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_inode_t *ip; 909 xfs_buf_t *ialloc_context = NULL; 910 int code; 911 void *dqinfo; 912 uint tflags; 913 914 tp = *tpp; 915 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 916 917 /* 918 * xfs_ialloc will return a pointer to an incore inode if 919 * the Space Manager has an available inode on the free 920 * list. Otherwise, it will do an allocation and replenish 921 * the freelist. Since we can only do one allocation per 922 * transaction without deadlocks, we will need to commit the 923 * current transaction and start a new one. We will then 924 * need to call xfs_ialloc again to get the inode. 925 * 926 * If xfs_ialloc did an allocation to replenish the freelist, 927 * it returns the bp containing the head of the freelist as 928 * ialloc_context. We will hold a lock on it across the 929 * transaction commit so that no other process can steal 930 * the inode(s) that we've just allocated. 931 */ 932 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, okalloc, 933 &ialloc_context, &ip); 934 935 /* 936 * Return an error if we were unable to allocate a new inode. 937 * This should only happen if we run out of space on disk or 938 * encounter a disk error. 939 */ 940 if (code) { 941 *ipp = NULL; 942 return code; 943 } 944 if (!ialloc_context && !ip) { 945 *ipp = NULL; 946 return -ENOSPC; 947 } 948 949 /* 950 * If the AGI buffer is non-NULL, then we were unable to get an 951 * inode in one operation. We need to commit the current 952 * transaction and call xfs_ialloc() again. It is guaranteed 953 * to succeed the second time. 954 */ 955 if (ialloc_context) { 956 /* 957 * Normally, xfs_trans_commit releases all the locks. 958 * We call bhold to hang on to the ialloc_context across 959 * the commit. Holding this buffer prevents any other 960 * processes from doing any allocations in this 961 * allocation group. 962 */ 963 xfs_trans_bhold(tp, ialloc_context); 964 965 /* 966 * We want the quota changes to be associated with the next 967 * transaction, NOT this one. So, detach the dqinfo from this 968 * and attach it to the next transaction. 969 */ 970 dqinfo = NULL; 971 tflags = 0; 972 if (tp->t_dqinfo) { 973 dqinfo = (void *)tp->t_dqinfo; 974 tp->t_dqinfo = NULL; 975 tflags = tp->t_flags & XFS_TRANS_DQ_DIRTY; 976 tp->t_flags &= ~(XFS_TRANS_DQ_DIRTY); 977 } 978 979 code = xfs_trans_roll(&tp, 0); 980 if (committed != NULL) 981 *committed = 1; 982 983 /* 984 * Re-attach the quota info that we detached from prev trx. 985 */ 986 if (dqinfo) { 987 tp->t_dqinfo = dqinfo; 988 tp->t_flags |= tflags; 989 } 990 991 if (code) { 992 xfs_buf_relse(ialloc_context); 993 *tpp = tp; 994 *ipp = NULL; 995 return code; 996 } 997 xfs_trans_bjoin(tp, ialloc_context); 998 999 /* 1000 * Call ialloc again. Since we've locked out all 1001 * other allocations in this allocation group, 1002 * this call should always succeed. 1003 */ 1004 code = xfs_ialloc(tp, dp, mode, nlink, rdev, prid, 1005 okalloc, &ialloc_context, &ip); 1006 1007 /* 1008 * If we get an error at this point, return to the caller 1009 * so that the current transaction can be aborted. 1010 */ 1011 if (code) { 1012 *tpp = tp; 1013 *ipp = NULL; 1014 return code; 1015 } 1016 ASSERT(!ialloc_context && ip); 1017 1018 } else { 1019 if (committed != NULL) 1020 *committed = 0; 1021 } 1022 1023 *ipp = ip; 1024 *tpp = tp; 1025 1026 return 0; 1027 } 1028 1029 /* 1030 * Decrement the link count on an inode & log the change. 1031 * If this causes the link count to go to zero, initiate the 1032 * logging activity required to truncate a file. 1033 */ 1034 int /* error */ 1035 xfs_droplink( 1036 xfs_trans_t *tp, 1037 xfs_inode_t *ip) 1038 { 1039 int error; 1040 1041 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1042 1043 ASSERT (ip->i_d.di_nlink > 0); 1044 ip->i_d.di_nlink--; 1045 drop_nlink(VFS_I(ip)); 1046 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1047 1048 error = 0; 1049 if (ip->i_d.di_nlink == 0) { 1050 /* 1051 * We're dropping the last link to this file. 1052 * Move the on-disk inode to the AGI unlinked list. 1053 * From xfs_inactive() we will pull the inode from 1054 * the list and free it. 1055 */ 1056 error = xfs_iunlink(tp, ip); 1057 } 1058 return error; 1059 } 1060 1061 /* 1062 * Increment the link count on an inode & log the change. 1063 */ 1064 int 1065 xfs_bumplink( 1066 xfs_trans_t *tp, 1067 xfs_inode_t *ip) 1068 { 1069 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG); 1070 1071 ASSERT(ip->i_d.di_version > 1); 1072 ASSERT(ip->i_d.di_nlink > 0 || (VFS_I(ip)->i_state & I_LINKABLE)); 1073 ip->i_d.di_nlink++; 1074 inc_nlink(VFS_I(ip)); 1075 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1076 return 0; 1077 } 1078 1079 int 1080 xfs_create( 1081 xfs_inode_t *dp, 1082 struct xfs_name *name, 1083 umode_t mode, 1084 xfs_dev_t rdev, 1085 xfs_inode_t **ipp) 1086 { 1087 int is_dir = S_ISDIR(mode); 1088 struct xfs_mount *mp = dp->i_mount; 1089 struct xfs_inode *ip = NULL; 1090 struct xfs_trans *tp = NULL; 1091 int error; 1092 xfs_bmap_free_t free_list; 1093 xfs_fsblock_t first_block; 1094 bool unlock_dp_on_error = false; 1095 int committed; 1096 prid_t prid; 1097 struct xfs_dquot *udqp = NULL; 1098 struct xfs_dquot *gdqp = NULL; 1099 struct xfs_dquot *pdqp = NULL; 1100 struct xfs_trans_res *tres; 1101 uint resblks; 1102 1103 trace_xfs_create(dp, name); 1104 1105 if (XFS_FORCED_SHUTDOWN(mp)) 1106 return -EIO; 1107 1108 prid = xfs_get_initial_prid(dp); 1109 1110 /* 1111 * Make sure that we have allocated dquot(s) on disk. 1112 */ 1113 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1114 xfs_kgid_to_gid(current_fsgid()), prid, 1115 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1116 &udqp, &gdqp, &pdqp); 1117 if (error) 1118 return error; 1119 1120 if (is_dir) { 1121 rdev = 0; 1122 resblks = XFS_MKDIR_SPACE_RES(mp, name->len); 1123 tres = &M_RES(mp)->tr_mkdir; 1124 tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR); 1125 } else { 1126 resblks = XFS_CREATE_SPACE_RES(mp, name->len); 1127 tres = &M_RES(mp)->tr_create; 1128 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE); 1129 } 1130 1131 /* 1132 * Initially assume that the file does not exist and 1133 * reserve the resources for that case. If that is not 1134 * the case we'll drop the one we have and get a more 1135 * appropriate transaction later. 1136 */ 1137 error = xfs_trans_reserve(tp, tres, resblks, 0); 1138 if (error == -ENOSPC) { 1139 /* flush outstanding delalloc blocks and retry */ 1140 xfs_flush_inodes(mp); 1141 error = xfs_trans_reserve(tp, tres, resblks, 0); 1142 } 1143 if (error == -ENOSPC) { 1144 /* No space at all so try a "no-allocation" reservation */ 1145 resblks = 0; 1146 error = xfs_trans_reserve(tp, tres, 0, 0); 1147 } 1148 if (error) 1149 goto out_trans_cancel; 1150 1151 1152 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 1153 unlock_dp_on_error = true; 1154 1155 xfs_bmap_init(&free_list, &first_block); 1156 1157 /* 1158 * Reserve disk quota and the inode. 1159 */ 1160 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1161 pdqp, resblks, 1, 0); 1162 if (error) 1163 goto out_trans_cancel; 1164 1165 if (!resblks) { 1166 error = xfs_dir_canenter(tp, dp, name); 1167 if (error) 1168 goto out_trans_cancel; 1169 } 1170 1171 /* 1172 * A newly created regular or special file just has one directory 1173 * entry pointing to them, but a directory also the "." entry 1174 * pointing to itself. 1175 */ 1176 error = xfs_dir_ialloc(&tp, dp, mode, is_dir ? 2 : 1, rdev, 1177 prid, resblks > 0, &ip, &committed); 1178 if (error) { 1179 if (error == -ENOSPC) 1180 goto out_trans_cancel; 1181 goto out_trans_cancel; 1182 } 1183 1184 /* 1185 * Now we join the directory inode to the transaction. We do not do it 1186 * earlier because xfs_dir_ialloc might commit the previous transaction 1187 * (and release all the locks). An error from here on will result in 1188 * the transaction cancel unlocking dp so don't do it explicitly in the 1189 * error path. 1190 */ 1191 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 1192 unlock_dp_on_error = false; 1193 1194 error = xfs_dir_createname(tp, dp, name, ip->i_ino, 1195 &first_block, &free_list, resblks ? 1196 resblks - XFS_IALLOC_SPACE_RES(mp) : 0); 1197 if (error) { 1198 ASSERT(error != -ENOSPC); 1199 goto out_trans_cancel; 1200 } 1201 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1202 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 1203 1204 if (is_dir) { 1205 error = xfs_dir_init(tp, ip, dp); 1206 if (error) 1207 goto out_bmap_cancel; 1208 1209 error = xfs_bumplink(tp, dp); 1210 if (error) 1211 goto out_bmap_cancel; 1212 } 1213 1214 /* 1215 * If this is a synchronous mount, make sure that the 1216 * create transaction goes to disk before returning to 1217 * the user. 1218 */ 1219 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 1220 xfs_trans_set_sync(tp); 1221 1222 /* 1223 * Attach the dquot(s) to the inodes and modify them incore. 1224 * These ids of the inode couldn't have changed since the new 1225 * inode has been locked ever since it was created. 1226 */ 1227 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1228 1229 error = xfs_bmap_finish(&tp, &free_list, &committed); 1230 if (error) 1231 goto out_bmap_cancel; 1232 1233 error = xfs_trans_commit(tp); 1234 if (error) 1235 goto out_release_inode; 1236 1237 xfs_qm_dqrele(udqp); 1238 xfs_qm_dqrele(gdqp); 1239 xfs_qm_dqrele(pdqp); 1240 1241 *ipp = ip; 1242 return 0; 1243 1244 out_bmap_cancel: 1245 xfs_bmap_cancel(&free_list); 1246 out_trans_cancel: 1247 xfs_trans_cancel(tp); 1248 out_release_inode: 1249 /* 1250 * Wait until after the current transaction is aborted to finish the 1251 * setup of the inode and release the inode. This prevents recursive 1252 * transactions and deadlocks from xfs_inactive. 