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_types.h" 23 #include "xfs_log.h" 24 #include "xfs_inum.h" 25 #include "xfs_trans.h" 26 #include "xfs_trans_priv.h" 27 #include "xfs_sb.h" 28 #include "xfs_ag.h" 29 #include "xfs_mount.h" 30 #include "xfs_bmap_btree.h" 31 #include "xfs_alloc_btree.h" 32 #include "xfs_ialloc_btree.h" 33 #include "xfs_attr_sf.h" 34 #include "xfs_dinode.h" 35 #include "xfs_inode.h" 36 #include "xfs_buf_item.h" 37 #include "xfs_inode_item.h" 38 #include "xfs_btree.h" 39 #include "xfs_alloc.h" 40 #include "xfs_ialloc.h" 41 #include "xfs_bmap.h" 42 #include "xfs_error.h" 43 #include "xfs_utils.h" 44 #include "xfs_quota.h" 45 #include "xfs_filestream.h" 46 #include "xfs_vnodeops.h" 47 #include "xfs_trace.h" 48 #include "xfs_icache.h" 49 50 kmem_zone_t *xfs_ifork_zone; 51 kmem_zone_t *xfs_inode_zone; 52 53 /* 54 * Used in xfs_itruncate_extents(). This is the maximum number of extents 55 * freed from a file in a single transaction. 56 */ 57 #define XFS_ITRUNC_MAX_EXTENTS 2 58 59 STATIC int xfs_iflush_int(xfs_inode_t *, xfs_buf_t *); 60 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int); 61 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int); 62 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int); 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 * This is a wrapper routine around the xfs_ilock() routine used to centralize 80 * some grungy code. It is used in places that wish to lock the inode solely 81 * for reading the extents. The reason these places can't just call 82 * xfs_ilock(SHARED) is that the inode lock also guards to bringing in of the 83 * extents from disk for a file in b-tree format. If the inode is in b-tree 84 * format, then we need to lock the inode exclusively until the extents are read 85 * in. Locking it exclusively all the time would limit our parallelism 86 * unnecessarily, though. What we do instead is check to see if the extents 87 * have been read in yet, and only lock the inode exclusively if they have not. 88 * 89 * The function returns a value which should be given to the corresponding 90 * xfs_iunlock_map_shared(). This value is the mode in which the lock was 91 * actually taken. 92 */ 93 uint 94 xfs_ilock_map_shared( 95 xfs_inode_t *ip) 96 { 97 uint lock_mode; 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 } else { 103 lock_mode = XFS_ILOCK_SHARED; 104 } 105 106 xfs_ilock(ip, lock_mode); 107 108 return lock_mode; 109 } 110 111 /* 112 * This is simply the unlock routine to go with xfs_ilock_map_shared(). 113 * All it does is call xfs_iunlock() with the given lock_mode. 114 */ 115 void 116 xfs_iunlock_map_shared( 117 xfs_inode_t *ip, 118 unsigned int lock_mode) 119 { 120 xfs_iunlock(ip, lock_mode); 121 } 122 123 /* 124 * The xfs inode contains 2 locks: a multi-reader lock called the 125 * i_iolock and a multi-reader lock called the i_lock. This routine 126 * allows either or both of the locks to be obtained. 127 * 128 * The 2 locks should always be ordered so that the IO lock is 129 * obtained first in order to prevent deadlock. 130 * 131 * ip -- the inode being locked 132 * lock_flags -- this parameter indicates the inode's locks 133 * to be locked. It can be: 134 * XFS_IOLOCK_SHARED, 135 * XFS_IOLOCK_EXCL, 136 * XFS_ILOCK_SHARED, 137 * XFS_ILOCK_EXCL, 138 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED, 139 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL, 140 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED, 141 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL 142 */ 143 void 144 xfs_ilock( 145 xfs_inode_t *ip, 146 uint lock_flags) 147 { 148 trace_xfs_ilock(ip, lock_flags, _RET_IP_); 149 150 /* 151 * You can't set both SHARED and EXCL for the same lock, 152 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 153 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 154 */ 155 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 156 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 157 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 158 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 159 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 160 161 if (lock_flags & XFS_IOLOCK_EXCL) 162 mrupdate_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 163 else if (lock_flags & XFS_IOLOCK_SHARED) 164 mraccess_nested(&ip->i_iolock, XFS_IOLOCK_DEP(lock_flags)); 165 166 if (lock_flags & XFS_ILOCK_EXCL) 167 mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 168 else if (lock_flags & XFS_ILOCK_SHARED) 169 mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags)); 170 } 171 172 /* 173 * This is just like xfs_ilock(), except that the caller 174 * is guaranteed not to sleep. It returns 1 if it gets 175 * the requested locks and 0 otherwise. If the IO lock is 176 * obtained but the inode lock cannot be, then the IO lock 177 * is dropped before returning. 178 * 179 * ip -- the inode being locked 180 * lock_flags -- this parameter indicates the inode's locks to be 181 * to be locked. See the comment for xfs_ilock() for a list 182 * of valid values. 183 */ 184 int 185 xfs_ilock_nowait( 186 xfs_inode_t *ip, 187 uint lock_flags) 188 { 189 trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_); 190 191 /* 192 * You can't set both SHARED and EXCL for the same lock, 193 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 194 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 195 */ 196 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 197 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 198 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 199 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 200 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 201 202 if (lock_flags & XFS_IOLOCK_EXCL) { 203 if (!mrtryupdate(&ip->i_iolock)) 204 goto out; 205 } else if (lock_flags & XFS_IOLOCK_SHARED) { 206 if (!mrtryaccess(&ip->i_iolock)) 207 goto out; 208 } 209 if (lock_flags & XFS_ILOCK_EXCL) { 210 if (!mrtryupdate(&ip->i_lock)) 211 goto out_undo_iolock; 212 } else if (lock_flags & XFS_ILOCK_SHARED) { 213 if (!mrtryaccess(&ip->i_lock)) 214 goto out_undo_iolock; 215 } 216 return 1; 217 218 out_undo_iolock: 219 if (lock_flags & XFS_IOLOCK_EXCL) 220 mrunlock_excl(&ip->i_iolock); 221 else if (lock_flags & XFS_IOLOCK_SHARED) 222 mrunlock_shared(&ip->i_iolock); 223 out: 224 return 0; 225 } 226 227 /* 228 * xfs_iunlock() is used to drop the inode locks acquired with 229 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass 230 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so 231 * that we know which locks to drop. 232 * 233 * ip -- the inode being unlocked 234 * lock_flags -- this parameter indicates the inode's locks to be 235 * to be unlocked. See the comment for xfs_ilock() for a list 236 * of valid values for this parameter. 237 * 238 */ 239 void 240 xfs_iunlock( 241 xfs_inode_t *ip, 242 uint lock_flags) 243 { 244 /* 245 * You can't set both SHARED and EXCL for the same lock, 246 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED, 247 * and XFS_ILOCK_EXCL are valid values to set in lock_flags. 248 */ 249 ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) != 250 (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)); 251 ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) != 252 (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 253 ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_DEP_MASK)) == 0); 254 ASSERT(lock_flags != 0); 255 256 if (lock_flags & XFS_IOLOCK_EXCL) 257 mrunlock_excl(&ip->i_iolock); 258 else if (lock_flags & XFS_IOLOCK_SHARED) 259 mrunlock_shared(&ip->i_iolock); 260 261 if (lock_flags & XFS_ILOCK_EXCL) 262 mrunlock_excl(&ip->i_lock); 263 else if (lock_flags & XFS_ILOCK_SHARED) 264 mrunlock_shared(&ip->i_lock); 265 266 trace_xfs_iunlock(ip, lock_flags, _RET_IP_); 267 } 268 269 /* 270 * give up write locks. the i/o lock cannot be held nested 271 * if it is being demoted. 272 */ 273 void 274 xfs_ilock_demote( 275 xfs_inode_t *ip, 276 uint lock_flags) 277 { 278 ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)); 279 ASSERT((lock_flags & ~(XFS_IOLOCK_EXCL|XFS_ILOCK_EXCL)) == 0); 280 281 if (lock_flags & XFS_ILOCK_EXCL) 282 mrdemote(&ip->i_lock); 283 if (lock_flags & XFS_IOLOCK_EXCL) 284 mrdemote(&ip->i_iolock); 285 286 trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_); 287 } 288 289 #ifdef DEBUG 290 int 291 xfs_isilocked( 292 xfs_inode_t *ip, 293 uint lock_flags) 294 { 295 if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) { 296 if (!(lock_flags & XFS_ILOCK_SHARED)) 297 return !!ip->i_lock.mr_writer; 298 return rwsem_is_locked(&ip->i_lock.mr_lock); 299 } 300 301 if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) { 302 if (!(lock_flags & XFS_IOLOCK_SHARED)) 303 return !!ip->i_iolock.mr_writer; 304 return rwsem_is_locked(&ip->i_iolock.mr_lock); 305 } 306 307 ASSERT(0); 308 return 0; 309 } 310 #endif 311 312 void 313 __xfs_iflock( 314 struct xfs_inode *ip) 315 { 316 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IFLOCK_BIT); 317 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IFLOCK_BIT); 318 319 do { 320 prepare_to_wait_exclusive(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 321 if (xfs_isiflocked(ip)) 322 io_schedule(); 323 } while (!xfs_iflock_nowait(ip)); 324 325 finish_wait(wq, &wait.wait); 326 } 327 328 #ifdef DEBUG 329 /* 330 * Make sure that the extents in the given memory buffer 331 * are valid. 332 */ 333 STATIC void 334 xfs_validate_extents( 335 xfs_ifork_t *ifp, 336 int nrecs, 337 xfs_exntfmt_t fmt) 338 { 339 xfs_bmbt_irec_t irec; 340 xfs_bmbt_rec_host_t rec; 341 int i; 342 343 for (i = 0; i < nrecs; i++) { 344 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 345 rec.l0 = get_unaligned(&ep->l0); 346 rec.l1 = get_unaligned(&ep->l1); 347 xfs_bmbt_get_all(&rec, &irec); 348 if (fmt == XFS_EXTFMT_NOSTATE) 349 ASSERT(irec.br_state == XFS_EXT_NORM); 350 } 351 } 352 #else /* DEBUG */ 353 #define xfs_validate_extents(ifp, nrecs, fmt) 354 #endif /* DEBUG */ 355 356 /* 357 * Check that none of the inode's in the buffer have a next 358 * unlinked field of 0. 359 */ 360 #if defined(DEBUG) 361 void 362 xfs_inobp_check( 363 xfs_mount_t *mp, 364 xfs_buf_t *bp) 365 { 366 int i; 367 int j; 368 xfs_dinode_t *dip; 369 370 j = mp->m_inode_cluster_size >> mp->m_sb.sb_inodelog; 371 372 for (i = 0; i < j; i++) { 373 dip = (xfs_dinode_t *)xfs_buf_offset(bp, 374 i * mp->m_sb.sb_inodesize); 375 if (!dip->di_next_unlinked) { 376 xfs_alert(mp, 377 "Detected bogus zero next_unlinked field in incore inode buffer 0x%p.", 378 bp); 379 ASSERT(dip->di_next_unlinked); 380 } 381 } 382 } 383 #endif 384 385 static void 386 xfs_inode_buf_verify( 387 struct xfs_buf *bp) 388 { 389 struct xfs_mount *mp = bp->b_target->bt_mount; 390 int i; 391 int ni; 392 393 /* 394 * Validate the magic number and version of every inode in the buffer 395 */ 396 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock; 397 for (i = 0; i < ni; i++) { 398 int di_ok; 399 xfs_dinode_t *dip; 400 401 dip = (struct xfs_dinode *)xfs_buf_offset(bp, 402 (i << mp->m_sb.sb_inodelog)); 403 di_ok = dip->di_magic == cpu_to_be16(XFS_DINODE_MAGIC) && 404 XFS_DINODE_GOOD_VERSION(dip->di_version); 405 if (unlikely(XFS_TEST_ERROR(!di_ok, mp, 406 XFS_ERRTAG_ITOBP_INOTOBP, 407 XFS_RANDOM_ITOBP_INOTOBP))) { 408 xfs_buf_ioerror(bp, EFSCORRUPTED); 409 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_HIGH, 410 mp, dip); 411 #ifdef DEBUG 412 xfs_emerg(mp, 413 "bad inode magic/vsn daddr %lld #%d (magic=%x)", 414 (unsigned long long)bp->b_bn, i, 415 be16_to_cpu(dip->di_magic)); 416 ASSERT(0); 417 #endif 418 } 419 } 420 xfs_inobp_check(mp, bp); 421 } 422 423 424 static void 425 xfs_inode_buf_read_verify( 426 struct xfs_buf *bp) 427 { 428 xfs_inode_buf_verify(bp); 429 } 430 431 static void 432 xfs_inode_buf_write_verify( 433 struct xfs_buf *bp) 434 { 435 xfs_inode_buf_verify(bp); 436 } 437 438 const struct xfs_buf_ops xfs_inode_buf_ops = { 439 .verify_read = xfs_inode_buf_read_verify, 440 .verify_write = xfs_inode_buf_write_verify, 441 }; 442 443 444 /* 445 * This routine is called to map an inode to the buffer containing the on-disk 446 * version of the inode. It returns a pointer to the buffer containing the 447 * on-disk inode in the bpp parameter, and in the dipp parameter it returns a 448 * pointer to the on-disk inode within that buffer. 449 * 450 * If a non-zero error is returned, then the contents of bpp and dipp are 451 * undefined. 452 */ 453 int 454 xfs_imap_to_bp( 455 struct xfs_mount *mp, 456 struct xfs_trans *tp, 457 struct xfs_imap *imap, 458 struct xfs_dinode **dipp, 459 struct xfs_buf **bpp, 460 uint buf_flags, 461 uint iget_flags) 462 { 463 struct xfs_buf *bp; 464 int error; 465 466 buf_flags |= XBF_UNMAPPED; 467 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno, 468 (int)imap->im_len, buf_flags, &bp, 469 &xfs_inode_buf_ops); 470 if (error) { 471 if (error == EAGAIN) { 472 ASSERT(buf_flags & XBF_TRYLOCK); 473 return error; 474 } 475 476 if (error == EFSCORRUPTED && 477 (iget_flags & XFS_IGET_UNTRUSTED)) 478 return XFS_ERROR(EINVAL); 479 480 xfs_warn(mp, "%s: xfs_trans_read_buf() returned error %d.", 481 __func__, error); 482 return error; 483 } 484 485 *bpp = bp; 486 *dipp = (struct xfs_dinode *)xfs_buf_offset(bp, imap->im_boffset); 487 return 0; 488 } 489 490 /* 491 * Move inode type and inode format specific information from the 492 * on-disk inode to the in-core inode. For fifos, devs, and sockets 493 * this means set if_rdev to the proper value. For files, directories, 494 * and symlinks this means to bring in the in-line data or extent 495 * pointers. For a file in B-tree format, only the root is immediately 496 * brought in-core. The rest will be in-lined in if_extents when it 497 * is first referenced (see xfs_iread_extents()). 