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_inode.h" 29 #include "xfs_da_format.h" 30 #include "xfs_da_btree.h" 31 #include "xfs_dir2.h" 32 #include "xfs_attr_sf.h" 33 #include "xfs_attr.h" 34 #include "xfs_trans_space.h" 35 #include "xfs_trans.h" 36 #include "xfs_buf_item.h" 37 #include "xfs_inode_item.h" 38 #include "xfs_ialloc.h" 39 #include "xfs_bmap.h" 40 #include "xfs_bmap_util.h" 41 #include "xfs_error.h" 42 #include "xfs_quota.h" 43 #include "xfs_filestream.h" 44 #include "xfs_cksum.h" 45 #include "xfs_trace.h" 46 #include "xfs_icache.h" 47 #include "xfs_symlink.h" 48 #include "xfs_trans_priv.h" 49 #include "xfs_log.h" 50 #include "xfs_bmap_btree.h" 51 52 kmem_zone_t *xfs_inode_zone; 53 54 /* 55 * Used in xfs_itruncate_extents(). This is the maximum number of extents 56 * freed from a file in a single transaction. 57 */ 58 #define XFS_ITRUNC_MAX_EXTENTS 2 59 60 STATIC int xfs_iflush_int(struct xfs_inode *, struct xfs_buf *); 61 STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *); 62 STATIC int xfs_iunlink_remove(struct xfs_trans *, struct xfs_inode *); 63 64 /* 65 * helper function to extract extent size hint from inode 66 */ 67 xfs_extlen_t 68 xfs_get_extsz_hint( 69 struct xfs_inode *ip) 70 { 71 if ((ip->i_d.di_flags & XFS_DIFLAG_EXTSIZE) && ip->i_d.di_extsize) 72 return ip->i_d.di_extsize; 73 if (XFS_IS_REALTIME_INODE(ip)) 74 return ip->i_mount->m_sb.sb_rextsize; 75 return 0; 76 } 77 78 /* 79 * These two are wrapper routines around the xfs_ilock() routine used to 80 * centralize some grungy code. They are used in places that wish to lock the 81 * inode solely for reading the extents. The reason these places can't just 82 * call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to 83 * bringing in of the extents from disk for a file in b-tree format. If the 84 * inode is in b-tree format, then we need to lock the inode exclusively until 85 * the extents are read in. Locking it exclusively all the time would limit 86 * our parallelism unnecessarily, though. What we do instead is check to see 87 * if the extents have been read in yet, and only lock the inode exclusively 88 * if they have not. 89 * 90 * The functions return a value which should be given to the corresponding 91 * xfs_iunlock() call. 92 */ 93 uint 94 xfs_ilock_data_map_shared( 95 struct xfs_inode *ip) 96 { 97 uint lock_mode = XFS_ILOCK_SHARED; 98 99 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 100 (ip->i_df.if_flags & XFS_IFEXTENTS) == 0) 101 lock_mode = XFS_ILOCK_EXCL; 102 xfs_ilock(ip, lock_mode); 103 return lock_mode; 104 } 105 106 uint 107 xfs_ilock_attr_map_shared( 108 struct xfs_inode *ip) 109 { 110 uint lock_mode = XFS_ILOCK_SHARED; 111 112 if (ip->i_d.di_aformat == XFS_DINODE_FMT_BTREE && 113 (ip->i_afp->if_flags & XFS_IFEXTENTS) == 0) 114 lock_mode = XFS_ILOCK_EXCL; 115 xfs_ilock(ip, lock_mode); 116 return lock_mode; 117 } 118 119 /* 120 * The xfs inode contains 3 multi-reader locks: the i_iolock the i_mmap_lock and 121 * the i_lock. This routine allows various combinations of the locks to be 122 * obtained. 123 * 124 * The 3 locks should always be ordered so that the IO lock is obtained first, 125 * the mmap lock second and the ilock last in order to prevent deadlock. 126 * 127 * Basic locking order: 128 * 129 * i_iolock -> i_mmap_lock -> page_lock -> i_ilock 130 * 131 * mmap_sem locking order: 132 * 133 * i_iolock -> page lock -> mmap_sem 134 * mmap_sem -> i_mmap_lock -> page_lock 135 * 136 * The difference in mmap_sem locking order mean that we cannot hold the 137 * i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can 138 * fault in pages during copy in/out (for buffered IO) or require the mmap_sem 139 * in get_user_pages() to map the user pages into the kernel address space for 140 * direct IO. Similarly the i_iolock cannot be taken inside a page fault because 141 * page faults already hold the mmap_sem. 142 * 143 * Hence to serialise fully against both syscall and mmap based IO, we need to 144 * take both the i_iolock and the i_mmap_lock. These locks should *only* be both 145 * taken in places where we need to invalidate the page cache in a race 146 * free manner (e.g. truncate, hole punch and other extent manipulation 147 * functions). 148 */ 149 void 150 xfs_ilock( 151 xfs_inode_t *ip, 152 uint lock_flags) 153 { 154 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 155 156 /* 157 * You can't set both SHARED and EXCL for the same lock, 158 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 159 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 160 */ 161 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 162 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 163 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 164 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 165 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 166 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 167 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 168 169 if (lock_flags & XFS_IOLOCK_EXCL) 170 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 171 else if (lock_flags & XFS_IOLOCK_SHARED) 172 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 173 174 if (lock_flags & XFS_MMAPLOCK_EXCL) 175 mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 176 else if (lock_flags & XFS_MMAPLOCK_SHARED) 177 mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags)); 178 179 if (lock_flags & XFS_ILOCK_EXCL) 180 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 181 else if (lock_flags & XFS_ILOCK_SHARED) 182 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 183 } 184 185 /* 186 * This is just like xfs_ilock(), except that the caller 187 * is guaranteed not to sleep. It returns 1 if it gets 188 * the requested locks and 0 otherwise. If the IO lock is 189 * obtained but the inode lock cannot be, then the IO lock 190 * is dropped before returning. 191 * 192 * ip -- the inode being locked 193 * lock_flags -- this parameter indicates the inode's locks to be 194 * to be locked. See the comment for xfs_ilock() for a list 195 * of valid values. 196 */ 197 int 198 xfs_ilock_nowait( 199 xfs_inode_t *ip, 200 uint lock_flags) 201 { 202 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 203 204 /* 205 * You can't set both SHARED and EXCL for the same lock, 206 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 207 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 208 */ 209 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 210 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 211 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 212 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 213 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 214 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 215 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 216 217 if (lock_flags & XFS_IOLOCK_EXCL) { 218 if (!mrtryupdate(&ip->i_iolock)) 219 goto out; 220 } else if (lock_flags & XFS_IOLOCK_SHARED) { 221 if (!mrtryaccess(&ip->i_iolock)) 222 goto out; 223 } 224 225 if (lock_flags & XFS_MMAPLOCK_EXCL) { 226 if (!mrtryupdate(&ip->i_mmaplock)) 227 goto out_undo_iolock; 228 } else if (lock_flags & XFS_MMAPLOCK_SHARED) { 229 if (!mrtryaccess(&ip->i_mmaplock)) 230 goto out_undo_iolock; 231 } 232 233 if (lock_flags & XFS_ILOCK_EXCL) { 234 if (!mrtryupdate(&ip->i_lock)) 235 goto out_undo_mmaplock; 236 } else if (lock_flags & XFS_ILOCK_SHARED) { 237 if (!mrtryaccess(&ip->i_lock)) 238 goto out_undo_mmaplock; 239 } 240 return 1; 241 242 out_undo_mmaplock: 243 if (lock_flags & XFS_MMAPLOCK_EXCL) 244 mrunlock_excl(&ip->i_mmaplock); 245 else if (lock_flags & XFS_MMAPLOCK_SHARED) 246 mrunlock_shared(&ip->i_mmaplock); 247 out_undo_iolock: 248 if (lock_flags & XFS_IOLOCK_EXCL) 249 mrunlock_excl(&ip->i_iolock); 250 else if (lock_flags & XFS_IOLOCK_SHARED) 251 mrunlock_shared(&ip->i_iolock); 252 out: 253 return 0; 254 } 255 256 /* 257 * xfs_iunlock() is used to drop the inode locks acquired with 258 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 259 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 260 * that we know which locks to drop. 261 * 262 * ip -- the inode being unlocked 263 * lock_flags -- this parameter indicates the inode's locks to be 264 * to be unlocked. See the comment for xfs_ilock() for a list 265 * of valid values for this parameter. 266 * 267 */ 268 void 269 xfs_iunlock( 270 xfs_inode_t *ip, 271 uint lock_flags) 272 { 273 /* 274 * You can't set both SHARED and EXCL for the same lock, 275 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 276 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 277 */ 278 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 279 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 280 ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) != 281 (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)); 282 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 283 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 284 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0); 285 ASSERT(lock_flags != 0); 286 287 if (lock_flags & XFS_IOLOCK_EXCL) 288 mrunlock_excl(&ip->i_iolock); 289 else if (lock_flags & XFS_IOLOCK_SHARED) 290 mrunlock_shared(&ip->i_iolock); 291 292 if (lock_flags & XFS_MMAPLOCK_EXCL) 293 mrunlock_excl(&ip->i_mmaplock); 294 else if (lock_flags & XFS_MMAPLOCK_SHARED) 295 mrunlock_shared(&ip->i_mmaplock); 296 297 if (lock_flags & XFS_ILOCK_EXCL) 298 mrunlock_excl(&ip->i_lock); 299 else if (lock_flags & XFS_ILOCK_SHARED) 300 mrunlock_shared(&ip->i_lock); 301 302 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 303 } 304 305 /* 306 * give up write locks. the i/o lock cannot be held nested 307 * if it is being demoted. 308 */ 309 void 310 xfs_ilock_demote( 311 xfs_inode_t *ip, 312 uint lock_flags) 313 { 314 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)); 315 ASSERT((lock_flags & 316 ~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 317 318 if (lock_flags & XFS_ILOCK_EXCL) 319 mrdemote(&ip->i_lock); 320 if (lock_flags & XFS_MMAPLOCK_EXCL) 321 mrdemote(&ip->i_mmaplock); 322 if (lock_flags & XFS_IOLOCK_EXCL) 323 mrdemote(&ip->i_iolock); 324 325 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 326 } 327 328 #if defined(DEBUG) || defined(XFS_WARN) 329 int 330 xfs_isilocked( 331 xfs_inode_t *ip, 332 uint lock_flags) 333 { 334 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 335 if (!(lock_flags & XFS_ILOCK_SHARED)) 336 return !!ip->i_lock.mr_writer; 337 return rwsem_is_locked(&ip->i_lock.mr_lock); 338 } 339 340 if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) { 341 if (!(lock_flags & XFS_MMAPLOCK_SHARED)) 342 return !!ip->i_mmaplock.mr_writer; 343 return rwsem_is_locked(&ip->i_mmaplock.mr_lock); 344 } 345 346 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 347 if (!(lock_flags & XFS_IOLOCK_SHARED)) 348 return !!ip->i_iolock.mr_writer; 349 return rwsem_is_locked(&ip->i_iolock.mr_lock); 350 } 351 352 ASSERT(0); 353 return 0; 354 } 355 #endif 356 357 #ifdef DEBUG 358 int xfs_locked_n; 359 int xfs_small_retries; 360 int xfs_middle_retries; 361 int xfs_lots_retries; 362 int xfs_lock_delays; 363 #endif 364 365 /* 366 * xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when 367 * DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined 368 * when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build 369 * errors and warnings. 370 */ 371 #if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP) 372 static bool 373 xfs_lockdep_subclass_ok( 374 int subclass) 375 { 376 return subclass < MAX_LOCKDEP_SUBCLASSES; 377 } 378 #else 379 #define xfs_lockdep_subclass_ok(subclass) (true) 380 #endif 381 382 /* 383 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 384 * value. This can be called for any type of inode lock combination, including 385 * parent locking. Care must be taken to ensure we don't overrun the subclass 386 * storage fields in the class mask we build. 387 */ 388 static inline int 389 xfs_lock_inumorder(int lock_mode, int subclass) 390 { 391 int class = 0; 392 393 ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP | 394 XFS_ILOCK_RTSUM))); 395 ASSERT(xfs_lockdep_subclass_ok(subclass)); 396 397 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 398 ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS); 399 ASSERT(xfs_lockdep_subclass_ok(subclass + 400 XFS_IOLOCK_PARENT_VAL)); 401 class += subclass << XFS_IOLOCK_SHIFT; 402 if (lock_mode & XFS_IOLOCK_PARENT) 403 class += XFS_IOLOCK_PARENT_VAL << XFS_IOLOCK_SHIFT; 404 } 405 406 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 407 ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS); 408 class += subclass << XFS_MMAPLOCK_SHIFT; 409 } 410 411 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) { 412 ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS); 413 class += subclass << XFS_ILOCK_SHIFT; 414 } 415 416 return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class; 417 } 418 419 /* 420 * The following routine will lock n inodes in exclusive mode. We assume the 421 * caller calls us with the inodes in i_ino order. 422 * 423 * We need to detect deadlock where an inode that we lock is in the AIL and we 424 * start waiting for another inode that is locked by a thread in a long running 425 * transaction (such as truncate). This can result in deadlock since the long 426 * running trans might need to wait for the inode we just locked in order to 427 * push the tail and free space in the log. 428 * 429 * xfs_lock_inodes() can only be used to lock one type of lock at a time - 430 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 431 * lock more than one at a time, lockdep will report false positives saying we 432 * have violated locking orders. 