1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 4 * Copyright (c) 2012 Red Hat, Inc. 5 * All Rights Reserved. 6 */ 7 #include "xfs.h" 8 #include "xfs_fs.h" 9 #include "xfs_shared.h" 10 #include "xfs_format.h" 11 #include "xfs_log_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_bit.h" 14 #include "xfs_mount.h" 15 #include "xfs_defer.h" 16 #include "xfs_inode.h" 17 #include "xfs_btree.h" 18 #include "xfs_trans.h" 19 #include "xfs_alloc.h" 20 #include "xfs_bmap.h" 21 #include "xfs_bmap_util.h" 22 #include "xfs_bmap_btree.h" 23 #include "xfs_rtalloc.h" 24 #include "xfs_error.h" 25 #include "xfs_quota.h" 26 #include "xfs_trans_space.h" 27 #include "xfs_trace.h" 28 #include "xfs_icache.h" 29 #include "xfs_iomap.h" 30 #include "xfs_reflink.h" 31 32 /* Kernel only BMAP related definitions and functions */ 33 34 /* 35 * Convert the given file system block to a disk block. We have to treat it 36 * differently based on whether the file is a real time file or not, because the 37 * bmap code does. 38 */ 39 xfs_daddr_t 40 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) 41 { 42 if (XFS_IS_REALTIME_INODE(ip)) 43 return XFS_FSB_TO_BB(ip->i_mount, fsb); 44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb); 45 } 46 47 /* 48 * Routine to zero an extent on disk allocated to the specific inode. 49 * 50 * The VFS functions take a linearised filesystem block offset, so we have to 51 * convert the sparse xfs fsb to the right format first. 52 * VFS types are real funky, too. 53 */ 54 int 55 xfs_zero_extent( 56 struct xfs_inode *ip, 57 xfs_fsblock_t start_fsb, 58 xfs_off_t count_fsb) 59 { 60 struct xfs_mount *mp = ip->i_mount; 61 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 62 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); 63 sector_t block = XFS_BB_TO_FSBT(mp, sector); 64 65 return blkdev_issue_zeroout(target->bt_bdev, 66 block << (mp->m_super->s_blocksize_bits - 9), 67 count_fsb << (mp->m_super->s_blocksize_bits - 9), 68 GFP_NOFS, 0); 69 } 70 71 #ifdef CONFIG_XFS_RT 72 int 73 xfs_bmap_rtalloc( 74 struct xfs_bmalloca *ap) /* bmap alloc argument struct */ 75 { 76 int error; /* error return value */ 77 xfs_mount_t *mp; /* mount point structure */ 78 xfs_extlen_t prod = 0; /* product factor for allocators */ 79 xfs_extlen_t mod = 0; /* product factor for allocators */ 80 xfs_extlen_t ralen = 0; /* realtime allocation length */ 81 xfs_extlen_t align; /* minimum allocation alignment */ 82 xfs_rtblock_t rtb; 83 84 mp = ap->ip->i_mount; 85 align = xfs_get_extsz_hint(ap->ip); 86 prod = align / mp->m_sb.sb_rextsize; 87 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, 88 align, 1, ap->eof, 0, 89 ap->conv, &ap->offset, &ap->length); 90 if (error) 91 return error; 92 ASSERT(ap->length); 93 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); 94 95 /* 96 * If the offset & length are not perfectly aligned 97 * then kill prod, it will just get us in trouble. 98 */ 99 div_u64_rem(ap->offset, align, &mod); 100 if (mod || ap->length % align) 101 prod = 1; 102 /* 103 * Set ralen to be the actual requested length in rtextents. 104 */ 105 ralen = ap->length / mp->m_sb.sb_rextsize; 106 /* 107 * If the old value was close enough to MAXEXTLEN that 108 * we rounded up to it, cut it back so it's valid again. 109 * Note that if it's a really large request (bigger than 110 * MAXEXTLEN), we don't hear about that number, and can't 111 * adjust the starting point to match it. 112 */ 113 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) 114 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; 115 116 /* 117 * Lock out modifications to both the RT bitmap and summary inodes 118 */ 119 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); 120 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); 121 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); 122 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); 123 124 /* 125 * If it's an allocation to an empty file at offset 0, 126 * pick an extent that will space things out in the rt area. 127 */ 128 if (ap->eof && ap->offset == 0) { 129 xfs_rtblock_t rtx; /* realtime extent no */ 130 131 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); 132 if (error) 133 return error; 134 ap->blkno = rtx * mp->m_sb.sb_rextsize; 135 } else { 136 ap->blkno = 0; 137 } 138 139 xfs_bmap_adjacent(ap); 140 141 /* 142 * Realtime allocation, done through xfs_rtallocate_extent. 143 */ 144 do_div(ap->blkno, mp->m_sb.sb_rextsize); 145 rtb = ap->blkno; 146 ap->length = ralen; 147 error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, 148 &ralen, ap->wasdel, prod, &rtb); 149 if (error) 150 return error; 151 152 ap->blkno = rtb; 153 if (ap->blkno != NULLFSBLOCK) { 154 ap->blkno *= mp->m_sb.sb_rextsize; 155 ralen *= mp->m_sb.sb_rextsize; 156 ap->length = ralen; 157 ap->ip->i_nblocks += ralen; 158 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); 159 if (ap->wasdel) 160 ap->ip->i_delayed_blks -= ralen; 161 /* 162 * Adjust the disk quota also. This was reserved 163 * earlier. 164 */ 165 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, 166 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : 167 XFS_TRANS_DQ_RTBCOUNT, (long) ralen); 168 } else { 169 ap->length = 0; 170 } 171 return 0; 172 } 173 #endif /* CONFIG_XFS_RT */ 174 175 /* 176 * Extent tree block counting routines. 177 */ 178 179 /* 180 * Count leaf blocks given a range of extent records. Delayed allocation 181 * extents are not counted towards the totals. 182 */ 183 xfs_extnum_t 184 xfs_bmap_count_leaves( 185 struct xfs_ifork *ifp, 186 xfs_filblks_t *count) 187 { 188 struct xfs_iext_cursor icur; 189 struct xfs_bmbt_irec got; 190 xfs_extnum_t numrecs = 0; 191 192 for_each_xfs_iext(ifp, &icur, &got) { 193 if (!isnullstartblock(got.br_startblock)) { 194 *count += got.br_blockcount; 195 numrecs++; 196 } 197 } 198 199 return numrecs; 200 } 201 202 /* 203 * Count fsblocks of the given fork. Delayed allocation extents are 204 * not counted towards the totals. 205 */ 206 int 207 xfs_bmap_count_blocks( 208 struct xfs_trans *tp, 209 struct xfs_inode *ip, 210 int whichfork, 211 xfs_extnum_t *nextents, 212 xfs_filblks_t *count) 213 { 214 struct xfs_mount *mp = ip->i_mount; 215 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork); 216 struct xfs_btree_cur *cur; 217 xfs_extlen_t btblocks = 0; 218 int error; 219 220 *nextents = 0; 221 *count = 0; 222 223 if (!ifp) 224 return 0; 225 226 switch (ifp->if_format) { 227 case XFS_DINODE_FMT_BTREE: 228 error = xfs_iread_extents(tp, ip, whichfork); 229 if (error) 230 return error; 231 232 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork); 233 error = xfs_btree_count_blocks(cur, &btblocks); 234 xfs_btree_del_cursor(cur, error); 235 if (error) 236 return error; 237 238 /* 239 * xfs_btree_count_blocks includes the root block contained in 240 * the inode fork in @btblocks, so subtract one because we're 241 * only interested in allocated disk blocks. 