498 */ 499 STATIC int 500 xfs_iformat( 501 xfs_inode_t *ip, 502 xfs_dinode_t *dip) 503 { 504 xfs_attr_shortform_t *atp; 505 int size; 506 int error = 0; 507 xfs_fsize_t di_size; 508 509 if (unlikely(be32_to_cpu(dip->di_nextents) + 510 be16_to_cpu(dip->di_anextents) > 511 be64_to_cpu(dip->di_nblocks))) { 512 xfs_warn(ip->i_mount, 513 "corrupt dinode %Lu, extent total = %d, nblocks = %Lu.", 514 (unsigned long long)ip->i_ino, 515 (int)(be32_to_cpu(dip->di_nextents) + 516 be16_to_cpu(dip->di_anextents)), 517 (unsigned long long) 518 be64_to_cpu(dip->di_nblocks)); 519 XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW, 520 ip->i_mount, dip); 521 return XFS_ERROR(EFSCORRUPTED); 522 } 523 524 if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) { 525 xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.", 526 (unsigned long long)ip->i_ino, 527 dip->di_forkoff); 528 XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW, 529 ip->i_mount, dip); 530 return XFS_ERROR(EFSCORRUPTED); 531 } 532 533 if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) && 534 !ip->i_mount->m_rtdev_targp)) { 535 xfs_warn(ip->i_mount, 536 "corrupt dinode %Lu, has realtime flag set.", 537 ip->i_ino); 538 XFS_CORRUPTION_ERROR("xfs_iformat(realtime)", 539 XFS_ERRLEVEL_LOW, ip->i_mount, dip); 540 return XFS_ERROR(EFSCORRUPTED); 541 } 542 543 switch (ip->i_d.di_mode & S_IFMT) { 544 case S_IFIFO: 545 case S_IFCHR: 546 case S_IFBLK: 547 case S_IFSOCK: 548 if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) { 549 XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW, 550 ip->i_mount, dip); 551 return XFS_ERROR(EFSCORRUPTED); 552 } 553 ip->i_d.di_size = 0; 554 ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip); 555 break; 556 557 case S_IFREG: 558 case S_IFLNK: 559 case S_IFDIR: 560 switch (dip->di_format) { 561 case XFS_DINODE_FMT_LOCAL: 562 /* 563 * no local regular files yet 564 */ 565 if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) { 566 xfs_warn(ip->i_mount, 567 "corrupt inode %Lu (local format for regular file).", 568 (unsigned long long) ip->i_ino); 569 XFS_CORRUPTION_ERROR("xfs_iformat(4)", 570 XFS_ERRLEVEL_LOW, 571 ip->i_mount, dip); 572 return XFS_ERROR(EFSCORRUPTED); 573 } 574 575 di_size = be64_to_cpu(dip->di_size); 576 if (unlikely(di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) { 577 xfs_warn(ip->i_mount, 578 "corrupt inode %Lu (bad size %Ld for local inode).", 579 (unsigned long long) ip->i_ino, 580 (long long) di_size); 581 XFS_CORRUPTION_ERROR("xfs_iformat(5)", 582 XFS_ERRLEVEL_LOW, 583 ip->i_mount, dip); 584 return XFS_ERROR(EFSCORRUPTED); 585 } 586 587 size = (int)di_size; 588 error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size); 589 break; 590 case XFS_DINODE_FMT_EXTENTS: 591 error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK); 592 break; 593 case XFS_DINODE_FMT_BTREE: 594 error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK); 595 break; 596 default: 597 XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW, 598 ip->i_mount); 599 return XFS_ERROR(EFSCORRUPTED); 600 } 601 break; 602 603 default: 604 XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount); 605 return XFS_ERROR(EFSCORRUPTED); 606 } 607 if (error) { 608 return error; 609 } 610 if (!XFS_DFORK_Q(dip)) 611 return 0; 612 613 ASSERT(ip->i_afp == NULL); 614 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS); 615 616 switch (dip->di_aformat) { 617 case XFS_DINODE_FMT_LOCAL: 618 atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip); 619 size = be16_to_cpu(atp->hdr.totsize); 620 621 if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) { 622 xfs_warn(ip->i_mount, 623 "corrupt inode %Lu (bad attr fork size %Ld).", 624 (unsigned long long) ip->i_ino, 625 (long long) size); 626 XFS_CORRUPTION_ERROR("xfs_iformat(8)", 627 XFS_ERRLEVEL_LOW, 628 ip->i_mount, dip); 629 return XFS_ERROR(EFSCORRUPTED); 630 } 631 632 error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size); 633 break; 634 case XFS_DINODE_FMT_EXTENTS: 635 error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK); 636 break; 637 case XFS_DINODE_FMT_BTREE: 638 error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK); 639 break; 640 default: 641 error = XFS_ERROR(EFSCORRUPTED); 642 break; 643 } 644 if (error) { 645 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 646 ip->i_afp = NULL; 647 xfs_idestroy_fork(ip, XFS_DATA_FORK); 648 } 649 return error; 650 } 651 652 /* 653 * The file is in-lined in the on-disk inode. 654 * If it fits into if_inline_data, then copy 655 * it there, otherwise allocate a buffer for it 656 * and copy the data there. Either way, set 657 * if_data to point at the data. 658 * If we allocate a buffer for the data, make 659 * sure that its size is a multiple of 4 and 660 * record the real size in i_real_bytes. 661 */ 662 STATIC int 663 xfs_iformat_local( 664 xfs_inode_t *ip, 665 xfs_dinode_t *dip, 666 int whichfork, 667 int size) 668 { 669 xfs_ifork_t *ifp; 670 int real_size; 671 672 /* 673 * If the size is unreasonable, then something 674 * is wrong and we just bail out rather than crash in 675 * kmem_alloc() or memcpy() below. 676 */ 677 if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 678 xfs_warn(ip->i_mount, 679 "corrupt inode %Lu (bad size %d for local fork, size = %d).", 680 (unsigned long long) ip->i_ino, size, 681 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork)); 682 XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW, 683 ip->i_mount, dip); 684 return XFS_ERROR(EFSCORRUPTED); 685 } 686 ifp = XFS_IFORK_PTR(ip, whichfork); 687 real_size = 0; 688 if (size == 0) 689 ifp->if_u1.if_data = NULL; 690 else if (size <= sizeof(ifp->if_u2.if_inline_data)) 691 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 692 else { 693 real_size = roundup(size, 4); 694 ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS); 695 } 696 ifp->if_bytes = size; 697 ifp->if_real_bytes = real_size; 698 if (size) 699 memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size); 700 ifp->if_flags &= ~XFS_IFEXTENTS; 701 ifp->if_flags |= XFS_IFINLINE; 702 return 0; 703 } 704 705 /* 706 * The file consists of a set of extents all 707 * of which fit into the on-disk inode. 708 * If there are few enough extents to fit into 709 * the if_inline_ext, then copy them there. 710 * Otherwise allocate a buffer for them and copy 711 * them into it. Either way, set if_extents 712 * to point at the extents. 713 */ 714 STATIC int 715 xfs_iformat_extents( 716 xfs_inode_t *ip, 717 xfs_dinode_t *dip, 718 int whichfork) 719 { 720 xfs_bmbt_rec_t *dp; 721 xfs_ifork_t *ifp; 722 int nex; 723 int size; 724 int i; 725 726 ifp = XFS_IFORK_PTR(ip, whichfork); 727 nex = XFS_DFORK_NEXTENTS(dip, whichfork); 728 size = nex * (uint)sizeof(xfs_bmbt_rec_t); 729 730 /* 731 * If the number of extents is unreasonable, then something 732 * is wrong and we just bail out rather than crash in 733 * kmem_alloc() or memcpy() below. 734 */ 735 if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) { 736 xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).", 737 (unsigned long long) ip->i_ino, nex); 738 XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW, 739 ip->i_mount, dip); 740 return XFS_ERROR(EFSCORRUPTED); 741 } 742 743 ifp->if_real_bytes = 0; 744 if (nex == 0) 745 ifp->if_u1.if_extents = NULL; 746 else if (nex <= XFS_INLINE_EXTS) 747 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 748 else 749 xfs_iext_add(ifp, 0, nex); 750 751 ifp->if_bytes = size; 752 if (size) { 753 dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork); 754 xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip)); 755 for (i = 0; i < nex; i++, dp++) { 756 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 757 ep->l0 = get_unaligned_be64(&dp->l0); 758 ep->l1 = get_unaligned_be64(&dp->l1); 759 } 760 XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork); 761 if (whichfork != XFS_DATA_FORK || 762 XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE) 763 if (unlikely(xfs_check_nostate_extents( 764 ifp, 0, nex))) { 765 XFS_ERROR_REPORT("xfs_iformat_extents(2)", 766 XFS_ERRLEVEL_LOW, 767 ip->i_mount); 768 return XFS_ERROR(EFSCORRUPTED); 769 } 770 } 771 ifp->if_flags |= XFS_IFEXTENTS; 772 return 0; 773 } 774 775 /* 776 * The file has too many extents to fit into 777 * the inode, so they are in B-tree format. 778 * Allocate a buffer for the root of the B-tree 779 * and copy the root into it. The i_extents 780 * field will remain NULL until all of the 781 * extents are read in (when they are needed). 782 */ 783 STATIC int 784 xfs_iformat_btree( 785 xfs_inode_t *ip, 786 xfs_dinode_t *dip, 787 int whichfork) 788 { 789 xfs_bmdr_block_t *dfp; 790 xfs_ifork_t *ifp; 791 /* REFERENCED */ 792 int nrecs; 793 int size; 794 795 ifp = XFS_IFORK_PTR(ip, whichfork); 796 dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork); 797 size = XFS_BMAP_BROOT_SPACE(dfp); 798 nrecs = be16_to_cpu(dfp->bb_numrecs); 799 800 /* 801 * blow out if -- fork has less extents than can fit in 802 * fork (fork shouldn't be a btree format), root btree 803 * block has more records than can fit into the fork, 804 * or the number of extents is greater than the number of 805 * blocks. 806 */ 807 if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <= 808 XFS_IFORK_MAXEXT(ip, whichfork) || 809 XFS_BMDR_SPACE_CALC(nrecs) > 810 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork) || 811 XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) { 812 xfs_warn(ip->i_mount, "corrupt inode %Lu (btree).", 813 (unsigned long long) ip->i_ino); 814 XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW, 815 ip->i_mount, dip); 816 return XFS_ERROR(EFSCORRUPTED); 817 } 818 819 ifp->if_broot_bytes = size; 820 ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS); 821 ASSERT(ifp->if_broot != NULL); 822 /* 823 * Copy and convert from the on-disk structure 824 * to the in-memory structure. 825 */ 826 xfs_bmdr_to_bmbt(ip->i_mount, dfp, 827 XFS_DFORK_SIZE(dip, ip->i_mount, whichfork), 828 ifp->if_broot, size); 829 ifp->if_flags &= ~XFS_IFEXTENTS; 830 ifp->if_flags |= XFS_IFBROOT; 831 832 return 0; 833 } 834 835 STATIC void 836 xfs_dinode_from_disk( 837 xfs_icdinode_t *to, 838 xfs_dinode_t *from) 839 { 840 to->di_magic = be16_to_cpu(from->di_magic); 841 to->di_mode = be16_to_cpu(from->di_mode); 842 to->di_version = from ->di_version; 843 to->di_format = from->di_format; 844 to->di_onlink = be16_to_cpu(from->di_onlink); 845 to->di_uid = be32_to_cpu(from->di_uid); 846 to->di_gid = be32_to_cpu(from->di_gid); 847 to->di_nlink = be32_to_cpu(from->di_nlink); 848 to->di_projid_lo = be16_to_cpu(from->di_projid_lo); 849 to->di_projid_hi = be16_to_cpu(from->di_projid_hi); 850 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); 851 to->di_flushiter = be16_to_cpu(from->di_flushiter); 852 to->di_atime.t_sec = be32_to_cpu(from->di_atime.t_sec); 853 to->di_atime.t_nsec = be32_to_cpu(from->di_atime.t_nsec); 854 to->di_mtime.t_sec = be32_to_cpu(from->di_mtime.t_sec); 855 to->di_mtime.t_nsec = be32_to_cpu(from->di_mtime.t_nsec); 856 to->di_ctime.t_sec = be32_to_cpu(from->di_ctime.t_sec); 857 to->di_ctime.t_nsec = be32_to_cpu(from->di_ctime.t_nsec); 858 to->di_size = be64_to_cpu(from->di_size); 859 to->di_nblocks = be64_to_cpu(from->di_nblocks); 860 to->di_extsize = be32_to_cpu(from->di_extsize); 861 to->di_nextents = be32_to_cpu(from->di_nextents); 862 to->di_anextents = be16_to_cpu(from->di_anextents); 863 to->di_forkoff = from->di_forkoff; 864 to->di_aformat = from->di_aformat; 865 to->di_dmevmask = be32_to_cpu(from->di_dmevmask); 866 to->di_dmstate = be16_to_cpu(from->di_dmstate); 867 to->di_flags = be16_to_cpu(from->di_flags); 868 to->di_gen = be32_to_cpu(from->di_gen); 869 } 870 871 void 872 xfs_dinode_to_disk( 873 xfs_dinode_t *to, 874 xfs_icdinode_t *from) 875 { 876 to->di_magic = cpu_to_be16(from->di_magic); 877 to->di_mode = cpu_to_be16(from->di_mode); 878 to->di_version = from ->di_version; 879 to->di_format = from->di_format; 880 to->di_onlink = cpu_to_be16(from->di_onlink); 881 to->di_uid = cpu_to_be32(from->di_uid); 882 to->di_gid = cpu_to_be32(from->di_gid); 883 to->di_nlink = cpu_to_be32(from->di_nlink); 884 to->di_projid_lo = cpu_to_be16(from->di_projid_lo); 885 to->di_projid_hi = cpu_to_be16(from->di_projid_hi); 886 memcpy(to->di_pad, from->di_pad, sizeof(to->di_pad)); 887 to->di_flushiter = cpu_to_be16(from->di_flushiter); 888 to->di_atime.t_sec = cpu_to_be32(from->di_atime.t_sec); 889 to->di_atime.t_nsec = cpu_to_be32(from->di_atime.t_nsec); 890 to->di_mtime.t_sec = cpu_to_be32(from->di_mtime.t_sec); 891 to->di_mtime.t_nsec = cpu_to_be32(from->di_mtime.t_nsec); 892 to->di_ctime.t_sec = cpu_to_be32(from->di_ctime.t_sec); 893 to->di_ctime.t_nsec = cpu_to_be32(from->di_ctime.t_nsec); 894 to->di_size = cpu_to_be64(from->di_size); 895 to->di_nblocks = cpu_to_be64(from->di_nblocks); 896 to->di_extsize = cpu_to_be32(from->di_extsize); 897 to->di_nextents = cpu_to_be32(from->di_nextents); 898 to->di_anextents = cpu_to_be16(from->di_anextents); 899 to->di_forkoff = from->di_forkoff; 900 to->di_aformat = from->di_aformat; 901 to->di_dmevmask = cpu_to_be32(from->di_dmevmask); 902 to->di_dmstate = cpu_to_be16(from->di_dmstate); 903 to->di_flags = cpu_to_be16(from->di_flags); 904 to->di_gen = cpu_to_be32(from->di_gen); 905 } 906 907 STATIC uint 908 _xfs_dic2xflags( 909 __uint16_t di_flags) 910 { 911 uint flags = 0; 912 913 if (di_flags & XFS_DIFLAG_ANY) { 914 if (di_flags & XFS_DIFLAG_REALTIME) 915 flags |= XFS_XFLAG_REALTIME; 916 if (di_flags & XFS_DIFLAG_PREALLOC) 917 flags |= XFS_XFLAG_PREALLOC; 918 if (di_flags & XFS_DIFLAG_IMMUTABLE) 919 flags |= XFS_XFLAG_IMMUTABLE; 920 if (di_flags & XFS_DIFLAG_APPEND) 921 flags |= XFS_XFLAG_APPEND; 922 if (di_flags & XFS_DIFLAG_SYNC) 923 flags |= XFS_XFLAG_SYNC; 924 if (di_flags & XFS_DIFLAG_NOATIME) 925 flags |= XFS_XFLAG_NOATIME; 926 if (di_flags & XFS_DIFLAG_NODUMP) 927 flags |= XFS_XFLAG_NODUMP; 928 if (di_flags & XFS_DIFLAG_RTINHERIT) 929 flags |= XFS_XFLAG_RTINHERIT; 930 if (di_flags & XFS_DIFLAG_PROJINHERIT) 931 flags |= XFS_XFLAG_PROJINHERIT; 932 if (di_flags & XFS_DIFLAG_NOSYMLINKS) 933 flags |= XFS_XFLAG_NOSYMLINKS; 934 if (di_flags & XFS_DIFLAG_EXTSIZE) 935 flags |= XFS_XFLAG_EXTSIZE; 936 if (di_flags & XFS_DIFLAG_EXTSZINHERIT) 937 flags |= XFS_XFLAG_EXTSZINHERIT; 938 if (di_flags & XFS_DIFLAG_NODEFRAG) 939 flags |= XFS_XFLAG_NODEFRAG; 940 if (di_flags & XFS_DIFLAG_FILESTREAM) 941 flags |= XFS_XFLAG_FILESTREAM; 942 } 943 944 return flags; 945 } 946 947 uint 948 xfs_ip2xflags( 949 xfs_inode_t *ip) 950 { 951 xfs_icdinode_t *dic = &ip->i_d; 952 953 return _xfs_dic2xflags(dic->di_flags) | 954 (XFS_IFORK_Q(ip) ? XFS_XFLAG_HASATTR : 0); 955 } 956 957 uint 958 xfs_dic2xflags( 959 xfs_dinode_t *dip) 960 { 961 return _xfs_dic2xflags(be16_to_cpu(dip->di_flags)) | 962 (XFS_DFORK_Q(dip) ? XFS_XFLAG_HASATTR : 0); 963 } 964 965 /* 966 * Read the disk inode attributes into the in-core inode structure. 967 */ 968 int 969 xfs_iread( 970 xfs_mount_t *mp, 971 xfs_trans_t *tp, 972 xfs_inode_t *ip, 973 uint iget_flags) 974 { 975 xfs_buf_t *bp; 976 xfs_dinode_t *dip; 977 int error; 978 979 /* 980 * Fill in the location information in the in-core inode. 981 */ 982 error = xfs_imap(mp, tp, ip->i_ino, &ip->i_imap, iget_flags); 983 if (error) 984 return error; 985 986 /* 987 * Get pointers to the on-disk inode and the buffer containing it. 988 */ 989 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &bp, 0, iget_flags); 990 if (error) 991 return error; 992 993 /* 994 * If we got something that isn't an inode it means someone 995 * (nfs or dmi) has a stale handle. 996 */ 997 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC)) { 998 #ifdef DEBUG 999 xfs_alert(mp, 1000 "%s: dip->di_magic (0x%x) != XFS_DINODE_MAGIC (0x%x)", 1001 __func__, be16_to_cpu(dip->di_magic), XFS_DINODE_MAGIC); 1002 #endif /* DEBUG */ 1003 error = XFS_ERROR(EINVAL); 1004 goto out_brelse; 1005 } 1006 1007 /* 1008 * If the on-disk inode is already linked to a directory 1009 * entry, copy all of the inode into the in-core inode. 1010 * xfs_iformat() handles copying in the inode format 1011 * specific information. 1012 * Otherwise, just get the truly permanent information. 1013 */ 1014 if (dip->di_mode) { 1015 xfs_dinode_from_disk(&ip->i_d, dip); 1016 error = xfs_iformat(ip, dip); 1017 if (error) { 1018 #ifdef DEBUG 1019 xfs_alert(mp, "%s: xfs_iformat() returned error %d", 1020 __func__, error); 1021 #endif /* DEBUG */ 1022 goto out_brelse; 1023 } 1024 } else { 1025 ip->i_d.di_magic = be16_to_cpu(dip->di_magic); 1026 ip->i_d.di_version = dip->di_version; 1027 ip->i_d.di_gen = be32_to_cpu(dip->di_gen); 1028 ip->i_d.di_flushiter = be16_to_cpu(dip->di_flushiter); 1029 /* 1030 * Make sure to pull in the mode here as well in 1031 * case the inode is released without being used. 1032 * This ensures that xfs_inactive() will see that 1033 * the inode is already free and not try to mess 1034 * with the uninitialized part of it. 1035 */ 1036 ip->i_d.di_mode = 0; 1037 } 1038 1039 /* 1040 * The inode format changed when we moved the link count and 1041 * made it 32 bits long. If this is an old format inode, 1042 * convert it in memory to look like a new one. If it gets 1043 * flushed to disk we will convert back before flushing or 1044 * logging it. We zero out the new projid field and the old link 1045 * count field. We'll handle clearing the pad field (the remains 1046 * of the old uuid field) when we actually convert the inode to 1047 * the new format. We don't change the version number so that we 1048 * can distinguish this from a real new format inode. 1049 */ 1050 if (ip->i_d.di_version == 1) { 1051 ip->i_d.di_nlink = ip->i_d.di_onlink; 1052 ip->i_d.di_onlink = 0; 1053 xfs_set_projid(ip, 0); 1054 } 1055 1056 ip->i_delayed_blks = 0; 1057 1058 /* 1059 * Mark the buffer containing the inode as something to keep 1060 * around for a while. This helps to keep recently accessed 1061 * meta-data in-core longer. 