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"
32 #define XFS_ALLOC_ALIGN(mp, off) \
33 (((off) >> mp->m_allocsize_log) << mp->m_allocsize_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,
49 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
54 xfs_iomap_inode_sequence(
60 if (iomap_flags & IOMAP_F_XATTR)
61 return READ_ONCE(ip->i_af.if_seq);
62 if ((iomap_flags & IOMAP_F_SHARED) && ip->i_cowfp)
63 cookie = (u64)READ_ONCE(ip->i_cowfp->if_seq) << 32;
64 return cookie | READ_ONCE(ip->i_df.if_seq);
68 * Check that the iomap passed to us is still valid for the given offset and
74 const struct iomap *iomap)
76 struct xfs_inode *ip = XFS_I(inode);
78 if (iomap->validity_cookie !=
79 xfs_iomap_inode_sequence(ip, iomap->flags)) {
80 trace_xfs_iomap_invalid(ip, iomap);
84 XFS_ERRORTAG_DELAY(ip->i_mount, XFS_ERRTAG_WRITE_DELAY_MS);
88 static const struct iomap_folio_ops xfs_iomap_folio_ops = {
89 .iomap_valid = xfs_iomap_valid,
96 struct xfs_bmbt_irec *imap,
97 unsigned int mapping_flags,
101 struct xfs_mount *mp = ip->i_mount;
102 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
104 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
105 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
106 return xfs_alert_fsblock_zero(ip, imap);
109 if (imap->br_startblock == HOLESTARTBLOCK) {
110 iomap->addr = IOMAP_NULL_ADDR;
111 iomap->type = IOMAP_HOLE;
112 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
113 isnullstartblock(imap->br_startblock)) {
114 iomap->addr = IOMAP_NULL_ADDR;
115 iomap->type = IOMAP_DELALLOC;
117 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
118 if (mapping_flags & IOMAP_DAX)
119 iomap->addr += target->bt_dax_part_off;
121 if (imap->br_state == XFS_EXT_UNWRITTEN)
122 iomap->type = IOMAP_UNWRITTEN;
124 iomap->type = IOMAP_MAPPED;
127 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
128 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
129 if (mapping_flags & IOMAP_DAX)
130 iomap->dax_dev = target->bt_daxdev;
132 iomap->bdev = target->bt_bdev;
133 iomap->flags = iomap_flags;
135 if (xfs_ipincount(ip) &&
136 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
137 iomap->flags |= IOMAP_F_DIRTY;
139 iomap->validity_cookie = sequence_cookie;
140 iomap->folio_ops = &xfs_iomap_folio_ops;
146 struct xfs_inode *ip,
148 xfs_fileoff_t offset_fsb,
149 xfs_fileoff_t end_fsb)
151 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
153 iomap->addr = IOMAP_NULL_ADDR;
154 iomap->type = IOMAP_HOLE;
155 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
156 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
157 iomap->bdev = target->bt_bdev;
158 iomap->dax_dev = target->bt_daxdev;
161 static inline xfs_fileoff_t
163 struct xfs_mount *mp,
167 ASSERT(offset <= mp->m_super->s_maxbytes);
168 return min(XFS_B_TO_FSB(mp, offset + count),
169 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
174 struct xfs_inode *ip)
176 struct xfs_mount *mp = ip->i_mount;
177 xfs_extlen_t align = 0;
179 if (!XFS_IS_REALTIME_INODE(ip)) {
181 * Round up the allocation request to a stripe unit
182 * (m_dalign) boundary if the file size is >= stripe unit
183 * size, and we are allocating past the allocation eof.
185 * If mounted with the "-o swalloc" option the alignment is
186 * increased from the strip unit size to the stripe width.
188 if (mp->m_swidth && xfs_has_swalloc(mp))
189 align = mp->m_swidth;
190 else if (mp->m_dalign)
191 align = mp->m_dalign;
193 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
201 * Check if last_fsb is outside the last extent, and if so grow it to the next
202 * stripe unit boundary.