242 */ 243 *count += btblocks - 1; 244 245 /* fall through */ 246 case XFS_DINODE_FMT_EXTENTS: 247 *nextents = xfs_bmap_count_leaves(ifp, count); 248 break; 249 } 250 251 return 0; 252 } 253 254 static int 255 xfs_getbmap_report_one( 256 struct xfs_inode *ip, 257 struct getbmapx *bmv, 258 struct kgetbmap *out, 259 int64_t bmv_end, 260 struct xfs_bmbt_irec *got) 261 { 262 struct kgetbmap *p = out + bmv->bmv_entries; 263 bool shared = false; 264 int error; 265 266 error = xfs_reflink_trim_around_shared(ip, got, &shared); 267 if (error) 268 return error; 269 270 if (isnullstartblock(got->br_startblock) || 271 got->br_startblock == DELAYSTARTBLOCK) { 272 /* 273 * Delalloc extents that start beyond EOF can occur due to 274 * speculative EOF allocation when the delalloc extent is larger 275 * than the largest freespace extent at conversion time. These 276 * extents cannot be converted by data writeback, so can exist 277 * here even if we are not supposed to be finding delalloc 278 * extents. 279 */ 280 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip))) 281 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0); 282 283 p->bmv_oflags |= BMV_OF_DELALLOC; 284 p->bmv_block = -2; 285 } else { 286 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock); 287 } 288 289 if (got->br_state == XFS_EXT_UNWRITTEN && 290 (bmv->bmv_iflags & BMV_IF_PREALLOC)) 291 p->bmv_oflags |= BMV_OF_PREALLOC; 292 293 if (shared) 294 p->bmv_oflags |= BMV_OF_SHARED; 295 296 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff); 297 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount); 298 299 bmv->bmv_offset = p->bmv_offset + p->bmv_length; 300 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 301 bmv->bmv_entries++; 302 return 0; 303 } 304 305 static void 306 xfs_getbmap_report_hole( 307 struct xfs_inode *ip, 308 struct getbmapx *bmv, 309 struct kgetbmap *out, 310 int64_t bmv_end, 311 xfs_fileoff_t bno, 312 xfs_fileoff_t end) 313 { 314 struct kgetbmap *p = out + bmv->bmv_entries; 315 316 if (bmv->bmv_iflags & BMV_IF_NO_HOLES) 317 return; 318 319 p->bmv_block = -1; 320 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno); 321 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno); 322 323 bmv->bmv_offset = p->bmv_offset + p->bmv_length; 324 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset); 325 bmv->bmv_entries++; 326 } 327 328 static inline bool 329 xfs_getbmap_full( 330 struct getbmapx *bmv) 331 { 332 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1; 333 } 334 335 static bool 336 xfs_getbmap_next_rec( 337 struct xfs_bmbt_irec *rec, 338 xfs_fileoff_t total_end) 339 { 340 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount; 341 342 if (end == total_end) 343 return false; 344 345 rec->br_startoff += rec->br_blockcount; 346 if (!isnullstartblock(rec->br_startblock) && 347 rec->br_startblock != DELAYSTARTBLOCK) 348 rec->br_startblock += rec->br_blockcount; 349 rec->br_blockcount = total_end - end; 350 return true; 351 } 352 353 /* 354 * Get inode's extents as described in bmv, and format for output. 355 * Calls formatter to fill the user's buffer until all extents 356 * are mapped, until the passed-in bmv->bmv_count slots have 357 * been filled, or until the formatter short-circuits the loop, 358 * if it is tracking filled-in extents on its own. 359 */ 360 int /* error code */ 361 xfs_getbmap( 362 struct xfs_inode *ip, 363 struct getbmapx *bmv, /* user bmap structure */ 364 struct kgetbmap *out) 365 { 366 struct xfs_mount *mp = ip->i_mount; 367 int iflags = bmv->bmv_iflags; 368 int whichfork, lock, error = 0; 369 int64_t bmv_end, max_len; 370 xfs_fileoff_t bno, first_bno; 371 struct xfs_ifork *ifp; 372 struct xfs_bmbt_irec got, rec; 373 xfs_filblks_t len; 374 struct xfs_iext_cursor icur; 375 376 if (bmv->bmv_iflags & ~BMV_IF_VALID) 377 return -EINVAL; 378 #ifndef DEBUG 379 /* Only allow CoW fork queries if we're debugging. */ 380 if (iflags & BMV_IF_COWFORK) 381 return -EINVAL; 382 #endif 383 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK)) 384 return -EINVAL; 385 386 if (bmv->bmv_length < -1) 387 return -EINVAL; 388 bmv->bmv_entries = 0; 389 if (bmv->bmv_length == 0) 390 return 0; 391 392 if (iflags & BMV_IF_ATTRFORK) 393 whichfork = XFS_ATTR_FORK; 394 else if (iflags & BMV_IF_COWFORK) 395 whichfork = XFS_COW_FORK; 396 else 397 whichfork = XFS_DATA_FORK; 398 ifp = XFS_IFORK_PTR(ip, whichfork); 399 400 xfs_ilock(ip, XFS_IOLOCK_SHARED); 401 switch (whichfork) { 402 case XFS_ATTR_FORK: 403 if (!XFS_IFORK_Q(ip)) 404 goto out_unlock_iolock; 405 406 max_len = 1LL << 32; 407 lock = xfs_ilock_attr_map_shared(ip); 408 break; 409 case XFS_COW_FORK: 410 /* No CoW fork? Just return */ 411 if (!ifp) 412 goto out_unlock_iolock; 413 414 if (xfs_get_cowextsz_hint(ip)) 415 max_len = mp->m_super->s_maxbytes; 416 else 417 max_len = XFS_ISIZE(ip); 418 419 lock = XFS_ILOCK_SHARED; 420 xfs_ilock(ip, lock); 421 break; 422 case XFS_DATA_FORK: 423 if (!(iflags & BMV_IF_DELALLOC) && 424 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) { 425 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 426 if (error) 427 goto out_unlock_iolock; 428 429 /* 430 * Even after flushing the inode, there can still be 431 * delalloc blocks on the inode beyond EOF due to 432 * speculative preallocation. These are not removed 433 * until the release function is called or the inode 434 * is inactivated. Hence we cannot assert here that 435 * ip->i_delayed_blks == 0. 436 */ 437 } 438 439 if (xfs_get_extsz_hint(ip) || 440 (ip->i_diflags & 441 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))) 442 max_len = mp->m_super->s_maxbytes; 443 else 444 max_len = XFS_ISIZE(ip); 445 446 lock = xfs_ilock_data_map_shared(ip); 447 break; 448 } 449 450 switch (ifp->if_format) { 451 case XFS_DINODE_FMT_EXTENTS: 452 case XFS_DINODE_FMT_BTREE: 453 break; 454 case XFS_DINODE_FMT_LOCAL: 455 /* Local format inode forks report no extents. */ 456 goto out_unlock_ilock; 457 default: 458 error = -EINVAL; 459 goto out_unlock_ilock; 460 } 461 462 if (bmv->bmv_length == -1) { 463 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len)); 464 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset); 465 } 466 467 bmv_end = bmv->bmv_offset + bmv->bmv_length; 468 469 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset); 470 len = XFS_BB_TO_FSB(mp, bmv->bmv_length); 471 472 error = xfs_iread_extents(NULL, ip, whichfork); 473 if (error) 474 goto out_unlock_ilock; 475 476 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) { 477 /* 478 * Report a whole-file hole if the delalloc flag is set to 479 * stay compatible with the old implementation. 