1062 */ 1063 xfs_buf_set_ref(bp, XFS_INO_REF); 1064 1065 /* 1066 * Use xfs_trans_brelse() to release the buffer containing the 1067 * on-disk inode, because it was acquired with xfs_trans_read_buf() 1068 * in xfs_imap_to_bp() above. If tp is NULL, this is just a normal 1069 * brelse(). If we're within a transaction, then xfs_trans_brelse() 1070 * will only release the buffer if it is not dirty within the 1071 * transaction. It will be OK to release the buffer in this case, 1072 * because inodes on disk are never destroyed and we will be 1073 * locking the new in-core inode before putting it in the hash 1074 * table where other processes can find it. Thus we don't have 1075 * to worry about the inode being changed just because we released 1076 * the buffer. 1077 */ 1078 out_brelse: 1079 xfs_trans_brelse(tp, bp); 1080 return error; 1081 } 1082 1083 /* 1084 * Read in extents from a btree-format inode. 1085 * Allocate and fill in if_extents. Real work is done in xfs_bmap.c. 1086 */ 1087 int 1088 xfs_iread_extents( 1089 xfs_trans_t *tp, 1090 xfs_inode_t *ip, 1091 int whichfork) 1092 { 1093 int error; 1094 xfs_ifork_t *ifp; 1095 xfs_extnum_t nextents; 1096 1097 if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) { 1098 XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW, 1099 ip->i_mount); 1100 return XFS_ERROR(EFSCORRUPTED); 1101 } 1102 nextents = XFS_IFORK_NEXTENTS(ip, whichfork); 1103 ifp = XFS_IFORK_PTR(ip, whichfork); 1104 1105 /* 1106 * We know that the size is valid (it's checked in iformat_btree) 1107 */ 1108 ifp->if_bytes = ifp->if_real_bytes = 0; 1109 ifp->if_flags |= XFS_IFEXTENTS; 1110 xfs_iext_add(ifp, 0, nextents); 1111 error = xfs_bmap_read_extents(tp, ip, whichfork); 1112 if (error) { 1113 xfs_iext_destroy(ifp); 1114 ifp->if_flags &= ~XFS_IFEXTENTS; 1115 return error; 1116 } 1117 xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip)); 1118 return 0; 1119 } 1120 1121 /* 1122 * Allocate an inode on disk and return a copy of its in-core version. 1123 * The in-core inode is locked exclusively. Set mode, nlink, and rdev 1124 * appropriately within the inode. The uid and gid for the inode are 1125 * set according to the contents of the given cred structure. 1126 * 1127 * Use xfs_dialloc() to allocate the on-disk inode. If xfs_dialloc() 1128 * has a free inode available, call xfs_iget() to obtain the in-core 1129 * version of the allocated inode. Finally, fill in the inode and 1130 * log its initial contents. In this case, ialloc_context would be 1131 * set to NULL. 1132 * 1133 * If xfs_dialloc() does not have an available inode, it will replenish 1134 * its supply by doing an allocation. Since we can only do one 1135 * allocation within a transaction without deadlocks, we must commit 1136 * the current transaction before returning the inode itself. 1137 * In this case, therefore, we will set ialloc_context and return. 1138 * The caller should then commit the current transaction, start a new 1139 * transaction, and call xfs_ialloc() again to actually get the inode. 1140 * 1141 * To ensure that some other process does not grab the inode that 1142 * was allocated during the first call to xfs_ialloc(), this routine 1143 * also returns the [locked] bp pointing to the head of the freelist 1144 * as ialloc_context. The caller should hold this buffer across 1145 * the commit and pass it back into this routine on the second call. 1146 * 1147 * If we are allocating quota inodes, we do not have a parent inode 1148 * to attach to or associate with (i.e. pip == NULL) because they 1149 * are not linked into the directory structure - they are attached 1150 * directly to the superblock - and so have no parent. 1151 */ 1152 int 1153 xfs_ialloc( 1154 xfs_trans_t *tp, 1155 xfs_inode_t *pip, 1156 umode_t mode, 1157 xfs_nlink_t nlink, 1158 xfs_dev_t rdev, 1159 prid_t prid, 1160 int okalloc, 1161 xfs_buf_t **ialloc_context, 1162 xfs_inode_t **ipp) 1163 { 1164 xfs_ino_t ino; 1165 xfs_inode_t *ip; 1166 uint flags; 1167 int error; 1168 timespec_t tv; 1169 int filestreams = 0; 1170 1171 /* 1172 * Call the space management code to pick 1173 * the on-disk inode to be allocated. 1174 */ 1175 error = xfs_dialloc(tp, pip ? pip->i_ino : 0, mode, okalloc, 1176 ialloc_context, &ino); 1177 if (error) 1178 return error; 1179 if (*ialloc_context || ino == NULLFSINO) { 1180 *ipp = NULL; 1181 return 0; 1182 } 1183 ASSERT(*ialloc_context == NULL); 1184 1185 /* 1186 * Get the in-core inode with the lock held exclusively. 1187 * This is because we're setting fields here we need 1188 * to prevent others from looking at until we're done. 1189 */ 1190 error = xfs_iget(tp->t_mountp, tp, ino, XFS_IGET_CREATE, 1191 XFS_ILOCK_EXCL, &ip); 1192 if (error) 1193 return error; 1194 ASSERT(ip != NULL); 1195 1196 ip->i_d.di_mode = mode; 1197 ip->i_d.di_onlink = 0; 1198 ip->i_d.di_nlink = nlink; 1199 ASSERT(ip->i_d.di_nlink == nlink); 1200 ip->i_d.di_uid = current_fsuid(); 1201 ip->i_d.di_gid = current_fsgid(); 1202 xfs_set_projid(ip, prid); 1203 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 1204 1205 /* 1206 * If the superblock version is up to where we support new format 1207 * inodes and this is currently an old format inode, then change 1208 * the inode version number now. This way we only do the conversion 1209 * here rather than here and in the flush/logging code. 1210 */ 1211 if (xfs_sb_version_hasnlink(&tp->t_mountp->m_sb) && 1212 ip->i_d.di_version == 1) { 1213 ip->i_d.di_version = 2; 1214 /* 1215 * We've already zeroed the old link count, the projid field, 1216 * and the pad field. 1217 */ 1218 } 1219 1220 /* 1221 * Project ids won't be stored on disk if we are using a version 1 inode. 1222 */ 1223 if ((prid != 0) && (ip->i_d.di_version == 1)) 1224 xfs_bump_ino_vers2(tp, ip); 1225 1226 if (pip && XFS_INHERIT_GID(pip)) { 1227 ip->i_d.di_gid = pip->i_d.di_gid; 1228 if ((pip->i_d.di_mode & S_ISGID) && S_ISDIR(mode)) { 1229 ip->i_d.di_mode |= S_ISGID; 1230 } 1231 } 1232 1233 /* 1234 * If the group ID of the new file does not match the effective group 1235 * ID or one of the supplementary group IDs, the S_ISGID bit is cleared 1236 * (and only if the irix_sgid_inherit compatibility variable is set). 1237 */ 1238 if ((irix_sgid_inherit) && 1239 (ip->i_d.di_mode & S_ISGID) && 1240 (!in_group_p((gid_t)ip->i_d.di_gid))) { 1241 ip->i_d.di_mode &= ~S_ISGID; 1242 } 1243 1244 ip->i_d.di_size = 0; 1245 ip->i_d.di_nextents = 0; 1246 ASSERT(ip->i_d.di_nblocks == 0); 1247 1248 nanotime(&tv); 1249 ip->i_d.di_mtime.t_sec = (__int32_t)tv.tv_sec; 1250 ip->i_d.di_mtime.t_nsec = (__int32_t)tv.tv_nsec; 1251 ip->i_d.di_atime = ip->i_d.di_mtime; 1252 ip->i_d.di_ctime = ip->i_d.di_mtime; 1253 1254 /* 1255 * di_gen will have been taken care of in xfs_iread. 1256 */ 1257 ip->i_d.di_extsize = 0; 1258 ip->i_d.di_dmevmask = 0; 1259 ip->i_d.di_dmstate = 0; 1260 ip->i_d.di_flags = 0; 1261 flags = XFS_ILOG_CORE; 1262 switch (mode & S_IFMT) { 1263 case S_IFIFO: 1264 case S_IFCHR: 1265 case S_IFBLK: 1266 case S_IFSOCK: 1267 ip->i_d.di_format = XFS_DINODE_FMT_DEV; 1268 ip->i_df.if_u2.if_rdev = rdev; 1269 ip->i_df.if_flags = 0; 1270 flags |= XFS_ILOG_DEV; 1271 break; 1272 case S_IFREG: 1273 /* 1274 * we can't set up filestreams until after the VFS inode 1275 * is set up properly. 1276 */ 1277 if (pip && xfs_inode_is_filestream(pip)) 1278 filestreams = 1; 1279 /* fall through */ 1280 case S_IFDIR: 1281 if (pip && (pip->i_d.di_flags & XFS_DIFLAG_ANY)) { 1282 uint di_flags = 0; 1283 1284 if (S_ISDIR(mode)) { 1285 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 1286 di_flags |= XFS_DIFLAG_RTINHERIT; 1287 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1288 di_flags |= XFS_DIFLAG_EXTSZINHERIT; 1289 ip->i_d.di_extsize = pip->i_d.di_extsize; 1290 } 1291 } else if (S_ISREG(mode)) { 1292 if (pip->i_d.di_flags & XFS_DIFLAG_RTINHERIT) 1293 di_flags |= XFS_DIFLAG_REALTIME; 1294 if (pip->i_d.di_flags & XFS_DIFLAG_EXTSZINHERIT) { 1295 di_flags |= XFS_DIFLAG_EXTSIZE; 1296 ip->i_d.di_extsize = pip->i_d.di_extsize; 1297 } 1298 } 1299 if ((pip->i_d.di_flags & XFS_DIFLAG_NOATIME) && 1300 xfs_inherit_noatime) 1301 di_flags |= XFS_DIFLAG_NOATIME; 1302 if ((pip->i_d.di_flags & XFS_DIFLAG_NODUMP) && 1303 xfs_inherit_nodump) 1304 di_flags |= XFS_DIFLAG_NODUMP; 1305 if ((pip->i_d.di_flags & XFS_DIFLAG_SYNC) && 1306 xfs_inherit_sync) 1307 di_flags |= XFS_DIFLAG_SYNC; 1308 if ((pip->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) && 1309 xfs_inherit_nosymlinks) 1310 di_flags |= XFS_DIFLAG_NOSYMLINKS; 1311 if (pip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) 1312 di_flags |= XFS_DIFLAG_PROJINHERIT; 1313 if ((pip->i_d.di_flags & XFS_DIFLAG_NODEFRAG) && 1314 xfs_inherit_nodefrag) 1315 di_flags |= XFS_DIFLAG_NODEFRAG; 1316 if (pip->i_d.di_flags & XFS_DIFLAG_FILESTREAM) 1317 di_flags |= XFS_DIFLAG_FILESTREAM; 1318 ip->i_d.di_flags |= di_flags; 1319 } 1320 /* FALLTHROUGH */ 1321 case S_IFLNK: 1322 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 1323 ip->i_df.if_flags = XFS_IFEXTENTS; 1324 ip->i_df.if_bytes = ip->i_df.if_real_bytes = 0; 1325 ip->i_df.if_u1.if_extents = NULL; 1326 break; 1327 default: 1328 ASSERT(0); 1329 } 1330 /* 1331 * Attribute fork settings for new inode. 1332 */ 1333 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 1334 ip->i_d.di_anextents = 0; 1335 1336 /* 1337 * Log the new values stuffed into the inode. 1338 */ 1339 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1340 xfs_trans_log_inode(tp, ip, flags); 1341 1342 /* now that we have an i_mode we can setup inode ops and unlock */ 1343 xfs_setup_inode(ip); 1344 1345 /* now we have set up the vfs inode we can associate the filestream */ 1346 if (filestreams) { 1347 error = xfs_filestream_associate(pip, ip); 1348 if (error < 0) 1349 return -error; 1350 if (!error) 1351 xfs_iflags_set(ip, XFS_IFILESTREAM); 1352 } 1353 1354 *ipp = ip; 1355 return 0; 1356 } 1357 1358 /* 1359 * Free up the underlying blocks past new_size. The new size must be smaller 1360 * than the current size. This routine can be used both for the attribute and 1361 * data fork, and does not modify the inode size, which is left to the caller. 1362 * 1363 * The transaction passed to this routine must have made a permanent log 1364 * reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the 1365 * given transaction and start new ones, so make sure everything involved in 1366 * the transaction is tidy before calling here. Some transaction will be 1367 * returned to the caller to be committed. The incoming transaction must 1368 * already include the inode, and both inode locks must be held exclusively. 1369 * The inode must also be "held" within the transaction. On return the inode 1370 * will be "held" within the returned transaction. This routine does NOT 1371 * require any disk space to be reserved for it within the transaction. 1372 * 1373 * If we get an error, we must return with the inode locked and linked into the 1374 * current transaction. This keeps things simple for the higher level code, 1375 * because it always knows that the inode is locked and held in the transaction 1376 * that returns to it whether errors occur or not. We don't mark the inode 1377 * dirty on error so that transactions can be easily aborted if possible. 1378 */ 1379 int 1380 xfs_itruncate_extents( 1381 struct xfs_trans **tpp, 1382 struct xfs_inode *ip, 1383 int whichfork, 1384 xfs_fsize_t new_size) 1385 { 1386 struct xfs_mount *mp = ip->i_mount; 1387 struct xfs_trans *tp = *tpp; 1388 struct xfs_trans *ntp; 1389 xfs_bmap_free_t free_list; 1390 xfs_fsblock_t first_block; 1391 xfs_fileoff_t first_unmap_block; 1392 xfs_fileoff_t last_block; 1393 xfs_filblks_t unmap_len; 1394 int committed; 1395 int error = 0; 1396 int done = 0; 1397 1398 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1399 ASSERT(!atomic_read(&VFS_I(ip)->i_count) || 1400 xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1401 ASSERT(new_size <= XFS_ISIZE(ip)); 1402 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); 1403 ASSERT(ip->i_itemp != NULL); 1404 ASSERT(ip->i_itemp->ili_lock_flags == 0); 1405 ASSERT(!XFS_NOT_DQATTACHED(mp, ip)); 1406 1407 trace_xfs_itruncate_extents_start(ip, new_size); 1408 1409 /* 1410 * Since it is possible for space to become allocated beyond 1411 * the end of the file (in a crash where the space is allocated 1412 * but the inode size is not yet updated), simply remove any 1413 * blocks which show up between the new EOF and the maximum 1414 * possible file size. If the first block to be removed is 1415 * beyond the maximum file size (ie it is the same as last_block), 1416 * then there is nothing to do. 1417 */ 1418 first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size); 1419 last_block = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 1420 if (first_unmap_block == last_block) 1421 return 0; 1422 1423 ASSERT(first_unmap_block < last_block); 1424 unmap_len = last_block - first_unmap_block + 1; 1425 while (!done) { 1426 xfs_bmap_init(&free_list, &first_block); 1427 error = xfs_bunmapi(tp, ip, 1428 first_unmap_block, unmap_len, 1429 xfs_bmapi_aflag(whichfork), 1430 XFS_ITRUNC_MAX_EXTENTS, 1431 &first_block, &free_list, 1432 &done); 1433 if (error) 1434 goto out_bmap_cancel; 1435 1436 /* 1437 * Duplicate the transaction that has the permanent 1438 * reservation and commit the old transaction. 1439 */ 1440 error = xfs_bmap_finish(&tp, &free_list, &committed); 1441 if (committed) 1442 xfs_trans_ijoin(tp, ip, 0); 1443 if (error) 1444 goto out_bmap_cancel; 1445 1446 if (committed) { 1447 /* 1448 * Mark the inode dirty so it will be logged and 1449 * moved forward in the log as part of every commit. 1450 */ 1451 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1452 } 1453 1454 ntp = xfs_trans_dup(tp); 1455 error = xfs_trans_commit(tp, 0); 1456 tp = ntp; 1457 1458 xfs_trans_ijoin(tp, ip, 0); 1459 1460 if (error) 1461 goto out; 1462 1463 /* 1464 * Transaction commit worked ok so we can drop the extra ticket 1465 * reference that we gained in xfs_trans_dup() 1466 */ 1467 xfs_log_ticket_put(tp->t_ticket); 1468 error = xfs_trans_reserve(tp, 0, 1469 XFS_ITRUNCATE_LOG_RES(mp), 0, 1470 XFS_TRANS_PERM_LOG_RES, 1471 XFS_ITRUNCATE_LOG_COUNT); 1472 if (error) 1473 goto out; 1474 } 1475 1476 /* 1477 * Always re-log the inode so that our permanent transaction can keep 1478 * on rolling it forward in the log. 1479 */ 1480 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1481 1482 trace_xfs_itruncate_extents_end(ip, new_size); 1483 1484 out: 1485 *tpp = tp; 1486 return error; 1487 out_bmap_cancel: 1488 /* 1489 * If the bunmapi call encounters an error, return to the caller where 1490 * the transaction can be properly aborted. We just need to make sure 1491 * we're not holding any resources that we were not when we came in. 1492 */ 1493 xfs_bmap_cancel(&free_list); 1494 goto out; 1495 } 1496 1497 /* 1498 * This is called when the inode's link count goes to 0. 1499 * We place the on-disk inode on a list in the AGI. It 1500 * will be pulled from this list when the inode is freed. 1501 */ 1502 int 1503 xfs_iunlink( 1504 xfs_trans_t *tp, 1505 xfs_inode_t *ip) 1506 { 1507 xfs_mount_t *mp; 1508 xfs_agi_t *agi; 1509 xfs_dinode_t *dip; 1510 xfs_buf_t *agibp; 1511 xfs_buf_t *ibp; 1512 xfs_agino_t agino; 1513 short bucket_index; 1514 int offset; 1515 int error; 1516 1517 ASSERT(ip->i_d.di_nlink == 0); 1518 ASSERT(ip->i_d.di_mode != 0); 1519 1520 mp = tp->t_mountp; 1521 1522 /* 1523 * Get the agi buffer first. It ensures lock ordering 1524 * on the list. 1525 */ 1526 error = xfs_read_agi(mp, tp, XFS_INO_TO_AGNO(mp, ip->i_ino), &agibp); 1527 if (error) 1528 return error; 1529 agi = XFS_BUF_TO_AGI(agibp); 1530 1531 /* 1532 * Get the index into the agi hash table for the 1533 * list this inode will go on. 1534 */ 1535 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1536 ASSERT(agino != 0); 1537 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1538 ASSERT(agi->agi_unlinked[bucket_index]); 1539 ASSERT(be32_to_cpu(agi->agi_unlinked[bucket_index]) != agino); 1540 1541 if (agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)) { 1542 /* 1543 * There is already another inode in the bucket we need 1544 * to add ourselves to. Add us at the front of the list. 1545 * Here we put the head pointer into our next pointer, 1546 * and then we fall through to point the head at us. 1547 */ 1548 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1549 0, 0); 1550 if (error) 1551 return error; 1552 1553 ASSERT(dip->di_next_unlinked == cpu_to_be32(NULLAGINO)); 1554 dip->di_next_unlinked = agi->agi_unlinked[bucket_index]; 1555 offset = ip->i_imap.im_boffset + 1556 offsetof(xfs_dinode_t, di_next_unlinked); 1557 xfs_trans_inode_buf(tp, ibp); 1558 xfs_trans_log_buf(tp, ibp, offset, 1559 (offset + sizeof(xfs_agino_t) - 1)); 1560 xfs_inobp_check(mp, ibp); 1561 } 1562 1563 /* 1564 * Point the bucket head pointer at the inode being inserted. 