1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap_btree.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_space.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_iomap.h"
25 #include "xfs_trace.h"
26 #include "xfs_quota.h"
27 #include "xfs_dquot_item.h"
28 #include "xfs_dquot.h"
29 #include "xfs_reflink.h"
32 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
36 xfs_alert_fsblock_zero(
38 xfs_bmbt_irec_t *imap)
40 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
41 "Access to block zero in inode %llu "
42 "start_block: %llx start_off: %llx "
43 "blkcnt: %llx extent-state: %x",
44 (unsigned long long)ip->i_ino,
45 (unsigned long long)imap->br_startblock,
46 (unsigned long long)imap->br_startoff,
47 (unsigned long long)imap->br_blockcount,
56 struct xfs_bmbt_irec *imap,
59 struct xfs_mount *mp = ip->i_mount;
61 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
62 return xfs_alert_fsblock_zero(ip, imap);
64 if (imap->br_startblock == HOLESTARTBLOCK) {
65 iomap->addr = IOMAP_NULL_ADDR;
66 iomap->type = IOMAP_HOLE;
67 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
68 isnullstartblock(imap->br_startblock)) {
69 iomap->addr = IOMAP_NULL_ADDR;
70 iomap->type = IOMAP_DELALLOC;
72 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
73 if (imap->br_state == XFS_EXT_UNWRITTEN)
74 iomap->type = IOMAP_UNWRITTEN;
76 iomap->type = IOMAP_MAPPED;
78 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
79 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
80 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
81 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
83 if (xfs_ipincount(ip) &&
84 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
85 iomap->flags |= IOMAP_F_DIRTY;
87 iomap->flags |= IOMAP_F_SHARED;
95 xfs_fileoff_t offset_fsb,
96 xfs_fileoff_t end_fsb)
98 iomap->addr = IOMAP_NULL_ADDR;
99 iomap->type = IOMAP_HOLE;
100 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
101 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
102 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
103 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
108 struct xfs_inode *ip,
109 xfs_extlen_t extsize)
111 struct xfs_mount *mp = ip->i_mount;
112 xfs_extlen_t align = 0;
114 if (!XFS_IS_REALTIME_INODE(ip)) {
116 * Round up the allocation request to a stripe unit
117 * (m_dalign) boundary if the file size is >= stripe unit
118 * size, and we are allocating past the allocation eof.
120 * If mounted with the "-o swalloc" option the alignment is
121 * increased from the strip unit size to the stripe width.
123 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
124 align = mp->m_swidth;
125 else if (mp->m_dalign)
126 align = mp->m_dalign;
128 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
133 * Always round up the allocation request to an extent boundary
134 * (when file on a real-time subvolume or has di_extsize hint).
138 align = roundup_64(align, extsize);
147 xfs_iomap_eof_align_last_fsb(
148 struct xfs_inode *ip,
149 xfs_extlen_t extsize,
150 xfs_fileoff_t *last_fsb)
152 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
155 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
158 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
162 *last_fsb = new_last_fsb;
168 xfs_iomap_write_direct(
172 xfs_bmbt_irec_t *imap,
175 xfs_mount_t *mp = ip->i_mount;
176 xfs_fileoff_t offset_fsb;
177 xfs_fileoff_t last_fsb;
178 xfs_filblks_t count_fsb, resaligned;
184 uint qblocks, resblks, resrtextents;
187 int bmapi_flags = XFS_BMAPI_PREALLOC;
190 rt = XFS_IS_REALTIME_INODE(ip);
191 extsz = xfs_get_extsz_hint(ip);
192 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
194 ASSERT(xfs_isilocked(ip, lockmode));
196 offset_fsb = XFS_B_TO_FSBT(mp, offset);
197 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
198 if ((offset + count) > XFS_ISIZE(ip)) {
200 * Assert that the in-core extent list is present since this can
201 * call xfs_iread_extents() and we only have the ilock shared.