1253 */ 1254 if (ip) { 1255 xfs_finish_inode_setup(ip); 1256 IRELE(ip); 1257 } 1258 1259 xfs_qm_dqrele(udqp); 1260 xfs_qm_dqrele(gdqp); 1261 xfs_qm_dqrele(pdqp); 1262 1263 if (unlock_dp_on_error) 1264 xfs_iunlock(dp, XFS_ILOCK_EXCL); 1265 return error; 1266 } 1267 1268 int 1269 xfs_create_tmpfile( 1270 struct xfs_inode *dp, 1271 struct dentry *dentry, 1272 umode_t mode, 1273 struct xfs_inode **ipp) 1274 { 1275 struct xfs_mount *mp = dp->i_mount; 1276 struct xfs_inode *ip = NULL; 1277 struct xfs_trans *tp = NULL; 1278 int error; 1279 prid_t prid; 1280 struct xfs_dquot *udqp = NULL; 1281 struct xfs_dquot *gdqp = NULL; 1282 struct xfs_dquot *pdqp = NULL; 1283 struct xfs_trans_res *tres; 1284 uint resblks; 1285 1286 if (XFS_FORCED_SHUTDOWN(mp)) 1287 return -EIO; 1288 1289 prid = xfs_get_initial_prid(dp); 1290 1291 /* 1292 * Make sure that we have allocated dquot(s) on disk. 1293 */ 1294 error = xfs_qm_vop_dqalloc(dp, xfs_kuid_to_uid(current_fsuid()), 1295 xfs_kgid_to_gid(current_fsgid()), prid, 1296 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 1297 &udqp, &gdqp, &pdqp); 1298 if (error) 1299 return error; 1300 1301 resblks = XFS_IALLOC_SPACE_RES(mp); 1302 tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE_TMPFILE); 1303 1304 tres = &M_RES(mp)->tr_create_tmpfile; 1305 error = xfs_trans_reserve(tp, tres, resblks, 0); 1306 if (error == -ENOSPC) { 1307 /* No space at all so try a "no-allocation" reservation */ 1308 resblks = 0; 1309 error = xfs_trans_reserve(tp, tres, 0, 0); 1310 } 1311 if (error) 1312 goto out_trans_cancel; 1313 1314 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 1315 pdqp, resblks, 1, 0); 1316 if (error) 1317 goto out_trans_cancel; 1318 1319 error = xfs_dir_ialloc(&tp, dp, mode, 1, 0, 1320 prid, resblks > 0, &ip, NULL); 1321 if (error) { 1322 if (error == -ENOSPC) 1323 goto out_trans_cancel; 1324 goto out_trans_cancel; 1325 } 1326 1327 if (mp->m_flags & XFS_MOUNT_WSYNC) 1328 xfs_trans_set_sync(tp); 1329 1330 /* 1331 * Attach the dquot(s) to the inodes and modify them incore. 1332 * These ids of the inode couldn't have changed since the new 1333 * inode has been locked ever since it was created. 1334 */ 1335 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 1336 1337 ip->i_d.di_nlink--; 1338 error = xfs_iunlink(tp, ip); 1339 if (error) 1340 goto out_trans_cancel; 1341 1342 error = xfs_trans_commit(tp); 1343 if (error) 1344 goto out_release_inode; 1345 1346 xfs_qm_dqrele(udqp); 1347 xfs_qm_dqrele(gdqp); 1348 xfs_qm_dqrele(pdqp); 1349 1350 *ipp = ip; 1351 return 0; 1352 1353 out_trans_cancel: 1354 xfs_trans_cancel(tp); 1355 out_release_inode: 1356 /* 1357 * Wait until after the current transaction is aborted to finish the 1358 * setup of the inode and release the inode. This prevents recursive 1359 * transactions and deadlocks from xfs_inactive. 1360 */ 1361 if (ip) { 1362 xfs_finish_inode_setup(ip); 1363 IRELE(ip); 1364 } 1365 1366 xfs_qm_dqrele(udqp); 1367 xfs_qm_dqrele(gdqp); 1368 xfs_qm_dqrele(pdqp); 1369 1370 return error; 1371 } 1372 1373 int 1374 xfs_link( 1375 xfs_inode_t *tdp, 1376 xfs_inode_t *sip, 1377 struct xfs_name *target_name) 1378 { 1379 xfs_mount_t *mp = tdp->i_mount; 1380 xfs_trans_t *tp; 1381 int error; 1382 xfs_bmap_free_t free_list; 1383 xfs_fsblock_t first_block; 1384 int committed; 1385 int resblks; 1386 1387 trace_xfs_link(tdp, target_name); 1388 1389 ASSERT(!S_ISDIR(sip->i_d.di_mode)); 1390 1391 if (XFS_FORCED_SHUTDOWN(mp)) 1392 return -EIO; 1393 1394 error = xfs_qm_dqattach(sip, 0); 1395 if (error) 1396 goto std_return; 1397 1398 error = xfs_qm_dqattach(tdp, 0); 1399 if (error) 1400 goto std_return; 1401 1402 tp = xfs_trans_alloc(mp, XFS_TRANS_LINK); 1403 resblks = XFS_LINK_SPACE_RES(mp, target_name->len); 1404 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, resblks, 0); 1405 if (error == -ENOSPC) { 1406 resblks = 0; 1407 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_link, 0, 0); 1408 } 1409 if (error) 1410 goto error_return; 1411 1412 xfs_lock_two_inodes(sip, tdp, XFS_ILOCK_EXCL); 1413 1414 xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL); 1415 xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL); 1416 1417 /* 1418 * If we are using project inheritance, we only allow hard link 1419 * creation in our tree when the project IDs are the same; else 1420 * the tree quota mechanism could be circumvented. 1421 */ 1422 if (unlikely((tdp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 1423 (xfs_get_projid(tdp) != xfs_get_projid(sip)))) { 1424 error = -EXDEV; 1425 goto error_return; 1426 } 1427 1428 if (!resblks) { 1429 error = xfs_dir_canenter(tp, tdp, target_name); 1430 if (error) 1431 goto error_return; 1432 } 1433 1434 xfs_bmap_init(&free_list, &first_block); 1435 1436 if (sip->i_d.di_nlink == 0) { 1437 error = xfs_iunlink_remove(tp, sip); 1438 if (error) 1439 goto error_return; 1440 } 1441 1442 error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino, 1443 &first_block, &free_list, resblks); 1444 if (error) 1445 goto error_return; 1446 xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 1447 xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE); 1448 1449 error = xfs_bumplink(tp, sip); 1450 if (error) 1451 goto error_return; 1452 1453 /* 1454 * If this is a synchronous mount, make sure that the 1455 * link transaction goes to disk before returning to 1456 * the user. 1457 */ 1458 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { 1459 xfs_trans_set_sync(tp); 1460 } 1461 1462 error = xfs_bmap_finish (&tp, &free_list, &committed); 1463 if (error) { 1464 xfs_bmap_cancel(&free_list); 1465 goto error_return; 1466 } 1467 1468 return xfs_trans_commit(tp); 1469 1470 error_return: 1471 xfs_trans_cancel(tp); 1472 std_return: 1473 return error; 1474 } 1475 1476 /* 1477 * Free up the underlying blocks past new_size. The new size must be smaller 1478 * than the current size. This routine can be used both for the attribute and 1479 * data fork, and does not modify the inode size, which is left to the caller. 1480 * 1481 * The transaction passed to this routine must have made a permanent log 1482 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1483 * given transaction and start new ones, so make sure everything involved in 1484 * the transaction is tidy before calling here. Some transaction will be 1485 * returned to the caller to be committed. The incoming transaction must 1486 * already include the inode, and both inode locks must be held exclusively. 1487 * The inode must also be "held" within the transaction. On return the inode 1488 * will be "held" within the returned transaction. This routine does NOT 1489 * require any disk space to be reserved for it within the transaction. 1490 * 1491 * If we get an error, we must return with the inode locked and linked into the 1492 * current transaction. This keeps things simple for the higher level code, 1493 * because it always knows that the inode is locked and held in the transaction 1494 * that returns to it whether errors occur or not. We don't mark the inode 1495 * dirty on error so that transactions can be easily aborted if possible. 1496 */ 1497 int 1498 xfs_itruncate_extents( 1499 struct xfs_trans **tpp, 1500 struct xfs_inode *ip, 1501 int whichfork, 1502 xfs_fsize_t new_size) 1503 { 1504 struct xfs_mount *mp = ip->i_mount; 1505 struct xfs_trans *tp = *tpp; 1506 xfs_bmap_free_t free_list; 1507 xfs_fsblock_t first_block; 1508 xfs_fileoff_t first_unmap_block; 1509 xfs_fileoff_t last_block; 1510 xfs_filblks_t unmap_len; 1511 int committed; 1512 int error = 0; 1513 int done = 0; 1514 1515 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1516 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1517 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1518 ASSERT(new_size <= XFS_ISIZE(ip)); 1519 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1520 ASSERT(ip->i_itemp != NULL); 1521 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1522 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1523 1524 trace_xfs_itruncate_extents_start(ip, new_size); 1525 1526 /* 1527 * Since it is possible for space to become allocated beyond 1528 * the end of the file (in a crash where the space is allocated 1529 * but the inode size is not yet updated), simply remove any 1530 * blocks which show up between the new EOF and the maximum 1531 * possible file size. If the first block to be removed is 1532 * beyond the maximum file size (ie it is the same as last_block), 1533 * then there is nothing to do. 1534 */ 1535 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1536 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1537 if (first_unmap_block == last_block) 1538 return 0; 1539 1540 ASSERT(first_unmap_block < last_block); 1541 unmap_len = last_block - first_unmap_block + 1; 1542 while (!done) { 1543 xfs_bmap_init(&free_list, &first_block); 1544 error = xfs_bunmapi(tp, ip, 1545 first_unmap_block, unmap_len, 1546 xfs_bmapi_aflag(whichfork), 1547 XFS_ITRUNC_MAX_EXTENTS, 1548 &first_block, &free_list, 1549 &done); 1550 if (error) 1551 goto out_bmap_cancel; 1552 1553 /* 1554 * Duplicate the transaction that has the permanent 1555 * reservation and commit the old transaction. 1556 */ 1557 error = xfs_bmap_finish(&tp, &free_list, &committed); 1558 if (committed) 1559 xfs_trans_ijoin(tp, ip, 0); 1560 if (error) 1561 goto out_bmap_cancel; 1562 1563 error = xfs_trans_roll(&tp, ip); 1564 if (error) 1565 goto out; 1566 } 1567 1568 /* 1569 * Always re-log the inode so that our permanent transaction can keep 1570 * on rolling it forward in the log. 1571 */ 1572 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1573 1574 trace_xfs_itruncate_extents_end(ip, new_size); 1575 1576 out: 1577 *tpp = tp; 1578 return error; 1579 out_bmap_cancel: 1580 /* 1581 * If the bunmapi call encounters an error, return to the caller where 1582 * the transaction can be properly aborted. We just need to make sure 1583 * we're not holding any resources that we were not when we came in. 1584 */ 1585 xfs_bmap_cancel(&free_list); 1586 goto out; 1587 } 1588 1589 int 1590 xfs_release( 1591 xfs_inode_t *ip) 1592 { 1593 xfs_mount_t *mp = ip->i_mount; 1594 int error; 1595 1596 if (!S_ISREG(ip->i_d.di_mode) || (ip->i_d.di_mode == 0)) 1597 return 0; 1598 1599 /* If this is a read-only mount, don't do this (would generate I/O) */ 1600 if (mp->m_flags & XFS_MOUNT_RDONLY) 1601 return 0; 1602 1603 if (!XFS_FORCED_SHUTDOWN(mp)) { 1604 int truncated; 1605 1606 /* 1607 * If we previously truncated this file and removed old data 1608 * in the process, we want to initiate "early" writeout on 1609 * the last close. This is an attempt to combat the notorious 1610 * NULL files problem which is particularly noticeable from a 1611 * truncate down, buffered (re-)write (delalloc), followed by 1612 * a crash. What we are effectively doing here is 1613 * significantly reducing the time window where we'd otherwise 1614 * be exposed to that problem. 