205 xfs_iomap_eof_align_last_fsb(
206 struct xfs_inode *ip,
207 xfs_fileoff_t end_fsb)
209 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
210 xfs_extlen_t extsz = xfs_get_extsz_hint(ip);
211 xfs_extlen_t align = xfs_eof_alignment(ip);
212 struct xfs_bmbt_irec irec;
213 struct xfs_iext_cursor icur;
215 ASSERT(!xfs_need_iread_extents(ifp));
218 * Always round up the allocation request to the extent hint boundary.
222 align = roundup_64(align, extsz);
228 xfs_fileoff_t aligned_end_fsb = roundup_64(end_fsb, align);
230 xfs_iext_last(ifp, &icur);
231 if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
232 aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
233 return aligned_end_fsb;
240 xfs_iomap_write_direct(
241 struct xfs_inode *ip,
242 xfs_fileoff_t offset_fsb,
243 xfs_fileoff_t count_fsb,
245 struct xfs_bmbt_irec *imap,
248 struct xfs_mount *mp = ip->i_mount;
249 struct xfs_trans *tp;
250 xfs_filblks_t resaligned;
252 unsigned int dblocks, rblocks;
255 int bmapi_flags = XFS_BMAPI_PREALLOC;
256 int nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
258 ASSERT(count_fsb > 0);
260 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
261 xfs_get_extsz_hint(ip));
262 if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
263 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
264 rblocks = resaligned;
266 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
270 error = xfs_qm_dqattach(ip);
275 * For DAX, we do not allocate unwritten extents, but instead we zero
276 * the block before we commit the transaction. Ideally we'd like to do
277 * this outside the transaction context, but if we commit and then crash
278 * we may not have zeroed the blocks and this will be exposed on
279 * recovery of the allocation. Hence we must zero before commit.
281 * Further, if we are mapping unwritten extents here, we need to zero
282 * and convert them to written so that we don't need an unwritten extent
283 * callback for DAX. This also means that we need to be able to dip into
284 * the reserve block pool for bmbt block allocation if there is no space
285 * left but we need to do unwritten extent conversion.
287 if (flags & IOMAP_DAX) {
288 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
289 if (imap->br_state == XFS_EXT_UNWRITTEN) {
291 nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
292 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
296 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
297 rblocks, force, &tp);
301 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, nr_exts);
303 error = xfs_iext_count_upgrade(tp, ip, nr_exts);
305 goto out_trans_cancel;
308 * From this point onwards we overwrite the imap pointer that the
312 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
315 goto out_trans_cancel;
318 * Complete the transaction
320 error = xfs_trans_commit(tp);
325 * Copy any maps to caller's array and return any error.
332 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock))) {
333 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
334 error = xfs_alert_fsblock_zero(ip, imap);
338 *seq = xfs_iomap_inode_sequence(ip, 0);
339 xfs_iunlock(ip, XFS_ILOCK_EXCL);
343 xfs_trans_cancel(tp);
348 xfs_quota_need_throttle(
349 struct xfs_inode *ip,
351 xfs_fsblock_t alloc_blocks)
353 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
355 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
358 /* no hi watermark, no throttle */
359 if (!dq->q_prealloc_hi_wmark)
362 /* under the lo watermark, no throttle */
363 if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
370 xfs_quota_calc_throttle(
371 struct xfs_inode *ip,
373 xfs_fsblock_t *qblocks,
377 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
381 /* no dq, or over hi wmark, squash the prealloc completely */
382 if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
388 freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
389 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
391 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
393 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
397 if (freesp < *qfreesp)
400 /* only overwrite the throttle values if we are more aggressive */
401 if ((freesp >> shift) < (*qblocks >> *qshift)) {
408 * If we don't have a user specified preallocation size, dynamically increase
409 * the preallocation size as the size of the file grows. Cap the maximum size
410 * at a single extent or less if the filesystem is near full. The closer the
411 * filesystem is to being full, the smaller the maximum preallocation.
