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(xfs_inode_t *, xfs_buf_t *); 61 62 STATIC int xfs_iunlink_remove(xfs_trans_t *, xfs_inode_t *); 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_DEP_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_DEP_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_DEP_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 * Bump the subclass so xfs_lock_inodes() acquires each lock with a different 367 * value. This shouldn't be called for page fault locking, but we also need to 368 * ensure we don't overrun the number of lockdep subclasses for the iolock or 369 * mmaplock as that is limited to 12 by the mmap lock lockdep annotations. 370 */ 371 static inline int 372 xfs_lock_inumorder(int lock_mode, int subclass) 373 { 374 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 375 ASSERT(subclass + XFS_LOCK_INUMORDER < 376 (1 << (XFS_MMAPLOCK_SHIFT - XFS_IOLOCK_SHIFT))); 377 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_IOLOCK_SHIFT; 378 } 379 380 if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) { 381 ASSERT(subclass + XFS_LOCK_INUMORDER < 382 (1 << (XFS_ILOCK_SHIFT - XFS_MMAPLOCK_SHIFT))); 383 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << 384 XFS_MMAPLOCK_SHIFT; 385 } 386 387 if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) 388 lock_mode |= (subclass + XFS_LOCK_INUMORDER) << XFS_ILOCK_SHIFT; 389 390 return lock_mode; 391 } 392 393 /* 394 * The following routine will lock n inodes in exclusive mode. We assume the 395 * caller calls us with the inodes in i_ino order. 396 * 397 * We need to detect deadlock where an inode that we lock is in the AIL and we 398 * start waiting for another inode that is locked by a thread in a long running 399 * transaction (such as truncate). This can result in deadlock since the long 400 * running trans might need to wait for the inode we just locked in order to 401 * push the tail and free space in the log. 402 */ 403 void 404 xfs_lock_inodes( 405 xfs_inode_t **ips, 406 int inodes, 407 uint lock_mode) 408 { 409 int attempts = 0, i, j, try_lock; 410 xfs_log_item_t *lp; 411 412 /* currently supports between 2 and 5 inodes */ 413 ASSERT(ips && inodes >= 2 && inodes <= 5); 414 415 try_lock = 0; 416 i = 0; 417 again: 418 for (; i < inodes; i++) { 419 ASSERT(ips[i]); 420 421 if (i && (ips[i] == ips[i - 1])) /* Already locked */ 422 continue; 423 424 /* 425 * If try_lock is not set yet, make sure all locked inodes are 426 * not in the AIL. If any are, set try_lock to be used later. 427 */ 428 if (!try_lock) { 429 for (j = (i - 1); j >= 0 && !try_lock; j--) { 430 lp = (xfs_log_item_t *)ips[j]->i_itemp; 431 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) 432 try_lock++; 433 } 434 } 435 436 /* 437 * If any of the previous locks we have locked is in the AIL, 438 * we must TRY to get the second and subsequent locks. If 439 * we can't get any, we must release all we have 440 * and try again. 441 */ 442 if (!try_lock) { 443 xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i)); 444 continue; 445 } 446 447 /* try_lock means we have an inode locked that is in the AIL. */ 448 ASSERT(i != 0); 449 if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i))) 450 continue; 451 452 /* 453 * Unlock all previous guys and try again. xfs_iunlock will try 454 * to push the tail if the inode is in the AIL. 455 */ 456 attempts++; 457 for (j = i - 1; j >= 0; j--) { 458 /* 459 * Check to see if we've already unlocked this one. Not 460 * the first one going back, and the inode ptr is the 461 * same. 462 */ 463 if (j != (i - 1) && ips[j] == ips[j + 1]) 464 continue; 465 466 xfs_iunlock(ips[j], lock_mode); 467 } 468 469 if ((attempts % 5) == 0) { 470 delay(1); /* Don't just spin the CPU */ 471 #ifdef DEBUG 472 xfs_lock_delays++; 473 #endif 474 } 475 i = 0; 476 try_lock = 0; 477 goto again; 478 } 479 480 #ifdef DEBUG 481 if (attempts) { 482 if (attempts < 5) xfs_small_retries++; 483 else if (attempts < 100) xfs_middle_retries++; 484 else xfs_lots_retries++; 485 } else { 486 xfs_locked_n++; 487 } 488 #endif 489 } 490 491 /* 492 * xfs_lock_two_inodes() can only be used to lock one type of lock at a time - 493 * the iolock, the mmaplock or the ilock, but not more than one at a time. If we 494 * lock more than one at a time, lockdep will report false positives saying we 495 * have violated locking orders. 496 */ 497 void 498 xfs_lock_two_inodes( 499 xfs_inode_t *ip0, 500 xfs_inode_t *ip1, 501 uint lock_mode) 502 { 503 xfs_inode_t *temp; 504 int attempts = 0; 505 xfs_log_item_t *lp; 506 507 if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) { 508 ASSERT(!(lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL))); 509 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 510 } else if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) 511 ASSERT(!(lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL))); 512 513 ASSERT(ip0->i_ino != ip1->i_ino); 514 515 if (ip0->i_ino > ip1->i_ino) { 516 temp = ip0; 517 ip0 = ip1; 518 ip1 = temp; 519 } 520 521 again: 522 xfs_ilock(ip0, xfs_lock_inumorder(lock_mode, 0)); 523 524 /* 525 * If the first lock we have locked is in the AIL, we must TRY to get 526 * the second lock. If we can't get it, we must release the first one 527 * and try again. 528 */ 529 lp = (xfs_log_item_t *)ip0->i_itemp; 530 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) { 531 if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(lock_mode, 1))) { 532 xfs_iunlock(ip0, lock_mode); 533 if ((++attempts % 5) == 0) 534 delay(1); /* Don't just spin the CPU */ 535 goto again; 536 } 537 } else { 538 xfs_ilock(ip1, xfs_lock_inumorder(lock_mode, 1)); 539 } 540 } 541 542 543 void 544 __xfs_iflock( 545 struct xfs_inode *ip) 546 { 547 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 548 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 549 550 do { 551 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 552 if (xfs_isiflocked(ip)) 553 io_schedule(); 554 } while (!xfs_iflock_nowait(ip)); 555 556 finish_wait(wq, &wait.wait); 557 } 558 559 STATIC uint 560 _xfs_dic2xflags( 561 __uint16_t di_flags) 562 { 563 uint flags = 0; 564 565 if (di_flags & XFS_DIFLAG_ANY) { 566 if (di_flags & XFS_DIFLAG_REALTIME) 567 flags |= XFS_XFLAG_REALTIME; 568 if (di_flags & XFS_DIFLAG_PREALLOC) 569 flags |= XFS_XFLAG_PREALLOC; 570 if (di_flags & XFS_DIFLAG_IMMUTABLE) 571 flags |= XFS_XFLAG_IMMUTABLE; 572 if (di_flags & XFS_DIFLAG_APPEND) 573 flags |= XFS_XFLAG_APPEND; 574 if (di_flags & XFS_DIFLAG_SYNC) 575 flags |= XFS_XFLAG_SYNC; 576 if (di_flags & XFS_DIFLAG_NOATIME) 577 flags |= XFS_XFLAG_NOATIME; 578 if (di_flags & XFS_DIFLAG_NODUMP) 579 flags |= XFS_XFLAG_NODUMP; 580 if (di_flags & XFS_DIFLAG_RTINHERIT) 581 flags |= XFS_XFLAG_RTINHERIT; 582 if (di_flags & XFS_DIFLAG_PROJINHERIT) 583 flags |= XFS_XFLAG_PROJINHERIT; 584 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 585 flags |= XFS_XFLAG_NOSYMLINKS; 586 if (di_flags & XFS_DIFLAG_EXTSIZE) 587 flags |= XFS_XFLAG_EXTSIZE; 588 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 589 flags |= XFS_XFLAG_EXTSZINHERIT; 590 if (di_flags & XFS_DIFLAG_NODEFRAG) 591 flags |= XFS_XFLAG_NODEFRAG; 592 if (di_flags & XFS_DIFLAG_FILESTREAM) 593 flags |= XFS_XFLAG_FILESTREAM; 594 } 595 596 return flags; 597 } 598 599 uint 600 xfs_ip2xflags( 601 xfs_inode_t *ip) 602 { 603 xfs_icdinode_t *dic = &ip->i_d; 604 605 return _xfs_dic2xflags(dic->di_flags) | 606 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); 607 } 608 609 uint 610 xfs_dic2xflags( 611 xfs_dinode_t *dip) 612 { 613 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | 614 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); 615 } 616 617 /* 618 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match 619 * is allowed, otherwise it has to be an exact match. If a CI match is found, 620 * ci_name->name will point to a the actual name (caller must free) or 621 * will be set to NULL if an exact match is found. 622 */ 623 int 624 xfs_lookup( 625 xfs_inode_t *dp, 626 struct xfs_name *name, 627 xfs_inode_t **ipp, 628 struct xfs_name *ci_name) 629 { 630 xfs_ino_t inum; 631 int error; 632 uint lock_mode; 633 634 trace_xfs_lookup(dp, name); 635 636 if (XFS_FORCED_SHUTDOWN(dp->i_mount)) 637 return -EIO; 638 639 lock_mode = xfs_ilock_data_map_shared(dp); 640 error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name); 641 xfs_iunlock(dp, lock_mode); 642 643 if (error) 644 goto out; 645 646 error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp); 647 if (error) 648 goto out_free_name; 649 650 return 0; 651 652 out_free_name: 653 if (ci_name) 654 kmem_free(ci_name->name); 655 out: 656 *ipp = NULL; 657 return error; 658 } 659 660 /* 661 * Allocate an inode on disk and return a copy of its in-core version. 662 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 663 * appropriately within the inode. The uid and gid for the inode are 664 * set according to the contents of the given cred structure. 665 * 666 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 667 * has a free inode available, call xfs_iget() to obtain the in-core 668 * version of the allocated inode. Finally, fill in the inode and 669 * log its initial contents. In this case, ialloc_context would be 670 * set to NULL. 671 * 672 * If xfs_dialloc() does not have an available inode, it will replenish 673 * its supply by doing an allocation. Since we can only do one 674 * allocation within a transaction without deadlocks, we must commit 675 * the current transaction before returning the inode itself. 676 * In this case, therefore, we will set ialloc_context and return. 677 * The caller should then commit the current transaction, start a new 678 * transaction, and call xfs_ialloc() again to actually get the inode. 679 * 680 * To ensure that some other process does not grab the inode that 681 * was allocated during the first call to xfs_ialloc(), this routine 682 * also returns the [locked] bp pointing to the head of the freelist 683 * as ialloc_context. The caller should hold this buffer across 684 * the commit and pass it back into this routine on the second call. 685 * 686 * If we are allocating quota inodes, we do not have a parent inode 687 * to attach to or associate with (i.e. pip == NULL) because they 688 * are not linked into the directory structure - they are attached 689 * directly to the superblock - and so have no parent. 690 */ 691 int 692 xfs_ialloc( 693 xfs_trans_t *tp, 694 xfs_inode_t *pip, 695 umode_t mode, 696 xfs_nlink_t nlink, 697 xfs_dev_t rdev, 698 prid_t prid, 699 int okalloc, 700 xfs_buf_t **ialloc_context, 701 xfs_inode_t **ipp) 702 { 703 struct xfs_mount *mp = tp->t_mountp; 704 xfs_ino_t ino; 705 xfs_inode_t *ip; 706 uint flags; 707 int error; 708 struct timespec tv; 709 710 /* 711 * Call the space management code to pick 712 * the on-disk inode to be allocated. 713 */ 714 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 715 ialloc_context, &ino); 716 if (error) 717 return error; 718 if (*ialloc_context || ino == NULLFSINO) { 719 *ipp = NULL; 720 return 0; 721 } 722 ASSERT(*ialloc_context == NULL); 723 724 /* 725 * Get the in-core inode with the lock held exclusively. 726 * This is because we're setting fields here we need 727 * to prevent others from looking at until we're done. 728 */ 729 error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, 730 XFS_ILOCK_EXCL, &ip); 731 if (error) 732 return error; 733 ASSERT(ip != NULL); 734 735 /* 736 * We always convert v1 inodes to v2 now - we only support filesystems 737 * with >= v2 inode capability, so there is no reason for ever leaving 738 * an inode in v1 format. 739 */ 740 if (ip->i_d.di_version == 1) 741 ip->i_d.di_version = 2; 742 743 ip->i_d.di_mode = mode; 744 ip->i_d.di_onlink = 0; 745 ip->i_d.di_nlink = nlink; 746 ASSERT(ip->i_d.di_nlink == nlink); 747 ip->i_d.di_uid = xfs_kuid_to_uid(current_fsuid()); 748 ip->i_d.di_gid = xfs_kgid_to_gid(current_fsgid()); 749 xfs_set_projid(ip, prid); 750 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 751 752 if (pip && XFS_INHERIT_GID(pip)) { 753 ip->i_d.di_gid = pip->i_d.di_gid; 754 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { 755 ip->i_d.di_mode |= S_ISGID; 756 } 757 } 758 759 /* 760 * If the group ID of the new file does not match the effective group 761 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 762 * (and only if the irix_sgid_inherit compatibility variable is set). 