1615 */ 1616 truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED); 1617 if (truncated) { 1618 xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE); 1619 if (ip->i_delayed_blks > 0) { 1620 error = filemap_flush(VFS_I(ip)->i_mapping); 1621 if (error) 1622 return error; 1623 } 1624 } 1625 } 1626 1627 if (ip->i_d.di_nlink == 0) 1628 return 0; 1629 1630 if (xfs_can_free_eofblocks(ip, false)) { 1631 1632 /* 1633 * If we can't get the iolock just skip truncating the blocks 1634 * past EOF because we could deadlock with the mmap_sem 1635 * otherwise. We'll get another chance to drop them once the 1636 * last reference to the inode is dropped, so we'll never leak 1637 * blocks permanently. 1638 * 1639 * Further, check if the inode is being opened, written and 1640 * closed frequently and we have delayed allocation blocks 1641 * outstanding (e.g. streaming writes from the NFS server), 1642 * truncating the blocks past EOF will cause fragmentation to 1643 * occur. 1644 * 1645 * In this case don't do the truncation, either, but we have to 1646 * be careful how we detect this case. Blocks beyond EOF show 1647 * up as i_delayed_blks even when the inode is clean, so we 1648 * need to truncate them away first before checking for a dirty 1649 * release. Hence on the first dirty close we will still remove 1650 * the speculative allocation, but after that we will leave it 1651 * in place. 1652 */ 1653 if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE)) 1654 return 0; 1655 1656 error = xfs_free_eofblocks(mp, ip, true); 1657 if (error && error != -EAGAIN) 1658 return error; 1659 1660 /* delalloc blocks after truncation means it really is dirty */ 1661 if (ip->i_delayed_blks) 1662 xfs_iflags_set(ip, XFS_IDIRTY_RELEASE); 1663 } 1664 return 0; 1665 } 1666 1667 /* 1668 * xfs_inactive_truncate 1669 * 1670 * Called to perform a truncate when an inode becomes unlinked. 1671 */ 1672 STATIC int 1673 xfs_inactive_truncate( 1674 struct xfs_inode *ip) 1675 { 1676 struct xfs_mount *mp = ip->i_mount; 1677 struct xfs_trans *tp; 1678 int error; 1679 1680 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); 1681 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); 1682 if (error) { 1683 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1684 xfs_trans_cancel(tp); 1685 return error; 1686 } 1687 1688 xfs_ilock(ip, XFS_ILOCK_EXCL); 1689 xfs_trans_ijoin(tp, ip, 0); 1690 1691 /* 1692 * Log the inode size first to prevent stale data exposure in the event 1693 * of a system crash before the truncate completes. See the related 1694 * comment in xfs_setattr_size() for details. 1695 */ 1696 ip->i_d.di_size = 0; 1697 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1698 1699 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0); 1700 if (error) 1701 goto error_trans_cancel; 1702 1703 ASSERT(ip->i_d.di_nextents == 0); 1704 1705 error = xfs_trans_commit(tp); 1706 if (error) 1707 goto error_unlock; 1708 1709 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1710 return 0; 1711 1712 error_trans_cancel: 1713 xfs_trans_cancel(tp); 1714 error_unlock: 1715 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1716 return error; 1717 } 1718 1719 /* 1720 * xfs_inactive_ifree() 1721 * 1722 * Perform the inode free when an inode is unlinked. 1723 */ 1724 STATIC int 1725 xfs_inactive_ifree( 1726 struct xfs_inode *ip) 1727 { 1728 xfs_bmap_free_t free_list; 1729 xfs_fsblock_t first_block; 1730 int committed; 1731 struct xfs_mount *mp = ip->i_mount; 1732 struct xfs_trans *tp; 1733 int error; 1734 1735 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE); 1736 1737 /* 1738 * The ifree transaction might need to allocate blocks for record 1739 * insertion to the finobt. We don't want to fail here at ENOSPC, so 1740 * allow ifree to dip into the reserved block pool if necessary. 1741 * 1742 * Freeing large sets of inodes generally means freeing inode chunks, 1743 * directory and file data blocks, so this should be relatively safe. 1744 * Only under severe circumstances should it be possible to free enough 1745 * inodes to exhaust the reserve block pool via finobt expansion while 1746 * at the same time not creating free space in the filesystem. 1747 * 1748 * Send a warning if the reservation does happen to fail, as the inode 1749 * now remains allocated and sits on the unlinked list until the fs is 1750 * repaired. 1751 */ 1752 tp->t_flags |= XFS_TRANS_RESERVE; 1753 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ifree, 1754 XFS_IFREE_SPACE_RES(mp), 0); 1755 if (error) { 1756 if (error == -ENOSPC) { 1757 xfs_warn_ratelimited(mp, 1758 "Failed to remove inode(s) from unlinked list. " 1759 "Please free space, unmount and run xfs_repair."); 1760 } else { 1761 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 1762 } 1763 xfs_trans_cancel(tp); 1764 return error; 1765 } 1766 1767 xfs_ilock(ip, XFS_ILOCK_EXCL); 1768 xfs_trans_ijoin(tp, ip, 0); 1769 1770 xfs_bmap_init(&free_list, &first_block); 1771 error = xfs_ifree(tp, ip, &free_list); 1772 if (error) { 1773 /* 1774 * If we fail to free the inode, shut down. The cancel 1775 * might do that, we need to make sure. Otherwise the 1776 * inode might be lost for a long time or forever. 1777 */ 1778 if (!XFS_FORCED_SHUTDOWN(mp)) { 1779 xfs_notice(mp, "%s: xfs_ifree returned error %d", 1780 __func__, error); 1781 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); 1782 } 1783 xfs_trans_cancel(tp); 1784 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1785 return error; 1786 } 1787 1788 /* 1789 * Credit the quota account(s). The inode is gone. 1790 */ 1791 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1); 1792 1793 /* 1794 * Just ignore errors at this point. There is nothing we can 1795 * do except to try to keep going. Make sure it's not a silent 1796 * error. 1797 */ 1798 error = xfs_bmap_finish(&tp, &free_list, &committed); 1799 if (error) 1800 xfs_notice(mp, "%s: xfs_bmap_finish returned error %d", 1801 __func__, error); 1802 error = xfs_trans_commit(tp); 1803 if (error) 1804 xfs_notice(mp, "%s: xfs_trans_commit returned error %d", 1805 __func__, error); 1806 1807 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1808 return 0; 1809 } 1810 1811 /* 1812 * xfs_inactive 1813 * 1814 * This is called when the vnode reference count for the vnode 1815 * goes to zero. If the file has been unlinked, then it must 1816 * now be truncated. Also, we clear all of the read-ahead state 1817 * kept for the inode here since the file is now closed. 1818 */ 1819 void 1820 xfs_inactive( 1821 xfs_inode_t *ip) 1822 { 1823 struct xfs_mount *mp; 1824 int error; 1825 int truncate = 0; 1826 1827 /* 1828 * If the inode is already free, then there can be nothing 1829 * to clean up here. 1830 */ 1831 if (ip->i_d.di_mode == 0) { 1832 ASSERT(ip->i_df.if_real_bytes == 0); 1833 ASSERT(ip->i_df.if_broot_bytes == 0); 1834 return; 1835 } 1836 1837 mp = ip->i_mount; 1838 1839 /* If this is a read-only mount, don't do this (would generate I/O) */ 1840 if (mp->m_flags & XFS_MOUNT_RDONLY) 1841 return; 1842 1843 if (ip->i_d.di_nlink != 0) { 1844 /* 1845 * force is true because we are evicting an inode from the 1846 * cache. Post-eof blocks must be freed, lest we end up with 1847 * broken free space accounting. 1848 */ 1849 if (xfs_can_free_eofblocks(ip, true)) 1850 xfs_free_eofblocks(mp, ip, false); 1851 1852 return; 1853 } 1854 1855 if (S_ISREG(ip->i_d.di_mode) && 1856 (ip->i_d.di_size != 0 || XFS_ISIZE(ip) != 0 || 1857 ip->i_d.di_nextents > 0 || ip->i_delayed_blks > 0)) 1858 truncate = 1; 1859 1860 error = xfs_qm_dqattach(ip, 0); 1861 if (error) 1862 return; 1863 1864 if (S_ISLNK(ip->i_d.di_mode)) 1865 error = xfs_inactive_symlink(ip); 1866 else if (truncate) 1867 error = xfs_inactive_truncate(ip); 1868 if (error) 1869 return; 1870 1871 /* 1872 * If there are attributes associated with the file then blow them away 1873 * now. The code calls a routine that recursively deconstructs the 1874 * attribute fork. If also blows away the in-core attribute fork. 1875 */ 1876 if (XFS_IFORK_Q(ip)) { 1877 error = xfs_attr_inactive(ip); 1878 if (error) 1879 return; 1880 } 1881 1882 ASSERT(!ip->i_afp); 1883 ASSERT(ip->i_d.di_anextents == 0); 1884 ASSERT(ip->i_d.di_forkoff == 0); 1885 1886 /* 1887 * Free the inode. 1888 */ 1889 error = xfs_inactive_ifree(ip); 1890 if (error) 1891 return; 1892 1893 /* 1894 * Release the dquots held by inode, if any. 1895 */ 1896 xfs_qm_dqdetach(ip); 1897 } 1898 1899 /* 1900 * This is called when the inode's link count goes to 0. 1901 * We place the on-disk inode on a list in the AGI. It 1902 * will be pulled from this list when the inode is freed. 