1 /* 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 3 * Copyright (c) 2016 Christoph Hellwig. 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 <linux/iomap.h> 20 #include "xfs.h" 21 #include "xfs_fs.h" 22 #include "xfs_shared.h" 23 #include "xfs_format.h" 24 #include "xfs_log_format.h" 25 #include "xfs_trans_resv.h" 26 #include "xfs_mount.h" 27 #include "xfs_defer.h" 28 #include "xfs_inode.h" 29 #include "xfs_btree.h" 30 #include "xfs_bmap_btree.h" 31 #include "xfs_bmap.h" 32 #include "xfs_bmap_util.h" 33 #include "xfs_errortag.h" 34 #include "xfs_error.h" 35 #include "xfs_trans.h" 36 #include "xfs_trans_space.h" 37 #include "xfs_inode_item.h" 38 #include "xfs_iomap.h" 39 #include "xfs_trace.h" 40 #include "xfs_icache.h" 41 #include "xfs_quota.h" 42 #include "xfs_dquot_item.h" 43 #include "xfs_dquot.h" 44 #include "xfs_reflink.h" 45 46 47 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ 48 << mp->m_writeio_log) 49 50 void 51 xfs_bmbt_to_iomap( 52 struct xfs_inode *ip, 53 struct iomap *iomap, 54 struct xfs_bmbt_irec *imap) 55 { 56 struct xfs_mount *mp = ip->i_mount; 57 58 if (imap->br_startblock == HOLESTARTBLOCK) { 59 iomap->addr = IOMAP_NULL_ADDR; 60 iomap->type = IOMAP_HOLE; 61 } else if (imap->br_startblock == DELAYSTARTBLOCK) { 62 iomap->addr = IOMAP_NULL_ADDR; 63 iomap->type = IOMAP_DELALLOC; 64 } else { 65 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock)); 66 if (imap->br_state == XFS_EXT_UNWRITTEN) 67 iomap->type = IOMAP_UNWRITTEN; 68 else 69 iomap->type = IOMAP_MAPPED; 70 } 71 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff); 72 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount); 73 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip)); 74 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip)); 75 } 76 77 xfs_extlen_t 78 xfs_eof_alignment( 79 struct xfs_inode *ip, 80 xfs_extlen_t extsize) 81 { 82 struct xfs_mount *mp = ip->i_mount; 83 xfs_extlen_t align = 0; 84 85 if (!XFS_IS_REALTIME_INODE(ip)) { 86 /* 87 * Round up the allocation request to a stripe unit 88 * (m_dalign) boundary if the file size is >= stripe unit 89 * size, and we are allocating past the allocation eof. 90 * 91 * If mounted with the "-o swalloc" option the alignment is 92 * increased from the strip unit size to the stripe width. 93 */ 94 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) 95 align = mp->m_swidth; 96 else if (mp->m_dalign) 97 align = mp->m_dalign; 98 99 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align)) 100 align = 0; 101 } 102 103 /* 104 * Always round up the allocation request to an extent boundary 105 * (when file on a real-time subvolume or has di_extsize hint). 106 */ 107 if (extsize) { 108 if (align) 109 align = roundup_64(align, extsize); 110 else 111 align = extsize; 112 } 113 114 return align; 115 } 116 117 STATIC int 118 xfs_iomap_eof_align_last_fsb( 119 struct xfs_inode *ip, 120 xfs_extlen_t extsize, 121 xfs_fileoff_t *last_fsb) 122 { 123 xfs_extlen_t align = xfs_eof_alignment(ip, extsize); 124 125 if (align) { 126 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align); 127 int eof, error; 128 129 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); 130 if (error) 131 return error; 132 if (eof) 133 *last_fsb = new_last_fsb; 134 } 135 return 0; 136 } 137 138 STATIC int 139 xfs_alert_fsblock_zero( 140 xfs_inode_t *ip, 141 xfs_bmbt_irec_t *imap) 142 { 143 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, 144 "Access to block zero in inode %llu " 145 "start_block: %llx start_off: %llx " 146 "blkcnt: %llx extent-state: %x", 147 (unsigned long long)ip->i_ino, 148 (unsigned long long)imap->br_startblock, 149 (unsigned long long)imap->br_startoff, 150 (unsigned long long)imap->br_blockcount, 151 imap->br_state); 152 return -EFSCORRUPTED; 153 } 154 155 int 156 xfs_iomap_write_direct( 157 xfs_inode_t *ip, 158 xfs_off_t offset, 159 size_t count, 160 xfs_bmbt_irec_t *imap, 161 int nmaps) 162 { 163 xfs_mount_t *mp = ip->i_mount; 164 xfs_fileoff_t offset_fsb; 165 xfs_fileoff_t last_fsb; 166 xfs_filblks_t count_fsb, resaligned; 167 xfs_fsblock_t firstfsb; 168 xfs_extlen_t extsz; 169 int nimaps; 170 int quota_flag; 171 int rt; 172 xfs_trans_t *tp; 173 struct xfs_defer_ops dfops; 174 uint qblocks, resblks, resrtextents; 175 int error; 176 int lockmode; 177 int bmapi_flags = XFS_BMAPI_PREALLOC; 178 uint tflags = 0; 179 180 rt = XFS_IS_REALTIME_INODE(ip); 181 extsz = xfs_get_extsz_hint(ip); 182 lockmode = XFS_ILOCK_SHARED; /* locked by caller */ 183 184 ASSERT(xfs_isilocked(ip, lockmode)); 185 186 offset_fsb = XFS_B_TO_FSBT(mp, offset); 187 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); 188 if ((offset + count) > XFS_ISIZE(ip)) { 189 /* 190 * Assert that the in-core extent list is present since this can 191 * call xfs_iread_extents() and we only have the ilock shared. 