480 */ 481 if (iflags & BMV_IF_DELALLOC) 482 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 483 XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); 484 goto out_unlock_ilock; 485 } 486 487 while (!xfs_getbmap_full(bmv)) { 488 xfs_trim_extent(&got, first_bno, len); 489 490 /* 491 * Report an entry for a hole if this extent doesn't directly 492 * follow the previous one. 493 */ 494 if (got.br_startoff > bno) { 495 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno, 496 got.br_startoff); 497 if (xfs_getbmap_full(bmv)) 498 break; 499 } 500 501 /* 502 * In order to report shared extents accurately, we report each 503 * distinct shared / unshared part of a single bmbt record with 504 * an individual getbmapx record. 505 */ 506 bno = got.br_startoff + got.br_blockcount; 507 rec = got; 508 do { 509 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end, 510 &rec); 511 if (error || xfs_getbmap_full(bmv)) 512 goto out_unlock_ilock; 513 } while (xfs_getbmap_next_rec(&rec, bno)); 514 515 if (!xfs_iext_next_extent(ifp, &icur, &got)) { 516 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); 517 518 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST; 519 520 if (whichfork != XFS_ATTR_FORK && bno < end && 521 !xfs_getbmap_full(bmv)) { 522 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, 523 bno, end); 524 } 525 break; 526 } 527 528 if (bno >= first_bno + len) 529 break; 530 } 531 532 out_unlock_ilock: 533 xfs_iunlock(ip, lock); 534 out_unlock_iolock: 535 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 536 return error; 537 } 538 539 /* 540 * Dead simple method of punching delalyed allocation blocks from a range in 541 * the inode. This will always punch out both the start and end blocks, even 542 * if the ranges only partially overlap them, so it is up to the caller to 543 * ensure that partial blocks are not passed in. 544 */ 545 int 546 xfs_bmap_punch_delalloc_range( 547 struct xfs_inode *ip, 548 xfs_fileoff_t start_fsb, 549 xfs_fileoff_t length) 550 { 551 struct xfs_ifork *ifp = &ip->i_df; 552 xfs_fileoff_t end_fsb = start_fsb + length; 553 struct xfs_bmbt_irec got, del; 554 struct xfs_iext_cursor icur; 555 int error = 0; 556 557 ASSERT(!xfs_need_iread_extents(ifp)); 558 559 xfs_ilock(ip, XFS_ILOCK_EXCL); 560 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got)) 561 goto out_unlock; 562 563 while (got.br_startoff + got.br_blockcount > start_fsb) { 564 del = got; 565 xfs_trim_extent(&del, start_fsb, length); 566 567 /* 568 * A delete can push the cursor forward. Step back to the 569 * previous extent on non-delalloc or extents outside the 570 * target range. 571 */ 572 if (!del.br_blockcount || 573 !isnullstartblock(del.br_startblock)) { 574 if (!xfs_iext_prev_extent(ifp, &icur, &got)) 575 break; 576 continue; 577 } 578 579 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur, 580 &got, &del); 581 if (error || !xfs_iext_get_extent(ifp, &icur, &got)) 582 break; 583 } 584 585 out_unlock: 586 xfs_iunlock(ip, XFS_ILOCK_EXCL); 587 return error; 588 } 589 590 /* 591 * Test whether it is appropriate to check an inode for and free post EOF 592 * blocks. The 'force' parameter determines whether we should also consider 593 * regular files that are marked preallocated or append-only. 594 */ 595 bool 596 xfs_can_free_eofblocks( 597 struct xfs_inode *ip, 598 bool force) 599 { 600 struct xfs_bmbt_irec imap; 601 struct xfs_mount *mp = ip->i_mount; 602 xfs_fileoff_t end_fsb; 603 xfs_fileoff_t last_fsb; 604 int nimaps = 1; 605 int error; 606 607 /* 608 * Caller must either hold the exclusive io lock; or be inactivating 609 * the inode, which guarantees there are no other users of the inode. 610 */ 611 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) || 612 (VFS_I(ip)->i_state & I_FREEING)); 613 614 /* prealloc/delalloc exists only on regular files */ 615 if (!S_ISREG(VFS_I(ip)->i_mode)) 616 return false; 617 618 /* 619 * Zero sized files with no cached pages and delalloc blocks will not 620 * have speculative prealloc/delalloc blocks to remove. 621 */ 622 if (VFS_I(ip)->i_size == 0 && 623 VFS_I(ip)->i_mapping->nrpages == 0 && 624 ip->i_delayed_blks == 0) 625 return false; 626 627 /* If we haven't read in the extent list, then don't do it now. */ 628 if (xfs_need_iread_extents(&ip->i_df)) 629 return false; 630 631 /* 632 * Do not free real preallocated or append-only files unless the file 633 * has delalloc blocks and we are forced to remove them. 634 */ 635 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 636 if (!force || ip->i_delayed_blks == 0) 637 return false; 638 639 /* 640 * Do not try to free post-EOF blocks if EOF is beyond the end of the 641 * range supported by the page cache, because the truncation will loop 642 * forever. 643 */ 644 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); 645 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 646 if (last_fsb <= end_fsb) 647 return false; 648 649 /* 650 * Look up the mapping for the first block past EOF. If we can't find 651 * it, there's nothing to free. 652 */ 653 xfs_ilock(ip, XFS_ILOCK_SHARED); 654 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps, 655 0); 656 xfs_iunlock(ip, XFS_ILOCK_SHARED); 657 if (error || nimaps == 0) 658 return false; 659 660 /* 661 * If there's a real mapping there or there are delayed allocation 662 * reservations, then we have post-EOF blocks to try to free. 663 */ 664 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks; 665 } 666 667 /* 668 * This is called to free any blocks beyond eof. The caller must hold 669 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only 670 * reference to the inode. 671 */ 672 int 673 xfs_free_eofblocks( 674 struct xfs_inode *ip) 675 { 676 struct xfs_trans *tp; 677 struct xfs_mount *mp = ip->i_mount; 678 int error; 679 680 /* Attach the dquots to the inode up front. */ 681 error = xfs_qm_dqattach(ip); 682 if (error) 683 return error; 684 685 /* Wait on dio to ensure i_size has settled. */ 686 inode_dio_wait(VFS_I(ip)); 687 688 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 689 if (error) { 690 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 691 return error; 692 } 693 694 xfs_ilock(ip, XFS_ILOCK_EXCL); 695 xfs_trans_ijoin(tp, ip, 0); 696 697 /* 698 * Do not update the on-disk file size. If we update the on-disk file 699 * size and then the system crashes before the contents of the file are 700 * flushed to disk then the files may be full of holes (ie NULL files 701 * bug). 