1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
7 #include <linux/iomap.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_bmap_btree.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_errortag.h"
22 #include "xfs_error.h"
23 #include "xfs_trans.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_inode_item.h"
26 #include "xfs_iomap.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_quota.h"
30 #include "xfs_dquot_item.h"
31 #include "xfs_dquot.h"
32 #include "xfs_reflink.h"
35 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
39 xfs_alert_fsblock_zero(
41 xfs_bmbt_irec_t *imap)
43 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
44 "Access to block zero in inode %llu "
45 "start_block: %llx start_off: %llx "
46 "blkcnt: %llx extent-state: %x",
47 (unsigned long long)ip->i_ino,
48 (unsigned long long)imap->br_startblock,
49 (unsigned long long)imap->br_startoff,
50 (unsigned long long)imap->br_blockcount,
59 struct xfs_bmbt_irec *imap,
62 struct xfs_mount *mp = ip->i_mount;
64 if (unlikely(!imap->br_startblock && !XFS_IS_REALTIME_INODE(ip)))
65 return xfs_alert_fsblock_zero(ip, imap);
67 if (imap->br_startblock == HOLESTARTBLOCK) {
68 iomap->addr = IOMAP_NULL_ADDR;
69 iomap->type = IOMAP_HOLE;
70 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
71 isnullstartblock(imap->br_startblock)) {
72 iomap->addr = IOMAP_NULL_ADDR;
73 iomap->type = IOMAP_DELALLOC;
75 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
76 if (imap->br_state == XFS_EXT_UNWRITTEN)
77 iomap->type = IOMAP_UNWRITTEN;
79 iomap->type = IOMAP_MAPPED;
81 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
82 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
83 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
84 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
86 if (xfs_ipincount(ip) &&
87 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
88 iomap->flags |= IOMAP_F_DIRTY;
90 iomap->flags |= IOMAP_F_SHARED;
98 xfs_fileoff_t offset_fsb,
99 xfs_fileoff_t end_fsb)
101 iomap->addr = IOMAP_NULL_ADDR;
102 iomap->type = IOMAP_HOLE;
103 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
104 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
105 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
106 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
111 struct xfs_inode *ip,
112 xfs_extlen_t extsize)
114 struct xfs_mount *mp = ip->i_mount;
115 xfs_extlen_t align = 0;
117 if (!XFS_IS_REALTIME_INODE(ip)) {
119 * Round up the allocation request to a stripe unit
120 * (m_dalign) boundary if the file size is >= stripe unit
121 * size, and we are allocating past the allocation eof.
123 * If mounted with the "-o swalloc" option the alignment is
124 * increased from the strip unit size to the stripe width.
126 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
127 align = mp->m_swidth;
128 else if (mp->m_dalign)
129 align = mp->m_dalign;
131 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
136 * Always round up the allocation request to an extent boundary
137 * (when file on a real-time subvolume or has di_extsize hint).
141 align = roundup_64(align, extsize);
150 xfs_iomap_eof_align_last_fsb(
151 struct xfs_inode *ip,
152 xfs_extlen_t extsize,
153 xfs_fileoff_t *last_fsb)
155 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
158 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
161 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
165 *last_fsb = new_last_fsb;
171 xfs_iomap_write_direct(
175 xfs_bmbt_irec_t *imap,
178 xfs_mount_t *mp = ip->i_mount;
179 xfs_fileoff_t offset_fsb;
180 xfs_fileoff_t last_fsb;
181 xfs_filblks_t count_fsb, resaligned;
187 uint qblocks, resblks, resrtextents;
190 int bmapi_flags = XFS_BMAPI_PREALLOC;
193 rt = XFS_IS_REALTIME_INODE(ip);
194 extsz = xfs_get_extsz_hint(ip);
195 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
197 ASSERT(xfs_isilocked(ip, lockmode));
199 offset_fsb = XFS_B_TO_FSBT(mp, offset);
200 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
201 if ((offset + count) > XFS_ISIZE(ip)) {
203 * Assert that the in-core extent list is present since this can
204 * call xfs_iread_extents() and we only have the ilock shared.