192 * This should be safe because the lock was held around a bmapi 193 * call in the caller and we only need it to access the in-core 194 * list. 195 */ 196 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags & 197 XFS_IFEXTENTS); 198 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb); 199 if (error) 200 goto out_unlock; 201 } else { 202 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) 203 last_fsb = MIN(last_fsb, (xfs_fileoff_t) 204 imap->br_blockcount + 205 imap->br_startoff); 206 } 207 count_fsb = last_fsb - offset_fsb; 208 ASSERT(count_fsb > 0); 209 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz); 210 211 if (unlikely(rt)) { 212 resrtextents = qblocks = resaligned; 213 resrtextents /= mp->m_sb.sb_rextsize; 214 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 215 quota_flag = XFS_QMOPT_RES_RTBLKS; 216 } else { 217 resrtextents = 0; 218 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); 219 quota_flag = XFS_QMOPT_RES_REGBLKS; 220 } 221 222 /* 223 * Drop the shared lock acquired by the caller, attach the dquot if 224 * necessary and move on to transaction setup. 225 */ 226 xfs_iunlock(ip, lockmode); 227 error = xfs_qm_dqattach(ip, 0); 228 if (error) 229 return error; 230 231 /* 232 * For DAX, we do not allocate unwritten extents, but instead we zero 233 * the block before we commit the transaction. Ideally we'd like to do 234 * this outside the transaction context, but if we commit and then crash 235 * we may not have zeroed the blocks and this will be exposed on 236 * recovery of the allocation. Hence we must zero before commit. 237 * 238 * Further, if we are mapping unwritten extents here, we need to zero 239 * and convert them to written so that we don't need an unwritten extent 240 * callback for DAX. This also means that we need to be able to dip into 241 * the reserve block pool for bmbt block allocation if there is no space 242 * left but we need to do unwritten extent conversion. 243 */ 244 if (IS_DAX(VFS_I(ip))) { 245 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO; 246 if (imap->br_state == XFS_EXT_UNWRITTEN) { 247 tflags |= XFS_TRANS_RESERVE; 248 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; 249 } 250 } 251 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents, 252 tflags, &tp); 253 if (error) 254 return error; 255 256 lockmode = XFS_ILOCK_EXCL; 257 xfs_ilock(ip, lockmode); 258 259 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); 260 if (error) 261 goto out_trans_cancel; 262 263 xfs_trans_ijoin(tp, ip, 0); 264 265 /* 266 * From this point onwards we overwrite the imap pointer that the 267 * caller gave to us. 268 */ 269 xfs_defer_init(&dfops, &firstfsb); 270 nimaps = 1; 271 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, 272 bmapi_flags, &firstfsb, resblks, imap, 273 &nimaps, &dfops); 274 if (error) 275 goto out_bmap_cancel; 276 277 /* 278 * Complete the transaction 279 */ 280 error = xfs_defer_finish(&tp, &dfops); 281 if (error) 282 goto out_bmap_cancel; 283 284 error = xfs_trans_commit(tp); 285 if (error) 286 goto out_unlock; 287 288 /* 289 * Copy any maps to caller's array and return any error. 290 */ 291 if (nimaps == 0) { 292 error = -ENOSPC; 293 goto out_unlock; 294 } 295 296 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) 297 error = xfs_alert_fsblock_zero(ip, imap); 298 299 out_unlock: 300 xfs_iunlock(ip, lockmode); 301 return error; 302 303 out_bmap_cancel: 304 xfs_defer_cancel(&dfops); 305 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); 306 out_trans_cancel: 307 xfs_trans_cancel(tp); 308 goto out_unlock; 309 } 310 311 STATIC bool 312 xfs_quota_need_throttle( 313 struct xfs_inode *ip, 314 int type, 315 xfs_fsblock_t alloc_blocks) 316 { 317 struct xfs_dquot *dq = xfs_inode_dquot(ip, type); 318 319 if (!dq || !xfs_this_quota_on(ip->i_mount, type)) 320 return false; 321 322 /* no hi watermark, no throttle */ 323 if (!dq->q_prealloc_hi_wmark) 324 return false; 325 326 /* under the lo watermark, no throttle */ 327 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) 328 return false; 329 330 return true; 331 } 332 333 STATIC void 334 xfs_quota_calc_throttle( 335 struct xfs_inode *ip, 336 int type, 337 xfs_fsblock_t *qblocks, 338 int *qshift, 339 int64_t *qfreesp) 340 { 341 int64_t freesp; 342 int shift = 0; 343 struct xfs_dquot *dq = xfs_inode_dquot(ip, type); 344 345 /* no dq, or over hi wmark, squash the prealloc