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"
30 #include "xfs_health.h"
31 #include "xfs_rtbitmap.h"
33 #define XFS_ALLOC_ALIGN(mp, off) \
34 (((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
37 xfs_alert_fsblock_zero(
39 xfs_bmbt_irec_t *imap)
41 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
42 "Access to block zero in inode %llu "
43 "start_block: %llx start_off: %llx "
44 "blkcnt: %llx extent-state: %x",
45 (unsigned long long)ip->i_ino,
46 (unsigned long long)imap->br_startblock,
47 (unsigned long long)imap->br_startoff,
48 (unsigned long long)imap->br_blockcount,
50 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
55 xfs_iomap_inode_sequence(
61 if (iomap_flags & IOMAP_F_XATTR)
62 return READ_ONCE(ip->i_af.if_seq);
63 if ((iomap_flags & IOMAP_F_SHARED) && ip->i_cowfp)
64 cookie = (u64)READ_ONCE(ip->i_cowfp->if_seq) << 32;
65 return cookie | READ_ONCE(ip->i_df.if_seq);
69 * Check that the iomap passed to us is still valid for the given offset and
75 const struct iomap *iomap)
77 struct xfs_inode *ip = XFS_I(inode);
79 if (iomap->validity_cookie !=
80 xfs_iomap_inode_sequence(ip, iomap->flags)) {
81 trace_xfs_iomap_invalid(ip, iomap);
85 XFS_ERRORTAG_DELAY(ip->i_mount, XFS_ERRTAG_WRITE_DELAY_MS);
89 static const struct iomap_folio_ops xfs_iomap_folio_ops = {
90 .iomap_valid = xfs_iomap_valid,
97 struct xfs_bmbt_irec *imap,
98 unsigned int mapping_flags,
102 struct xfs_mount *mp = ip->i_mount;
103 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
105 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
106 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
107 return xfs_alert_fsblock_zero(ip, imap);
110 if (imap->br_startblock == HOLESTARTBLOCK) {
111 iomap->addr = IOMAP_NULL_ADDR;
112 iomap->type = IOMAP_HOLE;
113 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
114 isnullstartblock(imap->br_startblock)) {
115 iomap->addr = IOMAP_NULL_ADDR;
116 iomap->type = IOMAP_DELALLOC;
118 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
119 if (mapping_flags & IOMAP_DAX)
120 iomap->addr += target->bt_dax_part_off;
122 if (imap->br_state == XFS_EXT_UNWRITTEN)
123 iomap->type = IOMAP_UNWRITTEN;
125 iomap->type = IOMAP_MAPPED;
128 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
129 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
130 if (mapping_flags & IOMAP_DAX)
131 iomap->dax_dev = target->bt_daxdev;
133 iomap->bdev = target->bt_bdev;
134 iomap->flags = iomap_flags;
136 if (xfs_ipincount(ip) &&
137 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
138 iomap->flags |= IOMAP_F_DIRTY;
140 iomap->validity_cookie = sequence_cookie;
141 iomap->folio_ops = &xfs_iomap_folio_ops;
147 struct xfs_inode *ip,
149 xfs_fileoff_t offset_fsb,
150 xfs_fileoff_t end_fsb)
152 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
154 iomap->addr = IOMAP_NULL_ADDR;
155 iomap->type = IOMAP_HOLE;
156 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
157 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
158 iomap->bdev = target->bt_bdev;
159 iomap->dax_dev = target->bt_daxdev;
162 static inline xfs_fileoff_t
164 struct xfs_mount *mp,
168 ASSERT(offset <= mp->m_super->s_maxbytes);
169 return min(XFS_B_TO_FSB(mp, offset + count),
170 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
175 struct xfs_inode *ip)
177 struct xfs_mount *mp = ip->i_mount;
178 xfs_extlen_t align = 0;
180 if (!XFS_IS_REALTIME_INODE(ip)) {
182 * Round up the allocation request to a stripe unit
183 * (m_dalign) boundary if the file size is >= stripe unit
184 * size, and we are allocating past the allocation eof.
186 * If mounted with the "-o swalloc" option the alignment is
187 * increased from the strip unit size to the stripe width.