202 * This should be safe because the lock was held around a bmapi
203 * call in the caller and we only need it to access the in-core
206 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
208 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
212 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
213 last_fsb = min(last_fsb, (xfs_fileoff_t)
214 imap->br_blockcount +
217 count_fsb = last_fsb - offset_fsb;
218 ASSERT(count_fsb > 0);
219 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
222 resrtextents = qblocks = resaligned;
223 resrtextents /= mp->m_sb.sb_rextsize;
224 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
225 quota_flag = XFS_QMOPT_RES_RTBLKS;
228 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
229 quota_flag = XFS_QMOPT_RES_REGBLKS;
233 * Drop the shared lock acquired by the caller, attach the dquot if
234 * necessary and move on to transaction setup.
236 xfs_iunlock(ip, lockmode);
237 error = xfs_qm_dqattach(ip);
242 * For DAX, we do not allocate unwritten extents, but instead we zero
243 * the block before we commit the transaction. Ideally we'd like to do
244 * this outside the transaction context, but if we commit and then crash
245 * we may not have zeroed the blocks and this will be exposed on
246 * recovery of the allocation. Hence we must zero before commit.
248 * Further, if we are mapping unwritten extents here, we need to zero
249 * and convert them to written so that we don't need an unwritten extent
250 * callback for DAX. This also means that we need to be able to dip into
251 * the reserve block pool for bmbt block allocation if there is no space
252 * left but we need to do unwritten extent conversion.
254 if (IS_DAX(VFS_I(ip))) {
255 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
256 if (imap->br_state == XFS_EXT_UNWRITTEN) {
257 tflags |= XFS_TRANS_RESERVE;
258 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
261 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
266 lockmode = XFS_ILOCK_EXCL;
267 xfs_ilock(ip, lockmode);
269 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
271 goto out_trans_cancel;
273 xfs_trans_ijoin(tp, ip, 0);
276 * From this point onwards we overwrite the imap pointer that the
280 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
281 bmapi_flags, resblks, imap, &nimaps);
286 * Complete the transaction
288 error = xfs_trans_commit(tp);
293 * Copy any maps to caller's array and return any error.
300 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
301 error = xfs_alert_fsblock_zero(ip, imap);
304 xfs_iunlock(ip, lockmode);
308 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
310 xfs_trans_cancel(tp);
315 xfs_quota_need_throttle(
316 struct xfs_inode *ip,
318 xfs_fsblock_t alloc_blocks)
320 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
322 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
325 /* no hi watermark, no throttle */
326 if (!dq->q_prealloc_hi_wmark)
329 /* under the lo watermark, no throttle */
330 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
337 xfs_quota_calc_throttle(
338 struct xfs_inode *ip,
340 xfs_fsblock_t *qblocks,
346 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
348 /* no dq, or over hi wmark, squash the prealloc completely */
349 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
355 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
356 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
358 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
360 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
364 if (freesp < *qfreesp)
367 /* only overwrite the throttle values if we are more aggressive */
368 if ((freesp >> shift) < (*qblocks >> *qshift)) {
375 * If we are doing a write at the end of the file and there are no allocations
376 * past this one, then extend the allocation out to the file system's write
379 * If we don't have a user specified preallocation size, dynamically increase
380 * the preallocation size as the size of the file grows. Cap the maximum size
381 * at a single extent or less if the filesystem is near full. The closer the
382 * filesystem is to full, the smaller the maximum prealocation.
384 * As an exception we don't do any preallocation at all if the file is smaller
385 * than the minimum preallocation and we are using the default dynamic
386 * preallocation scheme, as it is likely this is the only write to the file that
387 * is going to be done.
389 * We clean up any extra space left over when the file is closed in
393 xfs_iomap_prealloc_size(
394 struct xfs_inode *ip,
398 struct xfs_iext_cursor *icur)
400 struct xfs_mount *mp = ip->i_mount;
401 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
402 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
403 struct xfs_bmbt_irec prev;
406 xfs_fsblock_t qblocks;
408 xfs_fsblock_t alloc_blocks = 0;
410 if (offset + count <= XFS_ISIZE(ip))
413 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
414 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
418 * If an explicit allocsize is set, the file is small, or we
419 * are writing behind a hole, then use the minimum prealloc:
421 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
422 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
423 !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
424 prev.br_startoff + prev.br_blockcount < offset_fsb)
425 return mp->m_writeio_blocks;
428 * Determine the initial size of the preallocation. We are beyond the
429 * current EOF here, but we need to take into account whether this is
430 * a sparse write or an extending write when determining the
431 * preallocation size. Hence we need to look up the extent that ends
432 * at the current write offset and use the result to determine the
433 * preallocation size.