205 * This should be safe because the lock was held around a bmapi
206 * call in the caller and we only need it to access the in-core
209 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
211 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
215 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
216 last_fsb = min(last_fsb, (xfs_fileoff_t)
217 imap->br_blockcount +
220 count_fsb = last_fsb - offset_fsb;
221 ASSERT(count_fsb > 0);
222 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
225 resrtextents = qblocks = resaligned;
226 resrtextents /= mp->m_sb.sb_rextsize;
227 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
228 quota_flag = XFS_QMOPT_RES_RTBLKS;
231 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
232 quota_flag = XFS_QMOPT_RES_REGBLKS;
236 * Drop the shared lock acquired by the caller, attach the dquot if
237 * necessary and move on to transaction setup.
239 xfs_iunlock(ip, lockmode);
240 error = xfs_qm_dqattach(ip);
245 * For DAX, we do not allocate unwritten extents, but instead we zero
246 * the block before we commit the transaction. Ideally we'd like to do
247 * this outside the transaction context, but if we commit and then crash
248 * we may not have zeroed the blocks and this will be exposed on
249 * recovery of the allocation. Hence we must zero before commit.
251 * Further, if we are mapping unwritten extents here, we need to zero
252 * and convert them to written so that we don't need an unwritten extent
253 * callback for DAX. This also means that we need to be able to dip into
254 * the reserve block pool for bmbt block allocation if there is no space
255 * left but we need to do unwritten extent conversion.
257 if (IS_DAX(VFS_I(ip))) {
258 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
259 if (imap->br_state == XFS_EXT_UNWRITTEN) {
260 tflags |= XFS_TRANS_RESERVE;
261 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
264 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
269 lockmode = XFS_ILOCK_EXCL;
270 xfs_ilock(ip, lockmode);
272 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
274 goto out_trans_cancel;
276 xfs_trans_ijoin(tp, ip, 0);
279 * From this point onwards we overwrite the imap pointer that the
283 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
284 bmapi_flags, resblks, imap, &nimaps);
289 * Complete the transaction
291 error = xfs_trans_commit(tp);
296 * Copy any maps to caller's array and return any error.
303 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
304 error = xfs_alert_fsblock_zero(ip, imap);
307 xfs_iunlock(ip, lockmode);
311 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
313 xfs_trans_cancel(tp);
318 xfs_quota_need_throttle(
319 struct xfs_inode *ip,
321 xfs_fsblock_t alloc_blocks)
323 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
325 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
328 /* no hi watermark, no throttle */
329 if (!dq->q_prealloc_hi_wmark)
332 /* under the lo watermark, no throttle */
333 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
340 xfs_quota_calc_throttle(
341 struct xfs_inode *ip,
343 xfs_fsblock_t *qblocks,
349 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
351 /* no dq, or over hi wmark, squash the prealloc completely */
352 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
358 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
359 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
361 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
363 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
367 if (freesp < *qfreesp)
370 /* only overwrite the throttle values if we are more aggressive */
371 if ((freesp >> shift) < (*qblocks >> *qshift)) {
378 * If we are doing a write at the end of the file and there are no allocations
379 * past this one, then extend the allocation out to the file system's write
382 * If we don't have a user specified preallocation size, dynamically increase
383 * the preallocation size as the size of the file grows. Cap the maximum size
384 * at a single extent or less if the filesystem is near full. The closer the
385 * filesystem is to full, the smaller the maximum prealocation.
387 * As an exception we don't do any preallocation at all if the file is smaller
388 * than the minimum preallocation and we are using the default dynamic
389 * preallocation scheme, as it is likely this is the only write to the file that
390 * is going to be done.