1565 */ 1566 ASSERT(agino != 0); 1567 agi->agi_unlinked[bucket_index] = cpu_to_be32(agino); 1568 offset = offsetof(xfs_agi_t, agi_unlinked) + 1569 (sizeof(xfs_agino_t) * bucket_index); 1570 xfs_trans_log_buf(tp, agibp, offset, 1571 (offset + sizeof(xfs_agino_t) - 1)); 1572 return 0; 1573 } 1574 1575 /* 1576 * Pull the on-disk inode from the AGI unlinked list. 1577 */ 1578 STATIC int 1579 xfs_iunlink_remove( 1580 xfs_trans_t *tp, 1581 xfs_inode_t *ip) 1582 { 1583 xfs_ino_t next_ino; 1584 xfs_mount_t *mp; 1585 xfs_agi_t *agi; 1586 xfs_dinode_t *dip; 1587 xfs_buf_t *agibp; 1588 xfs_buf_t *ibp; 1589 xfs_agnumber_t agno; 1590 xfs_agino_t agino; 1591 xfs_agino_t next_agino; 1592 xfs_buf_t *last_ibp; 1593 xfs_dinode_t *last_dip = NULL; 1594 short bucket_index; 1595 int offset, last_offset = 0; 1596 int error; 1597 1598 mp = tp->t_mountp; 1599 agno = XFS_INO_TO_AGNO(mp, ip->i_ino); 1600 1601 /* 1602 * Get the agi buffer first. It ensures lock ordering 1603 * on the list. 1604 */ 1605 error = xfs_read_agi(mp, tp, agno, &agibp); 1606 if (error) 1607 return error; 1608 1609 agi = XFS_BUF_TO_AGI(agibp); 1610 1611 /* 1612 * Get the index into the agi hash table for the 1613 * list this inode will go on. 1614 */ 1615 agino = XFS_INO_TO_AGINO(mp, ip->i_ino); 1616 ASSERT(agino != 0); 1617 bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS; 1618 ASSERT(agi->agi_unlinked[bucket_index] != cpu_to_be32(NULLAGINO)); 1619 ASSERT(agi->agi_unlinked[bucket_index]); 1620 1621 if (be32_to_cpu(agi->agi_unlinked[bucket_index]) == agino) { 1622 /* 1623 * We're at the head of the list. Get the inode's on-disk 1624 * buffer to see if there is anyone after us on the list. 1625 * Only modify our next pointer if it is not already NULLAGINO. 1626 * This saves us the overhead of dealing with the buffer when 1627 * there is no need to change it. 1628 */ 1629 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1630 0, 0); 1631 if (error) { 1632 xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.", 1633 __func__, error); 1634 return error; 1635 } 1636 next_agino = be32_to_cpu(dip->di_next_unlinked); 1637 ASSERT(next_agino != 0); 1638 if (next_agino != NULLAGINO) { 1639 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 1640 offset = ip->i_imap.im_boffset + 1641 offsetof(xfs_dinode_t, di_next_unlinked); 1642 xfs_trans_inode_buf(tp, ibp); 1643 xfs_trans_log_buf(tp, ibp, offset, 1644 (offset + sizeof(xfs_agino_t) - 1)); 1645 xfs_inobp_check(mp, ibp); 1646 } else { 1647 xfs_trans_brelse(tp, ibp); 1648 } 1649 /* 1650 * Point the bucket head pointer at the next inode. 1651 */ 1652 ASSERT(next_agino != 0); 1653 ASSERT(next_agino != agino); 1654 agi->agi_unlinked[bucket_index] = cpu_to_be32(next_agino); 1655 offset = offsetof(xfs_agi_t, agi_unlinked) + 1656 (sizeof(xfs_agino_t) * bucket_index); 1657 xfs_trans_log_buf(tp, agibp, offset, 1658 (offset + sizeof(xfs_agino_t) - 1)); 1659 } else { 1660 /* 1661 * We need to search the list for the inode being freed. 1662 */ 1663 next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]); 1664 last_ibp = NULL; 1665 while (next_agino != agino) { 1666 struct xfs_imap imap; 1667 1668 if (last_ibp) 1669 xfs_trans_brelse(tp, last_ibp); 1670 1671 imap.im_blkno = 0; 1672 next_ino = XFS_AGINO_TO_INO(mp, agno, next_agino); 1673 1674 error = xfs_imap(mp, tp, next_ino, &imap, 0); 1675 if (error) { 1676 xfs_warn(mp, 1677 "%s: xfs_imap returned error %d.", 1678 __func__, error); 1679 return error; 1680 } 1681 1682 error = xfs_imap_to_bp(mp, tp, &imap, &last_dip, 1683 &last_ibp, 0, 0); 1684 if (error) { 1685 xfs_warn(mp, 1686 "%s: xfs_imap_to_bp returned error %d.", 1687 __func__, error); 1688 return error; 1689 } 1690 1691 last_offset = imap.im_boffset; 1692 next_agino = be32_to_cpu(last_dip->di_next_unlinked); 1693 ASSERT(next_agino != NULLAGINO); 1694 ASSERT(next_agino != 0); 1695 } 1696 1697 /* 1698 * Now last_ibp points to the buffer previous to us on the 1699 * unlinked list. Pull us from the list. 1700 */ 1701 error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &dip, &ibp, 1702 0, 0); 1703 if (error) { 1704 xfs_warn(mp, "%s: xfs_imap_to_bp(2) returned error %d.", 1705 __func__, error); 1706 return error; 1707 } 1708 next_agino = be32_to_cpu(dip->di_next_unlinked); 1709 ASSERT(next_agino != 0); 1710 ASSERT(next_agino != agino); 1711 if (next_agino != NULLAGINO) { 1712 dip->di_next_unlinked = cpu_to_be32(NULLAGINO); 1713 offset = ip->i_imap.im_boffset + 1714 offsetof(xfs_dinode_t, di_next_unlinked); 1715 xfs_trans_inode_buf(tp, ibp); 1716 xfs_trans_log_buf(tp, ibp, offset, 1717 (offset + sizeof(xfs_agino_t) - 1)); 1718 xfs_inobp_check(mp, ibp); 1719 } else { 1720 xfs_trans_brelse(tp, ibp); 1721 } 1722 /* 1723 * Point the previous inode on the list to the next inode. 1724 */ 1725 last_dip->di_next_unlinked = cpu_to_be32(next_agino); 1726 ASSERT(next_agino != 0); 1727 offset = last_offset + offsetof(xfs_dinode_t, di_next_unlinked); 1728 xfs_trans_inode_buf(tp, last_ibp); 1729 xfs_trans_log_buf(tp, last_ibp, offset, 1730 (offset + sizeof(xfs_agino_t) - 1)); 1731 xfs_inobp_check(mp, last_ibp); 1732 } 1733 return 0; 1734 } 1735 1736 /* 1737 * A big issue when freeing the inode cluster is is that we _cannot_ skip any 1738 * inodes that are in memory - they all must be marked stale and attached to 1739 * the cluster buffer. 1740 */ 1741 STATIC int 1742 xfs_ifree_cluster( 1743 xfs_inode_t *free_ip, 1744 xfs_trans_t *tp, 1745 xfs_ino_t inum) 1746 { 1747 xfs_mount_t *mp = free_ip->i_mount; 1748 int blks_per_cluster; 1749 int nbufs; 1750 int ninodes; 1751 int i, j; 1752 xfs_daddr_t blkno; 1753 xfs_buf_t *bp; 1754 xfs_inode_t *ip; 1755 xfs_inode_log_item_t *iip; 1756 xfs_log_item_t *lip; 1757 struct xfs_perag *pag; 1758 1759 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, inum)); 1760 if (mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp)) { 1761 blks_per_cluster = 1; 1762 ninodes = mp->m_sb.sb_inopblock; 1763 nbufs = XFS_IALLOC_BLOCKS(mp); 1764 } else { 1765 blks_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) / 1766 mp->m_sb.sb_blocksize; 1767 ninodes = blks_per_cluster * mp->m_sb.sb_inopblock; 1768 nbufs = XFS_IALLOC_BLOCKS(mp) / blks_per_cluster; 1769 } 1770 1771 for (j = 0; j < nbufs; j++, inum += ninodes) { 1772 blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum), 1773 XFS_INO_TO_AGBNO(mp, inum)); 1774 1775 /* 1776 * We obtain and lock the backing buffer first in the process 1777 * here, as we have to ensure that any dirty inode that we 1778 * can't get the flush lock on is attached to the buffer. 1779 * If we scan the in-memory inodes first, then buffer IO can 1780 * complete before we get a lock on it, and hence we may fail 1781 * to mark all the active inodes on the buffer stale. 1782 */ 1783 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno, 1784 mp->m_bsize * blks_per_cluster, 1785 XBF_UNMAPPED); 1786 1787 if (!bp) 1788 return ENOMEM; 1789 1790 /* 1791 * This buffer may not have been correctly initialised as we 1792 * didn't read it from disk. That's not important because we are 1793 * only using to mark the buffer as stale in the log, and to 1794 * attach stale cached inodes on it. That means it will never be 1795 * dispatched for IO. If it is, we want to know about it, and we 1796 * want it to fail. We can acheive this by adding a write 1797 * verifier to the buffer. 1798 */ 1799 bp->b_ops = &xfs_inode_buf_ops; 1800 1801 /* 1802 * Walk the inodes already attached to the buffer and mark them 1803 * stale. These will all have the flush locks held, so an 1804 * in-memory inode walk can't lock them. By marking them all 1805 * stale first, we will not attempt to lock them in the loop 1806 * below as the XFS_ISTALE flag will be set. 1807 */ 1808 lip = bp->b_fspriv; 1809 while (lip) { 1810 if (lip->li_type == XFS_LI_INODE) { 1811 iip = (xfs_inode_log_item_t *)lip; 1812 ASSERT(iip->ili_logged == 1); 1813 lip->li_cb = xfs_istale_done; 1814 xfs_trans_ail_copy_lsn(mp->m_ail, 1815 &iip->ili_flush_lsn, 1816 &iip->ili_item.li_lsn); 1817 xfs_iflags_set(iip->ili_inode, XFS_ISTALE); 1818 } 1819 lip = lip->li_bio_list; 1820 } 1821 1822 1823 /* 1824 * For each inode in memory attempt to add it to the inode 1825 * buffer and set it up for being staled on buffer IO 1826 * completion. This is safe as we've locked out tail pushing 1827 * and flushing by locking the buffer. 1828 * 1829 * We have already marked every inode that was part of a 1830 * transaction stale above, which means there is no point in 1831 * even trying to lock them. 1832 */ 1833 for (i = 0; i < ninodes; i++) { 1834 retry: 1835 rcu_read_lock(); 1836 ip = radix_tree_lookup(&pag->pag_ici_root, 1837 XFS_INO_TO_AGINO(mp, (inum + i))); 1838 1839 /* Inode not in memory, nothing to do */ 1840 if (!ip) { 1841 rcu_read_unlock(); 1842 continue; 1843 } 1844 1845 /* 1846 * because this is an RCU protected lookup, we could 1847 * find a recently freed or even reallocated inode 1848 * during the lookup. We need to check under the 1849 * i_flags_lock for a valid inode here. Skip it if it 1850 * is not valid, the wrong inode or stale. 1851 */ 1852 spin_lock(&ip->i_flags_lock); 1853 if (ip->i_ino != inum + i || 1854 __xfs_iflags_test(ip, XFS_ISTALE)) { 1855 spin_unlock(&ip->i_flags_lock); 1856 rcu_read_unlock(); 1857 continue; 1858 } 1859 spin_unlock(&ip->i_flags_lock); 1860 1861 /* 1862 * Don't try to lock/unlock the current inode, but we 1863 * _cannot_ skip the other inodes that we did not find 1864 * in the list attached to the buffer and are not 1865 * already marked stale. If we can't lock it, back off 1866 * and retry. 1867 */ 1868 if (ip != free_ip && 1869 !xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) { 1870 rcu_read_unlock(); 1871 delay(1); 1872 goto retry; 1873 } 1874 rcu_read_unlock(); 1875 1876 xfs_iflock(ip); 1877 xfs_iflags_set(ip, XFS_ISTALE); 1878 1879 /* 1880 * we don't need to attach clean inodes or those only 1881 * with unlogged changes (which we throw away, anyway). 1882 */ 1883 iip = ip->i_itemp; 1884 if (!iip || xfs_inode_clean(ip)) { 1885 ASSERT(ip != free_ip); 1886 xfs_ifunlock(ip); 1887 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1888 continue; 1889 } 1890 1891 iip->ili_last_fields = iip->ili_fields; 1892 iip->ili_fields = 0; 1893 iip->ili_logged = 1; 1894 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 1895 &iip->ili_item.li_lsn); 1896 1897 xfs_buf_attach_iodone(bp, xfs_istale_done, 1898 &iip->ili_item); 1899 1900 if (ip != free_ip) 1901 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1902 } 1903 1904 xfs_trans_stale_inode_buf(tp, bp); 1905 xfs_trans_binval(tp, bp); 1906 } 1907 1908 xfs_perag_put(pag); 1909 return 0; 1910 } 1911 1912 /* 1913 * This is called to return an inode to the inode free list. 1914 * The inode should already be truncated to 0 length and have 1915 * no pages associated with it. This routine also assumes that 1916 * the inode is already a part of the transaction. 1917 * 1918 * The on-disk copy of the inode will have been added to the list 1919 * of unlinked inodes in the AGI. We need to remove the inode from 1920 * that list atomically with respect to freeing it here. 1921 */ 1922 int 1923 xfs_ifree( 1924 xfs_trans_t *tp, 1925 xfs_inode_t *ip, 1926 xfs_bmap_free_t *flist) 1927 { 1928 int error; 1929 int delete; 1930 xfs_ino_t first_ino; 1931 xfs_dinode_t *dip; 1932 xfs_buf_t *ibp; 1933 1934 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 1935 ASSERT(ip->i_d.di_nlink == 0); 1936 ASSERT(ip->i_d.di_nextents == 0); 1937 ASSERT(ip->i_d.di_anextents == 0); 1938 ASSERT(ip->i_d.di_size == 0 || !S_ISREG(ip->i_d.di_mode)); 1939 ASSERT(ip->i_d.di_nblocks == 0); 1940 1941 /* 1942 * Pull the on-disk inode from the AGI unlinked list. 1943 */ 1944 error = xfs_iunlink_remove(tp, ip); 1945 if (error != 0) { 1946 return error; 1947 } 1948 1949 error = xfs_difree(tp, ip->i_ino, flist, &delete, &first_ino); 1950 if (error != 0) { 1951 return error; 1952 } 1953 ip->i_d.di_mode = 0; /* mark incore inode as free */ 1954 ip->i_d.di_flags = 0; 1955 ip->i_d.di_dmevmask = 0; 1956 ip->i_d.di_forkoff = 0; /* mark the attr fork not in use */ 1957 ip->i_d.di_format = XFS_DINODE_FMT_EXTENTS; 1958 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS; 1959 /* 1960 * Bump the generation count so no one will be confused 1961 * by reincarnations of this inode. 1962 */ 1963 ip->i_d.di_gen++; 1964 1965 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1966 1967 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &dip, &ibp, 1968 0, 0); 1969 if (error) 1970 return error; 1971 1972 /* 1973 * Clear the on-disk di_mode. This is to prevent xfs_bulkstat 1974 * from picking up this inode when it is reclaimed (its incore state 1975 * initialzed but not flushed to disk yet). The in-core di_mode is 1976 * already cleared and a corresponding transaction logged. 1977 * The hack here just synchronizes the in-core to on-disk 1978 * di_mode value in advance before the actual inode sync to disk. 1979 * This is OK because the inode is already unlinked and would never 1980 * change its di_mode again for this inode generation. 1981 * This is a temporary hack that would require a proper fix 1982 * in the future. 1983 */ 1984 dip->di_mode = 0; 1985 1986 if (delete) { 1987 error = xfs_ifree_cluster(ip, tp, first_ino); 1988 } 1989 1990 return error; 1991 } 1992 1993 /* 1994 * Reallocate the space for if_broot based on the number of records 1995 * being added or deleted as indicated in rec_diff. Move the records 1996 * and pointers in if_broot to fit the new size. When shrinking this 1997 * will eliminate holes between the records and pointers created by 1998 * the caller. When growing this will create holes to be filled in 1999 * by the caller. 2000 * 2001 * The caller must not request to add more records than would fit in 2002 * the on-disk inode root. If the if_broot is currently NULL, then 2003 * if we adding records one will be allocated. The caller must also 2004 * not request that the number of records go below zero, although 2005 * it can go to zero. 2006 * 2007 * ip -- the inode whose if_broot area is changing 2008 * ext_diff -- the change in the number of records, positive or negative, 2009 * requested for the if_broot array. 2010 */ 2011 void 2012 xfs_iroot_realloc( 2013 xfs_inode_t *ip, 2014 int rec_diff, 2015 int whichfork) 2016 { 2017 struct xfs_mount *mp = ip->i_mount; 2018 int cur_max; 2019 xfs_ifork_t *ifp; 2020 struct xfs_btree_block *new_broot; 2021 int new_max; 2022 size_t new_size; 2023 char *np; 2024 char *op; 2025 2026 /* 2027 * Handle the degenerate case quietly. 2028 */ 2029 if (rec_diff == 0) { 2030 return; 2031 } 2032 2033 ifp = XFS_IFORK_PTR(ip, whichfork); 2034 if (rec_diff > 0) { 2035 /* 2036 * If there wasn't any memory allocated before, just 2037 * allocate it now and get out. 2038 */ 2039 if (ifp->if_broot_bytes == 0) { 2040 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(rec_diff); 2041 ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); 2042 ifp->if_broot_bytes = (int)new_size; 2043 return; 2044 } 2045 2046 /* 2047 * If there is already an existing if_broot, then we need 2048 * to realloc() it and shift the pointers to their new 2049 * location. The records don't change location because 2050 * they are kept butted up against the btree block header. 