1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap_btree.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_space.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_iomap.h"
25 #include "xfs_trace.h"
26 #include "xfs_quota.h"
27 #include "xfs_dquot_item.h"
28 #include "xfs_dquot.h"
29 #include "xfs_reflink.h"
32 #define XFS_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,
56 struct xfs_bmbt_irec *imap,
59 struct xfs_mount *mp = ip->i_mount;
60 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
62 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
63 return xfs_alert_fsblock_zero(ip, imap);
65 if (imap->br_startblock == HOLESTARTBLOCK) {
66 iomap->addr = IOMAP_NULL_ADDR;
67 iomap->type = IOMAP_HOLE;
68 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
69 isnullstartblock(imap->br_startblock)) {
70 iomap->addr = IOMAP_NULL_ADDR;
71 iomap->type = IOMAP_DELALLOC;
73 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
74 if (imap->br_state == XFS_EXT_UNWRITTEN)
75 iomap->type = IOMAP_UNWRITTEN;
77 iomap->type = IOMAP_MAPPED;
79 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
80 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
81 iomap->bdev = target->bt_bdev;
82 iomap->dax_dev = target->bt_daxdev;
85 if (xfs_ipincount(ip) &&
86 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
87 iomap->flags |= IOMAP_F_DIRTY;
95 xfs_fileoff_t offset_fsb,
96 xfs_fileoff_t end_fsb)
98 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
100 iomap->addr = IOMAP_NULL_ADDR;
101 iomap->type = IOMAP_HOLE;
102 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
103 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
104 iomap->bdev = target->bt_bdev;
105 iomap->dax_dev = target->bt_daxdev;
108 static inline xfs_fileoff_t
110 struct xfs_mount *mp,
114 ASSERT(offset <= mp->m_super->s_maxbytes);
115 return min(XFS_B_TO_FSB(mp, offset + count),
116 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
121 struct xfs_inode *ip)
123 struct xfs_mount *mp = ip->i_mount;
124 xfs_extlen_t align = 0;
126 if (!XFS_IS_REALTIME_INODE(ip)) {
128 * Round up the allocation request to a stripe unit
129 * (m_dalign) boundary if the file size is >= stripe unit
130 * size, and we are allocating past the allocation eof.
132 * If mounted with the "-o swalloc" option the alignment is
133 * increased from the strip unit size to the stripe width.
135 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
136 align = mp->m_swidth;
137 else if (mp->m_dalign)
138 align = mp->m_dalign;
140 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
148 * Check if last_fsb is outside the last extent, and if so grow it to the next
149 * stripe unit boundary.
152 xfs_iomap_eof_align_last_fsb(
153 struct xfs_inode *ip,
154 xfs_fileoff_t end_fsb)
156 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
157 xfs_extlen_t extsz = xfs_get_extsz_hint(ip);
158 xfs_extlen_t align = xfs_eof_alignment(ip);
159 struct xfs_bmbt_irec irec;
160 struct xfs_iext_cursor icur;
162 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
165 * Always round up the allocation request to the extent hint boundary.
169 align = roundup_64(align, extsz);
175 xfs_fileoff_t aligned_end_fsb = roundup_64(end_fsb, align);
177 xfs_iext_last(ifp, &icur);
178 if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
179 aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
180 return aligned_end_fsb;
187 xfs_iomap_write_direct(
188 struct xfs_inode *ip,
189 xfs_fileoff_t offset_fsb,
190 xfs_fileoff_t count_fsb,
191 struct xfs_bmbt_irec *imap)
193 struct xfs_mount *mp = ip->i_mount;
194 struct xfs_trans *tp;
195 xfs_filblks_t resaligned;
197 unsigned int dblocks, rblocks;
200 int bmapi_flags = XFS_BMAPI_PREALLOC;
202 ASSERT(count_fsb > 0);
204 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
205 xfs_get_extsz_hint(ip));
206 if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
207 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
208 rblocks = resaligned;
210 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
214 error = xfs_qm_dqattach(ip);
219 * For DAX, we do not allocate unwritten extents, but instead we zero
220 * the block before we commit the transaction. Ideally we'd like to do
221 * this outside the transaction context, but if we commit and then crash
222 * we may not have zeroed the blocks and this will be exposed on
223 * recovery of the allocation. Hence we must zero before commit.