414 xfs_iomap_prealloc_size(
415 struct xfs_inode *ip,
419 struct xfs_iext_cursor *icur)
421 struct xfs_iext_cursor ncur = *icur;
422 struct xfs_bmbt_irec prev, got;
423 struct xfs_mount *mp = ip->i_mount;
424 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
425 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
427 xfs_fsblock_t qblocks;
428 xfs_fsblock_t alloc_blocks = 0;
434 * As an exception we don't do any preallocation at all if the file is
435 * smaller than the minimum preallocation and we are using the default
436 * dynamic preallocation scheme, as it is likely this is the only write
437 * to the file that is going to be done.
439 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
443 * Use the minimum preallocation size for small files or if we are
444 * writing right after a hole.
446 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
447 !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
448 prev.br_startoff + prev.br_blockcount < offset_fsb)
449 return mp->m_allocsize_blocks;
452 * Take the size of the preceding data extents as the basis for the
453 * preallocation size. Note that we don't care if the previous extents
454 * are written or not.
456 plen = prev.br_blockcount;
457 while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
458 if (plen > XFS_MAX_BMBT_EXTLEN / 2 ||
459 isnullstartblock(got.br_startblock) ||
460 got.br_startoff + got.br_blockcount != prev.br_startoff ||
461 got.br_startblock + got.br_blockcount != prev.br_startblock)
463 plen += got.br_blockcount;
468 * If the size of the extents is greater than half the maximum extent
469 * length, then use the current offset as the basis. This ensures that
470 * for large files the preallocation size always extends to
471 * XFS_BMBT_MAX_EXTLEN rather than falling short due to things like stripe
472 * unit/width alignment of real extents.
474 alloc_blocks = plen * 2;
475 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
476 alloc_blocks = XFS_B_TO_FSB(mp, offset);
477 qblocks = alloc_blocks;
480 * XFS_BMBT_MAX_EXTLEN is not a power of two value but we round the prealloc
481 * down to the nearest power of two value after throttling. To prevent
482 * the round down from unconditionally reducing the maximum supported
483 * prealloc size, we round up first, apply appropriate throttling, round
484 * down and cap the value to XFS_BMBT_MAX_EXTLEN.
486 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(XFS_MAX_BMBT_EXTLEN),
489 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
490 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
492 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
494 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
496 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
498 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
503 * Check each quota to cap the prealloc size, provide a shift value to
504 * throttle with and adjust amount of available space.
506 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
507 xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
509 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
510 xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
512 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
513 xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
517 * The final prealloc size is set to the minimum of free space available
518 * in each of the quotas and the overall filesystem.
520 * The shift throttle value is set to the maximum value as determined by
521 * the global low free space values and per-quota low free space values.
523 alloc_blocks = min(alloc_blocks, qblocks);
524 shift = max(shift, qshift);
527 alloc_blocks >>= shift;
529 * rounddown_pow_of_two() returns an undefined result if we pass in
533 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
534 if (alloc_blocks > XFS_MAX_BMBT_EXTLEN)
535 alloc_blocks = XFS_MAX_BMBT_EXTLEN;
538 * If we are still trying to allocate more space than is
539 * available, squash the prealloc hard. This can happen if we
540 * have a large file on a small filesystem and the above
541 * lowspace thresholds are smaller than XFS_BMBT_MAX_EXTLEN.
543 while (alloc_blocks && alloc_blocks >= freesp)
545 if (alloc_blocks < mp->m_allocsize_blocks)
546 alloc_blocks = mp->m_allocsize_blocks;
547 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
548 mp->m_allocsize_blocks);
553 xfs_iomap_write_unwritten(
559 xfs_mount_t *mp = ip->i_mount;
560 xfs_fileoff_t offset_fsb;
561 xfs_filblks_t count_fsb;
562 xfs_filblks_t numblks_fsb;
565 xfs_bmbt_irec_t imap;
566 struct inode *inode = VFS_I(ip);
571 trace_xfs_unwritten_convert(ip, offset, count);
573 offset_fsb = XFS_B_TO_FSBT(mp, offset);
574 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
575 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
578 * Reserve enough blocks in this transaction for two complete extent
579 * btree splits. We may be converting the middle part of an unwritten
580 * extent and in this case we will insert two new extents in the btree
581 * each of which could cause a full split.