433 */ 434 void 435 xfs_lock_inodes( 436 xfs_inode_t **ips, 437 int inodes, 438 uint lock_mode) 439 { 440 int attempts = 0, i, j, try_lock; 441 xfs_log_item_t *lp; 442 443 /* 444 * Currently supports between 2 and 5 inodes with exclusive locking. We 445 * support an arbitrary depth of locking here, but absolute limits on 446 * inodes depend on the the type of locking and the limits placed by 447 * lockdep annotations in xfs_lock_inumorder. These are all checked by 448 * the asserts. 449 */ 450 ASSERT(ips && inodes >= 2 && inodes <= 5); 451 ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL | 452 XFS_ILOCK_EXCL)); 453 ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED | 454 XFS_ILOCK_SHARED))); 455 ASSERT(!(lock_mode & XFS_IOLOCK_EXCL) || 456 inodes <= XFS_IOLOCK_MAX_SUBCLASS + 1); 457 ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) || 458 inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1); 459 ASSERT(!(lock_mode & XFS_ILOCK_EXCL) || 460 inodes <= XFS_ILOCK_MAX_SUBCLASS + 1); 461 462 if (lock_mode & XFS_IOLOCK_EXCL) { 463 ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL))); 464 } else if (lock_mode & XFS_MMAPLOCK_EXCL) 465 ASSERT(!(lock_mode & XFS_ILOCK_EXCL)); 466 467 try_lock = 0; 468 i = 0; 469 again: 470 for (; i < inodes; i++) { 471 ASSERT(ips[i]); 472 473 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 474 continue; 475 476 /* 477 * If try_lock is not set yet, make sure all locked inodes are 478 * not in the AIL. If any are, set try_lock to be used later. 479 */ 480 if (!try_lock) { 481 for (j = (i - 1); j >= 0 && !try_lock; j--) { 482 lp = (xfs_log_item_t *)ips[j]->i_itemp; 483 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) 484 try_lock++; 485 } 486 } 487 488 /* 489 * If any of the previous locks we have locked is in the AIL, 490 * we must TRY to get the second and subsequent locks. If 491 * we can't get any, we must release all we have 492 * and try again. 493 */ 494 if (!try_lock) { 495 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 496 continue; 497 } 498 499 /* try_lock means we have an inode locked that is in the AIL. */ 500 ASSERT(i != 0); 501 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 502 continue; 503 504 /* 505 * Unlock all previous guys and try again. xfs_iunlock will try 506 * to push the tail if the inode is in the AIL. 507 */ 508 attempts++; 509 for (j = i - 1; j >= 0; j--) { 510 /* 511 * Check to see if we've already unlocked this one. Not 512 * the first one going back, and the inode ptr is the 513 * same. 514 */ 515 if (j != (i - 1) && ips[j] == ips[j + 1]) 516 continue; 517 518 xfs_iunlock(ips[j], lock_mode); 519 } 520 521 if ((attempts % 5) == 0) { 522 delay(1); /* Don't just spin the CPU */ 523 #ifdef DEBUG 524 xfs_lock_delays++; 525 #endif 526 } 527 i = 0; 528 try_lock = 0; 529 goto again; 530 } 531 532 #ifdef DEBUG 533 if (attempts) { 534 if (attempts < 5) xfs_small_retries++; 535 else if (attempts < 100) xfs_middle_retries++; 536 else xfs_lots_retries++; 537 } else { 538 xfs_locked_n++; 539 } 540 #endif 541 } 542 543 /* 544 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 545 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 546 * lock more than one at a time, lockdep will report false positives saying we 547 * have violated locking orders. 548 */ 549 void 550 xfs_lock_two_inodes( 551 xfs_inode_t *ip0, 552 xfs_inode_t *ip1, 553 uint lock_mode) 554 { 555 xfs_inode_t *temp; 556 int attempts = 0; 557 xfs_log_item_t *lp; 558 559 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 560 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); 561 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 562 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) 563 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 564 565 ASSERT(ip0->i_ino != ip1->i_ino); 566 567 if (ip0->i_ino > ip1->i_ino) { 568 temp = ip0; 569 ip0 = ip1; 570 ip1 = temp; 571 } 572 573 again: 574 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); 575 576 /* 577 * If the first lock we have locked is in the AIL, we must TRY to get 578 * the second lock. If we can't get it, we must release the first one 579 * and try again. 580 */ 581 lp = (xfs_log_item_t *)ip0->i_itemp; 582 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { 583 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { 584 xfs_iunlock(ip0, lock_mode); 585 if ((++attempts % 5) == 0) 586 delay(1); /* Don't just spin the CPU */ 587 goto again; 588 } 589 } else { 590 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); 591 } 592 } 593 594 595 void 596 __xfs_iflock( 597 struct xfs_inode *ip) 598 { 599 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 600 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 601 602 do { 603 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 604 if (xfs_isiflocked(ip)) 605 io_schedule(); 606 } while (!xfs_iflock_nowait(ip)); 607 608 finish_wait(wq, &wait.wait); 609 } 610 611 STATIC uint 612 _xfs_dic2xflags( 613 __uint16_t di_flags, 614 uint64_t di_flags2, 615 bool has_attr) 616 { 617 uint flags = 0; 618 619 if (di_flags & XFS_DIFLAG_ANY) { 620 if (di_flags & XFS_DIFLAG_REALTIME) 621 flags |= FS_XFLAG_REALTIME; 622 if (di_flags & XFS_DIFLAG_PREALLOC) 623 flags |= FS_XFLAG_PREALLOC; 624 if (di_flags & XFS_DIFLAG_IMMUTABLE) 625 flags |= FS_XFLAG_IMMUTABLE; 626 if (di_flags & XFS_DIFLAG_APPEND) 627 flags |= FS_XFLAG_APPEND; 628 if (di_flags & XFS_DIFLAG_SYNC) 629 flags |= FS_XFLAG_SYNC; 630 if (di_flags & XFS_DIFLAG_NOATIME) 631 flags |= FS_XFLAG_NOATIME; 632 if (di_flags & XFS_DIFLAG_NODUMP) 633 flags |= FS_XFLAG_NODUMP; 634 if (di_flags & XFS_DIFLAG_RTINHERIT) 635 flags |= FS_XFLAG_RTINHERIT; 636 if (di_flags & XFS_DIFLAG_PROJINHERIT) 637 flags |= FS_XFLAG_PROJINHERIT; 638 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 639 flags |= FS_XFLAG_NOSYMLINKS; 640 if (di_flags & XFS_DIFLAG_EXTSIZE) 641 flags |= FS_XFLAG_EXTSIZE; 642 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 643 flags |= FS_XFLAG_EXTSZINHERIT; 644 if (di_flags & XFS_DIFLAG_NODEFRAG) 645 flags |= FS_XFLAG_NODEFRAG; 646 if (di_flags & XFS_DIFLAG_FILESTREAM) 647 flags |= FS_XFLAG_FILESTREAM; 648 } 649 650 if (di_flags2 & XFS_DIFLAG2_ANY) { 651 if (di_flags2 & XFS_DIFLAG2_DAX) 652 flags |= FS_XFLAG_DAX; 653 } 654 655 if (has_attr) 656 flags |= FS_XFLAG_HASATTR; 657 658 return flags; 659 } 660 661 uint 662 xfs_ip2xflags( 663 struct xfs_inode *ip) 664 { 665 struct xfs_icdinode *dic = &ip->i_d; 666 667 return _xfs_dic2xflags(dic->di_flags, dic->di_flags2, XFS_IFORK_Q(ip)); 668 } 669 670 uint 671 xfs_dic2xflags( 672 struct xfs_dinode *dip) 673 { 674 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags), 675 be64_to_cpu(dip->di_flags2), XFS_DFORK_Q(dip)); 676 } 677 678 /* 679 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 680 * is allowed, otherwise it has to be an exact match. If a CI match is found, 681 * ci_name->name will point to a the actual name (caller must free) or 682 * will be set to NULL if an exact match is found. 683 */ 684 int 685 xfs_lookup( 686 xfs_inode_t *dp, 687 struct xfs_name *name, 688 xfs_inode_t **ipp, 689 struct xfs_name *ci_name) 690 { 691 xfs_ino_t inum; 692 int error; 693 694 trace_xfs_lookup(dp, name); 695 696 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 697 return -EIO; 698 699 xfs_ilock(dp, XFS_IOLOCK_SHARED); 700 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 701 if (error) 702 goto out_unlock; 703 704 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 705 if (error) 706 goto out_free_name; 707 708 xfs_iunlock(dp, XFS_IOLOCK_SHARED); 709 return 0; 710 711 out_free_name: 712 if (ci_name) 713 kmem_free(ci_name->name); 714 out_unlock: 715 xfs_iunlock(dp, XFS_IOLOCK_SHARED); 716 *ipp = NULL; 717 return error; 718 } 719 720 /* 721 * Allocate an inode on disk and return a copy of its in-core version. 722 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 723 * appropriately within the inode. The uid and gid for the inode are 724 * set according to the contents of the given cred structure. 725 * 726 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 727 * has a free inode available, call xfs_iget() to obtain the in-core 728 * version of the allocated inode. Finally, fill in the inode and 729 * log its initial contents. In this case, ialloc_context would be 730 * set to NULL. 731 * 732 * If xfs_dialloc() does not have an available inode, it will replenish 733 * its supply by doing an allocation. Since we can only do one 734 * allocation within a transaction without deadlocks, we must commit 735 * the current transaction before returning the inode itself. 736 * In this case, therefore, we will set ialloc_context and return. 737 * The caller should then commit the current transaction, start a new 738 * transaction, and call xfs_ialloc() again to actually get the inode. 739 * 740 * To ensure that some other process does not grab the inode that 741 * was allocated during the first call to xfs_ialloc(), this routine 742 * also returns the [locked] bp pointing to the head of the freelist 743 * as ialloc_context. The caller should hold this buffer across 744 * the commit and pass it back into this routine on the second call. 745 * 746 * If we are allocating quota inodes, we do not have a parent inode 747 * to attach to or associate with (i.e. pip == NULL) because they 748 * are not linked into the directory structure - they are attached 749 * directly to the superblock - and so have no parent. 750 */ 751 int 752 xfs_ialloc( 753 xfs_trans_t *tp, 754 xfs_inode_t *pip, 755 umode_t mode, 756 xfs_nlink_t nlink, 757 xfs_dev_t rdev, 758 prid_t prid, 759 int okalloc, 760 xfs_buf_t **ialloc_context, 761 xfs_inode_t **ipp) 762 { 763 struct xfs_mount *mp = tp->t_mountp; 764 xfs_ino_t ino; 765 xfs_inode_t *ip; 766 uint flags; 767 int error; 768 struct timespec tv; 769 struct inode *inode; 770 771 /* 772 * Call the space management code to pick 773 * the on-disk inode to be allocated. 774 */ 775 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 776 ialloc_context, &ino); 777 if (error) 778 return error; 779 if (*ialloc_context || ino == NULLFSINO) { 780 *ipp = NULL; 781 return 0; 782 } 783 ASSERT(*ialloc_context == NULL); 784 785 /* 786 * Get the in-core inode with the lock held exclusively. 787 * This is because we're setting fields here we need 788 * to prevent others from looking at until we're done. 789 */ 790 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 791 XFS_ILOCK_EXCL, &ip); 792 if (error) 793 return error; 794 ASSERT(ip != NULL); 795 inode = VFS_I(ip); 796 797 /* 798 * We always convert v1 inodes to v2 now - we only support filesystems 799 * with >= v2 inode capability, so there is no reason for ever leaving 800 * an inode in v1 format. 801 */ 802 if (ip->i_d.di_version == 1) 803 ip->i_d.di_version = 2; 804 805 inode->i_mode = mode; 806 set_nlink(inode, nlink); 807 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 808 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 809 xfs_set_projid(ip, prid); 810 811 if (pip && XFS_INHERIT_GID(pip)) { 812 ip->i_d.di_gid = pip->i_d.di_gid; 813 if ((VFS_I(pip)->i_mode & S_ISGID) && S_ISDIR(mode)) 814 inode->i_mode |= S_ISGID; 815 } 816 817 /* 818 * If the group ID of the new file does not match the effective group 819 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 820 * (and only if the irix_sgid_inherit compatibility variable is set). 821 */ 822 if ((irix_sgid_inherit) && 823 (inode->i_mode & S_ISGID) && 824 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) 825 inode->i_mode &= ~S_ISGID; 826 827 ip->i_d.di_size = 0; 828 ip->i_d.di_nextents = 0; 829 ASSERT(ip->i_d.di_nblocks == 0); 830 831 tv = current_fs_time(mp->m_super); 832 inode->i_mtime = tv; 833 inode->i_atime = tv; 834 inode->i_ctime = tv; 835 836 ip->i_d.di_extsize = 0; 837 ip->i_d.di_dmevmask = 0; 838 ip->i_d.di_dmstate = 0; 839 ip->i_d.di_flags = 0; 840 841 if (ip->i_d.di_version == 3) { 842 inode->i_version = 1; 843 ip->i_d.di_flags2 = 0; 844 ip->i_d.di_crtime.t_sec = (__int32_t)tv.tv_sec; 845 ip->i_d.di_crtime.t_nsec = (__int32_t)tv.tv_nsec; 846 } 847 848 849 flags = XFS_ILOG_CORE; 850 switch (mode & S_IFMT) { 851 case S_IFIFO: 852 case S_IFCHR: 853 case S_IFBLK: 854 case S_IFSOCK: 855 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 856 ip->i_df.if_u2.if_rdev = rdev; 857 ip->i_df.if_flags = 0; 858 flags |= XFS_ILOG_DEV; 859 break; 860 case S_IFREG: 861 case S_IFDIR: 862 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 863 uint64_t di_flags2 = 0; 864 uint di_flags = 0; 865 866 if (S_ISDIR(mode)) { 867 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 868 di_flags |= XFS_DIFLAG_RTINHERIT; 869 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 870 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 871 ip->i_d.di_extsize = pip->i_d.di_extsize; 872 } 873 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 874 di_flags |= XFS_DIFLAG_PROJINHERIT; 875 } else if (S_ISREG(mode)) { 876 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 877 di_flags |= XFS_DIFLAG_REALTIME; 878 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 879 di_flags |= XFS_DIFLAG_EXTSIZE; 880 ip->i_d.di_extsize = pip->i_d.di_extsize; 881 } 882 } 883 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 884 xfs_inherit_noatime) 885 di_flags |= XFS_DIFLAG_NOATIME; 886 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 887 xfs_inherit_nodump) 888 di_flags |= XFS_DIFLAG_NODUMP; 889 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 890 xfs_inherit_sync) 891 di_flags |= XFS_DIFLAG_SYNC; 892 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 893 xfs_inherit_nosymlinks) 894 di_flags |= XFS_DIFLAG_NOSYMLINKS; 895 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 896 xfs_inherit_nodefrag) 897 di_flags |= XFS_DIFLAG_NODEFRAG; 898 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 899 di_flags |= XFS_DIFLAG_FILESTREAM; 900 if (pip->i_d.di_flags2 & XFS_DIFLAG2_DAX) 901 di_flags2 |= XFS_DIFLAG2_DAX; 902 903 ip->i_d.di_flags |= di_flags; 904 ip->i_d.di_flags2 |= di_flags2; 905 } 906 /* FALLTHROUGH */ 907 case S_IFLNK: 908 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 909 ip->i_df.if_flags = XFS_IFEXTENTS; 910 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 911 ip->i_df.if_u1.if_extents = NULL; 912 break; 913 default: 914 ASSERT(0); 915 } 916 /* 917 * Attribute fork settings for new inode. 