225 * Further, if we are mapping unwritten extents here, we need to zero
226 * and convert them to written so that we don't need an unwritten extent
227 * callback for DAX. This also means that we need to be able to dip into
228 * the reserve block pool for bmbt block allocation if there is no space
229 * left but we need to do unwritten extent conversion.
231 if (IS_DAX(VFS_I(ip))) {
232 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
233 if (imap->br_state == XFS_EXT_UNWRITTEN) {
235 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
239 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
240 rblocks, force, &tp);
244 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
245 XFS_IEXT_ADD_NOSPLIT_CNT);
247 goto out_trans_cancel;
250 * From this point onwards we overwrite the imap pointer that the
254 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
257 goto out_trans_cancel;
260 * Complete the transaction
262 error = xfs_trans_commit(tp);
267 * Copy any maps to caller's array and return any error.
274 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
275 error = xfs_alert_fsblock_zero(ip, imap);
278 xfs_iunlock(ip, XFS_ILOCK_EXCL);
282 xfs_trans_cancel(tp);
287 xfs_quota_need_throttle(
288 struct xfs_inode *ip,
290 xfs_fsblock_t alloc_blocks)
292 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
294 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
297 /* no hi watermark, no throttle */
298 if (!dq->q_prealloc_hi_wmark)
301 /* under the lo watermark, no throttle */
302 if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
309 xfs_quota_calc_throttle(
310 struct xfs_inode *ip,
312 xfs_fsblock_t *qblocks,
316 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
320 /* no dq, or over hi wmark, squash the prealloc completely */
321 if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
327 freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
328 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
330 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
332 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
336 if (freesp < *qfreesp)
339 /* only overwrite the throttle values if we are more aggressive */
340 if ((freesp >> shift) < (*qblocks >> *qshift)) {
347 * If we don't have a user specified preallocation size, dynamically increase
348 * the preallocation size as the size of the file grows. Cap the maximum size
349 * at a single extent or less if the filesystem is near full. The closer the
350 * filesystem is to being full, the smaller the maximum preallocation.
353 xfs_iomap_prealloc_size(
354 struct xfs_inode *ip,
358 struct xfs_iext_cursor *icur)
360 struct xfs_iext_cursor ncur = *icur;
361 struct xfs_bmbt_irec prev, got;
362 struct xfs_mount *mp = ip->i_mount;
363 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
364 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
366 xfs_fsblock_t qblocks;
367 xfs_fsblock_t alloc_blocks = 0;
373 * As an exception we don't do any preallocation at all if the file is
374 * smaller than the minimum preallocation and we are using the default
375 * dynamic preallocation scheme, as it is likely this is the only write
376 * to the file that is going to be done.
378 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
382 * Use the minimum preallocation size for small files or if we are
383 * writing right after a hole.
385 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
386 !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
387 prev.br_startoff + prev.br_blockcount < offset_fsb)
388 return mp->m_allocsize_blocks;
391 * Take the size of the preceding data extents as the basis for the
392 * preallocation size. Note that we don't care if the previous extents
393 * are written or not.
395 plen = prev.br_blockcount;
396 while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
397 if (plen > MAXEXTLEN / 2 ||
398 isnullstartblock(got.br_startblock) ||
399 got.br_startoff + got.br_blockcount != prev.br_startoff ||
400 got.br_startblock + got.br_blockcount != prev.br_startblock)
402 plen += got.br_blockcount;
407 * If the size of the extents is greater than half the maximum extent
408 * length, then use the current offset as the basis. This ensures that
409 * for large files the preallocation size always extends to MAXEXTLEN
410 * rather than falling short due to things like stripe unit/width
411 * alignment of real extents.