189 if (mp->m_swidth && xfs_has_swalloc(mp))
190 align = mp->m_swidth;
191 else if (mp->m_dalign)
192 align = mp->m_dalign;
194 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
202 * Check if last_fsb is outside the last extent, and if so grow it to the next
203 * stripe unit boundary.
206 xfs_iomap_eof_align_last_fsb(
207 struct xfs_inode *ip,
208 xfs_fileoff_t end_fsb)
210 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
211 xfs_extlen_t extsz = xfs_get_extsz_hint(ip);
212 xfs_extlen_t align = xfs_eof_alignment(ip);
213 struct xfs_bmbt_irec irec;
214 struct xfs_iext_cursor icur;
216 ASSERT(!xfs_need_iread_extents(ifp));
219 * Always round up the allocation request to the extent hint boundary.
223 align = roundup_64(align, extsz);
229 xfs_fileoff_t aligned_end_fsb = roundup_64(end_fsb, align);
231 xfs_iext_last(ifp, &icur);
232 if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
233 aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
234 return aligned_end_fsb;
241 xfs_iomap_write_direct(
242 struct xfs_inode *ip,
243 xfs_fileoff_t offset_fsb,
244 xfs_fileoff_t count_fsb,
246 struct xfs_bmbt_irec *imap,
249 struct xfs_mount *mp = ip->i_mount;
250 struct xfs_trans *tp;
251 xfs_filblks_t resaligned;
253 unsigned int dblocks, rblocks;
256 int bmapi_flags = XFS_BMAPI_PREALLOC;
257 int nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
259 ASSERT(count_fsb > 0);
261 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
262 xfs_get_extsz_hint(ip));
263 if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
264 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
265 rblocks = resaligned;
267 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
271 error = xfs_qm_dqattach(ip);
276 * For DAX, we do not allocate unwritten extents, but instead we zero
277 * the block before we commit the transaction. Ideally we'd like to do
278 * this outside the transaction context, but if we commit and then crash
279 * we may not have zeroed the blocks and this will be exposed on
280 * recovery of the allocation. Hence we must zero before commit.
282 * Further, if we are mapping unwritten extents here, we need to zero
283 * and convert them to written so that we don't need an unwritten extent
284 * callback for DAX. This also means that we need to be able to dip into
285 * the reserve block pool for bmbt block allocation if there is no space
286 * left but we need to do unwritten extent conversion.
288 if (flags & IOMAP_DAX) {
289 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
290 if (imap->br_state == XFS_EXT_UNWRITTEN) {
292 nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
293 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
297 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
298 rblocks, force, &tp);
302 error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK, nr_exts);
304 goto out_trans_cancel;
307 * From this point onwards we overwrite the imap pointer that the
311 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
314 goto out_trans_cancel;
317 * Complete the transaction
319 error = xfs_trans_commit(tp);
323 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
324 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
325 error = xfs_alert_fsblock_zero(ip, imap);
329 *seq = xfs_iomap_inode_sequence(ip, 0);
330 xfs_iunlock(ip, XFS_ILOCK_EXCL);
334 xfs_trans_cancel(tp);
339 xfs_quota_need_throttle(
340 struct xfs_inode *ip,
342 xfs_fsblock_t alloc_blocks)
344 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
346 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
349 /* no hi watermark, no throttle */
350 if (!dq->q_prealloc_hi_wmark)
353 /* under the lo watermark, no throttle */
354 if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
361 xfs_quota_calc_throttle(
362 struct xfs_inode *ip,
364 xfs_fsblock_t *qblocks,
368 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
372 /* no dq, or over hi wmark, squash the prealloc completely */
373 if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
379 freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
380 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
382 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
384 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
388 if (freesp < *qfreesp)
391 /* only overwrite the throttle values if we are more aggressive */
392 if ((freesp >> shift) < (*qblocks >> *qshift)) {
400 struct percpu_counter *counter,
401 uint64_t low_space[XFS_LOWSP_MAX],
406 freesp = percpu_counter_read_positive(counter);
407 if (freesp < low_space[XFS_LOWSP_5_PCNT]) {
409 if (freesp < low_space[XFS_LOWSP_4_PCNT])
411 if (freesp < low_space[XFS_LOWSP_3_PCNT])
413 if (freesp < low_space[XFS_LOWSP_2_PCNT])
415 if (freesp < low_space[XFS_LOWSP_1_PCNT])
422 * If we don't have a user specified preallocation size, dynamically increase
423 * the preallocation size as the size of the file grows. Cap the maximum size
424 * at a single extent or less if the filesystem is near full. The closer the
425 * filesystem is to being full, the smaller the maximum preallocation.