2051 */ 2052 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); 2053 new_max = cur_max + rec_diff; 2054 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); 2055 ifp->if_broot = kmem_realloc(ifp->if_broot, new_size, 2056 (size_t)XFS_BMAP_BROOT_SPACE_CALC(cur_max), /* old size */ 2057 KM_SLEEP | KM_NOFS); 2058 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2059 ifp->if_broot_bytes); 2060 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2061 (int)new_size); 2062 ifp->if_broot_bytes = (int)new_size; 2063 ASSERT(ifp->if_broot_bytes <= 2064 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); 2065 memmove(np, op, cur_max * (uint)sizeof(xfs_dfsbno_t)); 2066 return; 2067 } 2068 2069 /* 2070 * rec_diff is less than 0. In this case, we are shrinking the 2071 * if_broot buffer. It must already exist. If we go to zero 2072 * records, just get rid of the root and clear the status bit. 2073 */ 2074 ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0)); 2075 cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0); 2076 new_max = cur_max + rec_diff; 2077 ASSERT(new_max >= 0); 2078 if (new_max > 0) 2079 new_size = (size_t)XFS_BMAP_BROOT_SPACE_CALC(new_max); 2080 else 2081 new_size = 0; 2082 if (new_size > 0) { 2083 new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS); 2084 /* 2085 * First copy over the btree block header. 2086 */ 2087 memcpy(new_broot, ifp->if_broot, XFS_BTREE_LBLOCK_LEN); 2088 } else { 2089 new_broot = NULL; 2090 ifp->if_flags &= ~XFS_IFBROOT; 2091 } 2092 2093 /* 2094 * Only copy the records and pointers if there are any. 2095 */ 2096 if (new_max > 0) { 2097 /* 2098 * First copy the records. 2099 */ 2100 op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1); 2101 np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1); 2102 memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t)); 2103 2104 /* 2105 * Then copy the pointers. 2106 */ 2107 op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1, 2108 ifp->if_broot_bytes); 2109 np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1, 2110 (int)new_size); 2111 memcpy(np, op, new_max * (uint)sizeof(xfs_dfsbno_t)); 2112 } 2113 kmem_free(ifp->if_broot); 2114 ifp->if_broot = new_broot; 2115 ifp->if_broot_bytes = (int)new_size; 2116 ASSERT(ifp->if_broot_bytes <= 2117 XFS_IFORK_SIZE(ip, whichfork) + XFS_BROOT_SIZE_ADJ); 2118 return; 2119 } 2120 2121 2122 /* 2123 * This is called when the amount of space needed for if_data 2124 * is increased or decreased. The change in size is indicated by 2125 * the number of bytes that need to be added or deleted in the 2126 * byte_diff parameter. 2127 * 2128 * If the amount of space needed has decreased below the size of the 2129 * inline buffer, then switch to using the inline buffer. Otherwise, 2130 * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer 2131 * to what is needed. 2132 * 2133 * ip -- the inode whose if_data area is changing 2134 * byte_diff -- the change in the number of bytes, positive or negative, 2135 * requested for the if_data array. 2136 */ 2137 void 2138 xfs_idata_realloc( 2139 xfs_inode_t *ip, 2140 int byte_diff, 2141 int whichfork) 2142 { 2143 xfs_ifork_t *ifp; 2144 int new_size; 2145 int real_size; 2146 2147 if (byte_diff == 0) { 2148 return; 2149 } 2150 2151 ifp = XFS_IFORK_PTR(ip, whichfork); 2152 new_size = (int)ifp->if_bytes + byte_diff; 2153 ASSERT(new_size >= 0); 2154 2155 if (new_size == 0) { 2156 if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2157 kmem_free(ifp->if_u1.if_data); 2158 } 2159 ifp->if_u1.if_data = NULL; 2160 real_size = 0; 2161 } else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) { 2162 /* 2163 * If the valid extents/data can fit in if_inline_ext/data, 2164 * copy them from the malloc'd vector and free it. 2165 */ 2166 if (ifp->if_u1.if_data == NULL) { 2167 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2168 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2169 ASSERT(ifp->if_real_bytes != 0); 2170 memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data, 2171 new_size); 2172 kmem_free(ifp->if_u1.if_data); 2173 ifp->if_u1.if_data = ifp->if_u2.if_inline_data; 2174 } 2175 real_size = 0; 2176 } else { 2177 /* 2178 * Stuck with malloc/realloc. 2179 * For inline data, the underlying buffer must be 2180 * a multiple of 4 bytes in size so that it can be 2181 * logged and stay on word boundaries. We enforce 2182 * that here. 2183 */ 2184 real_size = roundup(new_size, 4); 2185 if (ifp->if_u1.if_data == NULL) { 2186 ASSERT(ifp->if_real_bytes == 0); 2187 ifp->if_u1.if_data = kmem_alloc(real_size, 2188 KM_SLEEP | KM_NOFS); 2189 } else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) { 2190 /* 2191 * Only do the realloc if the underlying size 2192 * is really changing. 2193 */ 2194 if (ifp->if_real_bytes != real_size) { 2195 ifp->if_u1.if_data = 2196 kmem_realloc(ifp->if_u1.if_data, 2197 real_size, 2198 ifp->if_real_bytes, 2199 KM_SLEEP | KM_NOFS); 2200 } 2201 } else { 2202 ASSERT(ifp->if_real_bytes == 0); 2203 ifp->if_u1.if_data = kmem_alloc(real_size, 2204 KM_SLEEP | KM_NOFS); 2205 memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data, 2206 ifp->if_bytes); 2207 } 2208 } 2209 ifp->if_real_bytes = real_size; 2210 ifp->if_bytes = new_size; 2211 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2212 } 2213 2214 void 2215 xfs_idestroy_fork( 2216 xfs_inode_t *ip, 2217 int whichfork) 2218 { 2219 xfs_ifork_t *ifp; 2220 2221 ifp = XFS_IFORK_PTR(ip, whichfork); 2222 if (ifp->if_broot != NULL) { 2223 kmem_free(ifp->if_broot); 2224 ifp->if_broot = NULL; 2225 } 2226 2227 /* 2228 * If the format is local, then we can't have an extents 2229 * array so just look for an inline data array. If we're 2230 * not local then we may or may not have an extents list, 2231 * so check and free it up if we do. 2232 */ 2233 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) { 2234 if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) && 2235 (ifp->if_u1.if_data != NULL)) { 2236 ASSERT(ifp->if_real_bytes != 0); 2237 kmem_free(ifp->if_u1.if_data); 2238 ifp->if_u1.if_data = NULL; 2239 ifp->if_real_bytes = 0; 2240 } 2241 } else if ((ifp->if_flags & XFS_IFEXTENTS) && 2242 ((ifp->if_flags & XFS_IFEXTIREC) || 2243 ((ifp->if_u1.if_extents != NULL) && 2244 (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) { 2245 ASSERT(ifp->if_real_bytes != 0); 2246 xfs_iext_destroy(ifp); 2247 } 2248 ASSERT(ifp->if_u1.if_extents == NULL || 2249 ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext); 2250 ASSERT(ifp->if_real_bytes == 0); 2251 if (whichfork == XFS_ATTR_FORK) { 2252 kmem_zone_free(xfs_ifork_zone, ip->i_afp); 2253 ip->i_afp = NULL; 2254 } 2255 } 2256 2257 /* 2258 * This is called to unpin an inode. The caller must have the inode locked 2259 * in at least shared mode so that the buffer cannot be subsequently pinned 2260 * once someone is waiting for it to be unpinned. 2261 */ 2262 static void 2263 xfs_iunpin( 2264 struct xfs_inode *ip) 2265 { 2266 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2267 2268 trace_xfs_inode_unpin_nowait(ip, _RET_IP_); 2269 2270 /* Give the log a push to start the unpinning I/O */ 2271 xfs_log_force_lsn(ip->i_mount, ip->i_itemp->ili_last_lsn, 0); 2272 2273 } 2274 2275 static void 2276 __xfs_iunpin_wait( 2277 struct xfs_inode *ip) 2278 { 2279 wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT); 2280 DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT); 2281 2282 xfs_iunpin(ip); 2283 2284 do { 2285 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); 2286 if (xfs_ipincount(ip)) 2287 io_schedule(); 2288 } while (xfs_ipincount(ip)); 2289 finish_wait(wq, &wait.wait); 2290 } 2291 2292 void 2293 xfs_iunpin_wait( 2294 struct xfs_inode *ip) 2295 { 2296 if (xfs_ipincount(ip)) 2297 __xfs_iunpin_wait(ip); 2298 } 2299 2300 /* 2301 * xfs_iextents_copy() 2302 * 2303 * This is called to copy the REAL extents (as opposed to the delayed 2304 * allocation extents) from the inode into the given buffer. It 2305 * returns the number of bytes copied into the buffer. 2306 * 2307 * If there are no delayed allocation extents, then we can just 2308 * memcpy() the extents into the buffer. Otherwise, we need to 2309 * examine each extent in turn and skip those which are delayed. 2310 */ 2311 int 2312 xfs_iextents_copy( 2313 xfs_inode_t *ip, 2314 xfs_bmbt_rec_t *dp, 2315 int whichfork) 2316 { 2317 int copied; 2318 int i; 2319 xfs_ifork_t *ifp; 2320 int nrecs; 2321 xfs_fsblock_t start_block; 2322 2323 ifp = XFS_IFORK_PTR(ip, whichfork); 2324 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2325 ASSERT(ifp->if_bytes > 0); 2326 2327 nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 2328 XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork); 2329 ASSERT(nrecs > 0); 2330 2331 /* 2332 * There are some delayed allocation extents in the 2333 * inode, so copy the extents one at a time and skip 2334 * the delayed ones. There must be at least one 2335 * non-delayed extent. 2336 */ 2337 copied = 0; 2338 for (i = 0; i < nrecs; i++) { 2339 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i); 2340 start_block = xfs_bmbt_get_startblock(ep); 2341 if (isnullstartblock(start_block)) { 2342 /* 2343 * It's a delayed allocation extent, so skip it. 2344 */ 2345 continue; 2346 } 2347 2348 /* Translate to on disk format */ 2349 put_unaligned(cpu_to_be64(ep->l0), &dp->l0); 2350 put_unaligned(cpu_to_be64(ep->l1), &dp->l1); 2351 dp++; 2352 copied++; 2353 } 2354 ASSERT(copied != 0); 2355 xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip)); 2356 2357 return (copied * (uint)sizeof(xfs_bmbt_rec_t)); 2358 } 2359 2360 /* 2361 * Each of the following cases stores data into the same region 2362 * of the on-disk inode, so only one of them can be valid at 2363 * any given time. While it is possible to have conflicting formats 2364 * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is 2365 * in EXTENTS format, this can only happen when the fork has 2366 * changed formats after being modified but before being flushed. 2367 * In these cases, the format always takes precedence, because the 2368 * format indicates the current state of the fork. 2369 */ 2370 /*ARGSUSED*/ 2371 STATIC void 2372 xfs_iflush_fork( 2373 xfs_inode_t *ip, 2374 xfs_dinode_t *dip, 2375 xfs_inode_log_item_t *iip, 2376 int whichfork, 2377 xfs_buf_t *bp) 2378 { 2379 char *cp; 2380 xfs_ifork_t *ifp; 2381 xfs_mount_t *mp; 2382 static const short brootflag[2] = 2383 { XFS_ILOG_DBROOT, XFS_ILOG_ABROOT }; 2384 static const short dataflag[2] = 2385 { XFS_ILOG_DDATA, XFS_ILOG_ADATA }; 2386 static const short extflag[2] = 2387 { XFS_ILOG_DEXT, XFS_ILOG_AEXT }; 2388 2389 if (!iip) 2390 return; 2391 ifp = XFS_IFORK_PTR(ip, whichfork); 2392 /* 2393 * This can happen if we gave up in iformat in an error path, 2394 * for the attribute fork. 2395 */ 2396 if (!ifp) { 2397 ASSERT(whichfork == XFS_ATTR_FORK); 2398 return; 2399 } 2400 cp = XFS_DFORK_PTR(dip, whichfork); 2401 mp = ip->i_mount; 2402 switch (XFS_IFORK_FORMAT(ip, whichfork)) { 2403 case XFS_DINODE_FMT_LOCAL: 2404 if ((iip->ili_fields & dataflag[whichfork]) && 2405 (ifp->if_bytes > 0)) { 2406 ASSERT(ifp->if_u1.if_data != NULL); 2407 ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork)); 2408 memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes); 2409 } 2410 break; 2411 2412 case XFS_DINODE_FMT_EXTENTS: 2413 ASSERT((ifp->if_flags & XFS_IFEXTENTS) || 2414 !(iip->ili_fields & extflag[whichfork])); 2415 if ((iip->ili_fields & extflag[whichfork]) && 2416 (ifp->if_bytes > 0)) { 2417 ASSERT(xfs_iext_get_ext(ifp, 0)); 2418 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0); 2419 (void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp, 2420 whichfork); 2421 } 2422 break; 2423 2424 case XFS_DINODE_FMT_BTREE: 2425 if ((iip->ili_fields & brootflag[whichfork]) && 2426 (ifp->if_broot_bytes > 0)) { 2427 ASSERT(ifp->if_broot != NULL); 2428 ASSERT(ifp->if_broot_bytes <= 2429 (XFS_IFORK_SIZE(ip, whichfork) + 2430 XFS_BROOT_SIZE_ADJ)); 2431 xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes, 2432 (xfs_bmdr_block_t *)cp, 2433 XFS_DFORK_SIZE(dip, mp, whichfork)); 2434 } 2435 break; 2436 2437 case XFS_DINODE_FMT_DEV: 2438 if (iip->ili_fields & XFS_ILOG_DEV) { 2439 ASSERT(whichfork == XFS_DATA_FORK); 2440 xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev); 2441 } 2442 break; 2443 2444 case XFS_DINODE_FMT_UUID: 2445 if (iip->ili_fields & XFS_ILOG_UUID) { 2446 ASSERT(whichfork == XFS_DATA_FORK); 2447 memcpy(XFS_DFORK_DPTR(dip), 2448 &ip->i_df.if_u2.if_uuid, 2449 sizeof(uuid_t)); 2450 } 2451 break; 2452 2453 default: 2454 ASSERT(0); 2455 break; 2456 } 2457 } 2458 2459 STATIC int 2460 xfs_iflush_cluster( 2461 xfs_inode_t *ip, 2462 xfs_buf_t *bp) 2463 { 2464 xfs_mount_t *mp = ip->i_mount; 2465 struct xfs_perag *pag; 2466 unsigned long first_index, mask; 2467 unsigned long inodes_per_cluster; 2468 int ilist_size; 2469 xfs_inode_t **ilist; 2470 xfs_inode_t *iq; 2471 int nr_found; 2472 int clcount = 0; 2473 int bufwasdelwri; 2474 int i; 2475 2476 pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino)); 2477 2478 inodes_per_cluster = XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog; 2479 ilist_size = inodes_per_cluster * sizeof(xfs_inode_t *); 2480 ilist = kmem_alloc(ilist_size, KM_MAYFAIL|KM_NOFS); 2481 if (!ilist) 2482 goto out_put; 2483 2484 mask = ~(((XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog)) - 1); 2485 first_index = XFS_INO_TO_AGINO(mp, ip->i_ino) & mask; 2486 rcu_read_lock(); 2487 /* really need a gang lookup range call here */ 2488 nr_found = radix_tree_gang_lookup(&pag->pag_ici_root, (void**)ilist, 2489 first_index, inodes_per_cluster); 2490 if (nr_found == 0) 2491 goto out_free; 2492 2493 for (i = 0; i < nr_found; i++) { 2494 iq = ilist[i]; 2495 if (iq == ip) 2496 continue; 2497 2498 /* 2499 * because this is an RCU protected lookup, we could find a 2500 * recently freed or even reallocated inode during the lookup. 2501 * We need to check under the i_flags_lock for a valid inode 2502 * here. Skip it if it is not valid or the wrong inode. 2503 */ 2504 spin_lock(&ip->i_flags_lock); 2505 if (!ip->i_ino || 2506 (XFS_INO_TO_AGINO(mp, iq->i_ino) & mask) != first_index) { 2507 spin_unlock(&ip->i_flags_lock); 2508 continue; 2509 } 2510 spin_unlock(&ip->i_flags_lock); 2511 2512 /* 2513 * Do an un-protected check to see if the inode is dirty and 2514 * is a candidate for flushing. These checks will be repeated 2515 * later after the appropriate locks are acquired. 2516 */ 2517 if (xfs_inode_clean(iq) && xfs_ipincount(iq) == 0) 2518 continue; 2519 2520 /* 2521 * Try to get locks. If any are unavailable or it is pinned, 2522 * then this inode cannot be flushed and is skipped. 2523 */ 2524 2525 if (!xfs_ilock_nowait(iq, XFS_ILOCK_SHARED)) 2526 continue; 2527 if (!xfs_iflock_nowait(iq)) { 2528 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2529 continue; 2530 } 2531 if (xfs_ipincount(iq)) { 2532 xfs_ifunlock(iq); 2533 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2534 continue; 2535 } 2536 2537 /* 2538 * arriving here means that this inode can be flushed. First 2539 * re-check that it's dirty before flushing. 2540 */ 2541 if (!xfs_inode_clean(iq)) { 2542 int error; 2543 error = xfs_iflush_int(iq, bp); 2544 if (error) { 2545 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2546 goto cluster_corrupt_out; 2547 } 2548 clcount++; 2549 } else { 2550 xfs_ifunlock(iq); 2551 } 2552 xfs_iunlock(iq, XFS_ILOCK_SHARED); 2553 } 2554 2555 if (clcount) { 2556 XFS_STATS_INC(xs_icluster_flushcnt); 2557 XFS_STATS_ADD(xs_icluster_flushinode, clcount); 2558 } 2559 2560 out_free: 2561 rcu_read_unlock(); 2562 kmem_free(ilist); 2563 out_put: 2564 xfs_perag_put(pag); 2565 return 0; 2566 2567 2568 cluster_corrupt_out: 2569 /* 2570 * Corruption detected in the clustering loop. Invalidate the 2571 * inode buffer and shut down the filesystem. 2572 */ 2573 rcu_read_unlock(); 2574 /* 2575 * Clean up the buffer. If it was delwri, just release it -- 2576 * brelse can handle it with no problems. If not, shut down the 2577 * filesystem before releasing the buffer. 