413 alloc_blocks = plen * 2;
414 if (alloc_blocks > MAXEXTLEN)
415 alloc_blocks = XFS_B_TO_FSB(mp, offset);
416 qblocks = alloc_blocks;
419 * MAXEXTLEN is not a power of two value but we round the prealloc down
420 * to the nearest power of two value after throttling. To prevent the
421 * round down from unconditionally reducing the maximum supported
422 * prealloc size, we round up first, apply appropriate throttling,
423 * round down and cap the value to MAXEXTLEN.
425 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
428 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
429 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
431 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
433 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
435 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
437 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
442 * Check each quota to cap the prealloc size, provide a shift value to
443 * throttle with and adjust amount of available space.
445 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
446 xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
448 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
449 xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
451 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
452 xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
456 * The final prealloc size is set to the minimum of free space available
457 * in each of the quotas and the overall filesystem.
459 * The shift throttle value is set to the maximum value as determined by
460 * the global low free space values and per-quota low free space values.
462 alloc_blocks = min(alloc_blocks, qblocks);
463 shift = max(shift, qshift);
466 alloc_blocks >>= shift;
468 * rounddown_pow_of_two() returns an undefined result if we pass in
472 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
473 if (alloc_blocks > MAXEXTLEN)
474 alloc_blocks = MAXEXTLEN;
477 * If we are still trying to allocate more space than is
478 * available, squash the prealloc hard. This can happen if we
479 * have a large file on a small filesystem and the above
480 * lowspace thresholds are smaller than MAXEXTLEN.
482 while (alloc_blocks && alloc_blocks >= freesp)
484 if (alloc_blocks < mp->m_allocsize_blocks)
485 alloc_blocks = mp->m_allocsize_blocks;
486 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
487 mp->m_allocsize_blocks);
492 xfs_iomap_write_unwritten(
498 xfs_mount_t *mp = ip->i_mount;
499 xfs_fileoff_t offset_fsb;
500 xfs_filblks_t count_fsb;
501 xfs_filblks_t numblks_fsb;
504 xfs_bmbt_irec_t imap;
505 struct inode *inode = VFS_I(ip);
510 trace_xfs_unwritten_convert(ip, offset, count);
512 offset_fsb = XFS_B_TO_FSBT(mp, offset);
513 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
514 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
517 * Reserve enough blocks in this transaction for two complete extent
518 * btree splits. We may be converting the middle part of an unwritten
519 * extent and in this case we will insert two new extents in the btree
520 * each of which could cause a full split.
522 * This reservation amount will be used in the first call to
523 * xfs_bmbt_split() to select an AG with enough space to satisfy the
524 * rest of the operation.
526 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
528 /* Attach dquots so that bmbt splits are accounted correctly. */
529 error = xfs_qm_dqattach(ip);
535 * Set up a transaction to convert the range of extents
536 * from unwritten to real. Do allocations in a loop until
537 * we have covered the range passed in.
539 * Note that we can't risk to recursing back into the filesystem
540 * here as we might be asked to write out the same inode that we
541 * complete here and might deadlock on the iolock.
543 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
548 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
549 XFS_IEXT_WRITE_UNWRITTEN_CNT);
551 goto error_on_bmapi_transaction;
554 * Modify the unwritten extent state of the buffer.
557 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
558 XFS_BMAPI_CONVERT, resblks, &imap,
561 goto error_on_bmapi_transaction;
564 * Log the updated inode size as we go. We have to be careful
565 * to only log it up to the actual write offset if it is
566 * halfway into a block.
568 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
569 if (i_size > offset + count)
570 i_size = offset + count;
571 if (update_isize && i_size > i_size_read(inode))
572 i_size_write(inode, i_size);
573 i_size = xfs_new_eof(ip, i_size);
575 ip->i_d.di_size = i_size;
576 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
579 error = xfs_trans_commit(tp);
580 xfs_iunlock(ip, XFS_ILOCK_EXCL);
584 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
585 return xfs_alert_fsblock_zero(ip, &imap);
587 if ((numblks_fsb = imap.br_blockcount) == 0) {
589 * The numblks_fsb value should always get
590 * smaller, otherwise the loop is stuck.
