1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * Copyright (c) 2012 Red Hat, Inc. 4 * All Rights Reserved. 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License as 8 * published by the Free Software Foundation. 9 * 10 * This program is distributed in the hope that it would be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 18 */ 19 #include "xfs.h" 20 #include "xfs_fs.h" 21 #include "xfs_shared.h" 22 #include "xfs_format.h" 23 #include "xfs_log_format.h" 24 #include "xfs_trans_resv.h" 25 #include "xfs_bit.h" 26 #include "xfs_mount.h" 27 #include "xfs_da_format.h" 28 #include "xfs_defer.h" 29 #include "xfs_inode.h" 30 #include "xfs_btree.h" 31 #include "xfs_trans.h" 32 #include "xfs_extfree_item.h" 33 #include "xfs_alloc.h" 34 #include "xfs_bmap.h" 35 #include "xfs_bmap_util.h" 36 #include "xfs_bmap_btree.h" 37 #include "xfs_rtalloc.h" 38 #include "xfs_error.h" 39 #include "xfs_quota.h" 40 #include "xfs_trans_space.h" 41 #include "xfs_trace.h" 42 #include "xfs_icache.h" 43 #include "xfs_log.h" 44 #include "xfs_rmap_btree.h" 45 #include "xfs_iomap.h" 46 #include "xfs_reflink.h" 47 #include "xfs_refcount.h" 48 49 /* Kernel only BMAP related definitions and functions */ 50 51 /* 52 * Convert the given file system block to a disk block. We have to treat it 53 * differently based on whether the file is a real time file or not, because the 54 * bmap code does. 55 */ 56 xfs_daddr_t 57 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb) 58 { 59 return (XFS_IS_REALTIME_INODE(ip) ? \ 60 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \ 61 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb))); 62 } 63 64 /* 65 * Routine to zero an extent on disk allocated to the specific inode. 66 * 67 * The VFS functions take a linearised filesystem block offset, so we have to 68 * convert the sparse xfs fsb to the right format first. 69 * VFS types are real funky, too. 70 */ 71 int 72 xfs_zero_extent( 73 struct xfs_inode *ip, 74 xfs_fsblock_t start_fsb, 75 xfs_off_t count_fsb) 76 { 77 struct xfs_mount *mp = ip->i_mount; 78 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb); 79 sector_t block = XFS_BB_TO_FSBT(mp, sector); 80 81 return blkdev_issue_zeroout(xfs_find_bdev_for_inode(VFS_I(ip)), 82 block << (mp->m_super->s_blocksize_bits - 9), 83 count_fsb << (mp->m_super->s_blocksize_bits - 9), 84 GFP_NOFS, true); 85 } 86 87 int 88 xfs_bmap_rtalloc( 89 struct xfs_bmalloca *ap) /* bmap alloc argument struct */ 90 { 91 xfs_alloctype_t atype = 0; /* type for allocation routines */ 92 int error; /* error return value */ 93 xfs_mount_t *mp; /* mount point structure */ 94 xfs_extlen_t prod = 0; /* product factor for allocators */ 95 xfs_extlen_t ralen = 0; /* realtime allocation length */ 96 xfs_extlen_t align; /* minimum allocation alignment */ 97 xfs_rtblock_t rtb; 98 99 mp = ap->ip->i_mount; 100 align = xfs_get_extsz_hint(ap->ip); 101 prod = align / mp->m_sb.sb_rextsize; 102 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev, 103 align, 1, ap->eof, 0, 104 ap->conv, &ap->offset, &ap->length); 105 if (error) 106 return error; 107 ASSERT(ap->length); 108 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0); 109 110 /* 111 * If the offset & length are not perfectly aligned 112 * then kill prod, it will just get us in trouble. 113 */ 114 if (do_mod(ap->offset, align) || ap->length % align) 115 prod = 1; 116 /* 117 * Set ralen to be the actual requested length in rtextents. 118 */ 119 ralen = ap->length / mp->m_sb.sb_rextsize; 120 /* 121 * If the old value was close enough to MAXEXTLEN that 122 * we rounded up to it, cut it back so it's valid again. 123 * Note that if it's a really large request (bigger than 124 * MAXEXTLEN), we don't hear about that number, and can't 125 * adjust the starting point to match it. 126 */ 127 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN) 128 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize; 129 130 /* 131 * Lock out modifications to both the RT bitmap and summary inodes 132 */ 133 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP); 134 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL); 135 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM); 136 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL); 137 138 /* 139 * If it's an allocation to an empty file at offset 0, 140 * pick an extent that will space things out in the rt area. 141 */ 142 if (ap->eof && ap->offset == 0) { 143 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */ 144 145 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx); 146 if (error) 147 return error; 148 ap->blkno = rtx * mp->m_sb.sb_rextsize; 149 } else { 150 ap->blkno = 0; 151 } 152 153 xfs_bmap_adjacent(ap); 154 155 /* 156 * Realtime allocation, done through xfs_rtallocate_extent. 157 */ 158 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO; 159 do_div(ap->blkno, mp->m_sb.sb_rextsize); 160 rtb = ap->blkno; 161 ap->length = ralen; 162 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length, 163 &ralen, atype, ap->wasdel, prod, &rtb))) 164 return error; 165 if (rtb == NULLFSBLOCK && prod > 1 && 166 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, 167 ap->length, &ralen, atype, 168 ap->wasdel, 1, &rtb))) 169 return error; 170 ap->blkno = rtb; 171 if (ap->blkno != NULLFSBLOCK) { 172 ap->blkno *= mp->m_sb.sb_rextsize; 173 ralen *= mp->m_sb.sb_rextsize; 174 ap->length = ralen; 175 ap->ip->i_d.di_nblocks += ralen; 176 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE); 177 if (ap->wasdel) 178 ap->ip->i_delayed_blks -= ralen; 179 /* 180 * Adjust the disk quota also. This was reserved 181 * earlier. 182 */ 183 xfs_trans_mod_dquot_byino(ap->tp, ap->ip, 184 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT : 185 XFS_TRANS_DQ_RTBCOUNT, (long) ralen); 186 187 /* Zero the extent if we were asked to do so */ 188 if (ap->datatype & XFS_ALLOC_USERDATA_ZERO) { 189 error = xfs_zero_extent(ap->ip, ap->blkno, ap->length); 190 if (error) 191 return error; 192 } 193 } else { 194 ap->length = 0; 195 } 196 return 0; 197 } 198 199 /* 200 * Check if the endoff is outside the last extent. If so the caller will grow 201 * the allocation to a stripe unit boundary. All offsets are considered outside 202 * the end of file for an empty fork, so 1 is returned in *eof in that case. 203 */ 204 int 205 xfs_bmap_eof( 206 struct xfs_inode *ip, 207 xfs_fileoff_t endoff, 208 int whichfork, 209 int *eof) 210 { 211 struct xfs_bmbt_irec rec; 212 int error; 213 214 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof); 215 if (error || *eof) 216 return error; 217 218 *eof = endoff >= rec.br_startoff + rec.br_blockcount; 219 return 0; 220 } 221 222 /* 223 * Extent tree block counting routines. 224 */ 225 226 /* 227 * Count leaf blocks given a range of extent records. 228 */ 229 STATIC void 230 xfs_bmap_count_leaves( 231 xfs_ifork_t *ifp, 232 xfs_extnum_t idx, 233 int numrecs, 234 int *count) 235 { 236 int b; 237 238 for (b = 0; b < numrecs; b++) { 239 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b); 240 *count += xfs_bmbt_get_blockcount(frp); 241 } 242 } 243 244 /* 245 * Count leaf blocks given a range of extent records originally 246 * in btree format. 