392 * We clean up any extra space left over when the file is closed in
396 xfs_iomap_prealloc_size(
397 struct xfs_inode *ip,
401 struct xfs_iext_cursor *icur)
403 struct xfs_mount *mp = ip->i_mount;
404 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
405 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
406 struct xfs_bmbt_irec prev;
409 xfs_fsblock_t qblocks;
411 xfs_fsblock_t alloc_blocks = 0;
413 if (offset + count <= XFS_ISIZE(ip))
416 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
417 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
421 * If an explicit allocsize is set, the file is small, or we
422 * are writing behind a hole, then use the minimum prealloc:
424 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
425 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
426 !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
427 prev.br_startoff + prev.br_blockcount < offset_fsb)
428 return mp->m_writeio_blocks;
431 * Determine the initial size of the preallocation. We are beyond the
432 * current EOF here, but we need to take into account whether this is
433 * a sparse write or an extending write when determining the
434 * preallocation size. Hence we need to look up the extent that ends
435 * at the current write offset and use the result to determine the
436 * preallocation size.
438 * If the extent is a hole, then preallocation is essentially disabled.
439 * Otherwise we take the size of the preceding data extent as the basis
440 * for the preallocation size. If the size of the extent is greater than
441 * half the maximum extent length, then use the current offset as the
442 * basis. This ensures that for large files the preallocation size
443 * always extends to MAXEXTLEN rather than falling short due to things
444 * like stripe unit/width alignment of real extents.
446 if (prev.br_blockcount <= (MAXEXTLEN >> 1))
447 alloc_blocks = prev.br_blockcount << 1;
449 alloc_blocks = XFS_B_TO_FSB(mp, offset);
452 qblocks = alloc_blocks;
455 * MAXEXTLEN is not a power of two value but we round the prealloc down
456 * to the nearest power of two value after throttling. To prevent the
457 * round down from unconditionally reducing the maximum supported prealloc
458 * size, we round up first, apply appropriate throttling, round down and
459 * cap the value to MAXEXTLEN.
461 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
464 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
465 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
467 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
469 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
471 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
473 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
478 * Check each quota to cap the prealloc size, provide a shift value to
479 * throttle with and adjust amount of available space.
481 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
482 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
484 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
485 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
487 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
488 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
492 * The final prealloc size is set to the minimum of free space available
493 * in each of the quotas and the overall filesystem.
495 * The shift throttle value is set to the maximum value as determined by
496 * the global low free space values and per-quota low free space values.
498 alloc_blocks = min(alloc_blocks, qblocks);
499 shift = max(shift, qshift);
502 alloc_blocks >>= shift;
504 * rounddown_pow_of_two() returns an undefined result if we pass in
508 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
509 if (alloc_blocks > MAXEXTLEN)
510 alloc_blocks = MAXEXTLEN;
513 * If we are still trying to allocate more space than is
514 * available, squash the prealloc hard. This can happen if we
515 * have a large file on a small filesystem and the above
516 * lowspace thresholds are smaller than MAXEXTLEN.
518 while (alloc_blocks && alloc_blocks >= freesp)
521 if (alloc_blocks < mp->m_writeio_blocks)
522 alloc_blocks = mp->m_writeio_blocks;
523 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
524 mp->m_writeio_blocks);
529 xfs_file_iomap_begin_delay(
536 struct xfs_inode *ip = XFS_I(inode);
537 struct xfs_mount *mp = ip->i_mount;
538 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
539 xfs_fileoff_t maxbytes_fsb =
540 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
541 xfs_fileoff_t end_fsb;
542 struct xfs_bmbt_irec imap, cmap;
543 struct xfs_iext_cursor icur, ccur;
544 xfs_fsblock_t prealloc_blocks = 0;
545 bool eof = false, cow_eof = false, shared = false;
546 int whichfork = XFS_DATA_FORK;
549 ASSERT(!XFS_IS_REALTIME_INODE(ip));
550 ASSERT(!xfs_get_extsz_hint(ip));
552 xfs_ilock(ip, XFS_ILOCK_EXCL);
554 if (unlikely(XFS_TEST_ERROR(
555 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
556 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
557 mp, XFS_ERRTAG_BMAPIFORMAT))) {
558 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
559 error = -EFSCORRUPTED;
563 XFS_STATS_INC(mp, xs_blk_mapw);
565 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
566 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
571 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
574 * Search the data fork fork first to look up our source mapping. We
575 * always need the data fork map, as we have to return it to the
576 * iomap code so that the higher level write code can read data in to
577 * perform read-modify-write cycles for unaligned writes.