592 ASSERT(imap.br_blockcount);
595 offset_fsb += numblks_fsb;
596 count_fsb -= numblks_fsb;
597 } while (count_fsb > 0);
601 error_on_bmapi_transaction:
602 xfs_trans_cancel(tp);
603 xfs_iunlock(ip, XFS_ILOCK_EXCL);
611 struct xfs_bmbt_irec *imap,
614 /* don't allocate blocks when just zeroing */
615 if (flags & IOMAP_ZERO)
618 imap->br_startblock == HOLESTARTBLOCK ||
619 imap->br_startblock == DELAYSTARTBLOCK)
621 /* we convert unwritten extents before copying the data for DAX */
622 if (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN)
629 struct xfs_inode *ip,
631 struct xfs_bmbt_irec *imap,
634 if (!xfs_is_cow_inode(ip))
637 /* when zeroing we don't have to COW holes or unwritten extents */
638 if (flags & IOMAP_ZERO) {
640 imap->br_startblock == HOLESTARTBLOCK ||
641 imap->br_state == XFS_EXT_UNWRITTEN)
650 struct xfs_inode *ip,
654 unsigned mode = XFS_ILOCK_SHARED;
655 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
658 * COW writes may allocate delalloc space or convert unwritten COW
659 * extents, so we need to make sure to take the lock exclusively here.
661 if (xfs_is_cow_inode(ip) && is_write)
662 mode = XFS_ILOCK_EXCL;
665 * Extents not yet cached requires exclusive access, don't block. This
666 * is an opencoded xfs_ilock_data_map_shared() call but with
667 * non-blocking behaviour.
669 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
670 if (flags & IOMAP_NOWAIT)
672 mode = XFS_ILOCK_EXCL;
676 if (flags & IOMAP_NOWAIT) {
677 if (!xfs_ilock_nowait(ip, mode))
684 * The reflink iflag could have changed since the earlier unlocked
685 * check, so if we got ILOCK_SHARED for a write and but we're now a
686 * reflink inode we have to switch to ILOCK_EXCL and relock.
688 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
689 xfs_iunlock(ip, mode);
690 mode = XFS_ILOCK_EXCL;
699 * Check that the imap we are going to return to the caller spans the entire
700 * range that the caller requested for the IO.
704 struct xfs_bmbt_irec *imap,
705 xfs_fileoff_t offset_fsb,
706 xfs_fileoff_t end_fsb)
708 if (imap->br_startoff > offset_fsb)
710 if (imap->br_startoff + imap->br_blockcount < end_fsb)
716 xfs_direct_write_iomap_begin(
722 struct iomap *srcmap)
724 struct xfs_inode *ip = XFS_I(inode);
725 struct xfs_mount *mp = ip->i_mount;
726 struct xfs_bmbt_irec imap, cmap;
727 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
728 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
729 int nimaps = 1, error = 0;
734 ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
736 if (XFS_FORCED_SHUTDOWN(mp))
740 * Writes that span EOF might trigger an IO size update on completion,
741 * so consider them to be dirty for the purposes of O_DSYNC even if
742 * there is no other metadata changes pending or have been made here.
744 if (offset + length > i_size_read(inode))
745 iomap_flags |= IOMAP_F_DIRTY;
747 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
751 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
756 if (imap_needs_cow(ip, flags, &imap, nimaps)) {
758 if (flags & IOMAP_NOWAIT)
761 /* may drop and re-acquire the ilock */
762 error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
763 &lockmode, flags & IOMAP_DIRECT);
768 end_fsb = imap.br_startoff + imap.br_blockcount;
769 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
772 if (imap_needs_alloc(inode, flags, &imap, nimaps))
773 goto allocate_blocks;
776 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
777 * a single map so that we avoid partial IO failures due to the rest of
778 * the I/O range not covered by this map triggering an EAGAIN condition
779 * when it is subsequently mapped and aborting the I/O.
781 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
783 if (!imap_spans_range(&imap, offset_fsb, end_fsb))
788 * For overwrite only I/O, we cannot convert unwritten extents without
789 * requiring sub-block zeroing. This can only be done under an
790 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
791 * extent to tell the caller to try again.