247 */ 248 STATIC void 249 xfs_bmap_disk_count_leaves( 250 struct xfs_mount *mp, 251 struct xfs_btree_block *block, 252 int numrecs, 253 int *count) 254 { 255 int b; 256 xfs_bmbt_rec_t *frp; 257 258 for (b = 1; b <= numrecs; b++) { 259 frp = XFS_BMBT_REC_ADDR(mp, block, b); 260 *count += xfs_bmbt_disk_get_blockcount(frp); 261 } 262 } 263 264 /* 265 * Recursively walks each level of a btree 266 * to count total fsblocks in use. 267 */ 268 STATIC int /* error */ 269 xfs_bmap_count_tree( 270 xfs_mount_t *mp, /* file system mount point */ 271 xfs_trans_t *tp, /* transaction pointer */ 272 xfs_ifork_t *ifp, /* inode fork pointer */ 273 xfs_fsblock_t blockno, /* file system block number */ 274 int levelin, /* level in btree */ 275 int *count) /* Count of blocks */ 276 { 277 int error; 278 xfs_buf_t *bp, *nbp; 279 int level = levelin; 280 __be64 *pp; 281 xfs_fsblock_t bno = blockno; 282 xfs_fsblock_t nextbno; 283 struct xfs_btree_block *block, *nextblock; 284 int numrecs; 285 286 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF, 287 &xfs_bmbt_buf_ops); 288 if (error) 289 return error; 290 *count += 1; 291 block = XFS_BUF_TO_BLOCK(bp); 292 293 if (--level) { 294 /* Not at node above leaves, count this level of nodes */ 295 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); 296 while (nextbno != NULLFSBLOCK) { 297 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp, 298 XFS_BMAP_BTREE_REF, 299 &xfs_bmbt_buf_ops); 300 if (error) 301 return error; 302 *count += 1; 303 nextblock = XFS_BUF_TO_BLOCK(nbp); 304 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib); 305 xfs_trans_brelse(tp, nbp); 306 } 307 308 /* Dive to the next level */ 309 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]); 310 bno = be64_to_cpu(*pp); 311 if (unlikely((error = 312 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) { 313 xfs_trans_brelse(tp, bp); 314 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)", 315 XFS_ERRLEVEL_LOW, mp); 316 return -EFSCORRUPTED; 317 } 318 xfs_trans_brelse(tp, bp); 319 } else { 320 /* count all level 1 nodes and their leaves */ 321 for (;;) { 322 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib); 323 numrecs = be16_to_cpu(block->bb_numrecs); 324 xfs_bmap_disk_count_leaves(mp, block, numrecs, count); 325 xfs_trans_brelse(tp, bp); 326 if (nextbno == NULLFSBLOCK) 327 break; 328 bno = nextbno; 329 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, 330 XFS_BMAP_BTREE_REF, 331 &xfs_bmbt_buf_ops); 332 if (error) 333 return error; 334 *count += 1; 335 block = XFS_BUF_TO_BLOCK(bp); 336 } 337 } 338 return 0; 339 } 340 341 /* 342 * Count fsblocks of the given fork. 343 */ 344 static int /* error */ 345 xfs_bmap_count_blocks( 346 xfs_trans_t *tp, /* transaction pointer */ 347 xfs_inode_t *ip, /* incore inode */ 348 int whichfork, /* data or attr fork */ 349 int *count) /* out: count of blocks */ 350 { 351 struct xfs_btree_block *block; /* current btree block */ 352 xfs_fsblock_t bno; /* block # of "block" */ 353 xfs_ifork_t *ifp; /* fork structure */ 354 int level; /* btree level, for checking */ 355 xfs_mount_t *mp; /* file system mount structure */ 356 __be64 *pp; /* pointer to block address */ 357 358 bno = NULLFSBLOCK; 359 mp = ip->i_mount; 360 ifp = XFS_IFORK_PTR(ip, whichfork); 361 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) { 362 xfs_bmap_count_leaves(ifp, 0, xfs_iext_count(ifp), count); 363 return 0; 364 } 365 366 /* 367 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out. 368 */ 369 block = ifp->if_broot; 370 level = be16_to_cpu(block->bb_level); 371 ASSERT(level > 0); 372 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes); 373 bno = be64_to_cpu(*pp); 374 ASSERT(bno != NULLFSBLOCK); 375 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount); 376 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks); 377 378 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) { 379 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW, 380 mp); 381 return -EFSCORRUPTED; 382 } 383 384 return 0; 385 } 386 387 /* 388 * returns 1 for success, 0 if we failed to map the extent. 389 */ 390 STATIC int 391 xfs_getbmapx_fix_eof_hole( 392 xfs_inode_t *ip, /* xfs incore inode pointer */ 393 int whichfork, 394 struct getbmapx *out, /* output structure */ 395 int prealloced, /* this is a file with 396 * preallocated data space */ 397 __int64_t end, /* last block requested */ 398 xfs_fsblock_t startblock, 399 bool moretocome) 400 { 401 __int64_t fixlen; 402 xfs_mount_t *mp; /* file system mount point */ 403 xfs_ifork_t *ifp; /* inode fork pointer */ 404 xfs_extnum_t lastx; /* last extent pointer */ 405 xfs_fileoff_t fileblock; 406 407 if (startblock == HOLESTARTBLOCK) { 408 mp = ip->i_mount; 409 out->bmv_block = -1; 410 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip))); 411 fixlen -= out->bmv_offset; 412 if (prealloced && out->bmv_offset + out->bmv_length == end) { 413 /* Came to hole at EOF. Trim it. */ 414 if (fixlen <= 0) 415 return 0; 416 out->bmv_length = fixlen; 417 } 418 } else { 419 if (startblock == DELAYSTARTBLOCK) 420 out->bmv_block = -2; 421 else 422 out->bmv_block = xfs_fsb_to_db(ip, startblock); 423 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset); 424 ifp = XFS_IFORK_PTR(ip, whichfork); 425 if (!moretocome && 426 xfs_iext_bno_to_ext(ifp, fileblock, &lastx) && 427 (lastx == xfs_iext_count(ifp) - 1)) 428 out->bmv_oflags |= BMV_OF_LAST; 429 } 430 431 return 1; 432 } 433 434 /* Adjust the reported bmap around shared/unshared extent transitions. */ 435 STATIC int 436 xfs_getbmap_adjust_shared( 437 struct xfs_inode *ip, 438 int whichfork, 439 struct xfs_bmbt_irec *map, 440 struct getbmapx *out, 441 struct xfs_bmbt_irec *next_map) 442 { 443 struct xfs_mount *mp = ip->i_mount; 444 xfs_agnumber_t agno; 445 xfs_agblock_t agbno; 446 xfs_agblock_t ebno; 447 xfs_extlen_t elen; 448 xfs_extlen_t nlen; 449 int error; 450 451 next_map->br_startblock = NULLFSBLOCK; 452 next_map->br_startoff = NULLFILEOFF; 453 next_map->br_blockcount = 0; 454 455 /* Only written data blocks can be shared. */ 456 if (!xfs_is_reflink_inode(ip) || whichfork != XFS_DATA_FORK || 457 map->br_startblock == DELAYSTARTBLOCK || 458 map->br_startblock == HOLESTARTBLOCK || 459 ISUNWRITTEN(map)) 460 return 0; 461 462 agno = XFS_FSB_TO_AGNO(mp, map->br_startblock); 463 agbno = XFS_FSB_TO_AGBNO(mp, map->br_startblock); 464 error = xfs_reflink_find_shared(mp, agno, agbno, map->br_blockcount, 465 &ebno, &elen, true); 466 if (error) 467 return error; 468 469 if (ebno == NULLAGBLOCK) { 470 /* No shared blocks at all. */ 471 return 0; 472 } else if (agbno == ebno) { 473 /* 474 * Shared extent at (agbno, elen). Shrink the reported 475 * extent length and prepare to move the start of map[i] 476 * to agbno+elen, with the aim of (re)formatting the new 477 * map[i] the next time through the inner loop. 478 */ 479 out->bmv_length = XFS_FSB_TO_BB(mp, elen); 480 out->bmv_oflags |= BMV_OF_SHARED; 481 if (elen != map->br_blockcount) { 482 *next_map = *map; 483 next_map->br_startblock += elen; 484 next_map->br_startoff += elen; 485 next_map->br_blockcount -= elen; 486 } 487 map->br_blockcount -= elen; 488 } else { 489 /* 490 * There's an unshared extent (agbno, ebno - agbno) 491 * followed by shared extent at (ebno, elen). Shrink 492 * the reported extent length to cover only the unshared 493 * extent and prepare to move up the start of map[i] to 494 * ebno, with the aim of (re)formatting the new map[i] 495 * the next time through the inner loop. 496 */ 497 *next_map = *map; 498 nlen = ebno - agbno; 499 out->bmv_length = XFS_FSB_TO_BB(mp, nlen); 500 next_map->br_startblock += nlen; 501 next_map->br_startoff += nlen; 502 next_map->br_blockcount -= nlen; 503 map->br_blockcount -= nlen; 504 } 505 506 return 0; 507 } 508 509 /* 510 * Get inode's extents as described in bmv, and format for output. 