2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_btree.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_itable.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_utils.h"
45 #include "xfs_iomap.h"
46 #include "xfs_trace.h"
49 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
51 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
54 xfs_iomap_eof_align_last_fsb(
58 xfs_fileoff_t *last_fsb)
60 xfs_fileoff_t new_last_fsb = 0;
61 xfs_extlen_t align = 0;
64 if (!XFS_IS_REALTIME_INODE(ip)) {
66 * Round up the allocation request to a stripe unit
67 * (m_dalign) boundary if the file size is >= stripe unit
68 * size, and we are allocating past the allocation eof.
70 * If mounted with the "-o swalloc" option the alignment is
71 * increased from the strip unit size to the stripe width.
73 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
75 else if (mp->m_dalign)
78 if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
79 new_last_fsb = roundup_64(*last_fsb, align);
83 * Always round up the allocation request to an extent boundary
84 * (when file on a real-time subvolume or has di_extsize hint).
88 align = roundup_64(new_last_fsb, extsize);
91 new_last_fsb = roundup_64(*last_fsb, align);
95 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
99 *last_fsb = new_last_fsb;
105 xfs_alert_fsblock_zero(
107 xfs_bmbt_irec_t *imap)
109 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
110 "Access to block zero in inode %llu "
111 "start_block: %llx start_off: %llx "
112 "blkcnt: %llx extent-state: %x\n",
113 (unsigned long long)ip->i_ino,
114 (unsigned long long)imap->br_startblock,
115 (unsigned long long)imap->br_startoff,
116 (unsigned long long)imap->br_blockcount,
122 xfs_iomap_write_direct(
126 xfs_bmbt_irec_t *imap,
129 xfs_mount_t *mp = ip->i_mount;
130 xfs_fileoff_t offset_fsb;
131 xfs_fileoff_t last_fsb;
132 xfs_filblks_t count_fsb, resaligned;
133 xfs_fsblock_t firstfsb;
134 xfs_extlen_t extsz, temp;
140 xfs_bmap_free_t free_list;
141 uint qblocks, resblks, resrtextents;
145 error = xfs_qm_dqattach(ip, 0);
147 return XFS_ERROR(error);
149 rt = XFS_IS_REALTIME_INODE(ip);
150 extsz = xfs_get_extsz_hint(ip);
152 offset_fsb = XFS_B_TO_FSBT(mp, offset);
153 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
154 if ((offset + count) > XFS_ISIZE(ip)) {
155 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
157 return XFS_ERROR(error);
159 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
160 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
161 imap->br_blockcount +
164 count_fsb = last_fsb - offset_fsb;
165 ASSERT(count_fsb > 0);
167 resaligned = count_fsb;
168 if (unlikely(extsz)) {
169 if ((temp = do_mod(offset_fsb, extsz)))
171 if ((temp = do_mod(resaligned, extsz)))
172 resaligned += extsz - temp;
176 resrtextents = qblocks = resaligned;
177 resrtextents /= mp->m_sb.sb_rextsize;
178 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
179 quota_flag = XFS_QMOPT_RES_RTBLKS;
182 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
183 quota_flag = XFS_QMOPT_RES_REGBLKS;
187 * Allocate and setup the transaction
189 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
190 error = xfs_trans_reserve(tp, resblks,
191 XFS_WRITE_LOG_RES(mp), resrtextents,
192 XFS_TRANS_PERM_LOG_RES,
193 XFS_WRITE_LOG_COUNT);
195 * Check for running out of space, note: need lock to return
198 xfs_trans_cancel(tp, 0);
199 return XFS_ERROR(error);
202 xfs_ilock(ip, XFS_ILOCK_EXCL);
204 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
206 goto out_trans_cancel;
208 xfs_trans_ijoin(tp, ip, 0);
211 if (offset < XFS_ISIZE(ip) || extsz)
212 bmapi_flag |= XFS_BMAPI_PREALLOC;
215 * From this point onwards we overwrite the imap pointer that the
218 xfs_bmap_init(&free_list, &firstfsb);
220 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
221 &firstfsb, 0, imap, &nimaps, &free_list);
223 goto out_bmap_cancel;
226 * Complete the transaction
228 error = xfs_bmap_finish(&tp, &free_list, &committed);
230 goto out_bmap_cancel;
231 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
236 * Copy any maps to caller's array and return any error.