918 */ 919 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 920 ip->i_d.di_anextents = 0; 921 922 /* 923 * Log the new values stuffed into the inode. 924 */ 925 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 926 xfs_trans_log_inode(tp, ip, flags); 927 928 /* now that we have an i_mode we can setup the inode structure */ 929 xfs_setup_inode(ip); 930 931 *ipp = ip; 932 return 0; 933 } 934 935 /* 936 * Allocates a new inode from disk and return a pointer to the 937 * incore copy. This routine will internally commit the current 938 * transaction and allocate a new one if the Space Manager needed 939 * to do an allocation to replenish the inode free-list. 940 * 941 * This routine is designed to be called from xfs_create and 942 * xfs_create_dir. 943 * 944 */ 945 int 946 xfs_dir_ialloc( 947 xfs_trans_t **tpp, /* input: current transaction; 948 output: may be a new transaction. */ 949 xfs_inode_t *dp, /* directory within whose allocate 950 the inode. */ 951 umode_t mode, 952 xfs_nlink_t nlink, 953 xfs_dev_t rdev, 954 prid_t prid, /* project id */ 955 int okalloc, /* ok to allocate new space */ 956 xfs_inode_t **ipp, /* pointer to inode; it will be 957 locked. */ 958 int *committed) 959 960 { 961 xfs_trans_t *tp; 962 xfs_inode_t *ip; 963 xfs_buf_t *ialloc_context = NULL; 964 int code; 965 void *dqinfo; 966 uint tflags; 967 968 tp = *tpp; 969 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 970 971 /* 972 * xfs_ialloc will return a pointer to an incore inode if 973 * the Space Manager has an available inode on the free 974 * list. Otherwise, it will do an allocation and replenish 975 * the freelist. Since we can only do one allocation per 976 * transaction without deadlocks, we will need to commit the 977 * current transaction and start a new one. We will then 978 * need to call xfs_ialloc again to get the inode. 979 * 980 * If xfs_ialloc did an allocation to replenish the freelist, 981 * it returns the bp containing the head of the freelist as 982 * ialloc_context. We will hold a lock on it across the 983 * transaction commit so that no other process can steal 984 * the inode(s) that we've just allocated. 985 */ 986 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 987 &ialloc_context, &ip); 988 989 /* 990 * Return an error if we were unable to allocate a new inode. 991 * This should only happen if we run out of space on disk or 992 * encounter a disk error. 993 */ 994 if (code) { 995 *ipp = NULL; 996 return code; 997 } 998 if (!ialloc_context && !ip) { 999 *ipp = NULL; 1000 return -ENOSPC; 1001 } 1002 1003 /* 1004 * If the AGI buffer is non-NULL, then we were unable to get an 1005 * inode in one operation. We need to commit the current 1006 * transaction and call xfs_ialloc() again. It is guaranteed 1007 * to succeed the second time. 1008 */ 1009 if (ialloc_context) { 1010 /* 1011 * Normally, xfs_trans_commit releases all the locks. 1012 * We call bhold to hang on to the ialloc_context across 1013 * the commit. Holding this buffer prevents any other 1014 * processes from doing any allocations in this 1015 * allocation group. 1016 */ 1017 xfs_trans_bhold(tp, ialloc_context); 1018 1019 /* 1020 * We want the quota changes to be associated with the next 1021 * transaction, NOT this one. So, detach the dqinfo from this 1022 * and attach it to the next transaction. 1023 */ 1024 dqinfo = NULL; 1025 tflags = 0; 1026 if (tp->t_dqinfo) { 1027 dqinfo = (void *)tp->t_dqinfo; 1028 tp->t_dqinfo = NULL; 1029 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 1030 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 1031 } 1032 1033 code = xfs_trans_roll(&tp, NULL); 1034 if (committed != NULL) 1035 *committed = 1; 1036 1037 /* 1038 * Re-attach the quota info that we detached from prev trx. 1039 */ 1040 if (dqinfo) { 1041 tp->t_dqinfo = dqinfo; 1042 tp->t_flags |= tflags; 1043 } 1044 1045 if (code) { 1046 xfs_buf_relse(ialloc_context); 1047 *tpp = tp; 1048 *ipp = NULL; 1049 return code; 1050 } 1051 xfs_trans_bjoin(tp, ialloc_context); 1052 1053 /* 1054 * Call ialloc again. Since we've locked out all 1055 * other allocations in this allocation group, 1056 * this call should always succeed. 1057 */ 1058 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1059 okalloc, &ialloc_context, &ip); 1060 1061 /* 1062 * If we get an error at this point, return to the caller 1063 * so that the current transaction can be aborted. 1064 */ 1065 if (code) { 1066 *tpp = tp; 1067 *ipp = NULL; 1068 return code; 1069 } 1070 ASSERT(!ialloc_context && ip); 1071 1072 } else { 1073 if (committed != NULL) 1074 *committed = 0; 1075 } 1076 1077 *ipp = ip; 1078 *tpp = tp; 1079 1080 return 0; 1081 } 1082 1083 /* 1084 * Decrement the link count on an inode & log the change. If this causes the 1085 * link count to go to zero, move the inode to AGI unlinked list so that it can 1086 * be freed when the last active reference goes away via xfs_inactive(). 1087 */ 1088 int /* error */ 1089 xfs_droplink( 1090 xfs_trans_t *tp, 1091 xfs_inode_t *ip) 1092 { 1093 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1094 1095 drop_nlink(VFS_I(ip)); 1096 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1097 1098 if (VFS_I(ip)->i_nlink) 1099 return 0; 1100 1101 return xfs_iunlink(tp, ip); 1102 } 1103 1104 /* 1105 * Increment the link count on an inode & log the change. 1106 */ 1107 int 1108 xfs_bumplink( 1109 xfs_trans_t *tp, 1110 xfs_inode_t *ip) 1111 { 1112 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1113 1114 ASSERT(ip->i_d.di_version > 1); 1115 inc_nlink(VFS_I(ip)); 1116 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1117 return 0; 1118 } 1119 1120 int 1121 xfs_create( 1122 xfs_inode_t *dp, 1123 struct xfs_name *name, 1124 umode_t mode, 1125 xfs_dev_t rdev, 1126 xfs_inode_t **ipp) 1127 { 1128 int is_dir = S_ISDIR(mode); 1129 struct xfs_mount *mp = dp->i_mount; 1130 struct xfs_inode *ip = NULL; 1131 struct xfs_trans *tp = NULL; 1132 int error; 1133 xfs_bmap_free_t free_list; 1134 xfs_fsblock_t first_block; 1135 bool unlock_dp_on_error = false; 1136 prid_t prid; 1137 struct xfs_dquot *udqp = NULL; 1138 struct xfs_dquot *gdqp = NULL; 1139 struct xfs_dquot *pdqp = NULL; 1140 struct xfs_trans_res *tres; 1141 uint resblks; 1142 1143 trace_xfs_create(dp, name); 1144 1145 if (XFS_FORCED_SHUTDOWN(mp)) 1146 return -EIO; 1147 1148 prid = xfs_get_initial_prid(dp); 1149 1150 /* 1151 * Make sure that we have allocated dquot(s) on disk. 1152 */ 1153 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1154 xfs_kgid_to_gid(current_fsgid()), prid, 1155 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1156 &udqp, &gdqp, &pdqp); 1157 if (error) 1158 return error; 1159 1160 if (is_dir) { 1161 rdev = 0; 1162 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1163 tres = &M_RES(mp)->tr_mkdir; 1164 } else { 1165 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1166 tres = &M_RES(mp)->tr_create; 1167 } 1168 1169 /* 1170 * Initially assume that the file does not exist and 1171 * reserve the resources for that case. If that is not 1172 * the case we'll drop the one we have and get a more 1173 * appropriate transaction later. 1174 */ 1175 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1176 if (error == -ENOSPC) { 1177 /* flush outstanding delalloc blocks and retry */ 1178 xfs_flush_inodes(mp); 1179 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1180 } 1181 if (error == -ENOSPC) { 1182 /* No space at all so try a "no-allocation" reservation */ 1183 resblks = 0; 1184 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1185 } 1186 if (error) 1187 goto out_release_inode; 1188 1189 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL | 1190 XFS_IOLOCK_PARENT | XFS_ILOCK_PARENT); 1191 unlock_dp_on_error = true; 1192 1193 xfs_bmap_init(&free_list, &first_block); 1194 1195 /* 1196 * Reserve disk quota and the inode. 1197 */ 1198 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1199 pdqp, resblks, 1, 0); 1200 if (error) 1201 goto out_trans_cancel; 1202 1203 if (!resblks) { 1204 error = xfs_dir_canenter(tp, dp, name); 1205 if (error) 1206 goto out_trans_cancel; 1207 } 1208 1209 /* 1210 * A newly created regular or special file just has one directory 1211 * entry pointing to them, but a directory also the "." entry 1212 * pointing to itself. 1213 */ 1214 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1215 prid, resblks > 0, &ip, NULL); 1216 if (error) 1217 goto out_trans_cancel; 1218 1219 /* 1220 * Now we join the directory inode to the transaction. We do not do it 1221 * earlier because xfs_dir_ialloc might commit the previous transaction 1222 * (and release all the locks). An error from here on will result in 1223 * the transaction cancel unlocking dp so don't do it explicitly in the 1224 * error path. 1225 */ 1226 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1227 unlock_dp_on_error = false; 1228 1229 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1230 &first_block, &free_list, resblks ? 1231 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1232 if (error) { 1233 ASSERT(error != -ENOSPC); 1234 goto out_trans_cancel; 1235 } 1236 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1237 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1238 1239 if (is_dir) { 1240 error = xfs_dir_init(tp, ip, dp); 1241 if (error) 1242 goto out_bmap_cancel; 1243 1244 error = xfs_bumplink(tp, dp); 1245 if (error) 1246 goto out_bmap_cancel; 1247 } 1248 1249 /* 1250 * If this is a synchronous mount, make sure that the 1251 * create transaction goes to disk before returning to 1252 * the user. 1253 */ 1254 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1255 xfs_trans_set_sync(tp); 1256 1257 /* 1258 * Attach the dquot(s) to the inodes and modify them incore. 1259 * These ids of the inode couldn't have changed since the new 1260 * inode has been locked ever since it was created. 1261 */ 1262 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1263 1264 error = xfs_bmap_finish(&tp, &free_list, NULL); 1265 if (error) 1266 goto out_bmap_cancel; 1267 1268 error = xfs_trans_commit(tp); 1269 if (error) 1270 goto out_release_inode; 1271 1272 xfs_qm_dqrele(udqp); 1273 xfs_qm_dqrele(gdqp); 1274 xfs_qm_dqrele(pdqp); 1275 1276 *ipp = ip; 1277 return 0; 1278 1279 out_bmap_cancel: 1280 xfs_bmap_cancel(&free_list); 1281 out_trans_cancel: 1282 xfs_trans_cancel(tp); 1283 out_release_inode: 1284 /* 1285 * Wait until after the current transaction is aborted to finish the 1286 * setup of the inode and release the inode. This prevents recursive 1287 * transactions and deadlocks from xfs_inactive. 1288 */ 1289 if (ip) { 1290 xfs_finish_inode_setup(ip); 1291 IRELE(ip); 1292 } 1293 1294 xfs_qm_dqrele(udqp); 1295 xfs_qm_dqrele(gdqp); 1296 xfs_qm_dqrele(pdqp); 1297 1298 if (unlock_dp_on_error) 1299 xfs_iunlock(dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1300 return error; 1301 } 1302 1303 int 1304 xfs_create_tmpfile( 1305 struct xfs_inode *dp, 1306 struct dentry *dentry, 1307 umode_t mode, 1308 struct xfs_inode **ipp) 1309 { 1310 struct xfs_mount *mp = dp->i_mount; 1311 struct xfs_inode *ip = NULL; 1312 struct xfs_trans *tp = NULL; 1313 int error; 1314 prid_t prid; 1315 struct xfs_dquot *udqp = NULL; 1316 struct xfs_dquot *gdqp = NULL; 1317 struct xfs_dquot *pdqp = NULL; 1318 struct xfs_trans_res *tres; 1319 uint resblks; 1320 1321 if (XFS_FORCED_SHUTDOWN(mp)) 1322 return -EIO; 1323 1324 prid = xfs_get_initial_prid(dp); 1325 1326 /* 1327 * Make sure that we have allocated dquot(s) on disk. 1328 */ 1329 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1330 xfs_kgid_to_gid(current_fsgid()), prid, 1331 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1332 &udqp, &gdqp, &pdqp); 1333 if (error) 1334 return error; 1335 1336 resblks = XFS_IALLOC_SPACE_RES(mp); 1337 tres = &M_RES(mp)->tr_create_tmpfile; 1338 1339 error = xfs_trans_alloc(mp, tres, resblks, 0, 0, &tp); 1340 if (error == -ENOSPC) { 1341 /* No space at all so try a "no-allocation" reservation */ 1342 resblks = 0; 1343 error = xfs_trans_alloc(mp, tres, 0, 0, 0, &tp); 1344 } 1345 if (error) 1346 goto out_release_inode; 1347 1348 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1349 pdqp, resblks, 1, 0); 1350 if (error) 1351 goto out_trans_cancel; 1352 1353 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1354 prid, resblks > 0, &ip, NULL); 1355 if (error) 1356 goto out_trans_cancel; 1357 1358 if (mp->m_flags & XFS_MOUNT_WSYNC) 1359 xfs_trans_set_sync(tp); 1360 1361 /* 1362 * Attach the dquot(s) to the inodes and modify them incore. 1363 * These ids of the inode couldn't have changed since the new 1364 * inode has been locked ever since it was created. 1365 */ 1366 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1367 1368 error = xfs_iunlink(tp, ip); 1369 if (error) 1370 goto out_trans_cancel; 1371 1372 error = xfs_trans_commit(tp); 1373 if (error) 1374 goto out_release_inode; 1375 1376 xfs_qm_dqrele(udqp); 1377 xfs_qm_dqrele(gdqp); 1378 xfs_qm_dqrele(pdqp); 1379 1380 *ipp = ip; 1381 return 0; 1382 1383 out_trans_cancel: 1384 xfs_trans_cancel(tp); 1385 out_release_inode: 1386 /* 1387 * Wait until after the current transaction is aborted to finish the 1388 * setup of the inode and release the inode. This prevents recursive 1389 * transactions and deadlocks from xfs_inactive. 