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"
31 #define XFS_ALLOC_ALIGN(mp, off) \
32 (((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
35 xfs_alert_fsblock_zero(
37 xfs_bmbt_irec_t *imap)
39 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
40 "Access to block zero in inode %llu "
41 "start_block: %llx start_off: %llx "
42 "blkcnt: %llx extent-state: %x",
43 (unsigned long long)ip->i_ino,
44 (unsigned long long)imap->br_startblock,
45 (unsigned long long)imap->br_startoff,
46 (unsigned long long)imap->br_blockcount,
52 xfs_iomap_inode_sequence(
58 if (iomap_flags & IOMAP_F_XATTR)
59 return READ_ONCE(ip->i_af.if_seq);
60 if ((iomap_flags & IOMAP_F_SHARED) && ip->i_cowfp)
61 cookie = (u64)READ_ONCE(ip->i_cowfp->if_seq) << 32;
62 return cookie | READ_ONCE(ip->i_df.if_seq);
66 * Check that the iomap passed to us is still valid for the given offset and
72 const struct iomap *iomap)
74 struct xfs_inode *ip = XFS_I(inode);
76 if (iomap->validity_cookie !=
77 xfs_iomap_inode_sequence(ip, iomap->flags)) {
78 trace_xfs_iomap_invalid(ip, iomap);
82 XFS_ERRORTAG_DELAY(ip->i_mount, XFS_ERRTAG_WRITE_DELAY_MS);
86 const struct iomap_page_ops xfs_iomap_page_ops = {
87 .iomap_valid = xfs_iomap_valid,
94 struct xfs_bmbt_irec *imap,
95 unsigned int mapping_flags,
99 struct xfs_mount *mp = ip->i_mount;
100 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
102 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
103 return xfs_alert_fsblock_zero(ip, imap);
105 if (imap->br_startblock == HOLESTARTBLOCK) {
106 iomap->addr = IOMAP_NULL_ADDR;
107 iomap->type = IOMAP_HOLE;
108 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
109 isnullstartblock(imap->br_startblock)) {
110 iomap->addr = IOMAP_NULL_ADDR;
111 iomap->type = IOMAP_DELALLOC;
113 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
114 if (mapping_flags & IOMAP_DAX)
115 iomap->addr += target->bt_dax_part_off;
117 if (imap->br_state == XFS_EXT_UNWRITTEN)
118 iomap->type = IOMAP_UNWRITTEN;
120 iomap->type = IOMAP_MAPPED;
123 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
124 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
125 if (mapping_flags & IOMAP_DAX)
126 iomap->dax_dev = target->bt_daxdev;
128 iomap->bdev = target->bt_bdev;
129 iomap->flags = iomap_flags;
131 if (xfs_ipincount(ip) &&
132 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
133 iomap->flags |= IOMAP_F_DIRTY;
135 iomap->validity_cookie = sequence_cookie;
136 iomap->page_ops = &xfs_iomap_page_ops;
142 struct xfs_inode *ip,
144 xfs_fileoff_t offset_fsb,
145 xfs_fileoff_t end_fsb)
147 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
149 iomap->addr = IOMAP_NULL_ADDR;
150 iomap->type = IOMAP_HOLE;
151 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
152 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
153 iomap->bdev = target->bt_bdev;
154 iomap->dax_dev = target->bt_daxdev;
157 static inline xfs_fileoff_t
159 struct xfs_mount *mp,
163 ASSERT(offset <= mp->m_super->s_maxbytes);
164 return min(XFS_B_TO_FSB(mp, offset + count),
165 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
170 struct xfs_inode *ip)
172 struct xfs_mount *mp = ip->i_mount;
173 xfs_extlen_t align = 0;
175 if (!XFS_IS_REALTIME_INODE(ip)) {
177 * Round up the allocation request to a stripe unit
178 * (m_dalign) boundary if the file size is >= stripe unit
179 * size, and we are allocating past the allocation eof.