1903 */ 1904 int 1905 xfs_iunlink( 1906 xfs_trans_t *tp, 1907 xfs_inode_t *ip) 1908 { 1909 xfs_mount_t *mp; 1910 xfs_agi_t *agi; 1911 xfs_dinode_t *dip; 1912 xfs_buf_t *agibp; 1913 xfs_buf_t *ibp; 1914 xfs_agino_t agino; 1915 short bucket_index; 1916 int offset; 1917 int error; 1918 1919 ASSERT(ip->i_d.di_nlink == 0); 1920 ASSERT(ip->i_d.di_mode != 0); 1921 1922 mp = tp->t_mountp; 1923 1924 /* 1925 * Get the agi buffer first. It ensures lock ordering 1926 * on the list. 1927 */ 1928 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1929 if (error) 1930 return error; 1931 agi = XFS_BUF_TO_AGI(agibp); 1932 1933 /* 1934 * Get the index into the agi hash table for the 1935 * list this inode will go on. 1936 */ 1937 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1938 ASSERT(agino != 0); 1939 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1940 ASSERT(agi->agi_unlinked[bucket_index]); 1941 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1942 1943 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1944 /* 1945 * There is already another inode in the bucket we need 1946 * to add ourselves to. Add us at the front of the list. 1947 * Here we put the head pointer into our next pointer, 1948 * and then we fall through to point the head at us. 1949 */ 1950 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1951 0, 0); 1952 if (error) 1953 return error; 1954 1955 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1956 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1957 offset = ip->i_imap.im_boffset + 1958 offsetof(xfs_dinode_t, di_next_unlinked); 1959 1960 /* need to recalc the inode CRC if appropriate */ 1961 xfs_dinode_calc_crc(mp, dip); 1962 1963 xfs_trans_inode_buf(tp, ibp); 1964 xfs_trans_log_buf(tp, ibp, offset, 1965 (offset + sizeof(xfs_agino_t) - 1)); 1966 xfs_inobp_check(mp, ibp); 1967 } 1968 1969 /* 1970 * Point the bucket head pointer at the inode being inserted. 1971 */ 1972 ASSERT(agino != 0); 1973 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1974 offset = offsetof(xfs_agi_t, agi_unlinked) + 1975 (sizeof(xfs_agino_t) * bucket_index); 1976 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 1977 xfs_trans_log_buf(tp, agibp, offset, 1978 (offset + sizeof(xfs_agino_t) - 1)); 1979 return 0; 1980 } 1981 1982 /* 1983 * Pull the on-disk inode from the AGI unlinked list. 1984 */ 1985 STATIC int 1986 xfs_iunlink_remove( 1987 xfs_trans_t *tp, 1988 xfs_inode_t *ip) 1989 { 1990 xfs_ino_t next_ino; 1991 xfs_mount_t *mp; 1992 xfs_agi_t *agi; 1993 xfs_dinode_t *dip; 1994 xfs_buf_t *agibp; 1995 xfs_buf_t *ibp; 1996 xfs_agnumber_t agno; 1997 xfs_agino_t agino; 1998 xfs_agino_t next_agino; 1999 xfs_buf_t *last_ibp; 2000 xfs_dinode_t *last_dip = NULL; 2001 short bucket_index; 2002 int offset, last_offset = 0; 2003 int error; 2004 2005 mp = tp->t_mountp; 2006 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 2007 2008 /* 2009 * Get the agi buffer first. It ensures lock ordering 2010 * on the list. 2011 */ 2012 error = xfs_read_agi(mp, tp, agno, &agibp); 2013 if (error) 2014 return error; 2015 2016 agi = XFS_BUF_TO_AGI(agibp); 2017 2018 /* 2019 * Get the index into the agi hash table for the 2020 * list this inode will go on. 2021 */ 2022 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 2023 ASSERT(agino != 0); 2024 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 2025 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 2026 ASSERT(agi->agi_unlinked[bucket_index]); 2027 2028 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 2029 /* 2030 * We're at the head of the list. Get the inode's on-disk 2031 * buffer to see if there is anyone after us on the list. 2032 * Only modify our next pointer if it is not already NULLAGINO. 2033 * This saves us the overhead of dealing with the buffer when 2034 * there is no need to change it. 2035 */ 2036 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2037 0, 0); 2038 if (error) { 2039 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 2040 __func__, error); 2041 return error; 2042 } 2043 next_agino = be32_to_cpu(dip->di_next_unlinked); 2044 ASSERT(next_agino != 0); 2045 if (next_agino != NULLAGINO) { 2046 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2047 offset = ip->i_imap.im_boffset + 2048 offsetof(xfs_dinode_t, di_next_unlinked); 2049 2050 /* need to recalc the inode CRC if appropriate */ 2051 xfs_dinode_calc_crc(mp, dip); 2052 2053 xfs_trans_inode_buf(tp, ibp); 2054 xfs_trans_log_buf(tp, ibp, offset, 2055 (offset + sizeof(xfs_agino_t) - 1)); 2056 xfs_inobp_check(mp, ibp); 2057 } else { 2058 xfs_trans_brelse(tp, ibp); 2059 } 2060 /* 2061 * Point the bucket head pointer at the next inode. 2062 */ 2063 ASSERT(next_agino != 0); 2064 ASSERT(next_agino != agino); 2065 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 2066 offset = offsetof(xfs_agi_t, agi_unlinked) + 2067 (sizeof(xfs_agino_t) * bucket_index); 2068 xfs_trans_buf_set_type(tp, agibp, XFS_BLFT_AGI_BUF); 2069 xfs_trans_log_buf(tp, agibp, offset, 2070 (offset + sizeof(xfs_agino_t) - 1)); 2071 } else { 2072 /* 2073 * We need to search the list for the inode being freed. 2074 */ 2075 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 2076 last_ibp = NULL; 2077 while (next_agino != agino) { 2078 struct xfs_imap imap; 2079 2080 if (last_ibp) 2081 xfs_trans_brelse(tp, last_ibp); 2082 2083 imap.im_blkno = 0; 2084 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 2085 2086 error = xfs_imap(mp, tp, next_ino, &imap, 0); 2087 if (error) { 2088 xfs_warn(mp, 2089 "%s: xfs_imap returned error %d.", 2090 __func__, error); 2091 return error; 2092 } 2093 2094 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 2095 &last_ibp, 0, 0); 2096 if (error) { 2097 xfs_warn(mp, 2098 "%s: xfs_imap_to_bp returned error %d.", 2099 __func__, error); 2100 return error; 2101 } 2102 2103 last_offset = imap.im_boffset; 2104 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 2105 ASSERT(next_agino != NULLAGINO); 2106 ASSERT(next_agino != 0); 2107 } 2108 2109 /* 2110 * Now last_ibp points to the buffer previous to us on the 2111 * unlinked list. Pull us from the list. 2112 */ 2113 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 2114 0, 0); 2115 if (error) { 2116 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 2117 __func__, error); 2118 return error; 2119 } 2120 next_agino = be32_to_cpu(dip->di_next_unlinked); 2121 ASSERT(next_agino != 0); 2122 ASSERT(next_agino != agino); 2123 if (next_agino != NULLAGINO) { 2124 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 2125 offset = ip->i_imap.im_boffset + 2126 offsetof(xfs_dinode_t, di_next_unlinked); 2127 2128 /* need to recalc the inode CRC if appropriate */ 2129 xfs_dinode_calc_crc(mp, dip); 2130 2131 xfs_trans_inode_buf(tp, ibp); 2132 xfs_trans_log_buf(tp, ibp, offset, 2133 (offset + sizeof(xfs_agino_t) - 1)); 2134 xfs_inobp_check(mp, ibp); 2135 } else { 2136 xfs_trans_brelse(tp, ibp); 2137 } 2138 /* 2139 * Point the previous inode on the list to the next inode. 2140 */ 2141 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 2142 ASSERT(next_agino != 0); 2143 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 2144 2145 /* need to recalc the inode CRC if appropriate */ 2146 xfs_dinode_calc_crc(mp, last_dip); 2147 2148 xfs_trans_inode_buf(tp, last_ibp); 2149 xfs_trans_log_buf(tp, last_ibp, offset, 2150 (offset + sizeof(xfs_agino_t) - 1)); 2151 xfs_inobp_check(mp, last_ibp); 2152 } 2153 return 0; 2154 } 2155 2156 /* 2157 * A big issue when freeing the inode cluster is that we _cannot_ skip any 2158 * inodes that are in memory - they all must be marked stale and attached to 2159 * the cluster buffer. 2160 */ 2161 STATIC int 2162 xfs_ifree_cluster( 2163 xfs_inode_t *free_ip, 2164 xfs_trans_t *tp, 2165 struct xfs_icluster *xic) 2166 { 2167 xfs_mount_t *mp = free_ip->i_mount; 2168 int blks_per_cluster; 2169 int inodes_per_cluster; 2170 int nbufs; 2171 int i, j; 2172 int ioffset; 2173 xfs_daddr_t blkno; 2174 xfs_buf_t *bp; 2175 xfs_inode_t *ip; 2176 xfs_inode_log_item_t *iip; 2177 xfs_log_item_t *lip; 2178 struct xfs_perag *pag; 2179 xfs_ino_t inum; 2180 2181 inum = xic->first_ino; 2182 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 2183 blks_per_cluster = xfs_icluster_size_fsb(mp); 2184 inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog; 2185 nbufs = mp->m_ialloc_blks / blks_per_cluster; 2186 2187 for (j = 0; j < nbufs; j++, inum += inodes_per_cluster) { 2188 /* 2189 * The allocation bitmap tells us which inodes of the chunk were 2190 * physically allocated. Skip the cluster if an inode falls into 2191 * a sparse region. 2192 */ 2193 ioffset = inum - xic->first_ino; 2194 if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) { 2195 ASSERT(do_mod(ioffset, inodes_per_cluster) == 0); 2196 continue; 2197 } 2198 2199 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 2200 XFS_INO_TO_AGBNO(mp, inum)); 2201 2202 /* 2203 * We obtain and lock the backing buffer first in the process 2204 * here, as we have to ensure that any dirty inode that we 2205 * can't get the flush lock on is attached to the buffer. 2206 * If we scan the in-memory inodes first, then buffer IO can 2207 * complete before we get a lock on it, and hence we may fail 2208 * to mark all the active inodes on the buffer stale. 2209 */ 2210 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 2211 mp->m_bsize * blks_per_cluster, 2212 XBF_UNMAPPED); 2213 2214 if (!bp) 2215 return -ENOMEM; 2216 2217 /* 2218 * This buffer may not have been correctly initialised as we 2219 * didn't read it from disk. That's not important because we are 2220 * only using to mark the buffer as stale in the log, and to 2221 * attach stale cached inodes on it. That means it will never be 2222 * dispatched for IO. If it is, we want to know about it, and we 2223 * want it to fail. We can acheive this by adding a write 2224 * verifier to the buffer. 2225 */ 2226 bp->b_ops = &xfs_inode_buf_ops; 2227 2228 /* 2229 * Walk the inodes already attached to the buffer and mark them 2230 * stale. These will all have the flush locks held, so an 2231 * in-memory inode walk can't lock them. By marking them all 2232 * stale first, we will not attempt to lock them in the loop 2233 * below as the XFS_ISTALE flag will be set. 2234 */ 2235 lip = bp->b_fspriv; 2236 while (lip) { 2237 if (lip->li_type == XFS_LI_INODE) { 2238 iip = (xfs_inode_log_item_t *)lip; 2239 ASSERT(iip->ili_logged == 1); 2240 lip->li_cb = xfs_istale_done; 2241 xfs_trans_ail_copy_lsn(mp->m_ail, 2242 &iip->ili_flush_lsn, 2243 &iip->ili_item.li_lsn); 2244 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 2245 } 2246 lip = lip->li_bio_list; 2247 } 2248 2249 2250 /* 2251 * For each inode in memory attempt to add it to the inode 2252 * buffer and set it up for being staled on buffer IO 2253 * completion. This is safe as we've locked out tail pushing 2254 * and flushing by locking the buffer. 