763 */ 764 if ((irix_sgid_inherit) && 765 (ip->i_d.di_mode & S_ISGID) && 766 (!in_group_p(xfs_gid_to_kgid(ip->i_d.di_gid)))) { 767 ip->i_d.di_mode &= ~S_ISGID; 768 } 769 770 ip->i_d.di_size = 0; 771 ip->i_d.di_nextents = 0; 772 ASSERT(ip->i_d.di_nblocks == 0); 773 774 tv = current_fs_time(mp->m_super); 775 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; 776 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; 777 ip->i_d.di_atime = ip->i_d.di_mtime; 778 ip->i_d.di_ctime = ip->i_d.di_mtime; 779 780 /* 781 * di_gen will have been taken care of in xfs_iread. 782 */ 783 ip->i_d.di_extsize = 0; 784 ip->i_d.di_dmevmask = 0; 785 ip->i_d.di_dmstate = 0; 786 ip->i_d.di_flags = 0; 787 788 if (ip->i_d.di_version == 3) { 789 ASSERT(ip->i_d.di_ino == ino); 790 ASSERT(uuid_equal(&ip->i_d.di_uuid, &mp->m_sb.sb_uuid)); 791 ip->i_d.di_crc = 0; 792 ip->i_d.di_changecount = 1; 793 ip->i_d.di_lsn = 0; 794 ip->i_d.di_flags2 = 0; 795 memset(&(ip->i_d.di_pad2[0]), 0, sizeof(ip->i_d.di_pad2)); 796 ip->i_d.di_crtime = ip->i_d.di_mtime; 797 } 798 799 800 flags = XFS_ILOG_CORE; 801 switch (mode & S_IFMT) { 802 case S_IFIFO: 803 case S_IFCHR: 804 case S_IFBLK: 805 case S_IFSOCK: 806 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 807 ip->i_df.if_u2.if_rdev = rdev; 808 ip->i_df.if_flags = 0; 809 flags |= XFS_ILOG_DEV; 810 break; 811 case S_IFREG: 812 case S_IFDIR: 813 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 814 uint di_flags = 0; 815 816 if (S_ISDIR(mode)) { 817 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 818 di_flags |= XFS_DIFLAG_RTINHERIT; 819 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 820 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 821 ip->i_d.di_extsize = pip->i_d.di_extsize; 822 } 823 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 824 di_flags |= XFS_DIFLAG_PROJINHERIT; 825 } else if (S_ISREG(mode)) { 826 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 827 di_flags |= XFS_DIFLAG_REALTIME; 828 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 829 di_flags |= XFS_DIFLAG_EXTSIZE; 830 ip->i_d.di_extsize = pip->i_d.di_extsize; 831 } 832 } 833 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 834 xfs_inherit_noatime) 835 di_flags |= XFS_DIFLAG_NOATIME; 836 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 837 xfs_inherit_nodump) 838 di_flags |= XFS_DIFLAG_NODUMP; 839 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 840 xfs_inherit_sync) 841 di_flags |= XFS_DIFLAG_SYNC; 842 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 843 xfs_inherit_nosymlinks) 844 di_flags |= XFS_DIFLAG_NOSYMLINKS; 845 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 846 xfs_inherit_nodefrag) 847 di_flags |= XFS_DIFLAG_NODEFRAG; 848 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 849 di_flags |= XFS_DIFLAG_FILESTREAM; 850 ip->i_d.di_flags |= di_flags; 851 } 852 /* FALLTHROUGH */ 853 case S_IFLNK: 854 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 855 ip->i_df.if_flags = XFS_IFEXTENTS; 856 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 857 ip->i_df.if_u1.if_extents = NULL; 858 break; 859 default: 860 ASSERT(0); 861 } 862 /* 863 * Attribute fork settings for new inode. 864 */ 865 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 866 ip->i_d.di_anextents = 0; 867 868 /* 869 * Log the new values stuffed into the inode. 870 */ 871 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 872 xfs_trans_log_inode(tp, ip, flags); 873 874 /* now that we have an i_mode we can setup the inode structure */ 875 xfs_setup_inode(ip); 876 877 *ipp = ip; 878 return 0; 879 } 880 881 /* 882 * Allocates a new inode from disk and return a pointer to the 883 * incore copy. This routine will internally commit the current 884 * transaction and allocate a new one if the Space Manager needed 885 * to do an allocation to replenish the inode free-list. 886 * 887 * This routine is designed to be called from xfs_create and 888 * xfs_create_dir. 889 * 890 */ 891 int 892 xfs_dir_ialloc( 893 xfs_trans_t **tpp, /* input: current transaction; 894 output: may be a new transaction. */ 895 xfs_inode_t *dp, /* directory within whose allocate 896 the inode. */ 897 umode_t mode, 898 xfs_nlink_t nlink, 899 xfs_dev_t rdev, 900 prid_t prid, /* project id */ 901 int okalloc, /* ok to allocate new space */ 902 xfs_inode_t **ipp, /* pointer to inode; it will be 903 locked. */ 904 int *committed) 905 906 { 907 xfs_trans_t *tp; 908 xfs_trans_t *ntp; 909 xfs_inode_t *ip; 910 xfs_buf_t *ialloc_context = NULL; 911 int code; 912 void *dqinfo; 913 uint tflags; 914 915 tp = *tpp; 916 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 917 918 /* 919 * xfs_ialloc will return a pointer to an incore inode if 920 * the Space Manager has an available inode on the free 921 * list. Otherwise, it will do an allocation and replenish 922 * the freelist. Since we can only do one allocation per 923 * transaction without deadlocks, we will need to commit the 924 * current transaction and start a new one. We will then 925 * need to call xfs_ialloc again to get the inode. 926 * 927 * If xfs_ialloc did an allocation to replenish the freelist, 928 * it returns the bp containing the head of the freelist as 929 * ialloc_context. We will hold a lock on it across the 930 * transaction commit so that no other process can steal 931 * the inode(s) that we've just allocated. 932 */ 933 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 934 &ialloc_context, &ip); 935 936 /* 937 * Return an error if we were unable to allocate a new inode. 938 * This should only happen if we run out of space on disk or 939 * encounter a disk error. 940 */ 941 if (code) { 942 *ipp = NULL; 943 return code; 944 } 945 if (!ialloc_context && !ip) { 946 *ipp = NULL; 947 return -ENOSPC; 948 } 949 950 /* 951 * If the AGI buffer is non-NULL, then we were unable to get an 952 * inode in one operation. We need to commit the current 953 * transaction and call xfs_ialloc() again. It is guaranteed 954 * to succeed the second time. 955 */ 956 if (ialloc_context) { 957 struct xfs_trans_res tres; 958 959 /* 960 * Normally, xfs_trans_commit releases all the locks. 961 * We call bhold to hang on to the ialloc_context across 962 * the commit. Holding this buffer prevents any other 963 * processes from doing any allocations in this 964 * allocation group. 965 */ 966 xfs_trans_bhold(tp, ialloc_context); 967 /* 968 * Save the log reservation so we can use 969 * them in the next transaction. 970 */ 971 tres.tr_logres = xfs_trans_get_log_res(tp); 972 tres.tr_logcount = xfs_trans_get_log_count(tp); 973 974 /* 975 * We want the quota changes to be associated with the next 976 * transaction, NOT this one. So, detach the dqinfo from this 977 * and attach it to the next transaction. 978 */ 979 dqinfo = NULL; 980 tflags = 0; 981 if (tp->t_dqinfo) { 982 dqinfo = (void *)tp->t_dqinfo; 983 tp->t_dqinfo = NULL; 984 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 985 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 986 } 987 988 ntp = xfs_trans_dup(tp); 989 code = xfs_trans_commit(tp, 0); 990 tp = ntp; 991 if (committed != NULL) { 992 *committed = 1; 993 } 994 /* 995 * If we get an error during the commit processing, 996 * release the buffer that is still held and return 997 * to the caller. 998 */ 999 if (code) { 1000 xfs_buf_relse(ialloc_context); 1001 if (dqinfo) { 1002 tp->t_dqinfo = dqinfo; 1003 xfs_trans_free_dqinfo(tp); 1004 } 1005 *tpp = ntp; 1006 *ipp = NULL; 1007 return code; 1008 } 1009 1010 /* 1011 * transaction commit worked ok so we can drop the extra ticket 1012 * reference that we gained in xfs_trans_dup() 1013 */ 1014 xfs_log_ticket_put(tp->t_ticket); 1015 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES; 1016 code = xfs_trans_reserve(tp, &tres, 0, 0); 1017 1018 /* 1019 * Re-attach the quota info that we detached from prev trx. 1020 */ 1021 if (dqinfo) { 1022 tp->t_dqinfo = dqinfo; 1023 tp->t_flags |= tflags; 1024 } 1025 1026 if (code) { 1027 xfs_buf_relse(ialloc_context); 1028 *tpp = ntp; 1029 *ipp = NULL; 1030 return code; 1031 } 1032 xfs_trans_bjoin(tp, ialloc_context); 1033 1034 /* 1035 * Call ialloc again. Since we've locked out all 1036 * other allocations in this allocation group, 1037 * this call should always succeed. 1038 */ 1039 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1040 okalloc, &ialloc_context, &ip); 1041 1042 /* 1043 * If we get an error at this point, return to the caller 1044 * so that the current transaction can be aborted. 1045 */ 1046 if (code) { 1047 *tpp = tp; 1048 *ipp = NULL; 1049 return code; 1050 } 1051 ASSERT(!ialloc_context && ip); 1052 1053 } else { 1054 if (committed != NULL) 1055 *committed = 0; 1056 } 1057 1058 *ipp = ip; 1059 *tpp = tp; 1060 1061 return 0; 1062 } 1063 1064 /* 1065 * Decrement the link count on an inode & log the change. 1066 * If this causes the link count to go to zero, initiate the 1067 * logging activity required to truncate a file. 1068 */ 1069 int /* error */ 1070 xfs_droplink( 1071 xfs_trans_t *tp, 1072 xfs_inode_t *ip) 1073 { 1074 int error; 1075 1076 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1077 1078 ASSERT (ip->i_d.di_nlink > 0); 1079 ip->i_d.di_nlink--; 1080 drop_nlink(VFS_I(ip)); 1081 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1082 1083 error = 0; 1084 if (ip->i_d.di_nlink == 0) { 1085 /* 1086 * We're dropping the last link to this file. 1087 * Move the on-disk inode to the AGI unlinked list. 1088 * From xfs_inactive() we will pull the inode from 1089 * the list and free it. 1090 */ 1091 error = xfs_iunlink(tp, ip); 1092 } 1093 return error; 1094 } 1095 1096 /* 1097 * Increment the link count on an inode & log the change. 1098 */ 1099 int 1100 xfs_bumplink( 1101 xfs_trans_t *tp, 1102 xfs_inode_t *ip) 1103 { 1104 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1105 1106 ASSERT(ip->i_d.di_version > 1); 1107 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE)); 1108 ip->i_d.di_nlink++; 1109 inc_nlink(VFS_I(ip)); 1110 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1111 return 0; 1112 } 1113 1114 int 1115 xfs_create( 1116 xfs_inode_t *dp, 1117 struct xfs_name *name, 1118 umode_t mode, 1119 xfs_dev_t rdev, 1120 xfs_inode_t **ipp) 1121 { 1122 int is_dir = S_ISDIR(mode); 1123 struct xfs_mount *mp = dp->i_mount; 1124 struct xfs_inode *ip = NULL; 1125 struct xfs_trans *tp = NULL; 1126 int error; 1127 xfs_bmap_free_t free_list; 1128 xfs_fsblock_t first_block; 1129 bool unlock_dp_on_error = false; 1130 uint cancel_flags; 1131 int committed; 1132 prid_t prid; 1133 struct xfs_dquot *udqp = NULL; 1134 struct xfs_dquot *gdqp = NULL; 1135 struct xfs_dquot *pdqp = NULL; 1136 struct xfs_trans_res *tres; 1137 uint resblks; 1138 1139 trace_xfs_create(dp, name); 1140 1141 if (XFS_FORCED_SHUTDOWN(mp)) 1142 return -EIO; 1143 1144 prid = xfs_get_initial_prid(dp); 1145 1146 /* 1147 * Make sure that we have allocated dquot(s) on disk. 1148 */ 1149 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1150 xfs_kgid_to_gid(current_fsgid()), prid, 1151 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1152 &udqp, &gdqp, &pdqp); 1153 if (error) 1154 return error; 1155 1156 if (is_dir) { 1157 rdev = 0; 1158 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1159 tres = &M_RES(mp)->tr_mkdir; 1160 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); 1161 } else { 1162 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1163 tres = &M_RES(mp)->tr_create; 1164 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); 1165 } 1166 1167 cancel_flags = XFS_TRANS_RELEASE_LOG_RES; 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_reserve(tp, tres, resblks, 0); 1176 if (error == -ENOSPC) { 1177 /* flush outstanding delalloc blocks and retry */ 1178 xfs_flush_inodes(mp); 1179 error = xfs_trans_reserve(tp, tres, resblks, 0); 1180 } 1181 if (error == -ENOSPC) { 1182 /* No space at all so try a "no-allocation" reservation */ 1183 resblks = 0; 1184 error = xfs_trans_reserve(tp, tres, 0, 0); 1185 } 1186 if (error) { 1187 cancel_flags = 0; 1188 goto out_trans_cancel; 1189 } 1190 1191 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1192 unlock_dp_on_error = true; 1193 1194 xfs_bmap_init(&free_list, &first_block); 1195 1196 /* 1197 * Reserve disk quota and the inode. 