435 * If the extent is a hole, then preallocation is essentially disabled.
436 * Otherwise we take the size of the preceding data extent as the basis
437 * for the preallocation size. If the size of the extent is greater than
438 * half the maximum extent length, then use the current offset as the
439 * basis. This ensures that for large files the preallocation size
440 * always extends to MAXEXTLEN rather than falling short due to things
441 * like stripe unit/width alignment of real extents.
443 if (prev.br_blockcount <= (MAXEXTLEN >> 1))
444 alloc_blocks = prev.br_blockcount << 1;
446 alloc_blocks = XFS_B_TO_FSB(mp, offset);
449 qblocks = alloc_blocks;
452 * MAXEXTLEN is not a power of two value but we round the prealloc down
453 * to the nearest power of two value after throttling. To prevent the
454 * round down from unconditionally reducing the maximum supported prealloc
455 * size, we round up first, apply appropriate throttling, round down and
456 * cap the value to MAXEXTLEN.
458 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
461 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
462 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
464 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
466 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
468 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
470 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
475 * Check each quota to cap the prealloc size, provide a shift value to
476 * throttle with and adjust amount of available space.
478 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
479 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
481 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
482 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
484 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
485 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
489 * The final prealloc size is set to the minimum of free space available
490 * in each of the quotas and the overall filesystem.
492 * The shift throttle value is set to the maximum value as determined by
493 * the global low free space values and per-quota low free space values.
495 alloc_blocks = min(alloc_blocks, qblocks);
496 shift = max(shift, qshift);
499 alloc_blocks >>= shift;
501 * rounddown_pow_of_two() returns an undefined result if we pass in
505 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
506 if (alloc_blocks > MAXEXTLEN)
507 alloc_blocks = MAXEXTLEN;
510 * If we are still trying to allocate more space than is
511 * available, squash the prealloc hard. This can happen if we
512 * have a large file on a small filesystem and the above
513 * lowspace thresholds are smaller than MAXEXTLEN.
515 while (alloc_blocks && alloc_blocks >= freesp)
518 if (alloc_blocks < mp->m_writeio_blocks)
519 alloc_blocks = mp->m_writeio_blocks;
520 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
521 mp->m_writeio_blocks);
526 xfs_file_iomap_begin_delay(
533 struct xfs_inode *ip = XFS_I(inode);
534 struct xfs_mount *mp = ip->i_mount;
535 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
536 xfs_fileoff_t maxbytes_fsb =
537 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
538 xfs_fileoff_t end_fsb;
539 struct xfs_bmbt_irec imap, cmap;
540 struct xfs_iext_cursor icur, ccur;
541 xfs_fsblock_t prealloc_blocks = 0;
542 bool eof = false, cow_eof = false, shared = false;
543 int whichfork = XFS_DATA_FORK;
546 ASSERT(!XFS_IS_REALTIME_INODE(ip));
547 ASSERT(!xfs_get_extsz_hint(ip));
549 xfs_ilock(ip, XFS_ILOCK_EXCL);
551 if (unlikely(XFS_TEST_ERROR(
552 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
553 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
554 mp, XFS_ERRTAG_BMAPIFORMAT))) {
555 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
556 error = -EFSCORRUPTED;
560 XFS_STATS_INC(mp, xs_blk_mapw);
562 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
563 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
568 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
571 * Search the data fork fork first to look up our source mapping. We
572 * always need the data fork map, as we have to return it to the
573 * iomap code so that the higher level write code can read data in to
574 * perform read-modify-write cycles for unaligned writes.
