 int dp2_flags = 0; 2709 2710 /* Swap inode number for dirent in first parent */ 2711 error = xfs_dir_replace(tp, dp1, name1, 2712 ip2->i_ino, 2713 first_block, dfops, spaceres); 2714 if (error) 2715 goto out_trans_abort; 2716 2717 /* Swap inode number for dirent in second parent */ 2718 error = xfs_dir_replace(tp, dp2, name2, 2719 ip1->i_ino, 2720 first_block, dfops, spaceres); 2721 if (error) 2722 goto out_trans_abort; 2723 2724 /* 2725 * If we're renaming one or more directories across different parents, 2726 * update the respective ".." entries (and link counts) to match the new 2727 * parents. 2728 */ 2729 if (dp1 != dp2) { 2730 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2731 2732 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2733 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2734 dp1->i_ino, first_block, 2735 dfops, spaceres); 2736 if (error) 2737 goto out_trans_abort; 2738 2739 /* transfer ip2 ".." reference to dp1 */ 2740 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2741 error = xfs_droplink(tp, dp2); 2742 if (error) 2743 goto out_trans_abort; 2744 error = xfs_bumplink(tp, dp1); 2745 if (error) 2746 goto out_trans_abort; 2747 } 2748 2749 /* 2750 * Although ip1 isn't changed here, userspace needs 2751 * to be warned about the change, so that applications 2752 * relying on it (like backup ones), will properly 2753 * notify the change 2754 */ 2755 ip1_flags |= XFS_ICHGTIME_CHG; 2756 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2757 } 2758 2759 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2760 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2761 dp2->i_ino, first_block, 2762 dfops, spaceres); 2763 if (error) 2764 goto out_trans_abort; 2765 2766 /* transfer ip1 ".." reference to dp2 */ 2767 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2768 error = xfs_droplink(tp, dp1); 2769 if (error) 2770 goto out_trans_abort; 2771 error = xfs_bumplink(tp, dp2); 2772 if (error) 2773 goto out_trans_abort; 2774 } 2775 2776 /* 2777 * Although ip2 isn't changed here, userspace needs 2778 * to be warned about the change, so that applications 2779 * relying on it (like backup ones), will properly 2780 * notify the change 2781 */ 2782 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2783 ip2_flags |= XFS_ICHGTIME_CHG; 2784 } 2785 } 2786 2787 if (ip1_flags) { 2788 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2789 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2790 } 2791 if (ip2_flags) { 2792 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2793 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2794 } 2795 if (dp2_flags) { 2796 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2797 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2798 } 2799 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2800 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2801 return xfs_finish_rename(tp, dfops); 2802 2803 out_trans_abort: 2804 xfs_defer_cancel(dfops); 2805 xfs_trans_cancel(tp); 2806 return error; 2807 } 2808 2809 /* 2810 * xfs_rename_alloc_whiteout() 2811 * 2812 * Return a referenced, unlinked, unlocked inode that that can be used as a 2813 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2814 * crash between allocating the inode and linking it into the rename transaction 2815 * recovery will free the inode and we won't leak it. 2816 */ 2817 static int 2818 xfs_rename_alloc_whiteout( 2819 struct xfs_inode *dp, 2820 struct xfs_inode **wip) 2821 { 2822 struct xfs_inode *tmpfile; 2823 int error; 2824 2825 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2826 if (error) 2827 return error; 2828 2829 /* 2830 * Prepare the tmpfile inode as if it were created through the VFS. 2831 * Otherwise, the link increment paths will complain about nlink 0->1. 2832 * Drop the link count as done by d_tmpfile(), complete the inode setup 2833 * and flag it as linkable. 2834 */ 2835 drop_nlink(VFS_I(tmpfile)); 2836 xfs_setup_iops(tmpfile); 2837 xfs_finish_inode_setup(tmpfile); 2838 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2839 2840 *wip = tmpfile; 2841 return 0; 2842 } 2843 2844 /* 2845 * xfs_rename 2846 */ 2847 int 2848 xfs_rename( 2849 struct xfs_inode *src_dp, 2850 struct xfs_name *src_name, 2851 struct xfs_inode *src_ip, 2852 struct xfs_inode *target_dp, 2853 struct xfs_name *target_name, 2854 struct xfs_inode *target_ip, 2855 unsigned int flags) 2856 { 2857 struct xfs_mount *mp = src_dp->i_mount; 2858 struct xfs_trans *tp; 2859 struct xfs_defer_ops dfops; 2860 xfs_fsblock_t first_block; 2861 struct xfs_inode *wip = NULL; /* whiteout inode */ 2862 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2863 int num_inodes = __XFS_SORT_INODES; 2864 bool new_parent = (src_dp != target_dp); 2865 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2866 int spaceres; 2867 int error; 2868 2869 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2870 2871 if ((flags & RENAME_EXCHANGE) && !target_ip) 2872 return -EINVAL; 2873 2874 /* 2875 * If we are doing a whiteout operation, allocate the whiteout inode 2876 * we will be placing at the target and ensure the type is set 2877 * appropriately. 