1198 */ 1199 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1200 pdqp, resblks, 1, 0); 1201 if (error) 1202 goto out_trans_cancel; 1203 1204 if (!resblks) { 1205 error = xfs_dir_canenter(tp, dp, name); 1206 if (error) 1207 goto out_trans_cancel; 1208 } 1209 1210 /* 1211 * A newly created regular or special file just has one directory 1212 * entry pointing to them, but a directory also the "." entry 1213 * pointing to itself. 1214 */ 1215 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1216 prid, resblks > 0, &ip, &committed); 1217 if (error) { 1218 if (error == -ENOSPC) 1219 goto out_trans_cancel; 1220 goto out_trans_abort; 1221 } 1222 1223 /* 1224 * Now we join the directory inode to the transaction. We do not do it 1225 * earlier because xfs_dir_ialloc might commit the previous transaction 1226 * (and release all the locks). An error from here on will result in 1227 * the transaction cancel unlocking dp so don't do it explicitly in the 1228 * error path. 1229 */ 1230 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1231 unlock_dp_on_error = false; 1232 1233 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1234 &first_block, &free_list, resblks ? 1235 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1236 if (error) { 1237 ASSERT(error != -ENOSPC); 1238 goto out_trans_abort; 1239 } 1240 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1241 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1242 1243 if (is_dir) { 1244 error = xfs_dir_init(tp, ip, dp); 1245 if (error) 1246 goto out_bmap_cancel; 1247 1248 error = xfs_bumplink(tp, dp); 1249 if (error) 1250 goto out_bmap_cancel; 1251 } 1252 1253 /* 1254 * If this is a synchronous mount, make sure that the 1255 * create transaction goes to disk before returning to 1256 * the user. 1257 */ 1258 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1259 xfs_trans_set_sync(tp); 1260 1261 /* 1262 * Attach the dquot(s) to the inodes and modify them incore. 1263 * These ids of the inode couldn't have changed since the new 1264 * inode has been locked ever since it was created. 1265 */ 1266 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1267 1268 error = xfs_bmap_finish(&tp, &free_list, &committed); 1269 if (error) 1270 goto out_bmap_cancel; 1271 1272 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 1273 if (error) 1274 goto out_release_inode; 1275 1276 xfs_qm_dqrele(udqp); 1277 xfs_qm_dqrele(gdqp); 1278 xfs_qm_dqrele(pdqp); 1279 1280 *ipp = ip; 1281 return 0; 1282 1283 out_bmap_cancel: 1284 xfs_bmap_cancel(&free_list); 1285 out_trans_abort: 1286 cancel_flags |= XFS_TRANS_ABORT; 1287 out_trans_cancel: 1288 xfs_trans_cancel(tp, cancel_flags); 1289 out_release_inode: 1290 /* 1291 * Wait until after the current transaction is aborted to finish the 1292 * setup of the inode and release the inode. This prevents recursive 1293 * transactions and deadlocks from xfs_inactive. 1294 */ 1295 if (ip) { 1296 xfs_finish_inode_setup(ip); 1297 IRELE(ip); 1298 } 1299 1300 xfs_qm_dqrele(udqp); 1301 xfs_qm_dqrele(gdqp); 1302 xfs_qm_dqrele(pdqp); 1303 1304 if (unlock_dp_on_error) 1305 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1306 return error; 1307 } 1308 1309 int 1310 xfs_create_tmpfile( 1311 struct xfs_inode *dp, 1312 struct dentry *dentry, 1313 umode_t mode, 1314 struct xfs_inode **ipp) 1315 { 1316 struct xfs_mount *mp = dp->i_mount; 1317 struct xfs_inode *ip = NULL; 1318 struct xfs_trans *tp = NULL; 1319 int error; 1320 uint cancel_flags = XFS_TRANS_RELEASE_LOG_RES; 1321 prid_t prid; 1322 struct xfs_dquot *udqp = NULL; 1323 struct xfs_dquot *gdqp = NULL; 1324 struct xfs_dquot *pdqp = NULL; 1325 struct xfs_trans_res *tres; 1326 uint resblks; 1327 1328 if (XFS_FORCED_SHUTDOWN(mp)) 1329 return -EIO; 1330 1331 prid = xfs_get_initial_prid(dp); 1332 1333 /* 1334 * Make sure that we have allocated dquot(s) on disk. 1335 */ 1336 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1337 xfs_kgid_to_gid(current_fsgid()), prid, 1338 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1339 &udqp, &gdqp, &pdqp); 1340 if (error) 1341 return error; 1342 1343 resblks = XFS_IALLOC_SPACE_RES(mp); 1344 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE); 1345 1346 tres = &M_RES(mp)->tr_create_tmpfile; 1347 error = xfs_trans_reserve(tp, tres, resblks, 0); 1348 if (error == -ENOSPC) { 1349 /* No space at all so try a "no-allocation" reservation */ 1350 resblks = 0; 1351 error = xfs_trans_reserve(tp, tres, 0, 0); 1352 } 1353 if (error) { 1354 cancel_flags = 0; 1355 goto out_trans_cancel; 1356 } 1357 1358 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1359 pdqp, resblks, 1, 0); 1360 if (error) 1361 goto out_trans_cancel; 1362 1363 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1364 prid, resblks > 0, &ip, NULL); 1365 if (error) { 1366 if (error == -ENOSPC) 1367 goto out_trans_cancel; 1368 goto out_trans_abort; 1369 } 1370 1371 if (mp->m_flags & XFS_MOUNT_WSYNC) 1372 xfs_trans_set_sync(tp); 1373 1374 /* 1375 * Attach the dquot(s) to the inodes and modify them incore. 1376 * These ids of the inode couldn't have changed since the new 1377 * inode has been locked ever since it was created. 1378 */ 1379 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1380 1381 ip->i_d.di_nlink--; 1382 error = xfs_iunlink(tp, ip); 1383 if (error) 1384 goto out_trans_abort; 1385 1386 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 1387 if (error) 1388 goto out_release_inode; 1389 1390 xfs_qm_dqrele(udqp); 1391 xfs_qm_dqrele(gdqp); 1392 xfs_qm_dqrele(pdqp); 1393 1394 *ipp = ip; 1395 return 0; 1396 1397 out_trans_abort: 1398 cancel_flags |= XFS_TRANS_ABORT; 1399 out_trans_cancel: 1400 xfs_trans_cancel(tp, cancel_flags); 1401 out_release_inode: 1402 /* 1403 * Wait until after the current transaction is aborted to finish the 1404 * setup of the inode and release the inode. This prevents recursive 1405 * transactions and deadlocks from xfs_inactive. 1406 */ 1407 if (ip) { 1408 xfs_finish_inode_setup(ip); 1409 IRELE(ip); 1410 } 1411 1412 xfs_qm_dqrele(udqp); 1413 xfs_qm_dqrele(gdqp); 1414 xfs_qm_dqrele(pdqp); 1415 1416 return error; 1417 } 1418 1419 int 1420 xfs_link( 1421 xfs_inode_t *tdp, 1422 xfs_inode_t *sip, 1423 struct xfs_name *target_name) 1424 { 1425 xfs_mount_t *mp = tdp->i_mount; 1426 xfs_trans_t *tp; 1427 int error; 1428 xfs_bmap_free_t free_list; 1429 xfs_fsblock_t first_block; 1430 int cancel_flags; 1431 int committed; 1432 int resblks; 1433 1434 trace_xfs_link(tdp, target_name); 1435 1436 ASSERT(!S_ISDIR(sip->i_d.di_mode)); 1437 1438 if (XFS_FORCED_SHUTDOWN(mp)) 1439 return -EIO; 1440 1441 error = xfs_qm_dqattach(sip, 0); 1442 if (error) 1443 goto std_return; 1444 1445 error = xfs_qm_dqattach(tdp, 0); 1446 if (error) 1447 goto std_return; 1448 1449 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); 1450 cancel_flags = XFS_TRANS_RELEASE_LOG_RES; 1451 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1452 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0); 1453 if (error == -ENOSPC) { 1454 resblks = 0; 1455 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0); 1456 } 1457 if (error) { 1458 cancel_flags = 0; 1459 goto error_return; 1460 } 1461 1462 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1463 1464 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1465 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1466 1467 /* 1468 * If we are using project inheritance, we only allow hard link 1469 * creation in our tree when the project IDs are the same; else 1470 * the tree quota mechanism could be circumvented. 1471 */ 1472 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1473 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1474 error = -EXDEV; 1475 goto error_return; 1476 } 1477 1478 if (!resblks) { 1479 error = xfs_dir_canenter(tp, tdp, target_name); 1480 if (error) 1481 goto error_return; 1482 } 1483 1484 xfs_bmap_init(&free_list, &first_block); 1485 1486 if (sip->i_d.di_nlink == 0) { 1487 error = xfs_iunlink_remove(tp, sip); 1488 if (error) 1489 goto abort_return; 1490 } 1491 1492 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1493 &first_block, &free_list, resblks); 1494 if (error) 1495 goto abort_return; 1496 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1497 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1498 1499 error = xfs_bumplink(tp, sip); 1500 if (error) 1501 goto abort_return; 1502 1503 /* 1504 * If this is a synchronous mount, make sure that the 1505 * link transaction goes to disk before returning to 1506 * the user. 1507 */ 1508 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { 1509 xfs_trans_set_sync(tp); 1510 } 1511 1512 error = xfs_bmap_finish (&tp, &free_list, &committed); 1513 if (error) { 1514 xfs_bmap_cancel(&free_list); 1515 goto abort_return; 1516 } 1517 1518 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 1519 1520 abort_return: 1521 cancel_flags |= XFS_TRANS_ABORT; 1522 error_return: 1523 xfs_trans_cancel(tp, cancel_flags); 1524 std_return: 1525 return error; 1526 } 1527 1528 /* 1529 * Free up the underlying blocks past new_size. The new size must be smaller 1530 * than the current size. This routine can be used both for the attribute and 1531 * data fork, and does not modify the inode size, which is left to the caller. 1532 * 1533 * The transaction passed to this routine must have made a permanent log 1534 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1535 * given transaction and start new ones, so make sure everything involved in 1536 * the transaction is tidy before calling here. Some transaction will be 1537 * returned to the caller to be committed. The incoming transaction must 1538 * already include the inode, and both inode locks must be held exclusively. 1539 * The inode must also be "held" within the transaction. On return the inode 1540 * will be "held" within the returned transaction. This routine does NOT 1541 * require any disk space to be reserved for it within the transaction. 1542 * 1543 * If we get an error, we must return with the inode locked and linked into the 1544 * current transaction. This keeps things simple for the higher level code, 1545 * because it always knows that the inode is locked and held in the transaction 1546 * that returns to it whether errors occur or not. We don't mark the inode 1547 * dirty on error so that transactions can be easily aborted if possible. 1548 */ 1549 int 1550 xfs_itruncate_extents( 1551 struct xfs_trans **tpp, 1552 struct xfs_inode *ip, 1553 int whichfork, 1554 xfs_fsize_t new_size) 1555 { 1556 struct xfs_mount *mp = ip->i_mount; 1557 struct xfs_trans *tp = *tpp; 1558 struct xfs_trans *ntp; 1559 xfs_bmap_free_t free_list; 1560 xfs_fsblock_t first_block; 1561 xfs_fileoff_t first_unmap_block; 1562 xfs_fileoff_t last_block; 1563 xfs_filblks_t unmap_len; 1564 int committed; 1565 int error = 0; 1566 int done = 0; 1567 1568 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1569 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1570 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1571 ASSERT(new_size <= XFS_ISIZE(ip)); 1572 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1573 ASSERT(ip->i_itemp != NULL); 1574 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1575 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1576 1577 trace_xfs_itruncate_extents_start(ip, new_size); 1578 1579 /* 1580 * Since it is possible for space to become allocated beyond 1581 * the end of the file (in a crash where the space is allocated 1582 * but the inode size is not yet updated), simply remove any 1583 * blocks which show up between the new EOF and the maximum 1584 * possible file size. If the first block to be removed is 1585 * beyond the maximum file size (ie it is the same as last_block), 1586 * then there is nothing to do. 1587 */ 1588 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1589 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1590 if (first_unmap_block == last_block) 1591 return 0; 1592 1593 ASSERT(first_unmap_block < last_block); 1594 unmap_len = last_block - first_unmap_block + 1; 1595 while (!done) { 1596 xfs_bmap_init(&free_list, &first_block); 1597 error = xfs_bunmapi(tp, ip, 1598 first_unmap_block, unmap_len, 1599 xfs_bmapi_aflag(whichfork), 1600 XFS_ITRUNC_MAX_EXTENTS, 1601 &first_block, &free_list, 1602 &done); 1603 if (error) 1604 goto out_bmap_cancel; 1605 1606 /* 1607 * Duplicate the transaction that has the permanent 1608 * reservation and commit the old transaction. 