428 xfs_iomap_prealloc_size(
429 struct xfs_inode *ip,
433 struct xfs_iext_cursor *icur)
435 struct xfs_iext_cursor ncur = *icur;
436 struct xfs_bmbt_irec prev, got;
437 struct xfs_mount *mp = ip->i_mount;
438 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
439 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
441 xfs_fsblock_t qblocks;
442 xfs_fsblock_t alloc_blocks = 0;
448 * As an exception we don't do any preallocation at all if the file is
449 * smaller than the minimum preallocation and we are using the default
450 * dynamic preallocation scheme, as it is likely this is the only write
451 * to the file that is going to be done.
453 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
457 * Use the minimum preallocation size for small files or if we are
458 * writing right after a hole.
460 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
461 !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
462 prev.br_startoff + prev.br_blockcount < offset_fsb)
463 return mp->m_allocsize_blocks;
466 * Take the size of the preceding data extents as the basis for the
467 * preallocation size. Note that we don't care if the previous extents
468 * are written or not.
470 plen = prev.br_blockcount;
471 while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
472 if (plen > XFS_MAX_BMBT_EXTLEN / 2 ||
473 isnullstartblock(got.br_startblock) ||
474 got.br_startoff + got.br_blockcount != prev.br_startoff ||
475 got.br_startblock + got.br_blockcount != prev.br_startblock)
477 plen += got.br_blockcount;
482 * If the size of the extents is greater than half the maximum extent
483 * length, then use the current offset as the basis. This ensures that
484 * for large files the preallocation size always extends to
485 * XFS_BMBT_MAX_EXTLEN rather than falling short due to things like stripe
486 * unit/width alignment of real extents.
488 alloc_blocks = plen * 2;
489 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
490 alloc_blocks = XFS_B_TO_FSB(mp, offset);
491 qblocks = alloc_blocks;
494 * XFS_BMBT_MAX_EXTLEN is not a power of two value but we round the prealloc
495 * down to the nearest power of two value after throttling. To prevent
496 * the round down from unconditionally reducing the maximum supported
497 * prealloc size, we round up first, apply appropriate throttling, round
498 * down and cap the value to XFS_BMBT_MAX_EXTLEN.
500 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(XFS_MAX_BMBT_EXTLEN),
503 if (unlikely(XFS_IS_REALTIME_INODE(ip)))
504 freesp = xfs_rtx_to_rtb(mp,
505 xfs_iomap_freesp(&mp->m_frextents,
506 mp->m_low_rtexts, &shift));
508 freesp = xfs_iomap_freesp(&mp->m_fdblocks, mp->m_low_space,
512 * Check each quota to cap the prealloc size, provide a shift value to
513 * throttle with and adjust amount of available space.
515 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
516 xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
518 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
519 xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
521 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
522 xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
526 * The final prealloc size is set to the minimum of free space available
527 * in each of the quotas and the overall filesystem.
529 * The shift throttle value is set to the maximum value as determined by
530 * the global low free space values and per-quota low free space values.
532 alloc_blocks = min(alloc_blocks, qblocks);
533 shift = max(shift, qshift);
536 alloc_blocks >>= shift;
538 * rounddown_pow_of_two() returns an undefined result if we pass in
542 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
543 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
544 alloc_blocks = XFS_MAX_BMBT_EXTLEN;
547 * If we are still trying to allocate more space than is
548 * available, squash the prealloc hard. This can happen if we
549 * have a large file on a small filesystem and the above
550 * lowspace thresholds are smaller than XFS_BMBT_MAX_EXTLEN.