2878 */ 2879 if (flags & RENAME_WHITEOUT) { 2880 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2881 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2882 if (error) 2883 return error; 2884 2885 /* setup target dirent info as whiteout */ 2886 src_name->type = XFS_DIR3_FT_CHRDEV; 2887 } 2888 2889 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2890 inodes, &num_inodes); 2891 2892 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2893 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2894 if (error == -ENOSPC) { 2895 spaceres = 0; 2896 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2897 &tp); 2898 } 2899 if (error) 2900 goto out_release_wip; 2901 2902 /* 2903 * Attach the dquots to the inodes 2904 */ 2905 error = xfs_qm_vop_rename_dqattach(inodes); 2906 if (error) 2907 goto out_trans_cancel; 2908 2909 /* 2910 * Lock all the participating inodes. Depending upon whether 2911 * the target_name exists in the target directory, and 2912 * whether the target directory is the same as the source 2913 * directory, we can lock from 2 to 4 inodes. 2914 */ 2915 if (!new_parent) 2916 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2917 else 2918 xfs_lock_two_inodes(src_dp, target_dp, 2919 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2920 2921 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2922 2923 /* 2924 * Join all the inodes to the transaction. From this point on, 2925 * we can rely on either trans_commit or trans_cancel to unlock 2926 * them. 2927 */ 2928 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2929 if (new_parent) 2930 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2931 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2932 if (target_ip) 2933 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2934 if (wip) 2935 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2936 2937 /* 2938 * If we are using project inheritance, we only allow renames 2939 * into our tree when the project IDs are the same; else the 2940 * tree quota mechanism would be circumvented. 2941 */ 2942 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2943 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2944 error = -EXDEV; 2945 goto out_trans_cancel; 2946 } 2947 2948 xfs_defer_init(&dfops, &first_block); 2949 2950 /* RENAME_EXCHANGE is unique from here on. */ 2951 if (flags & RENAME_EXCHANGE) 2952 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2953 target_dp, target_name, target_ip, 2954 &dfops, &first_block, spaceres); 2955 2956 /* 2957 * Set up the target. 2958 */ 2959 if (target_ip == NULL) { 2960 /* 2961 * If there's no space reservation, check the entry will 2962 * fit before actually inserting it. 2963 */ 2964 if (!spaceres) { 2965 error = xfs_dir_canenter(tp, target_dp, target_name); 2966 if (error) 2967 goto out_trans_cancel; 2968 } 2969 /* 2970 * If target does not exist and the rename crosses 2971 * directories, adjust the target directory link count 2972 * to account for the ".." reference from the new entry. 2973 */ 2974 error = xfs_dir_createname(tp, target_dp, target_name, 2975 src_ip->i_ino, &first_block, 2976 &dfops, spaceres); 2977 if (error) 2978 goto out_bmap_cancel; 2979 2980 xfs_trans_ichgtime(tp, target_dp, 2981 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2982 2983 if (new_parent && src_is_directory) { 2984 error = xfs_bumplink(tp, target_dp); 2985 if (error) 2986 goto out_bmap_cancel; 2987 } 2988 } else { /* target_ip != NULL */ 2989 /* 2990 * If target exists and it's a directory, check that both 2991 * target and source are directories and that target can be 2992 * destroyed, or that neither is a directory. 2993 */ 2994 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 2995 /* 2996 * Make sure target dir is empty. 2997 */ 2998 if (!(xfs_dir_isempty(target_ip)) || 2999 (VFS_I(target_ip)->i_nlink > 2)) { 3000 error = -EEXIST; 3001 goto out_trans_cancel; 3002 } 3003 } 3004 3005 /* 3006 * Link the source inode under the target name. 3007 * If the source inode is a directory and we are moving 3008 * it across directories, its ".." entry will be 3009 * inconsistent until we replace that down below. 