completely */ 346 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { 347 *qblocks = 0; 348 *qfreesp = 0; 349 return; 350 } 351 352 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; 353 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { 354 shift = 2; 355 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) 356 shift += 2; 357 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) 358 shift += 2; 359 } 360 361 if (freesp < *qfreesp) 362 *qfreesp = freesp; 363 364 /* only overwrite the throttle values if we are more aggressive */ 365 if ((freesp >> shift) < (*qblocks >> *qshift)) { 366 *qblocks = freesp; 367 *qshift = shift; 368 } 369 } 370 371 /* 372 * If we are doing a write at the end of the file and there are no allocations 373 * past this one, then extend the allocation out to the file system's write 374 * iosize. 375 * 376 * If we don't have a user specified preallocation size, dynamically increase 377 * the preallocation size as the size of the file grows. Cap the maximum size 378 * at a single extent or less if the filesystem is near full. The closer the 379 * filesystem is to full, the smaller the maximum prealocation. 380 * 381 * As an exception we don't do any preallocation at all if the file is smaller 382 * than the minimum preallocation and we are using the default dynamic 383 * preallocation scheme, as it is likely this is the only write to the file that 384 * is going to be done. 385 * 386 * We clean up any extra space left over when the file is closed in 387 * xfs_inactive(). 388 */ 389 STATIC xfs_fsblock_t 390 xfs_iomap_prealloc_size( 391 struct xfs_inode *ip, 392 loff_t offset, 393 loff_t count, 394 struct xfs_iext_cursor *icur) 395 { 396 struct xfs_mount *mp = ip->i_mount; 397 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 398 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); 399 struct xfs_bmbt_irec prev; 400 int shift = 0; 401 int64_t freesp; 402 xfs_fsblock_t qblocks; 403 int qshift = 0; 404 xfs_fsblock_t alloc_blocks = 0; 405 406 if (offset + count <= XFS_ISIZE(ip)) 407 return 0; 408 409 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && 410 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))) 411 return 0; 412 413 /* 414 * If an explicit allocsize is set, the file is small, or we 415 * are writing behind a hole, then use the minimum prealloc: 416 */ 417 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) || 418 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) || 419 !xfs_iext_peek_prev_extent(ifp, icur, &prev) || 420 prev.br_startoff + prev.br_blockcount < offset_fsb) 421 return mp->m_writeio_blocks; 422 423 /* 424 * Determine the initial size of the preallocation. We are beyond the 425 * current EOF here, but we need to take into account whether this is 426 * a sparse write or an extending write when determining the 427 * preallocation size. Hence we need to look up the extent that ends 428 * at the current write offset and use the result to determine the 429 * preallocation size. 430 * 431 * If the extent is a hole, then preallocation is essentially disabled. 432 * Otherwise we take the size of the preceding data extent as the basis 433 * for the preallocation size. If the size of the extent is greater than 434 * half the maximum extent length, then use the current offset as the 435 * basis. This ensures that for large files the preallocation size 436 * always extends to MAXEXTLEN rather than falling short due to things 437 * like stripe unit/width alignment of real extents. 438 */ 439 if (prev.br_blockcount <= (MAXEXTLEN >> 1)) 440 alloc_blocks = prev.br_blockcount << 1; 441 else 442 alloc_blocks = XFS_B_TO_FSB(mp, offset); 443 if (!alloc_blocks) 444 goto check_writeio; 445 qblocks = alloc_blocks; 446 447 /* 448 * MAXEXTLEN is not a power of two value but we round the prealloc down 449 * to the nearest power of two value after throttling. To prevent the 450 * round down from unconditionally reducing the maximum supported prealloc 451 * size, we round up first, apply appropriate throttling, round down and 452 * cap the value to MAXEXTLEN. 453 */ 454 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), 455 alloc_blocks); 456 457 freesp = percpu_counter_read_positive(&mp->m_fdblocks); 458 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { 459 shift = 2; 460 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) 461 shift++; 462 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) 463 shift++; 464 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) 465 shift++; 466 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) 467 shift++; 468 } 469 470 /* 471 * Check each quota to cap the prealloc size, provide a shift value to 472 * throttle with and adjust amount of available space. 