1390 */ 1391 if (ip) { 1392 xfs_finish_inode_setup(ip); 1393 IRELE(ip); 1394 } 1395 1396 xfs_qm_dqrele(udqp); 1397 xfs_qm_dqrele(gdqp); 1398 xfs_qm_dqrele(pdqp); 1399 1400 return error; 1401 } 1402 1403 int 1404 xfs_link( 1405 xfs_inode_t *tdp, 1406 xfs_inode_t *sip, 1407 struct xfs_name *target_name) 1408 { 1409 xfs_mount_t *mp = tdp->i_mount; 1410 xfs_trans_t *tp; 1411 int error; 1412 xfs_bmap_free_t free_list; 1413 xfs_fsblock_t first_block; 1414 int resblks; 1415 1416 trace_xfs_link(tdp, target_name); 1417 1418 ASSERT(!S_ISDIR(VFS_I(sip)->i_mode)); 1419 1420 if (XFS_FORCED_SHUTDOWN(mp)) 1421 return -EIO; 1422 1423 error = xfs_qm_dqattach(sip, 0); 1424 if (error) 1425 goto std_return; 1426 1427 error = xfs_qm_dqattach(tdp, 0); 1428 if (error) 1429 goto std_return; 1430 1431 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1432 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp); 1433 if (error == -ENOSPC) { 1434 resblks = 0; 1435 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp); 1436 } 1437 if (error) 1438 goto std_return; 1439 1440 xfs_ilock(tdp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 1441 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1442 1443 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1444 xfs_trans_ijoin(tp, tdp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 1445 1446 /* 1447 * If we are using project inheritance, we only allow hard link 1448 * creation in our tree when the project IDs are the same; else 1449 * the tree quota mechanism could be circumvented. 1450 */ 1451 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1452 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1453 error = -EXDEV; 1454 goto error_return; 1455 } 1456 1457 if (!resblks) { 1458 error = xfs_dir_canenter(tp, tdp, target_name); 1459 if (error) 1460 goto error_return; 1461 } 1462 1463 xfs_bmap_init(&free_list, &first_block); 1464 1465 /* 1466 * Handle initial link state of O_TMPFILE inode 1467 */ 1468 if (VFS_I(sip)->i_nlink == 0) { 1469 error = xfs_iunlink_remove(tp, sip); 1470 if (error) 1471 goto error_return; 1472 } 1473 1474 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1475 &first_block, &free_list, resblks); 1476 if (error) 1477 goto error_return; 1478 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1479 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1480 1481 error = xfs_bumplink(tp, sip); 1482 if (error) 1483 goto error_return; 1484 1485 /* 1486 * If this is a synchronous mount, make sure that the 1487 * link transaction goes to disk before returning to 1488 * the user. 1489 */ 1490 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1491 xfs_trans_set_sync(tp); 1492 1493 error = xfs_bmap_finish(&tp, &free_list, NULL); 1494 if (error) { 1495 xfs_bmap_cancel(&free_list); 1496 goto error_return; 1497 } 1498 1499 return xfs_trans_commit(tp); 1500 1501 error_return: 1502 xfs_trans_cancel(tp); 1503 std_return: 1504 return error; 1505 } 1506 1507 /* 1508 * Free up the underlying blocks past new_size. The new size must be smaller 1509 * than the current size. This routine can be used both for the attribute and 1510 * data fork, and does not modify the inode size, which is left to the caller. 1511 * 1512 * The transaction passed to this routine must have made a permanent log 1513 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1514 * given transaction and start new ones, so make sure everything involved in 1515 * the transaction is tidy before calling here. Some transaction will be 1516 * returned to the caller to be committed. The incoming transaction must 1517 * already include the inode, and both inode locks must be held exclusively. 1518 * The inode must also be "held" within the transaction. On return the inode 1519 * will be "held" within the returned transaction. This routine does NOT 1520 * require any disk space to be reserved for it within the transaction. 1521 * 1522 * If we get an error, we must return with the inode locked and linked into the 1523 * current transaction. This keeps things simple for the higher level code, 1524 * because it always knows that the inode is locked and held in the transaction 1525 * that returns to it whether errors occur or not. We don't mark the inode 1526 * dirty on error so that transactions can be easily aborted if possible. 1527 */ 1528 int 1529 xfs_itruncate_extents( 1530 struct xfs_trans **tpp, 1531 struct xfs_inode *ip, 1532 int whichfork, 1533 xfs_fsize_t new_size) 1534 { 1535 struct xfs_mount *mp = ip->i_mount; 1536 struct xfs_trans *tp = *tpp; 1537 xfs_bmap_free_t free_list; 1538 xfs_fsblock_t first_block; 1539 xfs_fileoff_t first_unmap_block; 1540 xfs_fileoff_t last_block; 1541 xfs_filblks_t unmap_len; 1542 int error = 0; 1543 int done = 0; 1544 1545 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1546 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1547 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1548 ASSERT(new_size <= XFS_ISIZE(ip)); 1549 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1550 ASSERT(ip->i_itemp != NULL); 1551 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1552 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1553 1554 trace_xfs_itruncate_extents_start(ip, new_size); 1555 1556 /* 1557 * Since it is possible for space to become allocated beyond 1558 * the end of the file (in a crash where the space is allocated 1559 * but the inode size is not yet updated), simply remove any 1560 * blocks which show up between the new EOF and the maximum 1561 * possible file size. If the first block to be removed is 1562 * beyond the maximum file size (ie it is the same as last_block), 1563 * then there is nothing to do. 1564 */ 1565 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1566 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1567 if (first_unmap_block == last_block) 1568 return 0; 1569 1570 ASSERT(first_unmap_block < last_block); 1571 unmap_len = last_block - first_unmap_block + 1; 1572 while (!done) { 1573 xfs_bmap_init(&free_list, &first_block); 1574 error = xfs_bunmapi(tp, ip, 1575 first_unmap_block, unmap_len, 1576 xfs_bmapi_aflag(whichfork), 1577 XFS_ITRUNC_MAX_EXTENTS, 1578 &first_block, &free_list, 1579 &done); 1580 if (error) 1581 goto out_bmap_cancel; 1582 1583 /* 1584 * Duplicate the transaction that has the permanent 1585 * reservation and commit the old transaction. 1586 */ 1587 error = xfs_bmap_finish(&tp, &free_list, ip); 1588 if (error) 1589 goto out_bmap_cancel; 1590 1591 error = xfs_trans_roll(&tp, ip); 1592 if (error) 1593 goto out; 1594 } 1595 1596 /* 1597 * Always re-log the inode so that our permanent transaction can keep 1598 * on rolling it forward in the log. 1599 */ 1600 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1601 1602 trace_xfs_itruncate_extents_end(ip, new_size); 1603 1604 out: 1605 *tpp = tp; 1606 return error; 1607 out_bmap_cancel: 1608 /* 1609 * If the bunmapi call encounters an error, return to the caller where 1610 * the transaction can be properly aborted. We just need to make sure 1611 * we're not holding any resources that we were not when we came in. 1612 */ 1613 xfs_bmap_cancel(&free_list); 1614 goto out; 1615 } 1616 1617 int 1618 xfs_release( 1619 xfs_inode_t *ip) 1620 { 1621 xfs_mount_t *mp = ip->i_mount; 1622 int error; 1623 1624 if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0)) 1625 return 0; 1626 1627 /* If this is a read-only mount, don't do this (would generate I/O) */ 1628 if (mp->m_flags & XFS_MOUNT_RDONLY) 1629 return 0; 1630 1631 if (!XFS_FORCED_SHUTDOWN(mp)) { 1632 int truncated; 1633 1634 /* 1635 * If we previously truncated this file and removed old data 1636 * in the process, we want to initiate "early" writeout on 1637 * the last close. This is an attempt to combat the notorious 1638 * NULL files problem which is particularly noticeable from a 1639 * truncate down, buffered (re-)write (delalloc), followed by 1640 * a crash. What we are effectively doing here is 1641 * significantly reducing the time window where we'd otherwise 1642 * be exposed to that problem. 1643 */ 1644 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1645 if (truncated) { 1646 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1647 if (ip->i_delayed_blks > 0) { 1648 error = filemap_flush(VFS_I(ip)->i_mapping); 1649 if (error) 1650 return error; 1651 } 1652 } 1653 } 1654 1655 if (VFS_I(ip)->i_nlink == 0) 1656 return 0; 1657 1658 if (xfs_can_free_eofblocks(ip, false)) { 1659 1660 /* 1661 * If we can't get the iolock just skip truncating the blocks 1662 * past EOF because we could deadlock with the mmap_sem 1663 * otherwise. We'll get another chance to drop them once the 1664 * last reference to the inode is dropped, so we'll never leak 1665 * blocks permanently. 1666 * 1667 * Further, check if the inode is being opened, written and 1668 * closed frequently and we have delayed allocation blocks 1669 * outstanding (e.g. streaming writes from the NFS server), 1670 * truncating the blocks past EOF will cause fragmentation to 1671 * occur. 1672 * 1673 * In this case don't do the truncation, either, but we have to 1674 * be careful how we detect this case. Blocks beyond EOF show 1675 * up as i_delayed_blks even when the inode is clean, so we 1676 * need to truncate them away first before checking for a dirty 1677 * release. Hence on the first dirty close we will still remove 1678 * the speculative allocation, but after that we will leave it 1679 * in place. 1680 */ 1681 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1682 return 0; 1683 1684 error = xfs_free_eofblocks(mp, ip, true); 1685 if (error && error != -EAGAIN) 1686 return error; 1687 1688 /* delalloc blocks after truncation means it really is dirty */ 1689 if (ip->i_delayed_blks) 1690 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1691 } 1692 return 0; 1693 } 1694 1695 /* 1696 * xfs_inactive_truncate 1697 * 1698 * Called to perform a truncate when an inode becomes unlinked. 1699 */ 1700 STATIC int 1701 xfs_inactive_truncate( 1702 struct xfs_inode *ip) 1703 { 1704 struct xfs_mount *mp = ip->i_mount; 1705 struct xfs_trans *tp; 1706 int error; 1707 1708 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 1709 if (error) { 1710 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1711 return error; 1712 } 1713 1714 xfs_ilock(ip, XFS_ILOCK_EXCL); 1715 xfs_trans_ijoin(tp, ip, 0); 1716 1717 /* 1718 * Log the inode size first to prevent stale data exposure in the event 1719 * of a system crash before the truncate completes. See the related 1720 * comment in xfs_setattr_size() for details. 1721 */ 1722 ip->i_d.di_size = 0; 1723 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1724 1725 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1726 if (error) 1727 goto error_trans_cancel; 1728 1729 ASSERT(ip->i_d.di_nextents == 0); 1730 1731 error = xfs_trans_commit(tp); 1732 if (error) 1733 goto error_unlock; 1734 1735 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1736 return 0; 1737 1738 error_trans_cancel: 1739 xfs_trans_cancel(tp); 1740 error_unlock: 1741 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1742 return error; 1743 } 1744 1745 /* 1746 * xfs_inactive_ifree() 1747 * 1748 * Perform the inode free when an inode is unlinked. 1749 */ 1750 STATIC int 1751 xfs_inactive_ifree( 1752 struct xfs_inode *ip) 1753 { 1754 xfs_bmap_free_t free_list; 1755 xfs_fsblock_t first_block; 1756 struct xfs_mount *mp = ip->i_mount; 1757 struct xfs_trans *tp; 1758 int error; 1759 1760 /* 1761 * The ifree transaction might need to allocate blocks for record 1762 * insertion to the finobt. We don't want to fail here at ENOSPC, so 1763 * allow ifree to dip into the reserved block pool if necessary. 1764 * 1765 * Freeing large sets of inodes generally means freeing inode chunks, 1766 * directory and file data blocks, so this should be relatively safe. 1767 * Only under severe circumstances should it be possible to free enough 1768 * inodes to exhaust the reserve block pool via finobt expansion while 1769 * at the same time not creating free space in the filesystem. 1770 * 1771 * Send a warning if the reservation does happen to fail, as the inode 1772 * now remains allocated and sits on the unlinked list until the fs is 1773 * repaired. 1774 */ 1775 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 1776 XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp); 1777 if (error) { 1778 if (error == -ENOSPC) { 1779 xfs_warn_ratelimited(mp, 1780 "Failed to remove inode(s) from unlinked list. " 1781 "Please free space, unmount and run xfs_repair."); 1782 } else { 1783 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1784 } 1785 return error; 1786 } 1787 1788 xfs_ilock(ip, XFS_ILOCK_EXCL); 1789 xfs_trans_ijoin(tp, ip, 0); 1790 1791 xfs_bmap_init(&free_list, &first_block); 1792 error = xfs_ifree(tp, ip, &free_list); 1793 if (error) { 1794 /* 1795 * If we fail to free the inode, shut down. The cancel 1796 * might do that, we need to make sure. Otherwise the 1797 * inode might be lost for a long time or forever. 1798 */ 1799 if (!XFS_FORCED_SHUTDOWN(mp)) { 1800 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1801 __func__, error); 1802 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1803 } 1804 xfs_trans_cancel(tp); 1805 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1806 return error; 1807 } 1808 1809 /* 1810 * Credit the quota account(s). The inode is gone. 1811 */ 1812 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1813 1814 /* 1815 * Just ignore errors at this point. There is nothing we can do except 1816 * to try to keep going. Make sure it's not a silent error. 