552 while (alloc_blocks && alloc_blocks >= freesp)
554 if (alloc_blocks < mp->m_allocsize_blocks)
555 alloc_blocks = mp->m_allocsize_blocks;
556 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
557 mp->m_allocsize_blocks);
562 xfs_iomap_write_unwritten(
568 xfs_mount_t *mp = ip->i_mount;
569 xfs_fileoff_t offset_fsb;
570 xfs_filblks_t count_fsb;
571 xfs_filblks_t numblks_fsb;
574 xfs_bmbt_irec_t imap;
575 struct inode *inode = VFS_I(ip);
580 trace_xfs_unwritten_convert(ip, offset, count);
582 offset_fsb = XFS_B_TO_FSBT(mp, offset);
583 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
584 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
587 * Reserve enough blocks in this transaction for two complete extent
588 * btree splits. We may be converting the middle part of an unwritten
589 * extent and in this case we will insert two new extents in the btree
590 * each of which could cause a full split.
592 * This reservation amount will be used in the first call to
593 * xfs_bmbt_split() to select an AG with enough space to satisfy the
594 * rest of the operation.
596 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
598 /* Attach dquots so that bmbt splits are accounted correctly. */
599 error = xfs_qm_dqattach(ip);
605 * Set up a transaction to convert the range of extents
606 * from unwritten to real. Do allocations in a loop until
607 * we have covered the range passed in.
609 * Note that we can't risk to recursing back into the filesystem
610 * here as we might be asked to write out the same inode that we
611 * complete here and might deadlock on the iolock.
613 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
618 error = xfs_iext_count_extend(tp, ip, XFS_DATA_FORK,
619 XFS_IEXT_WRITE_UNWRITTEN_CNT);
621 goto error_on_bmapi_transaction;
624 * Modify the unwritten extent state of the buffer.
627 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
628 XFS_BMAPI_CONVERT, resblks, &imap,
631 goto error_on_bmapi_transaction;
634 * Log the updated inode size as we go. We have to be careful
635 * to only log it up to the actual write offset if it is
636 * halfway into a block.
638 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
639 if (i_size > offset + count)
640 i_size = offset + count;
641 if (update_isize && i_size > i_size_read(inode))
642 i_size_write(inode, i_size);
643 i_size = xfs_new_eof(ip, i_size);
645 ip->i_disk_size = i_size;
646 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
649 error = xfs_trans_commit(tp);
650 xfs_iunlock(ip, XFS_ILOCK_EXCL);
654 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) {
655 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
656 return xfs_alert_fsblock_zero(ip, &imap);
659 if ((numblks_fsb = imap.br_blockcount) == 0) {
661 * The numblks_fsb value should always get
662 * smaller, otherwise the loop is stuck.
664 ASSERT(imap.br_blockcount);
667 offset_fsb += numblks_fsb;
668 count_fsb -= numblks_fsb;
669 } while (count_fsb > 0);
673 error_on_bmapi_transaction:
674 xfs_trans_cancel(tp);
675 xfs_iunlock(ip, XFS_ILOCK_EXCL);
683 struct xfs_bmbt_irec *imap,
686 /* don't allocate blocks when just zeroing */
687 if (flags & IOMAP_ZERO)
690 imap->br_startblock == HOLESTARTBLOCK ||
691 imap->br_startblock == DELAYSTARTBLOCK)
693 /* we convert unwritten extents before copying the data for DAX */
694 if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
701 struct xfs_inode *ip,
703 struct xfs_bmbt_irec *imap,
706 if (!xfs_is_cow_inode(ip))
709 /* when zeroing we don't have to COW holes or unwritten extents */
710 if (flags & IOMAP_ZERO) {
712 imap->br_startblock == HOLESTARTBLOCK ||
713 imap->br_state == XFS_EXT_UNWRITTEN)
722 struct xfs_inode *ip,
726 unsigned int mode = *lockmode;
727 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
730 * COW writes may allocate delalloc space or convert unwritten COW
731 * extents, so we need to make sure to take the lock exclusively here.
733 if (xfs_is_cow_inode(ip) && is_write)
734 mode = XFS_ILOCK_EXCL;
737 * Extents not yet cached requires exclusive access, don't block. This
738 * is an opencoded xfs_ilock_data_map_shared() call but with
739 * non-blocking behaviour.
741 if (xfs_need_iread_extents(&ip->i_df)) {
742 if (flags & IOMAP_NOWAIT)
744 mode = XFS_ILOCK_EXCL;
748 if (flags & IOMAP_NOWAIT) {
749 if (!xfs_ilock_nowait(ip, mode))
756 * The reflink iflag could have changed since the earlier unlocked
757 * check, so if we got ILOCK_SHARED for a write and but we're now a
758 * reflink inode we have to switch to ILOCK_EXCL and relock.