239 error = XFS_ERROR(ENOSPC);
243 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
244 error = xfs_alert_fsblock_zero(ip, imap);
247 xfs_iunlock(ip, XFS_ILOCK_EXCL);
251 xfs_bmap_cancel(&free_list);
252 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
254 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
259 * If the caller is doing a write at the end of the file, then extend the
260 * allocation out to the file system's write iosize. We clean up any extra
261 * space left over when the file is closed in xfs_inactive().
263 * If we find we already have delalloc preallocation beyond EOF, don't do more
264 * preallocation as it it not needed.
267 xfs_iomap_eof_want_preallocate(
272 xfs_bmbt_irec_t *imap,
276 xfs_fileoff_t start_fsb;
277 xfs_filblks_t count_fsb;
278 xfs_fsblock_t firstblock;
280 int found_delalloc = 0;
283 if (offset + count <= XFS_ISIZE(ip))
287 * If there are any real blocks past eof, then don't
288 * do any speculative allocation.
290 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
291 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
292 while (count_fsb > 0) {
294 firstblock = NULLFSBLOCK;
295 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
299 for (n = 0; n < imaps; n++) {
300 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
301 (imap[n].br_startblock != DELAYSTARTBLOCK))
303 start_fsb += imap[n].br_blockcount;
304 count_fsb -= imap[n].br_blockcount;
306 if (imap[n].br_startblock == DELAYSTARTBLOCK)
316 * If we don't have a user specified preallocation size, dynamically increase
317 * the preallocation size as the size of the file grows. Cap the maximum size
318 * at a single extent or less if the filesystem is near full. The closer the
319 * filesystem is to full, the smaller the maximum prealocation.
322 xfs_iomap_prealloc_size(
323 struct xfs_mount *mp,
324 struct xfs_inode *ip)
326 xfs_fsblock_t alloc_blocks = 0;
328 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
333 * rounddown_pow_of_two() returns an undefined result
334 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
335 * ensure we always pass in a non-zero value.
337 alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
338 alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
339 rounddown_pow_of_two(alloc_blocks));
341 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
342 freesp = mp->m_sb.sb_fdblocks;
343 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
345 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
347 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
349 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
351 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
355 alloc_blocks >>= shift;
358 if (alloc_blocks < mp->m_writeio_blocks)
359 alloc_blocks = mp->m_writeio_blocks;
365 xfs_iomap_write_delay(
369 xfs_bmbt_irec_t *ret_imap)
371 xfs_mount_t *mp = ip->i_mount;
372 xfs_fileoff_t offset_fsb;
373 xfs_fileoff_t last_fsb;
374 xfs_off_t aligned_offset;
375 xfs_fileoff_t ioalign;
378 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
379 int prealloc, flushed = 0;
382 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
385 * Make sure that the dquots are there. This doesn't hold
386 * the ilock across a disk read.
388 error = xfs_qm_dqattach_locked(ip, 0);
390 return XFS_ERROR(error);
392 extsz = xfs_get_extsz_hint(ip);
393 offset_fsb = XFS_B_TO_FSBT(mp, offset);
396 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
397 imap, XFS_WRITE_IMAPS, &prealloc);
403 xfs_fsblock_t alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
405 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
406 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
407 last_fsb = ioalign + alloc_blocks;
409 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
412 if (prealloc || extsz) {
413 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
419 * Make sure preallocation does not create extents beyond the range we
420 * actually support in this filesystem.
422 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_maxioffset))
423 last_fsb = XFS_B_TO_FSB(mp, mp->m_maxioffset);
425 ASSERT(last_fsb > offset_fsb);
427 nimaps = XFS_WRITE_IMAPS;
428 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
429 imap, &nimaps, XFS_BMAPI_ENTIRE);
436 return XFS_ERROR(error);
440 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
441 * ENOSPC, * flush all other inodes with delalloc blocks to free up
442 * some of the excess reserved metadata space. For both cases, retry
443 * without EOF preallocation.