2255 * 2256 * We have already marked every inode that was part of a 2257 * transaction stale above, which means there is no point in 2258 * even trying to lock them. 2259 */ 2260 for (i = 0; i < inodes_per_cluster; i++) { 2261 retry: 2262 rcu_read_lock(); 2263 ip = radix_tree_lookup(&pag->pag_ici_root, 2264 XFS_INO_TO_AGINO(mp, (inum + i))); 2265 2266 /* Inode not in memory, nothing to do */ 2267 if (!ip) { 2268 rcu_read_unlock(); 2269 continue; 2270 } 2271 2272 /* 2273 * because this is an RCU protected lookup, we could 2274 * find a recently freed or even reallocated inode 2275 * during the lookup. We need to check under the 2276 * i_flags_lock for a valid inode here. Skip it if it 2277 * is not valid, the wrong inode or stale. 2278 */ 2279 spin_lock(&ip->i_flags_lock); 2280 if (ip->i_ino != inum + i || 2281 __xfs_iflags_test(ip, XFS_ISTALE)) { 2282 spin_unlock(&ip->i_flags_lock); 2283 rcu_read_unlock(); 2284 continue; 2285 } 2286 spin_unlock(&ip->i_flags_lock); 2287 2288 /* 2289 * Don't try to lock/unlock the current inode, but we 2290 * _cannot_ skip the other inodes that we did not find 2291 * in the list attached to the buffer and are not 2292 * already marked stale. If we can't lock it, back off 2293 * and retry. 2294 */ 2295 if (ip != free_ip && 2296 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 2297 rcu_read_unlock(); 2298 delay(1); 2299 goto retry; 2300 } 2301 rcu_read_unlock(); 2302 2303 xfs_iflock(ip); 2304 xfs_iflags_set(ip, XFS_ISTALE); 2305 2306 /* 2307 * we don't need to attach clean inodes or those only 2308 * with unlogged changes (which we throw away, anyway). 2309 */ 2310 iip = ip->i_itemp; 2311 if (!iip || xfs_inode_clean(ip)) { 2312 ASSERT(ip != free_ip); 2313 xfs_ifunlock(ip); 2314 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2315 continue; 2316 } 2317 2318 iip->ili_last_fields = iip->ili_fields; 2319 iip->ili_fields = 0; 2320 iip->ili_logged = 1; 2321 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2322 &iip->ili_item.li_lsn); 2323 2324 xfs_buf_attach_iodone(bp, xfs_istale_done, 2325 &iip->ili_item); 2326 2327 if (ip != free_ip) 2328 xfs_iunlock(ip, XFS_ILOCK_EXCL); 2329 } 2330 2331 xfs_trans_stale_inode_buf(tp, bp); 2332 xfs_trans_binval(tp, bp); 2333 } 2334 2335 xfs_perag_put(pag); 2336 return 0; 2337 } 2338 2339 /* 2340 * This is called to return an inode to the inode free list. 2341 * The inode should already be truncated to 0 length and have 2342 * no pages associated with it. This routine also assumes that 2343 * the inode is already a part of the transaction. 2344 * 2345 * The on-disk copy of the inode will have been added to the list 2346 * of unlinked inodes in the AGI. We need to remove the inode from 2347 * that list atomically with respect to freeing it here. 2348 */ 2349 int 2350 xfs_ifree( 2351 xfs_trans_t *tp, 2352 xfs_inode_t *ip, 2353 xfs_bmap_free_t *flist) 2354 { 2355 int error; 2356 struct xfs_icluster xic = { 0 }; 2357 2358 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 2359 ASSERT(ip->i_d.di_nlink == 0); 2360 ASSERT(ip->i_d.di_nextents == 0); 2361 ASSERT(ip->i_d.di_anextents == 0); 2362 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); 2363 ASSERT(ip->i_d.di_nblocks == 0); 2364 2365 /* 2366 * Pull the on-disk inode from the AGI unlinked list. 2367 */ 2368 error = xfs_iunlink_remove(tp, ip); 2369 if (error) 2370 return error; 2371 2372 error = xfs_difree(tp, ip->i_ino, flist, &xic); 2373 if (error) 2374 return error; 2375 2376 ip->i_d.di_mode = 0; /* mark incore inode as free */ 2377 ip->i_d.di_flags = 0; 2378 ip->i_d.di_dmevmask = 0; 2379 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 2380 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 2381 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 2382 /* 2383 * Bump the generation count so no one will be confused 2384 * by reincarnations of this inode. 2385 */ 2386 ip->i_d.di_gen++; 2387 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 2388 2389 if (xic.deleted) 2390 error = xfs_ifree_cluster(ip, tp, &xic); 2391 2392 return error; 2393 } 2394 2395 /* 2396 * This is called to unpin an inode. The caller must have the inode locked 2397 * in at least shared mode so that the buffer cannot be subsequently pinned 2398 * once someone is waiting for it to be unpinned. 2399 */ 2400 static void 2401 xfs_iunpin( 2402 struct xfs_inode *ip) 2403 { 2404 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2405 2406 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2407 2408 /* Give the log a push to start the unpinning I/O */ 2409 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2410 2411 } 2412 2413 static void 2414 __xfs_iunpin_wait( 2415 struct xfs_inode *ip) 2416 { 2417 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2418 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2419 2420 xfs_iunpin(ip); 2421 2422 do { 2423 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2424 if (xfs_ipincount(ip)) 2425 io_schedule(); 2426 } while (xfs_ipincount(ip)); 2427 finish_wait(wq, &wait.wait); 2428 } 2429 2430 void 2431 xfs_iunpin_wait( 2432 struct xfs_inode *ip) 2433 { 2434 if (xfs_ipincount(ip)) 2435 __xfs_iunpin_wait(ip); 2436 } 2437 2438 /* 2439 * Removing an inode from the namespace involves removing the directory entry 2440 * and dropping the link count on the inode. Removing the directory entry can 2441 * result in locking an AGF (directory blocks were freed) and removing a link 2442 * count can result in placing the inode on an unlinked list which results in 2443 * locking an AGI. 2444 * 2445 * The big problem here is that we have an ordering constraint on AGF and AGI 2446 * locking - inode allocation locks the AGI, then can allocate a new extent for 2447 * new inodes, locking the AGF after the AGI. Similarly, freeing the inode 2448 * removes the inode from the unlinked list, requiring that we lock the AGI 2449 * first, and then freeing the inode can result in an inode chunk being freed 2450 * and hence freeing disk space requiring that we lock an AGF. 2451 * 2452 * Hence the ordering that is imposed by other parts of the code is AGI before 2453 * AGF. This means we cannot remove the directory entry before we drop the inode 2454 * reference count and put it on the unlinked list as this results in a lock 2455 * order of AGF then AGI, and this can deadlock against inode allocation and 2456 * freeing. Therefore we must drop the link counts before we remove the 2457 * directory entry. 2458 * 2459 * This is still safe from a transactional point of view - it is not until we 2460 * get to xfs_bmap_finish() that we have the possibility of multiple 2461 * transactions in this operation. Hence as long as we remove the directory 2462 * entry and drop the link count in the first transaction of the remove 2463 * operation, there are no transactional constraints on the ordering here. 2464 */ 2465 int 2466 xfs_remove( 2467 xfs_inode_t *dp, 2468 struct xfs_name *name, 2469 xfs_inode_t *ip) 2470 { 2471 xfs_mount_t *mp = dp->i_mount; 2472 xfs_trans_t *tp = NULL; 2473 int is_dir = S_ISDIR(ip->i_d.di_mode); 2474 int error = 0; 2475 xfs_bmap_free_t free_list; 2476 xfs_fsblock_t first_block; 2477 int committed; 2478 uint resblks; 2479 2480 trace_xfs_remove(dp, name); 2481 2482 if (XFS_FORCED_SHUTDOWN(mp)) 2483 return -EIO; 2484 2485 error = xfs_qm_dqattach(dp, 0); 2486 if (error) 2487 goto std_return; 2488 2489 error = xfs_qm_dqattach(ip, 0); 2490 if (error) 2491 goto std_return; 2492 2493 if (is_dir) 2494 tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR); 2495 else 2496 tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE); 2497 2498 /* 2499 * We try to get the real space reservation first, 2500 * allowing for directory btree deletion(s) implying 2501 * possible bmap insert(s). If we can't get the space 2502 * reservation then we use 0 instead, and avoid the bmap 2503 * btree insert(s) in the directory code by, if the bmap 2504 * insert tries to happen, instead trimming the LAST 2505 * block from the directory. 2506 */ 2507 resblks = XFS_REMOVE_SPACE_RES(mp); 2508 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, resblks, 0); 2509 if (error == -ENOSPC) { 2510 resblks = 0; 2511 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_remove, 0, 0); 2512 } 2513 if (error) { 2514 ASSERT(error != -ENOSPC); 2515 goto out_trans_cancel; 2516 } 2517 2518 xfs_lock_two_inodes(dp, ip, XFS_ILOCK_EXCL); 2519 2520 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 2521 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 2522 2523 /* 2524 * If we're removing a directory perform some additional validation. 2525 */ 2526 if (is_dir) { 2527 ASSERT(ip->i_d.di_nlink >= 2); 2528 if (ip->i_d.di_nlink != 2) { 2529 error = -ENOTEMPTY; 2530 goto out_trans_cancel; 2531 } 2532 if (!xfs_dir_isempty(ip)) { 2533 error = -ENOTEMPTY; 2534 goto out_trans_cancel; 2535 } 2536 2537 /* Drop the link from ip's "..". */ 2538 error = xfs_droplink(tp, dp); 2539 if (error) 2540 goto out_trans_cancel; 2541 2542 /* Drop the "." link from ip to self. */ 2543 error = xfs_droplink(tp, ip); 2544 if (error) 2545 goto out_trans_cancel; 2546 } else { 2547 /* 2548 * When removing a non-directory we need to log the parent 2549 * inode here. For a directory this is done implicitly 2550 * by the xfs_droplink call for the ".." entry. 