1609 */ 1610 error = xfs_bmap_finish(&tp, &free_list, &committed); 1611 if (committed) 1612 xfs_trans_ijoin(tp, ip, 0); 1613 if (error) 1614 goto out_bmap_cancel; 1615 1616 if (committed) { 1617 /* 1618 * Mark the inode dirty so it will be logged and 1619 * moved forward in the log as part of every commit. 1620 */ 1621 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1622 } 1623 1624 ntp = xfs_trans_dup(tp); 1625 error = xfs_trans_commit(tp, 0); 1626 tp = ntp; 1627 1628 xfs_trans_ijoin(tp, ip, 0); 1629 1630 if (error) 1631 goto out; 1632 1633 /* 1634 * Transaction commit worked ok so we can drop the extra ticket 1635 * reference that we gained in xfs_trans_dup() 1636 */ 1637 xfs_log_ticket_put(tp->t_ticket); 1638 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); 1639 if (error) 1640 goto out; 1641 } 1642 1643 /* 1644 * Always re-log the inode so that our permanent transaction can keep 1645 * on rolling it forward in the log. 1646 */ 1647 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1648 1649 trace_xfs_itruncate_extents_end(ip, new_size); 1650 1651 out: 1652 *tpp = tp; 1653 return error; 1654 out_bmap_cancel: 1655 /* 1656 * If the bunmapi call encounters an error, return to the caller where 1657 * the transaction can be properly aborted. We just need to make sure 1658 * we're not holding any resources that we were not when we came in. 1659 */ 1660 xfs_bmap_cancel(&free_list); 1661 goto out; 1662 } 1663 1664 int 1665 xfs_release( 1666 xfs_inode_t *ip) 1667 { 1668 xfs_mount_t *mp = ip->i_mount; 1669 int error; 1670 1671 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) 1672 return 0; 1673 1674 /* If this is a read-only mount, don't do this (would generate I/O) */ 1675 if (mp->m_flags & XFS_MOUNT_RDONLY) 1676 return 0; 1677 1678 if (!XFS_FORCED_SHUTDOWN(mp)) { 1679 int truncated; 1680 1681 /* 1682 * If we previously truncated this file and removed old data 1683 * in the process, we want to initiate "early" writeout on 1684 * the last close. This is an attempt to combat the notorious 1685 * NULL files problem which is particularly noticeable from a 1686 * truncate down, buffered (re-)write (delalloc), followed by 1687 * a crash. What we are effectively doing here is 1688 * significantly reducing the time window where we'd otherwise 1689 * be exposed to that problem. 1690 */ 1691 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1692 if (truncated) { 1693 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1694 if (ip->i_delayed_blks > 0) { 1695 error = filemap_flush(VFS_I(ip)->i_mapping); 1696 if (error) 1697 return error; 1698 } 1699 } 1700 } 1701 1702 if (ip->i_d.di_nlink == 0) 1703 return 0; 1704 1705 if (xfs_can_free_eofblocks(ip, false)) { 1706 1707 /* 1708 * If we can't get the iolock just skip truncating the blocks 1709 * past EOF because we could deadlock with the mmap_sem 1710 * otherwise. We'll get another chance to drop them once the 1711 * last reference to the inode is dropped, so we'll never leak 1712 * blocks permanently. 1713 * 1714 * Further, check if the inode is being opened, written and 1715 * closed frequently and we have delayed allocation blocks 1716 * outstanding (e.g. streaming writes from the NFS server), 1717 * truncating the blocks past EOF will cause fragmentation to 1718 * occur. 1719 * 1720 * In this case don't do the truncation, either, but we have to 1721 * be careful how we detect this case. Blocks beyond EOF show 1722 * up as i_delayed_blks even when the inode is clean, so we 1723 * need to truncate them away first before checking for a dirty 1724 * release. Hence on the first dirty close we will still remove 1725 * the speculative allocation, but after that we will leave it 1726 * in place. 1727 */ 1728 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1729 return 0; 1730 1731 error = xfs_free_eofblocks(mp, ip, true); 1732 if (error && error != -EAGAIN) 1733 return error; 1734 1735 /* delalloc blocks after truncation means it really is dirty */ 1736 if (ip->i_delayed_blks) 1737 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1738 } 1739 return 0; 1740 } 1741 1742 /* 1743 * xfs_inactive_truncate 1744 * 1745 * Called to perform a truncate when an inode becomes unlinked. 1746 */ 1747 STATIC int 1748 xfs_inactive_truncate( 1749 struct xfs_inode *ip) 1750 { 1751 struct xfs_mount *mp = ip->i_mount; 1752 struct xfs_trans *tp; 1753 int error; 1754 1755 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); 1756 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); 1757 if (error) { 1758 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1759 xfs_trans_cancel(tp, 0); 1760 return error; 1761 } 1762 1763 xfs_ilock(ip, XFS_ILOCK_EXCL); 1764 xfs_trans_ijoin(tp, ip, 0); 1765 1766 /* 1767 * Log the inode size first to prevent stale data exposure in the event 1768 * of a system crash before the truncate completes. See the related 1769 * comment in xfs_setattr_size() for details. 1770 */ 1771 ip->i_d.di_size = 0; 1772 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1773 1774 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1775 if (error) 1776 goto error_trans_cancel; 1777 1778 ASSERT(ip->i_d.di_nextents == 0); 1779 1780 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 1781 if (error) 1782 goto error_unlock; 1783 1784 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1785 return 0; 1786 1787 error_trans_cancel: 1788 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); 1789 error_unlock: 1790 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1791 return error; 1792 } 1793 1794 /* 1795 * xfs_inactive_ifree() 1796 * 1797 * Perform the inode free when an inode is unlinked. 1798 */ 1799 STATIC int 1800 xfs_inactive_ifree( 1801 struct xfs_inode *ip) 1802 { 1803 xfs_bmap_free_t free_list; 1804 xfs_fsblock_t first_block; 1805 int committed; 1806 struct xfs_mount *mp = ip->i_mount; 1807 struct xfs_trans *tp; 1808 int error; 1809 1810 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); 1811 1812 /* 1813 * The ifree transaction might need to allocate blocks for record 1814 * insertion to the finobt. We don't want to fail here at ENOSPC, so 1815 * allow ifree to dip into the reserved block pool if necessary. 1816 * 1817 * Freeing large sets of inodes generally means freeing inode chunks, 1818 * directory and file data blocks, so this should be relatively safe. 1819 * Only under severe circumstances should it be possible to free enough 1820 * inodes to exhaust the reserve block pool via finobt expansion while 1821 * at the same time not creating free space in the filesystem. 1822 * 1823 * Send a warning if the reservation does happen to fail, as the inode 1824 * now remains allocated and sits on the unlinked list until the fs is 1825 * repaired. 1826 */ 1827 tp->t_flags |= XFS_TRANS_RESERVE; 1828 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 1829 XFS_IFREE_SPACE_RES(mp), 0); 1830 if (error) { 1831 if (error == -ENOSPC) { 1832 xfs_warn_ratelimited(mp, 1833 "Failed to remove inode(s) from unlinked list. " 1834 "Please free space, unmount and run xfs_repair."); 1835 } else { 1836 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1837 } 1838 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES); 1839 return error; 1840 } 1841 1842 xfs_ilock(ip, XFS_ILOCK_EXCL); 1843 xfs_trans_ijoin(tp, ip, 0); 1844 1845 xfs_bmap_init(&free_list, &first_block); 1846 error = xfs_ifree(tp, ip, &free_list); 1847 if (error) { 1848 /* 1849 * If we fail to free the inode, shut down. The cancel 1850 * might do that, we need to make sure. Otherwise the 1851 * inode might be lost for a long time or forever. 1852 */ 1853 if (!XFS_FORCED_SHUTDOWN(mp)) { 1854 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1855 __func__, error); 1856 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1857 } 1858 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); 1859 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1860 return error; 1861 } 1862 1863 /* 1864 * Credit the quota account(s). The inode is gone. 1865 */ 1866 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1867 1868 /* 1869 * Just ignore errors at this point. There is nothing we can 1870 * do except to try to keep going. Make sure it's not a silent 1871 * error. 1872 */ 1873 error = xfs_bmap_finish(&tp, &free_list, &committed); 1874 if (error) 1875 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", 1876 __func__, error); 1877 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 1878 if (error) 1879 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1880 __func__, error); 1881 1882 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1883 return 0; 1884 } 1885 1886 /* 1887 * xfs_inactive 1888 * 1889 * This is called when the vnode reference count for the vnode 1890 * goes to zero. If the file has been unlinked, then it must 1891 * now be truncated. Also, we clear all of the read-ahead state 1892 * kept for the inode here since the file is now closed. 1893 */ 1894 void 1895 xfs_inactive( 1896 xfs_inode_t *ip) 1897 { 1898 struct xfs_mount *mp; 1899 int error; 1900 int truncate = 0; 1901 1902 /* 1903 * If the inode is already free, then there can be nothing 1904 * to clean up here. 1905 */ 1906 if (ip->i_d.di_mode == 0) { 1907 ASSERT(ip->i_df.if_real_bytes == 0); 1908 ASSERT(ip->i_df.if_broot_bytes == 0); 1909 return; 1910 } 1911 1912 mp = ip->i_mount; 1913 1914 /* If this is a read-only mount, don't do this (would generate I/O) */ 1915 if (mp->m_flags & XFS_MOUNT_RDONLY) 1916 return; 1917 1918 if (ip->i_d.di_nlink != 0) { 1919 /* 1920 * force is true because we are evicting an inode from the 1921 * cache. Post-eof blocks must be freed, lest we end up with 1922 * broken free space accounting. 1923 */ 1924 if (xfs_can_free_eofblocks(ip, true)) 1925 xfs_free_eofblocks(mp, ip, false); 1926 1927 return; 1928 } 1929 1930 if (S_ISREG(ip->i_d.di_mode) && 1931 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1932 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1933 truncate = 1; 1934 1935 error = xfs_qm_dqattach(ip, 0); 1936 if (error) 1937 return; 1938 1939 if (S_ISLNK(ip->i_d.di_mode)) 1940 error = xfs_inactive_symlink(ip); 1941 else if (truncate) 1942 error = xfs_inactive_truncate(ip); 1943 if (error) 1944 return; 1945 1946 /* 1947 * If there are attributes associated with the file then blow them away 1948 * now. The code calls a routine that recursively deconstructs the 1949 * attribute fork. We need to just commit the current transaction 1950 * because we can't use it for xfs_attr_inactive(). 1951 */ 1952 if (ip->i_d.di_anextents > 0) { 1953 ASSERT(ip->i_d.di_forkoff != 0); 1954 1955 error = xfs_attr_inactive(ip); 1956 if (error) 1957 return; 1958 } 1959 1960 if (ip->i_afp) 1961 xfs_idestroy_fork(ip, XFS_ATTR_FORK); 1962 1963 ASSERT(ip->i_d.di_anextents == 0); 1964 1965 /* 1966 * Free the inode. 1967 */ 1968 error = xfs_inactive_ifree(ip); 1969 if (error) 1970 return; 1971 1972 /* 1973 * Release the dquots held by inode, if any. 1974 */ 1975 xfs_qm_dqdetach(ip); 1976 } 1977 1978 /* 1979 * This is called when the inode's link count goes to 0. 1980 * We place the on-disk inode on a list in the AGI. It 1981 * will be pulled from this list when the inode is freed. 