576 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
578 imap.br_startoff = end_fsb; /* fake hole until the end */
580 /* We never need to allocate blocks for zeroing a hole. */
581 if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
582 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
587 * Search the COW fork extent list even if we did not find a data fork
588 * extent. This serves two purposes: first this implements the
589 * speculative preallocation using cowextsize, so that we also unshare
590 * block adjacent to shared blocks instead of just the shared blocks
591 * themselves. Second the lookup in the extent list is generally faster
592 * than going out to the shared extent tree.
594 if (xfs_is_cow_inode(ip)) {
596 ASSERT(!xfs_is_reflink_inode(ip));
597 xfs_ifork_init_cow(ip);
599 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
601 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
602 trace_xfs_reflink_cow_found(ip, &cmap);
603 whichfork = XFS_COW_FORK;
608 if (imap.br_startoff <= offset_fsb) {
610 * For reflink files we may need a delalloc reservation when
611 * overwriting shared extents. This includes zeroing of
612 * existing extents that contain data.
614 if (!xfs_is_cow_inode(ip) ||
615 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
616 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
621 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
623 /* Trim the mapping to the nearest shared extent boundary. */
624 error = xfs_inode_need_cow(ip, &imap, &shared);
628 /* Not shared? Just report the (potentially capped) extent. */
630 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
636 * Fork all the shared blocks from our write offset until the
639 whichfork = XFS_COW_FORK;
640 end_fsb = imap.br_startoff + imap.br_blockcount;
643 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
644 * pages to keep the chunks of work done where somewhat
645 * symmetric with the work writeback does. This is a completely
646 * arbitrary number pulled out of thin air.
648 * Note that the values needs to be less than 32-bits wide until
649 * the lower level functions are updated.
651 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
652 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
654 if (xfs_is_always_cow_inode(ip))
655 whichfork = XFS_COW_FORK;
658 error = xfs_qm_dqattach_locked(ip, false);
663 prealloc_blocks = xfs_iomap_prealloc_size(ip, whichfork, offset,
665 if (prealloc_blocks) {
667 xfs_off_t end_offset;
668 xfs_fileoff_t p_end_fsb;
670 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
671 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
674 align = xfs_eof_alignment(ip, 0);
676 p_end_fsb = roundup_64(p_end_fsb, align);
678 p_end_fsb = min(p_end_fsb, maxbytes_fsb);
679 ASSERT(p_end_fsb > offset_fsb);
680 prealloc_blocks = p_end_fsb - end_fsb;
685 error = xfs_bmapi_reserve_delalloc(ip, whichfork, offset_fsb,
686 end_fsb - offset_fsb, prealloc_blocks,
687 whichfork == XFS_DATA_FORK ? &imap : &cmap,
688 whichfork == XFS_DATA_FORK ? &icur : &ccur,
689 whichfork == XFS_DATA_FORK ? eof : cow_eof);
695 /* retry without any preallocation */
696 trace_xfs_delalloc_enospc(ip, offset, count);
697 if (prealloc_blocks) {
707 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
708 * them out if the write happens to fail.
710 iomap->flags |= IOMAP_F_NEW;
711 trace_xfs_iomap_alloc(ip, offset, count, whichfork,
712 whichfork == XFS_DATA_FORK ? &imap : &cmap);
714 if (whichfork == XFS_COW_FORK) {
715 if (imap.br_startoff > offset_fsb) {
716 xfs_trim_extent(&cmap, offset_fsb,
717 imap.br_startoff - offset_fsb);
718 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, true);
721 /* ensure we only report blocks we have a reservation for */
722 xfs_trim_extent(&imap, cmap.br_startoff, cmap.br_blockcount);
725 error = xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
727 xfs_iunlock(ip, XFS_ILOCK_EXCL);
732 xfs_iomap_write_unwritten(
738 xfs_mount_t *mp = ip->i_mount;
739 xfs_fileoff_t offset_fsb;
740 xfs_filblks_t count_fsb;
741 xfs_filblks_t numblks_fsb;
744 xfs_bmbt_irec_t imap;
745 struct inode *inode = VFS_I(ip);
750 trace_xfs_unwritten_convert(ip, offset, count);
752 offset_fsb = XFS_B_TO_FSBT(mp, offset);
753 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
754 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
757 * Reserve enough blocks in this transaction for two complete extent
758 * btree splits. We may be converting the middle part of an unwritten
759 * extent and in this case we will insert two new extents in the btree
760 * each of which could cause a full split.