511 * Calls formatter to fill the user's buffer until all extents 512 * are mapped, until the passed-in bmv->bmv_count slots have 513 * been filled, or until the formatter short-circuits the loop, 514 * if it is tracking filled-in extents on its own. 515 */ 516 int /* error code */ 517 xfs_getbmap( 518 xfs_inode_t *ip, 519 struct getbmapx *bmv, /* user bmap structure */ 520 xfs_bmap_format_t formatter, /* format to user */ 521 void *arg) /* formatter arg */ 522 { 523 __int64_t bmvend; /* last block requested */ 524 int error = 0; /* return value */ 525 __int64_t fixlen; /* length for -1 case */ 526 int i; /* extent number */ 527 int lock; /* lock state */ 528 xfs_bmbt_irec_t *map; /* buffer for user's data */ 529 xfs_mount_t *mp; /* file system mount point */ 530 int nex; /* # of user extents can do */ 531 int nexleft; /* # of user extents left */ 532 int subnex; /* # of bmapi's can do */ 533 int nmap; /* number of map entries */ 534 struct getbmapx *out; /* output structure */ 535 int whichfork; /* data or attr fork */ 536 int prealloced; /* this is a file with 537 * preallocated data space */ 538 int iflags; /* interface flags */ 539 int bmapi_flags; /* flags for xfs_bmapi */ 540 int cur_ext = 0; 541 struct xfs_bmbt_irec inject_map; 542 543 mp = ip->i_mount; 544 iflags = bmv->bmv_iflags; 545 546 #ifndef DEBUG 547 /* Only allow CoW fork queries if we're debugging. */ 548 if (iflags & BMV_IF_COWFORK) 549 return -EINVAL; 550 #endif 551 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK)) 552 return -EINVAL; 553 554 if (iflags & BMV_IF_ATTRFORK) 555 whichfork = XFS_ATTR_FORK; 556 else if (iflags & BMV_IF_COWFORK) 557 whichfork = XFS_COW_FORK; 558 else 559 whichfork = XFS_DATA_FORK; 560 561 switch (whichfork) { 562 case XFS_ATTR_FORK: 563 if (XFS_IFORK_Q(ip)) { 564 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS && 565 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE && 566 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) 567 return -EINVAL; 568 } else if (unlikely( 569 ip->i_d.di_aformat != 0 && 570 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) { 571 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW, 572 ip->i_mount); 573 return -EFSCORRUPTED; 574 } 575 576 prealloced = 0; 577 fixlen = 1LL << 32; 578 break; 579 case XFS_COW_FORK: 580 if (ip->i_cformat != XFS_DINODE_FMT_EXTENTS) 581 return -EINVAL; 582 583 if (xfs_get_cowextsz_hint(ip)) { 584 prealloced = 1; 585 fixlen = mp->m_super->s_maxbytes; 586 } else { 587 prealloced = 0; 588 fixlen = XFS_ISIZE(ip); 589 } 590 break; 591 default: 592 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS && 593 ip->i_d.di_format != XFS_DINODE_FMT_BTREE && 594 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL) 595 return -EINVAL; 596 597 if (xfs_get_extsz_hint(ip) || 598 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){ 599 prealloced = 1; 600 fixlen = mp->m_super->s_maxbytes; 601 } else { 602 prealloced = 0; 603 fixlen = XFS_ISIZE(ip); 604 } 605 break; 606 } 607 608 if (bmv->bmv_length == -1) { 609 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen)); 610 bmv->bmv_length = 611 max_t(__int64_t, fixlen - bmv->bmv_offset, 0); 612 } else if (bmv->bmv_length == 0) { 613 bmv->bmv_entries = 0; 614 return 0; 615 } else if (bmv->bmv_length < 0) { 616 return -EINVAL; 617 } 618 619 nex = bmv->bmv_count - 1; 620 if (nex <= 0) 621 return -EINVAL; 622 bmvend = bmv->bmv_offset + bmv->bmv_length; 623 624 625 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx)) 626 return -ENOMEM; 627 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0); 628 if (!out) 629 return -ENOMEM; 630 631 xfs_ilock(ip, XFS_IOLOCK_SHARED); 632 switch (whichfork) { 633 case XFS_DATA_FORK: 634 if (!(iflags & BMV_IF_DELALLOC) && 635 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) { 636 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 637 if (error) 638 goto out_unlock_iolock; 639 640 /* 641 * Even after flushing the inode, there can still be 642 * delalloc blocks on the inode beyond EOF due to 643 * speculative preallocation. These are not removed 644 * until the release function is called or the inode 645 * is inactivated. Hence we cannot assert here that 646 * ip->i_delayed_blks == 0. 647 */ 648 } 649 650 lock = xfs_ilock_data_map_shared(ip); 651 break; 652 case XFS_COW_FORK: 653 lock = XFS_ILOCK_SHARED; 654 xfs_ilock(ip, lock); 655 break; 656 case XFS_ATTR_FORK: 657 lock = xfs_ilock_attr_map_shared(ip); 658 break; 659 } 660 661 /* 662 * Don't let nex be bigger than the number of extents 663 * we can have assuming alternating holes and real extents. 664 */ 665 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1) 666 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1; 667 668 bmapi_flags = xfs_bmapi_aflag(whichfork); 669 if (!(iflags & BMV_IF_PREALLOC)) 670 bmapi_flags |= XFS_BMAPI_IGSTATE; 671 672 /* 673 * Allocate enough space to handle "subnex" maps at a time. 674 */ 675 error = -ENOMEM; 676 subnex = 16; 677 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS); 678 if (!map) 679 goto out_unlock_ilock; 680 681 bmv->bmv_entries = 0; 682 683 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 && 684 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) { 685 error = 0; 686 goto out_free_map; 687 } 688 689 nexleft = nex; 690 691 do { 692 nmap = (nexleft > subnex) ? subnex : nexleft; 693 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset), 694 XFS_BB_TO_FSB(mp, bmv->bmv_length), 695 map, &nmap, bmapi_flags); 696 if (error) 697 goto out_free_map; 698 ASSERT(nmap <= subnex); 699 700 for (i = 0; i < nmap && nexleft && bmv->bmv_length && 701 cur_ext < bmv->bmv_count; i++) { 702 out[cur_ext].bmv_oflags = 0; 703 if (map[i].br_state == XFS_EXT_UNWRITTEN) 704 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC; 705 else if (map[i].br_startblock == DELAYSTARTBLOCK) 706 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC; 707 out[cur_ext].bmv_offset = 708 XFS_FSB_TO_BB(mp, map[i].br_startoff); 709 out[cur_ext].bmv_length = 710 XFS_FSB_TO_BB(mp, map[i].br_blockcount); 711 out[cur_ext].bmv_unused1 = 0; 712 out[cur_ext].bmv_unused2 = 0; 713 714 /* 715 * delayed allocation extents that start beyond EOF can 716 * occur due to speculative EOF allocation when the 717 * delalloc extent is larger than the largest freespace 718 * extent at conversion time. These extents cannot be 719 * converted by data writeback, so can exist here even 720 * if we are not supposed to be finding delalloc 721 * extents. 722 */ 723 if (map[i].br_startblock == DELAYSTARTBLOCK && 724 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip))) 725 ASSERT((iflags & BMV_IF_DELALLOC) != 0); 726 727 if (map[i].br_startblock == HOLESTARTBLOCK && 728 whichfork == XFS_ATTR_FORK) { 729 /* came to the end of attribute fork */ 730 out[cur_ext].bmv_oflags |= BMV_OF_LAST; 731 goto out_free_map; 732 } 733 734 /* Is this a shared block? */ 735 error = xfs_getbmap_adjust_shared(ip, whichfork, 736 &map[i], &out[cur_ext], &inject_map); 737 if (error) 738 goto out_free_map; 739 740 if (!xfs_getbmapx_fix_eof_hole(ip, whichfork, 741 &out[cur_ext], prealloced, bmvend, 742 map[i].br_startblock, 743 inject_map.br_startblock != NULLFSBLOCK)) 744 goto out_free_map; 745 746 bmv->bmv_offset = 747 out[cur_ext].bmv_offset + 748 out[cur_ext].bmv_length; 749 bmv->bmv_length = 750 max_t(__int64_t, 0, bmvend - bmv->bmv_offset); 751 752 /* 753 * In case we don't want to return the hole, 754 * don't increase cur_ext so that we can reuse 755 * it in the next loop. 756 */ 757 if ((iflags & BMV_IF_NO_HOLES) && 758 map[i].br_startblock == HOLESTARTBLOCK) { 759 memset(&out[cur_ext], 0, sizeof(out[cur_ext])); 760 continue; 761 } 762 763 if (inject_map.br_startblock != NULLFSBLOCK) { 764 map[i] = inject_map; 765 i--; 766 } else 767 nexleft--; 768 bmv->bmv_entries++; 769 cur_ext++; 770 } 771 } while (nmap && nexleft && bmv->bmv_length && 772 cur_ext < bmv->bmv_count); 773 774 out_free_map: 775 kmem_free(map); 776 out_unlock_ilock: 777 xfs_iunlock(ip, lock); 778 out_unlock_iolock: 779 xfs_iunlock(ip, XFS_IOLOCK_SHARED); 780 781 for (i = 0; i < cur_ext; i++) { 782 int full = 0; /* user array is full */ 783 784 /* format results & advance arg */ 785 error = formatter(&arg, &out[i], &full); 786 if (error || full) 787 break; 788 } 789 790 kmem_free(out); 791 return error; 792 } 793 794 /* 795 * dead simple method of punching delalyed allocation blocks from a range in 796 * the