793 if (flags & IOMAP_OVERWRITE_ONLY) {
795 if (imap.br_state != XFS_EXT_NORM &&
796 ((offset | length) & mp->m_blockmask))
800 xfs_iunlock(ip, lockmode);
801 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
802 return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags);
806 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
810 * We cap the maximum length we map to a sane size to keep the chunks
811 * of work done where somewhat symmetric with the work writeback does.
812 * This is a completely arbitrary number pulled out of thin air as a
813 * best guess for initial testing.
815 * Note that the values needs to be less than 32-bits wide until the
816 * lower level functions are updated.
818 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
819 end_fsb = xfs_iomap_end_fsb(mp, offset, length);
821 if (offset + length > XFS_ISIZE(ip))
822 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
823 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
824 end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
825 xfs_iunlock(ip, lockmode);
827 error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
832 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
833 return xfs_bmbt_to_iomap(ip, iomap, &imap, iomap_flags | IOMAP_F_NEW);
836 xfs_iunlock(ip, lockmode);
837 length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
838 trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
839 if (imap.br_startblock != HOLESTARTBLOCK) {
840 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
844 return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
848 xfs_iunlock(ip, lockmode);
852 const struct iomap_ops xfs_direct_write_iomap_ops = {
853 .iomap_begin = xfs_direct_write_iomap_begin,
857 xfs_buffered_write_iomap_begin(
863 struct iomap *srcmap)
865 struct xfs_inode *ip = XFS_I(inode);
866 struct xfs_mount *mp = ip->i_mount;
867 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
868 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, count);
869 struct xfs_bmbt_irec imap, cmap;
870 struct xfs_iext_cursor icur, ccur;
871 xfs_fsblock_t prealloc_blocks = 0;
872 bool eof = false, cow_eof = false, shared = false;
873 int allocfork = XFS_DATA_FORK;
876 if (XFS_FORCED_SHUTDOWN(mp))
879 /* we can't use delayed allocations when using extent size hints */
880 if (xfs_get_extsz_hint(ip))
881 return xfs_direct_write_iomap_begin(inode, offset, count,
882 flags, iomap, srcmap);
884 ASSERT(!XFS_IS_REALTIME_INODE(ip));
886 xfs_ilock(ip, XFS_ILOCK_EXCL);
888 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
889 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
890 error = -EFSCORRUPTED;
894 XFS_STATS_INC(mp, xs_blk_mapw);
896 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
897 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
903 * Search the data fork first to look up our source mapping. We
904 * always need the data fork map, as we have to return it to the
905 * iomap code so that the higher level write code can read data in to
906 * perform read-modify-write cycles for unaligned writes.
908 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
910 imap.br_startoff = end_fsb; /* fake hole until the end */
912 /* We never need to allocate blocks for zeroing a hole. */
913 if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
914 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
919 * Search the COW fork extent list even if we did not find a data fork
920 * extent. This serves two purposes: first this implements the
921 * speculative preallocation using cowextsize, so that we also unshare
922 * block adjacent to shared blocks instead of just the shared blocks
923 * themselves. Second the lookup in the extent list is generally faster
924 * than going out to the shared extent tree.
926 if (xfs_is_cow_inode(ip)) {
928 ASSERT(!xfs_is_reflink_inode(ip));
929 xfs_ifork_init_cow(ip);
931 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
933 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
934 trace_xfs_reflink_cow_found(ip, &cmap);
939 if (imap.br_startoff <= offset_fsb) {
941 * For reflink files we may need a delalloc reservation when
942 * overwriting shared extents. This includes zeroing of
943 * existing extents that contain data.
