]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/xfs/xfs_iomap.c
xfs: factor unwritten extent map manipulations out of xfs_bmapi
[mirror_ubuntu-zesty-kernel.git] / fs / xfs / xfs_iomap.c
CommitLineData
1da177e4 1/*
3e57ecf6 2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
7b718769 3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
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
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
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.
1da177e4 13 *
7b718769
NS
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
1da177e4 17 */
1da177e4 18#include "xfs.h"
1da177e4 19#include "xfs_fs.h"
a844f451 20#include "xfs_bit.h"
1da177e4 21#include "xfs_log.h"
a844f451 22#include "xfs_inum.h"
1da177e4
LT
23#include "xfs_trans.h"
24#include "xfs_sb.h"
25#include "xfs_ag.h"
1da177e4 26#include "xfs_alloc.h"
1da177e4
LT
27#include "xfs_quota.h"
28#include "xfs_mount.h"
1da177e4 29#include "xfs_bmap_btree.h"
a844f451 30#include "xfs_alloc_btree.h"
1da177e4 31#include "xfs_ialloc_btree.h"
1da177e4
LT
32#include "xfs_dinode.h"
33#include "xfs_inode.h"
a844f451 34#include "xfs_btree.h"
1da177e4 35#include "xfs_bmap.h"
1da177e4
LT
36#include "xfs_rtalloc.h"
37#include "xfs_error.h"
38#include "xfs_itable.h"
39#include "xfs_rw.h"
1da177e4
LT
40#include "xfs_attr.h"
41#include "xfs_buf_item.h"
42#include "xfs_trans_space.h"
43#include "xfs_utils.h"
44#include "xfs_iomap.h"
0b1b213f 45#include "xfs_trace.h"
1da177e4 46
1da177e4
LT
47
48#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
49 << mp->m_writeio_log)
1da177e4
LT
50#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
51
dd9f438e
NS
52STATIC int
53xfs_iomap_eof_align_last_fsb(
54 xfs_mount_t *mp,
541d7d3c 55 xfs_inode_t *ip,
dd9f438e
NS
56 xfs_extlen_t extsize,
57 xfs_fileoff_t *last_fsb)
58{
59 xfs_fileoff_t new_last_fsb = 0;
60 xfs_extlen_t align;
61 int eof, error;
62
71ddabb9 63 if (XFS_IS_REALTIME_INODE(ip))
dd9f438e
NS
64 ;
65 /*
66 * If mounted with the "-o swalloc" option, roundup the allocation
67 * request to a stripe width boundary if the file size is >=
68 * stripe width and we are allocating past the allocation eof.
69 */
70 else if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC) &&
9f6c92b9 71 (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_swidth)))
dd9f438e
NS
72 new_last_fsb = roundup_64(*last_fsb, mp->m_swidth);
73 /*
74 * Roundup the allocation request to a stripe unit (m_dalign) boundary
75 * if the file size is >= stripe unit size, and we are allocating past
76 * the allocation eof.
77 */
9f6c92b9 78 else if (mp->m_dalign && (ip->i_size >= XFS_FSB_TO_B(mp, mp->m_dalign)))
dd9f438e
NS
79 new_last_fsb = roundup_64(*last_fsb, mp->m_dalign);
80
81 /*
82 * Always round up the allocation request to an extent boundary
83 * (when file on a real-time subvolume or has di_extsize hint).