inode. Walks a block at a time so will be slow, but is only executed in 797 * rare error cases so the overhead is not critical. This will always punch out 798 * both the start and end blocks, even if the ranges only partially overlap 799 * them, so it is up to the caller to ensure that partial blocks are not 800 * passed in. 801 */ 802 int 803 xfs_bmap_punch_delalloc_range( 804 struct xfs_inode *ip, 805 xfs_fileoff_t start_fsb, 806 xfs_fileoff_t length) 807 { 808 xfs_fileoff_t remaining = length; 809 int error = 0; 810 811 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 812 813 do { 814 int done; 815 xfs_bmbt_irec_t imap; 816 int nimaps = 1; 817 xfs_fsblock_t firstblock; 818 struct xfs_defer_ops dfops; 819 820 /* 821 * Map the range first and check that it is a delalloc extent 822 * before trying to unmap the range. Otherwise we will be 823 * trying to remove a real extent (which requires a 824 * transaction) or a hole, which is probably a bad idea... 825 */ 826 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps, 827 XFS_BMAPI_ENTIRE); 828 829 if (error) { 830 /* something screwed, just bail */ 831 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { 832 xfs_alert(ip->i_mount, 833 "Failed delalloc mapping lookup ino %lld fsb %lld.", 834 ip->i_ino, start_fsb); 835 } 836 break; 837 } 838 if (!nimaps) { 839 /* nothing there */ 840 goto next_block; 841 } 842 if (imap.br_startblock != DELAYSTARTBLOCK) { 843 /* been converted, ignore */ 844 goto next_block; 845 } 846 WARN_ON(imap.br_blockcount == 0); 847 848 /* 849 * Note: while we initialise the firstblock/dfops pair, they 850 * should never be used because blocks should never be 851 * allocated or freed for a delalloc extent and hence we need 852 * don't cancel or finish them after the xfs_bunmapi() call. 853 */ 854 xfs_defer_init(&dfops, &firstblock); 855 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock, 856 &dfops, &done); 857 if (error) 858 break; 859 860 ASSERT(!xfs_defer_has_unfinished_work(&dfops)); 861 next_block: 862 start_fsb++; 863 remaining--; 864 } while(remaining > 0); 865 866 return error; 867 } 868 869 /* 870 * Test whether it is appropriate to check an inode for and free post EOF 871 * blocks. The 'force' parameter determines whether we should also consider 872 * regular files that are marked preallocated or append-only. 873 */ 874 bool 875 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force) 876 { 877 /* prealloc/delalloc exists only on regular files */ 878 if (!S_ISREG(VFS_I(ip)->i_mode)) 879 return false; 880 881 /* 882 * Zero sized files with no cached pages and delalloc blocks will not 883 * have speculative prealloc/delalloc blocks to remove. 884 */ 885 if (VFS_I(ip)->i_size == 0 && 886 VFS_I(ip)->i_mapping->nrpages == 0 && 887 ip->i_delayed_blks == 0) 888 return false; 889 890 /* If we haven't read in the extent list, then don't do it now. */ 891 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) 892 return false; 893 894 /* 895 * Do not free real preallocated or append-only files unless the file 896 * has delalloc blocks and we are forced to remove them. 897 */ 898 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)) 899 if (!force || ip->i_delayed_blks == 0) 900 return false; 901 902 return true; 903 } 904 905 /* 906 * This is called by xfs_inactive to free any blocks beyond eof 907 * when the link count isn't zero and by xfs_dm_punch_hole() when 908 * punching a hole to EOF. 909 */ 910 int 911 xfs_free_eofblocks( 912 xfs_mount_t *mp, 913 xfs_inode_t *ip, 914 bool need_iolock) 915 { 916 xfs_trans_t *tp; 917 int error; 918 xfs_fileoff_t end_fsb; 919 xfs_fileoff_t last_fsb; 920 xfs_filblks_t map_len; 921 int nimaps; 922 xfs_bmbt_irec_t imap; 923 924 /* 925 * Figure out if there are any blocks beyond the end 926 * of the file. If not, then there is nothing to do. 927 */ 928 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip)); 929 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 930 if (last_fsb <= end_fsb) 931 return 0; 932 map_len = last_fsb - end_fsb; 933 934 nimaps = 1; 935 xfs_ilock(ip, XFS_ILOCK_SHARED); 936 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0); 937 xfs_iunlock(ip, XFS_ILOCK_SHARED); 938 939 if (!error && (nimaps != 0) && 940 (imap.br_startblock != HOLESTARTBLOCK || 941 ip->i_delayed_blks)) { 942 /* 943 * Attach the dquots to the inode up front. 944 */ 945 error = xfs_qm_dqattach(ip, 0); 946 if (error) 947 return error; 948 949 /* 950 * There are blocks after the end of file. 951 * Free them up now by truncating the file to 952 * its current size. 953 */ 954 if (need_iolock) { 955 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) 956 return -EAGAIN; 957 } 958 959 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, 960 &tp); 961 if (error) { 962 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 963 if (need_iolock) 964 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 965 return error; 966 } 967 968 xfs_ilock(ip, XFS_ILOCK_EXCL); 969 xfs_trans_ijoin(tp, ip, 0); 970 971 /* 972 * Do not update the on-disk file size. If we update the 973 * on-disk file size and then the system crashes before the 974 * contents of the file are flushed to disk then the files 975 * may be full of holes (ie NULL files bug). 976 */ 977 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 978 XFS_ISIZE(ip)); 979 if (error) { 980 /* 981 * If we get an error at this point we simply don't 982 * bother truncating the file. 983 */ 984 xfs_trans_cancel(tp); 985 } else { 986 error = xfs_trans_commit(tp); 987 if (!error) 988 xfs_inode_clear_eofblocks_tag(ip); 989 } 990 991 xfs_iunlock(ip, XFS_ILOCK_EXCL); 992 if (need_iolock) 993 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 994 } 995 return error; 996 } 997 998 int 999 xfs_alloc_file_space( 1000 struct xfs_inode *ip, 1001 xfs_off_t offset, 1002 xfs_off_t len, 1003 int alloc_type) 1004 { 1005 xfs_mount_t *mp = ip->i_mount; 1006 xfs_off_t count; 1007 xfs_filblks_t allocated_fsb; 1008 xfs_filblks_t allocatesize_fsb; 1009 xfs_extlen_t extsz, temp; 1010 xfs_fileoff_t startoffset_fsb; 1011 xfs_fsblock_t firstfsb; 1012 int nimaps; 1013 int quota_flag; 1014 int rt; 1015 xfs_trans_t *tp; 1016 xfs_bmbt_irec_t imaps[1], *imapp; 1017 struct xfs_defer_ops dfops; 1018 uint qblocks, resblks, resrtextents; 1019 int error; 1020 1021 trace_xfs_alloc_file_space(ip); 1022 1023 if (XFS_FORCED_SHUTDOWN(mp)) 1024 return -EIO; 1025 1026 error = xfs_qm_dqattach(ip, 0); 1027 if (error) 1028 return error; 1029 1030 if (len <= 0) 1031 return -EINVAL; 1032 1033 rt = XFS_IS_REALTIME_INODE(ip); 1034 extsz = xfs_get_extsz_hint(ip); 1035 1036 count = len; 1037 imapp = &imaps[0]; 1038 nimaps = 1; 1039 startoffset_fsb = XFS_B_TO_FSBT(mp, offset); 1040 allocatesize_fsb = XFS_B_TO_FSB(mp, count); 1041 1042 /* 1043 * Allocate file space until done or until there is an error 1044 */ 1045 while (allocatesize_fsb && !error) { 1046 xfs_fileoff_t s, e; 1047 1048 /* 1049 * Determine space reservations for data/realtime. 1050 */ 1051 if (unlikely(extsz)) { 1052 s = startoffset_fsb; 1053 do_div(s, extsz); 1054 s *= extsz; 1055 e = startoffset_fsb + allocatesize_fsb; 1056 if ((temp = do_mod(startoffset_fsb, extsz))) 1057 e += temp; 1058 if ((temp = do_mod(e, extsz))) 1059 e += extsz - temp; 1060 } else { 1061 s = 0; 1062 e = allocatesize_fsb; 1063 } 1064 1065 /* 1066 * The transaction reservation is limited to a 32-bit block 1067 * count, hence we need to limit the number of blocks we are 1068 * trying to reserve to avoid an overflow. We can't allocate 1069 * more than @nimaps extents, and an extent is limited on disk 1070 * to MAXEXTLEN (21 bits), so use that to enforce the limit. 