702 */ 703 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK, 704 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD); 705 if (error) 706 goto err_cancel; 707 708 error = xfs_trans_commit(tp); 709 if (error) 710 goto out_unlock; 711 712 xfs_inode_clear_eofblocks_tag(ip); 713 goto out_unlock; 714 715 err_cancel: 716 /* 717 * If we get an error at this point we simply don't 718 * bother truncating the file. 719 */ 720 xfs_trans_cancel(tp); 721 out_unlock: 722 xfs_iunlock(ip, XFS_ILOCK_EXCL); 723 return error; 724 } 725 726 int 727 xfs_alloc_file_space( 728 struct xfs_inode *ip, 729 xfs_off_t offset, 730 xfs_off_t len, 731 int alloc_type) 732 { 733 xfs_mount_t *mp = ip->i_mount; 734 xfs_off_t count; 735 xfs_filblks_t allocated_fsb; 736 xfs_filblks_t allocatesize_fsb; 737 xfs_extlen_t extsz, temp; 738 xfs_fileoff_t startoffset_fsb; 739 xfs_fileoff_t endoffset_fsb; 740 int nimaps; 741 int rt; 742 xfs_trans_t *tp; 743 xfs_bmbt_irec_t imaps[1], *imapp; 744 int error; 745 746 trace_xfs_alloc_file_space(ip); 747 748 if (XFS_FORCED_SHUTDOWN(mp)) 749 return -EIO; 750 751 error = xfs_qm_dqattach(ip); 752 if (error) 753 return error; 754 755 if (len <= 0) 756 return -EINVAL; 757 758 rt = XFS_IS_REALTIME_INODE(ip); 759 extsz = xfs_get_extsz_hint(ip); 760 761 count = len; 762 imapp = &imaps[0]; 763 nimaps = 1; 764 startoffset_fsb = XFS_B_TO_FSBT(mp, offset); 765 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count); 766 allocatesize_fsb = endoffset_fsb - startoffset_fsb; 767 768 /* 769 * Allocate file space until done or until there is an error 770 */ 771 while (allocatesize_fsb && !error) { 772 xfs_fileoff_t s, e; 773 unsigned int dblocks, rblocks, resblks; 774 775 /* 776 * Determine space reservations for data/realtime. 777 */ 778 if (unlikely(extsz)) { 779 s = startoffset_fsb; 780 do_div(s, extsz); 781 s *= extsz; 782 e = startoffset_fsb + allocatesize_fsb; 783 div_u64_rem(startoffset_fsb, extsz, &temp); 784 if (temp) 785 e += temp; 786 div_u64_rem(e, extsz, &temp); 787 if (temp) 788 e += extsz - temp; 789 } else { 790 s = 0; 791 e = allocatesize_fsb; 792 } 793 794 /* 795 * The transaction reservation is limited to a 32-bit block 796 * count, hence we need to limit the number of blocks we are 797 * trying to reserve to avoid an overflow. We can't allocate 798 * more than @nimaps extents, and an extent is limited on disk 799 * to MAXEXTLEN (21 bits), so use that to enforce the limit. 800 */ 801 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); 802 if (unlikely(rt)) { 803 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 804 rblocks = resblks; 805 } else { 806 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); 807 rblocks = 0; 808 } 809 810 /* 811 * Allocate and setup the transaction. 812 */ 813 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, 814 dblocks, rblocks, false, &tp); 815 if (error) 816 break; 817 818 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 819 XFS_IEXT_ADD_NOSPLIT_CNT); 820 if (error) 821 goto error; 822 823 error = xfs_bmapi_write(tp, ip, startoffset_fsb, 824 allocatesize_fsb, alloc_type, 0, imapp, 825 &nimaps); 826 if (error) 827 goto error; 828 829 /* 830 * Complete the transaction 831 */ 832 error = xfs_trans_commit(tp); 833 xfs_iunlock(ip, XFS_ILOCK_EXCL); 834 if (error) 835 break; 836 837 allocated_fsb = imapp->br_blockcount; 838 839 if (nimaps == 0) { 840 error = -ENOSPC; 841 break; 842 } 843 844 startoffset_fsb += allocated_fsb; 845 allocatesize_fsb -= allocated_fsb; 846 } 847 848 return error; 849 850 error: 851 xfs_trans_cancel(tp); 852 xfs_iunlock(ip, XFS_ILOCK_EXCL); 853 return error; 854 } 855 856 static int 857 xfs_unmap_extent( 858 struct xfs_inode *ip, 859 xfs_fileoff_t startoffset_fsb, 860 xfs_filblks_t len_fsb, 861 int *done) 862 { 863 struct xfs_mount *mp = ip->i_mount; 864 struct xfs_trans *tp; 865 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 866 int error; 867 868 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0, 869 false, &tp); 870 if (error) 871 return error; 872 873 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 874 XFS_IEXT_PUNCH_HOLE_CNT); 875 if (error) 876 goto out_trans_cancel; 877 878 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done); 879 if (error) 880 goto out_trans_cancel; 881 882 error = xfs_trans_commit(tp); 883 out_unlock: 884 xfs_iunlock(ip, XFS_ILOCK_EXCL); 885 return error; 886 887 out_trans_cancel: 888 xfs_trans_cancel(tp); 889 goto out_unlock; 890 } 891 892 /* Caller must first wait for the completion of any pending DIOs if required. */ 893 int 894 xfs_flush_unmap_range( 895 struct xfs_inode *ip, 896 xfs_off_t offset, 897 xfs_off_t len) 898 { 899 struct xfs_mount *mp = ip->i_mount; 900 struct inode *inode = VFS_I(ip); 901 xfs_off_t rounding, start, end; 902 int error; 903 904 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE); 905 start = round_down(offset, rounding); 906 end = round_up(offset + len, rounding) - 1; 907 908 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 909 if (error) 910 return error; 911 truncate_pagecache_range(inode, start, end); 912 return 0; 913 } 914 915 int 916 xfs_free_file_space( 917 struct xfs_inode *ip, 918 xfs_off_t offset, 919 xfs_off_t len) 920 { 921 struct xfs_mount *mp = ip->i_mount; 922 xfs_fileoff_t startoffset_fsb; 923 xfs_fileoff_t endoffset_fsb; 924 int done = 0, error; 925 926 trace_xfs_free_file_space(ip); 927 928 error = xfs_qm_dqattach(ip); 929 if (error) 930 return error; 931 932 if (len <= 0) /* if nothing being freed */ 933 return 0; 934 935 startoffset_fsb = XFS_B_TO_FSB(mp, offset); 936 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); 937 938 /* We can only free complete realtime extents. */ 939 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) { 940 startoffset_fsb = roundup_64(startoffset_fsb, 941 mp->m_sb.sb_rextsize); 942 endoffset_fsb = rounddown_64(endoffset_fsb, 943 mp->m_sb.sb_rextsize); 944 } 945 946 /* 947 * Need to zero the stuff we're not freeing, on disk. 948 */ 949 if (endoffset_fsb > startoffset_fsb) { 950 while (!done) { 951 error = xfs_unmap_extent(ip, startoffset_fsb, 952 endoffset_fsb - startoffset_fsb, &done); 953 if (error) 954 return error; 955 } 956 } 957 958 /* 959 * Now that we've unmap all full blocks we'll have to zero out any 960 * partial block at the beginning and/or end. iomap_zero_range is smart 961 * enough to skip any holes, including those we just created, but we 962 * must take care not to zero beyond EOF and enlarge i_size. 963 */ 964 if (offset >= XFS_ISIZE(ip)) 965 return 0; 966 if (offset + len > XFS_ISIZE(ip)) 967 len = XFS_ISIZE(ip) - offset; 968 error = iomap_zero_range(VFS_I(ip), offset, len, NULL, 969 &xfs_buffered_write_iomap_ops); 970 if (error) 971 return error; 972 973 /* 974 * If we zeroed right up to EOF and EOF straddles a page boundary we 975 * must make sure that the post-EOF area is also zeroed because the 976 * page could be mmap'd and iomap_zero_range doesn't do that for us. 977 * Writeback of the eof page will do this, albeit clumsily. 