473 */ 474 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) 475 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift, 476 &freesp); 477 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) 478 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift, 479 &freesp); 480 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) 481 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift, 482 &freesp); 483 484 /* 485 * The final prealloc size is set to the minimum of free space available 486 * in each of the quotas and the overall filesystem. 487 * 488 * The shift throttle value is set to the maximum value as determined by 489 * the global low free space values and per-quota low free space values. 490 */ 491 alloc_blocks = MIN(alloc_blocks, qblocks); 492 shift = MAX(shift, qshift); 493 494 if (shift) 495 alloc_blocks >>= shift; 496 /* 497 * rounddown_pow_of_two() returns an undefined result if we pass in 498 * alloc_blocks = 0. 499 */ 500 if (alloc_blocks) 501 alloc_blocks = rounddown_pow_of_two(alloc_blocks); 502 if (alloc_blocks > MAXEXTLEN) 503 alloc_blocks = MAXEXTLEN; 504 505 /* 506 * If we are still trying to allocate more space than is 507 * available, squash the prealloc hard. This can happen if we 508 * have a large file on a small filesystem and the above 509 * lowspace thresholds are smaller than MAXEXTLEN. 510 */ 511 while (alloc_blocks && alloc_blocks >= freesp) 512 alloc_blocks >>= 4; 513 check_writeio: 514 if (alloc_blocks < mp->m_writeio_blocks) 515 alloc_blocks = mp->m_writeio_blocks; 516 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, 517 mp->m_writeio_blocks); 518 return alloc_blocks; 519 } 520 521 static int 522 xfs_file_iomap_begin_delay( 523 struct inode *inode, 524 loff_t offset, 525 loff_t count, 526 struct iomap *iomap) 527 { 528 struct xfs_inode *ip = XFS_I(inode); 529 struct xfs_mount *mp = ip->i_mount; 530 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); 531 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); 532 xfs_fileoff_t maxbytes_fsb = 533 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); 534 xfs_fileoff_t end_fsb; 535 int error = 0, eof = 0; 536 struct xfs_bmbt_irec got; 537 struct xfs_iext_cursor icur; 538 xfs_fsblock_t prealloc_blocks = 0; 539 540 ASSERT(!XFS_IS_REALTIME_INODE(ip)); 541 ASSERT(!xfs_get_extsz_hint(ip)); 542 543 xfs_ilock(ip, XFS_ILOCK_EXCL); 544 545 if (unlikely(XFS_TEST_ERROR( 546 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS && 547 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE), 548 mp, XFS_ERRTAG_BMAPIFORMAT))) { 549 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); 550 error = -EFSCORRUPTED; 551 goto out_unlock; 552 } 553 554 XFS_STATS_INC(mp, xs_blk_mapw); 555 556 if (!(ifp->if_flags & XFS_IFEXTENTS)) { 557 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK); 558 if (error) 559 goto out_unlock; 560 } 561 562 eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got); 563 if (!eof && got.br_startoff <= offset_fsb) { 564 if (xfs_is_reflink_inode(ip)) { 565 bool shared; 566 567 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), 568 maxbytes_fsb); 569 xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb); 570 error = xfs_reflink_reserve_cow(ip, &got, &shared); 571 if (error) 572 goto out_unlock; 573 } 574 575 trace_xfs_iomap_found(ip, offset, count, 0, &got); 576 goto done; 577 } 578 579 error = xfs_qm_dqattach_locked(ip, 0); 580 if (error) 581 goto out_unlock; 582 583 /* 584 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages 585 * to keep the chunks of work done where somewhat symmetric with the 586 * work writeback does. This is a completely arbitrary number pulled 587 * out of thin air as a best guess for initial testing. 588 * 589 * Note that the values needs to be less than 32-bits wide until 590 * the lower level functions are updated. 