1071 */ 1072 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps)); 1073 if (unlikely(rt)) { 1074 resrtextents = qblocks = resblks; 1075 resrtextents /= mp->m_sb.sb_rextsize; 1076 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 1077 quota_flag = XFS_QMOPT_RES_RTBLKS; 1078 } else { 1079 resrtextents = 0; 1080 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks); 1081 quota_flag = XFS_QMOPT_RES_REGBLKS; 1082 } 1083 1084 /* 1085 * Allocate and setup the transaction. 1086 */ 1087 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 1088 resrtextents, 0, &tp); 1089 1090 /* 1091 * Check for running out of space 1092 */ 1093 if (error) { 1094 /* 1095 * Free the transaction structure. 1096 */ 1097 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); 1098 break; 1099 } 1100 xfs_ilock(ip, XFS_ILOCK_EXCL); 1101 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 1102 0, quota_flag); 1103 if (error) 1104 goto error1; 1105 1106 xfs_trans_ijoin(tp, ip, 0); 1107 1108 xfs_defer_init(&dfops, &firstfsb); 1109 error = xfs_bmapi_write(tp, ip, startoffset_fsb, 1110 allocatesize_fsb, alloc_type, &firstfsb, 1111 resblks, imapp, &nimaps, &dfops); 1112 if (error) 1113 goto error0; 1114 1115 /* 1116 * Complete the transaction 1117 */ 1118 error = xfs_defer_finish(&tp, &dfops, NULL); 1119 if (error) 1120 goto error0; 1121 1122 error = xfs_trans_commit(tp); 1123 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1124 if (error) 1125 break; 1126 1127 allocated_fsb = imapp->br_blockcount; 1128 1129 if (nimaps == 0) { 1130 error = -ENOSPC; 1131 break; 1132 } 1133 1134 startoffset_fsb += allocated_fsb; 1135 allocatesize_fsb -= allocated_fsb; 1136 } 1137 1138 return error; 1139 1140 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */ 1141 xfs_defer_cancel(&dfops); 1142 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); 1143 1144 error1: /* Just cancel transaction */ 1145 xfs_trans_cancel(tp); 1146 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1147 return error; 1148 } 1149 1150 static int 1151 xfs_unmap_extent( 1152 struct xfs_inode *ip, 1153 xfs_fileoff_t startoffset_fsb, 1154 xfs_filblks_t len_fsb, 1155 int *done) 1156 { 1157 struct xfs_mount *mp = ip->i_mount; 1158 struct xfs_trans *tp; 1159 struct xfs_defer_ops dfops; 1160 xfs_fsblock_t firstfsb; 1161 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 1162 int error; 1163 1164 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp); 1165 if (error) { 1166 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp)); 1167 return error; 1168 } 1169 1170 xfs_ilock(ip, XFS_ILOCK_EXCL); 1171 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, ip->i_gdquot, 1172 ip->i_pdquot, resblks, 0, XFS_QMOPT_RES_REGBLKS); 1173 if (error) 1174 goto out_trans_cancel; 1175 1176 xfs_trans_ijoin(tp, ip, 0); 1177 1178 xfs_defer_init(&dfops, &firstfsb); 1179 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, &firstfsb, 1180 &dfops, done); 1181 if (error) 1182 goto out_bmap_cancel; 1183 1184 error = xfs_defer_finish(&tp, &dfops, ip); 1185 if (error) 1186 goto out_bmap_cancel; 1187 1188 error = xfs_trans_commit(tp); 1189 out_unlock: 1190 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1191 return error; 1192 1193 out_bmap_cancel: 1194 xfs_defer_cancel(&dfops); 1195 out_trans_cancel: 1196 xfs_trans_cancel(tp); 1197 goto out_unlock; 1198 } 1199 1200 static int 1201 xfs_adjust_extent_unmap_boundaries( 1202 struct xfs_inode *ip, 1203 xfs_fileoff_t *startoffset_fsb, 1204 xfs_fileoff_t *endoffset_fsb) 1205 { 1206 struct xfs_mount *mp = ip->i_mount; 1207 struct xfs_bmbt_irec imap; 1208 int nimap, error; 1209 xfs_extlen_t mod = 0; 1210 1211 nimap = 1; 1212 error = xfs_bmapi_read(ip, *startoffset_fsb, 1, &imap, &nimap, 0); 1213 if (error) 1214 return error; 1215 1216 if (nimap && imap.br_startblock != HOLESTARTBLOCK) { 1217 xfs_daddr_t block; 1218 1219 ASSERT(imap.br_startblock != DELAYSTARTBLOCK); 1220 block = imap.br_startblock; 1221 mod = do_div(block, mp->m_sb.sb_rextsize); 1222 if (mod) 1223 *startoffset_fsb += mp->m_sb.sb_rextsize - mod; 1224 } 1225 1226 nimap = 1; 1227 error = xfs_bmapi_read(ip, *endoffset_fsb - 1, 1, &imap, &nimap, 0); 1228 if (error) 1229 return error; 1230 1231 if (nimap && imap.br_startblock != HOLESTARTBLOCK) { 1232 ASSERT(imap.br_startblock != DELAYSTARTBLOCK); 1233 mod++; 1234 if (mod && mod != mp->m_sb.sb_rextsize) 1235 *endoffset_fsb -= mod; 1236 } 1237 1238 return 0; 1239 } 1240 1241 static int 1242 xfs_flush_unmap_range( 1243 struct xfs_inode *ip, 1244 xfs_off_t offset, 1245 xfs_off_t len) 1246 { 1247 struct xfs_mount *mp = ip->i_mount; 1248 struct inode *inode = VFS_I(ip); 1249 xfs_off_t rounding, start, end; 1250 int error; 1251 1252 /* wait for the completion of any pending DIOs */ 1253 inode_dio_wait(inode); 1254 1255 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE); 1256 start = round_down(offset, rounding); 1257 end = round_up(offset + len, rounding) - 1; 1258 1259 error = filemap_write_and_wait_range(inode->i_mapping, start, end); 1260 if (error) 1261 return error; 1262 truncate_pagecache_range(inode, start, end); 1263 return 0; 1264 } 1265 1266 int 1267 xfs_free_file_space( 1268 struct xfs_inode *ip, 1269 xfs_off_t offset, 1270 xfs_off_t len) 1271 { 1272 struct xfs_mount *mp = ip->i_mount; 1273 xfs_fileoff_t startoffset_fsb; 1274 xfs_fileoff_t endoffset_fsb; 1275 int done = 0, error; 1276 1277 trace_xfs_free_file_space(ip); 1278 1279 error = xfs_qm_dqattach(ip, 0); 1280 if (error) 1281 return error; 1282 1283 if (len <= 0) /* if nothing being freed */ 1284 return 0; 1285 1286 error = xfs_flush_unmap_range(ip, offset, len); 1287 if (error) 1288 return error; 1289 1290 startoffset_fsb = XFS_B_TO_FSB(mp, offset); 1291 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len); 1292 1293 /* 1294 * Need to zero the stuff we're not freeing, on disk. If it's a RT file 1295 * and we can't use unwritten extents then we actually need to ensure 1296 * to zero the whole extent, otherwise we just need to take of block 1297 * boundaries, and xfs_bunmapi will handle the rest. 1298 */ 1299 if (XFS_IS_REALTIME_INODE(ip) && 1300 !xfs_sb_version_hasextflgbit(&mp->m_sb)) { 1301 error = xfs_adjust_extent_unmap_boundaries(ip, &startoffset_fsb, 1302 &endoffset_fsb); 1303 if (error) 1304 return error; 1305 } 1306 1307 if (endoffset_fsb > startoffset_fsb) { 1308 while (!done) { 1309 error = xfs_unmap_extent(ip, startoffset_fsb, 1310 endoffset_fsb - startoffset_fsb, &done); 1311 if (error) 1312 return error; 1313 } 1314 } 1315 1316 /* 1317 * Now that we've unmap all full blocks we'll have to zero out any 1318 * partial block at the beginning and/or end. xfs_zero_range is 1319 * smart enough to skip any holes, including those we just created. 1320 */ 1321 return xfs_zero_range(ip, offset, len, NULL); 1322 } 1323 1324 /* 1325 * Preallocate and zero a range of a file. This mechanism has the allocation 1326 * semantics of fallocate and in addition converts data in the range to zeroes. 1327 */ 1328 int 1329 xfs_zero_file_space( 1330 struct xfs_inode *ip, 1331 xfs_off_t offset, 1332 xfs_off_t len) 1333 { 1334 struct xfs_mount *mp = ip->i_mount; 1335 uint blksize; 1336 int error; 1337 1338 trace_xfs_zero_file_space(ip); 1339 1340 blksize = 1 << mp->m_sb.sb_blocklog; 1341 1342 /* 1343 * Punch a hole and prealloc the range. We use hole punch rather than 1344 * unwritten extent conversion for two reasons: 1345 * 1346 * 1.) Hole punch handles partial block zeroing for us. 1347 * 1348 * 2.) If prealloc returns ENOSPC, the file range is still zero-valued 1349 * by virtue of the hole punch. 1350 */ 1351 error = xfs_free_file_space(ip, offset, len); 1352 if (error) 1353 goto out; 1354 1355 error = xfs_alloc_file_space(ip, round_down(offset, blksize), 1356 round_up(offset + len, blksize) - 1357 round_down(offset, blksize), 1358 XFS_BMAPI_PREALLOC); 1359 out: 1360 return error; 1361 1362 } 1363 1364 /* 1365 * @next_fsb will keep track of the extent currently undergoing shift. 