583 * This reservation amount will be used in the first call to
584 * xfs_bmbt_split() to select an AG with enough space to satisfy the
585 * rest of the operation.
587 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
589 /* Attach dquots so that bmbt splits are accounted correctly. */
590 error = xfs_qm_dqattach(ip);
596 * Set up a transaction to convert the range of extents
597 * from unwritten to real. Do allocations in a loop until
598 * we have covered the range passed in.
600 * Note that we can't risk to recursing back into the filesystem
601 * here as we might be asked to write out the same inode that we
602 * complete here and might deadlock on the iolock.
604 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
609 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
610 XFS_IEXT_WRITE_UNWRITTEN_CNT);
612 error = xfs_iext_count_upgrade(tp, ip,
613 XFS_IEXT_WRITE_UNWRITTEN_CNT);
615 goto error_on_bmapi_transaction;
618 * Modify the unwritten extent state of the buffer.
621 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
622 XFS_BMAPI_CONVERT, resblks, &imap,
625 goto error_on_bmapi_transaction;
628 * Log the updated inode size as we go. We have to be careful
629 * to only log it up to the actual write offset if it is
630 * halfway into a block.
632 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
633 if (i_size > offset + count)
634 i_size = offset + count;
635 if (update_isize && i_size > i_size_read(inode))
636 i_size_write(inode, i_size);
637 i_size = xfs_new_eof(ip, i_size);
639 ip->i_disk_size = i_size;
640 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
643 error = xfs_trans_commit(tp);
644 xfs_iunlock(ip, XFS_ILOCK_EXCL);
648 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock))) {
649 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
650 return xfs_alert_fsblock_zero(ip, &imap);
653 if ((numblks_fsb = imap.br_blockcount) == 0) {
655 * The numblks_fsb value should always get
656 * smaller, otherwise the loop is stuck.
658 ASSERT(imap.br_blockcount);
661 offset_fsb += numblks_fsb;
662 count_fsb -= numblks_fsb;
663 } while (count_fsb > 0);
667 error_on_bmapi_transaction:
668 xfs_trans_cancel(tp);
669 xfs_iunlock(ip, XFS_ILOCK_EXCL);
677 struct xfs_bmbt_irec *imap,
680 /* don't allocate blocks when just zeroing */
681 if (flags & IOMAP_ZERO)
684 imap->br_startblock == HOLESTARTBLOCK ||
685 imap->br_startblock == DELAYSTARTBLOCK)
687 /* we convert unwritten extents before copying the data for DAX */
688 if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
695 struct xfs_inode *ip,
697 struct xfs_bmbt_irec *imap,
700 if (!xfs_is_cow_inode(ip))
703 /* when zeroing we don't have to COW holes or unwritten extents */
704 if (flags & IOMAP_ZERO) {
706 imap->br_startblock == HOLESTARTBLOCK ||
707 imap->br_state == XFS_EXT_UNWRITTEN)
716 struct xfs_inode *ip,
720 unsigned int mode = *lockmode;
721 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
724 * COW writes may allocate delalloc space or convert unwritten COW
725 * extents, so we need to make sure to take the lock exclusively here.
727 if (xfs_is_cow_inode(ip) && is_write)
728 mode = XFS_ILOCK_EXCL;
731 * Extents not yet cached requires exclusive access, don't block. This
732 * is an opencoded xfs_ilock_data_map_shared() call but with
733 * non-blocking behaviour.
735 if (xfs_need_iread_extents(&ip->i_df)) {
736 if (flags & IOMAP_NOWAIT)
738 mode = XFS_ILOCK_EXCL;
742 if (flags & IOMAP_NOWAIT) {
743 if (!xfs_ilock_nowait(ip, mode))
750 * The reflink iflag could have changed since the earlier unlocked
751 * check, so if we got ILOCK_SHARED for a write and but we're now a
752 * reflink inode we have to switch to ILOCK_EXCL and relock.