1982 */ 1983 int 1984 xfs_iunlink( 1985 xfs_trans_t *tp, 1986 xfs_inode_t *ip) 1987 { 1988 xfs_mount_t *mp; 1989 xfs_agi_t *agi; 1990 xfs_dinode_t *dip; 1991 xfs_buf_t *agibp; 1992 xfs_buf_t *ibp; 1993 xfs_agino_t agino; 1994 short bucket_index; 1995 int offset; 1996 int error; 1997 1998 ASSERT(ip->i_d.di_nlink == 0); 1999 ASSERT(ip->i_d.di_mode != 0); 2000 2001 mp = tp->t_mountp; 2002 2003 /* 2004 * Get the agi buffer first. It ensures lock ordering 2005 * on the list. 2006 */ 2007 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 2008 if (error) 2009 return error; 2010 agi = XFS_BUF_TO_AGI(agibp); 2011 2012 /* 2013 * Get the index into the agi hash table for the 2014 * list this inode will go on. 2015 */ 2016 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2017 ASSERT(agino != 0); 2018 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2019 ASSERT(agi->agi_unlinked[bucket_index]); 2020 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 2021 2022 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 2023 /* 2024 * There is already another inode in the bucket we need 2025 * to add ourselves to. Add us at the front of the list. 2026 * Here we put the head pointer into our next pointer, 2027 * and then we fall through to point the head at us. 2028 */ 2029 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2030 0, 0); 2031 if (error) 2032 return error; 2033 2034 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 2035 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 2036 offset = ip->i_imap.im_boffset + 2037 offsetof(xfs_dinode_t, di_next_unlinked); 2038 2039 /* need to recalc the inode CRC if appropriate */ 2040 xfs_dinode_calc_crc(mp, dip); 2041 2042 xfs_trans_inode_buf(tp, ibp); 2043 xfs_trans_log_buf(tp, ibp, offset, 2044 (offset + sizeof(xfs_agino_t) - 1)); 2045 xfs_inobp_check(mp, ibp); 2046 } 2047 2048 /* 2049 * Point the bucket head pointer at the inode being inserted. 2050 */ 2051 ASSERT(agino != 0); 2052 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 2053 offset = offsetof(xfs_agi_t, agi_unlinked) + 2054 (sizeof(xfs_agino_t) * bucket_index); 2055 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2056 xfs_trans_log_buf(tp, agibp, offset, 2057 (offset + sizeof(xfs_agino_t) - 1)); 2058 return 0; 2059 } 2060 2061 /* 2062 * Pull the on-disk inode from the AGI unlinked list. 2063 */ 2064 STATIC int 2065 xfs_iunlink_remove( 2066 xfs_trans_t *tp, 2067 xfs_inode_t *ip) 2068 { 2069 xfs_ino_t next_ino; 2070 xfs_mount_t *mp; 2071 xfs_agi_t *agi; 2072 xfs_dinode_t *dip; 2073 xfs_buf_t *agibp; 2074 xfs_buf_t *ibp; 2075 xfs_agnumber_t agno; 2076 xfs_agino_t agino; 2077 xfs_agino_t next_agino; 2078 xfs_buf_t *last_ibp; 2079 xfs_dinode_t *last_dip = NULL; 2080 short bucket_index; 2081 int offset, last_offset = 0; 2082 int error; 2083 2084 mp = tp->t_mountp; 2085 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2086 2087 /* 2088 * Get the agi buffer first. It ensures lock ordering 2089 * on the list. 2090 */ 2091 error = xfs_read_agi(mp, tp, agno, &agibp); 2092 if (error) 2093 return error; 2094 2095 agi = XFS_BUF_TO_AGI(agibp); 2096 2097 /* 2098 * Get the index into the agi hash table for the 2099 * list this inode will go on. 2100 */ 2101 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2102 ASSERT(agino != 0); 2103 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2104 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2105 ASSERT(agi->agi_unlinked[bucket_index]); 2106 2107 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2108 /* 2109 * We're at the head of the list. Get the inode's on-disk 2110 * buffer to see if there is anyone after us on the list. 2111 * Only modify our next pointer if it is not already NULLAGINO. 2112 * This saves us the overhead of dealing with the buffer when 2113 * there is no need to change it. 2114 */ 2115 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2116 0, 0); 2117 if (error) { 2118 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2119 __func__, error); 2120 return error; 2121 } 2122 next_agino = be32_to_cpu(dip->di_next_unlinked); 2123 ASSERT(next_agino != 0); 2124 if (next_agino != NULLAGINO) { 2125 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2126 offset = ip->i_imap.im_boffset + 2127 offsetof(xfs_dinode_t, di_next_unlinked); 2128 2129 /* need to recalc the inode CRC if appropriate */ 2130 xfs_dinode_calc_crc(mp, dip); 2131 2132 xfs_trans_inode_buf(tp, ibp); 2133 xfs_trans_log_buf(tp, ibp, offset, 2134 (offset + sizeof(xfs_agino_t) - 1)); 2135 xfs_inobp_check(mp, ibp); 2136 } else { 2137 xfs_trans_brelse(tp, ibp); 2138 } 2139 /* 2140 * Point the bucket head pointer at the next inode. 2141 */ 2142 ASSERT(next_agino != 0); 2143 ASSERT(next_agino != agino); 2144 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2145 offset = offsetof(xfs_agi_t, agi_unlinked) + 2146 (sizeof(xfs_agino_t) * bucket_index); 2147 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2148 xfs_trans_log_buf(tp, agibp, offset, 2149 (offset + sizeof(xfs_agino_t) - 1)); 2150 } else { 2151 /* 2152 * We need to search the list for the inode being freed. 2153 */ 2154 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2155 last_ibp = NULL; 2156 while (next_agino != agino) { 2157 struct xfs_imap imap; 2158 2159 if (last_ibp) 2160 xfs_trans_brelse(tp, last_ibp); 2161 2162 imap.im_blkno = 0; 2163 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2164 2165 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2166 if (error) { 2167 xfs_warn(mp, 2168 "%s: xfs_imap returned error %d.", 2169 __func__, error); 2170 return error; 2171 } 2172 2173 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2174 &last_ibp, 0, 0); 2175 if (error) { 2176 xfs_warn(mp, 2177 "%s: xfs_imap_to_bp returned error %d.", 2178 __func__, error); 2179 return error; 2180 } 2181 2182 last_offset = imap.im_boffset; 2183 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2184 ASSERT(next_agino != NULLAGINO); 2185 ASSERT(next_agino != 0); 2186 } 2187 2188 /* 2189 * Now last_ibp points to the buffer previous to us on the 2190 * unlinked list. Pull us from the list. 2191 */ 2192 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2193 0, 0); 2194 if (error) { 2195 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2196 __func__, error); 2197 return error; 2198 } 2199 next_agino = be32_to_cpu(dip->di_next_unlinked); 2200 ASSERT(next_agino != 0); 2201 ASSERT(next_agino != agino); 2202 if (next_agino != NULLAGINO) { 2203 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2204 offset = ip->i_imap.im_boffset + 2205 offsetof(xfs_dinode_t, di_next_unlinked); 2206 2207 /* need to recalc the inode CRC if appropriate */ 2208 xfs_dinode_calc_crc(mp, dip); 2209 2210 xfs_trans_inode_buf(tp, ibp); 2211 xfs_trans_log_buf(tp, ibp, offset, 2212 (offset + sizeof(xfs_agino_t) - 1)); 2213 xfs_inobp_check(mp, ibp); 2214 } else { 2215 xfs_trans_brelse(tp, ibp); 2216 } 2217 /* 2218 * Point the previous inode on the list to the next inode. 2219 */ 2220 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2221 ASSERT(next_agino != 0); 2222 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2223 2224 /* need to recalc the inode CRC if appropriate */ 2225 xfs_dinode_calc_crc(mp, last_dip); 2226 2227 xfs_trans_inode_buf(tp, last_ibp); 2228 xfs_trans_log_buf(tp, last_ibp, offset, 2229 (offset + sizeof(xfs_agino_t) - 1)); 2230 xfs_inobp_check(mp, last_ibp); 2231 } 2232 return 0; 2233 } 2234 2235 /* 2236 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2237 * inodes that are in memory - they all must be marked stale and attached to 2238 * the cluster buffer. 2239 */ 2240 STATIC int 2241 xfs_ifree_cluster( 2242 xfs_inode_t *free_ip, 2243 xfs_trans_t *tp, 2244 xfs_ino_t inum) 2245 { 2246 xfs_mount_t *mp = free_ip->i_mount; 2247 int blks_per_cluster; 2248 int inodes_per_cluster; 2249 int nbufs; 2250 int i, j; 2251 xfs_daddr_t blkno; 2252 xfs_buf_t *bp; 2253 xfs_inode_t *ip; 2254 xfs_inode_log_item_t *iip; 2255 xfs_log_item_t *lip; 2256 struct xfs_perag *pag; 2257 2258 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2259 blks_per_cluster = xfs_icluster_size_fsb(mp); 2260 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2261 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2262 2263 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2264 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2265 XFS_INO_TO_AGBNO(mp, inum)); 2266 2267 /* 2268 * We obtain and lock the backing buffer first in the process 2269 * here, as we have to ensure that any dirty inode that we 2270 * can't get the flush lock on is attached to the buffer. 2271 * If we scan the in-memory inodes first, then buffer IO can 2272 * complete before we get a lock on it, and hence we may fail 2273 * to mark all the active inodes on the buffer stale. 2274 */ 2275 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2276 mp->m_bsize * blks_per_cluster, 2277 XBF_UNMAPPED); 2278 2279 if (!bp) 2280 return -ENOMEM; 2281 2282 /* 2283 * This buffer may not have been correctly initialised as we 2284 * didn't read it from disk. That's not important because we are 2285 * only using to mark the buffer as stale in the log, and to 2286 * attach stale cached inodes on it. That means it will never be 2287 * dispatched for IO. If it is, we want to know about it, and we 2288 * want it to fail. We can acheive this by adding a write 2289 * verifier to the buffer. 2290 */ 2291 bp->b_ops = &xfs_inode_buf_ops; 2292 2293 /* 2294 * Walk the inodes already attached to the buffer and mark them 2295 * stale. These will all have the flush locks held, so an 2296 * in-memory inode walk can't lock them. By marking them all 2297 * stale first, we will not attempt to lock them in the loop 2298 * below as the XFS_ISTALE flag will be set. 2299 */ 2300 lip = bp->b_fspriv; 2301 while (lip) { 2302 if (lip->li_type == XFS_LI_INODE) { 2303 iip = (xfs_inode_log_item_t *)lip; 2304 ASSERT(iip->ili_logged == 1); 2305 lip->li_cb = xfs_istale_done; 2306 xfs_trans_ail_copy_lsn(mp->m_ail, 2307 &iip->ili_flush_lsn, 2308 &iip->ili_item.li_lsn); 2309 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2310 } 2311 lip = lip->li_bio_list; 2312 } 2313 2314 2315 /* 2316 * For each inode in memory attempt to add it to the inode 2317 * buffer and set it up for being staled on buffer IO 2318 * completion. This is safe as we've locked out tail pushing 2319 * and flushing by locking the buffer. 2320 * 2321 * We have already marked every inode that was part of a 2322 * transaction stale above, which means there is no point in 2323 * even trying to lock them. 2324 */ 2325 for (i = 0; i < inodes_per_cluster; i++) { 2326 retry: 2327 rcu_read_lock(); 2328 ip = radix_tree_lookup(&pag->pag_ici_root, 2329 XFS_INO_TO_AGINO(mp, (inum + i))); 2330 2331 /* Inode not in memory, nothing to do */ 2332 if (!ip) { 2333 rcu_read_unlock(); 2334 continue; 2335 } 2336 2337 /* 2338 * because this is an RCU protected lookup, we could 2339 * find a recently freed or even reallocated inode 2340 * during the lookup. We need to check under the 2341 * i_flags_lock for a valid inode here. Skip it if it 2342 * is not valid, the wrong inode or stale. 2343 */ 2344 spin_lock(&ip->i_flags_lock); 2345 if (ip->i_ino != inum + i || 2346 __xfs_iflags_test(ip, XFS_ISTALE)) { 2347 spin_unlock(&ip->i_flags_lock); 2348 rcu_read_unlock(); 2349 continue; 2350 } 2351 spin_unlock(&ip->i_flags_lock); 2352 2353 /* 2354 * Don't try to lock/unlock the current inode, but we 2355 * _cannot_ skip the other inodes that we did not find 2356 * in the list attached to the buffer and are not 2357 * already marked stale. If we can't lock it, back off 2358 * and retry. 2359 */ 2360 if (ip != free_ip && 2361 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2362 rcu_read_unlock(); 2363 delay(1); 2364 goto retry; 2365 } 2366 rcu_read_unlock(); 2367 2368 xfs_iflock(ip); 2369 xfs_iflags_set(ip, XFS_ISTALE); 2370 2371 /* 2372 * we don't need to attach clean inodes or those only 2373 * with unlogged changes (which we throw away, anyway). 2374 */ 2375 iip = ip->i_itemp; 2376 if (!iip || xfs_inode_clean(ip)) { 2377 ASSERT(ip != free_ip); 2378 xfs_ifunlock(ip); 2379 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2380 continue; 2381 } 2382 2383 iip->ili_last_fields = iip->ili_fields; 2384 iip->ili_fields = 0; 2385 iip->ili_logged = 1; 2386 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2387 &iip->ili_item.li_lsn); 2388 2389 xfs_buf_attach_iodone(bp, xfs_istale_done, 2390 &iip->ili_item); 2391 2392 if (ip != free_ip) 2393 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2394 } 2395 2396 xfs_trans_stale_inode_buf(tp, bp); 2397 xfs_trans_binval(tp, bp); 2398 } 2399 2400 xfs_perag_put(pag); 2401 return 0; 2402 } 2403 2404 /* 2405 * This is called to return an inode to the inode free list. 2406 * The inode should already be truncated to 0 length and have 2407 * no pages associated with it. This routine also assumes that 2408 * the inode is already a part of the transaction. 