181 * If mounted with the "-o swalloc" option the alignment is
182 * increased from the strip unit size to the stripe width.
184 if (mp->m_swidth && xfs_has_swalloc(mp))
185 align = mp->m_swidth;
186 else if (mp->m_dalign)
187 align = mp->m_dalign;
189 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
197 * Check if last_fsb is outside the last extent, and if so grow it to the next
198 * stripe unit boundary.
201 xfs_iomap_eof_align_last_fsb(
202 struct xfs_inode *ip,
203 xfs_fileoff_t end_fsb)
205 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
206 xfs_extlen_t extsz = xfs_get_extsz_hint(ip);
207 xfs_extlen_t align = xfs_eof_alignment(ip);
208 struct xfs_bmbt_irec irec;
209 struct xfs_iext_cursor icur;
211 ASSERT(!xfs_need_iread_extents(ifp));
214 * Always round up the allocation request to the extent hint boundary.
218 align = roundup_64(align, extsz);
224 xfs_fileoff_t aligned_end_fsb = roundup_64(end_fsb, align);
226 xfs_iext_last(ifp, &icur);
227 if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
228 aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
229 return aligned_end_fsb;
236 xfs_iomap_write_direct(
237 struct xfs_inode *ip,
238 xfs_fileoff_t offset_fsb,
239 xfs_fileoff_t count_fsb,
241 struct xfs_bmbt_irec *imap,
244 struct xfs_mount *mp = ip->i_mount;
245 struct xfs_trans *tp;
246 xfs_filblks_t resaligned;
248 unsigned int dblocks, rblocks;
251 int bmapi_flags = XFS_BMAPI_PREALLOC;
252 int nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
254 ASSERT(count_fsb > 0);
256 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
257 xfs_get_extsz_hint(ip));
258 if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
259 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
260 rblocks = resaligned;
262 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
266 error = xfs_qm_dqattach(ip);
271 * For DAX, we do not allocate unwritten extents, but instead we zero
272 * the block before we commit the transaction. Ideally we'd like to do
273 * this outside the transaction context, but if we commit and then crash
274 * we may not have zeroed the blocks and this will be exposed on
275 * recovery of the allocation. Hence we must zero before commit.
277 * Further, if we are mapping unwritten extents here, we need to zero
278 * and convert them to written so that we don't need an unwritten extent
279 * callback for DAX. This also means that we need to be able to dip into
280 * the reserve block pool for bmbt block allocation if there is no space
281 * left but we need to do unwritten extent conversion.
283 if (flags & IOMAP_DAX) {
284 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
285 if (imap->br_state == XFS_EXT_UNWRITTEN) {
287 nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
288 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
292 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
293 rblocks, force, &tp);
297 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, nr_exts);
299 error = xfs_iext_count_upgrade(tp, ip, nr_exts);
301 goto out_trans_cancel;
304 * From this point onwards we overwrite the imap pointer that the
308 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
311 goto out_trans_cancel;
314 * Complete the transaction
316 error = xfs_trans_commit(tp);
321 * Copy any maps to caller's array and return any error.
328 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
329 error = xfs_alert_fsblock_zero(ip, imap);
332 *seq = xfs_iomap_inode_sequence(ip, 0);
333 xfs_iunlock(ip, XFS_ILOCK_EXCL);
337 xfs_trans_cancel(tp);
342 xfs_quota_need_throttle(
343 struct xfs_inode *ip,
345 xfs_fsblock_t alloc_blocks)
347 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
349 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
352 /* no hi watermark, no throttle */
353 if (!dq->q_prealloc_hi_wmark)
356 /* under the lo watermark, no throttle */
357 if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
364 xfs_quota_calc_throttle(
365 struct xfs_inode *ip,
367 xfs_fsblock_t *qblocks,
371 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
375 /* no dq, or over hi wmark, squash the prealloc completely */
376 if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
382 freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
383 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
385 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
387 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
391 if (freesp < *qfreesp)
394 /* only overwrite the throttle values if we are more aggressive */
395 if ((freesp >> shift) < (*qblocks >> *qshift)) {
402 * If we don't have a user specified preallocation size, dynamically increase
403 * the preallocation size as the size of the file grows. Cap the maximum size
404 * at a single extent or less if the filesystem is near full. The closer the
405 * filesystem is to being full, the smaller the maximum preallocation.