754 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
755 xfs_iunlock(ip, mode);
756 mode = XFS_ILOCK_EXCL;
765 * Check that the imap we are going to return to the caller spans the entire
766 * range that the caller requested for the IO.
770 struct xfs_bmbt_irec *imap,
771 xfs_fileoff_t offset_fsb,
772 xfs_fileoff_t end_fsb)
774 if (imap->br_startoff > offset_fsb)
776 if (imap->br_startoff + imap->br_blockcount < end_fsb)
782 xfs_direct_write_iomap_begin(
788 struct iomap *srcmap)
790 struct xfs_inode *ip = XFS_I(inode);
791 struct xfs_mount *mp = ip->i_mount;
792 struct xfs_bmbt_irec imap, cmap;
793 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
794 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
795 int nimaps = 1, error = 0;
798 unsigned int lockmode = XFS_ILOCK_SHARED;
801 ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
803 if (xfs_is_shutdown(mp))
807 * Writes that span EOF might trigger an IO size update on completion,
808 * so consider them to be dirty for the purposes of O_DSYNC even if
809 * there is no other metadata changes pending or have been made here.
811 if (offset + length > i_size_read(inode))
812 iomap_flags |= IOMAP_F_DIRTY;
814 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
818 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
823 if (imap_needs_cow(ip, flags, &imap, nimaps)) {
825 if (flags & IOMAP_NOWAIT)
828 /* may drop and re-acquire the ilock */
829 error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
831 (flags & IOMAP_DIRECT) || IS_DAX(inode));
836 end_fsb = imap.br_startoff + imap.br_blockcount;
837 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
840 if (imap_needs_alloc(inode, flags, &imap, nimaps))
841 goto allocate_blocks;
844 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
845 * a single map so that we avoid partial IO failures due to the rest of
846 * the I/O range not covered by this map triggering an EAGAIN condition
847 * when it is subsequently mapped and aborting the I/O.
849 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
851 if (!imap_spans_range(&imap, offset_fsb, end_fsb))
856 * For overwrite only I/O, we cannot convert unwritten extents without
857 * requiring sub-block zeroing. This can only be done under an
858 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
859 * extent to tell the caller to try again.
861 if (flags & IOMAP_OVERWRITE_ONLY) {
863 if (imap.br_state != XFS_EXT_NORM &&
864 ((offset | length) & mp->m_blockmask))
868 seq = xfs_iomap_inode_sequence(ip, iomap_flags);
869 xfs_iunlock(ip, lockmode);
870 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
871 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags, seq);
875 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
879 * We cap the maximum length we map to a sane size to keep the chunks
880 * of work done where somewhat symmetric with the work writeback does.
881 * This is a completely arbitrary number pulled out of thin air as a
882 * best guess for initial testing.
884 * Note that the values needs to be less than 32-bits wide until the
885 * lower level functions are updated.