1817 */ 1818 error = xfs_bmap_finish(&tp, &free_list, NULL); 1819 if (error) { 1820 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", 1821 __func__, error); 1822 xfs_bmap_cancel(&free_list); 1823 } 1824 error = xfs_trans_commit(tp); 1825 if (error) 1826 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1827 __func__, error); 1828 1829 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1830 return 0; 1831 } 1832 1833 /* 1834 * xfs_inactive 1835 * 1836 * This is called when the vnode reference count for the vnode 1837 * goes to zero. If the file has been unlinked, then it must 1838 * now be truncated. Also, we clear all of the read-ahead state 1839 * kept for the inode here since the file is now closed. 1840 */ 1841 void 1842 xfs_inactive( 1843 xfs_inode_t *ip) 1844 { 1845 struct xfs_mount *mp; 1846 int error; 1847 int truncate = 0; 1848 1849 /* 1850 * If the inode is already free, then there can be nothing 1851 * to clean up here. 1852 */ 1853 if (VFS_I(ip)->i_mode == 0) { 1854 ASSERT(ip->i_df.if_real_bytes == 0); 1855 ASSERT(ip->i_df.if_broot_bytes == 0); 1856 return; 1857 } 1858 1859 mp = ip->i_mount; 1860 1861 /* If this is a read-only mount, don't do this (would generate I/O) */ 1862 if (mp->m_flags & XFS_MOUNT_RDONLY) 1863 return; 1864 1865 if (VFS_I(ip)->i_nlink != 0) { 1866 /* 1867 * force is true because we are evicting an inode from the 1868 * cache. Post-eof blocks must be freed, lest we end up with 1869 * broken free space accounting. 1870 */ 1871 if (xfs_can_free_eofblocks(ip, true)) 1872 xfs_free_eofblocks(mp, ip, false); 1873 1874 return; 1875 } 1876 1877 if (S_ISREG(VFS_I(ip)->i_mode) && 1878 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1879 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1880 truncate = 1; 1881 1882 error = xfs_qm_dqattach(ip, 0); 1883 if (error) 1884 return; 1885 1886 if (S_ISLNK(VFS_I(ip)->i_mode)) 1887 error = xfs_inactive_symlink(ip); 1888 else if (truncate) 1889 error = xfs_inactive_truncate(ip); 1890 if (error) 1891 return; 1892 1893 /* 1894 * If there are attributes associated with the file then blow them away 1895 * now. The code calls a routine that recursively deconstructs the 1896 * attribute fork. If also blows away the in-core attribute fork. 1897 */ 1898 if (XFS_IFORK_Q(ip)) { 1899 error = xfs_attr_inactive(ip); 1900 if (error) 1901 return; 1902 } 1903 1904 ASSERT(!ip->i_afp); 1905 ASSERT(ip->i_d.di_anextents == 0); 1906 ASSERT(ip->i_d.di_forkoff == 0); 1907 1908 /* 1909 * Free the inode. 1910 */ 1911 error = xfs_inactive_ifree(ip); 1912 if (error) 1913 return; 1914 1915 /* 1916 * Release the dquots held by inode, if any. 1917 */ 1918 xfs_qm_dqdetach(ip); 1919 } 1920 1921 /* 1922 * This is called when the inode's link count goes to 0 or we are creating a 1923 * tmpfile via O_TMPFILE. In the case of a tmpfile, @ignore_linkcount will be 1924 * set to true as the link count is dropped to zero by the VFS after we've 1925 * created the file successfully, so we have to add it to the unlinked list 1926 * while the link count is non-zero. 1927 * 1928 * We place the on-disk inode on a list in the AGI. It will be pulled from this 1929 * list when the inode is freed. 1930 */ 1931 STATIC int 1932 xfs_iunlink( 1933 struct xfs_trans *tp, 1934 struct xfs_inode *ip) 1935 { 1936 xfs_mount_t *mp = tp->t_mountp; 1937 xfs_agi_t *agi; 1938 xfs_dinode_t *dip; 1939 xfs_buf_t *agibp; 1940 xfs_buf_t *ibp; 1941 xfs_agino_t agino; 1942 short bucket_index; 1943 int offset; 1944 int error; 1945 1946 ASSERT(VFS_I(ip)->i_mode != 0); 1947 1948 /* 1949 * Get the agi buffer first. It ensures lock ordering 1950 * on the list. 1951 */ 1952 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1953 if (error) 1954 return error; 1955 agi = XFS_BUF_TO_AGI(agibp); 1956 1957 /* 1958 * Get the index into the agi hash table for the 1959 * list this inode will go on. 1960 */ 1961 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1962 ASSERT(agino != 0); 1963 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1964 ASSERT(agi->agi_unlinked[bucket_index]); 1965 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1966 1967 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1968 /* 1969 * There is already another inode in the bucket we need 1970 * to add ourselves to. Add us at the front of the list. 1971 * Here we put the head pointer into our next pointer, 1972 * and then we fall through to point the head at us. 1973 */ 1974 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1975 0, 0); 1976 if (error) 1977 return error; 1978 1979 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1980 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1981 offset = ip->i_imap.im_boffset + 1982 offsetof(xfs_dinode_t, di_next_unlinked); 1983 1984 /* need to recalc the inode CRC if appropriate */ 1985 xfs_dinode_calc_crc(mp, dip); 1986 1987 xfs_trans_inode_buf(tp, ibp); 1988 xfs_trans_log_buf(tp, ibp, offset, 1989 (offset + sizeof(xfs_agino_t) - 1)); 1990 xfs_inobp_check(mp, ibp); 1991 } 1992 1993 /* 1994 * Point the bucket head pointer at the inode being inserted. 1995 */ 1996 ASSERT(agino != 0); 1997 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1998 offset = offsetof(xfs_agi_t, agi_unlinked) + 1999 (sizeof(xfs_agino_t) * bucket_index); 2000 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2001 xfs_trans_log_buf(tp, agibp, offset, 2002 (offset + sizeof(xfs_agino_t) - 1)); 2003 return 0; 2004 } 2005 2006 /* 2007 * Pull the on-disk inode from the AGI unlinked list. 2008 */ 2009 STATIC int 2010 xfs_iunlink_remove( 2011 xfs_trans_t *tp, 2012 xfs_inode_t *ip) 2013 { 2014 xfs_ino_t next_ino; 2015 xfs_mount_t *mp; 2016 xfs_agi_t *agi; 2017 xfs_dinode_t *dip; 2018 xfs_buf_t *agibp; 2019 xfs_buf_t *ibp; 2020 xfs_agnumber_t agno; 2021 xfs_agino_t agino; 2022 xfs_agino_t next_agino; 2023 xfs_buf_t *last_ibp; 2024 xfs_dinode_t *last_dip = NULL; 2025 short bucket_index; 2026 int offset, last_offset = 0; 2027 int error; 2028 2029 mp = tp->t_mountp; 2030 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2031 2032 /* 2033 * Get the agi buffer first. It ensures lock ordering 2034 * on the list. 2035 */ 2036 error = xfs_read_agi(mp, tp, agno, &agibp); 2037 if (error) 2038 return error; 2039 2040 agi = XFS_BUF_TO_AGI(agibp); 2041 2042 /* 2043 * Get the index into the agi hash table for the 2044 * list this inode will go on. 2045 */ 2046 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2047 ASSERT(agino != 0); 2048 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2049 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2050 ASSERT(agi->agi_unlinked[bucket_index]); 2051 2052 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2053 /* 2054 * We're at the head of the list. Get the inode's on-disk 2055 * buffer to see if there is anyone after us on the list. 2056 * Only modify our next pointer if it is not already NULLAGINO. 2057 * This saves us the overhead of dealing with the buffer when 2058 * there is no need to change it. 2059 */ 2060 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2061 0, 0); 2062 if (error) { 2063 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2064 __func__, error); 2065 return error; 2066 } 2067 next_agino = be32_to_cpu(dip->di_next_unlinked); 2068 ASSERT(next_agino != 0); 2069 if (next_agino != NULLAGINO) { 2070 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2071 offset = ip->i_imap.im_boffset + 2072 offsetof(xfs_dinode_t, di_next_unlinked); 2073 2074 /* need to recalc the inode CRC if appropriate */ 2075 xfs_dinode_calc_crc(mp, dip); 2076 2077 xfs_trans_inode_buf(tp, ibp); 2078 xfs_trans_log_buf(tp, ibp, offset, 2079 (offset + sizeof(xfs_agino_t) - 1)); 2080 xfs_inobp_check(mp, ibp); 2081 } else { 2082 xfs_trans_brelse(tp, ibp); 2083 } 2084 /* 2085 * Point the bucket head pointer at the next inode. 2086 */ 2087 ASSERT(next_agino != 0); 2088 ASSERT(next_agino != agino); 2089 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2090 offset = offsetof(xfs_agi_t, agi_unlinked) + 2091 (sizeof(xfs_agino_t) * bucket_index); 2092 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2093 xfs_trans_log_buf(tp, agibp, offset, 2094 (offset + sizeof(xfs_agino_t) - 1)); 2095 } else { 2096 /* 2097 * We need to search the list for the inode being freed. 2098 */ 2099 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2100 last_ibp = NULL; 2101 while (next_agino != agino) { 2102 struct xfs_imap imap; 2103 2104 if (last_ibp) 2105 xfs_trans_brelse(tp, last_ibp); 2106 2107 imap.im_blkno = 0; 2108 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2109 2110 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2111 if (error) { 2112 xfs_warn(mp, 2113 "%s: xfs_imap returned error %d.", 2114 __func__, error); 2115 return error; 2116 } 2117 2118 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2119 &last_ibp, 0, 0); 2120 if (error) { 2121 xfs_warn(mp, 2122 "%s: xfs_imap_to_bp returned error %d.", 2123 __func__, error); 2124 return error; 2125 } 2126 2127 last_offset = imap.im_boffset; 2128 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2129 ASSERT(next_agino != NULLAGINO); 2130 ASSERT(next_agino != 0); 2131 } 2132 2133 /* 2134 * Now last_ibp points to the buffer previous to us on the 2135 * unlinked list. Pull us from the list. 2136 */ 2137 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2138 0, 0); 2139 if (error) { 2140 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2141 __func__, error); 2142 return error; 2143 } 2144 next_agino = be32_to_cpu(dip->di_next_unlinked); 2145 ASSERT(next_agino != 0); 2146 ASSERT(next_agino != agino); 2147 if (next_agino != NULLAGINO) { 2148 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2149 offset = ip->i_imap.im_boffset + 2150 offsetof(xfs_dinode_t, di_next_unlinked); 2151 2152 /* need to recalc the inode CRC if appropriate */ 2153 xfs_dinode_calc_crc(mp, dip); 2154 2155 xfs_trans_inode_buf(tp, ibp); 2156 xfs_trans_log_buf(tp, ibp, offset, 2157 (offset + sizeof(xfs_agino_t) - 1)); 2158 xfs_inobp_check(mp, ibp); 2159 } else { 2160 xfs_trans_brelse(tp, ibp); 2161 } 2162 /* 2163 * Point the previous inode on the list to the next inode. 2164 */ 2165 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2166 ASSERT(next_agino != 0); 2167 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2168 2169 /* need to recalc the inode CRC if appropriate */ 2170 xfs_dinode_calc_crc(mp, last_dip); 2171 2172 xfs_trans_inode_buf(tp, last_ibp); 2173 xfs_trans_log_buf(tp, last_ibp, offset, 2174 (offset + sizeof(xfs_agino_t) - 1)); 2175 xfs_inobp_check(mp, last_ibp); 2176 } 2177 return 0; 2178 } 2179 2180 /* 2181 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2182 * inodes that are in memory - they all must be marked stale and attached to 2183 * the cluster buffer. 2184 */ 2185 STATIC int 2186 xfs_ifree_cluster( 2187 xfs_inode_t *free_ip, 2188 xfs_trans_t *tp, 2189 struct xfs_icluster *xic) 2190 { 2191 xfs_mount_t *mp = free_ip->i_mount; 2192 int blks_per_cluster; 2193 int inodes_per_cluster; 2194 int nbufs; 2195 int i, j; 2196 int ioffset; 2197 xfs_daddr_t blkno; 2198 xfs_buf_t *bp; 2199 xfs_inode_t *ip; 2200 xfs_inode_log_item_t *iip; 2201 xfs_log_item_t *lip; 2202 struct xfs_perag *pag; 2203 xfs_ino_t inum; 2204 2205 inum = xic->first_ino; 2206 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2207 blks_per_cluster = xfs_icluster_size_fsb(mp); 2208 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2209 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2210 2211 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2212 /* 2213 * The allocation bitmap tells us which inodes of the chunk were 2214 * physically allocated. Skip the cluster if an inode falls into 2215 * a sparse region. 2216 */ 2217 ioffset = inum - xic->first_ino; 2218 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2219 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2220 continue; 2221 } 2222 2223 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2224 XFS_INO_TO_AGBNO(mp, inum)); 2225 2226 /* 2227 * We obtain and lock the backing buffer first in the process 2228 * here, as we have to ensure that any dirty inode that we 2229 * can't get the flush lock on is attached to the buffer. 2230 * If we scan the in-memory inodes first, then buffer IO can 2231 * complete before we get a lock on it, and hence we may fail 2232 * to mark all the active inodes on the buffer stale. 2233 */ 2234 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2235 mp->m_bsize * blks_per_cluster, 2236 XBF_UNMAPPED); 2237 2238 if (!bp) 2239 return -ENOMEM; 2240 2241 /* 2242 * This buffer may not have been correctly initialised as we 2243 * didn't read it from disk. That's not important because we are 2244 * only using to mark the buffer as stale in the log, and to 2245 * attach stale cached inodes on it. That means it will never be 2246 * dispatched for IO. If it is, we want to know about it, and we 2247 * want it to fail. We can acheive this by adding a write 2248 * verifier to the buffer. 