408 xfs_iomap_prealloc_size(
409 struct xfs_inode *ip,
413 struct xfs_iext_cursor *icur)
415 struct xfs_iext_cursor ncur = *icur;
416 struct xfs_bmbt_irec prev, got;
417 struct xfs_mount *mp = ip->i_mount;
418 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
419 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
421 xfs_fsblock_t qblocks;
422 xfs_fsblock_t alloc_blocks = 0;
428 * As an exception we don't do any preallocation at all if the file is
429 * smaller than the minimum preallocation and we are using the default
430 * dynamic preallocation scheme, as it is likely this is the only write
431 * to the file that is going to be done.
433 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
437 * Use the minimum preallocation size for small files or if we are
438 * writing right after a hole.
440 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
441 !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
442 prev.br_startoff + prev.br_blockcount < offset_fsb)
443 return mp->m_allocsize_blocks;
446 * Take the size of the preceding data extents as the basis for the
447 * preallocation size. Note that we don't care if the previous extents
448 * are written or not.
450 plen = prev.br_blockcount;
451 while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
452 if (plen > XFS_MAX_BMBT_EXTLEN / 2 ||
453 isnullstartblock(got.br_startblock) ||
454 got.br_startoff + got.br_blockcount != prev.br_startoff ||
455 got.br_startblock + got.br_blockcount != prev.br_startblock)
457 plen += got.br_blockcount;
462 * If the size of the extents is greater than half the maximum extent
463 * length, then use the current offset as the basis. This ensures that
464 * for large files the preallocation size always extends to
465 * XFS_BMBT_MAX_EXTLEN rather than falling short due to things like stripe
466 * unit/width alignment of real extents.
468 alloc_blocks = plen * 2;
469 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
470 alloc_blocks = XFS_B_TO_FSB(mp, offset);
471 qblocks = alloc_blocks;
474 * XFS_BMBT_MAX_EXTLEN is not a power of two value but we round the prealloc
475 * down to the nearest power of two value after throttling. To prevent
476 * the round down from unconditionally reducing the maximum supported
477 * prealloc size, we round up first, apply appropriate throttling, round
478 * down and cap the value to XFS_BMBT_MAX_EXTLEN.
480 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(XFS_MAX_BMBT_EXTLEN),
483 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
484 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
486 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
488 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
490 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
492 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
497 * Check each quota to cap the prealloc size, provide a shift value to
498 * throttle with and adjust amount of available space.
500 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
501 xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
503 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
504 xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
506 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
507 xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
511 * The final prealloc size is set to the minimum of free space available
512 * in each of the quotas and the overall filesystem.
514 * The shift throttle value is set to the maximum value as determined by
515 * the global low free space values and per-quota low free space values.
517 alloc_blocks = min(alloc_blocks, qblocks);
518 shift = max(shift, qshift);
521 alloc_blocks >>= shift;
523 * rounddown_pow_of_two() returns an undefined result if we pass in
527 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
528 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
529 alloc_blocks = XFS_MAX_BMBT_EXTLEN;
532 * If we are still trying to allocate more space than is
533 * available, squash the prealloc hard. This can happen if we
534 * have a large file on a small filesystem and the above
535 * lowspace thresholds are smaller than XFS_BMBT_MAX_EXTLEN.