579 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
581 imap.br_startoff = end_fsb; /* fake hole until the end */
583 /* We never need to allocate blocks for zeroing a hole. */
584 if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
585 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
590 * Search the COW fork extent list even if we did not find a data fork
591 * extent. This serves two purposes: first this implements the
592 * speculative preallocation using cowextsize, so that we also unshare
593 * block adjacent to shared blocks instead of just the shared blocks
594 * themselves. Second the lookup in the extent list is generally faster
595 * than going out to the shared extent tree.
597 if (xfs_is_cow_inode(ip)) {
599 ASSERT(!xfs_is_reflink_inode(ip));
600 xfs_ifork_init_cow(ip);
602 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
604 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
605 trace_xfs_reflink_cow_found(ip, &cmap);
606 whichfork = XFS_COW_FORK;
611 if (imap.br_startoff <= offset_fsb) {
613 * For reflink files we may need a delalloc reservation when
614 * overwriting shared extents. This includes zeroing of
615 * existing extents that contain data.
617 if (!xfs_is_cow_inode(ip) ||
618 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
619 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
624 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
626 /* Trim the mapping to the nearest shared extent boundary. */
627 error = xfs_inode_need_cow(ip, &imap, &shared);
631 /* Not shared? Just report the (potentially capped) extent. */
633 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
639 * Fork all the shared blocks from our write offset until the
642 whichfork = XFS_COW_FORK;
643 end_fsb = imap.br_startoff + imap.br_blockcount;
646 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
647 * pages to keep the chunks of work done where somewhat
648 * symmetric with the work writeback does. This is a completely
649 * arbitrary number pulled out of thin air.
651 * Note that the values needs to be less than 32-bits wide until
652 * the lower level functions are updated.
654 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
655 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
657 if (xfs_is_always_cow_inode(ip))
658 whichfork = XFS_COW_FORK;
661 error = xfs_qm_dqattach_locked(ip, false);
666 prealloc_blocks = xfs_iomap_prealloc_size(ip, whichfork, offset,
668 if (prealloc_blocks) {
670 xfs_off_t end_offset;
671 xfs_fileoff_t p_end_fsb;
673 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
674 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
677 align = xfs_eof_alignment(ip, 0);
679 p_end_fsb = roundup_64(p_end_fsb, align);
681 p_end_fsb = min(p_end_fsb, maxbytes_fsb);
682 ASSERT(p_end_fsb > offset_fsb);
683 prealloc_blocks = p_end_fsb - end_fsb;
688 error = xfs_bmapi_reserve_delalloc(ip, whichfork, offset_fsb,
689 end_fsb - offset_fsb, prealloc_blocks,
690 whichfork == XFS_DATA_FORK ? &imap : &cmap,
691 whichfork == XFS_DATA_FORK ? &icur : &ccur,
692 whichfork == XFS_DATA_FORK ? eof : cow_eof);
698 /* retry without any preallocation */
699 trace_xfs_delalloc_enospc(ip, offset, count);
700 if (prealloc_blocks) {
710 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
711 * them out if the write happens to fail.
713 iomap->flags |= IOMAP_F_NEW;
714 trace_xfs_iomap_alloc(ip, offset, count, whichfork,
715 whichfork == XFS_DATA_FORK ? &imap : &cmap);
717 if (whichfork == XFS_COW_FORK) {
718 if (imap.br_startoff > offset_fsb) {
719 xfs_trim_extent(&cmap, offset_fsb,
720 imap.br_startoff - offset_fsb);
721 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, true);
724 /* ensure we only report blocks we have a reservation for */
725 xfs_trim_extent(&imap, cmap.br_startoff, cmap.br_blockcount);
728 error = xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
730 xfs_iunlock(ip, XFS_ILOCK_EXCL);
735 xfs_iomap_write_unwritten(
741 xfs_mount_t *mp = ip->i_mount;
742 xfs_fileoff_t offset_fsb;
743 xfs_filblks_t count_fsb;
744 xfs_filblks_t numblks_fsb;
747 xfs_bmbt_irec_t imap;
748 struct inode *inode = VFS_I(ip);
753 trace_xfs_unwritten_convert(ip, offset, count);
755 offset_fsb = XFS_B_TO_FSBT(mp, offset);
756 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
757 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
760 * Reserve enough blocks in this transaction for two complete extent
761 * btree splits. We may be converting the middle part of an unwritten
762 * extent and in this case we will insert two new extents in the btree
763 * each of which could cause a full split.