591 */ 592 count = min_t(loff_t, count, 1024 * PAGE_SIZE); 593 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb); 594 595 if (eof) { 596 prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count, 597 &icur); 598 if (prealloc_blocks) { 599 xfs_extlen_t align; 600 xfs_off_t end_offset; 601 xfs_fileoff_t p_end_fsb; 602 603 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1); 604 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) + 605 prealloc_blocks; 606 607 align = xfs_eof_alignment(ip, 0); 608 if (align) 609 p_end_fsb = roundup_64(p_end_fsb, align); 610 611 p_end_fsb = min(p_end_fsb, maxbytes_fsb); 612 ASSERT(p_end_fsb > offset_fsb); 613 prealloc_blocks = p_end_fsb - end_fsb; 614 } 615 } 616 617 retry: 618 error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb, 619 end_fsb - offset_fsb, prealloc_blocks, &got, &icur, 620 eof); 621 switch (error) { 622 case 0: 623 break; 624 case -ENOSPC: 625 case -EDQUOT: 626 /* retry without any preallocation */ 627 trace_xfs_delalloc_enospc(ip, offset, count); 628 if (prealloc_blocks) { 629 prealloc_blocks = 0; 630 goto retry; 631 } 632 /*FALLTHRU*/ 633 default: 634 goto out_unlock; 635 } 636 637 /* 638 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch 639 * them out if the write happens to fail. 640 */ 641 iomap->flags = IOMAP_F_NEW; 642 trace_xfs_iomap_alloc(ip, offset, count, 0, &got); 643 done: 644 if (isnullstartblock(got.br_startblock)) 645 got.br_startblock = DELAYSTARTBLOCK; 646 647 if (!got.br_startblock) { 648 error = xfs_alert_fsblock_zero(ip, &got); 649 if (error) 650 goto out_unlock; 651 } 652 653 xfs_bmbt_to_iomap(ip, iomap, &got); 654 655 out_unlock: 656 xfs_iunlock(ip, XFS_ILOCK_EXCL); 657 return error; 658 } 659 660 /* 661 * Pass in a delayed allocate extent, convert it to real extents; 662 * return to the caller the extent we create which maps on top of 663 * the originating callers request. 664 * 665 * Called without a lock on the inode. 666 * 667 * We no longer bother to look at the incoming map - all we have to 668 * guarantee is that whatever we allocate fills the required range. 669 */ 670 int 671 xfs_iomap_write_allocate( 672 xfs_inode_t *ip, 673 int whichfork, 674 xfs_off_t offset, 675 xfs_bmbt_irec_t *imap) 676 { 677 xfs_mount_t *mp = ip->i_mount; 678 xfs_fileoff_t offset_fsb, last_block; 679 xfs_fileoff_t end_fsb, map_start_fsb; 680 xfs_fsblock_t first_block; 681 struct xfs_defer_ops dfops; 682 xfs_filblks_t count_fsb; 683 xfs_trans_t *tp; 684 int nimaps; 685 int error = 0; 686 int flags = XFS_BMAPI_DELALLOC; 687 int nres; 688 689 if (whichfork == XFS_COW_FORK) 690 flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC; 691 692 /* 693 * Make sure that the dquots are there. 694 */ 695 error = xfs_qm_dqattach(ip, 0); 696 if (error) 697 return error; 698 699 offset_fsb = XFS_B_TO_FSBT(mp, offset); 700 count_fsb = imap->br_blockcount; 701 map_start_fsb = imap->br_startoff; 702 703 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); 704 705 while (count_fsb != 0) { 706 /* 707 * Set up a transaction with which to allocate the 708 * backing store for the file. Do allocations in a 709 * loop until we get some space in the range we are 710 * interested in. The other space that might be allocated 711 * is in the delayed allocation extent on which we sit 712 * but before our buffer starts. 713 */ 714 nimaps = 0; 715 while (nimaps == 0) { 716 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); 717 /* 718 * We have already reserved space for the extent and any 719 * indirect blocks when creating the delalloc extent, 720 * there is no need to reserve space in this transaction 721 * again. 722 */ 723 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 724 0, XFS_TRANS_RESERVE, &tp); 725 if (error) 726 return error; 727 728 xfs_ilock(ip, XFS_ILOCK_EXCL); 729 xfs_trans_ijoin(tp, ip, 0); 730 731 xfs_defer_init(&dfops, &first_block); 732 733 /* 734 * it is possible that the extents have changed since 735 * we did the read call as we dropped the ilock for a 736 * while. We have to be careful about truncates or hole 737 * punchs here - we are not allowed to allocate 738 * non-delalloc blocks here. 739 * 740 * The only protection against truncation is the pages 741 * for the range we are being asked to convert are 742 * locked and hence a truncate will block on them 743 * first. 744 * 745 * As a result, if we go beyond the range we really 746 * need and hit an delalloc extent boundary followed by 747 * a hole while we have excess blocks in the map, we 748 * will fill the hole incorrectly and overrun the 749 * transaction reservation. 750 * 751 * Using a single map prevents this as we are forced to 752 * check each map we look for overlap with the desired 753 * range and abort as soon as we find it. Also, given 754 * that we only return a single map, having one beyond 755 * what we can return is probably a bit silly. 