978 */ 979 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) { 980 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 981 round_down(offset + len, PAGE_SIZE), LLONG_MAX); 982 } 983 984 return error; 985 } 986 987 static int 988 xfs_prepare_shift( 989 struct xfs_inode *ip, 990 loff_t offset) 991 { 992 struct xfs_mount *mp = ip->i_mount; 993 int error; 994 995 /* 996 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation 997 * into the accessible region of the file. 998 */ 999 if (xfs_can_free_eofblocks(ip, true)) { 1000 error = xfs_free_eofblocks(ip); 1001 if (error) 1002 return error; 1003 } 1004 1005 /* 1006 * Shift operations must stabilize the start block offset boundary along 1007 * with the full range of the operation. If we don't, a COW writeback 1008 * completion could race with an insert, front merge with the start 1009 * extent (after split) during the shift and corrupt the file. Start 1010 * with the block just prior to the start to stabilize the boundary. 1011 */ 1012 offset = round_down(offset, 1 << mp->m_sb.sb_blocklog); 1013 if (offset) 1014 offset -= (1 << mp->m_sb.sb_blocklog); 1015 1016 /* 1017 * Writeback and invalidate cache for the remainder of the file as we're 1018 * about to shift down every extent from offset to EOF. 1019 */ 1020 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip)); 1021 if (error) 1022 return error; 1023 1024 /* 1025 * Clean out anything hanging around in the cow fork now that 1026 * we've flushed all the dirty data out to disk to avoid having 1027 * CoW extents at the wrong offsets. 1028 */ 1029 if (xfs_inode_has_cow_data(ip)) { 1030 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF, 1031 true); 1032 if (error) 1033 return error; 1034 } 1035 1036 return 0; 1037 } 1038 1039 /* 1040 * xfs_collapse_file_space() 1041 * This routine frees disk space and shift extent for the given file. 1042 * The first thing we do is to free data blocks in the specified range 1043 * by calling xfs_free_file_space(). It would also sync dirty data 1044 * and invalidate page cache over the region on which collapse range 1045 * is working. And Shift extent records to the left to cover a hole. 1046 * RETURNS: 1047 * 0 on success 1048 * errno on error 1049 * 1050 */ 1051 int 1052 xfs_collapse_file_space( 1053 struct xfs_inode *ip, 1054 xfs_off_t offset, 1055 xfs_off_t len) 1056 { 1057 struct xfs_mount *mp = ip->i_mount; 1058 struct xfs_trans *tp; 1059 int error; 1060 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len); 1061 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1062 bool done = false; 1063 1064 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1065 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 1066 1067 trace_xfs_collapse_file_space(ip); 1068 1069 error = xfs_free_file_space(ip, offset, len); 1070 if (error) 1071 return error; 1072 1073 error = xfs_prepare_shift(ip, offset); 1074 if (error) 1075 return error; 1076 1077 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp); 1078 if (error) 1079 return error; 1080 1081 xfs_ilock(ip, XFS_ILOCK_EXCL); 1082 xfs_trans_ijoin(tp, ip, 0); 1083 1084 while (!done) { 1085 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb, 1086 &done); 1087 if (error) 1088 goto out_trans_cancel; 1089 if (done) 1090 break; 1091 1092 /* finish any deferred frees and roll the transaction */ 1093 error = xfs_defer_finish(&tp); 1094 if (error) 1095 goto out_trans_cancel; 1096 } 1097 1098 error = xfs_trans_commit(tp); 1099 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1100 return error; 1101 1102 out_trans_cancel: 1103 xfs_trans_cancel(tp); 1104 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1105 return error; 1106 } 1107 1108 /* 1109 * xfs_insert_file_space() 1110 * This routine create hole space by shifting extents for the given file. 1111 * The first thing we do is to sync dirty data and invalidate page cache 1112 * over the region on which insert range is working. And split an extent 1113 * to two extents at given offset by calling xfs_bmap_split_extent. 1114 * And shift all extent records which are laying between [offset, 1115 * last allocated extent] to the right to reserve hole range. 1116 * RETURNS: 1117 * 0 on success 1118 * errno on error 1119 */ 1120 int 1121 xfs_insert_file_space( 1122 struct xfs_inode *ip, 1123 loff_t offset, 1124 loff_t len) 1125 { 1126 struct xfs_mount *mp = ip->i_mount; 1127 struct xfs_trans *tp; 1128 int error; 1129 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset); 1130 xfs_fileoff_t next_fsb = NULLFSBLOCK; 1131 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len); 1132 bool done = false; 1133 1134 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1135 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 1136 1137 trace_xfs_insert_file_space(ip); 1138 1139 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb); 1140 if (error) 1141 return error; 1142 1143 error = xfs_prepare_shift(ip, offset); 1144 if (error) 1145 return error; 1146 1147 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 1148 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); 1149 if (error) 1150 return error; 1151 1152 xfs_ilock(ip, XFS_ILOCK_EXCL); 1153 xfs_trans_ijoin(tp, ip, 0); 1154 1155 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, 1156 XFS_IEXT_PUNCH_HOLE_CNT); 1157 if (error) 1158 goto out_trans_cancel; 1159 1160 /* 1161 * The extent shifting code works on extent granularity. So, if stop_fsb 1162 * is not the starting block of extent, we need to split the extent at 1163 * stop_fsb. 1164 */ 1165 error = xfs_bmap_split_extent(tp, ip, stop_fsb); 1166 if (error) 1167 goto out_trans_cancel; 1168 1169 do { 1170 error = xfs_defer_finish(&tp); 1171 if (error) 1172 goto out_trans_cancel; 1173 1174 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb, 1175 &done, stop_fsb); 1176 if (error) 1177 goto out_trans_cancel; 1178 } while (!done); 1179 1180 error = xfs_trans_commit(tp); 1181 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1182 return error; 1183 1184 out_trans_cancel: 1185 xfs_trans_cancel(tp); 1186 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1187 return error; 1188 } 1189 1190 /* 1191 * We need to check that the format of the data fork in the temporary inode is 1192 * valid for the target inode before doing the swap. This is not a problem with 1193 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized 1194 * data fork depending on the space the attribute fork is taking so we can get 1195 * invalid formats on the target inode. 