762 * This reservation amount will be used in the first call to
763 * xfs_bmbt_split() to select an AG with enough space to satisfy the
764 * rest of the operation.
766 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
770 * Set up a transaction to convert the range of extents
771 * from unwritten to real. Do allocations in a loop until
772 * we have covered the range passed in.
774 * Note that we can't risk to recursing back into the filesystem
775 * here as we might be asked to write out the same inode that we
776 * complete here and might deadlock on the iolock.
778 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
779 XFS_TRANS_RESERVE, &tp);
783 xfs_ilock(ip, XFS_ILOCK_EXCL);
784 xfs_trans_ijoin(tp, ip, 0);
787 * Modify the unwritten extent state of the buffer.
790 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
791 XFS_BMAPI_CONVERT, resblks, &imap,
794 goto error_on_bmapi_transaction;
797 * Log the updated inode size as we go. We have to be careful
798 * to only log it up to the actual write offset if it is
799 * halfway into a block.
801 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
802 if (i_size > offset + count)
803 i_size = offset + count;
804 if (update_isize && i_size > i_size_read(inode))
805 i_size_write(inode, i_size);
806 i_size = xfs_new_eof(ip, i_size);
808 ip->i_d.di_size = i_size;
809 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
812 error = xfs_trans_commit(tp);
813 xfs_iunlock(ip, XFS_ILOCK_EXCL);
817 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
818 return xfs_alert_fsblock_zero(ip, &imap);
820 if ((numblks_fsb = imap.br_blockcount) == 0) {
822 * The numblks_fsb value should always get
823 * smaller, otherwise the loop is stuck.
825 ASSERT(imap.br_blockcount);
828 offset_fsb += numblks_fsb;
829 count_fsb -= numblks_fsb;
830 } while (count_fsb > 0);
834 error_on_bmapi_transaction:
835 xfs_trans_cancel(tp);
836 xfs_iunlock(ip, XFS_ILOCK_EXCL);
843 struct xfs_bmbt_irec *imap,
847 imap->br_startblock == HOLESTARTBLOCK ||
848 imap->br_startblock == DELAYSTARTBLOCK ||
849 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
853 needs_cow_for_zeroing(
854 struct xfs_bmbt_irec *imap,
858 imap->br_startblock != HOLESTARTBLOCK &&
859 imap->br_state != XFS_EXT_UNWRITTEN;
864 struct xfs_inode *ip,
868 unsigned mode = XFS_ILOCK_SHARED;
869 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
872 * COW writes may allocate delalloc space or convert unwritten COW
873 * extents, so we need to make sure to take the lock exclusively here.
875 if (xfs_is_cow_inode(ip) && is_write) {
877 * FIXME: It could still overwrite on unshared extents and not
880 if (flags & IOMAP_NOWAIT)
882 mode = XFS_ILOCK_EXCL;
886 * Extents not yet cached requires exclusive access, don't block. This
887 * is an opencoded xfs_ilock_data_map_shared() call but with
888 * non-blocking behaviour.
890 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
891 if (flags & IOMAP_NOWAIT)
893 mode = XFS_ILOCK_EXCL;
897 if (flags & IOMAP_NOWAIT) {
898 if (!xfs_ilock_nowait(ip, mode))
905 * The reflink iflag could have changed since the earlier unlocked
906 * check, so if we got ILOCK_SHARED for a write and but we're now a
907 * reflink inode we have to switch to ILOCK_EXCL and relock.