756 * 757 * We also need to check that we don't go beyond EOF; 758 * this is a truncate optimisation as a truncate sets 759 * the new file size before block on the pages we 760 * currently have locked under writeback. Because they 761 * are about to be tossed, we don't need to write them 762 * back.... 763 */ 764 nimaps = 1; 765 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); 766 error = xfs_bmap_last_offset(ip, &last_block, 767 XFS_DATA_FORK); 768 if (error) 769 goto trans_cancel; 770 771 last_block = XFS_FILEOFF_MAX(last_block, end_fsb); 772 if ((map_start_fsb + count_fsb) > last_block) { 773 count_fsb = last_block - map_start_fsb; 774 if (count_fsb == 0) { 775 error = -EAGAIN; 776 goto trans_cancel; 777 } 778 } 779 780 /* 781 * From this point onwards we overwrite the imap 782 * pointer that the caller gave to us. 783 */ 784 error = xfs_bmapi_write(tp, ip, map_start_fsb, 785 count_fsb, flags, &first_block, 786 nres, imap, &nimaps, 787 &dfops); 788 if (error) 789 goto trans_cancel; 790 791 error = xfs_defer_finish(&tp, &dfops); 792 if (error) 793 goto trans_cancel; 794 795 error = xfs_trans_commit(tp); 796 if (error) 797 goto error0; 798 799 xfs_iunlock(ip, XFS_ILOCK_EXCL); 800 } 801 802 /* 803 * See if we were able to allocate an extent that 804 * covers at least part of the callers request 805 */ 806 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) 807 return xfs_alert_fsblock_zero(ip, imap); 808 809 if ((offset_fsb >= imap->br_startoff) && 810 (offset_fsb < (imap->br_startoff + 811 imap->br_blockcount))) { 812 XFS_STATS_INC(mp, xs_xstrat_quick); 813 return 0; 814 } 815 816 /* 817 * So far we have not mapped the requested part of the 818 * file, just surrounding data, try again. 819 */ 820 count_fsb -= imap->br_blockcount; 821 map_start_fsb = imap->br_startoff + imap->br_blockcount; 822 } 823 824 trans_cancel: 825 xfs_defer_cancel(&dfops); 826 xfs_trans_cancel(tp); 827 error0: 828 xfs_iunlock(ip, XFS_ILOCK_EXCL); 829 return error; 830 } 831 832 int 833 xfs_iomap_write_unwritten( 834 xfs_inode_t *ip, 835 xfs_off_t offset, 836 xfs_off_t count, 837 bool update_isize) 838 { 839 xfs_mount_t *mp = ip->i_mount; 840 xfs_fileoff_t offset_fsb; 841 xfs_filblks_t count_fsb; 842 xfs_filblks_t numblks_fsb; 843 xfs_fsblock_t firstfsb; 844 int nimaps; 845 xfs_trans_t *tp; 846 xfs_bmbt_irec_t imap; 847 struct xfs_defer_ops dfops; 848 struct inode *inode = VFS_I(ip); 849 xfs_fsize_t i_size; 850 uint resblks; 851 int error; 852 853 trace_xfs_unwritten_convert(ip, offset, count); 854 855 offset_fsb = XFS_B_TO_FSBT(mp, offset); 856 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); 857 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); 858 859 /* 860 * Reserve enough blocks in this transaction for two complete extent 861 * btree splits. We may be converting the middle part of an unwritten 862 * extent and in this case we will insert two new extents in the btree 863 * each of which could cause a full split. 864 * 865 * This reservation amount will be used in the first call to 866 * xfs_bmbt_split() to select an AG with enough space to satisfy the 867 * rest of the operation. 868 */ 869 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; 870 871 do { 872 /* 873 * Set up a transaction to convert the range of extents 874 * from unwritten to real. Do allocations in a loop until 875 * we have covered the range passed in. 876 * 877 * Note that we can't risk to recursing back into the filesystem 878 * here as we might be asked to write out the same inode that we 879 * complete here and might deadlock on the iolock. 880 */ 881 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 882 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp); 883 if (error) 884 return error; 885 886 xfs_ilock(ip, XFS_ILOCK_EXCL); 887 xfs_trans_ijoin(tp, ip, 0); 888 889 /* 890 * Modify the unwritten extent state of the buffer. 891 */ 892 xfs_defer_init(&dfops, &firstfsb); 893 nimaps = 1; 894 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, 895 XFS_BMAPI_CONVERT, &firstfsb, resblks, 896 &imap, &nimaps, &dfops); 897 if (error) 898 goto error_on_bmapi_transaction; 899 900 /* 901 * Log the updated inode size as we go. We have to be careful 902 * to only log it up to the actual write offset if it is 903 * halfway into a block. 904 */ 905 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); 906 if (i_size > offset + count) 907 i_size = offset + count; 908 if (update_isize && i_size > i_size_read(inode)) 909 i_size_write(inode, i_size); 910 i_size = xfs_new_eof(ip, i_size); 911 if (i_size) { 912 ip->i_d.di_size = i_size; 913 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 914 } 915 916 error = xfs_defer_finish(&tp, &dfops); 917 if (error) 918 goto error_on_bmapi_transaction; 919 920 error = xfs_trans_commit(tp); 921 xfs_iunlock(ip, XFS_ILOCK_EXCL); 922 if (error) 923 return error; 924 925 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) 926 return xfs_alert_fsblock_zero(ip, &imap); 927 928 if ((numblks_fsb = imap.br_blockcount) == 0) { 929 /* 930 * The numblks_fsb value should always get 931 * smaller, otherwise the loop is stuck. 