1196 * 1197 * E.g. target has space for 7 extents in extent format, temp inode only has 1198 * space for 6. If we defragment down to 7 extents, then the tmp format is a 1199 * btree, but when swapped it needs to be in extent format. Hence we can't just 1200 * blindly swap data forks on attr2 filesystems. 1201 * 1202 * Note that we check the swap in both directions so that we don't end up with 1203 * a corrupt temporary inode, either. 1204 * 1205 * Note that fixing the way xfs_fsr sets up the attribute fork in the source 1206 * inode will prevent this situation from occurring, so all we do here is 1207 * reject and log the attempt. basically we are putting the responsibility on 1208 * userspace to get this right. 1209 */ 1210 static int 1211 xfs_swap_extents_check_format( 1212 struct xfs_inode *ip, /* target inode */ 1213 struct xfs_inode *tip) /* tmp inode */ 1214 { 1215 struct xfs_ifork *ifp = &ip->i_df; 1216 struct xfs_ifork *tifp = &tip->i_df; 1217 1218 /* User/group/project quota ids must match if quotas are enforced. */ 1219 if (XFS_IS_QUOTA_ON(ip->i_mount) && 1220 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) || 1221 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) || 1222 ip->i_projid != tip->i_projid)) 1223 return -EINVAL; 1224 1225 /* Should never get a local format */ 1226 if (ifp->if_format == XFS_DINODE_FMT_LOCAL || 1227 tifp->if_format == XFS_DINODE_FMT_LOCAL) 1228 return -EINVAL; 1229 1230 /* 1231 * if the target inode has less extents that then temporary inode then 1232 * why did userspace call us? 1233 */ 1234 if (ifp->if_nextents < tifp->if_nextents) 1235 return -EINVAL; 1236 1237 /* 1238 * If we have to use the (expensive) rmap swap method, we can 1239 * handle any number of extents and any format. 1240 */ 1241 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb)) 1242 return 0; 1243 1244 /* 1245 * if the target inode is in extent form and the temp inode is in btree 1246 * form then we will end up with the target inode in the wrong format 1247 * as we already know there are less extents in the temp inode. 1248 */ 1249 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1250 tifp->if_format == XFS_DINODE_FMT_BTREE) 1251 return -EINVAL; 1252 1253 /* Check temp in extent form to max in target */ 1254 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS && 1255 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1256 return -EINVAL; 1257 1258 /* Check target in extent form to max in temp */ 1259 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && 1260 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1261 return -EINVAL; 1262 1263 /* 1264 * If we are in a btree format, check that the temp root block will fit 1265 * in the target and that it has enough extents to be in btree format 1266 * in the target. 1267 * 1268 * Note that we have to be careful to allow btree->extent conversions 1269 * (a common defrag case) which will occur when the temp inode is in 1270 * extent format... 1271 */ 1272 if (tifp->if_format == XFS_DINODE_FMT_BTREE) { 1273 if (XFS_IFORK_Q(ip) && 1274 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > XFS_IFORK_BOFF(ip)) 1275 return -EINVAL; 1276 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1277 return -EINVAL; 1278 } 1279 1280 /* Reciprocal target->temp btree format checks */ 1281 if (ifp->if_format == XFS_DINODE_FMT_BTREE) { 1282 if (XFS_IFORK_Q(tip) && 1283 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) 1284 return -EINVAL; 1285 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1286 return -EINVAL; 1287 } 1288 1289 return 0; 1290 } 1291 1292 static int 1293 xfs_swap_extent_flush( 1294 struct xfs_inode *ip) 1295 { 1296 int error; 1297 1298 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 1299 if (error) 1300 return error; 1301 truncate_pagecache_range(VFS_I(ip), 0, -1); 1302 1303 /* Verify O_DIRECT for ftmp */ 1304 if (VFS_I(ip)->i_mapping->nrpages) 1305 return -EINVAL; 1306 return 0; 1307 } 1308 1309 /* 1310 * Move extents from one file to another, when rmap is enabled. 1311 */ 1312 STATIC int 1313 xfs_swap_extent_rmap( 1314 struct xfs_trans **tpp, 1315 struct xfs_inode *ip, 1316 struct xfs_inode *tip) 1317 { 1318 struct xfs_trans *tp = *tpp; 1319 struct xfs_bmbt_irec irec; 1320 struct xfs_bmbt_irec uirec; 1321 struct xfs_bmbt_irec tirec; 1322 xfs_fileoff_t offset_fsb; 1323 xfs_fileoff_t end_fsb; 1324 xfs_filblks_t count_fsb; 1325 int error; 1326 xfs_filblks_t ilen; 1327 xfs_filblks_t rlen; 1328 int nimaps; 1329 uint64_t tip_flags2; 1330 1331 /* 1332 * If the source file has shared blocks, we must flag the donor 1333 * file as having shared blocks so that we get the shared-block 1334 * rmap functions when we go to fix up the rmaps. The flags 1335 * will be switch for reals later. 1336 */ 1337 tip_flags2 = tip->i_diflags2; 1338 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK) 1339 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK; 1340 1341 offset_fsb = 0; 1342 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip))); 1343 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb); 1344 1345 while (count_fsb) { 1346 /* Read extent from the donor file */ 1347 nimaps = 1; 1348 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec, 1349 &nimaps, 0); 1350 if (error) 1351 goto out; 1352 ASSERT(nimaps == 1); 1353 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK); 1354 1355 trace_xfs_swap_extent_rmap_remap(tip, &tirec); 1356 ilen = tirec.br_blockcount; 1357 1358 /* Unmap the old blocks in the source file. */ 1359 while (tirec.br_blockcount) { 1360 ASSERT(tp->t_firstblock == NULLFSBLOCK); 1361 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec); 1362 1363 /* Read extent from the source file */ 1364 nimaps = 1; 1365 error = xfs_bmapi_read(ip, tirec.br_startoff, 1366 tirec.br_blockcount, &irec, 1367 &nimaps, 0); 1368 if (error) 1369 goto out; 1370 ASSERT(nimaps == 1); 1371 ASSERT(tirec.br_startoff == irec.br_startoff); 1372 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec); 1373 1374 /* Trim the extent. */ 1375 uirec = tirec; 1376 uirec.br_blockcount = rlen = min_t(xfs_filblks_t, 1377 tirec.br_blockcount, 1378 irec.br_blockcount); 1379 