760 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
761 xfs_iunlock(ip, mode);
762 mode = XFS_ILOCK_EXCL;
771 * Check that the imap we are going to return to the caller spans the entire
772 * range that the caller requested for the IO.
776 struct xfs_bmbt_irec *imap,
777 xfs_fileoff_t offset_fsb,
778 xfs_fileoff_t end_fsb)
780 if (imap->br_startoff > offset_fsb)
782 if (imap->br_startoff + imap->br_blockcount < end_fsb)
788 xfs_direct_write_iomap_begin(
794 struct iomap *srcmap)
796 struct xfs_inode *ip = XFS_I(inode);
797 struct xfs_mount *mp = ip->i_mount;
798 struct xfs_bmbt_irec imap, cmap;
799 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
800 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
801 int nimaps = 1, error = 0;
804 unsigned int lockmode = XFS_ILOCK_SHARED;
807 ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
809 if (xfs_is_shutdown(mp))
813 * Writes that span EOF might trigger an IO size update on completion,
814 * so consider them to be dirty for the purposes of O_DSYNC even if
815 * there is no other metadata changes pending or have been made here.
817 if (offset + length > i_size_read(inode))
818 iomap_flags |= IOMAP_F_DIRTY;
820 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
824 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
829 if (imap_needs_cow(ip, flags, &imap, nimaps)) {
831 if (flags & IOMAP_NOWAIT)
834 /* may drop and re-acquire the ilock */
835 error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
837 (flags & IOMAP_DIRECT) || IS_DAX(inode));
842 end_fsb = imap.br_startoff + imap.br_blockcount;
843 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
846 if (imap_needs_alloc(inode, flags, &imap, nimaps))
847 goto allocate_blocks;
850 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
851 * a single map so that we avoid partial IO failures due to the rest of
852 * the I/O range not covered by this map triggering an EAGAIN condition
853 * when it is subsequently mapped and aborting the I/O.
855 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
857 if (!imap_spans_range(&imap, offset_fsb, end_fsb))
862 * For overwrite only I/O, we cannot convert unwritten extents without
863 * requiring sub-block zeroing. This can only be done under an
864 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
865 * extent to tell the caller to try again.
867 if (flags & IOMAP_OVERWRITE_ONLY) {
869 if (imap.br_state != XFS_EXT_NORM &&
870 ((offset | length) & mp->m_blockmask))
874 seq = xfs_iomap_inode_sequence(ip, iomap_flags);
875 xfs_iunlock(ip, lockmode);
876 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
877 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
881 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
885 * We cap the maximum length we map to a sane size to keep the chunks
886 * of work done where somewhat symmetric with the work writeback does.
887 * This is a completely arbitrary number pulled out of thin air as a
888 * best guess for initial testing.
890 * Note that the values needs to be less than 32-bits wide until the
891 * lower level functions are updated.