765 * This reservation amount will be used in the first call to
766 * xfs_bmbt_split() to select an AG with enough space to satisfy the
767 * rest of the operation.
769 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
773 * Set up a transaction to convert the range of extents
774 * from unwritten to real. Do allocations in a loop until
775 * we have covered the range passed in.
777 * Note that we can't risk to recursing back into the filesystem
778 * here as we might be asked to write out the same inode that we
779 * complete here and might deadlock on the iolock.
781 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
782 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
786 xfs_ilock(ip, XFS_ILOCK_EXCL);
787 xfs_trans_ijoin(tp, ip, 0);
790 * Modify the unwritten extent state of the buffer.
793 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
794 XFS_BMAPI_CONVERT, resblks, &imap,
797 goto error_on_bmapi_transaction;
800 * Log the updated inode size as we go. We have to be careful
801 * to only log it up to the actual write offset if it is
802 * halfway into a block.
804 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
805 if (i_size > offset + count)
806 i_size = offset + count;
807 if (update_isize && i_size > i_size_read(inode))
808 i_size_write(inode, i_size);
809 i_size = xfs_new_eof(ip, i_size);
811 ip->i_d.di_size = i_size;
812 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
815 error = xfs_trans_commit(tp);
816 xfs_iunlock(ip, XFS_ILOCK_EXCL);
820 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
821 return xfs_alert_fsblock_zero(ip, &imap);
823 if ((numblks_fsb = imap.br_blockcount) == 0) {
825 * The numblks_fsb value should always get
826 * smaller, otherwise the loop is stuck.
828 ASSERT(imap.br_blockcount);
831 offset_fsb += numblks_fsb;
832 count_fsb -= numblks_fsb;
833 } while (count_fsb > 0);
837 error_on_bmapi_transaction:
838 xfs_trans_cancel(tp);
839 xfs_iunlock(ip, XFS_ILOCK_EXCL);
846 struct xfs_bmbt_irec *imap,
850 imap->br_startblock == HOLESTARTBLOCK ||
851 imap->br_startblock == DELAYSTARTBLOCK ||
852 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
856 needs_cow_for_zeroing(
857 struct xfs_bmbt_irec *imap,
861 imap->br_startblock != HOLESTARTBLOCK &&
862 imap->br_state != XFS_EXT_UNWRITTEN;
867 struct xfs_inode *ip,
871 unsigned mode = XFS_ILOCK_SHARED;
872 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
875 * COW writes may allocate delalloc space or convert unwritten COW
876 * extents, so we need to make sure to take the lock exclusively here.
878 if (xfs_is_cow_inode(ip) && is_write) {
880 * FIXME: It could still overwrite on unshared extents and not
883 if (flags & IOMAP_NOWAIT)
885 mode = XFS_ILOCK_EXCL;
889 * Extents not yet cached requires exclusive access, don't block. This
890 * is an opencoded xfs_ilock_data_map_shared() call but with
891 * non-blocking behaviour.
893 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
894 if (flags & IOMAP_NOWAIT)
896 mode = XFS_ILOCK_EXCL;
900 if (flags & IOMAP_NOWAIT) {
901 if (!xfs_ilock_nowait(ip, mode))
908 * The reflink iflag could have changed since the earlier unlocked
909 * check, so if we got ILOCK_SHARED for a write and but we're now a
910 * reflink inode we have to switch to ILOCK_EXCL and relock.