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec); 1380 1381 if (xfs_bmap_is_real_extent(&uirec)) { 1382 error = xfs_iext_count_may_overflow(ip, 1383 XFS_DATA_FORK, 1384 XFS_IEXT_SWAP_RMAP_CNT); 1385 if (error) 1386 goto out; 1387 } 1388 1389 if (xfs_bmap_is_real_extent(&irec)) { 1390 error = xfs_iext_count_may_overflow(tip, 1391 XFS_DATA_FORK, 1392 XFS_IEXT_SWAP_RMAP_CNT); 1393 if (error) 1394 goto out; 1395 } 1396 1397 /* Remove the mapping from the donor file. */ 1398 xfs_bmap_unmap_extent(tp, tip, &uirec); 1399 1400 /* Remove the mapping from the source file. */ 1401 xfs_bmap_unmap_extent(tp, ip, &irec); 1402 1403 /* Map the donor file's blocks into the source file. */ 1404 xfs_bmap_map_extent(tp, ip, &uirec); 1405 1406 /* Map the source file's blocks into the donor file. */ 1407 xfs_bmap_map_extent(tp, tip, &irec); 1408 1409 error = xfs_defer_finish(tpp); 1410 tp = *tpp; 1411 if (error) 1412 goto out; 1413 1414 tirec.br_startoff += rlen; 1415 if (tirec.br_startblock != HOLESTARTBLOCK && 1416 tirec.br_startblock != DELAYSTARTBLOCK) 1417 tirec.br_startblock += rlen; 1418 tirec.br_blockcount -= rlen; 1419 } 1420 1421 /* Roll on... */ 1422 count_fsb -= ilen; 1423 offset_fsb += ilen; 1424 } 1425 1426 tip->i_diflags2 = tip_flags2; 1427 return 0; 1428 1429 out: 1430 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_); 1431 tip->i_diflags2 = tip_flags2; 1432 return error; 1433 } 1434 1435 /* Swap the extents of two files by swapping data forks. */ 1436 STATIC int 1437 xfs_swap_extent_forks( 1438 struct xfs_trans *tp, 1439 struct xfs_inode *ip, 1440 struct xfs_inode *tip, 1441 int *src_log_flags, 1442 int *target_log_flags) 1443 { 1444 xfs_filblks_t aforkblks = 0; 1445 xfs_filblks_t taforkblks = 0; 1446 xfs_extnum_t junk; 1447 uint64_t tmp; 1448 int error; 1449 1450 /* 1451 * Count the number of extended attribute blocks 1452 */ 1453 if (XFS_IFORK_Q(ip) && ip->i_afp->if_nextents > 0 && 1454 ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) { 1455 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk, 1456 &aforkblks); 1457 if (error) 1458 return error; 1459 } 1460 if (XFS_IFORK_Q(tip) && tip->i_afp->if_nextents > 0 && 1461 tip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) { 1462 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk, 1463 &taforkblks); 1464 if (error) 1465 return error; 1466 } 1467 1468 /* 1469 * Btree format (v3) inodes have the inode number stamped in the bmbt 1470 * block headers. We can't start changing the bmbt blocks until the 1471 * inode owner change is logged so recovery does the right thing in the 1472 * event of a crash. Set the owner change log flags now and leave the 1473 * bmbt scan as the last step. 1474 */ 1475 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) { 1476 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE) 1477 (*target_log_flags) |= XFS_ILOG_DOWNER; 1478 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE) 1479 (*src_log_flags) |= XFS_ILOG_DOWNER; 1480 } 1481 1482 /* 1483 * Swap the data forks of the inodes 1484 */ 1485 swap(ip->i_df, tip->i_df); 1486 1487 /* 1488 * Fix the on-disk inode values 1489 */ 1490 tmp = (uint64_t)ip->i_nblocks; 1491 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks; 1492 tip->i_nblocks = tmp + taforkblks - aforkblks; 1493 1494 /* 1495 * The extents in the source inode could still contain speculative 1496 * preallocation beyond EOF (e.g. the file is open but not modified 1497 * while defrag is in progress). In that case, we need to copy over the 1498 * number of delalloc blocks the data fork in the source inode is 1499 * tracking beyond EOF so that when the fork is truncated away when the 1500 * temporary inode is unlinked we don't underrun the i_delayed_blks 1501 * counter on that inode. 1502 */ 1503 ASSERT(tip->i_delayed_blks == 0); 1504 tip->i_delayed_blks = ip->i_delayed_blks; 1505 ip->i_delayed_blks = 0; 1506 1507 switch (ip->i_df.if_format) { 1508 case XFS_DINODE_FMT_EXTENTS: 1509 (*src_log_flags) |= XFS_ILOG_DEXT; 1510 break; 1511 case XFS_DINODE_FMT_BTREE: 1512 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) || 1513 (*src_log_flags & XFS_ILOG_DOWNER)); 1514 (*src_log_flags) |= XFS_ILOG_DBROOT; 1515 break; 1516 } 1517 1518 switch (tip->i_df.if_format) { 1519 case XFS_DINODE_FMT_EXTENTS: 1520 (*target_log_flags) |= XFS_ILOG_DEXT; 1521 break; 1522 case XFS_DINODE_FMT_BTREE: 1523 (*target_log_flags) |= XFS_ILOG_DBROOT; 1524 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) || 1525 (*target_log_flags & XFS_ILOG_DOWNER)); 1526 break; 1527 } 1528 1529 return 0; 1530 } 1531 1532 /* 1533 * Fix up the owners of the bmbt blocks to refer to the current inode. The 1534 * change owner scan attempts to order all modified buffers in the current 1535 * transaction. In the event of ordered buffer failure, the offending buffer is 1536 * physically logged as a fallback and the scan returns -EAGAIN. We must roll 1537 * the transaction in this case to replenish the fallback log reservation and 1538 * restart the scan. This process repeats until the scan completes. 1539 */ 1540 static int 1541 xfs_swap_change_owner( 1542 struct xfs_trans **tpp, 1543 struct xfs_inode *ip, 1544 struct xfs_inode *tmpip) 1545 { 1546 int error; 1547 struct xfs_trans *tp = *tpp; 1548 1549 do { 1550 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino, 1551 NULL); 1552 /* success or fatal error */ 1553 if (error != -EAGAIN) 1554 break; 1555 1556 error = xfs_trans_roll(tpp); 1557 if (error) 1558 break; 1559 tp = *tpp; 1560 1561 /* 1562 * Redirty both inodes so they can relog and keep the log tail 1563 * moving forward. 1564 */ 1565 xfs_trans_ijoin(tp, ip, 0); 1566 xfs_trans_ijoin(tp, tmpip, 0); 1567 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1568 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE); 1569 } while (true); 1570 1571 return error; 1572 } 1573 1574 int 1575 xfs_swap_extents( 1576 struct xfs_inode *ip, /* target inode */ 1577 struct xfs_inode *tip, /* tmp inode */ 1578 struct xfs_swapext *sxp) 1579 { 1580 struct xfs_mount *mp = ip->i_mount; 1581 struct xfs_trans *tp; 1582 struct xfs_bstat *sbp = &sxp->sx_stat; 1583 int src_log_flags, target_log_flags; 1584 int error = 0; 1585 int lock_flags; 1586 uint64_t f; 1587 int resblks = 0; 1588 unsigned int flags = 0; 1589 1590 /* 1591 * Lock the inodes against other IO, page faults and truncate to 1592 * begin with. Then we can ensure the inodes are flushed and have no 1593 * page cache safely. Once we have done this we can take the ilocks and 1594 * do the rest of the checks. 