945 if (!xfs_is_cow_inode(ip) ||
946 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
947 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
952 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
954 /* Trim the mapping to the nearest shared extent boundary. */
955 error = xfs_bmap_trim_cow(ip, &imap, &shared);
959 /* Not shared? Just report the (potentially capped) extent. */
961 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
967 * Fork all the shared blocks from our write offset until the
970 allocfork = XFS_COW_FORK;
971 end_fsb = imap.br_startoff + imap.br_blockcount;
974 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
975 * pages to keep the chunks of work done where somewhat
976 * symmetric with the work writeback does. This is a completely
977 * arbitrary number pulled out of thin air.
979 * Note that the values needs to be less than 32-bits wide until
980 * the lower level functions are updated.
982 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
983 end_fsb = xfs_iomap_end_fsb(mp, offset, count);
985 if (xfs_is_always_cow_inode(ip))
986 allocfork = XFS_COW_FORK;
989 error = xfs_qm_dqattach_locked(ip, false);
993 if (eof && offset + count > XFS_ISIZE(ip)) {
995 * Determine the initial size of the preallocation.
996 * We clean up any extra preallocation when the file is closed.
998 if (mp->m_flags & XFS_MOUNT_ALLOCSIZE)
999 prealloc_blocks = mp->m_allocsize_blocks;
1001 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1002 offset, count, &icur);
1003 if (prealloc_blocks) {
1005 xfs_off_t end_offset;
1006 xfs_fileoff_t p_end_fsb;
1008 end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1009 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1012 align = xfs_eof_alignment(ip);
1014 p_end_fsb = roundup_64(p_end_fsb, align);
1016 p_end_fsb = min(p_end_fsb,
1017 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1018 ASSERT(p_end_fsb > offset_fsb);
1019 prealloc_blocks = p_end_fsb - end_fsb;
1024 error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1025 end_fsb - offset_fsb, prealloc_blocks,
1026 allocfork == XFS_DATA_FORK ? &imap : &cmap,
1027 allocfork == XFS_DATA_FORK ? &icur : &ccur,
1028 allocfork == XFS_DATA_FORK ? eof : cow_eof);
1034 /* retry without any preallocation */
1035 trace_xfs_delalloc_enospc(ip, offset, count);
1036 if (prealloc_blocks) {
1037 prealloc_blocks = 0;
1045 if (allocfork == XFS_COW_FORK) {
1046 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1051 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1052 * them out if the write happens to fail.
1054 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1055 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1056 return xfs_bmbt_to_iomap(ip, iomap, &imap, IOMAP_F_NEW);
1059 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1060 return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1063 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1064 if (imap.br_startoff <= offset_fsb) {
1065 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, 0);
1069 xfs_trim_extent(&cmap, offset_fsb,
1070 imap.br_startoff - offset_fsb);
1072 return xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1075 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1080 xfs_buffered_write_iomap_end(
1081 struct inode *inode,
1086 struct iomap *iomap)
1088 struct xfs_inode *ip = XFS_I(inode);
1089 struct xfs_mount *mp = ip->i_mount;
1090 xfs_fileoff_t start_fsb;
1091 xfs_fileoff_t end_fsb;
1094 if (iomap->type != IOMAP_DELALLOC)
1098 * Behave as if the write failed if drop writes is enabled. Set the NEW
1099 * flag to force delalloc cleanup.
1101 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1102 iomap->flags |= IOMAP_F_NEW;
1107 * start_fsb refers to the first unused block after a short write. If
1108 * nothing was written, round offset down to point at the first block in
1111 if (unlikely(!written))
1112 start_fsb = XFS_B_TO_FSBT(mp, offset);
1114 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1115 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1118 * Trim delalloc blocks if they were allocated by this write and we
1119 * didn't manage to write the whole range.
1121 * We don't need to care about racing delalloc as we hold i_mutex
1122 * across the reserve/allocate/unreserve calls. If there are delalloc
1123 * blocks in the range, they are ours.