84 */
85 if (extsize) {
86 if (new_last_fsb)
87 align = roundup_64(new_last_fsb, extsize);
88 else
89 align = extsize;
90 new_last_fsb = roundup_64(*last_fsb, align);
91 }
92
93 if (new_last_fsb) {
541d7d3c 94 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
dd9f438e
NS
95 if (error)
96 return error;
97 if (eof)
98 *last_fsb = new_last_fsb;
99 }
100 return 0;
101}
102
572d95f4 103STATIC int
6d4a8ecb 104xfs_alert_fsblock_zero(
572d95f4
NS
105 xfs_inode_t *ip,
106 xfs_bmbt_irec_t *imap)
107{
6a19d939 108 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
572d95f4
NS
109 "Access to block zero in inode %llu "
110 "start_block: %llx start_off: %llx "
111 "blkcnt: %llx extent-state: %x\n",
112 (unsigned long long)ip->i_ino,
113 (unsigned long long)imap->br_startblock,
114 (unsigned long long)imap->br_startoff,
115 (unsigned long long)imap->br_blockcount,
116 imap->br_state);
117 return EFSCORRUPTED;
118}
119
a206c817 120int
1da177e4
LT
121xfs_iomap_write_direct(
122 xfs_inode_t *ip,
f403b7f4 123 xfs_off_t offset,
1da177e4 124 size_t count,
3070451e 125 xfs_bmbt_irec_t *imap,
405f8042 126 int nmaps)
1da177e4
LT
127{
128 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
129 xfs_fileoff_t offset_fsb;
130 xfs_fileoff_t last_fsb;
dd9f438e 131 xfs_filblks_t count_fsb, resaligned;
1da177e4 132 xfs_fsblock_t firstfsb;
dd9f438e 133 xfs_extlen_t extsz, temp;
0116d935 134 int nimaps;
1da177e4 135 int bmapi_flag;
06d10dd9 136 int quota_flag;
1da177e4
LT
137 int rt;
138 xfs_trans_t *tp;
1da177e4 139 xfs_bmap_free_t free_list;
dd9f438e 140 uint qblocks, resblks, resrtextents;
1da177e4 141 int committed;
dd9f438e 142 int error;
1da177e4
LT
143
144 /*
145 * Make sure that the dquots are there. This doesn't hold
146 * the ilock across a disk read.
147 */
7d095257 148 error = xfs_qm_dqattach_locked(ip, 0);
1da177e4
LT
149 if (error)
150 return XFS_ERROR(error);
151
dd9f438e 152 rt = XFS_IS_REALTIME_INODE(ip);
957d0ebe 153 extsz = xfs_get_extsz_hint(ip);
1da177e4 154
957d0ebe
DC
155 offset_fsb = XFS_B_TO_FSBT(mp, offset);
156 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
9f6c92b9
LM
157 if ((offset + count) > ip->i_size) {
158 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
dd9f438e
NS
159 if (error)
160 goto error_out;
1da177e4 161 } else {
405f8042 162 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
dd9f438e 163 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
3070451e
CH
164 imap->br_blockcount +
165 imap->br_startoff);
1da177e4 166 }
dd9f438e
NS
167 count_fsb = last_fsb - offset_fsb;
168 ASSERT(count_fsb > 0);
169
170 resaligned = count_fsb;
171 if (unlikely(extsz)) {
172 if ((temp = do_mod(offset_fsb, extsz)))
173 resaligned += temp;
174 if ((temp = do_mod(resaligned, extsz)))
175 resaligned += extsz - temp;
176 }
177
178 if (unlikely(rt)) {
179 resrtextents = qblocks = resaligned;
180 resrtextents /= mp->m_sb.sb_rextsize;
84e1e99f
DC
181 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
182 quota_flag = XFS_QMOPT_RES_RTBLKS;
183 } else {
184 resrtextents = 0;
dd9f438e 185 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
84e1e99f
DC
186 quota_flag = XFS_QMOPT_RES_REGBLKS;
187 }
1da177e4
LT
188
189 /*
06d10dd9 190 * Allocate and setup the transaction
1da177e4
LT
191 */
192 xfs_iunlock(ip, XFS_ILOCK_EXCL);
193 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1da177e4 194 error = xfs_trans_reserve(tp, resblks,
d52b44d0 195 XFS_WRITE_LOG_RES(mp), resrtextents,
1da177e4
LT
196 XFS_TRANS_PERM_LOG_RES,
197 XFS_WRITE_LOG_COUNT);
1da177e4 198 /*
06d10dd9 199 * Check for running out of space, note: need lock to return
1da177e4
LT
200 */
201 if (error)
1da177e4 202 xfs_trans_cancel(tp, 0);
1da177e4 203 xfs_ilock(ip, XFS_ILOCK_EXCL);
1da177e4 204 if (error)
06d10dd9 205 goto error_out;
1da177e4 206
7d095257 207 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
dd9f438e 208 if (error)
1da177e4 209 goto error1;
1da177e4 210
898621d5 211 xfs_trans_ijoin(tp, ip);
1da177e4 212
dd9f438e 213 bmapi_flag = XFS_BMAPI_WRITE;
405f8042 214 if (offset < ip->i_size || extsz)
1da177e4
LT
215 bmapi_flag |= XFS_BMAPI_PREALLOC;
216
217 /*
3070451e
CH
218 * Issue the xfs_bmapi() call to allocate the blocks.
219 *
220 * From this point onwards we overwrite the imap pointer that the
221 * caller gave to us.