887 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
888 end_fsb = xfs_iomap_end_fsb(mp, offset, length);
890 if (offset + length > XFS_ISIZE(ip))
891 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
892 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
893 end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
894 xfs_iunlock(ip, lockmode);
896 error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
901 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
902 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
903 iomap_flags | IOMAP_F_NEW, seq);
906 length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
907 trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
908 if (imap.br_startblock != HOLESTARTBLOCK) {
909 seq = xfs_iomap_inode_sequence(ip, 0);
910 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
914 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
915 xfs_iunlock(ip, lockmode);
916 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED, seq);
920 xfs_iunlock(ip, lockmode);
924 const struct iomap_ops xfs_direct_write_iomap_ops = {
925 .iomap_begin = xfs_direct_write_iomap_begin,
929 xfs_dax_write_iomap_end(
937 struct xfs_inode *ip = XFS_I(inode);
939 if (!xfs_is_cow_inode(ip))
943 xfs_reflink_cancel_cow_range(ip, pos, length, true);
947 return xfs_reflink_end_cow(ip, pos, written);
950 const struct iomap_ops xfs_dax_write_iomap_ops = {
951 .iomap_begin = xfs_direct_write_iomap_begin,
952 .iomap_end = xfs_dax_write_iomap_end,
956 xfs_buffered_write_iomap_begin(
962 struct iomap *srcmap)
964 struct xfs_inode *ip = XFS_I(inode);
965 struct xfs_mount *mp = ip->i_mount;
966 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
967 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, count);
968 struct xfs_bmbt_irec imap, cmap;
969 struct xfs_iext_cursor icur, ccur;
970 xfs_fsblock_t prealloc_blocks = 0;
971 bool eof = false, cow_eof = false, shared = false;
972 int allocfork = XFS_DATA_FORK;
974 unsigned int lockmode = XFS_ILOCK_EXCL;
977 if (xfs_is_shutdown(mp))
980 /* we can't use delayed allocations when using extent size hints */
981 if (xfs_get_extsz_hint(ip))
982 return xfs_direct_write_iomap_begin(inode, offset, count,
983 flags, iomap, srcmap);
985 ASSERT(!XFS_IS_REALTIME_INODE(ip));
987 error = xfs_qm_dqattach(ip);
991 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
995 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
996 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
997 xfs_bmap_mark_sick(ip, XFS_DATA_FORK);
998 error = -EFSCORRUPTED;
1002 XFS_STATS_INC(mp, xs_blk_mapw);
1004 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1009 * Search the data fork first to look up our source mapping. We
1010 * always need the data fork map, as we have to return it to the
1011 * iomap code so that the higher level write code can read data in to
1012 * perform read-modify-write cycles for unaligned writes.
1014 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
1016 imap.br_startoff = end_fsb; /* fake hole until the end */
1018 /* We never need to allocate blocks for zeroing or unsharing a hole. */
1019 if ((flags & (IOMAP_UNSHARE | IOMAP_ZERO)) &&
1020 imap.br_startoff > offset_fsb) {
1021 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
1026 * Search the COW fork extent list even if we did not find a data fork
1027 * extent. This serves two purposes: first this implements the
1028 * speculative preallocation using cowextsize, so that we also unshare
1029 * block adjacent to shared blocks instead of just the shared blocks
1030 * themselves. Second the lookup in the extent list is generally faster
1031 * than going out to the shared extent tree.
1033 if (xfs_is_cow_inode(ip)) {
1035 ASSERT(!xfs_is_reflink_inode(ip));
1036 xfs_ifork_init_cow(ip);
1038 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
1040 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
1041 trace_xfs_reflink_cow_found(ip, &cmap);
1046 if (imap.br_startoff <= offset_fsb) {
1048 * For reflink files we may need a delalloc reservation when
1049 * overwriting shared extents. This includes zeroing of
1050 * existing extents that contain data.
1052 if (!xfs_is_cow_inode(ip) ||
1053 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
1054 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1059 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1061 /* Trim the mapping to the nearest shared extent boundary. */
1062 error = xfs_bmap_trim_cow(ip, &imap, &shared);
1066 /* Not shared? Just report the (potentially capped) extent. */
1068 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
1074 * Fork all the shared blocks from our write offset until the
1075 * end of the extent.
1077 allocfork = XFS_COW_FORK;
1078 end_fsb = imap.br_startoff + imap.br_blockcount;
1081 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1082 * pages to keep the chunks of work done where somewhat
1083 * symmetric with the work writeback does. This is a completely
1084 * arbitrary number pulled out of thin air.
1086 * Note that the values needs to be less than 32-bits wide until
1087 * the lower level functions are updated.
1089 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
1090 end_fsb = xfs_iomap_end_fsb(mp, offset, count);
1092 if (xfs_is_always_cow_inode(ip))
1093 allocfork = XFS_COW_FORK;
1096 if (eof && offset + count > XFS_ISIZE(ip)) {
1098 * Determine the initial size of the preallocation.
1099 * We clean up any extra preallocation when the file is closed.