932 */ 933 ASSERT(imap.br_blockcount); 934 break; 935 } 936 offset_fsb += numblks_fsb; 937 count_fsb -= numblks_fsb; 938 } while (count_fsb > 0); 939 940 return 0; 941 942 error_on_bmapi_transaction: 943 xfs_defer_cancel(&dfops); 944 xfs_trans_cancel(tp); 945 xfs_iunlock(ip, XFS_ILOCK_EXCL); 946 return error; 947 } 948 949 static inline bool imap_needs_alloc(struct inode *inode, 950 struct xfs_bmbt_irec *imap, int nimaps) 951 { 952 return !nimaps || 953 imap->br_startblock == HOLESTARTBLOCK || 954 imap->br_startblock == DELAYSTARTBLOCK || 955 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN); 956 } 957 958 static inline bool needs_cow_for_zeroing(struct xfs_bmbt_irec *imap, int nimaps) 959 { 960 return nimaps && 961 imap->br_startblock != HOLESTARTBLOCK && 962 imap->br_state != XFS_EXT_UNWRITTEN; 963 } 964 965 static inline bool need_excl_ilock(struct xfs_inode *ip, unsigned flags) 966 { 967 /* 968 * COW writes may allocate delalloc space or convert unwritten COW 969 * extents, so we need to make sure to take the lock exclusively here. 970 */ 971 if (xfs_is_reflink_inode(ip) && (flags & (IOMAP_WRITE | IOMAP_ZERO))) 972 return true; 973 974 /* 975 * Extents not yet cached requires exclusive access, don't block. 976 * This is an opencoded xfs_ilock_data_map_shared() to cater for the 977 * non-blocking behaviour. 978 */ 979 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE && 980 !(ip->i_df.if_flags & XFS_IFEXTENTS)) 981 return true; 982 return false; 983 } 984 985 static int 986 xfs_file_iomap_begin( 987 struct inode *inode, 988 loff_t offset, 989 loff_t length, 990 unsigned flags, 991 struct iomap *iomap) 992 { 993 struct xfs_inode *ip = XFS_I(inode); 994 struct xfs_mount *mp = ip->i_mount; 995 struct xfs_bmbt_irec imap; 996 xfs_fileoff_t offset_fsb, end_fsb; 997 int nimaps = 1, error = 0; 998 bool shared = false, trimmed = false; 999 unsigned lockmode; 1000 1001 if (XFS_FORCED_SHUTDOWN(mp)) 1002 return -EIO; 1003 1004 if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) && 1005 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) { 1006 /* Reserve delalloc blocks for regular writeback. */ 1007 return xfs_file_iomap_begin_delay(inode, offset, length, iomap); 1008 } 1009 1010 if (need_excl_ilock(ip, flags)) 1011 lockmode = XFS_ILOCK_EXCL; 1012 else 1013 lockmode = XFS_ILOCK_SHARED; 1014 1015 if (flags & IOMAP_NOWAIT) { 1016 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) 1017 return -EAGAIN; 1018 if (!xfs_ilock_nowait(ip, lockmode)) 1019 return -EAGAIN; 1020 } else { 1021 xfs_ilock(ip, lockmode); 1022 } 1023 1024 ASSERT(offset <= mp->m_super->s_maxbytes); 1025 if (offset > mp->m_super->s_maxbytes - length) 1026 length = mp->m_super->s_maxbytes - offset; 1027 offset_fsb = XFS_B_TO_FSBT(mp, offset); 1028 end_fsb = XFS_B_TO_FSB(mp, offset + length); 1029 1030 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, 1031 &nimaps, 0); 1032 if (error) 1033 goto out_unlock; 1034 1035 if (flags & IOMAP_REPORT) { 1036 /* Trim the mapping to the nearest shared extent boundary. */ 1037 error = xfs_reflink_trim_around_shared(ip, &imap, &shared, 1038 &trimmed); 1039 if (error) 1040 goto out_unlock; 1041 } 1042 1043 if (xfs_is_reflink_inode(ip) && 1044 ((flags & IOMAP_WRITE) || 1045 ((flags & IOMAP_ZERO) && needs_cow_for_zeroing(&imap, nimaps)))) { 1046 if (flags & IOMAP_DIRECT) { 1047 /* 1048 * A reflinked inode will result in CoW alloc. 1049 * FIXME: It could still overwrite on unshared extents 1050 * and not need allocation. 1051 */ 1052 if (flags & IOMAP_NOWAIT) { 1053 error = -EAGAIN; 1054 goto out_unlock; 1055 } 1056 /* may drop and re-acquire the ilock */ 1057 error = xfs_reflink_allocate_cow(ip, &imap, &shared, 1058 &lockmode); 1059 if (error) 1060 goto out_unlock; 1061 } else { 1062 error = xfs_reflink_reserve_cow(ip, &imap, &shared); 1063 if (error) 1064 goto out_unlock; 1065 } 1066 1067 end_fsb = imap.br_startoff + imap.br_blockcount; 1068 length = XFS_FSB_TO_B(mp, end_fsb) - offset; 1069 } 1070 1071 if ((flags & IOMAP_WRITE) && imap_needs_alloc(inode, &imap, nimaps)) { 1072 /* 1073 * If nowait is set bail since we are going to make 1074 * allocations. 1075 */ 1076 if (flags & IOMAP_NOWAIT) { 1077 error = -EAGAIN; 1078 goto out_unlock; 1079 } 1080 /* 1081 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES 1082 * pages to keep the chunks of work done where somewhat symmetric 1083 * with the work writeback does. This is a completely arbitrary 1084 * number pulled out of thin air as a best guess for initial 1085 * testing. 