3010 * 3011 * In case there is already an entry with the same 3012 * name at the destination directory, remove it first. 3013 */ 3014 error = xfs_dir_replace(tp, target_dp, target_name, 3015 src_ip->i_ino, 3016 &first_block, &dfops, spaceres); 3017 if (error) 3018 goto out_bmap_cancel; 3019 3020 xfs_trans_ichgtime(tp, target_dp, 3021 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3022 3023 /* 3024 * Decrement the link count on the target since the target 3025 * dir no longer points to it. 3026 */ 3027 error = xfs_droplink(tp, target_ip); 3028 if (error) 3029 goto out_bmap_cancel; 3030 3031 if (src_is_directory) { 3032 /* 3033 * Drop the link from the old "." entry. 3034 */ 3035 error = xfs_droplink(tp, target_ip); 3036 if (error) 3037 goto out_bmap_cancel; 3038 } 3039 } /* target_ip != NULL */ 3040 3041 /* 3042 * Remove the source. 3043 */ 3044 if (new_parent && src_is_directory) { 3045 /* 3046 * Rewrite the ".." entry to point to the new 3047 * directory. 3048 */ 3049 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3050 target_dp->i_ino, 3051 &first_block, &dfops, spaceres); 3052 ASSERT(error != -EEXIST); 3053 if (error) 3054 goto out_bmap_cancel; 3055 } 3056 3057 /* 3058 * We always want to hit the ctime on the source inode. 3059 * 3060 * This isn't strictly required by the standards since the source 3061 * inode isn't really being changed, but old unix file systems did 3062 * it and some incremental backup programs won't work without it. 3063 */ 3064 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3065 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3066 3067 /* 3068 * Adjust the link count on src_dp. This is necessary when 3069 * renaming a directory, either within one parent when 3070 * the target existed, or across two parent directories. 3071 */ 3072 if (src_is_directory && (new_parent || target_ip != NULL)) { 3073 3074 /* 3075 * Decrement link count on src_directory since the 3076 * entry that's moved no longer points to it. 3077 */ 3078 error = xfs_droplink(tp, src_dp); 3079 if (error) 3080 goto out_bmap_cancel; 3081 } 3082 3083 /* 3084 * For whiteouts, we only need to update the source dirent with the 3085 * inode number of the whiteout inode rather than removing it 3086 * altogether. 3087 */ 3088 if (wip) { 3089 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3090 &first_block, &dfops, spaceres); 3091 } else 3092 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3093 &first_block, &dfops, spaceres); 3094 if (error) 3095 goto out_bmap_cancel; 3096 3097 /* 3098 * For whiteouts, we need to bump the link count on the whiteout inode. 3099 * This means that failures all the way up to this point leave the inode 3100 * on the unlinked list and so cleanup is a simple matter of dropping 3101 * the remaining reference to it. If we fail here after bumping the link 3102 * count, we're shutting down the filesystem so we'll never see the 3103 * intermediate state on disk. 3104 */ 3105 if (wip) { 3106 ASSERT(VFS_I(wip)->i_nlink == 0); 3107 error = xfs_bumplink(tp, wip); 3108 if (error) 3109 goto out_bmap_cancel; 3110 error = xfs_iunlink_remove(tp, wip); 3111 if (error) 3112 goto out_bmap_cancel; 3113 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3114 3115 /* 3116 * Now we have a real link, clear the "I'm a tmpfile" state 3117 * flag from the inode so it doesn't accidentally get misused in 3118 * future. 3119 */ 3120 VFS_I(wip)->i_state &= ~I_LINKABLE; 3121 } 3122 3123 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3124 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3125 if (new_parent) 3126 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3127 3128 error = xfs_finish_rename(tp, &dfops); 3129 if (wip) 3130 IRELE(wip); 3131 return error; 3132 3133 out_bmap_cancel: 3134 xfs_defer_cancel(&dfops); 3135 out_trans_cancel: 3136 xfs_trans_cancel(tp); 3137 out_release_wip: 3138 if (wip) 3139 IRELE(wip); 3140 return error; 3141 } 3142 3143 STATIC int 3144 xfs_iflush_cluster( 3145 struct xfs_inode *ip, 3146 struct xfs_buf *bp) 3147 { 3148 struct xfs_mount *mp = ip->i_mount; 3149 struct xfs_perag *pag; 3150 unsigned long first_index, mask; 3151 unsigned long inodes_per_cluster; 3152 int cilist_size; 3153 struct xfs_inode **cilist; 3154 struct xfs_inode *cip; 3155 int nr_found; 3156 int clcount = 0; 3157 int bufwasdelwri; 3158 int i; 3159 3160 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3161 