537 while (alloc_blocks && alloc_blocks >= freesp)
539 if (alloc_blocks < mp->m_allocsize_blocks)
540 alloc_blocks = mp->m_allocsize_blocks;
541 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
542 mp->m_allocsize_blocks);
547 xfs_iomap_write_unwritten(
553 xfs_mount_t *mp = ip->i_mount;
554 xfs_fileoff_t offset_fsb;
555 xfs_filblks_t count_fsb;
556 xfs_filblks_t numblks_fsb;
559 xfs_bmbt_irec_t imap;
560 struct inode *inode = VFS_I(ip);
565 trace_xfs_unwritten_convert(ip, offset, count);
567 offset_fsb = XFS_B_TO_FSBT(mp, offset);
568 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
569 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
572 * Reserve enough blocks in this transaction for two complete extent
573 * btree splits. We may be converting the middle part of an unwritten
574 * extent and in this case we will insert two new extents in the btree
575 * each of which could cause a full split.
577 * This reservation amount will be used in the first call to
578 * xfs_bmbt_split() to select an AG with enough space to satisfy the
579 * rest of the operation.
581 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
583 /* Attach dquots so that bmbt splits are accounted correctly. */
584 error = xfs_qm_dqattach(ip);
590 * Set up a transaction to convert the range of extents
591 * from unwritten to real. Do allocations in a loop until
592 * we have covered the range passed in.
594 * Note that we can't risk to recursing back into the filesystem
595 * here as we might be asked to write out the same inode that we
596 * complete here and might deadlock on the iolock.
598 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
603 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
604 XFS_IEXT_WRITE_UNWRITTEN_CNT);
606 error = xfs_iext_count_upgrade(tp, ip,
607 XFS_IEXT_WRITE_UNWRITTEN_CNT);
609 goto error_on_bmapi_transaction;
612 * Modify the unwritten extent state of the buffer.
615 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
616 XFS_BMAPI_CONVERT, resblks, &imap,
619 goto error_on_bmapi_transaction;
622 * Log the updated inode size as we go. We have to be careful
623 * to only log it up to the actual write offset if it is
624 * halfway into a block.
626 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
627 if (i_size > offset + count)
628 i_size = offset + count;
629 if (update_isize && i_size > i_size_read(inode))
630 i_size_write(inode, i_size);
631 i_size = xfs_new_eof(ip, i_size);
633 ip->i_disk_size = i_size;
634 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
637 error = xfs_trans_commit(tp);
638 xfs_iunlock(ip, XFS_ILOCK_EXCL);
642 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
643 return xfs_alert_fsblock_zero(ip, &imap);
645 if ((numblks_fsb = imap.br_blockcount) == 0) {
647 * The numblks_fsb value should always get
648 * smaller, otherwise the loop is stuck.
650 ASSERT(imap.br_blockcount);
653 offset_fsb += numblks_fsb;
654 count_fsb -= numblks_fsb;
655 } while (count_fsb > 0);
659 error_on_bmapi_transaction:
660 xfs_trans_cancel(tp);
661 xfs_iunlock(ip, XFS_ILOCK_EXCL);
669 struct xfs_bmbt_irec *imap,
672 /* don't allocate blocks when just zeroing */
673 if (flags & IOMAP_ZERO)
676 imap->br_startblock == HOLESTARTBLOCK ||
677 imap->br_startblock == DELAYSTARTBLOCK)
679 /* we convert unwritten extents before copying the data for DAX */
680 if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
687 struct xfs_inode *ip,
689 struct xfs_bmbt_irec *imap,
692 if (!xfs_is_cow_inode(ip))
695 /* when zeroing we don't have to COW holes or unwritten extents */
696 if (flags & IOMAP_ZERO) {
698 imap->br_startblock == HOLESTARTBLOCK ||
699 imap->br_state == XFS_EXT_UNWRITTEN)
708 struct xfs_inode *ip,
712 unsigned int mode = *lockmode;
713 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
716 * COW writes may allocate delalloc space or convert unwritten COW
717 * extents, so we need to make sure to take the lock exclusively here.
719 if (xfs_is_cow_inode(ip) && is_write)
720 mode = XFS_ILOCK_EXCL;
723 * Extents not yet cached requires exclusive access, don't block. This
724 * is an opencoded xfs_ilock_data_map_shared() call but with
725 * non-blocking behaviour.