2578 */ 2579 bufwasdelwri = (bp->b_flags & _XBF_DELWRI_Q); 2580 if (bufwasdelwri) 2581 xfs_buf_relse(bp); 2582 2583 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 2584 2585 if (!bufwasdelwri) { 2586 /* 2587 * Just like incore_relse: if we have b_iodone functions, 2588 * mark the buffer as an error and call them. Otherwise 2589 * mark it as stale and brelse. 2590 */ 2591 if (bp->b_iodone) { 2592 XFS_BUF_UNDONE(bp); 2593 xfs_buf_stale(bp); 2594 xfs_buf_ioerror(bp, EIO); 2595 xfs_buf_ioend(bp, 0); 2596 } else { 2597 xfs_buf_stale(bp); 2598 xfs_buf_relse(bp); 2599 } 2600 } 2601 2602 /* 2603 * Unlocks the flush lock 2604 */ 2605 xfs_iflush_abort(iq, false); 2606 kmem_free(ilist); 2607 xfs_perag_put(pag); 2608 return XFS_ERROR(EFSCORRUPTED); 2609 } 2610 2611 /* 2612 * Flush dirty inode metadata into the backing buffer. 2613 * 2614 * The caller must have the inode lock and the inode flush lock held. The 2615 * inode lock will still be held upon return to the caller, and the inode 2616 * flush lock will be released after the inode has reached the disk. 2617 * 2618 * The caller must write out the buffer returned in *bpp and release it. 2619 */ 2620 int 2621 xfs_iflush( 2622 struct xfs_inode *ip, 2623 struct xfs_buf **bpp) 2624 { 2625 struct xfs_mount *mp = ip->i_mount; 2626 struct xfs_buf *bp; 2627 struct xfs_dinode *dip; 2628 int error; 2629 2630 XFS_STATS_INC(xs_iflush_count); 2631 2632 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2633 ASSERT(xfs_isiflocked(ip)); 2634 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 2635 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 2636 2637 *bpp = NULL; 2638 2639 xfs_iunpin_wait(ip); 2640 2641 /* 2642 * For stale inodes we cannot rely on the backing buffer remaining 2643 * stale in cache for the remaining life of the stale inode and so 2644 * xfs_imap_to_bp() below may give us a buffer that no longer contains 2645 * inodes below. We have to check this after ensuring the inode is 2646 * unpinned so that it is safe to reclaim the stale inode after the 2647 * flush call. 2648 */ 2649 if (xfs_iflags_test(ip, XFS_ISTALE)) { 2650 xfs_ifunlock(ip); 2651 return 0; 2652 } 2653 2654 /* 2655 * This may have been unpinned because the filesystem is shutting 2656 * down forcibly. If that's the case we must not write this inode 2657 * to disk, because the log record didn't make it to disk. 2658 * 2659 * We also have to remove the log item from the AIL in this case, 2660 * as we wait for an empty AIL as part of the unmount process. 2661 */ 2662 if (XFS_FORCED_SHUTDOWN(mp)) { 2663 error = XFS_ERROR(EIO); 2664 goto abort_out; 2665 } 2666 2667 /* 2668 * Get the buffer containing the on-disk inode. 2669 */ 2670 error = xfs_imap_to_bp(mp, NULL, &ip->i_imap, &dip, &bp, XBF_TRYLOCK, 2671 0); 2672 if (error || !bp) { 2673 xfs_ifunlock(ip); 2674 return error; 2675 } 2676 2677 /* 2678 * First flush out the inode that xfs_iflush was called with. 2679 */ 2680 error = xfs_iflush_int(ip, bp); 2681 if (error) 2682 goto corrupt_out; 2683 2684 /* 2685 * If the buffer is pinned then push on the log now so we won't 2686 * get stuck waiting in the write for too long. 2687 */ 2688 if (xfs_buf_ispinned(bp)) 2689 xfs_log_force(mp, 0); 2690 2691 /* 2692 * inode clustering: 2693 * see if other inodes can be gathered into this write 2694 */ 2695 error = xfs_iflush_cluster(ip, bp); 2696 if (error) 2697 goto cluster_corrupt_out; 2698 2699 *bpp = bp; 2700 return 0; 2701 2702 corrupt_out: 2703 xfs_buf_relse(bp); 2704 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); 2705 cluster_corrupt_out: 2706 error = XFS_ERROR(EFSCORRUPTED); 2707 abort_out: 2708 /* 2709 * Unlocks the flush lock 2710 */ 2711 xfs_iflush_abort(ip, false); 2712 return error; 2713 } 2714 2715 2716 STATIC int 2717 xfs_iflush_int( 2718 xfs_inode_t *ip, 2719 xfs_buf_t *bp) 2720 { 2721 xfs_inode_log_item_t *iip; 2722 xfs_dinode_t *dip; 2723 xfs_mount_t *mp; 2724 2725 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)); 2726 ASSERT(xfs_isiflocked(ip)); 2727 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || 2728 ip->i_d.di_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)); 2729 2730 iip = ip->i_itemp; 2731 mp = ip->i_mount; 2732 2733 /* set *dip = inode's place in the buffer */ 2734 dip = (xfs_dinode_t *)xfs_buf_offset(bp, ip->i_imap.im_boffset); 2735 2736 if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC), 2737 mp, XFS_ERRTAG_IFLUSH_1, XFS_RANDOM_IFLUSH_1)) { 2738 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2739 "%s: Bad inode %Lu magic number 0x%x, ptr 0x%p", 2740 __func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip); 2741 goto corrupt_out; 2742 } 2743 if (XFS_TEST_ERROR(ip->i_d.di_magic != XFS_DINODE_MAGIC, 2744 mp, XFS_ERRTAG_IFLUSH_2, XFS_RANDOM_IFLUSH_2)) { 2745 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2746 "%s: Bad inode %Lu, ptr 0x%p, magic number 0x%x", 2747 __func__, ip->i_ino, ip, ip->i_d.di_magic); 2748 goto corrupt_out; 2749 } 2750 if (S_ISREG(ip->i_d.di_mode)) { 2751 if (XFS_TEST_ERROR( 2752 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 2753 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE), 2754 mp, XFS_ERRTAG_IFLUSH_3, XFS_RANDOM_IFLUSH_3)) { 2755 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2756 "%s: Bad regular inode %Lu, ptr 0x%p", 2757 __func__, ip->i_ino, ip); 2758 goto corrupt_out; 2759 } 2760 } else if (S_ISDIR(ip->i_d.di_mode)) { 2761 if (XFS_TEST_ERROR( 2762 (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS) && 2763 (ip->i_d.di_format != XFS_DINODE_FMT_BTREE) && 2764 (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL), 2765 mp, XFS_ERRTAG_IFLUSH_4, XFS_RANDOM_IFLUSH_4)) { 2766 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2767 "%s: Bad directory inode %Lu, ptr 0x%p", 2768 __func__, ip->i_ino, ip); 2769 goto corrupt_out; 2770 } 2771 } 2772 if (XFS_TEST_ERROR(ip->i_d.di_nextents + ip->i_d.di_anextents > 2773 ip->i_d.di_nblocks, mp, XFS_ERRTAG_IFLUSH_5, 2774 XFS_RANDOM_IFLUSH_5)) { 2775 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2776 "%s: detected corrupt incore inode %Lu, " 2777 "total extents = %d, nblocks = %Ld, ptr 0x%p", 2778 __func__, ip->i_ino, 2779 ip->i_d.di_nextents + ip->i_d.di_anextents, 2780 ip->i_d.di_nblocks, ip); 2781 goto corrupt_out; 2782 } 2783 if (XFS_TEST_ERROR(ip->i_d.di_forkoff > mp->m_sb.sb_inodesize, 2784 mp, XFS_ERRTAG_IFLUSH_6, XFS_RANDOM_IFLUSH_6)) { 2785 xfs_alert_tag(mp, XFS_PTAG_IFLUSH, 2786 "%s: bad inode %Lu, forkoff 0x%x, ptr 0x%p", 2787 __func__, ip->i_ino, ip->i_d.di_forkoff, ip); 2788 goto corrupt_out; 2789 } 2790 /* 2791 * bump the flush iteration count, used to detect flushes which 2792 * postdate a log record during recovery. 2793 */ 2794 2795 ip->i_d.di_flushiter++; 2796 2797 /* 2798 * Copy the dirty parts of the inode into the on-disk 2799 * inode. We always copy out the core of the inode, 2800 * because if the inode is dirty at all the core must 2801 * be. 2802 */ 2803 xfs_dinode_to_disk(dip, &ip->i_d); 2804 2805 /* Wrap, we never let the log put out DI_MAX_FLUSH */ 2806 if (ip->i_d.di_flushiter == DI_MAX_FLUSH) 2807 ip->i_d.di_flushiter = 0; 2808 2809 /* 2810 * If this is really an old format inode and the superblock version 2811 * has not been updated to support only new format inodes, then 2812 * convert back to the old inode format. If the superblock version 2813 * has been updated, then make the conversion permanent. 2814 */ 2815 ASSERT(ip->i_d.di_version == 1 || xfs_sb_version_hasnlink(&mp->m_sb)); 2816 if (ip->i_d.di_version == 1) { 2817 if (!xfs_sb_version_hasnlink(&mp->m_sb)) { 2818 /* 2819 * Convert it back. 2820 */ 2821 ASSERT(ip->i_d.di_nlink <= XFS_MAXLINK_1); 2822 dip->di_onlink = cpu_to_be16(ip->i_d.di_nlink); 2823 } else { 2824 /* 2825 * The superblock version has already been bumped, 2826 * so just make the conversion to the new inode 2827 * format permanent. 2828 */ 2829 ip->i_d.di_version = 2; 2830 dip->di_version = 2; 2831 ip->i_d.di_onlink = 0; 2832 dip->di_onlink = 0; 2833 memset(&(ip->i_d.di_pad[0]), 0, sizeof(ip->i_d.di_pad)); 2834 memset(&(dip->di_pad[0]), 0, 2835 sizeof(dip->di_pad)); 2836 ASSERT(xfs_get_projid(ip) == 0); 2837 } 2838 } 2839 2840 xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK, bp); 2841 if (XFS_IFORK_Q(ip)) 2842 xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK, bp); 2843 xfs_inobp_check(mp, bp); 2844 2845 /* 2846 * We've recorded everything logged in the inode, so we'd like to clear 2847 * the ili_fields bits so we don't log and flush things unnecessarily. 2848 * However, we can't stop logging all this information until the data 2849 * we've copied into the disk buffer is written to disk. If we did we 2850 * might overwrite the copy of the inode in the log with all the data 2851 * after re-logging only part of it, and in the face of a crash we 2852 * wouldn't have all the data we need to recover. 2853 * 2854 * What we do is move the bits to the ili_last_fields field. When 2855 * logging the inode, these bits are moved back to the ili_fields field. 2856 * In the xfs_iflush_done() routine we clear ili_last_fields, since we 2857 * know that the information those bits represent is permanently on 2858 * disk. As long as the flush completes before the inode is logged 2859 * again, then both ili_fields and ili_last_fields will be cleared. 2860 * 2861 * We can play with the ili_fields bits here, because the inode lock 2862 * must be held exclusively in order to set bits there and the flush 2863 * lock protects the ili_last_fields bits. Set ili_logged so the flush 2864 * done routine can tell whether or not to look in the AIL. Also, store 2865 * the current LSN of the inode so that we can tell whether the item has 2866 * moved in the AIL from xfs_iflush_done(). In order to read the lsn we 2867 * need the AIL lock, because it is a 64 bit value that cannot be read 2868 * atomically. 2869 */ 2870 if (iip != NULL && iip->ili_fields != 0) { 2871 iip->ili_last_fields = iip->ili_fields; 2872 iip->ili_fields = 0; 2873 iip->ili_logged = 1; 2874 2875 xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn, 2876 &iip->ili_item.li_lsn); 2877 2878 /* 2879 * Attach the function xfs_iflush_done to the inode's 2880 * buffer. This will remove the inode from the AIL 2881 * and unlock the inode's flush lock when the inode is 2882 * completely written to disk. 2883 */ 2884 xfs_buf_attach_iodone(bp, xfs_iflush_done, &iip->ili_item); 2885 2886 ASSERT(bp->b_fspriv != NULL); 2887 ASSERT(bp->b_iodone != NULL); 2888 } else { 2889 /* 2890 * We're flushing an inode which is not in the AIL and has 2891 * not been logged. For this case we can immediately drop 2892 * the inode flush lock because we can avoid the whole 2893 * AIL state thing. It's OK to drop the flush lock now, 2894 * because we've already locked the buffer and to do anything 2895 * you really need both. 2896 */ 2897 if (iip != NULL) { 2898 ASSERT(iip->ili_logged == 0); 2899 ASSERT(iip->ili_last_fields == 0); 2900 ASSERT((iip->ili_item.li_flags & XFS_LI_IN_AIL) == 0); 2901 } 2902 xfs_ifunlock(ip); 2903 } 2904 2905 return 0; 2906 2907 corrupt_out: 2908 return XFS_ERROR(EFSCORRUPTED); 2909 } 2910 2911 /* 2912 * Return a pointer to the extent record at file index idx. 2913 */ 2914 xfs_bmbt_rec_host_t * 2915 xfs_iext_get_ext( 2916 xfs_ifork_t *ifp, /* inode fork pointer */ 2917 xfs_extnum_t idx) /* index of target extent */ 2918 { 2919 ASSERT(idx >= 0); 2920 ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); 2921 2922 if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) { 2923 return ifp->if_u1.if_ext_irec->er_extbuf; 2924 } else if (ifp->if_flags & XFS_IFEXTIREC) { 2925 xfs_ext_irec_t *erp; /* irec pointer */ 2926 int erp_idx = 0; /* irec index */ 2927 xfs_extnum_t page_idx = idx; /* ext index in target list */ 2928 2929 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 2930 return &erp->er_extbuf[page_idx]; 2931 } else if (ifp->if_bytes) { 2932 return &ifp->if_u1.if_extents[idx]; 2933 } else { 2934 return NULL; 2935 } 2936 } 2937 2938 /* 2939 * Insert new item(s) into the extent records for incore inode 2940 * fork 'ifp'. 'count' new items are inserted at index 'idx'. 2941 */ 2942 void 2943 xfs_iext_insert( 2944 xfs_inode_t *ip, /* incore inode pointer */ 2945 xfs_extnum_t idx, /* starting index of new items */ 2946 xfs_extnum_t count, /* number of inserted items */ 2947 xfs_bmbt_irec_t *new, /* items to insert */ 2948 int state) /* type of extent conversion */ 2949 { 2950 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; 2951 xfs_extnum_t i; /* extent record index */ 2952 2953 trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_); 2954 2955 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 2956 xfs_iext_add(ifp, idx, count); 2957 for (i = idx; i < idx + count; i++, new++) 2958 xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new); 2959 } 2960 2961 /* 2962 * This is called when the amount of space required for incore file 2963 * extents needs to be increased. The ext_diff parameter stores the 2964 * number of new extents being added and the idx parameter contains 2965 * the extent index where the new extents will be added. If the new 2966 * extents are being appended, then we just need to (re)allocate and 2967 * initialize the space. Otherwise, if the new extents are being 2968 * inserted into the middle of the existing entries, a bit more work 2969 * is required to make room for the new extents to be inserted. The 2970 * caller is responsible for filling in the new extent entries upon 2971 * return. 2972 */ 2973 void 2974 xfs_iext_add( 2975 xfs_ifork_t *ifp, /* inode fork pointer */ 2976 xfs_extnum_t idx, /* index to begin adding exts */ 2977 int ext_diff) /* number of extents to add */ 2978 { 2979 int byte_diff; /* new bytes being added */ 2980 int new_size; /* size of extents after adding */ 2981 xfs_extnum_t nextents; /* number of extents in file */ 2982 2983 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 2984 ASSERT((idx >= 0) && (idx <= nextents)); 2985 byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t); 2986 new_size = ifp->if_bytes + byte_diff; 2987 /* 2988 * If the new number of extents (nextents + ext_diff) 2989 * fits inside the inode, then continue to use the inline 2990 * extent buffer. 2991 */ 2992 if (nextents + ext_diff <= XFS_INLINE_EXTS) { 2993 if (idx < nextents) { 2994 memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff], 2995 &ifp->if_u2.if_inline_ext[idx], 2996 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 2997 memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff); 2998 } 2999 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 3000 ifp->if_real_bytes = 0; 3001 } 3002 /* 3003 * Otherwise use a linear (direct) extent list. 3004 * If the extents are currently inside the inode, 3005 * xfs_iext_realloc_direct will switch us from 3006 * inline to direct extent allocation mode. 3007 */ 3008 else if (nextents + ext_diff <= XFS_LINEAR_EXTS) { 3009 xfs_iext_realloc_direct(ifp, new_size); 3010 if (idx < nextents) { 3011 memmove(&ifp->if_u1.if_extents[idx + ext_diff], 3012 &ifp->if_u1.if_extents[idx], 3013 (nextents - idx) * sizeof(xfs_bmbt_rec_t)); 3014 memset(&ifp->if_u1.if_extents[idx], 0, byte_diff); 3015 } 3016 } 3017 /* Indirection array */ 3018 else { 3019 xfs_ext_irec_t *erp; 3020 int erp_idx = 0; 3021 int page_idx = idx; 3022 3023 ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS); 3024 if (ifp->if_flags & XFS_IFEXTIREC) { 3025 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1); 3026 } else { 3027 xfs_iext_irec_init(ifp); 3028 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3029 erp = ifp->if_u1.if_ext_irec; 3030 } 3031 /* Extents fit in target extent page */ 3032 if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) { 3033 if (page_idx < erp->er_extcount) { 3034 memmove(&erp->er_extbuf[page_idx + ext_diff], 3035 &erp->er_extbuf[page_idx], 3036 (erp->er_extcount - page_idx) * 3037 sizeof(xfs_bmbt_rec_t)); 3038 memset(&erp->er_extbuf[page_idx], 0, byte_diff); 3039 } 3040 erp->er_extcount += ext_diff; 3041 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3042 } 3043 /* Insert a new extent page */ 3044 else if (erp) { 3045 xfs_iext_add_indirect_multi(ifp, 3046 erp_idx, page_idx, ext_diff); 3047 } 3048 /* 3049 * If extent(s) are being appended to the last page in 3050 * the indirection array and the new extent(s) don't fit 3051 * in the page, then erp is NULL and erp_idx is set to 3052 * the next index needed in the indirection array. 