2551 */ 2552 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 2553 } 2554 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2555 2556 /* Drop the link from dp to ip. */ 2557 error = xfs_droplink(tp, ip); 2558 if (error) 2559 goto out_trans_cancel; 2560 2561 xfs_bmap_init(&free_list, &first_block); 2562 error = xfs_dir_removename(tp, dp, name, ip->i_ino, 2563 &first_block, &free_list, resblks); 2564 if (error) { 2565 ASSERT(error != -ENOENT); 2566 goto out_bmap_cancel; 2567 } 2568 2569 /* 2570 * If this is a synchronous mount, make sure that the 2571 * remove transaction goes to disk before returning to 2572 * the user. 2573 */ 2574 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2575 xfs_trans_set_sync(tp); 2576 2577 error = xfs_bmap_finish(&tp, &free_list, &committed); 2578 if (error) 2579 goto out_bmap_cancel; 2580 2581 error = xfs_trans_commit(tp); 2582 if (error) 2583 goto std_return; 2584 2585 if (is_dir && xfs_inode_is_filestream(ip)) 2586 xfs_filestream_deassociate(ip); 2587 2588 return 0; 2589 2590 out_bmap_cancel: 2591 xfs_bmap_cancel(&free_list); 2592 out_trans_cancel: 2593 xfs_trans_cancel(tp); 2594 std_return: 2595 return error; 2596 } 2597 2598 /* 2599 * Enter all inodes for a rename transaction into a sorted array. 2600 */ 2601 #define __XFS_SORT_INODES 5 2602 STATIC void 2603 xfs_sort_for_rename( 2604 struct xfs_inode *dp1, /* in: old (source) directory inode */ 2605 struct xfs_inode *dp2, /* in: new (target) directory inode */ 2606 struct xfs_inode *ip1, /* in: inode of old entry */ 2607 struct xfs_inode *ip2, /* in: inode of new entry */ 2608 struct xfs_inode *wip, /* in: whiteout inode */ 2609 struct xfs_inode **i_tab,/* out: sorted array of inodes */ 2610 int *num_inodes) /* in/out: inodes in array */ 2611 { 2612 int i, j; 2613 2614 ASSERT(*num_inodes == __XFS_SORT_INODES); 2615 memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *)); 2616 2617 /* 2618 * i_tab contains a list of pointers to inodes. We initialize 2619 * the table here & we'll sort it. We will then use it to 2620 * order the acquisition of the inode locks. 2621 * 2622 * Note that the table may contain duplicates. e.g., dp1 == dp2. 2623 */ 2624 i = 0; 2625 i_tab[i++] = dp1; 2626 i_tab[i++] = dp2; 2627 i_tab[i++] = ip1; 2628 if (ip2) 2629 i_tab[i++] = ip2; 2630 if (wip) 2631 i_tab[i++] = wip; 2632 *num_inodes = i; 2633 2634 /* 2635 * Sort the elements via bubble sort. (Remember, there are at 2636 * most 5 elements to sort, so this is adequate.) 2637 */ 2638 for (i = 0; i < *num_inodes; i++) { 2639 for (j = 1; j < *num_inodes; j++) { 2640 if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) { 2641 struct xfs_inode *temp = i_tab[j]; 2642 i_tab[j] = i_tab[j-1]; 2643 i_tab[j-1] = temp; 2644 } 2645 } 2646 } 2647 } 2648 2649 static int 2650 xfs_finish_rename( 2651 struct xfs_trans *tp, 2652 struct xfs_bmap_free *free_list) 2653 { 2654 int committed = 0; 2655 int error; 2656 2657 /* 2658 * If this is a synchronous mount, make sure that the rename transaction 2659 * goes to disk before returning to the user. 2660 */ 2661 if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) 2662 xfs_trans_set_sync(tp); 2663 2664 error = xfs_bmap_finish(&tp, free_list, &committed); 2665 if (error) { 2666 xfs_bmap_cancel(free_list); 2667 xfs_trans_cancel(tp); 2668 return error; 2669 } 2670 2671 return xfs_trans_commit(tp); 2672 } 2673 2674 /* 2675 * xfs_cross_rename() 2676 * 2677 * responsible for handling RENAME_EXCHANGE flag in renameat2() sytemcall 2678 */ 2679 STATIC int 2680 xfs_cross_rename( 2681 struct xfs_trans *tp, 2682 struct xfs_inode *dp1, 2683 struct xfs_name *name1, 2684 struct xfs_inode *ip1, 2685 struct xfs_inode *dp2, 2686 struct xfs_name *name2, 2687 struct xfs_inode *ip2, 2688 struct xfs_bmap_free *free_list, 2689 xfs_fsblock_t *first_block, 2690 int spaceres) 2691 { 2692 int error = 0; 2693 int ip1_flags = 0; 2694 int ip2_flags = 0; 2695 int dp2_flags = 0; 2696 2697 /* Swap inode number for dirent in first parent */ 2698 error = xfs_dir_replace(tp, dp1, name1, 2699 ip2->i_ino, 2700 first_block, free_list, spaceres); 2701 if (error) 2702 goto out_trans_abort; 2703 2704 /* Swap inode number for dirent in second parent */ 2705 error = xfs_dir_replace(tp, dp2, name2, 2706 ip1->i_ino, 2707 first_block, free_list, spaceres); 2708 if (error) 2709 goto out_trans_abort; 2710 2711 /* 2712 * If we're renaming one or more directories across different parents, 2713 * update the respective ".." entries (and link counts) to match the new 2714 * parents. 2715 */ 2716 if (dp1 != dp2) { 2717 dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2718 2719 if (S_ISDIR(ip2->i_d.di_mode)) { 2720 error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot, 2721 dp1->i_ino, first_block, 2722 free_list, spaceres); 2723 if (error) 2724 goto out_trans_abort; 2725 2726 /* transfer ip2 ".." reference to dp1 */ 2727 if (!S_ISDIR(ip1->i_d.di_mode)) { 2728 error = xfs_droplink(tp, dp2); 2729 if (error) 2730 goto out_trans_abort; 2731 error = xfs_bumplink(tp, dp1); 2732 if (error) 2733 goto out_trans_abort; 2734 } 2735 2736 /* 2737 * Although ip1 isn't changed here, userspace needs 2738 * to be warned about the change, so that applications 2739 * relying on it (like backup ones), will properly 2740 * notify the change 2741 */ 2742 ip1_flags |= XFS_ICHGTIME_CHG; 2743 ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2744 } 2745 2746 if (S_ISDIR(ip1->i_d.di_mode)) { 2747 error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot, 2748 dp2->i_ino, first_block, 2749 free_list, spaceres); 2750 if (error) 2751 goto out_trans_abort; 2752 2753 /* transfer ip1 ".." reference to dp2 */ 2754 if (!S_ISDIR(ip2->i_d.di_mode)) { 2755 error = xfs_droplink(tp, dp1); 2756 if (error) 2757 goto out_trans_abort; 2758 error = xfs_bumplink(tp, dp2); 2759 if (error) 2760 goto out_trans_abort; 2761 } 2762 2763 /* 2764 * Although ip2 isn't changed here, userspace needs 2765 * to be warned about the change, so that applications 2766 * relying on it (like backup ones), will properly 2767 * notify the change 2768 */ 2769 ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG; 2770 ip2_flags |= XFS_ICHGTIME_CHG; 2771 } 2772 } 2773 2774 if (ip1_flags) { 2775 xfs_trans_ichgtime(tp, ip1, ip1_flags); 2776 xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE); 2777 } 2778 if (ip2_flags) { 2779 xfs_trans_ichgtime(tp, ip2, ip2_flags); 2780 xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE); 2781 } 2782 if (dp2_flags) { 2783 xfs_trans_ichgtime(tp, dp2, dp2_flags); 2784 xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE); 2785 } 2786 xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2787 xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE); 2788 return xfs_finish_rename(tp, free_list); 2789 2790 out_trans_abort: 2791 xfs_bmap_cancel(free_list); 2792 xfs_trans_cancel(tp); 2793 return error; 2794 } 2795 2796 /* 2797 * xfs_rename_alloc_whiteout() 2798 * 2799 * Return a referenced, unlinked, unlocked inode that that can be used as a 2800 * whiteout in a rename transaction. We use a tmpfile inode here so that if we 2801 * crash between allocating the inode and linking it into the rename transaction 2802 * recovery will free the inode and we won't leak it. 2803 */ 2804 static int 2805 xfs_rename_alloc_whiteout( 2806 struct xfs_inode *dp, 2807 struct xfs_inode **wip) 2808 { 2809 struct xfs_inode *tmpfile; 2810 int error; 2811 2812 error = xfs_create_tmpfile(dp, NULL, S_IFCHR | WHITEOUT_MODE, &tmpfile); 2813 if (error) 2814 return error; 2815 2816 /* 2817 * Prepare the tmpfile inode as if it were created through the VFS. 2818 * Otherwise, the link increment paths will complain about nlink 0->1. 2819 * Drop the link count as done by d_tmpfile(), complete the inode setup 2820 * and flag it as linkable. 2821 */ 2822 drop_nlink(VFS_I(tmpfile)); 2823 xfs_finish_inode_setup(tmpfile); 2824 VFS_I(tmpfile)->i_state |= I_LINKABLE; 2825 2826 *wip = tmpfile; 2827 return 0; 2828 } 2829 2830 /* 2831 * xfs_rename 2832 */ 2833 int 2834 xfs_rename( 2835 struct xfs_inode *src_dp, 2836 struct xfs_name *src_name, 2837 struct xfs_inode *src_ip, 2838 struct xfs_inode *target_dp, 2839 struct xfs_name *target_name, 2840 struct xfs_inode *target_ip, 2841 unsigned int flags) 2842 { 2843 struct xfs_mount *mp = src_dp->i_mount; 2844 struct xfs_trans *tp; 2845 struct xfs_bmap_free free_list; 2846 xfs_fsblock_t first_block; 2847 struct xfs_inode *wip = NULL; /* whiteout inode */ 2848 struct xfs_inode *inodes[__XFS_SORT_INODES]; 2849 int num_inodes = __XFS_SORT_INODES; 2850 bool new_parent = (src_dp != target_dp); 2851 bool src_is_directory = S_ISDIR(src_ip->i_d.di_mode); 2852 int spaceres; 2853 int error; 2854 2855 trace_xfs_rename(src_dp, target_dp, src_name, target_name); 2856 2857 if ((flags & RENAME_EXCHANGE) && !target_ip) 2858 return -EINVAL; 2859 2860 /* 2861 * If we are doing a whiteout operation, allocate the whiteout inode 2862 * we will be placing at the target and ensure the type is set 2863 * appropriately. 2864 */ 2865 if (flags & RENAME_WHITEOUT) { 2866 ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE))); 2867 error = xfs_rename_alloc_whiteout(target_dp, &wip); 2868 if (error) 2869 return error; 2870 2871 /* setup target dirent info as whiteout */ 2872 src_name->type = XFS_DIR3_FT_CHRDEV; 2873 } 2874 2875 xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip, 2876 inodes, &num_inodes); 2877 2878 tp = xfs_trans_alloc(mp, XFS_TRANS_RENAME); 2879 spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len); 2880 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, spaceres, 0); 2881 if (error == -ENOSPC) { 2882 spaceres = 0; 2883 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_rename, 0, 0); 2884 } 2885 if (error) 2886 goto out_trans_cancel; 2887 2888 /* 2889 * Attach the dquots to the inodes 2890 */ 2891 error = xfs_qm_vop_rename_dqattach(inodes); 2892 if (error) 2893 goto out_trans_cancel; 2894 2895 /* 2896 * Lock all the participating inodes. Depending upon whether 2897 * the target_name exists in the target directory, and 2898 * whether the target directory is the same as the source 2899 * directory, we can lock from 2 to 4 inodes. 2900 */ 2901 xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL); 2902 2903 /* 2904 * Join all the inodes to the transaction. From this point on, 2905 * we can rely on either trans_commit or trans_cancel to unlock 2906 * them. 2907 */ 2908 xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL); 2909 if (new_parent) 2910 xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL); 2911 xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL); 2912 if (target_ip) 2913 xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL); 2914 if (wip) 2915 xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL); 2916 2917 /* 2918 * If we are using project inheritance, we only allow renames 2919 * into our tree when the project IDs are the same; else the 2920 * tree quota mechanism would be circumvented. 