1da177e4 222 */
9d87c319 223 xfs_bmap_init(&free_list, &firstfsb);
06d10dd9 224 nimaps = 1;
541d7d3c 225 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb, bmapi_flag,
3070451e 226 &firstfsb, 0, imap, &nimaps, &free_list);
06d10dd9 227 if (error)
1da177e4 228 goto error0;
1da177e4
LT
229
230 /*
06d10dd9 231 * Complete the transaction
1da177e4 232 */
f7c99b6f 233 error = xfs_bmap_finish(&tp, &free_list, &committed);
06d10dd9 234 if (error)
1da177e4 235 goto error0;
1c72bf90 236 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
06d10dd9 237 if (error)
1da177e4 238 goto error_out;
1da177e4 239
06d10dd9
NS
240 /*
241 * Copy any maps to caller's array and return any error.
242 */
1da177e4 243 if (nimaps == 0) {
572d95f4
NS
244 error = ENOSPC;
245 goto error_out;
246 }
247
3070451e 248 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) {
6d4a8ecb 249 error = xfs_alert_fsblock_zero(ip, imap);
1da177e4
LT
250 goto error_out;
251 }
252
1da177e4
LT
253 return 0;
254
06d10dd9 255error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1da177e4 256 xfs_bmap_cancel(&free_list);
7d095257 257 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
1da177e4 258
06d10dd9 259error1: /* Just cancel transaction */
1da177e4 260 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1da177e4
LT
261
262error_out:
263 return XFS_ERROR(error);
264}
265
dd9f438e 266/*
8de2bf93
DC
267 * If the caller is doing a write at the end of the file, then extend the
268 * allocation out to the file system's write iosize. We clean up any extra
269 * space left over when the file is closed in xfs_inactive().
055388a3
DC
270 *
271 * If we find we already have delalloc preallocation beyond EOF, don't do more
272 * preallocation as it it not needed.
dd9f438e
NS
273 */
274STATIC int
275xfs_iomap_eof_want_preallocate(
276 xfs_mount_t *mp,
541d7d3c 277 xfs_inode_t *ip,
dd9f438e
NS
278 xfs_off_t offset,
279 size_t count,
dd9f438e
NS
280 xfs_bmbt_irec_t *imap,
281 int nimaps,
282 int *prealloc)
283{
284 xfs_fileoff_t start_fsb;
285 xfs_filblks_t count_fsb;
286 xfs_fsblock_t firstblock;
287 int n, error, imaps;
055388a3 288 int found_delalloc = 0;
dd9f438e
NS
289
290 *prealloc = 0;
8de2bf93 291 if ((offset + count) <= ip->i_size)
dd9f438e
NS
292 return 0;
293
294 /*
295 * If there are any real blocks past eof, then don't
296 * do any speculative allocation.
297 */
298 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
299 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
300 while (count_fsb > 0) {
301 imaps = nimaps;
3ddb8fa9 302 firstblock = NULLFSBLOCK;
5c8ed202
DC
303 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
304 0);
dd9f438e
NS
305 if (error)
306 return error;
307 for (n = 0; n < imaps; n++) {
308 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
309 (imap[n].br_startblock != DELAYSTARTBLOCK))
310 return 0;
311 start_fsb += imap[n].br_blockcount;
312 count_fsb -= imap[n].br_blockcount;
055388a3
DC
313
314 if (imap[n].br_startblock == DELAYSTARTBLOCK)
315 found_delalloc = 1;
dd9f438e
NS
316 }
317 }
055388a3
DC
318 if (!found_delalloc)
319 *prealloc = 1;
dd9f438e
NS
320 return 0;
321}
322
055388a3
DC
323/*
324 * If we don't have a user specified preallocation size, dynamically increase
325 * the preallocation size as the size of the file grows. Cap the maximum size
326 * at a single extent or less if the filesystem is near full. The closer the
327 * filesystem is to full, the smaller the maximum prealocation.
328 */
329STATIC xfs_fsblock_t
330xfs_iomap_prealloc_size(
331 struct xfs_mount *mp,
332 struct xfs_inode *ip)
333{
334 xfs_fsblock_t alloc_blocks = 0;
335
336 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
337 int shift = 0;
338 int64_t freesp;
339
b8fc8263
DC
340 /*
341 * rounddown_pow_of_two() returns an undefined result
342 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
343 * ensure we always pass in a non-zero value.