3162 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3163 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3164 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3165 if (!cilist) 3166 goto out_put; 3167 3168 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3169 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3170 rcu_read_lock(); 3171 /* really need a gang lookup range call here */ 3172 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3173 first_index, inodes_per_cluster); 3174 if (nr_found == 0) 3175 goto out_free; 3176 3177 for (i = 0; i < nr_found; i++) { 3178 cip = cilist[i]; 3179 if (cip == ip) 3180 continue; 3181 3182 /* 3183 * because this is an RCU protected lookup, we could find a 3184 * recently freed or even reallocated inode during the lookup. 3185 * We need to check under the i_flags_lock for a valid inode 3186 * here. Skip it if it is not valid or the wrong inode. 3187 */ 3188 spin_lock(&cip->i_flags_lock); 3189 if (!cip->i_ino || 3190 __xfs_iflags_test(cip, XFS_ISTALE)) { 3191 spin_unlock(&cip->i_flags_lock); 3192 continue; 3193 } 3194 3195 /* 3196 * Once we fall off the end of the cluster, no point checking 3197 * any more inodes in the list because they will also all be 3198 * outside the cluster. 3199 */ 3200 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3201 spin_unlock(&cip->i_flags_lock); 3202 break; 3203 } 3204 spin_unlock(&cip->i_flags_lock); 3205 3206 /* 3207 * Do an un-protected check to see if the inode is dirty and 3208 * is a candidate for flushing. These checks will be repeated 3209 * later after the appropriate locks are acquired. 3210 */ 3211 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3212 continue; 3213 3214 /* 3215 * Try to get locks. If any are unavailable or it is pinned, 3216 * then this inode cannot be flushed and is skipped. 3217 */ 3218 3219 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3220 continue; 3221 if (!xfs_iflock_nowait(cip)) { 3222 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3223 continue; 3224 } 3225 if (xfs_ipincount(cip)) { 3226 xfs_ifunlock(cip); 3227 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3228 continue; 3229 } 3230 3231 3232 /* 3233 * Check the inode number again, just to be certain we are not 3234 * racing with freeing in xfs_reclaim_inode(). See the comments 3235 * in that function for more information as to why the initial 3236 * check is not sufficient. 3237 */ 3238 if (!cip->i_ino) { 3239 xfs_ifunlock(cip); 3240 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3241 continue; 3242 } 3243 3244 /* 3245 * arriving here means that this inode can be flushed. First 3246 * re-check that it's dirty before flushing. 3247 */ 3248 if (!xfs_inode_clean(cip)) { 3249 int error; 3250 error = xfs_iflush_int(cip, bp); 3251 if (error) { 3252 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3253 goto cluster_corrupt_out; 3254 } 3255 clcount++; 3256 } else { 3257 xfs_ifunlock(cip); 3258 } 3259 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3260 } 3261 3262 if (clcount) { 3263 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3264 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3265 } 3266 3267 out_free: 3268 rcu_read_unlock(); 3269 kmem_free(cilist); 3270 out_put: 3271 xfs_perag_put(pag); 3272 return 0; 3273 3274 3275 cluster_corrupt_out: 3276 /* 3277 * Corruption detected in the clustering loop. Invalidate the 3278 * inode buffer and shut down the filesystem. 3279 */ 3280 rcu_read_unlock(); 3281 /* 3282 * Clean up the buffer. If it was delwri, just release it -- 3283 * brelse can handle it with no problems. If not, shut down the 3284 * filesystem before releasing the buffer. 3285 */ 3286 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3287 if (bufwasdelwri) 3288 xfs_buf_relse(bp); 3289 3290 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3291 3292 if (!bufwasdelwri) { 3293 /* 3294 * Just like incore_relse: if we have b_iodone functions, 3295 * mark the buffer as an error and call them. Otherwise 3296 * mark it as stale and brelse. 3297 */ 3298 if (bp->b_iodone) { 3299 bp->b_flags &= ~XBF_DONE; 3300 xfs_buf_stale(bp); 3301 xfs_buf_ioerror(bp, -EIO); 3302 xfs_buf_ioend(bp); 3303 } else { 3304 xfs_buf_stale(bp); 3305 xfs_buf_relse(bp); 3306 } 3307 } 3308 3309 /* 3310 * Unlocks the flush lock 3311 */ 3312 xfs_iflush_abort(cip, false); 3313 kmem_free(cilist); 3314 xfs_perag_put(pag); 3315 return -EFSCORRUPTED; 3316 } 3317 3318 /* 3319 * Flush dirty inode metadata into the backing buffer. 