2409 * 2410 * The on-disk copy of the inode will have been added to the list 2411 * of unlinked inodes in the AGI. We need to remove the inode from 2412 * that list atomically with respect to freeing it here. 2413 */ 2414 int 2415 xfs_ifree( 2416 xfs_trans_t *tp, 2417 xfs_inode_t *ip, 2418 xfs_bmap_free_t *flist) 2419 { 2420 int error; 2421 int delete; 2422 xfs_ino_t first_ino; 2423 2424 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2425 ASSERT(ip->i_d.di_nlink == 0); 2426 ASSERT(ip->i_d.di_nextents == 0); 2427 ASSERT(ip->i_d.di_anextents == 0); 2428 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); 2429 ASSERT(ip->i_d.di_nblocks == 0); 2430 2431 /* 2432 * Pull the on-disk inode from the AGI unlinked list. 2433 */ 2434 error = xfs_iunlink_remove(tp, ip); 2435 if (error) 2436 return error; 2437 2438 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); 2439 if (error) 2440 return error; 2441 2442 ip->i_d.di_mode = 0; /* mark incore inode as free */ 2443 ip->i_d.di_flags = 0; 2444 ip->i_d.di_dmevmask = 0; 2445 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2446 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2447 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2448 /* 2449 * Bump the generation count so no one will be confused 2450 * by reincarnations of this inode. 2451 */ 2452 ip->i_d.di_gen++; 2453 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2454 2455 if (delete) 2456 error = xfs_ifree_cluster(ip, tp, first_ino); 2457 2458 return error; 2459 } 2460 2461 /* 2462 * This is called to unpin an inode. The caller must have the inode locked 2463 * in at least shared mode so that the buffer cannot be subsequently pinned 2464 * once someone is waiting for it to be unpinned. 2465 */ 2466 static void 2467 xfs_iunpin( 2468 struct xfs_inode *ip) 2469 { 2470 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2471 2472 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2473 2474 /* Give the log a push to start the unpinning I/O */ 2475 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2476 2477 } 2478 2479 static void 2480 __xfs_iunpin_wait( 2481 struct xfs_inode *ip) 2482 { 2483 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2484 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2485 2486 xfs_iunpin(ip); 2487 2488 do { 2489 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2490 if (xfs_ipincount(ip)) 2491 io_schedule(); 2492 } while (xfs_ipincount(ip)); 2493 finish_wait(wq, &wait.wait); 2494 } 2495 2496 void 2497 xfs_iunpin_wait( 2498 struct xfs_inode *ip) 2499 { 2500 if (xfs_ipincount(ip)) 2501 __xfs_iunpin_wait(ip); 2502 } 2503 2504 /* 2505 * Removing an inode from the namespace involves removing the directory entry 2506 * and dropping the link count on the inode. Removing the directory entry can 2507 * result in locking an AGF (directory blocks were freed) and removing a link 2508 * count can result in placing the inode on an unlinked list which results in 2509 * locking an AGI. 2510 * 2511 * The big problem here is that we have an ordering constraint on AGF and AGI 2512 * locking - inode allocation locks the AGI, then can allocate a new extent for 2513 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2514 * removes the inode from the unlinked list, requiring that we lock the AGI 2515 * first, and then freeing the inode can result in an inode chunk being freed 2516 * and hence freeing disk space requiring that we lock an AGF. 2517 * 2518 * Hence the ordering that is imposed by other parts of the code is AGI before 2519 * AGF. This means we cannot remove the directory entry before we drop the inode 2520 * reference count and put it on the unlinked list as this results in a lock 2521 * order of AGF then AGI, and this can deadlock against inode allocation and 2522 * freeing. Therefore we must drop the link counts before we remove the 2523 * directory entry. 2524 * 2525 * This is still safe from a transactional point of view - it is not until we 2526 * get to xfs_bmap_finish() that we have the possibility of multiple 2527 * transactions in this operation. Hence as long as we remove the directory 2528 * entry and drop the link count in the first transaction of the remove 2529 * operation, there are no transactional constraints on the ordering here. 2530 */ 2531 int 2532 xfs_remove( 2533 xfs_inode_t *dp, 2534 struct xfs_name *name, 2535 xfs_inode_t *ip) 2536 { 2537 xfs_mount_t *mp = dp->i_mount; 2538 xfs_trans_t *tp = NULL; 2539 int is_dir = S_ISDIR(ip->i_d.di_mode); 2540 int error = 0; 2541 xfs_bmap_free_t free_list; 2542 xfs_fsblock_t first_block; 2543 int cancel_flags; 2544 int committed; 2545 uint resblks; 2546 2547 trace_xfs_remove(dp, name); 2548 2549 if (XFS_FORCED_SHUTDOWN(mp)) 2550 return -EIO; 2551 2552 error = xfs_qm_dqattach(dp, 0); 2553 if (error) 2554 goto std_return; 2555 2556 error = xfs_qm_dqattach(ip, 0); 2557 if (error) 2558 goto std_return; 2559 2560 if (is_dir) 2561 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); 2562 else 2563 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); 2564 cancel_flags = XFS_TRANS_RELEASE_LOG_RES; 2565 2566 /* 2567 * We try to get the real space reservation first, 2568 * allowing for directory btree deletion(s) implying 2569 * possible bmap insert(s). If we can't get the space 2570 * reservation then we use 0 instead, and avoid the bmap 2571 * btree insert(s) in the directory code by, if the bmap 2572 * insert tries to happen, instead trimming the LAST 2573 * block from the directory. 2574 */ 2575 resblks = XFS_REMOVE_SPACE_RES(mp); 2576 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0); 2577 if (error == -ENOSPC) { 2578 resblks = 0; 2579 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0); 2580 } 2581 if (error) { 2582 ASSERT(error != -ENOSPC); 2583 cancel_flags = 0; 2584 goto out_trans_cancel; 2585 } 2586 2587 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2588 2589 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2590 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2591 2592 /* 2593 * If we're removing a directory perform some additional validation. 2594 */ 2595 cancel_flags |= XFS_TRANS_ABORT; 2596 if (is_dir) { 2597 ASSERT(ip->i_d.di_nlink >= 2); 2598 if (ip->i_d.di_nlink != 2) { 2599 error = -ENOTEMPTY; 2600 goto out_trans_cancel; 2601 } 2602 if (!xfs_dir_isempty(ip)) { 2603 error = -ENOTEMPTY; 2604 goto out_trans_cancel; 2605 } 2606 2607 /* Drop the link from ip's "..". */ 2608 error = xfs_droplink(tp, dp); 2609 if (error) 2610 goto out_trans_cancel; 2611 2612 /* Drop the "." link from ip to self. */ 2613 error = xfs_droplink(tp, ip); 2614 if (error) 2615 goto out_trans_cancel; 2616 } else { 2617 /* 2618 * When removing a non-directory we need to log the parent 2619 * inode here. For a directory this is done implicitly 2620 * by the xfs_droplink call for the ".." entry. 2621 */ 2622 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2623 } 2624 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2625 2626 /* Drop the link from dp to ip. */ 2627 error = xfs_droplink(tp, ip); 2628 if (error) 2629 goto out_trans_cancel; 2630 2631 xfs_bmap_init(&free_list, &first_block); 2632 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2633 &first_block, &free_list, resblks); 2634 if (error) { 2635 ASSERT(error != -ENOENT); 2636 goto out_bmap_cancel; 2637 } 2638 2639 /* 2640 * If this is a synchronous mount, make sure that the 2641 * remove transaction goes to disk before returning to 2642 * the user. 2643 */ 2644 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2645 xfs_trans_set_sync(tp); 2646 2647 error = xfs_bmap_finish(&tp, &free_list, &committed); 2648 if (error) 2649 goto out_bmap_cancel; 2650 2651 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 2652 if (error) 2653 goto std_return; 2654 2655 if (is_dir && xfs_inode_is_filestream(ip)) 2656 xfs_filestream_deassociate(ip); 2657 2658 return 0; 2659 2660 out_bmap_cancel: 2661 xfs_bmap_cancel(&free_list); 2662 out_trans_cancel: 2663 xfs_trans_cancel(tp, cancel_flags); 2664 std_return: 2665 return error; 2666 } 2667 2668 /* 2669 * Enter all inodes for a rename transaction into a sorted array. 2670 */ 2671 #define __XFS_SORT_INODES 5 2672 STATIC void 2673 xfs_sort_for_rename( 2674 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2675 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2676 struct xfs_inode *ip1, /* in: inode of old entry */ 2677 struct xfs_inode *ip2, /* in: inode of new entry */ 2678 struct xfs_inode *wip, /* in: whiteout inode */ 2679 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2680 int *num_inodes) /* in/out: inodes in array */ 2681 { 2682 int i, j; 2683 2684 ASSERT(*num_inodes == __XFS_SORT_INODES); 2685 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2686 2687 /* 2688 * i_tab contains a list of pointers to inodes. We initialize 2689 * the table here & we'll sort it. We will then use it to 2690 * order the acquisition of the inode locks. 2691 * 2692 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2693 */ 2694 i = 0; 2695 i_tab[i++] = dp1; 2696 i_tab[i++] = dp2; 2697 i_tab[i++] = ip1; 2698 if (ip2) 2699 i_tab[i++] = ip2; 2700 if (wip) 2701 i_tab[i++] = wip; 2702 *num_inodes = i; 2703 2704 /* 2705 * Sort the elements via bubble sort. (Remember, there are at 2706 * most 5 elements to sort, so this is adequate.) 2707 */ 2708 for (i = 0; i < *num_inodes; i++) { 2709 for (j = 1; j < *num_inodes; j++) { 2710 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2711 struct xfs_inode *temp = i_tab[j]; 2712 i_tab[j] = i_tab[j-1]; 2713 i_tab[j-1] = temp; 2714 } 2715 } 2716 } 2717 } 2718 2719 static int 2720 xfs_finish_rename( 2721 struct xfs_trans *tp, 2722 struct xfs_bmap_free *free_list) 2723 { 2724 int committed = 0; 2725 int error; 2726 2727 /* 2728 * If this is a synchronous mount, make sure that the rename transaction 2729 * goes to disk before returning to the user. 2730 */ 2731 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2732 xfs_trans_set_sync(tp); 2733 2734 error = xfs_bmap_finish(&tp, free_list, &committed); 2735 if (error) { 2736 xfs_bmap_cancel(free_list); 2737 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); 2738 return error; 2739 } 2740 2741 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 2742 } 2743 2744 /* 2745 * xfs_cross_rename() 2746 * 2747 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2748 */ 2749 STATIC int 2750 xfs_cross_rename( 2751 struct xfs_trans *tp, 2752 struct xfs_inode *dp1, 2753 struct xfs_name *name1, 2754 struct xfs_inode *ip1, 2755 struct xfs_inode *dp2, 2756 struct xfs_name *name2, 2757 struct xfs_inode *ip2, 2758 struct xfs_bmap_free *free_list, 2759 xfs_fsblock_t *first_block, 2760 int spaceres) 2761 { 2762 int error = 0; 2763 int ip1_flags = 0; 2764 int ip2_flags = 0; 2765 int dp2_flags = 0; 2766 2767 /* Swap inode number for dirent in first parent */ 2768 error = xfs_dir_replace(tp, dp1, name1, 2769 ip2->i_ino, 2770 first_block, free_list, spaceres); 2771 if (error) 2772 goto out_trans_abort; 2773 2774 /* Swap inode number for dirent in second parent */ 2775 error = xfs_dir_replace(tp, dp2, name2, 2776 ip1->i_ino, 2777 first_block, free_list, spaceres); 2778 if (error) 2779 goto out_trans_abort; 2780 2781 /* 2782 * If we're renaming one or more directories across different parents, 2783 * update the respective ".." entries (and link counts) to match the new 2784 * parents. 