344 */
345 alloc_blocks = XFS_B_TO_FSB(mp, ip->i_size) + 1;
055388a3
DC
346 alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
347 rounddown_pow_of_two(alloc_blocks));
348
349 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
350 freesp = mp->m_sb.sb_fdblocks;
351 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
352 shift = 2;
353 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
354 shift++;
355 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
356 shift++;
357 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
358 shift++;
359 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
360 shift++;
361 }
362 if (shift)
363 alloc_blocks >>= shift;
364 }
365
366 if (alloc_blocks < mp->m_writeio_blocks)
367 alloc_blocks = mp->m_writeio_blocks;
368
369 return alloc_blocks;
370}
371
a206c817 372int
1da177e4
LT
373xfs_iomap_write_delay(
374 xfs_inode_t *ip,
f403b7f4 375 xfs_off_t offset,
1da177e4 376 size_t count,
405f8042 377 xfs_bmbt_irec_t *ret_imap)
1da177e4
LT
378{
379 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
380 xfs_fileoff_t offset_fsb;
381 xfs_fileoff_t last_fsb;
dd9f438e
NS
382 xfs_off_t aligned_offset;
383 xfs_fileoff_t ioalign;
dd9f438e 384 xfs_extlen_t extsz;
1da177e4 385 int nimaps;
1da177e4 386 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
8de2bf93 387 int prealloc, flushed = 0;
dd9f438e 388 int error;
1da177e4 389
579aa9ca 390 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
391
392 /*
393 * Make sure that the dquots are there. This doesn't hold
394 * the ilock across a disk read.
395 */
7d095257 396 error = xfs_qm_dqattach_locked(ip, 0);
1da177e4
LT
397 if (error)
398 return XFS_ERROR(error);
399
957d0ebe 400 extsz = xfs_get_extsz_hint(ip);
dd9f438e
NS
401 offset_fsb = XFS_B_TO_FSBT(mp, offset);
402
055388a3 403
9f6c92b9 404 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
405f8042 405 imap, XFS_WRITE_IMAPS, &prealloc);
dd9f438e
NS
406 if (error)
407 return error;
1da177e4 408
8de2bf93 409retry:
dd9f438e 410 if (prealloc) {
055388a3
DC
411 xfs_fsblock_t alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
412
1da177e4
LT
413 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
414 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
055388a3 415 last_fsb = ioalign + alloc_blocks;
dd9f438e
NS
416 } else {
417 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
1da177e4 418 }
1da177e4 419
dd9f438e 420 if (prealloc || extsz) {
9f6c92b9 421 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
dd9f438e 422 if (error)
1da177e4 423 return error;
1da177e4 424 }
dd9f438e
NS
425
426 nimaps = XFS_WRITE_IMAPS;
4403280a
CH
427 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
428 imap, &nimaps, XFS_BMAPI_ENTIRE);
055388a3
DC
429 switch (error) {
430 case 0:
431 case ENOSPC:
432 case EDQUOT:
433 break;
434 default:
1da177e4 435 return XFS_ERROR(error);
055388a3 436 }
dd9f438e 437
1da177e4 438 /*
055388a3
DC
439 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
440 * ENOSPC, * flush all other inodes with delalloc blocks to free up
441 * some of the excess reserved metadata space. For both cases, retry
442 * without EOF preallocation.
1da177e4
LT
443 */
444 if (nimaps == 0) {
0b1b213f 445 trace_xfs_delalloc_enospc(ip, offset, count);
8de2bf93 446 if (flushed)
055388a3 447 return XFS_ERROR(error ? error : ENOSPC);
1da177e4 448
055388a3
DC
449 if (error == ENOSPC) {
450 xfs_iunlock(ip, XFS_ILOCK_EXCL);
451 xfs_flush_inodes(ip);
452 xfs_ilock(ip, XFS_ILOCK_EXCL);
453 }
8de2bf93
DC
454
455 flushed = 1;
1da177e4 456 error = 0;
8de2bf93 457 prealloc = 0;
1da177e4
LT
458 goto retry;
459 }
460
86c4d623 461 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 462 return xfs_alert_fsblock_zero(ip, &imap[0]);
dd9f438e
NS
463
464 *ret_imap = imap[0];
1da177e4
LT
465 return 0;
466}
467
468/*
469 * Pass in a delayed allocate extent, convert it to real extents;
470 * return to the caller the extent we create which maps on top of
471 * the originating callers request.