3320 * 3321 * The caller must have the inode lock and the inode flush lock held. The 3322 * inode lock will still be held upon return to the caller, and the inode 3323 * flush lock will be released after the inode has reached the disk. 3324 * 3325 * The caller must write out the buffer returned in *bpp and release it. 3326 */ 3327 int 3328 xfs_iflush( 3329 struct xfs_inode *ip, 3330 struct xfs_buf **bpp) 3331 { 3332 struct xfs_mount *mp = ip->i_mount; 3333 struct xfs_buf *bp = NULL; 3334 struct xfs_dinode *dip; 3335 int error; 3336 3337 XFS_STATS_INC(mp, xs_iflush_count); 3338 3339 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3340 ASSERT(xfs_isiflocked(ip)); 3341 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3342 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3343 3344 *bpp = NULL; 3345 3346 xfs_iunpin_wait(ip); 3347 3348 /* 3349 * For stale inodes we cannot rely on the backing buffer remaining 3350 * stale in cache for the remaining life of the stale inode and so 3351 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3352 * inodes below. We have to check this after ensuring the inode is 3353 * unpinned so that it is safe to reclaim the stale inode after the 3354 * flush call. 3355 */ 3356 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3357 xfs_ifunlock(ip); 3358 return 0; 3359 } 3360 3361 /* 3362 * This may have been unpinned because the filesystem is shutting 3363 * down forcibly. If that's the case we must not write this inode 3364 * to disk, because the log record didn't make it to disk. 3365 * 3366 * We also have to remove the log item from the AIL in this case, 3367 * as we wait for an empty AIL as part of the unmount process. 3368 */ 3369 if (XFS_FORCED_SHUTDOWN(mp)) { 3370 error = -EIO; 3371 goto abort_out; 3372 } 3373 3374 /* 3375 * Get the buffer containing the on-disk inode. We are doing a try-lock 3376 * operation here, so we may get an EAGAIN error. In that case, we 3377 * simply want to return with the inode still dirty. 3378 * 3379 * If we get any other error, we effectively have a corruption situation 3380 * and we cannot flush the inode, so we treat it the same as failing 3381 * xfs_iflush_int(). 3382 */ 3383 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3384 0); 3385 if (error == -EAGAIN) { 3386 xfs_ifunlock(ip); 3387 return error; 3388 } 3389 if (error) 3390 goto corrupt_out; 3391 3392 /* 3393 * First flush out the inode that xfs_iflush was called with. 3394 */ 3395 error = xfs_iflush_int(ip, bp); 3396 if (error) 3397 goto corrupt_out; 3398 3399 /* 3400 * If the buffer is pinned then push on the log now so we won't 3401 * get stuck waiting in the write for too long. 3402 */ 3403 if (xfs_buf_ispinned(bp)) 3404 xfs_log_force(mp, 0); 3405 3406 /* 3407 * inode clustering: 3408 * see if other inodes can be gathered into this write 3409 */ 3410 error = xfs_iflush_cluster(ip, bp); 3411 if (error) 3412 goto cluster_corrupt_out; 3413 3414 *bpp = bp; 3415 return 0; 3416 3417 corrupt_out: 3418 if (bp) 3419 xfs_buf_relse(bp); 3420 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3421 cluster_corrupt_out: 3422 error = -EFSCORRUPTED; 3423 abort_out: 3424 /* 3425 * Unlocks the flush lock 3426 */ 3427 xfs_iflush_abort(ip, false); 3428 return error; 3429 } 3430 3431 STATIC int 3432 xfs_iflush_int( 3433 struct xfs_inode *ip, 3434 struct xfs_buf *bp) 3435 { 3436 struct xfs_inode_log_item *iip = ip->i_itemp; 3437 struct xfs_dinode *dip; 3438 struct xfs_mount *mp = ip->i_mount; 3439 3440 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3441 ASSERT(xfs_isiflocked(ip)); 3442 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3443 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3444 ASSERT(iip != NULL && iip->ili_fields != 0); 3445 ASSERT(ip->i_d.di_version > 1); 3446 3447 /* set *dip = inode's place in the buffer */ 3448 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3449 3450 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3451 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 3452 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3453 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3454 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3455 goto corrupt_out; 3456 } 3457 if (S_ISREG(VFS_I(ip)->i_mode)) { 3458 if (XFS_TEST_ERROR( 3459 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3460 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3461 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 3462 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3463 "%s: Bad regular inode %Lu, ptr 0x%p", 3464 __func__, ip->i_ino, ip); 3465 goto corrupt_out; 