2249 */ 2250 bp->b_ops = &xfs_inode_buf_ops; 2251 2252 /* 2253 * Walk the inodes already attached to the buffer and mark them 2254 * stale. These will all have the flush locks held, so an 2255 * in-memory inode walk can't lock them. By marking them all 2256 * stale first, we will not attempt to lock them in the loop 2257 * below as the XFS_ISTALE flag will be set. 2258 */ 2259 lip = bp->b_fspriv; 2260 while (lip) { 2261 if (lip->li_type == XFS_LI_INODE) { 2262 iip = (xfs_inode_log_item_t *)lip; 2263 ASSERT(iip->ili_logged == 1); 2264 lip->li_cb = xfs_istale_done; 2265 xfs_trans_ail_copy_lsn(mp->m_ail, 2266 &iip->ili_flush_lsn, 2267 &iip->ili_item.li_lsn); 2268 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2269 } 2270 lip = lip->li_bio_list; 2271 } 2272 2273 2274 /* 2275 * For each inode in memory attempt to add it to the inode 2276 * buffer and set it up for being staled on buffer IO 2277 * completion. This is safe as we've locked out tail pushing 2278 * and flushing by locking the buffer. 2279 * 2280 * We have already marked every inode that was part of a 2281 * transaction stale above, which means there is no point in 2282 * even trying to lock them. 2283 */ 2284 for (i = 0; i < inodes_per_cluster; i++) { 2285 retry: 2286 rcu_read_lock(); 2287 ip = radix_tree_lookup(&pag->pag_ici_root, 2288 XFS_INO_TO_AGINO(mp, (inum + i))); 2289 2290 /* Inode not in memory, nothing to do */ 2291 if (!ip) { 2292 rcu_read_unlock(); 2293 continue; 2294 } 2295 2296 /* 2297 * because this is an RCU protected lookup, we could 2298 * find a recently freed or even reallocated inode 2299 * during the lookup. We need to check under the 2300 * i_flags_lock for a valid inode here. Skip it if it 2301 * is not valid, the wrong inode or stale. 2302 */ 2303 spin_lock(&ip->i_flags_lock); 2304 if (ip->i_ino != inum + i || 2305 __xfs_iflags_test(ip, XFS_ISTALE)) { 2306 spin_unlock(&ip->i_flags_lock); 2307 rcu_read_unlock(); 2308 continue; 2309 } 2310 spin_unlock(&ip->i_flags_lock); 2311 2312 /* 2313 * Don't try to lock/unlock the current inode, but we 2314 * _cannot_ skip the other inodes that we did not find 2315 * in the list attached to the buffer and are not 2316 * already marked stale. If we can't lock it, back off 2317 * and retry. 2318 */ 2319 if (ip != free_ip && 2320 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2321 rcu_read_unlock(); 2322 delay(1); 2323 goto retry; 2324 } 2325 rcu_read_unlock(); 2326 2327 xfs_iflock(ip); 2328 xfs_iflags_set(ip, XFS_ISTALE); 2329 2330 /* 2331 * we don't need to attach clean inodes or those only 2332 * with unlogged changes (which we throw away, anyway). 2333 */ 2334 iip = ip->i_itemp; 2335 if (!iip || xfs_inode_clean(ip)) { 2336 ASSERT(ip != free_ip); 2337 xfs_ifunlock(ip); 2338 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2339 continue; 2340 } 2341 2342 iip->ili_last_fields = iip->ili_fields; 2343 iip->ili_fields = 0; 2344 iip->ili_fsync_fields = 0; 2345 iip->ili_logged = 1; 2346 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2347 &iip->ili_item.li_lsn); 2348 2349 xfs_buf_attach_iodone(bp, xfs_istale_done, 2350 &iip->ili_item); 2351 2352 if (ip != free_ip) 2353 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2354 } 2355 2356 xfs_trans_stale_inode_buf(tp, bp); 2357 xfs_trans_binval(tp, bp); 2358 } 2359 2360 xfs_perag_put(pag); 2361 return 0; 2362 } 2363 2364 /* 2365 * This is called to return an inode to the inode free list. 2366 * The inode should already be truncated to 0 length and have 2367 * no pages associated with it. This routine also assumes that 2368 * the inode is already a part of the transaction. 2369 * 2370 * The on-disk copy of the inode will have been added to the list 2371 * of unlinked inodes in the AGI. We need to remove the inode from 2372 * that list atomically with respect to freeing it here. 2373 */ 2374 int 2375 xfs_ifree( 2376 xfs_trans_t *tp, 2377 xfs_inode_t *ip, 2378 xfs_bmap_free_t *flist) 2379 { 2380 int error; 2381 struct xfs_icluster xic = { 0 }; 2382 2383 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2384 ASSERT(VFS_I(ip)->i_nlink == 0); 2385 ASSERT(ip->i_d.di_nextents == 0); 2386 ASSERT(ip->i_d.di_anextents == 0); 2387 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(VFS_I(ip)->i_mode)); 2388 ASSERT(ip->i_d.di_nblocks == 0); 2389 2390 /* 2391 * Pull the on-disk inode from the AGI unlinked list. 2392 */ 2393 error = xfs_iunlink_remove(tp, ip); 2394 if (error) 2395 return error; 2396 2397 error = xfs_difree(tp, ip->i_ino, flist, &xic); 2398 if (error) 2399 return error; 2400 2401 VFS_I(ip)->i_mode = 0; /* mark incore inode as free */ 2402 ip->i_d.di_flags = 0; 2403 ip->i_d.di_dmevmask = 0; 2404 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2405 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2406 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2407 /* 2408 * Bump the generation count so no one will be confused 2409 * by reincarnations of this inode. 2410 */ 2411 VFS_I(ip)->i_generation++; 2412 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2413 2414 if (xic.deleted) 2415 error = xfs_ifree_cluster(ip, tp, &xic); 2416 2417 return error; 2418 } 2419 2420 /* 2421 * This is called to unpin an inode. The caller must have the inode locked 2422 * in at least shared mode so that the buffer cannot be subsequently pinned 2423 * once someone is waiting for it to be unpinned. 2424 */ 2425 static void 2426 xfs_iunpin( 2427 struct xfs_inode *ip) 2428 { 2429 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2430 2431 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2432 2433 /* Give the log a push to start the unpinning I/O */ 2434 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2435 2436 } 2437 2438 static void 2439 __xfs_iunpin_wait( 2440 struct xfs_inode *ip) 2441 { 2442 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2443 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2444 2445 xfs_iunpin(ip); 2446 2447 do { 2448 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2449 if (xfs_ipincount(ip)) 2450 io_schedule(); 2451 } while (xfs_ipincount(ip)); 2452 finish_wait(wq, &wait.wait); 2453 } 2454 2455 void 2456 xfs_iunpin_wait( 2457 struct xfs_inode *ip) 2458 { 2459 if (xfs_ipincount(ip)) 2460 __xfs_iunpin_wait(ip); 2461 } 2462 2463 /* 2464 * Removing an inode from the namespace involves removing the directory entry 2465 * and dropping the link count on the inode. Removing the directory entry can 2466 * result in locking an AGF (directory blocks were freed) and removing a link 2467 * count can result in placing the inode on an unlinked list which results in 2468 * locking an AGI. 2469 * 2470 * The big problem here is that we have an ordering constraint on AGF and AGI 2471 * locking - inode allocation locks the AGI, then can allocate a new extent for 2472 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2473 * removes the inode from the unlinked list, requiring that we lock the AGI 2474 * first, and then freeing the inode can result in an inode chunk being freed 2475 * and hence freeing disk space requiring that we lock an AGF. 2476 * 2477 * Hence the ordering that is imposed by other parts of the code is AGI before 2478 * AGF. This means we cannot remove the directory entry before we drop the inode 2479 * reference count and put it on the unlinked list as this results in a lock 2480 * order of AGF then AGI, and this can deadlock against inode allocation and 2481 * freeing. Therefore we must drop the link counts before we remove the 2482 * directory entry. 2483 * 2484 * This is still safe from a transactional point of view - it is not until we 2485 * get to xfs_bmap_finish() that we have the possibility of multiple 2486 * transactions in this operation. Hence as long as we remove the directory 2487 * entry and drop the link count in the first transaction of the remove 2488 * operation, there are no transactional constraints on the ordering here. 2489 */ 2490 int 2491 xfs_remove( 2492 xfs_inode_t *dp, 2493 struct xfs_name *name, 2494 xfs_inode_t *ip) 2495 { 2496 xfs_mount_t *mp = dp->i_mount; 2497 xfs_trans_t *tp = NULL; 2498 int is_dir = S_ISDIR(VFS_I(ip)->i_mode); 2499 int error = 0; 2500 xfs_bmap_free_t free_list; 2501 xfs_fsblock_t first_block; 2502 uint resblks; 2503 2504 trace_xfs_remove(dp, name); 2505 2506 if (XFS_FORCED_SHUTDOWN(mp)) 2507 return -EIO; 2508 2509 error = xfs_qm_dqattach(dp, 0); 2510 if (error) 2511 goto std_return; 2512 2513 error = xfs_qm_dqattach(ip, 0); 2514 if (error) 2515 goto std_return; 2516 2517 /* 2518 * We try to get the real space reservation first, 2519 * allowing for directory btree deletion(s) implying 2520 * possible bmap insert(s). If we can't get the space 2521 * reservation then we use 0 instead, and avoid the bmap 2522 * btree insert(s) in the directory code by, if the bmap 2523 * insert tries to happen, instead trimming the LAST 2524 * block from the directory. 2525 */ 2526 resblks = XFS_REMOVE_SPACE_RES(mp); 2527 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp); 2528 if (error == -ENOSPC) { 2529 resblks = 0; 2530 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0, 2531 &tp); 2532 } 2533 if (error) { 2534 ASSERT(error != -ENOSPC); 2535 goto std_return; 2536 } 2537 2538 xfs_ilock(dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2539 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2540 2541 xfs_trans_ijoin(tp, dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2542 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2543 2544 /* 2545 * If we're removing a directory perform some additional validation. 2546 */ 2547 if (is_dir) { 2548 ASSERT(VFS_I(ip)->i_nlink >= 2); 2549 if (VFS_I(ip)->i_nlink != 2) { 2550 error = -ENOTEMPTY; 2551 goto out_trans_cancel; 2552 } 2553 if (!xfs_dir_isempty(ip)) { 2554 error = -ENOTEMPTY; 2555 goto out_trans_cancel; 2556 } 2557 2558 /* Drop the link from ip's "..". */ 2559 error = xfs_droplink(tp, dp); 2560 if (error) 2561 goto out_trans_cancel; 2562 2563 /* Drop the "." link from ip to self. */ 2564 error = xfs_droplink(tp, ip); 2565 if (error) 2566 goto out_trans_cancel; 2567 } else { 2568 /* 2569 * When removing a non-directory we need to log the parent 2570 * inode here. For a directory this is done implicitly 2571 * by the xfs_droplink call for the ".." entry. 2572 */ 2573 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2574 } 2575 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2576 2577 /* Drop the link from dp to ip. */ 2578 error = xfs_droplink(tp, ip); 2579 if (error) 2580 goto out_trans_cancel; 2581 2582 xfs_bmap_init(&free_list, &first_block); 2583 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2584 &first_block, &free_list, resblks); 2585 if (error) { 2586 ASSERT(error != -ENOENT); 2587 goto out_bmap_cancel; 2588 } 2589 2590 /* 2591 * If this is a synchronous mount, make sure that the 2592 * remove transaction goes to disk before returning to 2593 * the user. 2594 */ 2595 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2596 xfs_trans_set_sync(tp); 2597 2598 error = xfs_bmap_finish(&tp, &free_list, NULL); 2599 if (error) 2600 goto out_bmap_cancel; 2601 2602 error = xfs_trans_commit(tp); 2603 if (error) 2604 goto std_return; 2605 2606 if (is_dir && xfs_inode_is_filestream(ip)) 2607 xfs_filestream_deassociate(ip); 2608 2609 return 0; 2610 2611 out_bmap_cancel: 2612 xfs_bmap_cancel(&free_list); 2613 out_trans_cancel: 2614 xfs_trans_cancel(tp); 2615 std_return: 2616 return error; 2617 } 2618 2619 /* 2620 * Enter all inodes for a rename transaction into a sorted array. 2621 */ 2622 #define __XFS_SORT_INODES 5 2623 STATIC void 2624 xfs_sort_for_rename( 2625 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2626 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2627 struct xfs_inode *ip1, /* in: inode of old entry */ 2628 struct xfs_inode *ip2, /* in: inode of new entry */ 2629 struct xfs_inode *wip, /* in: whiteout inode */ 2630 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2631 int *num_inodes) /* in/out: inodes in array */ 2632 { 2633 int i, j; 2634 2635 ASSERT(*num_inodes == __XFS_SORT_INODES); 2636 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2637 2638 /* 2639 * i_tab contains a list of pointers to inodes. We initialize 2640 * the table here & we'll sort it. We will then use it to 2641 * order the acquisition of the inode locks. 2642 * 2643 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2644 */ 2645 i = 0; 2646 i_tab[i++] = dp1; 2647 i_tab[i++] = dp2; 2648 i_tab[i++] = ip1; 2649 if (ip2) 2650 i_tab[i++] = ip2; 2651 if (wip) 2652 i_tab[i++] = wip; 2653 *num_inodes = i; 2654 2655 /* 2656 * Sort the elements via bubble sort. (Remember, there are at 2657 * most 5 elements to sort, so this is adequate.) 2658 */ 2659 for (i = 0; i < *num_inodes; i++) { 2660 for (j = 1; j < *num_inodes; j++) { 2661 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2662 struct xfs_inode *temp = i_tab[j]; 2663 i_tab[j] = i_tab[j-1]; 2664 i_tab[j-1] = temp; 2665 } 2666 } 2667 } 2668 } 2669 2670 static int 2671 xfs_finish_rename( 2672 struct xfs_trans *tp, 2673 struct xfs_bmap_free *free_list) 2674 { 2675 int error; 2676 2677 /* 2678 * If this is a synchronous mount, make sure that the rename transaction 2679 * goes to disk before returning to the user. 2680 */ 2681 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2682 xfs_trans_set_sync(tp); 2683 2684 error = xfs_bmap_finish(&tp, free_list, NULL); 2685 if (error) { 2686 xfs_bmap_cancel(free_list); 2687 xfs_trans_cancel(tp); 2688 return error; 2689 } 2690 2691 return xfs_trans_commit(tp); 2692 } 2693 2694 /* 2695 * xfs_cross_rename() 2696 * 2697 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2698 */ 2699 STATIC int 2700 xfs_cross_rename( 2701 struct xfs_trans *tp, 2702 struct xfs_inode *dp1, 2703 struct xfs_name *name1, 2704 struct xfs_inode *ip1, 2705 struct xfs_inode *dp2, 2706 struct xfs_name *name2, 2707 struct xfs_inode *ip2, 2708 struct xfs_bmap_free *free_list, 2709 xfs_fsblock_t *first_block, 2710 int spaceres) 2711 { 2712 int error = 0; 2713 int ip1_flags = 0; 2714 int ip2_flags = 0; 2715 int dp2_flags = 0; 2716 2717 /* Swap inode number for dirent in first parent */ 2718 error = xfs_dir_replace(tp, dp1, name1, 2719 ip2->i_ino, 2720 first_block, free_list, spaceres); 2721 if (error) 2722 goto out_trans_abort; 2723 2724 /* Swap inode number for dirent in second parent */ 2725 error = xfs_dir_replace(tp, dp2, name2, 2726 ip1->i_ino, 2727 first_block, free_list, spaceres); 2728 if (error) 2729 goto out_trans_abort; 2730 2731 /* 2732 * If we're renaming one or more directories across different parents, 2733 * update the respective ".." entries (and link counts) to match the new 2734 * parents. 