1086 * 1087 * Note that the values needs to be less than 32-bits wide until 1088 * the lower level functions are updated. 1089 */ 1090 length = min_t(loff_t, length, 1024 * PAGE_SIZE); 1091 /* 1092 * xfs_iomap_write_direct() expects the shared lock. It 1093 * is unlocked on return. 1094 */ 1095 if (lockmode == XFS_ILOCK_EXCL) 1096 xfs_ilock_demote(ip, lockmode); 1097 error = xfs_iomap_write_direct(ip, offset, length, &imap, 1098 nimaps); 1099 if (error) 1100 return error; 1101 1102 iomap->flags = IOMAP_F_NEW; 1103 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap); 1104 } else { 1105 ASSERT(nimaps); 1106 1107 xfs_iunlock(ip, lockmode); 1108 trace_xfs_iomap_found(ip, offset, length, 0, &imap); 1109 } 1110 1111 if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields 1112 & ~XFS_ILOG_TIMESTAMP)) 1113 iomap->flags |= IOMAP_F_DIRTY; 1114 1115 xfs_bmbt_to_iomap(ip, iomap, &imap); 1116 1117 if (shared) 1118 iomap->flags |= IOMAP_F_SHARED; 1119 return 0; 1120 out_unlock: 1121 xfs_iunlock(ip, lockmode); 1122 return error; 1123 } 1124 1125 static int 1126 xfs_file_iomap_end_delalloc( 1127 struct xfs_inode *ip, 1128 loff_t offset, 1129 loff_t length, 1130 ssize_t written, 1131 struct iomap *iomap) 1132 { 1133 struct xfs_mount *mp = ip->i_mount; 1134 xfs_fileoff_t start_fsb; 1135 xfs_fileoff_t end_fsb; 1136 int error = 0; 1137 1138 /* 1139 * Behave as if the write failed if drop writes is enabled. Set the NEW 1140 * flag to force delalloc cleanup. 1141 */ 1142 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) { 1143 iomap->flags |= IOMAP_F_NEW; 1144 written = 0; 1145 } 1146 1147 /* 1148 * start_fsb refers to the first unused block after a short write. If 1149 * nothing was written, round offset down to point at the first block in 1150 * the range. 1151 */ 1152 if (unlikely(!written)) 1153 start_fsb = XFS_B_TO_FSBT(mp, offset); 1154 else 1155 start_fsb = XFS_B_TO_FSB(mp, offset + written); 1156 end_fsb = XFS_B_TO_FSB(mp, offset + length); 1157 1158 /* 1159 * Trim delalloc blocks if they were allocated by this write and we 1160 * didn't manage to write the whole range. 1161 * 1162 * We don't need to care about racing delalloc as we hold i_mutex 1163 * across the reserve/allocate/unreserve calls. If there are delalloc 1164 * blocks in the range, they are ours. 1165 */ 1166 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) { 1167 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb), 1168 XFS_FSB_TO_B(mp, end_fsb) - 1); 1169 1170 xfs_ilock(ip, XFS_ILOCK_EXCL); 1171 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1172 end_fsb - start_fsb); 1173 xfs_iunlock(ip, XFS_ILOCK_EXCL); 1174 1175 if (error && !XFS_FORCED_SHUTDOWN(mp)) { 1176 xfs_alert(mp, "%s: unable to clean up ino %lld", 1177 __func__, ip->i_ino); 1178 return error; 1179 } 1180 } 1181 1182 return 0; 1183 } 1184 1185 static int 1186 xfs_file_iomap_end( 1187 struct inode *inode, 1188 loff_t offset, 1189 loff_t length, 1190 ssize_t written, 1191 unsigned flags, 1192 struct iomap *iomap) 1193 { 1194 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC) 1195 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset, 1196 length, written, iomap); 1197 return 0; 1198 } 1199 1200 const struct iomap_ops xfs_iomap_ops = { 1201 .iomap_begin = xfs_file_iomap_begin, 1202 .iomap_end = xfs_file_iomap_end, 1203 }; 1204 1205 static int 1206 xfs_xattr_iomap_begin( 1207 struct inode *inode, 1208 loff_t offset, 1209 loff_t length, 1210 unsigned flags, 1211 struct iomap *iomap) 1212 { 1213 struct xfs_inode *ip = XFS_I(inode); 1214 struct xfs_mount *mp = ip->i_mount; 1215 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); 1216 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length); 1217 struct xfs_bmbt_irec imap; 1218 int nimaps = 1, error = 0; 1219 unsigned lockmode; 1220 1221 if (XFS_FORCED_SHUTDOWN(mp)) 1222 return -EIO; 1223 1224 lockmode = xfs_ilock_attr_map_shared(ip); 1225 1226 /* if there are no attribute fork or extents, return ENOENT */ 1227 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) { 1228 error = -ENOENT; 1229 goto out_unlock; 1230 } 1231 1232 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL); 1233 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, 1234 &nimaps, XFS_BMAPI_ATTRFORK); 1235 out_unlock: 1236 xfs_iunlock(ip, lockmode); 1237 1238 if (!error) { 1239 ASSERT(nimaps); 1240 xfs_bmbt_to_iomap(ip, iomap, &imap); 1241 } 1242 1243 return error; 1244 } 1245 1246 const struct iomap_ops xfs_xattr_iomap_ops = { 1247 .iomap_begin = xfs_xattr_iomap_begin, 1248 }; 1249