1101 if (xfs_has_allocsize(mp))
1102 prealloc_blocks = mp->m_allocsize_blocks;
1103 else if (allocfork == XFS_DATA_FORK)
1104 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1105 offset, count, &icur);
1107 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1108 offset, count, &ccur);
1109 if (prealloc_blocks) {
1111 xfs_off_t end_offset;
1112 xfs_fileoff_t p_end_fsb;
1114 end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1115 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1118 align = xfs_eof_alignment(ip);
1120 p_end_fsb = roundup_64(p_end_fsb, align);
1122 p_end_fsb = min(p_end_fsb,
1123 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1124 ASSERT(p_end_fsb > offset_fsb);
1125 prealloc_blocks = p_end_fsb - end_fsb;
1130 error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1131 end_fsb - offset_fsb, prealloc_blocks,
1132 allocfork == XFS_DATA_FORK ? &imap : &cmap,
1133 allocfork == XFS_DATA_FORK ? &icur : &ccur,
1134 allocfork == XFS_DATA_FORK ? eof : cow_eof);
1140 /* retry without any preallocation */
1141 trace_xfs_delalloc_enospc(ip, offset, count);
1142 if (prealloc_blocks) {
1143 prealloc_blocks = 0;
1151 if (allocfork == XFS_COW_FORK) {
1152 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1157 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1158 * them out if the write happens to fail.
1160 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_NEW);
1161 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1162 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1163 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_NEW, seq);
1166 seq = xfs_iomap_inode_sequence(ip, 0);
1167 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1168 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1171 seq = xfs_iomap_inode_sequence(ip, 0);
1172 if (imap.br_startoff <= offset_fsb) {
1173 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0, seq);
1176 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1177 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1178 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1179 IOMAP_F_SHARED, seq);
1182 xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1183 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1184 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, 0, seq);
1187 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1192 xfs_buffered_write_delalloc_punch(
1193 struct inode *inode,
1197 return xfs_bmap_punch_delalloc_range(XFS_I(inode), offset,
1202 xfs_buffered_write_iomap_end(
1203 struct inode *inode,
1208 struct iomap *iomap)
1211 struct xfs_mount *mp = XFS_M(inode->i_sb);
1214 error = iomap_file_buffered_write_punch_delalloc(inode, iomap, offset,
1215 length, written, &xfs_buffered_write_delalloc_punch);
1216 if (error && !xfs_is_shutdown(mp)) {
1217 xfs_alert(mp, "%s: unable to clean up ino 0x%llx",
1218 __func__, XFS_I(inode)->i_ino);
1224 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1225 .iomap_begin = xfs_buffered_write_iomap_begin,
1226 .iomap_end = xfs_buffered_write_iomap_end,
1230 * iomap_page_mkwrite() will never fail in a way that requires delalloc extents
1231 * that it allocated to be revoked. Hence we do not need an .iomap_end method
1232 * for this operation.
1234 const struct iomap_ops xfs_page_mkwrite_iomap_ops = {
1235 .iomap_begin = xfs_buffered_write_iomap_begin,
1239 xfs_read_iomap_begin(
1240 struct inode *inode,
1244 struct iomap *iomap,
1245 struct iomap *srcmap)
1247 struct xfs_inode *ip = XFS_I(inode);
1248 struct xfs_mount *mp = ip->i_mount;
1249 struct xfs_bmbt_irec imap;
1250 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1251 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1252 int nimaps = 1, error = 0;
1253 bool shared = false;
1254 unsigned int lockmode = XFS_ILOCK_SHARED;
1257 ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1259 if (xfs_is_shutdown(mp))
1262 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1265 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1267 if (!error && ((flags & IOMAP_REPORT) || IS_DAX(inode)))
1268 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1269 seq = xfs_iomap_inode_sequence(ip, shared ? IOMAP_F_SHARED : 0);
1270 xfs_iunlock(ip, lockmode);
1274 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1275 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1276 shared ? IOMAP_F_SHARED : 0, seq);
1279 const struct iomap_ops xfs_read_iomap_ops = {
1280 .iomap_begin = xfs_read_iomap_begin,
1284 xfs_seek_iomap_begin(
1285 struct inode *inode,
1289 struct iomap *iomap,
1290 struct iomap *srcmap)
1292 struct xfs_inode *ip = XFS_I(inode);
1293 struct xfs_mount *mp = ip->i_mount;
1294 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1295 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1296 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1297 struct xfs_iext_cursor icur;
1298 struct xfs_bmbt_irec imap, cmap;
1303 if (xfs_is_shutdown(mp))
1306 lockmode = xfs_ilock_data_map_shared(ip);
1307 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1311 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1313 * If we found a data extent we are done.