909 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
910 xfs_iunlock(ip, mode);
911 mode = XFS_ILOCK_EXCL;
920 xfs_file_iomap_begin(
927 struct xfs_inode *ip = XFS_I(inode);
928 struct xfs_mount *mp = ip->i_mount;
929 struct xfs_bmbt_irec imap;
930 xfs_fileoff_t offset_fsb, end_fsb;
931 int nimaps = 1, error = 0;
935 if (XFS_FORCED_SHUTDOWN(mp))
938 if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && !(flags & IOMAP_DIRECT) &&
939 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
940 /* Reserve delalloc blocks for regular writeback. */
941 return xfs_file_iomap_begin_delay(inode, offset, length, flags,
946 * Lock the inode in the manner required for the specified operation and
947 * check for as many conditions that would result in blocking as
948 * possible. This removes most of the non-blocking checks from the
949 * mapping code below.
951 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
955 ASSERT(offset <= mp->m_super->s_maxbytes);
956 if (offset > mp->m_super->s_maxbytes - length)
957 length = mp->m_super->s_maxbytes - offset;
958 offset_fsb = XFS_B_TO_FSBT(mp, offset);
959 end_fsb = XFS_B_TO_FSB(mp, offset + length);
961 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
966 if (flags & IOMAP_REPORT) {
967 /* Trim the mapping to the nearest shared extent boundary. */
968 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
973 /* Non-modifying mapping requested, so we are done */
974 if (!(flags & (IOMAP_WRITE | IOMAP_ZERO)))
978 * Break shared extents if necessary. Checks for non-blocking IO have
979 * been done up front, so we don't need to do them here.
981 if (xfs_is_cow_inode(ip)) {
982 struct xfs_bmbt_irec cmap;
983 bool directio = (flags & IOMAP_DIRECT);
985 /* if zeroing doesn't need COW allocation, then we are done. */
986 if ((flags & IOMAP_ZERO) &&
987 !needs_cow_for_zeroing(&imap, nimaps))
990 /* may drop and re-acquire the ilock */
992 error = xfs_reflink_allocate_cow(ip, &cmap, &shared, &lockmode,
998 * For buffered writes we need to report the address of the
999 * previous block (if there was any) so that the higher level
1000 * write code can perform read-modify-write operations; we
1001 * won't need the CoW fork mapping until writeback. For direct
1002 * I/O, which must be block aligned, we need to report the
1003 * newly allocated address. If the data fork has a hole, copy
1004 * the COW fork mapping to avoid allocating to the data fork.
1006 if (directio || imap.br_startblock == HOLESTARTBLOCK)
1009 end_fsb = imap.br_startoff + imap.br_blockcount;
1010 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1013 /* Don't need to allocate over holes when doing zeroing operations. */
1014 if (flags & IOMAP_ZERO)
1017 if (!imap_needs_alloc(inode, &imap, nimaps))
1020 /* If nowait is set bail since we are going to make allocations. */
1021 if (flags & IOMAP_NOWAIT) {
1027 * We cap the maximum length we map to a sane size to keep the chunks
1028 * of work done where somewhat symmetric with the work writeback does.
1029 * This is a completely arbitrary number pulled out of thin air as a
1030 * best guess for initial testing.
1032 * Note that the values needs to be less than 32-bits wide until the
1033 * lower level functions are updated.
1035 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1038 * xfs_iomap_write_direct() expects the shared lock. It is unlocked on
1041 if (lockmode == XFS_ILOCK_EXCL)
1042 xfs_ilock_demote(ip, lockmode);
1043 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1048 iomap->flags |= IOMAP_F_NEW;
1049 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
1052 return xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
1056 xfs_iunlock(ip, lockmode);
1057 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1061 xfs_iunlock(ip, lockmode);
1066 xfs_file_iomap_end_delalloc(
1067 struct xfs_inode *ip,
1071 struct iomap *iomap)
1073 struct xfs_mount *mp = ip->i_mount;
1074 xfs_fileoff_t start_fsb;
1075 xfs_fileoff_t end_fsb;
1079 * Behave as if the write failed if drop writes is enabled. Set the NEW
1080 * flag to force delalloc cleanup.