1125 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1126 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1127 XFS_FSB_TO_B(mp, end_fsb) - 1);
1129 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1130 end_fsb - start_fsb);
1131 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1132 xfs_alert(mp, "%s: unable to clean up ino %lld",
1133 __func__, ip->i_ino);
1141 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1142 .iomap_begin = xfs_buffered_write_iomap_begin,
1143 .iomap_end = xfs_buffered_write_iomap_end,
1147 xfs_read_iomap_begin(
1148 struct inode *inode,
1152 struct iomap *iomap,
1153 struct iomap *srcmap)
1155 struct xfs_inode *ip = XFS_I(inode);
1156 struct xfs_mount *mp = ip->i_mount;
1157 struct xfs_bmbt_irec imap;
1158 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1159 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1160 int nimaps = 1, error = 0;
1161 bool shared = false;
1164 ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1166 if (XFS_FORCED_SHUTDOWN(mp))
1169 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1172 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1174 if (!error && (flags & IOMAP_REPORT))
1175 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1176 xfs_iunlock(ip, lockmode);
1180 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1181 return xfs_bmbt_to_iomap(ip, iomap, &imap, shared ? IOMAP_F_SHARED : 0);
1184 const struct iomap_ops xfs_read_iomap_ops = {
1185 .iomap_begin = xfs_read_iomap_begin,
1189 xfs_seek_iomap_begin(
1190 struct inode *inode,
1194 struct iomap *iomap,
1195 struct iomap *srcmap)
1197 struct xfs_inode *ip = XFS_I(inode);
1198 struct xfs_mount *mp = ip->i_mount;
1199 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1200 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1201 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1202 struct xfs_iext_cursor icur;
1203 struct xfs_bmbt_irec imap, cmap;
1207 if (XFS_FORCED_SHUTDOWN(mp))
1210 lockmode = xfs_ilock_data_map_shared(ip);
1211 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1212 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1217 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1219 * If we found a data extent we are done.
1221 if (imap.br_startoff <= offset_fsb)
1223 data_fsb = imap.br_startoff;
1226 * Fake a hole until the end of the file.
1228 data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1232 * If a COW fork extent covers the hole, report it - capped to the next
1235 if (xfs_inode_has_cow_data(ip) &&
1236 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1237 cow_fsb = cmap.br_startoff;
1238 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1239 if (data_fsb < cow_fsb + cmap.br_blockcount)
1240 end_fsb = min(end_fsb, data_fsb);
1241 xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1242 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, IOMAP_F_SHARED);
1244 * This is a COW extent, so we must probe the page cache
1245 * because there could be dirty page cache being backed
1248 iomap->type = IOMAP_UNWRITTEN;
1253 * Else report a hole, capped to the next found data or COW extent.
1255 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1256 imap.br_blockcount = cow_fsb - offset_fsb;
1258 imap.br_blockcount = data_fsb - offset_fsb;
1259 imap.br_startoff = offset_fsb;
1260 imap.br_startblock = HOLESTARTBLOCK;
1261 imap.br_state = XFS_EXT_NORM;
1263 xfs_trim_extent(&imap, offset_fsb, end_fsb);
1264 error = xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1266 xfs_iunlock(ip, lockmode);
1270 const struct iomap_ops xfs_seek_iomap_ops = {
1271 .iomap_begin = xfs_seek_iomap_begin,
1275 xfs_xattr_iomap_begin(
1276 struct inode *inode,
1280 struct iomap *iomap,
1281 struct iomap *srcmap)
1283 struct xfs_inode *ip = XFS_I(inode);
1284 struct xfs_mount *mp = ip->i_mount;
1285 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1286 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1287 struct xfs_bmbt_irec imap;
1288 int nimaps = 1, error = 0;
1291 if (XFS_FORCED_SHUTDOWN(mp))
1294 lockmode = xfs_ilock_attr_map_shared(ip);
1296 /* if there are no attribute fork or extents, return ENOENT */
1297 if (!XFS_IFORK_Q(ip) || !ip->i_afp->if_nextents) {
1302 ASSERT(ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL);
1303 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1304 &nimaps, XFS_BMAPI_ATTRFORK);
1306 xfs_iunlock(ip, lockmode);
1311 return xfs_bmbt_to_iomap(ip, iomap, &imap, 0);
1314 const struct iomap_ops xfs_xattr_iomap_ops = {
1315 .iomap_begin = xfs_xattr_iomap_begin,