893 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
894 end_fsb = xfs_iomap_end_fsb(mp, offset, length);
896 if (offset + length > XFS_ISIZE(ip))
897 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
898 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
899 end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
900 xfs_iunlock(ip, lockmode);
902 error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
907 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
908 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
909 iomap_flags | IOMAP_F_NEW, seq);
912 length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
913 trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
914 if (imap.br_startblock != HOLESTARTBLOCK) {
915 seq = xfs_iomap_inode_sequence(ip, 0);
916 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
920 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
921 xfs_iunlock(ip, lockmode);
922 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED, seq);
926 xfs_iunlock(ip, lockmode);
930 const struct iomap_ops xfs_direct_write_iomap_ops = {
931 .iomap_begin = xfs_direct_write_iomap_begin,
935 xfs_dax_write_iomap_end(
943 struct xfs_inode *ip = XFS_I(inode);
945 if (!xfs_is_cow_inode(ip))
949 xfs_reflink_cancel_cow_range(ip, pos, length, true);
953 return xfs_reflink_end_cow(ip, pos, written);
956 const struct iomap_ops xfs_dax_write_iomap_ops = {
957 .iomap_begin = xfs_direct_write_iomap_begin,
958 .iomap_end = xfs_dax_write_iomap_end,
962 xfs_buffered_write_iomap_begin(
968 struct iomap *srcmap)
970 struct xfs_inode *ip = XFS_I(inode);
971 struct xfs_mount *mp = ip->i_mount;
972 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
973 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, count);
974 struct xfs_bmbt_irec imap, cmap;
975 struct xfs_iext_cursor icur, ccur;
976 xfs_fsblock_t prealloc_blocks = 0;
977 bool eof = false, cow_eof = false, shared = false;
978 int allocfork = XFS_DATA_FORK;
980 unsigned int lockmode = XFS_ILOCK_EXCL;
983 if (xfs_is_shutdown(mp))
986 /* we can't use delayed allocations when using extent size hints */
987 if (xfs_get_extsz_hint(ip))
988 return xfs_direct_write_iomap_begin(inode, offset, count,
989 flags, iomap, srcmap);
991 error = xfs_qm_dqattach(ip);
995 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
999 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
1000 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
1001 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
1002 error = -EFSCORRUPTED;
1006 XFS_STATS_INC(mp, xs_blk_mapw);
1008 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1013 * Search the data fork first to look up our source mapping. We
1014 * always need the data fork map, as we have to return it to the
1015 * iomap code so that the higher level write code can read data in to
1016 * perform read-modify-write cycles for unaligned writes.
1018 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
1020 imap.br_startoff = end_fsb; /* fake hole until the end */
1022 /* We never need to allocate blocks for zeroing or unsharing a hole. */
1023 if ((flags & (IOMAP_UNSHARE | IOMAP_ZERO)) &&
1024 imap.br_startoff > offset_fsb) {
1025 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
1030 * For zeroing, trim a delalloc extent that extends beyond the EOF
1031 * block. If it starts beyond the EOF block, convert it to an
1034 if ((flags & IOMAP_ZERO) && imap.br_startoff <= offset_fsb &&
1035 isnullstartblock(imap.br_startblock)) {
1036 xfs_fileoff_t eof_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
1038 if (offset_fsb >= eof_fsb)
1040 if (end_fsb > eof_fsb) {
1042 xfs_trim_extent(&imap, offset_fsb,
1043 end_fsb - offset_fsb);
1048 * Search the COW fork extent list even if we did not find a data fork
1049 * extent. This serves two purposes: first this implements the
1050 * speculative preallocation using cowextsize, so that we also unshare
1051 * block adjacent to shared blocks instead of just the shared blocks
1052 * themselves. Second the lookup in the extent list is generally faster
1053 * than going out to the shared extent tree.
1055 if (xfs_is_cow_inode(ip)) {
1057 ASSERT(!xfs_is_reflink_inode(ip));
1058 xfs_ifork_init_cow(ip);
1060 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
1062 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
1063 trace_xfs_reflink_cow_found(ip, &cmap);
1068 if (imap.br_startoff <= offset_fsb) {
1070 * For reflink files we may need a delalloc reservation when
1071 * overwriting shared extents. This includes zeroing of
1072 * existing extents that contain data.
1074 if (!xfs_is_cow_inode(ip) ||
1075 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
1076 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1081 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1083 /* Trim the mapping to the nearest shared extent boundary. */
1084 error = xfs_bmap_trim_cow(ip, &imap, &shared);
1088 /* Not shared? Just report the (potentially capped) extent. */
1090 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1096 * Fork all the shared blocks from our write offset until the
1097 * end of the extent.
1099 allocfork = XFS_COW_FORK;
1100 end_fsb = imap.br_startoff + imap.br_blockcount;
1103 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1104 * pages to keep the chunks of work done where somewhat
1105 * symmetric with the work writeback does. This is a completely
1106 * arbitrary number pulled out of thin air.
1108 * Note that the values needs to be less than 32-bits wide until
1109 * the lower level functions are updated.
1111 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
1112 end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1114 if (xfs_is_always_cow_inode(ip))
1115 allocfork = XFS_COW_FORK;
1118 if (eof && offset + count > XFS_ISIZE(ip)) {
1120 * Determine the initial size of the preallocation.
1121 * We clean up any extra preallocation when the file is closed.