727 if (xfs_need_iread_extents(&ip->i_df)) {
728 if (flags & IOMAP_NOWAIT)
730 mode = XFS_ILOCK_EXCL;
734 if (flags & IOMAP_NOWAIT) {
735 if (!xfs_ilock_nowait(ip, mode))
742 * The reflink iflag could have changed since the earlier unlocked
743 * check, so if we got ILOCK_SHARED for a write and but we're now a
744 * reflink inode we have to switch to ILOCK_EXCL and relock.
746 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
747 xfs_iunlock(ip, mode);
748 mode = XFS_ILOCK_EXCL;
757 * Check that the imap we are going to return to the caller spans the entire
758 * range that the caller requested for the IO.
762 struct xfs_bmbt_irec *imap,
763 xfs_fileoff_t offset_fsb,
764 xfs_fileoff_t end_fsb)
766 if (imap->br_startoff > offset_fsb)
768 if (imap->br_startoff + imap->br_blockcount < end_fsb)
774 xfs_direct_write_iomap_begin(
780 struct iomap *srcmap)
782 struct xfs_inode *ip = XFS_I(inode);
783 struct xfs_mount *mp = ip->i_mount;
784 struct xfs_bmbt_irec imap, cmap;
785 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
786 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
787 int nimaps = 1, error = 0;
790 unsigned int lockmode = XFS_ILOCK_SHARED;
793 ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
795 if (xfs_is_shutdown(mp))
799 * Writes that span EOF might trigger an IO size update on completion,
800 * so consider them to be dirty for the purposes of O_DSYNC even if
801 * there is no other metadata changes pending or have been made here.
803 if (offset + length > i_size_read(inode))
804 iomap_flags |= IOMAP_F_DIRTY;
806 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
810 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
815 if (imap_needs_cow(ip, flags, &imap, nimaps)) {
817 if (flags & IOMAP_NOWAIT)
820 /* may drop and re-acquire the ilock */
821 error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
823 (flags & IOMAP_DIRECT) || IS_DAX(inode));
828 end_fsb = imap.br_startoff + imap.br_blockcount;
829 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
832 if (imap_needs_alloc(inode, flags, &imap, nimaps))
833 goto allocate_blocks;
836 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
837 * a single map so that we avoid partial IO failures due to the rest of
838 * the I/O range not covered by this map triggering an EAGAIN condition
839 * when it is subsequently mapped and aborting the I/O.
841 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
843 if (!imap_spans_range(&imap, offset_fsb, end_fsb))
848 * For overwrite only I/O, we cannot convert unwritten extents without
849 * requiring sub-block zeroing. This can only be done under an
850 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
851 * extent to tell the caller to try again.
853 if (flags & IOMAP_OVERWRITE_ONLY) {
855 if (imap.br_state != XFS_EXT_NORM &&
856 ((offset | length) & mp->m_blockmask))
860 seq = xfs_iomap_inode_sequence(ip, iomap_flags);
861 xfs_iunlock(ip, lockmode);
862 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
863 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
867 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
871 * We cap the maximum length we map to a sane size to keep the chunks
872 * of work done where somewhat symmetric with the work writeback does.
873 * This is a completely arbitrary number pulled out of thin air as a
874 * best guess for initial testing.
876 * Note that the values needs to be less than 32-bits wide until the
877 * lower level functions are updated.