3053 */ 3054 else { 3055 int count = ext_diff; 3056 3057 while (count) { 3058 erp = xfs_iext_irec_new(ifp, erp_idx); 3059 erp->er_extcount = count; 3060 count -= MIN(count, (int)XFS_LINEAR_EXTS); 3061 if (count) { 3062 erp_idx++; 3063 } 3064 } 3065 } 3066 } 3067 ifp->if_bytes = new_size; 3068 } 3069 3070 /* 3071 * This is called when incore extents are being added to the indirection 3072 * array and the new extents do not fit in the target extent list. The 3073 * erp_idx parameter contains the irec index for the target extent list 3074 * in the indirection array, and the idx parameter contains the extent 3075 * index within the list. The number of extents being added is stored 3076 * in the count parameter. 3077 * 3078 * |-------| |-------| 3079 * | | | | idx - number of extents before idx 3080 * | idx | | count | 3081 * | | | | count - number of extents being inserted at idx 3082 * |-------| |-------| 3083 * | count | | nex2 | nex2 - number of extents after idx + count 3084 * |-------| |-------| 3085 */ 3086 void 3087 xfs_iext_add_indirect_multi( 3088 xfs_ifork_t *ifp, /* inode fork pointer */ 3089 int erp_idx, /* target extent irec index */ 3090 xfs_extnum_t idx, /* index within target list */ 3091 int count) /* new extents being added */ 3092 { 3093 int byte_diff; /* new bytes being added */ 3094 xfs_ext_irec_t *erp; /* pointer to irec entry */ 3095 xfs_extnum_t ext_diff; /* number of extents to add */ 3096 xfs_extnum_t ext_cnt; /* new extents still needed */ 3097 xfs_extnum_t nex2; /* extents after idx + count */ 3098 xfs_bmbt_rec_t *nex2_ep = NULL; /* temp list for nex2 extents */ 3099 int nlists; /* number of irec's (lists) */ 3100 3101 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3102 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3103 nex2 = erp->er_extcount - idx; 3104 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3105 3106 /* 3107 * Save second part of target extent list 3108 * (all extents past */ 3109 if (nex2) { 3110 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3111 nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS); 3112 memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff); 3113 erp->er_extcount -= nex2; 3114 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2); 3115 memset(&erp->er_extbuf[idx], 0, byte_diff); 3116 } 3117 3118 /* 3119 * Add the new extents to the end of the target 3120 * list, then allocate new irec record(s) and 3121 * extent buffer(s) as needed to store the rest 3122 * of the new extents. 3123 */ 3124 ext_cnt = count; 3125 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount); 3126 if (ext_diff) { 3127 erp->er_extcount += ext_diff; 3128 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3129 ext_cnt -= ext_diff; 3130 } 3131 while (ext_cnt) { 3132 erp_idx++; 3133 erp = xfs_iext_irec_new(ifp, erp_idx); 3134 ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS); 3135 erp->er_extcount = ext_diff; 3136 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff); 3137 ext_cnt -= ext_diff; 3138 } 3139 3140 /* Add nex2 extents back to indirection array */ 3141 if (nex2) { 3142 xfs_extnum_t ext_avail; 3143 int i; 3144 3145 byte_diff = nex2 * sizeof(xfs_bmbt_rec_t); 3146 ext_avail = XFS_LINEAR_EXTS - erp->er_extcount; 3147 i = 0; 3148 /* 3149 * If nex2 extents fit in the current page, append 3150 * nex2_ep after the new extents. 3151 */ 3152 if (nex2 <= ext_avail) { 3153 i = erp->er_extcount; 3154 } 3155 /* 3156 * Otherwise, check if space is available in the 3157 * next page. 3158 */ 3159 else if ((erp_idx < nlists - 1) && 3160 (nex2 <= (ext_avail = XFS_LINEAR_EXTS - 3161 ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) { 3162 erp_idx++; 3163 erp++; 3164 /* Create a hole for nex2 extents */ 3165 memmove(&erp->er_extbuf[nex2], erp->er_extbuf, 3166 erp->er_extcount * sizeof(xfs_bmbt_rec_t)); 3167 } 3168 /* 3169 * Final choice, create a new extent page for 3170 * nex2 extents. 3171 */ 3172 else { 3173 erp_idx++; 3174 erp = xfs_iext_irec_new(ifp, erp_idx); 3175 } 3176 memmove(&erp->er_extbuf[i], nex2_ep, byte_diff); 3177 kmem_free(nex2_ep); 3178 erp->er_extcount += nex2; 3179 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2); 3180 } 3181 } 3182 3183 /* 3184 * This is called when the amount of space required for incore file 3185 * extents needs to be decreased. The ext_diff parameter stores the 3186 * number of extents to be removed and the idx parameter contains 3187 * the extent index where the extents will be removed from. 3188 * 3189 * If the amount of space needed has decreased below the linear 3190 * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous 3191 * extent array. Otherwise, use kmem_realloc() to adjust the 3192 * size to what is needed. 3193 */ 3194 void 3195 xfs_iext_remove( 3196 xfs_inode_t *ip, /* incore inode pointer */ 3197 xfs_extnum_t idx, /* index to begin removing exts */ 3198 int ext_diff, /* number of extents to remove */ 3199 int state) /* type of extent conversion */ 3200 { 3201 xfs_ifork_t *ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df; 3202 xfs_extnum_t nextents; /* number of extents in file */ 3203 int new_size; /* size of extents after removal */ 3204 3205 trace_xfs_iext_remove(ip, idx, state, _RET_IP_); 3206 3207 ASSERT(ext_diff > 0); 3208 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3209 new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t); 3210 3211 if (new_size == 0) { 3212 xfs_iext_destroy(ifp); 3213 } else if (ifp->if_flags & XFS_IFEXTIREC) { 3214 xfs_iext_remove_indirect(ifp, idx, ext_diff); 3215 } else if (ifp->if_real_bytes) { 3216 xfs_iext_remove_direct(ifp, idx, ext_diff); 3217 } else { 3218 xfs_iext_remove_inline(ifp, idx, ext_diff); 3219 } 3220 ifp->if_bytes = new_size; 3221 } 3222 3223 /* 3224 * This removes ext_diff extents from the inline buffer, beginning 3225 * at extent index idx. 3226 */ 3227 void 3228 xfs_iext_remove_inline( 3229 xfs_ifork_t *ifp, /* inode fork pointer */ 3230 xfs_extnum_t idx, /* index to begin removing exts */ 3231 int ext_diff) /* number of extents to remove */ 3232 { 3233 int nextents; /* number of extents in file */ 3234 3235 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3236 ASSERT(idx < XFS_INLINE_EXTS); 3237 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3238 ASSERT(((nextents - ext_diff) > 0) && 3239 (nextents - ext_diff) < XFS_INLINE_EXTS); 3240 3241 if (idx + ext_diff < nextents) { 3242 memmove(&ifp->if_u2.if_inline_ext[idx], 3243 &ifp->if_u2.if_inline_ext[idx + ext_diff], 3244 (nextents - (idx + ext_diff)) * 3245 sizeof(xfs_bmbt_rec_t)); 3246 memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff], 3247 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 3248 } else { 3249 memset(&ifp->if_u2.if_inline_ext[idx], 0, 3250 ext_diff * sizeof(xfs_bmbt_rec_t)); 3251 } 3252 } 3253 3254 /* 3255 * This removes ext_diff extents from a linear (direct) extent list, 3256 * beginning at extent index idx. If the extents are being removed 3257 * from the end of the list (ie. truncate) then we just need to re- 3258 * allocate the list to remove the extra space. Otherwise, if the 3259 * extents are being removed from the middle of the existing extent 3260 * entries, then we first need to move the extent records beginning 3261 * at idx + ext_diff up in the list to overwrite the records being 3262 * removed, then remove the extra space via kmem_realloc. 3263 */ 3264 void 3265 xfs_iext_remove_direct( 3266 xfs_ifork_t *ifp, /* inode fork pointer */ 3267 xfs_extnum_t idx, /* index to begin removing exts */ 3268 int ext_diff) /* number of extents to remove */ 3269 { 3270 xfs_extnum_t nextents; /* number of extents in file */ 3271 int new_size; /* size of extents after removal */ 3272 3273 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3274 new_size = ifp->if_bytes - 3275 (ext_diff * sizeof(xfs_bmbt_rec_t)); 3276 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3277 3278 if (new_size == 0) { 3279 xfs_iext_destroy(ifp); 3280 return; 3281 } 3282 /* Move extents up in the list (if needed) */ 3283 if (idx + ext_diff < nextents) { 3284 memmove(&ifp->if_u1.if_extents[idx], 3285 &ifp->if_u1.if_extents[idx + ext_diff], 3286 (nextents - (idx + ext_diff)) * 3287 sizeof(xfs_bmbt_rec_t)); 3288 } 3289 memset(&ifp->if_u1.if_extents[nextents - ext_diff], 3290 0, ext_diff * sizeof(xfs_bmbt_rec_t)); 3291 /* 3292 * Reallocate the direct extent list. If the extents 3293 * will fit inside the inode then xfs_iext_realloc_direct 3294 * will switch from direct to inline extent allocation 3295 * mode for us. 3296 */ 3297 xfs_iext_realloc_direct(ifp, new_size); 3298 ifp->if_bytes = new_size; 3299 } 3300 3301 /* 3302 * This is called when incore extents are being removed from the 3303 * indirection array and the extents being removed span multiple extent 3304 * buffers. The idx parameter contains the file extent index where we 3305 * want to begin removing extents, and the count parameter contains 3306 * how many extents need to be removed. 3307 * 3308 * |-------| |-------| 3309 * | nex1 | | | nex1 - number of extents before idx 3310 * |-------| | count | 3311 * | | | | count - number of extents being removed at idx 3312 * | count | |-------| 3313 * | | | nex2 | nex2 - number of extents after idx + count 3314 * |-------| |-------| 3315 */ 3316 void 3317 xfs_iext_remove_indirect( 3318 xfs_ifork_t *ifp, /* inode fork pointer */ 3319 xfs_extnum_t idx, /* index to begin removing extents */ 3320 int count) /* number of extents to remove */ 3321 { 3322 xfs_ext_irec_t *erp; /* indirection array pointer */ 3323 int erp_idx = 0; /* indirection array index */ 3324 xfs_extnum_t ext_cnt; /* extents left to remove */ 3325 xfs_extnum_t ext_diff; /* extents to remove in current list */ 3326 xfs_extnum_t nex1; /* number of extents before idx */ 3327 xfs_extnum_t nex2; /* extents after idx + count */ 3328 int page_idx = idx; /* index in target extent list */ 3329 3330 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3331 erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0); 3332 ASSERT(erp != NULL); 3333 nex1 = page_idx; 3334 ext_cnt = count; 3335 while (ext_cnt) { 3336 nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0); 3337 ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1)); 3338 /* 3339 * Check for deletion of entire list; 3340 * xfs_iext_irec_remove() updates extent offsets. 3341 */ 3342 if (ext_diff == erp->er_extcount) { 3343 xfs_iext_irec_remove(ifp, erp_idx); 3344 ext_cnt -= ext_diff; 3345 nex1 = 0; 3346 if (ext_cnt) { 3347 ASSERT(erp_idx < ifp->if_real_bytes / 3348 XFS_IEXT_BUFSZ); 3349 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3350 nex1 = 0; 3351 continue; 3352 } else { 3353 break; 3354 } 3355 } 3356 /* Move extents up (if needed) */ 3357 if (nex2) { 3358 memmove(&erp->er_extbuf[nex1], 3359 &erp->er_extbuf[nex1 + ext_diff], 3360 nex2 * sizeof(xfs_bmbt_rec_t)); 3361 } 3362 /* Zero out rest of page */ 3363 memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ - 3364 ((nex1 + nex2) * sizeof(xfs_bmbt_rec_t)))); 3365 /* Update remaining counters */ 3366 erp->er_extcount -= ext_diff; 3367 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff); 3368 ext_cnt -= ext_diff; 3369 nex1 = 0; 3370 erp_idx++; 3371 erp++; 3372 } 3373 ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t); 3374 xfs_iext_irec_compact(ifp); 3375 } 3376 3377 /* 3378 * Create, destroy, or resize a linear (direct) block of extents. 3379 */ 3380 void 3381 xfs_iext_realloc_direct( 3382 xfs_ifork_t *ifp, /* inode fork pointer */ 3383 int new_size) /* new size of extents */ 3384 { 3385 int rnew_size; /* real new size of extents */ 3386 3387 rnew_size = new_size; 3388 3389 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) || 3390 ((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) && 3391 (new_size != ifp->if_real_bytes))); 3392 3393 /* Free extent records */ 3394 if (new_size == 0) { 3395 xfs_iext_destroy(ifp); 3396 } 3397 /* Resize direct extent list and zero any new bytes */ 3398 else if (ifp->if_real_bytes) { 3399 /* Check if extents will fit inside the inode */ 3400 if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) { 3401 xfs_iext_direct_to_inline(ifp, new_size / 3402 (uint)sizeof(xfs_bmbt_rec_t)); 3403 ifp->if_bytes = new_size; 3404 return; 3405 } 3406 if (!is_power_of_2(new_size)){ 3407 rnew_size = roundup_pow_of_two(new_size); 3408 } 3409 if (rnew_size != ifp->if_real_bytes) { 3410 ifp->if_u1.if_extents = 3411 kmem_realloc(ifp->if_u1.if_extents, 3412 rnew_size, 3413 ifp->if_real_bytes, KM_NOFS); 3414 } 3415 if (rnew_size > ifp->if_real_bytes) { 3416 memset(&ifp->if_u1.if_extents[ifp->if_bytes / 3417 (uint)sizeof(xfs_bmbt_rec_t)], 0, 3418 rnew_size - ifp->if_real_bytes); 3419 } 3420 } 3421 /* 3422 * Switch from the inline extent buffer to a direct 3423 * extent list. Be sure to include the inline extent 3424 * bytes in new_size. 3425 */ 3426 else { 3427 new_size += ifp->if_bytes; 3428 if (!is_power_of_2(new_size)) { 3429 rnew_size = roundup_pow_of_two(new_size); 3430 } 3431 xfs_iext_inline_to_direct(ifp, rnew_size); 3432 } 3433 ifp->if_real_bytes = rnew_size; 3434 ifp->if_bytes = new_size; 3435 } 3436 3437 /* 3438 * Switch from linear (direct) extent records to inline buffer. 3439 */ 3440 void 3441 xfs_iext_direct_to_inline( 3442 xfs_ifork_t *ifp, /* inode fork pointer */ 3443 xfs_extnum_t nextents) /* number of extents in file */ 3444 { 3445 ASSERT(ifp->if_flags & XFS_IFEXTENTS); 3446 ASSERT(nextents <= XFS_INLINE_EXTS); 3447 /* 3448 * The inline buffer was zeroed when we switched 3449 * from inline to direct extent allocation mode, 3450 * so we don't need to clear it here. 3451 */ 3452 memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents, 3453 nextents * sizeof(xfs_bmbt_rec_t)); 3454 kmem_free(ifp->if_u1.if_extents); 3455 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 3456 ifp->if_real_bytes = 0; 3457 } 3458 3459 /* 3460 * Switch from inline buffer to linear (direct) extent records. 3461 * new_size should already be rounded up to the next power of 2 3462 * by the caller (when appropriate), so use new_size as it is. 3463 * However, since new_size may be rounded up, we can't update 3464 * if_bytes here. It is the caller's responsibility to update 3465 * if_bytes upon return. 3466 */ 3467 void 3468 xfs_iext_inline_to_direct( 3469 xfs_ifork_t *ifp, /* inode fork pointer */ 3470 int new_size) /* number of extents in file */ 3471 { 3472 ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS); 3473 memset(ifp->if_u1.if_extents, 0, new_size); 3474 if (ifp->if_bytes) { 3475 memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext, 3476 ifp->if_bytes); 3477 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 3478 sizeof(xfs_bmbt_rec_t)); 3479 } 3480 ifp->if_real_bytes = new_size; 3481 } 3482 3483 /* 3484 * Resize an extent indirection array to new_size bytes. 3485 */ 3486 STATIC void 3487 xfs_iext_realloc_indirect( 3488 xfs_ifork_t *ifp, /* inode fork pointer */ 3489 int new_size) /* new indirection array size */ 3490 { 3491 int nlists; /* number of irec's (ex lists) */ 3492 int size; /* current indirection array size */ 3493 3494 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3495 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3496 size = nlists * sizeof(xfs_ext_irec_t); 3497 ASSERT(ifp->if_real_bytes); 3498 ASSERT((new_size >= 0) && (new_size != size)); 3499 if (new_size == 0) { 3500 xfs_iext_destroy(ifp); 3501 } else { 3502 ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *) 3503 kmem_realloc(ifp->if_u1.if_ext_irec, 3504 new_size, size, KM_NOFS); 3505 } 3506 } 3507 3508 /* 3509 * Switch from indirection array to linear (direct) extent allocations. 