1123 if (xfs_has_allocsize(mp))
1124 prealloc_blocks = mp->m_allocsize_blocks;
1125 else if (allocfork == XFS_DATA_FORK)
1126 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1127 offset, count, &icur);
1129 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1130 offset, count, &ccur);
1131 if (prealloc_blocks) {
1133 xfs_off_t end_offset;
1134 xfs_fileoff_t p_end_fsb;
1136 end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1137 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1140 align = xfs_eof_alignment(ip);
1142 p_end_fsb = roundup_64(p_end_fsb, align);
1144 p_end_fsb = min(p_end_fsb,
1145 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1146 ASSERT(p_end_fsb > offset_fsb);
1147 prealloc_blocks = p_end_fsb - end_fsb;
1152 error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1153 end_fsb - offset_fsb, prealloc_blocks,
1154 allocfork == XFS_DATA_FORK ? &imap : &cmap,
1155 allocfork == XFS_DATA_FORK ? &icur : &ccur,
1156 allocfork == XFS_DATA_FORK ? eof : cow_eof);
1162 /* retry without any preallocation */
1163 trace_xfs_delalloc_enospc(ip, offset, count);
1164 if (prealloc_blocks) {
1165 prealloc_blocks = 0;
1173 if (allocfork == XFS_COW_FORK) {
1174 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1179 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1180 * them out if the write happens to fail.
1182 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_NEW);
1183 xfs_iunlock(ip, lockmode);
1184 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1185 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_NEW, seq);
1188 seq = xfs_iomap_inode_sequence(ip, 0);
1189 xfs_iunlock(ip, lockmode);
1190 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1193 xfs_iunlock(ip, lockmode);
1194 truncate_pagecache(inode, offset);
1195 error = xfs_bmapi_convert_delalloc(ip, XFS_DATA_FORK, offset,
1200 trace_xfs_iomap_alloc(ip, offset, count, XFS_DATA_FORK, &imap);
1204 seq = xfs_iomap_inode_sequence(ip, 0);
1205 if (imap.br_startoff <= offset_fsb) {
1206 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
1209 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1210 xfs_iunlock(ip, lockmode);
1211 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1212 IOMAP_F_SHARED, seq);
1215 xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1216 xfs_iunlock(ip, lockmode);
1217 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, 0, seq);
1220 xfs_iunlock(ip, lockmode);
1225 xfs_buffered_write_delalloc_punch(
1226 struct inode *inode,
1230 xfs_bmap_punch_delalloc_range(XFS_I(inode), offset, offset + length);
1235 xfs_buffered_write_iomap_end(
1236 struct inode *inode,
1241 struct iomap *iomap)
1244 struct xfs_mount *mp = XFS_M(inode->i_sb);
1247 error = iomap_file_buffered_write_punch_delalloc(inode, iomap, offset,
1248 length, written, &xfs_buffered_write_delalloc_punch);
1249 if (error && !xfs_is_shutdown(mp)) {
1250 xfs_alert(mp, "%s: unable to clean up ino 0x%llx",
1251 __func__, XFS_I(inode)->i_ino);
1257 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1258 .iomap_begin = xfs_buffered_write_iomap_begin,
1259 .iomap_end = xfs_buffered_write_iomap_end,
1263 * iomap_page_mkwrite() will never fail in a way that requires delalloc extents
1264 * that it allocated to be revoked. Hence we do not need an .iomap_end method
1265 * for this operation.