912 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
913 xfs_iunlock(ip, mode);
914 mode = XFS_ILOCK_EXCL;
923 xfs_file_iomap_begin(
930 struct xfs_inode *ip = XFS_I(inode);
931 struct xfs_mount *mp = ip->i_mount;
932 struct xfs_bmbt_irec imap;
933 xfs_fileoff_t offset_fsb, end_fsb;
934 int nimaps = 1, error = 0;
938 if (XFS_FORCED_SHUTDOWN(mp))
941 if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && !(flags & IOMAP_DIRECT) &&
942 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
943 /* Reserve delalloc blocks for regular writeback. */
944 return xfs_file_iomap_begin_delay(inode, offset, length, flags,
949 * Lock the inode in the manner required for the specified operation and
950 * check for as many conditions that would result in blocking as
951 * possible. This removes most of the non-blocking checks from the
952 * mapping code below.
954 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
958 ASSERT(offset <= mp->m_super->s_maxbytes);
959 if (offset > mp->m_super->s_maxbytes - length)
960 length = mp->m_super->s_maxbytes - offset;
961 offset_fsb = XFS_B_TO_FSBT(mp, offset);
962 end_fsb = XFS_B_TO_FSB(mp, offset + length);
964 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
969 if (flags & IOMAP_REPORT) {
970 /* Trim the mapping to the nearest shared extent boundary. */
971 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
976 /* Non-modifying mapping requested, so we are done */
977 if (!(flags & (IOMAP_WRITE | IOMAP_ZERO)))
981 * Break shared extents if necessary. Checks for non-blocking IO have
982 * been done up front, so we don't need to do them here.
984 if (xfs_is_cow_inode(ip)) {
985 struct xfs_bmbt_irec cmap;
986 bool directio = (flags & IOMAP_DIRECT);
988 /* if zeroing doesn't need COW allocation, then we are done. */
989 if ((flags & IOMAP_ZERO) &&
990 !needs_cow_for_zeroing(&imap, nimaps))
993 /* may drop and re-acquire the ilock */
995 error = xfs_reflink_allocate_cow(ip, &cmap, &shared, &lockmode,
1001 * For buffered writes we need to report the address of the
1002 * previous block (if there was any) so that the higher level
1003 * write code can perform read-modify-write operations; we
1004 * won't need the CoW fork mapping until writeback. For direct
1005 * I/O, which must be block aligned, we need to report the
1006 * newly allocated address. If the data fork has a hole, copy
1007 * the COW fork mapping to avoid allocating to the data fork.
1009 if (directio || imap.br_startblock == HOLESTARTBLOCK)
1012 end_fsb = imap.br_startoff + imap.br_blockcount;
1013 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1016 /* Don't need to allocate over holes when doing zeroing operations. */
1017 if (flags & IOMAP_ZERO)
1020 if (!imap_needs_alloc(inode, &imap, nimaps))
1023 /* If nowait is set bail since we are going to make allocations. */
1024 if (flags & IOMAP_NOWAIT) {
1030 * We cap the maximum length we map to a sane size to keep the chunks
1031 * of work done where somewhat symmetric with the work writeback does.
1032 * This is a completely arbitrary number pulled out of thin air as a
1033 * best guess for initial testing.
1035 * Note that the values needs to be less than 32-bits wide until the
1036 * lower level functions are updated.
1038 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1041 * xfs_iomap_write_direct() expects the shared lock. It is unlocked on
1044 if (lockmode == XFS_ILOCK_EXCL)
1045 xfs_ilock_demote(ip, lockmode);
1046 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1051 iomap->flags |= IOMAP_F_NEW;
1052 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
1055 return xfs_bmbt_to_iomap(ip, iomap, &imap, shared);
1059 xfs_iunlock(ip, lockmode);
1060 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1064 xfs_iunlock(ip, lockmode);
1069 xfs_file_iomap_end_delalloc(
1070 struct xfs_inode *ip,
1074 struct iomap *iomap)
1076 struct xfs_mount *mp = ip->i_mount;
1077 xfs_fileoff_t start_fsb;
1078 xfs_fileoff_t end_fsb;
1082 * Behave as if the write failed if drop writes is enabled. Set the NEW
1083 * flag to force delalloc cleanup.