2735 */ 2736 if (dp1 != dp2) { 2737 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2738 2739 if (S_ISDIR(VFS_I(ip2)->i_mode)) { 2740 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2741 dp1->i_ino, first_block, 2742 free_list, spaceres); 2743 if (error) 2744 goto out_trans_abort; 2745 2746 /* transfer ip2 ".." reference to dp1 */ 2747 if (!S_ISDIR(VFS_I(ip1)->i_mode)) { 2748 error = xfs_droplink(tp, dp2); 2749 if (error) 2750 goto out_trans_abort; 2751 error = xfs_bumplink(tp, dp1); 2752 if (error) 2753 goto out_trans_abort; 2754 } 2755 2756 /* 2757 * Although ip1 isn't changed here, userspace needs 2758 * to be warned about the change, so that applications 2759 * relying on it (like backup ones), will properly 2760 * notify the change 2761 */ 2762 ip1_flags |= XFS_ICHGTIME_CHG; 2763 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2764 } 2765 2766 if (S_ISDIR(VFS_I(ip1)->i_mode)) { 2767 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2768 dp2->i_ino, first_block, 2769 free_list, spaceres); 2770 if (error) 2771 goto out_trans_abort; 2772 2773 /* transfer ip1 ".." reference to dp2 */ 2774 if (!S_ISDIR(VFS_I(ip2)->i_mode)) { 2775 error = xfs_droplink(tp, dp1); 2776 if (error) 2777 goto out_trans_abort; 2778 error = xfs_bumplink(tp, dp2); 2779 if (error) 2780 goto out_trans_abort; 2781 } 2782 2783 /* 2784 * Although ip2 isn't changed here, userspace needs 2785 * to be warned about the change, so that applications 2786 * relying on it (like backup ones), will properly 2787 * notify the change 2788 */ 2789 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2790 ip2_flags |= XFS_ICHGTIME_CHG; 2791 } 2792 } 2793 2794 if (ip1_flags) { 2795 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2796 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2797 } 2798 if (ip2_flags) { 2799 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2800 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2801 } 2802 if (dp2_flags) { 2803 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2804 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2805 } 2806 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2807 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2808 return xfs_finish_rename(tp, free_list); 2809 2810 out_trans_abort: 2811 xfs_bmap_cancel(free_list); 2812 xfs_trans_cancel(tp); 2813 return error; 2814 } 2815 2816 /* 2817 * xfs_rename_alloc_whiteout() 2818 * 2819 * Return a referenced, unlinked, unlocked inode that that can be used as a 2820 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2821 * crash between allocating the inode and linking it into the rename transaction 2822 * recovery will free the inode and we won't leak it. 2823 */ 2824 static int 2825 xfs_rename_alloc_whiteout( 2826 struct xfs_inode *dp, 2827 struct xfs_inode **wip) 2828 { 2829 struct xfs_inode *tmpfile; 2830 int error; 2831 2832 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2833 if (error) 2834 return error; 2835 2836 /* 2837 * Prepare the tmpfile inode as if it were created through the VFS. 2838 * Otherwise, the link increment paths will complain about nlink 0->1. 2839 * Drop the link count as done by d_tmpfile(), complete the inode setup 2840 * and flag it as linkable. 2841 */ 2842 drop_nlink(VFS_I(tmpfile)); 2843 xfs_setup_iops(tmpfile); 2844 xfs_finish_inode_setup(tmpfile); 2845 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2846 2847 *wip = tmpfile; 2848 return 0; 2849 } 2850 2851 /* 2852 * xfs_rename 2853 */ 2854 int 2855 xfs_rename( 2856 struct xfs_inode *src_dp, 2857 struct xfs_name *src_name, 2858 struct xfs_inode *src_ip, 2859 struct xfs_inode *target_dp, 2860 struct xfs_name *target_name, 2861 struct xfs_inode *target_ip, 2862 unsigned int flags) 2863 { 2864 struct xfs_mount *mp = src_dp->i_mount; 2865 struct xfs_trans *tp; 2866 struct xfs_bmap_free free_list; 2867 xfs_fsblock_t first_block; 2868 struct xfs_inode *wip = NULL; /* whiteout inode */ 2869 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2870 int num_inodes = __XFS_SORT_INODES; 2871 bool new_parent = (src_dp != target_dp); 2872 bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode); 2873 int spaceres; 2874 int error; 2875 2876 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2877 2878 if ((flags & RENAME_EXCHANGE) && !target_ip) 2879 return -EINVAL; 2880 2881 /* 2882 * If we are doing a whiteout operation, allocate the whiteout inode 2883 * we will be placing at the target and ensure the type is set 2884 * appropriately. 2885 */ 2886 if (flags & RENAME_WHITEOUT) { 2887 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2888 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2889 if (error) 2890 return error; 2891 2892 /* setup target dirent info as whiteout */ 2893 src_name->type = XFS_DIR3_FT_CHRDEV; 2894 } 2895 2896 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2897 inodes, &num_inodes); 2898 2899 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2900 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp); 2901 if (error == -ENOSPC) { 2902 spaceres = 0; 2903 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0, 2904 &tp); 2905 } 2906 if (error) 2907 goto out_release_wip; 2908 2909 /* 2910 * Attach the dquots to the inodes 2911 */ 2912 error = xfs_qm_vop_rename_dqattach(inodes); 2913 if (error) 2914 goto out_trans_cancel; 2915 2916 /* 2917 * Lock all the participating inodes. Depending upon whether 2918 * the target_name exists in the target directory, and 2919 * whether the target directory is the same as the source 2920 * directory, we can lock from 2 to 4 inodes. 2921 */ 2922 if (!new_parent) 2923 xfs_ilock(src_dp, XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2924 else 2925 xfs_lock_two_inodes(src_dp, target_dp, 2926 XFS_IOLOCK_EXCL | XFS_IOLOCK_PARENT); 2927 2928 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2929 2930 /* 2931 * Join all the inodes to the transaction. From this point on, 2932 * we can rely on either trans_commit or trans_cancel to unlock 2933 * them. 2934 */ 2935 xfs_trans_ijoin(tp, src_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2936 if (new_parent) 2937 xfs_trans_ijoin(tp, target_dp, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL); 2938 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2939 if (target_ip) 2940 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2941 if (wip) 2942 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2943 2944 /* 2945 * If we are using project inheritance, we only allow renames 2946 * into our tree when the project IDs are the same; else the 2947 * tree quota mechanism would be circumvented. 2948 */ 2949 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2950 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2951 error = -EXDEV; 2952 goto out_trans_cancel; 2953 } 2954 2955 xfs_bmap_init(&free_list, &first_block); 2956 2957 /* RENAME_EXCHANGE is unique from here on. */ 2958 if (flags & RENAME_EXCHANGE) 2959 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2960 target_dp, target_name, target_ip, 2961 &free_list, &first_block, spaceres); 2962 2963 /* 2964 * Set up the target. 2965 */ 2966 if (target_ip == NULL) { 2967 /* 2968 * If there's no space reservation, check the entry will 2969 * fit before actually inserting it. 2970 */ 2971 if (!spaceres) { 2972 error = xfs_dir_canenter(tp, target_dp, target_name); 2973 if (error) 2974 goto out_trans_cancel; 2975 } 2976 /* 2977 * If target does not exist and the rename crosses 2978 * directories, adjust the target directory link count 2979 * to account for the ".." reference from the new entry. 2980 */ 2981 error = xfs_dir_createname(tp, target_dp, target_name, 2982 src_ip->i_ino, &first_block, 2983 &free_list, spaceres); 2984 if (error) 2985 goto out_bmap_cancel; 2986 2987 xfs_trans_ichgtime(tp, target_dp, 2988 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2989 2990 if (new_parent && src_is_directory) { 2991 error = xfs_bumplink(tp, target_dp); 2992 if (error) 2993 goto out_bmap_cancel; 2994 } 2995 } else { /* target_ip != NULL */ 2996 /* 2997 * If target exists and it's a directory, check that both 2998 * target and source are directories and that target can be 2999 * destroyed, or that neither is a directory. 3000 */ 3001 if (S_ISDIR(VFS_I(target_ip)->i_mode)) { 3002 /* 3003 * Make sure target dir is empty. 3004 */ 3005 if (!(xfs_dir_isempty(target_ip)) || 3006 (VFS_I(target_ip)->i_nlink > 2)) { 3007 error = -EEXIST; 3008 goto out_trans_cancel; 3009 } 3010 } 3011 3012 /* 3013 * Link the source inode under the target name. 3014 * If the source inode is a directory and we are moving 3015 * it across directories, its ".." entry will be 3016 * inconsistent until we replace that down below. 3017 * 3018 * In case there is already an entry with the same 3019 * name at the destination directory, remove it first. 3020 */ 3021 error = xfs_dir_replace(tp, target_dp, target_name, 3022 src_ip->i_ino, 3023 &first_block, &free_list, spaceres); 3024 if (error) 3025 goto out_bmap_cancel; 3026 3027 xfs_trans_ichgtime(tp, target_dp, 3028 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3029 3030 /* 3031 * Decrement the link count on the target since the target 3032 * dir no longer points to it. 3033 */ 3034 error = xfs_droplink(tp, target_ip); 3035 if (error) 3036 goto out_bmap_cancel; 3037 3038 if (src_is_directory) { 3039 /* 3040 * Drop the link from the old "." entry. 3041 */ 3042 error = xfs_droplink(tp, target_ip); 3043 if (error) 3044 goto out_bmap_cancel; 3045 } 3046 } /* target_ip != NULL */ 3047 3048 /* 3049 * Remove the source. 3050 */ 3051 if (new_parent && src_is_directory) { 3052 /* 3053 * Rewrite the ".." entry to point to the new 3054 * directory. 3055 */ 3056 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3057 target_dp->i_ino, 3058 &first_block, &free_list, spaceres); 3059 ASSERT(error != -EEXIST); 3060 if (error) 3061 goto out_bmap_cancel; 3062 } 3063 3064 /* 3065 * We always want to hit the ctime on the source inode. 3066 * 3067 * This isn't strictly required by the standards since the source 3068 * inode isn't really being changed, but old unix file systems did 3069 * it and some incremental backup programs won't work without it. 3070 */ 3071 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3072 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3073 3074 /* 3075 * Adjust the link count on src_dp. This is necessary when 3076 * renaming a directory, either within one parent when 3077 * the target existed, or across two parent directories. 3078 */ 3079 if (src_is_directory && (new_parent || target_ip != NULL)) { 3080 3081 /* 3082 * Decrement link count on src_directory since the 3083 * entry that's moved no longer points to it. 3084 */ 3085 error = xfs_droplink(tp, src_dp); 3086 if (error) 3087 goto out_bmap_cancel; 3088 } 3089 3090 /* 3091 * For whiteouts, we only need to update the source dirent with the 3092 * inode number of the whiteout inode rather than removing it 3093 * altogether. 3094 */ 3095 if (wip) { 3096 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3097 &first_block, &free_list, spaceres); 3098 } else 3099 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3100 &first_block, &free_list, spaceres); 3101 if (error) 3102 goto out_bmap_cancel; 3103 3104 /* 3105 * For whiteouts, we need to bump the link count on the whiteout inode. 3106 * This means that failures all the way up to this point leave the inode 3107 * on the unlinked list and so cleanup is a simple matter of dropping 3108 * the remaining reference to it. If we fail here after bumping the link 3109 * count, we're shutting down the filesystem so we'll never see the 3110 * intermediate state on disk. 3111 */ 3112 if (wip) { 3113 ASSERT(VFS_I(wip)->i_nlink == 0); 3114 error = xfs_bumplink(tp, wip); 3115 if (error) 3116 goto out_bmap_cancel; 3117 error = xfs_iunlink_remove(tp, wip); 3118 if (error) 3119 goto out_bmap_cancel; 3120 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3121 3122 /* 3123 * Now we have a real link, clear the "I'm a tmpfile" state 3124 * flag from the inode so it doesn't accidentally get misused in 3125 * future. 3126 */ 3127 VFS_I(wip)->i_state &= ~I_LINKABLE; 3128 } 3129 3130 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3131 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3132 if (new_parent) 3133 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3134 3135 error = xfs_finish_rename(tp, &free_list); 3136 if (wip) 3137 IRELE(wip); 3138 return error; 3139 3140 out_bmap_cancel: 3141 xfs_bmap_cancel(&free_list); 3142 out_trans_cancel: 3143 xfs_trans_cancel(tp); 3144 out_release_wip: 3145 if (wip) 3146 IRELE(wip); 3147 return error; 3148 } 3149 3150 STATIC int 3151 xfs_iflush_cluster( 3152 struct xfs_inode *ip, 3153 struct xfs_buf *bp) 3154 { 3155 struct xfs_mount *mp = ip->i_mount; 3156 struct xfs_perag *pag; 3157 unsigned long first_index, mask; 3158 unsigned long inodes_per_cluster; 3159 int cilist_size; 3160 struct xfs_inode **cilist; 3161 struct xfs_inode *cip; 3162 int nr_found; 3163 int clcount = 0; 3164 int bufwasdelwri; 3165 int i; 3166 3167 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3168 3169 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3170 cilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3171 cilist = kmem_alloc(cilist_size, KM_MAYFAIL|KM_NOFS); 3172 if (!cilist) 3173 goto out_put; 3174 3175 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3176 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3177 rcu_read_lock(); 3178 /* really need a gang lookup range call here */ 3179 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)cilist, 3180 first_index, inodes_per_cluster); 3181 if (nr_found == 0) 3182 goto out_free; 3183 3184 for (i = 0; i < nr_found; i++) { 3185 cip = cilist[i]; 3186 if (cip == ip) 3187 continue; 3188 3189 /* 3190 * because this is an RCU protected lookup, we could find a 3191 * recently freed or even reallocated inode during the lookup. 