1267 const struct iomap_ops xfs_page_mkwrite_iomap_ops = {
1268 .iomap_begin = xfs_buffered_write_iomap_begin,
1272 xfs_read_iomap_begin(
1273 struct inode *inode,
1277 struct iomap *iomap,
1278 struct iomap *srcmap)
1280 struct xfs_inode *ip = XFS_I(inode);
1281 struct xfs_mount *mp = ip->i_mount;
1282 struct xfs_bmbt_irec imap;
1283 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1284 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1285 int nimaps = 1, error = 0;
1286 bool shared = false;
1287 unsigned int lockmode = XFS_ILOCK_SHARED;
1290 ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1292 if (xfs_is_shutdown(mp))
1295 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1298 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1300 if (!error && ((flags & IOMAP_REPORT) || IS_DAX(inode)))
1301 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1302 seq = xfs_iomap_inode_sequence(ip, shared ? IOMAP_F_SHARED : 0);
1303 xfs_iunlock(ip, lockmode);
1307 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1308 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1309 shared ? IOMAP_F_SHARED : 0, seq);
1312 const struct iomap_ops xfs_read_iomap_ops = {
1313 .iomap_begin = xfs_read_iomap_begin,
1317 xfs_seek_iomap_begin(
1318 struct inode *inode,
1322 struct iomap *iomap,
1323 struct iomap *srcmap)
1325 struct xfs_inode *ip = XFS_I(inode);
1326 struct xfs_mount *mp = ip->i_mount;
1327 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1328 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1329 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1330 struct xfs_iext_cursor icur;
1331 struct xfs_bmbt_irec imap, cmap;
1336 if (xfs_is_shutdown(mp))
1339 lockmode = xfs_ilock_data_map_shared(ip);
1340 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1344 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1346 * If we found a data extent we are done.
1348 if (imap.br_startoff <= offset_fsb)
1350 data_fsb = imap.br_startoff;
1353 * Fake a hole until the end of the file.
1355 data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1359 * If a COW fork extent covers the hole, report it - capped to the next
1362 if (xfs_inode_has_cow_data(ip) &&
1363 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1364 cow_fsb = cmap.br_startoff;
1365 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1366 if (data_fsb < cow_fsb + cmap.br_blockcount)
1367 end_fsb = min(end_fsb, data_fsb);
1368 xfs_trim_extent(&cmap, offset_fsb, end_fsb - offset_fsb);
1369 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1370 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1371 IOMAP_F_SHARED, seq);
1373 * This is a COW extent, so we must probe the page cache
1374 * because there could be dirty page cache being backed
1377 iomap->type = IOMAP_UNWRITTEN;
1382 * Else report a hole, capped to the next found data or COW extent.
1384 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1385 imap.br_blockcount = cow_fsb - offset_fsb;
1387 imap.br_blockcount = data_fsb - offset_fsb;
1388 imap.br_startoff = offset_fsb;
1389 imap.br_startblock = HOLESTARTBLOCK;
1390 imap.br_state = XFS_EXT_NORM;
1392 seq = xfs_iomap_inode_sequence(ip, 0);
1393 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1394 error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1396 xfs_iunlock(ip, lockmode);
1400 const struct iomap_ops xfs_seek_iomap_ops = {
1401 .iomap_begin = xfs_seek_iomap_begin,
1405 xfs_xattr_iomap_begin(
1406 struct inode *inode,
1410 struct iomap *iomap,
1411 struct iomap *srcmap)
1413 struct xfs_inode *ip = XFS_I(inode);
1414 struct xfs_mount *mp = ip->i_mount;
1415 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1416 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1417 struct xfs_bmbt_irec imap;
1418 int nimaps = 1, error = 0;
1422 if (xfs_is_shutdown(mp))
1425 lockmode = xfs_ilock_attr_map_shared(ip);
1427 /* if there are no attribute fork or extents, return ENOENT */
1428 if (!xfs_inode_has_attr_fork(ip) || !ip->i_af.if_nextents) {
1433 ASSERT(ip->i_af.if_format != XFS_DINODE_FMT_LOCAL);
1434 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1435 &nimaps, XFS_BMAPI_ATTRFORK);
1438 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_XATTR);
1439 xfs_iunlock(ip, lockmode);
1444 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_XATTR, seq);
1447 const struct iomap_ops xfs_xattr_iomap_ops = {
1448 .iomap_begin = xfs_xattr_iomap_begin,
1453 struct xfs_inode *ip,
1458 struct inode *inode = VFS_I(ip);
1461 return dax_zero_range(inode, pos, len, did_zero,
1462 &xfs_dax_write_iomap_ops);
1463 return iomap_zero_range(inode, pos, len, did_zero,
1464 &xfs_buffered_write_iomap_ops);
1469 struct xfs_inode *ip,
1473 struct inode *inode = VFS_I(ip);
1476 return dax_truncate_page(inode, pos, did_zero,
1477 &xfs_dax_write_iomap_ops);
1478 return iomap_truncate_page(inode, pos, did_zero,
1479 &xfs_buffered_write_iomap_ops);