2785 */ 2786 if (dp1 != dp2) { 2787 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2788 2789 if (S_ISDIR(ip2->i_d.di_mode)) { 2790 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2791 dp1->i_ino, first_block, 2792 free_list, spaceres); 2793 if (error) 2794 goto out_trans_abort; 2795 2796 /* transfer ip2 ".." reference to dp1 */ 2797 if (!S_ISDIR(ip1->i_d.di_mode)) { 2798 error = xfs_droplink(tp, dp2); 2799 if (error) 2800 goto out_trans_abort; 2801 error = xfs_bumplink(tp, dp1); 2802 if (error) 2803 goto out_trans_abort; 2804 } 2805 2806 /* 2807 * Although ip1 isn't changed here, userspace needs 2808 * to be warned about the change, so that applications 2809 * relying on it (like backup ones), will properly 2810 * notify the change 2811 */ 2812 ip1_flags |= XFS_ICHGTIME_CHG; 2813 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2814 } 2815 2816 if (S_ISDIR(ip1->i_d.di_mode)) { 2817 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2818 dp2->i_ino, first_block, 2819 free_list, spaceres); 2820 if (error) 2821 goto out_trans_abort; 2822 2823 /* transfer ip1 ".." reference to dp2 */ 2824 if (!S_ISDIR(ip2->i_d.di_mode)) { 2825 error = xfs_droplink(tp, dp1); 2826 if (error) 2827 goto out_trans_abort; 2828 error = xfs_bumplink(tp, dp2); 2829 if (error) 2830 goto out_trans_abort; 2831 } 2832 2833 /* 2834 * Although ip2 isn't changed here, userspace needs 2835 * to be warned about the change, so that applications 2836 * relying on it (like backup ones), will properly 2837 * notify the change 2838 */ 2839 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2840 ip2_flags |= XFS_ICHGTIME_CHG; 2841 } 2842 } 2843 2844 if (ip1_flags) { 2845 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2846 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2847 } 2848 if (ip2_flags) { 2849 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2850 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2851 } 2852 if (dp2_flags) { 2853 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2854 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2855 } 2856 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2857 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2858 return xfs_finish_rename(tp, free_list); 2859 2860 out_trans_abort: 2861 xfs_bmap_cancel(free_list); 2862 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT); 2863 return error; 2864 } 2865 2866 /* 2867 * xfs_rename_alloc_whiteout() 2868 * 2869 * Return a referenced, unlinked, unlocked inode that that can be used as a 2870 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2871 * crash between allocating the inode and linking it into the rename transaction 2872 * recovery will free the inode and we won't leak it. 2873 */ 2874 static int 2875 xfs_rename_alloc_whiteout( 2876 struct xfs_inode *dp, 2877 struct xfs_inode **wip) 2878 { 2879 struct xfs_inode *tmpfile; 2880 int error; 2881 2882 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2883 if (error) 2884 return error; 2885 2886 /* Satisfy xfs_bumplink that this is a real tmpfile */ 2887 xfs_finish_inode_setup(tmpfile); 2888 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2889 2890 *wip = tmpfile; 2891 return 0; 2892 } 2893 2894 /* 2895 * xfs_rename 2896 */ 2897 int 2898 xfs_rename( 2899 struct xfs_inode *src_dp, 2900 struct xfs_name *src_name, 2901 struct xfs_inode *src_ip, 2902 struct xfs_inode *target_dp, 2903 struct xfs_name *target_name, 2904 struct xfs_inode *target_ip, 2905 unsigned int flags) 2906 { 2907 struct xfs_mount *mp = src_dp->i_mount; 2908 struct xfs_trans *tp; 2909 struct xfs_bmap_free free_list; 2910 xfs_fsblock_t first_block; 2911 struct xfs_inode *wip = NULL; /* whiteout inode */ 2912 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2913 int num_inodes = __XFS_SORT_INODES; 2914 bool new_parent = (src_dp != target_dp); 2915 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode); 2916 int cancel_flags = 0; 2917 int spaceres; 2918 int error; 2919 2920 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2921 2922 if ((flags & RENAME_EXCHANGE) && !target_ip) 2923 return -EINVAL; 2924 2925 /* 2926 * If we are doing a whiteout operation, allocate the whiteout inode 2927 * we will be placing at the target and ensure the type is set 2928 * appropriately. 2929 */ 2930 if (flags & RENAME_WHITEOUT) { 2931 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2932 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2933 if (error) 2934 return error; 2935 2936 /* setup target dirent info as whiteout */ 2937 src_name->type = XFS_DIR3_FT_CHRDEV; 2938 } 2939 2940 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2941 inodes, &num_inodes); 2942 2943 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME); 2944 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2945 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0); 2946 if (error == -ENOSPC) { 2947 spaceres = 0; 2948 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0); 2949 } 2950 if (error) 2951 goto out_trans_cancel; 2952 cancel_flags = XFS_TRANS_RELEASE_LOG_RES; 2953 2954 /* 2955 * Attach the dquots to the inodes 2956 */ 2957 error = xfs_qm_vop_rename_dqattach(inodes); 2958 if (error) 2959 goto out_trans_cancel; 2960 2961 /* 2962 * Lock all the participating inodes. Depending upon whether 2963 * the target_name exists in the target directory, and 2964 * whether the target directory is the same as the source 2965 * directory, we can lock from 2 to 4 inodes. 2966 */ 2967 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2968 2969 /* 2970 * Join all the inodes to the transaction. From this point on, 2971 * we can rely on either trans_commit or trans_cancel to unlock 2972 * them. 2973 */ 2974 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 2975 if (new_parent) 2976 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 2977 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2978 if (target_ip) 2979 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2980 if (wip) 2981 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2982 2983 /* 2984 * If we are using project inheritance, we only allow renames 2985 * into our tree when the project IDs are the same; else the 2986 * tree quota mechanism would be circumvented. 2987 */ 2988 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2989 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2990 error = -EXDEV; 2991 goto out_trans_cancel; 2992 } 2993 2994 xfs_bmap_init(&free_list, &first_block); 2995 2996 /* RENAME_EXCHANGE is unique from here on. */ 2997 if (flags & RENAME_EXCHANGE) 2998 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2999 target_dp, target_name, target_ip, 3000 &free_list, &first_block, spaceres); 3001 3002 /* 3003 * Set up the target. 3004 */ 3005 if (target_ip == NULL) { 3006 /* 3007 * If there's no space reservation, check the entry will 3008 * fit before actually inserting it. 3009 */ 3010 if (!spaceres) { 3011 error = xfs_dir_canenter(tp, target_dp, target_name); 3012 if (error) 3013 goto out_trans_cancel; 3014 } 3015 /* 3016 * If target does not exist and the rename crosses 3017 * directories, adjust the target directory link count 3018 * to account for the ".." reference from the new entry. 3019 */ 3020 error = xfs_dir_createname(tp, target_dp, target_name, 3021 src_ip->i_ino, &first_block, 3022 &free_list, spaceres); 3023 if (error == -ENOSPC) 3024 goto out_bmap_cancel; 3025 if (error) 3026 goto out_trans_abort; 3027 3028 xfs_trans_ichgtime(tp, target_dp, 3029 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3030 3031 if (new_parent && src_is_directory) { 3032 error = xfs_bumplink(tp, target_dp); 3033 if (error) 3034 goto out_trans_abort; 3035 } 3036 } else { /* target_ip != NULL */ 3037 /* 3038 * If target exists and it's a directory, check that both 3039 * target and source are directories and that target can be 3040 * destroyed, or that neither is a directory. 3041 */ 3042 if (S_ISDIR(target_ip->i_d.di_mode)) { 3043 /* 3044 * Make sure target dir is empty. 3045 */ 3046 if (!(xfs_dir_isempty(target_ip)) || 3047 (target_ip->i_d.di_nlink > 2)) { 3048 error = -EEXIST; 3049 goto out_trans_cancel; 3050 } 3051 } 3052 3053 /* 3054 * Link the source inode under the target name. 3055 * If the source inode is a directory and we are moving 3056 * it across directories, its ".." entry will be 3057 * inconsistent until we replace that down below. 3058 * 3059 * In case there is already an entry with the same 3060 * name at the destination directory, remove it first. 3061 */ 3062 error = xfs_dir_replace(tp, target_dp, target_name, 3063 src_ip->i_ino, 3064 &first_block, &free_list, spaceres); 3065 if (error) 3066 goto out_trans_abort; 3067 3068 xfs_trans_ichgtime(tp, target_dp, 3069 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3070 3071 /* 3072 * Decrement the link count on the target since the target 3073 * dir no longer points to it. 3074 */ 3075 error = xfs_droplink(tp, target_ip); 3076 if (error) 3077 goto out_trans_abort; 3078 3079 if (src_is_directory) { 3080 /* 3081 * Drop the link from the old "." entry. 3082 */ 3083 error = xfs_droplink(tp, target_ip); 3084 if (error) 3085 goto out_trans_abort; 3086 } 3087 } /* target_ip != NULL */ 3088 3089 /* 3090 * Remove the source. 3091 */ 3092 if (new_parent && src_is_directory) { 3093 /* 3094 * Rewrite the ".." entry to point to the new 3095 * directory. 3096 */ 3097 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3098 target_dp->i_ino, 3099 &first_block, &free_list, spaceres); 3100 ASSERT(error != -EEXIST); 3101 if (error) 3102 goto out_trans_abort; 3103 } 3104 3105 /* 3106 * We always want to hit the ctime on the source inode. 3107 * 3108 * This isn't strictly required by the standards since the source 3109 * inode isn't really being changed, but old unix file systems did 3110 * it and some incremental backup programs won't work without it. 3111 */ 3112 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3113 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3114 3115 /* 3116 * Adjust the link count on src_dp. This is necessary when 3117 * renaming a directory, either within one parent when 3118 * the target existed, or across two parent directories. 3119 */ 3120 if (src_is_directory && (new_parent || target_ip != NULL)) { 3121 3122 /* 3123 * Decrement link count on src_directory since the 3124 * entry that's moved no longer points to it. 3125 */ 3126 error = xfs_droplink(tp, src_dp); 3127 if (error) 3128 goto out_trans_abort; 3129 } 3130 3131 /* 3132 * For whiteouts, we only need to update the source dirent with the 3133 * inode number of the whiteout inode rather than removing it 3134 * altogether. 3135 */ 3136 if (wip) { 3137 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3138 &first_block, &free_list, spaceres); 3139 } else 3140 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3141 &first_block, &free_list, spaceres); 3142 if (error) 3143 goto out_trans_abort; 3144 3145 /* 3146 * For whiteouts, we need to bump the link count on the whiteout inode. 3147 * This means that failures all the way up to this point leave the inode 3148 * on the unlinked list and so cleanup is a simple matter of dropping 3149 * the remaining reference to it. If we fail here after bumping the link 3150 * count, we're shutting down the filesystem so we'll never see the 3151 * intermediate state on disk. 3152 */ 3153 if (wip) { 3154 ASSERT(wip->i_d.di_nlink == 0); 3155 error = xfs_bumplink(tp, wip); 3156 if (error) 3157 goto out_trans_abort; 3158 error = xfs_iunlink_remove(tp, wip); 3159 if (error) 3160 goto out_trans_abort; 3161 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3162 3163 /* 3164 * Now we have a real link, clear the "I'm a tmpfile" state 3165 * flag from the inode so it doesn't accidentally get misused in 3166 * future. 3167 */ 3168 VFS_I(wip)->i_state &= ~I_LINKABLE; 3169 } 3170 3171 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3172 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3173 if (new_parent) 3174 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3175 3176 error = xfs_finish_rename(tp, &free_list); 3177 if (wip) 3178 IRELE(wip); 3179 return error; 3180 3181 out_trans_abort: 3182 cancel_flags |= XFS_TRANS_ABORT; 3183 out_bmap_cancel: 3184 xfs_bmap_cancel(&free_list); 3185 out_trans_cancel: 3186 xfs_trans_cancel(tp, cancel_flags); 3187 if (wip) 3188 IRELE(wip); 3189 return error; 3190 } 3191 3192 STATIC int 3193 xfs_iflush_cluster( 3194 xfs_inode_t *ip, 3195 xfs_buf_t *bp) 3196 { 3197 xfs_mount_t *mp = ip->i_mount; 3198 struct xfs_perag *pag; 3199 unsigned long first_index, mask; 3200 unsigned long inodes_per_cluster; 3201 int ilist_size; 3202 xfs_inode_t **ilist; 3203 xfs_inode_t *iq; 3204 int nr_found; 3205 int clcount = 0; 3206 int bufwasdelwri; 3207 int i; 3208 3209 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3210 3211 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3212 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3213 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); 3214 if (!ilist) 3215 goto out_put; 3216 3217 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3218 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3219 rcu_read_lock(); 3220 /* really need a gang lookup range call here */ 3221 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, 3222 first_index, inodes_per_cluster); 3223 if (nr_found == 0) 3224 goto out_free; 3225 3226 for (i = 0; i < nr_found; i++) { 3227 iq = ilist[i]; 3228 if (iq == ip) 3229 continue; 3230 3231 /* 3232 * because this is an RCU protected lookup, we could find a 3233 * recently freed or even reallocated inode during the lookup. 