1366 * @stop_fsb will keep track of the extent at which we have to stop. 1367 * If we are shifting left, we will start with block (offset + len) and 1368 * shift each extent till last extent. 1369 * If we are shifting right, we will start with last extent inside file space 1370 * and continue until we reach the block corresponding to offset. 1371 */ 1372 static int 1373 xfs_shift_file_space( 1374 struct xfs_inode *ip, 1375 xfs_off_t offset, 1376 xfs_off_t len, 1377 enum shift_direction direction) 1378 { 1379 int done = 0; 1380 struct xfs_mount *mp = ip->i_mount; 1381 struct xfs_trans *tp; 1382 int error; 1383 struct xfs_defer_ops dfops; 1384 xfs_fsblock_t first_block; 1385 xfs_fileoff_t stop_fsb; 1386 xfs_fileoff_t next_fsb; 1387 xfs_fileoff_t shift_fsb; 1388 1389 ASSERT(direction == SHIFT_LEFT || direction == SHIFT_RIGHT); 1390 1391 if (direction == SHIFT_LEFT) { 1392 next_fsb = XFS_B_TO_FSB(mp, offset + len); 1393 stop_fsb = XFS_B_TO_FSB(mp, VFS_I(ip)->i_size); 1394 } else { 1395 /* 1396 * If right shift, delegate the work of initialization of 1397 * next_fsb to xfs_bmap_shift_extent as it has ilock held. 1398 */ 1399 next_fsb = NULLFSBLOCK; 1400 stop_fsb = XFS_B_TO_FSB(mp, offset); 1401 } 1402 1403 shift_fsb = XFS_B_TO_FSB(mp, len); 1404 1405 /* 1406 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation 1407 * into the accessible region of the file. 1408 */ 1409 if (xfs_can_free_eofblocks(ip, true)) { 1410 error = xfs_free_eofblocks(mp, ip, false); 1411 if (error) 1412 return error; 1413 } 1414 1415 /* 1416 * Writeback and invalidate cache for the remainder of the file as we're 1417 * about to shift down every extent from offset to EOF. 1418 */ 1419 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 1420 offset, -1); 1421 if (error) 1422 return error; 1423 error = invalidate_inode_pages2_range(VFS_I(ip)->i_mapping, 1424 offset >> PAGE_SHIFT, -1); 1425 if (error) 1426 return error; 1427 1428 /* 1429 * The extent shiting code works on extent granularity. So, if 1430 * stop_fsb is not the starting block of extent, we need to split 1431 * the extent at stop_fsb. 1432 */ 1433 if (direction == SHIFT_RIGHT) { 1434 error = xfs_bmap_split_extent(ip, stop_fsb); 1435 if (error) 1436 return error; 1437 } 1438 1439 while (!error && !done) { 1440 /* 1441 * We would need to reserve permanent block for transaction. 1442 * This will come into picture when after shifting extent into 1443 * hole we found that adjacent extents can be merged which 1444 * may lead to freeing of a block during record update. 1445 */ 1446 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 1447 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp); 1448 if (error) 1449 break; 1450 1451 xfs_ilock(ip, XFS_ILOCK_EXCL); 1452 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot, 1453 ip->i_gdquot, ip->i_pdquot, 1454 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 1455 XFS_QMOPT_RES_REGBLKS); 1456 if (error) 1457 goto out_trans_cancel; 1458 1459 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1460 1461 xfs_defer_init(&dfops, &first_block); 1462 1463 /* 1464 * We are using the write transaction in which max 2 bmbt 1465 * updates are allowed 1466 */ 1467 error = xfs_bmap_shift_extents(tp, ip, &next_fsb, shift_fsb, 1468 &done, stop_fsb, &first_block, &dfops, 1469 direction, XFS_BMAP_MAX_SHIFT_EXTENTS); 1470 if (error) 1471 goto out_bmap_cancel; 1472 1473 error = xfs_defer_finish(&tp, &dfops, NULL); 1474 if (error) 1475 goto out_bmap_cancel; 1476 1477 error = xfs_trans_commit(tp); 1478 } 1479 1480 return error; 1481 1482 out_bmap_cancel: 1483 xfs_defer_cancel(&dfops); 1484 out_trans_cancel: 1485 xfs_trans_cancel(tp); 1486 return error; 1487 } 1488 1489 /* 1490 * xfs_collapse_file_space() 1491 * This routine frees disk space and shift extent for the given file. 1492 * The first thing we do is to free data blocks in the specified range 1493 * by calling xfs_free_file_space(). It would also sync dirty data 1494 * and invalidate page cache over the region on which collapse range 1495 * is working. And Shift extent records to the left to cover a hole. 1496 * RETURNS: 1497 * 0 on success 1498 * errno on error 1499 * 1500 */ 1501 int 1502 xfs_collapse_file_space( 1503 struct xfs_inode *ip, 1504 xfs_off_t offset, 1505 xfs_off_t len) 1506 { 1507 int error; 1508 1509 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1510 trace_xfs_collapse_file_space(ip); 1511 1512 error = xfs_free_file_space(ip, offset, len); 1513 if (error) 1514 return error; 1515 1516 return xfs_shift_file_space(ip, offset, len, SHIFT_LEFT); 1517 } 1518 1519 /* 1520 * xfs_insert_file_space() 1521 * This routine create hole space by shifting extents for the given file. 1522 * The first thing we do is to sync dirty data and invalidate page cache 1523 * over the region on which insert range is working. And split an extent 1524 * to two extents at given offset by calling xfs_bmap_split_extent. 1525 * And shift all extent records which are laying between [offset, 1526 * last allocated extent] to the right to reserve hole range. 1527 * RETURNS: 1528 * 0 on success 1529 * errno on error 1530 */ 1531 int 1532 xfs_insert_file_space( 1533 struct xfs_inode *ip, 1534 loff_t offset, 1535 loff_t len) 1536 { 1537 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 1538 trace_xfs_insert_file_space(ip); 1539 1540 return xfs_shift_file_space(ip, offset, len, SHIFT_RIGHT); 1541 } 1542 1543 /* 1544 * We need to check that the format of the data fork in the temporary inode is 1545 * valid for the target inode before doing the swap. This is not a problem with 1546 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized 1547 * data fork depending on the space the attribute fork is taking so we can get 1548 * invalid formats on the target inode. 1549 * 1550 * E.g. target has space for 7 extents in extent format, temp inode only has 1551 * space for 6. If we defragment down to 7 extents, then the tmp format is a 1552 * btree, but when swapped it needs to be in extent format. Hence we can't just 1553 * blindly swap data forks on attr2 filesystems. 1554 * 1555 * Note that we check the swap in both directions so that we don't end up with 1556 * a corrupt temporary inode, either. 1557 * 1558 * Note that fixing the way xfs_fsr sets up the attribute fork in the source 1559 * inode will prevent this situation from occurring, so all we do here is 1560 * reject and log the attempt. basically we are putting the responsibility on 1561 * userspace to get this right. 1562 */ 1563 static int 1564 xfs_swap_extents_check_format( 1565 struct xfs_inode *ip, /* target inode */ 1566 struct xfs_inode *tip) /* tmp inode */ 1567 { 1568 1569 /* Should never get a local format */ 1570 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL || 1571 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL) 1572 return -EINVAL; 1573 1574 /* 1575 * if the target inode has less extents that then temporary inode then 1576 * why did userspace call us? 1577 */ 1578 if (ip->i_d.di_nextents < tip->i_d.di_nextents) 1579 return -EINVAL; 1580 1581 /* 1582 * If we have to use the (expensive) rmap swap method, we can 1583 * handle any number of extents and any format. 1584 */ 1585 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb)) 1586 return 0; 1587 1588 /* 1589 * if the target inode is in extent form and the temp inode is in btree 1590 * form then we will end up with the target inode in the wrong format 1591 * as we already know there are less extents in the temp inode. 