1085 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1086 iomap->flags |= IOMAP_F_NEW;
1091 * start_fsb refers to the first unused block after a short write. If
1092 * nothing was written, round offset down to point at the first block in
1095 if (unlikely(!written))
1096 start_fsb = XFS_B_TO_FSBT(mp, offset);
1098 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1099 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1102 * Trim delalloc blocks if they were allocated by this write and we
1103 * didn't manage to write the whole range.
1105 * We don't need to care about racing delalloc as we hold i_mutex
1106 * across the reserve/allocate/unreserve calls. If there are delalloc
1107 * blocks in the range, they are ours.
1109 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1110 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1111 XFS_FSB_TO_B(mp, end_fsb) - 1);
1113 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1114 end_fsb - start_fsb);
1115 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1116 xfs_alert(mp, "%s: unable to clean up ino %lld",
1117 __func__, ip->i_ino);
1127 struct inode *inode,
1132 struct iomap *iomap)
1134 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1135 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1136 length, written, iomap);
1140 const struct iomap_ops xfs_iomap_ops = {
1141 .iomap_begin = xfs_file_iomap_begin,
1142 .iomap_end = xfs_file_iomap_end,
1146 xfs_seek_iomap_begin(
1147 struct inode *inode,
1151 struct iomap *iomap)
1153 struct xfs_inode *ip = XFS_I(inode);
1154 struct xfs_mount *mp = ip->i_mount;
1155 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1156 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1157 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1158 struct xfs_iext_cursor icur;
1159 struct xfs_bmbt_irec imap, cmap;
1163 if (XFS_FORCED_SHUTDOWN(mp))
1166 lockmode = xfs_ilock_data_map_shared(ip);
1167 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
1168 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1173 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1175 * If we found a data extent we are done.
1177 if (imap.br_startoff <= offset_fsb)
1179 data_fsb = imap.br_startoff;
1182 * Fake a hole until the end of the file.
1184 data_fsb = min(XFS_B_TO_FSB(mp, offset + length),
1185 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1189 * If a COW fork extent covers the hole, report it - capped to the next
1192 if (xfs_inode_has_cow_data(ip) &&
1193 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1194 cow_fsb = cmap.br_startoff;
1195 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1196 if (data_fsb < cow_fsb + cmap.br_blockcount)
1197 end_fsb = min(end_fsb, data_fsb);
1198 xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1199 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, true);
1201 * This is a COW extent, so we must probe the page cache
1202 * because there could be dirty page cache being backed
1205 iomap->type = IOMAP_UNWRITTEN;
1210 * Else report a hole, capped to the next found data or COW extent.
1212 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1213 imap.br_blockcount = cow_fsb - offset_fsb;
1215 imap.br_blockcount = data_fsb - offset_fsb;
1216 imap.br_startoff = offset_fsb;
1217 imap.br_startblock = HOLESTARTBLOCK;
1218 imap.br_state = XFS_EXT_NORM;
1220 xfs_trim_extent(&imap, offset_fsb, end_fsb);
1221 error = xfs_bmbt_to_iomap(ip, iomap, &imap, false);
1223 xfs_iunlock(ip, lockmode);
1227 const struct iomap_ops xfs_seek_iomap_ops = {
1228 .iomap_begin = xfs_seek_iomap_begin,
1232 xfs_xattr_iomap_begin(
1233 struct inode *inode,
1237 struct iomap *iomap)
1239 struct xfs_inode *ip = XFS_I(inode);
1240 struct xfs_mount *mp = ip->i_mount;
1241 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1242 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1243 struct xfs_bmbt_irec imap;
1244 int nimaps = 1, error = 0;
1247 if (XFS_FORCED_SHUTDOWN(mp))
1250 lockmode = xfs_ilock_attr_map_shared(ip);
1252 /* if there are no attribute fork or extents, return ENOENT */
1253 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1258 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1259 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1260 &nimaps, XFS_BMAPI_ATTRFORK);
1262 xfs_iunlock(ip, lockmode);
1267 return xfs_bmbt_to_iomap(ip, iomap, &imap, false);
1270 const struct iomap_ops xfs_xattr_iomap_ops = {
1271 .iomap_begin = xfs_xattr_iomap_begin,