1595 */ 1596 lock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1597 lock_flags = XFS_MMAPLOCK_EXCL; 1598 xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL); 1599 1600 /* Verify that both files have the same format */ 1601 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { 1602 error = -EINVAL; 1603 goto out_unlock; 1604 } 1605 1606 /* Verify both files are either real-time or non-realtime */ 1607 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { 1608 error = -EINVAL; 1609 goto out_unlock; 1610 } 1611 1612 error = xfs_qm_dqattach(ip); 1613 if (error) 1614 goto out_unlock; 1615 1616 error = xfs_qm_dqattach(tip); 1617 if (error) 1618 goto out_unlock; 1619 1620 error = xfs_swap_extent_flush(ip); 1621 if (error) 1622 goto out_unlock; 1623 error = xfs_swap_extent_flush(tip); 1624 if (error) 1625 goto out_unlock; 1626 1627 if (xfs_inode_has_cow_data(tip)) { 1628 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true); 1629 if (error) 1630 goto out_unlock; 1631 } 1632 1633 /* 1634 * Extent "swapping" with rmap requires a permanent reservation and 1635 * a block reservation because it's really just a remap operation 1636 * performed with log redo items! 1637 */ 1638 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) { 1639 int w = XFS_DATA_FORK; 1640 uint32_t ipnext = ip->i_df.if_nextents; 1641 uint32_t tipnext = tip->i_df.if_nextents; 1642 1643 /* 1644 * Conceptually this shouldn't affect the shape of either bmbt, 1645 * but since we atomically move extents one by one, we reserve 1646 * enough space to rebuild both trees. 1647 */ 1648 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w); 1649 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w); 1650 1651 /* 1652 * If either inode straddles a bmapbt block allocation boundary, 1653 * the rmapbt algorithm triggers repeated allocs and frees as 1654 * extents are remapped. This can exhaust the block reservation 1655 * prematurely and cause shutdown. Return freed blocks to the 1656 * transaction reservation to counter this behavior. 1657 */ 1658 flags |= XFS_TRANS_RES_FDBLKS; 1659 } 1660 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags, 1661 &tp); 1662 if (error) 1663 goto out_unlock; 1664 1665 /* 1666 * Lock and join the inodes to the tansaction so that transaction commit 1667 * or cancel will unlock the inodes from this point onwards. 1668 */ 1669 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL); 1670 lock_flags |= XFS_ILOCK_EXCL; 1671 xfs_trans_ijoin(tp, ip, 0); 1672 xfs_trans_ijoin(tp, tip, 0); 1673 1674 1675 /* Verify all data are being swapped */ 1676 if (sxp->sx_offset != 0 || 1677 sxp->sx_length != ip->i_disk_size || 1678 sxp->sx_length != tip->i_disk_size) { 1679 error = -EFAULT; 1680 goto out_trans_cancel; 1681 } 1682 1683 trace_xfs_swap_extent_before(ip, 0); 1684 trace_xfs_swap_extent_before(tip, 1); 1685 1686 /* check inode formats now that data is flushed */ 1687 error = xfs_swap_extents_check_format(ip, tip); 1688 if (error) { 1689 xfs_notice(mp, 1690 "%s: inode 0x%llx format is incompatible for exchanging.", 1691 __func__, ip->i_ino); 1692 goto out_trans_cancel; 1693 } 1694 1695 /* 1696 * Compare the current change & modify times with that 1697 * passed in. If they differ, we abort this swap. 1698 * This is the mechanism used to ensure the calling 1699 * process that the file was not changed out from 1700 * under it. 1701 */ 1702 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || 1703 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || 1704 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || 1705 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { 1706 error = -EBUSY; 1707 goto out_trans_cancel; 1708 } 1709 1710 /* 1711 * Note the trickiness in setting the log flags - we set the owner log 1712 * flag on the opposite inode (i.e. the inode we are setting the new 1713 * owner to be) because once we swap the forks and log that, log 1714 * recovery is going to see the fork as owned by the swapped inode, 1715 * not the pre-swapped inodes. 1716 */ 1717 src_log_flags = XFS_ILOG_CORE; 1718 target_log_flags = XFS_ILOG_CORE; 1719 1720 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) 1721 error = xfs_swap_extent_rmap(&tp, ip, tip); 1722 else 1723 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags, 1724 &target_log_flags); 1725 if (error) 1726 goto out_trans_cancel; 1727 1728 /* Do we have to swap reflink flags? */ 1729 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^ 1730 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) { 1731 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK; 1732 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 1733 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK; 1734 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK; 1735 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK; 1736 } 1737 1738 /* Swap the cow forks. */ 1739 if (xfs_sb_version_hasreflink(&mp->m_sb)) { 1740 ASSERT(!ip->i_cowfp || 1741 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 1742 ASSERT(!tip->i_cowfp || 1743 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS); 1744 1745 swap(ip->i_cowfp, tip->i_cowfp); 1746 1747 if (ip->i_cowfp && ip->i_cowfp->if_bytes) 1748 xfs_inode_set_cowblocks_tag(ip); 1749 else 1750 xfs_inode_clear_cowblocks_tag(ip); 1751 if (tip->i_cowfp && tip->i_cowfp->if_bytes) 1752 xfs_inode_set_cowblocks_tag(tip); 1753 else 1754 xfs_inode_clear_cowblocks_tag(tip); 1755 } 1756 1757 xfs_trans_log_inode(tp, ip, src_log_flags); 1758 xfs_trans_log_inode(tp, tip, target_log_flags); 1759 1760 /* 1761 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems 1762 * have inode number owner values in the bmbt blocks that still refer to 1763 * the old inode. Scan each bmbt to fix up the owner values with the 1764 * inode number of the current inode. 1765 */ 1766 if (src_log_flags & XFS_ILOG_DOWNER) { 1767 error = xfs_swap_change_owner(&tp, ip, tip); 1768 if (error) 1769 goto out_trans_cancel; 1770 } 1771 if (target_log_flags & XFS_ILOG_DOWNER) { 1772 error = xfs_swap_change_owner(&tp, tip, ip); 1773 if (error) 1774 goto out_trans_cancel; 1775 } 1776 1777 /* 1778 * If this is a synchronous mount, make sure that the 1779 * transaction goes to disk before returning to the user. 1780 */ 1781 if (mp->m_flags & XFS_MOUNT_WSYNC) 1782 xfs_trans_set_sync(tp); 1783 1784 error = xfs_trans_commit(tp); 1785 1786 trace_xfs_swap_extent_after(ip, 0); 1787 trace_xfs_swap_extent_after(tip, 1); 1788 1789 out_unlock: 1790 xfs_iunlock(ip, lock_flags); 1791 xfs_iunlock(tip, lock_flags); 1792 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1793 return error; 1794 1795 out_trans_cancel: 1796 xfs_trans_cancel(tp); 1797 goto out_unlock; 1798 } 1799