3466 } 3467 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3468 if (XFS_TEST_ERROR( 3469 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3470 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3471 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3472 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 3473 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3474 "%s: Bad directory inode %Lu, ptr 0x%p", 3475 __func__, ip->i_ino, ip); 3476 goto corrupt_out; 3477 } 3478 } 3479 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3480 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 3481 XFS_RANDOM_IFLUSH_5)) { 3482 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3483 "%s: detected corrupt incore inode %Lu, " 3484 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3485 __func__, ip->i_ino, 3486 ip->i_d.di_nextents + ip->i_d.di_anextents, 3487 ip->i_d.di_nblocks, ip); 3488 goto corrupt_out; 3489 } 3490 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3491 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 3492 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3493 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3494 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3495 goto corrupt_out; 3496 } 3497 3498 /* 3499 * Inode item log recovery for v2 inodes are dependent on the 3500 * di_flushiter count for correct sequencing. We bump the flush 3501 * iteration count so we can detect flushes which postdate a log record 3502 * during recovery. This is redundant as we now log every change and 3503 * hence this can't happen but we need to still do it to ensure 3504 * backwards compatibility with old kernels that predate logging all 3505 * inode changes. 3506 */ 3507 if (ip->i_d.di_version < 3) 3508 ip->i_d.di_flushiter++; 3509 3510 /* 3511 * Copy the dirty parts of the inode into the on-disk inode. We always 3512 * copy out the core of the inode, because if the inode is dirty at all 3513 * the core must be. 3514 */ 3515 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3516 3517 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3518 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3519 ip->i_d.di_flushiter = 0; 3520 3521 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3522 if (XFS_IFORK_Q(ip)) 3523 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3524 xfs_inobp_check(mp, bp); 3525 3526 /* 3527 * We've recorded everything logged in the inode, so we'd like to clear 3528 * the ili_fields bits so we don't log and flush things unnecessarily. 3529 * However, we can't stop logging all this information until the data 3530 * we've copied into the disk buffer is written to disk. If we did we 3531 * might overwrite the copy of the inode in the log with all the data 3532 * after re-logging only part of it, and in the face of a crash we 3533 * wouldn't have all the data we need to recover. 3534 * 3535 * What we do is move the bits to the ili_last_fields field. When 3536 * logging the inode, these bits are moved back to the ili_fields field. 3537 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3538 * know that the information those bits represent is permanently on 3539 * disk. As long as the flush completes before the inode is logged 3540 * again, then both ili_fields and ili_last_fields will be cleared. 3541 * 3542 * We can play with the ili_fields bits here, because the inode lock 3543 * must be held exclusively in order to set bits there and the flush 3544 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3545 * done routine can tell whether or not to look in the AIL. Also, store 3546 * the current LSN of the inode so that we can tell whether the item has 3547 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3548 * need the AIL lock, because it is a 64 bit value that cannot be read 3549 * atomically. 3550 */ 3551 iip->ili_last_fields = iip->ili_fields; 3552 iip->ili_fields = 0; 3553 iip->ili_fsync_fields = 0; 3554 iip->ili_logged = 1; 3555 3556 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3557 &iip->ili_item.li_lsn); 3558 3559 /* 3560 * Attach the function xfs_iflush_done to the inode's 3561 * buffer. This will remove the inode from the AIL 3562 * and unlock the inode's flush lock when the inode is 3563 * completely written to disk. 3564 */ 3565 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3566 3567 /* generate the checksum. */ 3568 xfs_dinode_calc_crc(mp, dip); 3569 3570 ASSERT(bp->b_fspriv != NULL); 3571 ASSERT(bp->b_iodone != NULL); 3572 return 0; 3573 3574 corrupt_out: 3575 return -EFSCORRUPTED; 3576 } 3577