446 trace_xfs_delalloc_enospc(ip, offset, count);
448 return XFS_ERROR(error ? error : ENOSPC);
450 if (error == ENOSPC) {
451 xfs_iunlock(ip, XFS_ILOCK_EXCL);
452 xfs_flush_inodes(ip);
453 xfs_ilock(ip, XFS_ILOCK_EXCL);
462 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
463 return xfs_alert_fsblock_zero(ip, &imap[0]);
470 * Pass in a delayed allocate extent, convert it to real extents;
471 * return to the caller the extent we create which maps on top of
472 * the originating callers request.
474 * Called without a lock on the inode.
476 * We no longer bother to look at the incoming map - all we have to
477 * guarantee is that whatever we allocate fills the required range.
480 xfs_iomap_write_allocate(
484 xfs_bmbt_irec_t *imap)
486 xfs_mount_t *mp = ip->i_mount;
487 xfs_fileoff_t offset_fsb, last_block;
488 xfs_fileoff_t end_fsb, map_start_fsb;
489 xfs_fsblock_t first_block;
490 xfs_bmap_free_t free_list;
491 xfs_filblks_t count_fsb;
493 int nimaps, committed;
498 * Make sure that the dquots are there.
500 error = xfs_qm_dqattach(ip, 0);
502 return XFS_ERROR(error);
504 offset_fsb = XFS_B_TO_FSBT(mp, offset);
505 count_fsb = imap->br_blockcount;
506 map_start_fsb = imap->br_startoff;
508 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
510 while (count_fsb != 0) {
512 * Set up a transaction with which to allocate the
513 * backing store for the file. Do allocations in a
514 * loop until we get some space in the range we are
515 * interested in. The other space that might be allocated
516 * is in the delayed allocation extent on which we sit
517 * but before our buffer starts.
521 while (nimaps == 0) {
522 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
523 tp->t_flags |= XFS_TRANS_RESERVE;
524 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
525 error = xfs_trans_reserve(tp, nres,
526 XFS_WRITE_LOG_RES(mp),
527 0, XFS_TRANS_PERM_LOG_RES,
528 XFS_WRITE_LOG_COUNT);
530 xfs_trans_cancel(tp, 0);
531 return XFS_ERROR(error);
533 xfs_ilock(ip, XFS_ILOCK_EXCL);
534 xfs_trans_ijoin(tp, ip, 0);
536 xfs_bmap_init(&free_list, &first_block);
539 * it is possible that the extents have changed since
540 * we did the read call as we dropped the ilock for a
541 * while. We have to be careful about truncates or hole
542 * punchs here - we are not allowed to allocate
543 * non-delalloc blocks here.
545 * The only protection against truncation is the pages
546 * for the range we are being asked to convert are
547 * locked and hence a truncate will block on them
550 * As a result, if we go beyond the range we really
551 * need and hit an delalloc extent boundary followed by
552 * a hole while we have excess blocks in the map, we
553 * will fill the hole incorrectly and overrun the
554 * transaction reservation.
556 * Using a single map prevents this as we are forced to
557 * check each map we look for overlap with the desired
558 * range and abort as soon as we find it. Also, given
559 * that we only return a single map, having one beyond
560 * what we can return is probably a bit silly.
562 * We also need to check that we don't go beyond EOF;
563 * this is a truncate optimisation as a truncate sets
564 * the new file size before block on the pages we
565 * currently have locked under writeback. Because they
566 * are about to be tossed, we don't need to write them
570 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
571 error = xfs_bmap_last_offset(NULL, ip, &last_block,
576 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
577 if ((map_start_fsb + count_fsb) > last_block) {
578 count_fsb = last_block - map_start_fsb;
579 if (count_fsb == 0) {
586 * From this point onwards we overwrite the imap
587 * pointer that the caller gave to us.