3510 */ 3511 STATIC void 3512 xfs_iext_indirect_to_direct( 3513 xfs_ifork_t *ifp) /* inode fork pointer */ 3514 { 3515 xfs_bmbt_rec_host_t *ep; /* extent record pointer */ 3516 xfs_extnum_t nextents; /* number of extents in file */ 3517 int size; /* size of file extents */ 3518 3519 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3520 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3521 ASSERT(nextents <= XFS_LINEAR_EXTS); 3522 size = nextents * sizeof(xfs_bmbt_rec_t); 3523 3524 xfs_iext_irec_compact_pages(ifp); 3525 ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ); 3526 3527 ep = ifp->if_u1.if_ext_irec->er_extbuf; 3528 kmem_free(ifp->if_u1.if_ext_irec); 3529 ifp->if_flags &= ~XFS_IFEXTIREC; 3530 ifp->if_u1.if_extents = ep; 3531 ifp->if_bytes = size; 3532 if (nextents < XFS_LINEAR_EXTS) { 3533 xfs_iext_realloc_direct(ifp, size); 3534 } 3535 } 3536 3537 /* 3538 * Free incore file extents. 3539 */ 3540 void 3541 xfs_iext_destroy( 3542 xfs_ifork_t *ifp) /* inode fork pointer */ 3543 { 3544 if (ifp->if_flags & XFS_IFEXTIREC) { 3545 int erp_idx; 3546 int nlists; 3547 3548 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3549 for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) { 3550 xfs_iext_irec_remove(ifp, erp_idx); 3551 } 3552 ifp->if_flags &= ~XFS_IFEXTIREC; 3553 } else if (ifp->if_real_bytes) { 3554 kmem_free(ifp->if_u1.if_extents); 3555 } else if (ifp->if_bytes) { 3556 memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS * 3557 sizeof(xfs_bmbt_rec_t)); 3558 } 3559 ifp->if_u1.if_extents = NULL; 3560 ifp->if_real_bytes = 0; 3561 ifp->if_bytes = 0; 3562 } 3563 3564 /* 3565 * Return a pointer to the extent record for file system block bno. 3566 */ 3567 xfs_bmbt_rec_host_t * /* pointer to found extent record */ 3568 xfs_iext_bno_to_ext( 3569 xfs_ifork_t *ifp, /* inode fork pointer */ 3570 xfs_fileoff_t bno, /* block number to search for */ 3571 xfs_extnum_t *idxp) /* index of target extent */ 3572 { 3573 xfs_bmbt_rec_host_t *base; /* pointer to first extent */ 3574 xfs_filblks_t blockcount = 0; /* number of blocks in extent */ 3575 xfs_bmbt_rec_host_t *ep = NULL; /* pointer to target extent */ 3576 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 3577 int high; /* upper boundary in search */ 3578 xfs_extnum_t idx = 0; /* index of target extent */ 3579 int low; /* lower boundary in search */ 3580 xfs_extnum_t nextents; /* number of file extents */ 3581 xfs_fileoff_t startoff = 0; /* start offset of extent */ 3582 3583 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3584 if (nextents == 0) { 3585 *idxp = 0; 3586 return NULL; 3587 } 3588 low = 0; 3589 if (ifp->if_flags & XFS_IFEXTIREC) { 3590 /* Find target extent list */ 3591 int erp_idx = 0; 3592 erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx); 3593 base = erp->er_extbuf; 3594 high = erp->er_extcount - 1; 3595 } else { 3596 base = ifp->if_u1.if_extents; 3597 high = nextents - 1; 3598 } 3599 /* Binary search extent records */ 3600 while (low <= high) { 3601 idx = (low + high) >> 1; 3602 ep = base + idx; 3603 startoff = xfs_bmbt_get_startoff(ep); 3604 blockcount = xfs_bmbt_get_blockcount(ep); 3605 if (bno < startoff) { 3606 high = idx - 1; 3607 } else if (bno >= startoff + blockcount) { 3608 low = idx + 1; 3609 } else { 3610 /* Convert back to file-based extent index */ 3611 if (ifp->if_flags & XFS_IFEXTIREC) { 3612 idx += erp->er_extoff; 3613 } 3614 *idxp = idx; 3615 return ep; 3616 } 3617 } 3618 /* Convert back to file-based extent index */ 3619 if (ifp->if_flags & XFS_IFEXTIREC) { 3620 idx += erp->er_extoff; 3621 } 3622 if (bno >= startoff + blockcount) { 3623 if (++idx == nextents) { 3624 ep = NULL; 3625 } else { 3626 ep = xfs_iext_get_ext(ifp, idx); 3627 } 3628 } 3629 *idxp = idx; 3630 return ep; 3631 } 3632 3633 /* 3634 * Return a pointer to the indirection array entry containing the 3635 * extent record for filesystem block bno. Store the index of the 3636 * target irec in *erp_idxp. 3637 */ 3638 xfs_ext_irec_t * /* pointer to found extent record */ 3639 xfs_iext_bno_to_irec( 3640 xfs_ifork_t *ifp, /* inode fork pointer */ 3641 xfs_fileoff_t bno, /* block number to search for */ 3642 int *erp_idxp) /* irec index of target ext list */ 3643 { 3644 xfs_ext_irec_t *erp = NULL; /* indirection array pointer */ 3645 xfs_ext_irec_t *erp_next; /* next indirection array entry */ 3646 int erp_idx; /* indirection array index */ 3647 int nlists; /* number of extent irec's (lists) */ 3648 int high; /* binary search upper limit */ 3649 int low; /* binary search lower limit */ 3650 3651 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3652 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3653 erp_idx = 0; 3654 low = 0; 3655 high = nlists - 1; 3656 while (low <= high) { 3657 erp_idx = (low + high) >> 1; 3658 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3659 erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL; 3660 if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) { 3661 high = erp_idx - 1; 3662 } else if (erp_next && bno >= 3663 xfs_bmbt_get_startoff(erp_next->er_extbuf)) { 3664 low = erp_idx + 1; 3665 } else { 3666 break; 3667 } 3668 } 3669 *erp_idxp = erp_idx; 3670 return erp; 3671 } 3672 3673 /* 3674 * Return a pointer to the indirection array entry containing the 3675 * extent record at file extent index *idxp. Store the index of the 3676 * target irec in *erp_idxp and store the page index of the target 3677 * extent record in *idxp. 3678 */ 3679 xfs_ext_irec_t * 3680 xfs_iext_idx_to_irec( 3681 xfs_ifork_t *ifp, /* inode fork pointer */ 3682 xfs_extnum_t *idxp, /* extent index (file -> page) */ 3683 int *erp_idxp, /* pointer to target irec */ 3684 int realloc) /* new bytes were just added */ 3685 { 3686 xfs_ext_irec_t *prev; /* pointer to previous irec */ 3687 xfs_ext_irec_t *erp = NULL; /* pointer to current irec */ 3688 int erp_idx; /* indirection array index */ 3689 int nlists; /* number of irec's (ex lists) */ 3690 int high; /* binary search upper limit */ 3691 int low; /* binary search lower limit */ 3692 xfs_extnum_t page_idx = *idxp; /* extent index in target list */ 3693 3694 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3695 ASSERT(page_idx >= 0); 3696 ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t)); 3697 ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc); 3698 3699 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3700 erp_idx = 0; 3701 low = 0; 3702 high = nlists - 1; 3703 3704 /* Binary search extent irec's */ 3705 while (low <= high) { 3706 erp_idx = (low + high) >> 1; 3707 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3708 prev = erp_idx > 0 ? erp - 1 : NULL; 3709 if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff && 3710 realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) { 3711 high = erp_idx - 1; 3712 } else if (page_idx > erp->er_extoff + erp->er_extcount || 3713 (page_idx == erp->er_extoff + erp->er_extcount && 3714 !realloc)) { 3715 low = erp_idx + 1; 3716 } else if (page_idx == erp->er_extoff + erp->er_extcount && 3717 erp->er_extcount == XFS_LINEAR_EXTS) { 3718 ASSERT(realloc); 3719 page_idx = 0; 3720 erp_idx++; 3721 erp = erp_idx < nlists ? erp + 1 : NULL; 3722 break; 3723 } else { 3724 page_idx -= erp->er_extoff; 3725 break; 3726 } 3727 } 3728 *idxp = page_idx; 3729 *erp_idxp = erp_idx; 3730 return(erp); 3731 } 3732 3733 /* 3734 * Allocate and initialize an indirection array once the space needed 3735 * for incore extents increases above XFS_IEXT_BUFSZ. 3736 */ 3737 void 3738 xfs_iext_irec_init( 3739 xfs_ifork_t *ifp) /* inode fork pointer */ 3740 { 3741 xfs_ext_irec_t *erp; /* indirection array pointer */ 3742 xfs_extnum_t nextents; /* number of extents in file */ 3743 3744 ASSERT(!(ifp->if_flags & XFS_IFEXTIREC)); 3745 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3746 ASSERT(nextents <= XFS_LINEAR_EXTS); 3747 3748 erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS); 3749 3750 if (nextents == 0) { 3751 ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); 3752 } else if (!ifp->if_real_bytes) { 3753 xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ); 3754 } else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) { 3755 xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ); 3756 } 3757 erp->er_extbuf = ifp->if_u1.if_extents; 3758 erp->er_extcount = nextents; 3759 erp->er_extoff = 0; 3760 3761 ifp->if_flags |= XFS_IFEXTIREC; 3762 ifp->if_real_bytes = XFS_IEXT_BUFSZ; 3763 ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t); 3764 ifp->if_u1.if_ext_irec = erp; 3765 3766 return; 3767 } 3768 3769 /* 3770 * Allocate and initialize a new entry in the indirection array. 3771 */ 3772 xfs_ext_irec_t * 3773 xfs_iext_irec_new( 3774 xfs_ifork_t *ifp, /* inode fork pointer */ 3775 int erp_idx) /* index for new irec */ 3776 { 3777 xfs_ext_irec_t *erp; /* indirection array pointer */ 3778 int i; /* loop counter */ 3779 int nlists; /* number of irec's (ex lists) */ 3780 3781 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3782 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3783 3784 /* Resize indirection array */ 3785 xfs_iext_realloc_indirect(ifp, ++nlists * 3786 sizeof(xfs_ext_irec_t)); 3787 /* 3788 * Move records down in the array so the 3789 * new page can use erp_idx. 3790 */ 3791 erp = ifp->if_u1.if_ext_irec; 3792 for (i = nlists - 1; i > erp_idx; i--) { 3793 memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t)); 3794 } 3795 ASSERT(i == erp_idx); 3796 3797 /* Initialize new extent record */ 3798 erp = ifp->if_u1.if_ext_irec; 3799 erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS); 3800 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 3801 memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ); 3802 erp[erp_idx].er_extcount = 0; 3803 erp[erp_idx].er_extoff = erp_idx > 0 ? 3804 erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0; 3805 return (&erp[erp_idx]); 3806 } 3807 3808 /* 3809 * Remove a record from the indirection array. 3810 */ 3811 void 3812 xfs_iext_irec_remove( 3813 xfs_ifork_t *ifp, /* inode fork pointer */ 3814 int erp_idx) /* irec index to remove */ 3815 { 3816 xfs_ext_irec_t *erp; /* indirection array pointer */ 3817 int i; /* loop counter */ 3818 int nlists; /* number of irec's (ex lists) */ 3819 3820 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3821 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3822 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3823 if (erp->er_extbuf) { 3824 xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, 3825 -erp->er_extcount); 3826 kmem_free(erp->er_extbuf); 3827 } 3828 /* Compact extent records */ 3829 erp = ifp->if_u1.if_ext_irec; 3830 for (i = erp_idx; i < nlists - 1; i++) { 3831 memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t)); 3832 } 3833 /* 3834 * Manually free the last extent record from the indirection 3835 * array. A call to xfs_iext_realloc_indirect() with a size 3836 * of zero would result in a call to xfs_iext_destroy() which 3837 * would in turn call this function again, creating a nasty 3838 * infinite loop. 3839 */ 3840 if (--nlists) { 3841 xfs_iext_realloc_indirect(ifp, 3842 nlists * sizeof(xfs_ext_irec_t)); 3843 } else { 3844 kmem_free(ifp->if_u1.if_ext_irec); 3845 } 3846 ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ; 3847 } 3848 3849 /* 3850 * This is called to clean up large amounts of unused memory allocated 3851 * by the indirection array. Before compacting anything though, verify 3852 * that the indirection array is still needed and switch back to the 3853 * linear extent list (or even the inline buffer) if possible. The 3854 * compaction policy is as follows: 3855 * 3856 * Full Compaction: Extents fit into a single page (or inline buffer) 3857 * Partial Compaction: Extents occupy less than 50% of allocated space 3858 * No Compaction: Extents occupy at least 50% of allocated space 3859 */ 3860 void 3861 xfs_iext_irec_compact( 3862 xfs_ifork_t *ifp) /* inode fork pointer */ 3863 { 3864 xfs_extnum_t nextents; /* number of extents in file */ 3865 int nlists; /* number of irec's (ex lists) */ 3866 3867 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3868 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3869 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t); 3870 3871 if (nextents == 0) { 3872 xfs_iext_destroy(ifp); 3873 } else if (nextents <= XFS_INLINE_EXTS) { 3874 xfs_iext_indirect_to_direct(ifp); 3875 xfs_iext_direct_to_inline(ifp, nextents); 3876 } else if (nextents <= XFS_LINEAR_EXTS) { 3877 xfs_iext_indirect_to_direct(ifp); 3878 } else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) { 3879 xfs_iext_irec_compact_pages(ifp); 3880 } 3881 } 3882 3883 /* 3884 * Combine extents from neighboring extent pages. 3885 */ 3886 void 3887 xfs_iext_irec_compact_pages( 3888 xfs_ifork_t *ifp) /* inode fork pointer */ 3889 { 3890 xfs_ext_irec_t *erp, *erp_next;/* pointers to irec entries */ 3891 int erp_idx = 0; /* indirection array index */ 3892 int nlists; /* number of irec's (ex lists) */ 3893 3894 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3895 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3896 while (erp_idx < nlists - 1) { 3897 erp = &ifp->if_u1.if_ext_irec[erp_idx]; 3898 erp_next = erp + 1; 3899 if (erp_next->er_extcount <= 3900 (XFS_LINEAR_EXTS - erp->er_extcount)) { 3901 memcpy(&erp->er_extbuf[erp->er_extcount], 3902 erp_next->er_extbuf, erp_next->er_extcount * 3903 sizeof(xfs_bmbt_rec_t)); 3904 erp->er_extcount += erp_next->er_extcount; 3905 /* 3906 * Free page before removing extent record 3907 * so er_extoffs don't get modified in 3908 * xfs_iext_irec_remove. 3909 */ 3910 kmem_free(erp_next->er_extbuf); 3911 erp_next->er_extbuf = NULL; 3912 xfs_iext_irec_remove(ifp, erp_idx + 1); 3913 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3914 } else { 3915 erp_idx++; 3916 } 3917 } 3918 } 3919 3920 /* 3921 * This is called to update the er_extoff field in the indirection 3922 * array when extents have been added or removed from one of the 3923 * extent lists. erp_idx contains the irec index to begin updating 3924 * at and ext_diff contains the number of extents that were added 3925 * or removed. 3926 */ 3927 void 3928 xfs_iext_irec_update_extoffs( 3929 xfs_ifork_t *ifp, /* inode fork pointer */ 3930 int erp_idx, /* irec index to update */ 3931 int ext_diff) /* number of new extents */ 3932 { 3933 int i; /* loop counter */ 3934 int nlists; /* number of irec's (ex lists */ 3935 3936 ASSERT(ifp->if_flags & XFS_IFEXTIREC); 3937 nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ; 3938 for (i = erp_idx; i < nlists; i++) { 3939 ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff; 3940 } 3941 } 3942 3943 /* 3944 * Test whether it is appropriate to check an inode for and free post EOF 3945 * blocks. The 'force' parameter determines whether we should also consider 3946 * regular files that are marked preallocated or append-only. 3947 */ 3948 bool 3949 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) 3950 { 3951 /* prealloc/delalloc exists only on regular files */ 3952 if (!S_ISREG(ip->i_d.di_mode)) 3953 return false; 3954 3955 /* 3956 * Zero sized files with no cached pages and delalloc blocks will not 3957 * have speculative prealloc/delalloc blocks to remove. 3958 */ 3959 if (VFS_I(ip)->i_size == 0 && 3960 VN_CACHED(VFS_I(ip)) == 0 && 3961 ip->i_delayed_blks == 0) 3962 return false; 3963 3964 /* If we haven't read in the extent list, then don't do it now. */ 3965 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) 3966 return false; 3967 3968 /* 3969 * Do not free real preallocated or append-only files unless the file 3970 * has delalloc blocks and we are forced to remove them. 3971 */ 3972 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 3973 if (!force || ip->i_delayed_blks == 0) 3974 return false; 3975 3976 return true; 3977 } 3978 3979