1082 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1083 iomap->flags |= IOMAP_F_NEW;
1088 * start_fsb refers to the first unused block after a short write. If
1089 * nothing was written, round offset down to point at the first block in
1092 if (unlikely(!written))
1093 start_fsb = XFS_B_TO_FSBT(mp, offset);
1095 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1096 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1099 * Trim delalloc blocks if they were allocated by this write and we
1100 * didn't manage to write the whole range.
1102 * We don't need to care about racing delalloc as we hold i_mutex
1103 * across the reserve/allocate/unreserve calls. If there are delalloc
1104 * blocks in the range, they are ours.
1106 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1107 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1108 XFS_FSB_TO_B(mp, end_fsb) - 1);
1110 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1111 end_fsb - start_fsb);
1112 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1113 xfs_alert(mp, "%s: unable to clean up ino %lld",
1114 __func__, ip->i_ino);
1124 struct inode *inode,
1129 struct iomap *iomap)
1131 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1132 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1133 length, written, iomap);
1137 const struct iomap_ops xfs_iomap_ops = {
1138 .iomap_begin = xfs_file_iomap_begin,
1139 .iomap_end = xfs_file_iomap_end,
1143 xfs_seek_iomap_begin(
1144 struct inode *inode,
1148 struct iomap *iomap)
1150 struct xfs_inode *ip = XFS_I(inode);
1151 struct xfs_mount *mp = ip->i_mount;
1152 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1153 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1154 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1155 struct xfs_iext_cursor icur;
1156 struct xfs_bmbt_irec imap, cmap;
1160 if (XFS_FORCED_SHUTDOWN(mp))
1163 lockmode = xfs_ilock_data_map_shared(ip);
1164 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1165 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1170 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1172 * If we found a data extent we are done.
1174 if (imap.br_startoff <= offset_fsb)
1176 data_fsb = imap.br_startoff;
1179 * Fake a hole until the end of the file.
1181 data_fsb = min(XFS_B_TO_FSB(mp, offset + length),
1182 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1186 * If a COW fork extent covers the hole, report it - capped to the next
1189 if (xfs_inode_has_cow_data(ip) &&
1190 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1191 cow_fsb = cmap.br_startoff;
1192 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1193 if (data_fsb < cow_fsb + cmap.br_blockcount)
1194 end_fsb = min(end_fsb, data_fsb);
1195 xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1196 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, true);
1198 * This is a COW extent, so we must probe the page cache
1199 * because there could be dirty page cache being backed
1202 iomap->type = IOMAP_UNWRITTEN;
1207 * Else report a hole, capped to the next found data or COW extent.
1209 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1210 imap.br_blockcount = cow_fsb - offset_fsb;
1212 imap.br_blockcount = data_fsb - offset_fsb;
1213 imap.br_startoff = offset_fsb;
1214 imap.br_startblock = HOLESTARTBLOCK;
1215 imap.br_state = XFS_EXT_NORM;
1217 xfs_trim_extent(&imap, offset_fsb, end_fsb);
1218 error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
1220 xfs_iunlock(ip, lockmode);
1224 const struct iomap_ops xfs_seek_iomap_ops = {
1225 .iomap_begin = xfs_seek_iomap_begin,
1229 xfs_xattr_iomap_begin(
1230 struct inode *inode,
1234 struct iomap *iomap)
1236 struct xfs_inode *ip = XFS_I(inode);
1237 struct xfs_mount *mp = ip->i_mount;
1238 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1239 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1240 struct xfs_bmbt_irec imap;
1241 int nimaps = 1, error = 0;
1244 if (XFS_FORCED_SHUTDOWN(mp))
1247 lockmode = xfs_ilock_attr_map_shared(ip);
1249 /* if there are no attribute fork or extents, return ENOENT */
1250 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1255 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1256 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1257 &nimaps, XFS_BMAPI_ATTRFORK);
1259 xfs_iunlock(ip, lockmode);
1264 return xfs_bmbt_to_iomap(ip, iomap, &imap, false);
1267 const struct iomap_ops xfs_xattr_iomap_ops = {
1268 .iomap_begin = xfs_xattr_iomap_begin,