3234 * We need to check under the i_flags_lock for a valid inode 3235 * here. Skip it if it is not valid or the wrong inode. 3236 */ 3237 spin_lock(&ip->i_flags_lock); 3238 if (!ip->i_ino || 3239 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { 3240 spin_unlock(&ip->i_flags_lock); 3241 continue; 3242 } 3243 spin_unlock(&ip->i_flags_lock); 3244 3245 /* 3246 * Do an un-protected check to see if the inode is dirty and 3247 * is a candidate for flushing. These checks will be repeated 3248 * later after the appropriate locks are acquired. 3249 */ 3250 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) 3251 continue; 3252 3253 /* 3254 * Try to get locks. If any are unavailable or it is pinned, 3255 * then this inode cannot be flushed and is skipped. 3256 */ 3257 3258 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) 3259 continue; 3260 if (!xfs_iflock_nowait(iq)) { 3261 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3262 continue; 3263 } 3264 if (xfs_ipincount(iq)) { 3265 xfs_ifunlock(iq); 3266 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3267 continue; 3268 } 3269 3270 /* 3271 * arriving here means that this inode can be flushed. First 3272 * re-check that it's dirty before flushing. 3273 */ 3274 if (!xfs_inode_clean(iq)) { 3275 int error; 3276 error = xfs_iflush_int(iq, bp); 3277 if (error) { 3278 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3279 goto cluster_corrupt_out; 3280 } 3281 clcount++; 3282 } else { 3283 xfs_ifunlock(iq); 3284 } 3285 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3286 } 3287 3288 if (clcount) { 3289 XFS_STATS_INC(xs_icluster_flushcnt); 3290 XFS_STATS_ADD(xs_icluster_flushinode, clcount); 3291 } 3292 3293 out_free: 3294 rcu_read_unlock(); 3295 kmem_free(ilist); 3296 out_put: 3297 xfs_perag_put(pag); 3298 return 0; 3299 3300 3301 cluster_corrupt_out: 3302 /* 3303 * Corruption detected in the clustering loop. Invalidate the 3304 * inode buffer and shut down the filesystem. 3305 */ 3306 rcu_read_unlock(); 3307 /* 3308 * Clean up the buffer. If it was delwri, just release it -- 3309 * brelse can handle it with no problems. If not, shut down the 3310 * filesystem before releasing the buffer. 3311 */ 3312 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3313 if (bufwasdelwri) 3314 xfs_buf_relse(bp); 3315 3316 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3317 3318 if (!bufwasdelwri) { 3319 /* 3320 * Just like incore_relse: if we have b_iodone functions, 3321 * mark the buffer as an error and call them. Otherwise 3322 * mark it as stale and brelse. 3323 */ 3324 if (bp->b_iodone) { 3325 XFS_BUF_UNDONE(bp); 3326 xfs_buf_stale(bp); 3327 xfs_buf_ioerror(bp, -EIO); 3328 xfs_buf_ioend(bp); 3329 } else { 3330 xfs_buf_stale(bp); 3331 xfs_buf_relse(bp); 3332 } 3333 } 3334 3335 /* 3336 * Unlocks the flush lock 3337 */ 3338 xfs_iflush_abort(iq, false); 3339 kmem_free(ilist); 3340 xfs_perag_put(pag); 3341 return -EFSCORRUPTED; 3342 } 3343 3344 /* 3345 * Flush dirty inode metadata into the backing buffer. 3346 * 3347 * The caller must have the inode lock and the inode flush lock held. The 3348 * inode lock will still be held upon return to the caller, and the inode 3349 * flush lock will be released after the inode has reached the disk. 3350 * 3351 * The caller must write out the buffer returned in *bpp and release it. 3352 */ 3353 int 3354 xfs_iflush( 3355 struct xfs_inode *ip, 3356 struct xfs_buf **bpp) 3357 { 3358 struct xfs_mount *mp = ip->i_mount; 3359 struct xfs_buf *bp; 3360 struct xfs_dinode *dip; 3361 int error; 3362 3363 XFS_STATS_INC(xs_iflush_count); 3364 3365 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3366 ASSERT(xfs_isiflocked(ip)); 3367 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3368 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3369 3370 *bpp = NULL; 3371 3372 xfs_iunpin_wait(ip); 3373 3374 /* 3375 * For stale inodes we cannot rely on the backing buffer remaining 3376 * stale in cache for the remaining life of the stale inode and so 3377 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3378 * inodes below. We have to check this after ensuring the inode is 3379 * unpinned so that it is safe to reclaim the stale inode after the 3380 * flush call. 3381 */ 3382 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3383 xfs_ifunlock(ip); 3384 return 0; 3385 } 3386 3387 /* 3388 * This may have been unpinned because the filesystem is shutting 3389 * down forcibly. If that's the case we must not write this inode 3390 * to disk, because the log record didn't make it to disk. 3391 * 3392 * We also have to remove the log item from the AIL in this case, 3393 * as we wait for an empty AIL as part of the unmount process. 3394 */ 3395 if (XFS_FORCED_SHUTDOWN(mp)) { 3396 error = -EIO; 3397 goto abort_out; 3398 } 3399 3400 /* 3401 * Get the buffer containing the on-disk inode. 3402 */ 3403 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3404 0); 3405 if (error || !bp) { 3406 xfs_ifunlock(ip); 3407 return error; 3408 } 3409 3410 /* 3411 * First flush out the inode that xfs_iflush was called with. 3412 */ 3413 error = xfs_iflush_int(ip, bp); 3414 if (error) 3415 goto corrupt_out; 3416 3417 /* 3418 * If the buffer is pinned then push on the log now so we won't 3419 * get stuck waiting in the write for too long. 3420 */ 3421 if (xfs_buf_ispinned(bp)) 3422 xfs_log_force(mp, 0); 3423 3424 /* 3425 * inode clustering: 3426 * see if other inodes can be gathered into this write 3427 */ 3428 error = xfs_iflush_cluster(ip, bp); 3429 if (error) 3430 goto cluster_corrupt_out; 3431 3432 *bpp = bp; 3433 return 0; 3434 3435 corrupt_out: 3436 xfs_buf_relse(bp); 3437 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3438 cluster_corrupt_out: 3439 error = -EFSCORRUPTED; 3440 abort_out: 3441 /* 3442 * Unlocks the flush lock 3443 */ 3444 xfs_iflush_abort(ip, false); 3445 return error; 3446 } 3447 3448 STATIC int 3449 xfs_iflush_int( 3450 struct xfs_inode *ip, 3451 struct xfs_buf *bp) 3452 { 3453 struct xfs_inode_log_item *iip = ip->i_itemp; 3454 struct xfs_dinode *dip; 3455 struct xfs_mount *mp = ip->i_mount; 3456 3457 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3458 ASSERT(xfs_isiflocked(ip)); 3459 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3460 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3461 ASSERT(iip != NULL && iip->ili_fields != 0); 3462 ASSERT(ip->i_d.di_version > 1); 3463 3464 /* set *dip = inode's place in the buffer */ 3465 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); 3466 3467 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3468 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 3469 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3470 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3471 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3472 goto corrupt_out; 3473 } 3474 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, 3475 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { 3476 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3477 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", 3478 __func__, ip->i_ino, ip, ip->i_d.di_magic); 3479 goto corrupt_out; 3480 } 3481 if (S_ISREG(ip->i_d.di_mode)) { 3482 if (XFS_TEST_ERROR( 3483 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3484 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3485 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 3486 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3487 "%s: Bad regular inode %Lu, ptr 0x%p", 3488 __func__, ip->i_ino, ip); 3489 goto corrupt_out; 3490 } 3491 } else if (S_ISDIR(ip->i_d.di_mode)) { 3492 if (XFS_TEST_ERROR( 3493 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3494 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3495 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3496 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 3497 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3498 "%s: Bad directory inode %Lu, ptr 0x%p", 3499 __func__, ip->i_ino, ip); 3500 goto corrupt_out; 3501 } 3502 } 3503 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3504 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 3505 XFS_RANDOM_IFLUSH_5)) { 3506 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3507 "%s: detected corrupt incore inode %Lu, " 3508 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3509 __func__, ip->i_ino, 3510 ip->i_d.di_nextents + ip->i_d.di_anextents, 3511 ip->i_d.di_nblocks, ip); 3512 goto corrupt_out; 3513 } 3514 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3515 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 3516 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3517 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3518 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3519 goto corrupt_out; 3520 } 3521 3522 /* 3523 * Inode item log recovery for v2 inodes are dependent on the 3524 * di_flushiter count for correct sequencing. We bump the flush 3525 * iteration count so we can detect flushes which postdate a log record 3526 * during recovery. This is redundant as we now log every change and 3527 * hence this can't happen but we need to still do it to ensure 3528 * backwards compatibility with old kernels that predate logging all 3529 * inode changes. 3530 */ 3531 if (ip->i_d.di_version < 3) 3532 ip->i_d.di_flushiter++; 3533 3534 /* 3535 * Copy the dirty parts of the inode into the on-disk 3536 * inode. We always copy out the core of the inode, 3537 * because if the inode is dirty at all the core must 3538 * be. 3539 */ 3540 xfs_dinode_to_disk(dip, &ip->i_d); 3541 3542 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3543 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3544 ip->i_d.di_flushiter = 0; 3545 3546 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3547 if (XFS_IFORK_Q(ip)) 3548 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3549 xfs_inobp_check(mp, bp); 3550 3551 /* 3552 * We've recorded everything logged in the inode, so we'd like to clear 3553 * the ili_fields bits so we don't log and flush things unnecessarily. 3554 * However, we can't stop logging all this information until the data 3555 * we've copied into the disk buffer is written to disk. If we did we 3556 * might overwrite the copy of the inode in the log with all the data 3557 * after re-logging only part of it, and in the face of a crash we 3558 * wouldn't have all the data we need to recover. 3559 * 3560 * What we do is move the bits to the ili_last_fields field. When 3561 * logging the inode, these bits are moved back to the ili_fields field. 3562 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3563 * know that the information those bits represent is permanently on 3564 * disk. As long as the flush completes before the inode is logged 3565 * again, then both ili_fields and ili_last_fields will be cleared. 3566 * 3567 * We can play with the ili_fields bits here, because the inode lock 3568 * must be held exclusively in order to set bits there and the flush 3569 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3570 * done routine can tell whether or not to look in the AIL. Also, store 3571 * the current LSN of the inode so that we can tell whether the item has 3572 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3573 * need the AIL lock, because it is a 64 bit value that cannot be read 3574 * atomically. 3575 */ 3576 iip->ili_last_fields = iip->ili_fields; 3577 iip->ili_fields = 0; 3578 iip->ili_logged = 1; 3579 3580 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3581 &iip->ili_item.li_lsn); 3582 3583 /* 3584 * Attach the function xfs_iflush_done to the inode's 3585 * buffer. This will remove the inode from the AIL 3586 * and unlock the inode's flush lock when the inode is 3587 * completely written to disk. 3588 */ 3589 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3590 3591 /* update the lsn in the on disk inode if required */ 3592 if (ip->i_d.di_version == 3) 3593 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); 3594 3595 /* generate the checksum. */ 3596 xfs_dinode_calc_crc(mp, dip); 3597 3598 ASSERT(bp->b_fspriv != NULL); 3599 ASSERT(bp->b_iodone != NULL); 3600 return 0; 3601 3602 corrupt_out: 3603 return -EFSCORRUPTED; 3604 } 3605