589 error = xfs_bmapi_write(tp, ip, map_start_fsb,
590 count_fsb, 0, &first_block, 1,
591 imap, &nimaps, &free_list);
595 error = xfs_bmap_finish(&tp, &free_list, &committed);
599 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
603 xfs_iunlock(ip, XFS_ILOCK_EXCL);
607 * See if we were able to allocate an extent that
608 * covers at least part of the callers request
610 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
611 return xfs_alert_fsblock_zero(ip, imap);
613 if ((offset_fsb >= imap->br_startoff) &&
614 (offset_fsb < (imap->br_startoff +
615 imap->br_blockcount))) {
616 XFS_STATS_INC(xs_xstrat_quick);
621 * So far we have not mapped the requested part of the
622 * file, just surrounding data, try again.
624 count_fsb -= imap->br_blockcount;
625 map_start_fsb = imap->br_startoff + imap->br_blockcount;
629 xfs_bmap_cancel(&free_list);
630 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
632 xfs_iunlock(ip, XFS_ILOCK_EXCL);
633 return XFS_ERROR(error);
637 xfs_iomap_write_unwritten(
642 xfs_mount_t *mp = ip->i_mount;
643 xfs_fileoff_t offset_fsb;
644 xfs_filblks_t count_fsb;
645 xfs_filblks_t numblks_fsb;
646 xfs_fsblock_t firstfsb;
649 xfs_bmbt_irec_t imap;
650 xfs_bmap_free_t free_list;
656 trace_xfs_unwritten_convert(ip, offset, count);
658 offset_fsb = XFS_B_TO_FSBT(mp, offset);
659 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
660 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
663 * Reserve enough blocks in this transaction for two complete extent
664 * btree splits. We may be converting the middle part of an unwritten
665 * extent and in this case we will insert two new extents in the btree
666 * each of which could cause a full split.
668 * This reservation amount will be used in the first call to
669 * xfs_bmbt_split() to select an AG with enough space to satisfy the
670 * rest of the operation.
672 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
676 * set up a transaction to convert the range of extents
677 * from unwritten to real. Do allocations in a loop until
678 * we have covered the range passed in.
680 * Note that we open code the transaction allocation here
681 * to pass KM_NOFS--we can't risk to recursing back into
682 * the filesystem here as we might be asked to write out
683 * the same inode that we complete here and might deadlock
686 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
687 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
688 tp->t_flags |= XFS_TRANS_RESERVE;
689 error = xfs_trans_reserve(tp, resblks,
690 XFS_WRITE_LOG_RES(mp), 0,
691 XFS_TRANS_PERM_LOG_RES,
692 XFS_WRITE_LOG_COUNT);
694 xfs_trans_cancel(tp, 0);
695 return XFS_ERROR(error);
698 xfs_ilock(ip, XFS_ILOCK_EXCL);
699 xfs_trans_ijoin(tp, ip, 0);
702 * Modify the unwritten extent state of the buffer.
704 xfs_bmap_init(&free_list, &firstfsb);
706 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
707 XFS_BMAPI_CONVERT, &firstfsb,
708 1, &imap, &nimaps, &free_list);
710 goto error_on_bmapi_transaction;
713 * Log the updated inode size as we go. We have to be careful
714 * to only log it up to the actual write offset if it is
715 * halfway into a block.
717 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
718 if (i_size > offset + count)
719 i_size = offset + count;
721 i_size = xfs_new_eof(ip, i_size);
723 ip->i_d.di_size = i_size;
724 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
727 error = xfs_bmap_finish(&tp, &free_list, &committed);
729 goto error_on_bmapi_transaction;
731 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
732 xfs_iunlock(ip, XFS_ILOCK_EXCL);
734 return XFS_ERROR(error);
736 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
737 return xfs_alert_fsblock_zero(ip, &imap);
739 if ((numblks_fsb = imap.br_blockcount) == 0) {
741 * The numblks_fsb value should always get
742 * smaller, otherwise the loop is stuck.
744 ASSERT(imap.br_blockcount);
747 offset_fsb += numblks_fsb;
748 count_fsb -= numblks_fsb;
749 } while (count_fsb > 0);
753 error_on_bmapi_transaction:
754 xfs_bmap_cancel(&free_list);
755 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
756 xfs_iunlock(ip, XFS_ILOCK_EXCL);
757 return XFS_ERROR(error);