879 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
880 end_fsb = xfs_iomap_end_fsb(mp, offset, length);
882 if (offset + length > XFS_ISIZE(ip))
883 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
884 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
885 end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
886 xfs_iunlock(ip, lockmode);
888 error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
893 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
894 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
895 iomap_flags | IOMAP_F_NEW, seq);
898 length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
899 trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
900 if (imap.br_startblock != HOLESTARTBLOCK) {
901 seq = xfs_iomap_inode_sequence(ip, 0);
902 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
906 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
907 xfs_iunlock(ip, lockmode);
908 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED, seq);
912 xfs_iunlock(ip, lockmode);
916 const struct iomap_ops xfs_direct_write_iomap_ops = {
917 .iomap_begin = xfs_direct_write_iomap_begin,
921 xfs_dax_write_iomap_end(
929 struct xfs_inode *ip = XFS_I(inode);
931 if (!xfs_is_cow_inode(ip))
935 xfs_reflink_cancel_cow_range(ip, pos, length, true);
939 return xfs_reflink_end_cow(ip, pos, written);
942 const struct iomap_ops xfs_dax_write_iomap_ops = {
943 .iomap_begin = xfs_direct_write_iomap_begin,
944 .iomap_end = xfs_dax_write_iomap_end,
948 xfs_buffered_write_iomap_begin(
954 struct iomap *srcmap)
956 struct xfs_inode *ip = XFS_I(inode);
957 struct xfs_mount *mp = ip->i_mount;
958 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
959 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, count);
960 struct xfs_bmbt_irec imap, cmap;
961 struct xfs_iext_cursor icur, ccur;
962 xfs_fsblock_t prealloc_blocks = 0;
963 bool eof = false, cow_eof = false, shared = false;
964 int allocfork = XFS_DATA_FORK;
966 unsigned int lockmode = XFS_ILOCK_EXCL;
969 if (xfs_is_shutdown(mp))
972 /* we can't use delayed allocations when using extent size hints */
973 if (xfs_get_extsz_hint(ip))
974 return xfs_direct_write_iomap_begin(inode, offset, count,
975 flags, iomap, srcmap);
977 ASSERT(!XFS_IS_REALTIME_INODE(ip));
979 error = xfs_qm_dqattach(ip);
983 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
987 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
988 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
989 error = -EFSCORRUPTED;
993 XFS_STATS_INC(mp, xs_blk_mapw);
995 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1000 * Search the data fork first to look up our source mapping. We
1001 * always need the data fork map, as we have to return it to the
1002 * iomap code so that the higher level write code can read data in to
1003 * perform read-modify-write cycles for unaligned writes.
1005 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
1007 imap.br_startoff = end_fsb; /* fake hole until the end */
1009 /* We never need to allocate blocks for zeroing a hole. */
1010 if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
1011 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
1016 * Search the COW fork extent list even if we did not find a data fork
1017 * extent. This serves two purposes: first this implements the
1018 * speculative preallocation using cowextsize, so that we also unshare
1019 * block adjacent to shared blocks instead of just the shared blocks
1020 * themselves. Second the lookup in the extent list is generally faster
1021 * than going out to the shared extent tree.
1023 if (xfs_is_cow_inode(ip)) {
1025 ASSERT(!xfs_is_reflink_inode(ip));
1026 xfs_ifork_init_cow(ip);
1028 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
1030 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
1031 trace_xfs_reflink_cow_found(ip, &cmap);
1036 if (imap.br_startoff <= offset_fsb) {
1038 * For reflink files we may need a delalloc reservation when
1039 * overwriting shared extents. This includes zeroing of
1040 * existing extents that contain data.
1042 if (!xfs_is_cow_inode(ip) ||
1043 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
1044 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1049 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1051 /* Trim the mapping to the nearest shared extent boundary. */
1052 error = xfs_bmap_trim_cow(ip, &imap, &shared);
1056 /* Not shared? Just report the (potentially capped) extent. */
1058 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1064 * Fork all the shared blocks from our write offset until the
1065 * end of the extent.
1067 allocfork = XFS_COW_FORK;
1068 end_fsb = imap.br_startoff + imap.br_blockcount;
1071 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1072 * pages to keep the chunks of work done where somewhat
1073 * symmetric with the work writeback does. This is a completely
1074 * arbitrary number pulled out of thin air.
1076 * Note that the values needs to be less than 32-bits wide until
1077 * the lower level functions are updated.
1079 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
1080 end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1082 if (xfs_is_always_cow_inode(ip))
1083 allocfork = XFS_COW_FORK;
1086 if (eof && offset + count > XFS_ISIZE(ip)) {
1088 * Determine the initial size of the preallocation.