3192 * We need to check under the i_flags_lock for a valid inode 3193 * here. Skip it if it is not valid or the wrong inode. 3194 */ 3195 spin_lock(&cip->i_flags_lock); 3196 if (!cip->i_ino || 3197 __xfs_iflags_test(cip, XFS_ISTALE)) { 3198 spin_unlock(&cip->i_flags_lock); 3199 continue; 3200 } 3201 3202 /* 3203 * Once we fall off the end of the cluster, no point checking 3204 * any more inodes in the list because they will also all be 3205 * outside the cluster. 3206 */ 3207 if ((XFS_INO_TO_AGINO(mp, cip->i_ino) & mask) != first_index) { 3208 spin_unlock(&cip->i_flags_lock); 3209 break; 3210 } 3211 spin_unlock(&cip->i_flags_lock); 3212 3213 /* 3214 * Do an un-protected check to see if the inode is dirty and 3215 * is a candidate for flushing. These checks will be repeated 3216 * later after the appropriate locks are acquired. 3217 */ 3218 if (xfs_inode_clean(cip) && xfs_ipincount(cip) == 0) 3219 continue; 3220 3221 /* 3222 * Try to get locks. If any are unavailable or it is pinned, 3223 * then this inode cannot be flushed and is skipped. 3224 */ 3225 3226 if (!xfs_ilock_nowait(cip, XFS_ILOCK_SHARED)) 3227 continue; 3228 if (!xfs_iflock_nowait(cip)) { 3229 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3230 continue; 3231 } 3232 if (xfs_ipincount(cip)) { 3233 xfs_ifunlock(cip); 3234 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3235 continue; 3236 } 3237 3238 3239 /* 3240 * Check the inode number again, just to be certain we are not 3241 * racing with freeing in xfs_reclaim_inode(). See the comments 3242 * in that function for more information as to why the initial 3243 * check is not sufficient. 3244 */ 3245 if (!cip->i_ino) { 3246 xfs_ifunlock(cip); 3247 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3248 continue; 3249 } 3250 3251 /* 3252 * arriving here means that this inode can be flushed. First 3253 * re-check that it's dirty before flushing. 3254 */ 3255 if (!xfs_inode_clean(cip)) { 3256 int error; 3257 error = xfs_iflush_int(cip, bp); 3258 if (error) { 3259 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3260 goto cluster_corrupt_out; 3261 } 3262 clcount++; 3263 } else { 3264 xfs_ifunlock(cip); 3265 } 3266 xfs_iunlock(cip, XFS_ILOCK_SHARED); 3267 } 3268 3269 if (clcount) { 3270 XFS_STATS_INC(mp, xs_icluster_flushcnt); 3271 XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount); 3272 } 3273 3274 out_free: 3275 rcu_read_unlock(); 3276 kmem_free(cilist); 3277 out_put: 3278 xfs_perag_put(pag); 3279 return 0; 3280 3281 3282 cluster_corrupt_out: 3283 /* 3284 * Corruption detected in the clustering loop. Invalidate the 3285 * inode buffer and shut down the filesystem. 3286 */ 3287 rcu_read_unlock(); 3288 /* 3289 * Clean up the buffer. If it was delwri, just release it -- 3290 * brelse can handle it with no problems. If not, shut down the 3291 * filesystem before releasing the buffer. 3292 */ 3293 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3294 if (bufwasdelwri) 3295 xfs_buf_relse(bp); 3296 3297 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3298 3299 if (!bufwasdelwri) { 3300 /* 3301 * Just like incore_relse: if we have b_iodone functions, 3302 * mark the buffer as an error and call them. Otherwise 3303 * mark it as stale and brelse. 3304 */ 3305 if (bp->b_iodone) { 3306 bp->b_flags &= ~XBF_DONE; 3307 xfs_buf_stale(bp); 3308 xfs_buf_ioerror(bp, -EIO); 3309 xfs_buf_ioend(bp); 3310 } else { 3311 xfs_buf_stale(bp); 3312 xfs_buf_relse(bp); 3313 } 3314 } 3315 3316 /* 3317 * Unlocks the flush lock 3318 */ 3319 xfs_iflush_abort(cip, false); 3320 kmem_free(cilist); 3321 xfs_perag_put(pag); 3322 return -EFSCORRUPTED; 3323 } 3324 3325 /* 3326 * Flush dirty inode metadata into the backing buffer. 3327 * 3328 * The caller must have the inode lock and the inode flush lock held. The 3329 * inode lock will still be held upon return to the caller, and the inode 3330 * flush lock will be released after the inode has reached the disk. 3331 * 3332 * The caller must write out the buffer returned in *bpp and release it. 3333 */ 3334 int 3335 xfs_iflush( 3336 struct xfs_inode *ip, 3337 struct xfs_buf **bpp) 3338 { 3339 struct xfs_mount *mp = ip->i_mount; 3340 struct xfs_buf *bp = NULL; 3341 struct xfs_dinode *dip; 3342 int error; 3343 3344 XFS_STATS_INC(mp, xs_iflush_count); 3345 3346 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3347 ASSERT(xfs_isiflocked(ip)); 3348 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3349 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3350 3351 *bpp = NULL; 3352 3353 xfs_iunpin_wait(ip); 3354 3355 /* 3356 * For stale inodes we cannot rely on the backing buffer remaining 3357 * stale in cache for the remaining life of the stale inode and so 3358 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3359 * inodes below. We have to check this after ensuring the inode is 3360 * unpinned so that it is safe to reclaim the stale inode after the 3361 * flush call. 3362 */ 3363 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3364 xfs_ifunlock(ip); 3365 return 0; 3366 } 3367 3368 /* 3369 * This may have been unpinned because the filesystem is shutting 3370 * down forcibly. If that's the case we must not write this inode 3371 * to disk, because the log record didn't make it to disk. 3372 * 3373 * We also have to remove the log item from the AIL in this case, 3374 * as we wait for an empty AIL as part of the unmount process. 3375 */ 3376 if (XFS_FORCED_SHUTDOWN(mp)) { 3377 error = -EIO; 3378 goto abort_out; 3379 } 3380 3381 /* 3382 * Get the buffer containing the on-disk inode. We are doing a try-lock 3383 * operation here, so we may get an EAGAIN error. In that case, we 3384 * simply want to return with the inode still dirty. 3385 * 3386 * If we get any other error, we effectively have a corruption situation 3387 * and we cannot flush the inode, so we treat it the same as failing 3388 * xfs_iflush_int(). 3389 */ 3390 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3391 0); 3392 if (error == -EAGAIN) { 3393 xfs_ifunlock(ip); 3394 return error; 3395 } 3396 if (error) 3397 goto corrupt_out; 3398 3399 /* 3400 * First flush out the inode that xfs_iflush was called with. 3401 */ 3402 error = xfs_iflush_int(ip, bp); 3403 if (error) 3404 goto corrupt_out; 3405 3406 /* 3407 * If the buffer is pinned then push on the log now so we won't 3408 * get stuck waiting in the write for too long. 3409 */ 3410 if (xfs_buf_ispinned(bp)) 3411 xfs_log_force(mp, 0); 3412 3413 /* 3414 * inode clustering: 3415 * see if other inodes can be gathered into this write 3416 */ 3417 error = xfs_iflush_cluster(ip, bp); 3418 if (error) 3419 goto cluster_corrupt_out; 3420 3421 *bpp = bp; 3422 return 0; 3423 3424 corrupt_out: 3425 if (bp) 3426 xfs_buf_relse(bp); 3427 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3428 cluster_corrupt_out: 3429 error = -EFSCORRUPTED; 3430 abort_out: 3431 /* 3432 * Unlocks the flush lock 3433 */ 3434 xfs_iflush_abort(ip, false); 3435 return error; 3436 } 3437 3438 STATIC int 3439 xfs_iflush_int( 3440 struct xfs_inode *ip, 3441 struct xfs_buf *bp) 3442 { 3443 struct xfs_inode_log_item *iip = ip->i_itemp; 3444 struct xfs_dinode *dip; 3445 struct xfs_mount *mp = ip->i_mount; 3446 3447 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3448 ASSERT(xfs_isiflocked(ip)); 3449 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3450 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3451 ASSERT(iip != NULL && iip->ili_fields != 0); 3452 ASSERT(ip->i_d.di_version > 1); 3453 3454 /* set *dip = inode's place in the buffer */ 3455 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3456 3457 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3458 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 3459 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3460 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3461 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3462 goto corrupt_out; 3463 } 3464 if (S_ISREG(VFS_I(ip)->i_mode)) { 3465 if (XFS_TEST_ERROR( 3466 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3467 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3468 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 3469 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3470 "%s: Bad regular inode %Lu, ptr 0x%p", 3471 __func__, ip->i_ino, ip); 3472 goto corrupt_out; 3473 } 3474 } else if (S_ISDIR(VFS_I(ip)->i_mode)) { 3475 if (XFS_TEST_ERROR( 3476 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3477 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3478 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3479 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 3480 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3481 "%s: Bad directory inode %Lu, ptr 0x%p", 3482 __func__, ip->i_ino, ip); 3483 goto corrupt_out; 3484 } 3485 } 3486 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3487 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 3488 XFS_RANDOM_IFLUSH_5)) { 3489 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3490 "%s: detected corrupt incore inode %Lu, " 3491 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3492 __func__, ip->i_ino, 3493 ip->i_d.di_nextents + ip->i_d.di_anextents, 3494 ip->i_d.di_nblocks, ip); 3495 goto corrupt_out; 3496 } 3497 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3498 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 3499 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3500 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3501 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3502 goto corrupt_out; 3503 } 3504 3505 /* 3506 * Inode item log recovery for v2 inodes are dependent on the 3507 * di_flushiter count for correct sequencing. We bump the flush 3508 * iteration count so we can detect flushes which postdate a log record 3509 * during recovery. This is redundant as we now log every change and 3510 * hence this can't happen but we need to still do it to ensure 3511 * backwards compatibility with old kernels that predate logging all 3512 * inode changes. 3513 */ 3514 if (ip->i_d.di_version < 3) 3515 ip->i_d.di_flushiter++; 3516 3517 /* 3518 * Copy the dirty parts of the inode into the on-disk inode. We always 3519 * copy out the core of the inode, because if the inode is dirty at all 3520 * the core must be. 3521 */ 3522 xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn); 3523 3524 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3525 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3526 ip->i_d.di_flushiter = 0; 3527 3528 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3529 if (XFS_IFORK_Q(ip)) 3530 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3531 xfs_inobp_check(mp, bp); 3532 3533 /* 3534 * We've recorded everything logged in the inode, so we'd like to clear 3535 * the ili_fields bits so we don't log and flush things unnecessarily. 3536 * However, we can't stop logging all this information until the data 3537 * we've copied into the disk buffer is written to disk. If we did we 3538 * might overwrite the copy of the inode in the log with all the data 3539 * after re-logging only part of it, and in the face of a crash we 3540 * wouldn't have all the data we need to recover. 3541 * 3542 * What we do is move the bits to the ili_last_fields field. When 3543 * logging the inode, these bits are moved back to the ili_fields field. 3544 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3545 * know that the information those bits represent is permanently on 3546 * disk. As long as the flush completes before the inode is logged 3547 * again, then both ili_fields and ili_last_fields will be cleared. 3548 * 3549 * We can play with the ili_fields bits here, because the inode lock 3550 * must be held exclusively in order to set bits there and the flush 3551 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3552 * done routine can tell whether or not to look in the AIL. Also, store 3553 * the current LSN of the inode so that we can tell whether the item has 3554 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3555 * need the AIL lock, because it is a 64 bit value that cannot be read 3556 * atomically. 3557 */ 3558 iip->ili_last_fields = iip->ili_fields; 3559 iip->ili_fields = 0; 3560 iip->ili_fsync_fields = 0; 3561 iip->ili_logged = 1; 3562 3563 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3564 &iip->ili_item.li_lsn); 3565 3566 /* 3567 * Attach the function xfs_iflush_done to the inode's 3568 * buffer. This will remove the inode from the AIL 3569 * and unlock the inode's flush lock when the inode is 3570 * completely written to disk. 3571 */ 3572 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3573 3574 /* generate the checksum. */ 3575 xfs_dinode_calc_crc(mp, dip); 3576 3577 ASSERT(bp->b_fspriv != NULL); 3578 ASSERT(bp->b_iodone != NULL); 3579 return 0; 3580 3581 corrupt_out: 3582 return -EFSCORRUPTED; 3583 } 3584