2921 */ 2922 if (unlikely((target_dp->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) && 2923 (xfs_get_projid(target_dp) != xfs_get_projid(src_ip)))) { 2924 error = -EXDEV; 2925 goto out_trans_cancel; 2926 } 2927 2928 xfs_bmap_init(&free_list, &first_block); 2929 2930 /* RENAME_EXCHANGE is unique from here on. */ 2931 if (flags & RENAME_EXCHANGE) 2932 return xfs_cross_rename(tp, src_dp, src_name, src_ip, 2933 target_dp, target_name, target_ip, 2934 &free_list, &first_block, spaceres); 2935 2936 /* 2937 * Set up the target. 2938 */ 2939 if (target_ip == NULL) { 2940 /* 2941 * If there's no space reservation, check the entry will 2942 * fit before actually inserting it. 2943 */ 2944 if (!spaceres) { 2945 error = xfs_dir_canenter(tp, target_dp, target_name); 2946 if (error) 2947 goto out_trans_cancel; 2948 } 2949 /* 2950 * If target does not exist and the rename crosses 2951 * directories, adjust the target directory link count 2952 * to account for the ".." reference from the new entry. 2953 */ 2954 error = xfs_dir_createname(tp, target_dp, target_name, 2955 src_ip->i_ino, &first_block, 2956 &free_list, spaceres); 2957 if (error) 2958 goto out_bmap_cancel; 2959 2960 xfs_trans_ichgtime(tp, target_dp, 2961 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 2962 2963 if (new_parent && src_is_directory) { 2964 error = xfs_bumplink(tp, target_dp); 2965 if (error) 2966 goto out_bmap_cancel; 2967 } 2968 } else { /* target_ip != NULL */ 2969 /* 2970 * If target exists and it's a directory, check that both 2971 * target and source are directories and that target can be 2972 * destroyed, or that neither is a directory. 2973 */ 2974 if (S_ISDIR(target_ip->i_d.di_mode)) { 2975 /* 2976 * Make sure target dir is empty. 2977 */ 2978 if (!(xfs_dir_isempty(target_ip)) || 2979 (target_ip->i_d.di_nlink > 2)) { 2980 error = -EEXIST; 2981 goto out_trans_cancel; 2982 } 2983 } 2984 2985 /* 2986 * Link the source inode under the target name. 2987 * If the source inode is a directory and we are moving 2988 * it across directories, its ".." entry will be 2989 * inconsistent until we replace that down below. 2990 * 2991 * In case there is already an entry with the same 2992 * name at the destination directory, remove it first. 2993 */ 2994 error = xfs_dir_replace(tp, target_dp, target_name, 2995 src_ip->i_ino, 2996 &first_block, &free_list, spaceres); 2997 if (error) 2998 goto out_bmap_cancel; 2999 3000 xfs_trans_ichgtime(tp, target_dp, 3001 XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3002 3003 /* 3004 * Decrement the link count on the target since the target 3005 * dir no longer points to it. 3006 */ 3007 error = xfs_droplink(tp, target_ip); 3008 if (error) 3009 goto out_bmap_cancel; 3010 3011 if (src_is_directory) { 3012 /* 3013 * Drop the link from the old "." entry. 3014 */ 3015 error = xfs_droplink(tp, target_ip); 3016 if (error) 3017 goto out_bmap_cancel; 3018 } 3019 } /* target_ip != NULL */ 3020 3021 /* 3022 * Remove the source. 3023 */ 3024 if (new_parent && src_is_directory) { 3025 /* 3026 * Rewrite the ".." entry to point to the new 3027 * directory. 3028 */ 3029 error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot, 3030 target_dp->i_ino, 3031 &first_block, &free_list, spaceres); 3032 ASSERT(error != -EEXIST); 3033 if (error) 3034 goto out_bmap_cancel; 3035 } 3036 3037 /* 3038 * We always want to hit the ctime on the source inode. 3039 * 3040 * This isn't strictly required by the standards since the source 3041 * inode isn't really being changed, but old unix file systems did 3042 * it and some incremental backup programs won't work without it. 3043 */ 3044 xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG); 3045 xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE); 3046 3047 /* 3048 * Adjust the link count on src_dp. This is necessary when 3049 * renaming a directory, either within one parent when 3050 * the target existed, or across two parent directories. 3051 */ 3052 if (src_is_directory && (new_parent || target_ip != NULL)) { 3053 3054 /* 3055 * Decrement link count on src_directory since the 3056 * entry that's moved no longer points to it. 3057 */ 3058 error = xfs_droplink(tp, src_dp); 3059 if (error) 3060 goto out_bmap_cancel; 3061 } 3062 3063 /* 3064 * For whiteouts, we only need to update the source dirent with the 3065 * inode number of the whiteout inode rather than removing it 3066 * altogether. 3067 */ 3068 if (wip) { 3069 error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino, 3070 &first_block, &free_list, spaceres); 3071 } else 3072 error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino, 3073 &first_block, &free_list, spaceres); 3074 if (error) 3075 goto out_bmap_cancel; 3076 3077 /* 3078 * For whiteouts, we need to bump the link count on the whiteout inode. 3079 * This means that failures all the way up to this point leave the inode 3080 * on the unlinked list and so cleanup is a simple matter of dropping 3081 * the remaining reference to it. If we fail here after bumping the link 3082 * count, we're shutting down the filesystem so we'll never see the 3083 * intermediate state on disk. 3084 */ 3085 if (wip) { 3086 ASSERT(VFS_I(wip)->i_nlink == 0 && wip->i_d.di_nlink == 0); 3087 error = xfs_bumplink(tp, wip); 3088 if (error) 3089 goto out_bmap_cancel; 3090 error = xfs_iunlink_remove(tp, wip); 3091 if (error) 3092 goto out_bmap_cancel; 3093 xfs_trans_log_inode(tp, wip, XFS_ILOG_CORE); 3094 3095 /* 3096 * Now we have a real link, clear the "I'm a tmpfile" state 3097 * flag from the inode so it doesn't accidentally get misused in 3098 * future. 3099 */ 3100 VFS_I(wip)->i_state &= ~I_LINKABLE; 3101 } 3102 3103 xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 3104 xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE); 3105 if (new_parent) 3106 xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE); 3107 3108 error = xfs_finish_rename(tp, &free_list); 3109 if (wip) 3110 IRELE(wip); 3111 return error; 3112 3113 out_bmap_cancel: 3114 xfs_bmap_cancel(&free_list); 3115 out_trans_cancel: 3116 xfs_trans_cancel(tp); 3117 if (wip) 3118 IRELE(wip); 3119 return error; 3120 } 3121 3122 STATIC int 3123 xfs_iflush_cluster( 3124 xfs_inode_t *ip, 3125 xfs_buf_t *bp) 3126 { 3127 xfs_mount_t *mp = ip->i_mount; 3128 struct xfs_perag *pag; 3129 unsigned long first_index, mask; 3130 unsigned long inodes_per_cluster; 3131 int ilist_size; 3132 xfs_inode_t **ilist; 3133 xfs_inode_t *iq; 3134 int nr_found; 3135 int clcount = 0; 3136 int bufwasdelwri; 3137 int i; 3138 3139 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 3140 3141 inodes_per_cluster = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 3142 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 3143 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); 3144 if (!ilist) 3145 goto out_put; 3146 3147 mask = ~(((mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog)) - 1); 3148 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 3149 rcu_read_lock(); 3150 /* really need a gang lookup range call here */ 3151 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, 3152 first_index, inodes_per_cluster); 3153 if (nr_found == 0) 3154 goto out_free; 3155 3156 for (i = 0; i < nr_found; i++) { 3157 iq = ilist[i]; 3158 if (iq == ip) 3159 continue; 3160 3161 /* 3162 * because this is an RCU protected lookup, we could find a 3163 * recently freed or even reallocated inode during the lookup. 3164 * We need to check under the i_flags_lock for a valid inode 3165 * here. Skip it if it is not valid or the wrong inode. 3166 */ 3167 spin_lock(&ip->i_flags_lock); 3168 if (!ip->i_ino || 3169 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { 3170 spin_unlock(&ip->i_flags_lock); 3171 continue; 3172 } 3173 spin_unlock(&ip->i_flags_lock); 3174 3175 /* 3176 * Do an un-protected check to see if the inode is dirty and 3177 * is a candidate for flushing. These checks will be repeated 3178 * later after the appropriate locks are acquired. 3179 */ 3180 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) 3181 continue; 3182 3183 /* 3184 * Try to get locks. If any are unavailable or it is pinned, 3185 * then this inode cannot be flushed and is skipped. 3186 */ 3187 3188 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) 3189 continue; 3190 if (!xfs_iflock_nowait(iq)) { 3191 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3192 continue; 3193 } 3194 if (xfs_ipincount(iq)) { 3195 xfs_ifunlock(iq); 3196 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3197 continue; 3198 } 3199 3200 /* 3201 * arriving here means that this inode can be flushed. First 3202 * re-check that it's dirty before flushing. 3203 */ 3204 if (!xfs_inode_clean(iq)) { 3205 int error; 3206 error = xfs_iflush_int(iq, bp); 3207 if (error) { 3208 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3209 goto cluster_corrupt_out; 3210 } 3211 clcount++; 3212 } else { 3213 xfs_ifunlock(iq); 3214 } 3215 xfs_iunlock(iq, XFS_ILOCK_SHARED); 3216 } 3217 3218 if (clcount) { 3219 XFS_STATS_INC(xs_icluster_flushcnt); 3220 XFS_STATS_ADD(xs_icluster_flushinode, clcount); 3221 } 3222 3223 out_free: 3224 rcu_read_unlock(); 3225 kmem_free(ilist); 3226 out_put: 3227 xfs_perag_put(pag); 3228 return 0; 3229 3230 3231 cluster_corrupt_out: 3232 /* 3233 * Corruption detected in the clustering loop. Invalidate the 3234 * inode buffer and shut down the filesystem. 3235 */ 3236 rcu_read_unlock(); 3237 /* 3238 * Clean up the buffer. If it was delwri, just release it -- 3239 * brelse can handle it with no problems. If not, shut down the 3240 * filesystem before releasing the buffer. 