1089 * We clean up any extra preallocation when the file is closed.
1091 if (xfs_has_allocsize(mp))
1092 prealloc_blocks = mp->m_allocsize_blocks;
1094 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1095 offset, count, &icur);
1096 if (prealloc_blocks) {
1098 xfs_off_t end_offset;
1099 xfs_fileoff_t p_end_fsb;
1101 end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1102 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1105 align = xfs_eof_alignment(ip);
1107 p_end_fsb = roundup_64(p_end_fsb, align);
1109 p_end_fsb = min(p_end_fsb,
1110 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1111 ASSERT(p_end_fsb > offset_fsb);
1112 prealloc_blocks = p_end_fsb - end_fsb;
1117 error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1118 end_fsb - offset_fsb, prealloc_blocks,
1119 allocfork == XFS_DATA_FORK ? &imap : &cmap,
1120 allocfork == XFS_DATA_FORK ? &icur : &ccur,
1121 allocfork == XFS_DATA_FORK ? eof : cow_eof);
1127 /* retry without any preallocation */
1128 trace_xfs_delalloc_enospc(ip, offset, count);
1129 if (prealloc_blocks) {
1130 prealloc_blocks = 0;
1138 if (allocfork == XFS_COW_FORK) {
1139 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1144 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1145 * them out if the write happens to fail.
1147 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_NEW);
1148 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1149 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1150 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_NEW, seq);
1153 seq = xfs_iomap_inode_sequence(ip, 0);
1154 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1155 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1158 seq = xfs_iomap_inode_sequence(ip, 0);
1159 if (imap.br_startoff <= offset_fsb) {
1160 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
1163 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1164 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1165 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1166 IOMAP_F_SHARED, seq);
1169 xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1170 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1171 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, 0, seq);
1174 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1179 xfs_buffered_write_delalloc_punch(
1180 struct inode *inode,
1184 return xfs_bmap_punch_delalloc_range(XFS_I(inode), offset,
1189 xfs_buffered_write_iomap_end(
1190 struct inode *inode,
1195 struct iomap *iomap)
1198 struct xfs_mount *mp = XFS_M(inode->i_sb);
1201 error = iomap_file_buffered_write_punch_delalloc(inode, iomap, offset,
1202 length, written, &xfs_buffered_write_delalloc_punch);
1203 if (error && !xfs_is_shutdown(mp)) {
1204 xfs_alert(mp, "%s: unable to clean up ino 0x%llx",
1205 __func__, XFS_I(inode)->i_ino);
1211 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1212 .iomap_begin = xfs_buffered_write_iomap_begin,
1213 .iomap_end = xfs_buffered_write_iomap_end,
1217 * iomap_page_mkwrite() will never fail in a way that requires delalloc extents
1218 * that it allocated to be revoked. Hence we do not need an .iomap_end method
1219 * for this operation.
1221 const struct iomap_ops xfs_page_mkwrite_iomap_ops = {
1222 .iomap_begin = xfs_buffered_write_iomap_begin,
1226 xfs_read_iomap_begin(
1227 struct inode *inode,
1231 struct iomap *iomap,
1232 struct iomap *srcmap)
1234 struct xfs_inode *ip = XFS_I(inode);
1235 struct xfs_mount *mp = ip->i_mount;
1236 struct xfs_bmbt_irec imap;
1237 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1238 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1239 int nimaps = 1, error = 0;
1240 bool shared = false;
1241 unsigned int lockmode = XFS_ILOCK_SHARED;
1244 ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1246 if (xfs_is_shutdown(mp))
1249 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1252 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1254 if (!error && ((flags & IOMAP_REPORT) || IS_DAX(inode)))
1255 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1256 seq = xfs_iomap_inode_sequence(ip, shared ? IOMAP_F_SHARED : 0);
1257 xfs_iunlock(ip, lockmode);
1261 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1262 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1263 shared ? IOMAP_F_SHARED : 0, seq);
1266 const struct iomap_ops xfs_read_iomap_ops = {
1267 .iomap_begin = xfs_read_iomap_begin,
1271 xfs_seek_iomap_begin(
1272 struct inode *inode,
1276 struct iomap *iomap,
1277 struct iomap *srcmap)
1279 struct xfs_inode *ip = XFS_I(inode);
1280 struct xfs_mount *mp = ip->i_mount;
1281 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1282 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1283 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1284 struct xfs_iext_cursor icur;
1285 struct xfs_bmbt_irec imap, cmap;
1290 if (xfs_is_shutdown(mp))
1293 lockmode = xfs_ilock_data_map_shared(ip);
1294 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1298 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1300 * If we found a data extent we are done.