1315 if (imap.br_startoff <= offset_fsb)
1317 data_fsb = imap.br_startoff;
1320 * Fake a hole until the end of the file.
1322 data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1326 * If a COW fork extent covers the hole, report it - capped to the next
1329 if (xfs_inode_has_cow_data(ip) &&
1330 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1331 cow_fsb = cmap.br_startoff;
1332 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1333 if (data_fsb < cow_fsb + cmap.br_blockcount)
1334 end_fsb = min(end_fsb, data_fsb);
1335 xfs_trim_extent(&cmap, offset_fsb, end_fsb - offset_fsb);
1336 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_SHARED);
1337 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1338 IOMAP_F_SHARED, seq);
1340 * This is a COW extent, so we must probe the page cache
1341 * because there could be dirty page cache being backed
1344 iomap->type = IOMAP_UNWRITTEN;
1349 * Else report a hole, capped to the next found data or COW extent.
1351 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1352 imap.br_blockcount = cow_fsb - offset_fsb;
1354 imap.br_blockcount = data_fsb - offset_fsb;
1355 imap.br_startoff = offset_fsb;
1356 imap.br_startblock = HOLESTARTBLOCK;
1357 imap.br_state = XFS_EXT_NORM;
1359 seq = xfs_iomap_inode_sequence(ip, 0);
1360 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
1361 error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0, seq);
1363 xfs_iunlock(ip, lockmode);
1367 const struct iomap_ops xfs_seek_iomap_ops = {
1368 .iomap_begin = xfs_seek_iomap_begin,
1372 xfs_xattr_iomap_begin(
1373 struct inode *inode,
1377 struct iomap *iomap,
1378 struct iomap *srcmap)
1380 struct xfs_inode *ip = XFS_I(inode);
1381 struct xfs_mount *mp = ip->i_mount;
1382 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1383 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1384 struct xfs_bmbt_irec imap;
1385 int nimaps = 1, error = 0;
1389 if (xfs_is_shutdown(mp))
1392 lockmode = xfs_ilock_attr_map_shared(ip);
1394 /* if there are no attribute fork or extents, return ENOENT */
1395 if (!xfs_inode_has_attr_fork(ip) || !ip->i_af.if_nextents) {
1400 ASSERT(ip->i_af.if_format != XFS_DINODE_FMT_LOCAL);
1401 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1402 &nimaps, XFS_BMAPI_ATTRFORK);
1405 seq = xfs_iomap_inode_sequence(ip, IOMAP_F_XATTR);
1406 xfs_iunlock(ip, lockmode);
1411 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_XATTR, seq);
1414 const struct iomap_ops xfs_xattr_iomap_ops = {
1415 .iomap_begin = xfs_xattr_iomap_begin,
1420 struct xfs_inode *ip,
1425 struct inode *inode = VFS_I(ip);
1428 return dax_zero_range(inode, pos, len, did_zero,
1429 &xfs_dax_write_iomap_ops);
1430 return iomap_zero_range(inode, pos, len, did_zero,
1431 &xfs_buffered_write_iomap_ops);
1436 struct xfs_inode *ip,
1440 struct inode *inode = VFS_I(ip);
1443 return dax_truncate_page(inode, pos, did_zero,
1444 &xfs_dax_write_iomap_ops);
1445 return iomap_truncate_page(inode, pos, did_zero,
1446 &xfs_buffered_write_iomap_ops);