1592 */ 1593 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && 1594 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) 1595 return -EINVAL; 1596 1597 /* Check temp in extent form to max in target */ 1598 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && 1599 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) > 1600 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1601 return -EINVAL; 1602 1603 /* Check target in extent form to max in temp */ 1604 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS && 1605 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) > 1606 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1607 return -EINVAL; 1608 1609 /* 1610 * If we are in a btree format, check that the temp root block will fit 1611 * in the target and that it has enough extents to be in btree format 1612 * in the target. 1613 * 1614 * Note that we have to be careful to allow btree->extent conversions 1615 * (a common defrag case) which will occur when the temp inode is in 1616 * extent format... 1617 */ 1618 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { 1619 if (XFS_IFORK_BOFF(ip) && 1620 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip)) 1621 return -EINVAL; 1622 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <= 1623 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK)) 1624 return -EINVAL; 1625 } 1626 1627 /* Reciprocal target->temp btree format checks */ 1628 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { 1629 if (XFS_IFORK_BOFF(tip) && 1630 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip)) 1631 return -EINVAL; 1632 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <= 1633 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK)) 1634 return -EINVAL; 1635 } 1636 1637 return 0; 1638 } 1639 1640 static int 1641 xfs_swap_extent_flush( 1642 struct xfs_inode *ip) 1643 { 1644 int error; 1645 1646 error = filemap_write_and_wait(VFS_I(ip)->i_mapping); 1647 if (error) 1648 return error; 1649 truncate_pagecache_range(VFS_I(ip), 0, -1); 1650 1651 /* Verify O_DIRECT for ftmp */ 1652 if (VFS_I(ip)->i_mapping->nrpages) 1653 return -EINVAL; 1654 return 0; 1655 } 1656 1657 /* 1658 * Move extents from one file to another, when rmap is enabled. 1659 */ 1660 STATIC int 1661 xfs_swap_extent_rmap( 1662 struct xfs_trans **tpp, 1663 struct xfs_inode *ip, 1664 struct xfs_inode *tip) 1665 { 1666 struct xfs_bmbt_irec irec; 1667 struct xfs_bmbt_irec uirec; 1668 struct xfs_bmbt_irec tirec; 1669 xfs_fileoff_t offset_fsb; 1670 xfs_fileoff_t end_fsb; 1671 xfs_filblks_t count_fsb; 1672 xfs_fsblock_t firstfsb; 1673 struct xfs_defer_ops dfops; 1674 int error; 1675 xfs_filblks_t ilen; 1676 xfs_filblks_t rlen; 1677 int nimaps; 1678 __uint64_t tip_flags2; 1679 1680 /* 1681 * If the source file has shared blocks, we must flag the donor 1682 * file as having shared blocks so that we get the shared-block 1683 * rmap functions when we go to fix up the rmaps. The flags 1684 * will be switch for reals later. 1685 */ 1686 tip_flags2 = tip->i_d.di_flags2; 1687 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) 1688 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK; 1689 1690 offset_fsb = 0; 1691 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip))); 1692 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb); 1693 1694 while (count_fsb) { 1695 /* Read extent from the donor file */ 1696 nimaps = 1; 1697 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec, 1698 &nimaps, 0); 1699 if (error) 1700 goto out; 1701 ASSERT(nimaps == 1); 1702 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK); 1703 1704 trace_xfs_swap_extent_rmap_remap(tip, &tirec); 1705 ilen = tirec.br_blockcount; 1706 1707 /* Unmap the old blocks in the source file. */ 1708 while (tirec.br_blockcount) { 1709 xfs_defer_init(&dfops, &firstfsb); 1710 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec); 1711 1712 /* Read extent from the source file */ 1713 nimaps = 1; 1714 error = xfs_bmapi_read(ip, tirec.br_startoff, 1715 tirec.br_blockcount, &irec, 1716 &nimaps, 0); 1717 if (error) 1718 goto out_defer; 1719 ASSERT(nimaps == 1); 1720 ASSERT(tirec.br_startoff == irec.br_startoff); 1721 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec); 1722 1723 /* Trim the extent. */ 1724 uirec = tirec; 1725 uirec.br_blockcount = rlen = min_t(xfs_filblks_t, 1726 tirec.br_blockcount, 1727 irec.br_blockcount); 1728 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec); 1729 1730 /* Remove the mapping from the donor file. */ 1731 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops, 1732 tip, &uirec); 1733 if (error) 1734 goto out_defer; 1735 1736 /* Remove the mapping from the source file. */ 1737 error = xfs_bmap_unmap_extent((*tpp)->t_mountp, &dfops, 1738 ip, &irec); 1739 if (error) 1740 goto out_defer; 1741 1742 /* Map the donor file's blocks into the source file. */ 1743 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops, 1744 ip, &uirec); 1745 if (error) 1746 goto out_defer; 1747 1748 /* Map the source file's blocks into the donor file. */ 1749 error = xfs_bmap_map_extent((*tpp)->t_mountp, &dfops, 1750 tip, &irec); 1751 if (error) 1752 goto out_defer; 1753 1754 error = xfs_defer_finish(tpp, &dfops, ip); 1755 if (error) 1756 goto out_defer; 1757 1758 tirec.br_startoff += rlen; 1759 if (tirec.br_startblock != HOLESTARTBLOCK && 1760 tirec.br_startblock != DELAYSTARTBLOCK) 1761 tirec.br_startblock += rlen; 1762 tirec.br_blockcount -= rlen; 1763 } 1764 1765 /* Roll on... */ 1766 count_fsb -= ilen; 1767 offset_fsb += ilen; 1768 } 1769 1770 tip->i_d.di_flags2 = tip_flags2; 1771 return 0; 1772 1773 out_defer: 1774 xfs_defer_cancel(&dfops); 1775 out: 1776 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_); 1777 tip->i_d.di_flags2 = tip_flags2; 1778 return error; 1779 } 1780 1781 /* Swap the extents of two files by swapping data forks. */ 1782 STATIC int 1783 xfs_swap_extent_forks( 1784 struct xfs_trans *tp, 1785 struct xfs_inode *ip, 1786 struct xfs_inode *tip, 1787 int *src_log_flags, 1788 int *target_log_flags) 1789 { 1790 struct xfs_ifork tempifp, *ifp, *tifp; 1791 int aforkblks = 0; 1792 int taforkblks = 0; 1793 xfs_extnum_t nextents; 1794 __uint64_t tmp; 1795 int error; 1796 1797 /* 1798 * Count the number of extended attribute blocks 1799 */ 1800 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) && 1801 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { 1802 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, 1803 &aforkblks); 1804 if (error) 1805 return error; 1806 } 1807 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) && 1808 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) { 1809 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, 1810 &taforkblks); 1811 if (error) 1812 return error; 1813 } 1814 1815 /* 1816 * Before we've swapped the forks, lets set the owners of the forks 1817 * appropriately. We have to do this as we are demand paging the btree 1818 * buffers, and so the validation done on read will expect the owner 1819 * field to be correctly set. Once we change the owners, we can swap the 1820 * inode forks. 1821 */ 1822 if (ip->i_d.di_version == 3 && 1823 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) { 1824 (*target_log_flags) |= XFS_ILOG_DOWNER; 1825 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, 1826 tip->i_ino, NULL); 1827 if (error) 1828 return error; 1829 } 1830 1831 if (tip->i_d.di_version == 3 && 1832 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) { 1833 (*src_log_flags) |= XFS_ILOG_DOWNER; 1834 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK, 1835 ip->i_ino, NULL); 1836 if (error) 1837 return error; 1838 } 1839 1840 /* 1841 * Swap the data forks of the inodes 1842 */ 1843 ifp = &ip->i_df; 1844 tifp = &tip->i_df; 1845 tempifp = *ifp; /* struct copy */ 1846 *ifp = *tifp; /* struct copy */ 1847 *tifp = tempifp; /* struct copy */ 1848 1849 /* 1850 * Fix the on-disk inode values 1851 */ 1852 tmp = (__uint64_t)ip->i_d.di_nblocks; 1853 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks; 1854 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks; 1855 1856 tmp = (__uint64_t) ip->i_d.di_nextents; 1857 ip->i_d.di_nextents = tip->i_d.di_nextents; 1858 tip->i_d.di_nextents = tmp; 1859 1860 tmp = (__uint64_t) ip->i_d.di_format; 1861 ip->i_d.di_format = tip->i_d.di_format; 1862 tip->i_d.di_format = tmp; 1863 1864 /* 1865 * The extents in the source inode could still contain speculative 1866 * preallocation beyond EOF (e.g. the file is open but not modified 1867 * while defrag is in progress). In that case, we need to copy over the 1868 * number of delalloc blocks the data fork in the source inode is 1869 * tracking beyond EOF so that when the fork is truncated away when the 1870 * temporary inode is unlinked we don't underrun the i_delayed_blks 1871 * counter on that inode. 