1302 if (imap.br_startoff <= offset_fsb)
1304 data_fsb = imap.br_startoff;
1307 * Fake a hole until the end of the file.
1309 data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1313 * If a COW fork extent covers the hole, report it - capped to the next
1316 if (xfs_inode_has_cow_data(ip) &&
1317 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1318 cow_fsb = cmap.br_startoff;
1319 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1320 if (data_fsb < cow_fsb + cmap.br_blockcount)
1321 end_fsb = min(end_fsb, data_fsb);
1322 xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1323 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1324 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1325 IOMAP_F_SHARED, seq);
1327 * This is a COW extent, so we must probe the page cache
1328 * because there could be dirty page cache being backed
1331 iomap->type = IOMAP_UNWRITTEN;
1336 * Else report a hole, capped to the next found data or COW extent.
1338 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1339 imap.br_blockcount = cow_fsb - offset_fsb;
1341 imap.br_blockcount = data_fsb - offset_fsb;
1342 imap.br_startoff = offset_fsb;
1343 imap.br_startblock = HOLESTARTBLOCK;
1344 imap.br_state = XFS_EXT_NORM;
1346 seq = xfs_iomap_inode_sequence(ip, 0);
1347 xfs_trim_extent(&imap, offset_fsb, end_fsb);
1348 error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1350 xfs_iunlock(ip, lockmode);
1354 const struct iomap_ops xfs_seek_iomap_ops = {
1355 .iomap_begin = xfs_seek_iomap_begin,
1359 xfs_xattr_iomap_begin(
1360 struct inode *inode,
1364 struct iomap *iomap,
1365 struct iomap *srcmap)
1367 struct xfs_inode *ip = XFS_I(inode);
1368 struct xfs_mount *mp = ip->i_mount;
1369 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1370 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1371 struct xfs_bmbt_irec imap;
1372 int nimaps = 1, error = 0;
1376 if (xfs_is_shutdown(mp))
1379 lockmode = xfs_ilock_attr_map_shared(ip);
1381 /* if there are no attribute fork or extents, return ENOENT */
1382 if (!xfs_inode_has_attr_fork(ip) || !ip->i_af.if_nextents) {
1387 ASSERT(ip->i_af.if_format != XFS_DINODE_FMT_LOCAL);
1388 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1389 &nimaps, XFS_BMAPI_ATTRFORK);
1392 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_XATTR);
1393 xfs_iunlock(ip, lockmode);
1398 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_XATTR, seq);
1401 const struct iomap_ops xfs_xattr_iomap_ops = {
1402 .iomap_begin = xfs_xattr_iomap_begin,
1407 struct xfs_inode *ip,
1412 struct inode *inode = VFS_I(ip);
1415 return dax_zero_range(inode, pos, len, did_zero,
1416 &xfs_dax_write_iomap_ops);
1417 return iomap_zero_range(inode, pos, len, did_zero,
1418 &xfs_buffered_write_iomap_ops);
1423 struct xfs_inode *ip,
1427 struct inode *inode = VFS_I(ip);
1430 return dax_truncate_page(inode, pos, did_zero,
1431 &xfs_dax_write_iomap_ops);
1432 return iomap_truncate_page(inode, pos, did_zero,
1433 &xfs_buffered_write_iomap_ops);