3241 */ 3242 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 3243 if (bufwasdelwri) 3244 xfs_buf_relse(bp); 3245 3246 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3247 3248 if (!bufwasdelwri) { 3249 /* 3250 * Just like incore_relse: if we have b_iodone functions, 3251 * mark the buffer as an error and call them. Otherwise 3252 * mark it as stale and brelse. 3253 */ 3254 if (bp->b_iodone) { 3255 XFS_BUF_UNDONE(bp); 3256 xfs_buf_stale(bp); 3257 xfs_buf_ioerror(bp, -EIO); 3258 xfs_buf_ioend(bp); 3259 } else { 3260 xfs_buf_stale(bp); 3261 xfs_buf_relse(bp); 3262 } 3263 } 3264 3265 /* 3266 * Unlocks the flush lock 3267 */ 3268 xfs_iflush_abort(iq, false); 3269 kmem_free(ilist); 3270 xfs_perag_put(pag); 3271 return -EFSCORRUPTED; 3272 } 3273 3274 /* 3275 * Flush dirty inode metadata into the backing buffer. 3276 * 3277 * The caller must have the inode lock and the inode flush lock held. The 3278 * inode lock will still be held upon return to the caller, and the inode 3279 * flush lock will be released after the inode has reached the disk. 3280 * 3281 * The caller must write out the buffer returned in *bpp and release it. 3282 */ 3283 int 3284 xfs_iflush( 3285 struct xfs_inode *ip, 3286 struct xfs_buf **bpp) 3287 { 3288 struct xfs_mount *mp = ip->i_mount; 3289 struct xfs_buf *bp; 3290 struct xfs_dinode *dip; 3291 int error; 3292 3293 XFS_STATS_INC(xs_iflush_count); 3294 3295 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3296 ASSERT(xfs_isiflocked(ip)); 3297 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3298 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3299 3300 *bpp = NULL; 3301 3302 xfs_iunpin_wait(ip); 3303 3304 /* 3305 * For stale inodes we cannot rely on the backing buffer remaining 3306 * stale in cache for the remaining life of the stale inode and so 3307 * xfs_imap_to_bp() below may give us a buffer that no longer contains 3308 * inodes below. We have to check this after ensuring the inode is 3309 * unpinned so that it is safe to reclaim the stale inode after the 3310 * flush call. 3311 */ 3312 if (xfs_iflags_test(ip, XFS_ISTALE)) { 3313 xfs_ifunlock(ip); 3314 return 0; 3315 } 3316 3317 /* 3318 * This may have been unpinned because the filesystem is shutting 3319 * down forcibly. If that's the case we must not write this inode 3320 * to disk, because the log record didn't make it to disk. 3321 * 3322 * We also have to remove the log item from the AIL in this case, 3323 * as we wait for an empty AIL as part of the unmount process. 3324 */ 3325 if (XFS_FORCED_SHUTDOWN(mp)) { 3326 error = -EIO; 3327 goto abort_out; 3328 } 3329 3330 /* 3331 * Get the buffer containing the on-disk inode. 3332 */ 3333 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 3334 0); 3335 if (error || !bp) { 3336 xfs_ifunlock(ip); 3337 return error; 3338 } 3339 3340 /* 3341 * First flush out the inode that xfs_iflush was called with. 3342 */ 3343 error = xfs_iflush_int(ip, bp); 3344 if (error) 3345 goto corrupt_out; 3346 3347 /* 3348 * If the buffer is pinned then push on the log now so we won't 3349 * get stuck waiting in the write for too long. 3350 */ 3351 if (xfs_buf_ispinned(bp)) 3352 xfs_log_force(mp, 0); 3353 3354 /* 3355 * inode clustering: 3356 * see if other inodes can be gathered into this write 3357 */ 3358 error = xfs_iflush_cluster(ip, bp); 3359 if (error) 3360 goto cluster_corrupt_out; 3361 3362 *bpp = bp; 3363 return 0; 3364 3365 corrupt_out: 3366 xfs_buf_relse(bp); 3367 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 3368 cluster_corrupt_out: 3369 error = -EFSCORRUPTED; 3370 abort_out: 3371 /* 3372 * Unlocks the flush lock 3373 */ 3374 xfs_iflush_abort(ip, false); 3375 return error; 3376 } 3377 3378 STATIC int 3379 xfs_iflush_int( 3380 struct xfs_inode *ip, 3381 struct xfs_buf *bp) 3382 { 3383 struct xfs_inode_log_item *iip = ip->i_itemp; 3384 struct xfs_dinode *dip; 3385 struct xfs_mount *mp = ip->i_mount; 3386 3387 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 3388 ASSERT(xfs_isiflocked(ip)); 3389 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 3390 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 3391 ASSERT(iip != NULL && iip->ili_fields != 0); 3392 ASSERT(ip->i_d.di_version > 1); 3393 3394 /* set *dip = inode's place in the buffer */ 3395 dip = xfs_buf_offset(bp, ip->i_imap.im_boffset); 3396 3397 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 3398 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 3399 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3400 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 3401 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 3402 goto corrupt_out; 3403 } 3404 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, 3405 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { 3406 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3407 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", 3408 __func__, ip->i_ino, ip, ip->i_d.di_magic); 3409 goto corrupt_out; 3410 } 3411 if (S_ISREG(ip->i_d.di_mode)) { 3412 if (XFS_TEST_ERROR( 3413 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3414 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 3415 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 3416 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3417 "%s: Bad regular inode %Lu, ptr 0x%p", 3418 __func__, ip->i_ino, ip); 3419 goto corrupt_out; 3420 } 3421 } else if (S_ISDIR(ip->i_d.di_mode)) { 3422 if (XFS_TEST_ERROR( 3423 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 3424 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 3425 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 3426 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 3427 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3428 "%s: Bad directory inode %Lu, ptr 0x%p", 3429 __func__, ip->i_ino, ip); 3430 goto corrupt_out; 3431 } 3432 } 3433 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 3434 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 3435 XFS_RANDOM_IFLUSH_5)) { 3436 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3437 "%s: detected corrupt incore inode %Lu, " 3438 "total extents = %d, nblocks = %Ld, ptr 0x%p", 3439 __func__, ip->i_ino, 3440 ip->i_d.di_nextents + ip->i_d.di_anextents, 3441 ip->i_d.di_nblocks, ip); 3442 goto corrupt_out; 3443 } 3444 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 3445 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 3446 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 3447 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 3448 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 3449 goto corrupt_out; 3450 } 3451 3452 /* 3453 * Inode item log recovery for v2 inodes are dependent on the 3454 * di_flushiter count for correct sequencing. We bump the flush 3455 * iteration count so we can detect flushes which postdate a log record 3456 * during recovery. This is redundant as we now log every change and 3457 * hence this can't happen but we need to still do it to ensure 3458 * backwards compatibility with old kernels that predate logging all 3459 * inode changes. 3460 */ 3461 if (ip->i_d.di_version < 3) 3462 ip->i_d.di_flushiter++; 3463 3464 /* 3465 * Copy the dirty parts of the inode into the on-disk 3466 * inode. We always copy out the core of the inode, 3467 * because if the inode is dirty at all the core must 3468 * be. 3469 */ 3470 xfs_dinode_to_disk(dip, &ip->i_d); 3471 3472 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 3473 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 3474 ip->i_d.di_flushiter = 0; 3475 3476 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK); 3477 if (XFS_IFORK_Q(ip)) 3478 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK); 3479 xfs_inobp_check(mp, bp); 3480 3481 /* 3482 * We've recorded everything logged in the inode, so we'd like to clear 3483 * the ili_fields bits so we don't log and flush things unnecessarily. 3484 * However, we can't stop logging all this information until the data 3485 * we've copied into the disk buffer is written to disk. If we did we 3486 * might overwrite the copy of the inode in the log with all the data 3487 * after re-logging only part of it, and in the face of a crash we 3488 * wouldn't have all the data we need to recover. 3489 * 3490 * What we do is move the bits to the ili_last_fields field. When 3491 * logging the inode, these bits are moved back to the ili_fields field. 3492 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 3493 * know that the information those bits represent is permanently on 3494 * disk. As long as the flush completes before the inode is logged 3495 * again, then both ili_fields and ili_last_fields will be cleared. 3496 * 3497 * We can play with the ili_fields bits here, because the inode lock 3498 * must be held exclusively in order to set bits there and the flush 3499 * lock protects the ili_last_fields bits. Set ili_logged so the flush 3500 * done routine can tell whether or not to look in the AIL. Also, store 3501 * the current LSN of the inode so that we can tell whether the item has 3502 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 3503 * need the AIL lock, because it is a 64 bit value that cannot be read 3504 * atomically. 3505 */ 3506 iip->ili_last_fields = iip->ili_fields; 3507 iip->ili_fields = 0; 3508 iip->ili_logged = 1; 3509 3510 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 3511 &iip->ili_item.li_lsn); 3512 3513 /* 3514 * Attach the function xfs_iflush_done to the inode's 3515 * buffer. This will remove the inode from the AIL 3516 * and unlock the inode's flush lock when the inode is 3517 * completely written to disk. 3518 */ 3519 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 3520 3521 /* update the lsn in the on disk inode if required */ 3522 if (ip->i_d.di_version == 3) 3523 dip->di_lsn = cpu_to_be64(iip->ili_item.li_lsn); 3524 3525 /* generate the checksum. */ 3526 xfs_dinode_calc_crc(mp, dip); 3527 3528 ASSERT(bp->b_fspriv != NULL); 3529 ASSERT(bp->b_iodone != NULL); 3530 return 0; 3531 3532 corrupt_out: 3533 return -EFSCORRUPTED; 3534 } 3535