1872 */ 1873 ASSERT(tip->i_delayed_blks == 0); 1874 tip->i_delayed_blks = ip->i_delayed_blks; 1875 ip->i_delayed_blks = 0; 1876 1877 switch (ip->i_d.di_format) { 1878 case XFS_DINODE_FMT_EXTENTS: 1879 /* 1880 * If the extents fit in the inode, fix the pointer. Otherwise 1881 * it's already NULL or pointing to the extent. 1882 */ 1883 nextents = xfs_iext_count(&ip->i_df); 1884 if (nextents <= XFS_INLINE_EXTS) 1885 ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext; 1886 (*src_log_flags) |= XFS_ILOG_DEXT; 1887 break; 1888 case XFS_DINODE_FMT_BTREE: 1889 ASSERT(ip->i_d.di_version < 3 || 1890 (*src_log_flags & XFS_ILOG_DOWNER)); 1891 (*src_log_flags) |= XFS_ILOG_DBROOT; 1892 break; 1893 } 1894 1895 switch (tip->i_d.di_format) { 1896 case XFS_DINODE_FMT_EXTENTS: 1897 /* 1898 * If the extents fit in the inode, fix the pointer. Otherwise 1899 * it's already NULL or pointing to the extent. 1900 */ 1901 nextents = xfs_iext_count(&tip->i_df); 1902 if (nextents <= XFS_INLINE_EXTS) 1903 tifp->if_u1.if_extents = tifp->if_u2.if_inline_ext; 1904 (*target_log_flags) |= XFS_ILOG_DEXT; 1905 break; 1906 case XFS_DINODE_FMT_BTREE: 1907 (*target_log_flags) |= XFS_ILOG_DBROOT; 1908 ASSERT(tip->i_d.di_version < 3 || 1909 (*target_log_flags & XFS_ILOG_DOWNER)); 1910 break; 1911 } 1912 1913 return 0; 1914 } 1915 1916 int 1917 xfs_swap_extents( 1918 struct xfs_inode *ip, /* target inode */ 1919 struct xfs_inode *tip, /* tmp inode */ 1920 struct xfs_swapext *sxp) 1921 { 1922 struct xfs_mount *mp = ip->i_mount; 1923 struct xfs_trans *tp; 1924 struct xfs_bstat *sbp = &sxp->sx_stat; 1925 int src_log_flags, target_log_flags; 1926 int error = 0; 1927 int lock_flags; 1928 struct xfs_ifork *cowfp; 1929 __uint64_t f; 1930 int resblks; 1931 1932 /* 1933 * Lock the inodes against other IO, page faults and truncate to 1934 * begin with. Then we can ensure the inodes are flushed and have no 1935 * page cache safely. Once we have done this we can take the ilocks and 1936 * do the rest of the checks. 1937 */ 1938 lock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 1939 lock_flags = XFS_MMAPLOCK_EXCL; 1940 xfs_lock_two_inodes(ip, tip, XFS_MMAPLOCK_EXCL); 1941 1942 /* Verify that both files have the same format */ 1943 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) { 1944 error = -EINVAL; 1945 goto out_unlock; 1946 } 1947 1948 /* Verify both files are either real-time or non-realtime */ 1949 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) { 1950 error = -EINVAL; 1951 goto out_unlock; 1952 } 1953 1954 error = xfs_swap_extent_flush(ip); 1955 if (error) 1956 goto out_unlock; 1957 error = xfs_swap_extent_flush(tip); 1958 if (error) 1959 goto out_unlock; 1960 1961 /* 1962 * Extent "swapping" with rmap requires a permanent reservation and 1963 * a block reservation because it's really just a remap operation 1964 * performed with log redo items! 1965 */ 1966 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) { 1967 /* 1968 * Conceptually this shouldn't affect the shape of either 1969 * bmbt, but since we atomically move extents one by one, 1970 * we reserve enough space to rebuild both trees. 1971 */ 1972 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, 1973 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK), 1974 XFS_DATA_FORK) + 1975 XFS_SWAP_RMAP_SPACE_RES(mp, 1976 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK), 1977 XFS_DATA_FORK); 1978 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 1979 0, 0, &tp); 1980 } else 1981 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 1982 0, 0, &tp); 1983 if (error) 1984 goto out_unlock; 1985 1986 /* 1987 * Lock and join the inodes to the tansaction so that transaction commit 1988 * or cancel will unlock the inodes from this point onwards. 1989 */ 1990 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL); 1991 lock_flags |= XFS_ILOCK_EXCL; 1992 xfs_trans_ijoin(tp, ip, 0); 1993 xfs_trans_ijoin(tp, tip, 0); 1994 1995 1996 /* Verify all data are being swapped */ 1997 if (sxp->sx_offset != 0 || 1998 sxp->sx_length != ip->i_d.di_size || 1999 sxp->sx_length != tip->i_d.di_size) { 2000 error = -EFAULT; 2001 goto out_trans_cancel; 2002 } 2003 2004 trace_xfs_swap_extent_before(ip, 0); 2005 trace_xfs_swap_extent_before(tip, 1); 2006 2007 /* check inode formats now that data is flushed */ 2008 error = xfs_swap_extents_check_format(ip, tip); 2009 if (error) { 2010 xfs_notice(mp, 2011 "%s: inode 0x%llx format is incompatible for exchanging.", 2012 __func__, ip->i_ino); 2013 goto out_trans_cancel; 2014 } 2015 2016 /* 2017 * Compare the current change & modify times with that 2018 * passed in. If they differ, we abort this swap. 2019 * This is the mechanism used to ensure the calling 2020 * process that the file was not changed out from 2021 * under it. 2022 */ 2023 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) || 2024 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) || 2025 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) || 2026 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) { 2027 error = -EBUSY; 2028 goto out_trans_cancel; 2029 } 2030 2031 /* 2032 * Note the trickiness in setting the log flags - we set the owner log 2033 * flag on the opposite inode (i.e. the inode we are setting the new 2034 * owner to be) because once we swap the forks and log that, log 2035 * recovery is going to see the fork as owned by the swapped inode, 2036 * not the pre-swapped inodes. 2037 */ 2038 src_log_flags = XFS_ILOG_CORE; 2039 target_log_flags = XFS_ILOG_CORE; 2040 2041 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) 2042 error = xfs_swap_extent_rmap(&tp, ip, tip); 2043 else 2044 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags, 2045 &target_log_flags); 2046 if (error) 2047 goto out_trans_cancel; 2048 2049 /* Do we have to swap reflink flags? */ 2050 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^ 2051 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) { 2052 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK; 2053 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 2054 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK; 2055 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK; 2056 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK; 2057 cowfp = ip->i_cowfp; 2058 ip->i_cowfp = tip->i_cowfp; 2059 tip->i_cowfp = cowfp; 2060 xfs_inode_set_cowblocks_tag(ip); 2061 xfs_inode_set_cowblocks_tag(tip); 2062 } 2063 2064 xfs_trans_log_inode(tp, ip, src_log_flags); 2065 xfs_trans_log_inode(tp, tip, target_log_flags); 2066 2067 /* 2068 * If this is a synchronous mount, make sure that the 2069 * transaction goes to disk before returning to the user. 2070 */ 2071 if (mp->m_flags & XFS_MOUNT_WSYNC) 2072 xfs_trans_set_sync(tp); 2073 2074 error = xfs_trans_commit(tp); 2075 2076 trace_xfs_swap_extent_after(ip, 0); 2077 trace_xfs_swap_extent_after(tip, 1); 2078 2079 out_unlock: 2080 xfs_iunlock(ip, lock_flags); 2081 xfs_iunlock(tip, lock_flags); 2082 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip)); 2083 return error; 2084 2085 out_trans_cancel: 2086 xfs_trans_cancel(tp); 2087 goto out_unlock; 2088 } 2089