472 *
473 * Called without a lock on the inode.
e4143a1c
DC
474 *
475 * We no longer bother to look at the incoming map - all we have to
476 * guarantee is that whatever we allocate fills the required range.
1da177e4 477 */
a206c817 478int
1da177e4
LT
479xfs_iomap_write_allocate(
480 xfs_inode_t *ip,
f403b7f4 481 xfs_off_t offset,
24e17b5f 482 size_t count,
405f8042 483 xfs_bmbt_irec_t *imap)
1da177e4
LT
484{
485 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
486 xfs_fileoff_t offset_fsb, last_block;
487 xfs_fileoff_t end_fsb, map_start_fsb;
488 xfs_fsblock_t first_block;
489 xfs_bmap_free_t free_list;
490 xfs_filblks_t count_fsb;
1da177e4 491 xfs_trans_t *tp;
e4143a1c 492 int nimaps, committed;
1da177e4
LT
493 int error = 0;
494 int nres;
495
1da177e4
LT
496 /*
497 * Make sure that the dquots are there.
498 */
7d095257
CH
499 error = xfs_qm_dqattach(ip, 0);
500 if (error)
1da177e4
LT
501 return XFS_ERROR(error);
502
24e17b5f 503 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3070451e
CH
504 count_fsb = imap->br_blockcount;
505 map_start_fsb = imap->br_startoff;
1da177e4
LT
506
507 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
508
509 while (count_fsb != 0) {
510 /*
511 * Set up a transaction with which to allocate the
512 * backing store for the file. Do allocations in a
513 * loop until we get some space in the range we are
514 * interested in. The other space that might be allocated
515 * is in the delayed allocation extent on which we sit
516 * but before our buffer starts.
517 */
518
519 nimaps = 0;
520 while (nimaps == 0) {
521 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
84e1e99f 522 tp->t_flags |= XFS_TRANS_RESERVE;
1da177e4
LT
523 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
524 error = xfs_trans_reserve(tp, nres,
525 XFS_WRITE_LOG_RES(mp),
526 0, XFS_TRANS_PERM_LOG_RES,
527 XFS_WRITE_LOG_COUNT);
1da177e4
LT
528 if (error) {
529 xfs_trans_cancel(tp, 0);
530 return XFS_ERROR(error);
531 }
532 xfs_ilock(ip, XFS_ILOCK_EXCL);
898621d5 533 xfs_trans_ijoin(tp, ip);
1da177e4 534
9d87c319 535 xfs_bmap_init(&free_list, &first_block);
1da177e4 536
1da177e4 537 /*
e4143a1c
DC
538 * it is possible that the extents have changed since
539 * we did the read call as we dropped the ilock for a
540 * while. We have to be careful about truncates or hole
541 * punchs here - we are not allowed to allocate
542 * non-delalloc blocks here.
543 *
544 * The only protection against truncation is the pages
545 * for the range we are being asked to convert are
546 * locked and hence a truncate will block on them
547 * first.
548 *
549 * As a result, if we go beyond the range we really
550 * need and hit an delalloc extent boundary followed by
551 * a hole while we have excess blocks in the map, we
552 * will fill the hole incorrectly and overrun the
553 * transaction reservation.
554 *
555 * Using a single map prevents this as we are forced to
556 * check each map we look for overlap with the desired
557 * range and abort as soon as we find it. Also, given
558 * that we only return a single map, having one beyond
559 * what we can return is probably a bit silly.
560 *
561 * We also need to check that we don't go beyond EOF;
562 * this is a truncate optimisation as a truncate sets
563 * the new file size before block on the pages we
564 * currently have locked under writeback. Because they
565 * are about to be tossed, we don't need to write them
566 * back....
1da177e4 567 */
e4143a1c 568 nimaps = 1;
ba87ea69 569 end_fsb = XFS_B_TO_FSB(mp, ip->i_size);
7c9ef85c
DC
570 error = xfs_bmap_last_offset(NULL, ip, &last_block,
571 XFS_DATA_FORK);
572 if (error)
573 goto trans_cancel;
574
1da177e4
LT
575 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
576 if ((map_start_fsb + count_fsb) > last_block) {
577 count_fsb = last_block - map_start_fsb;
578 if (count_fsb == 0) {
579 error = EAGAIN;
580 goto trans_cancel;
581 }
582 }
583
3070451e
CH
584 /*
585 * Go get the actual blocks.
586 *
587 * From this point onwards we overwrite the imap
588 * pointer that the caller gave to us.
589 */
541d7d3c 590 error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
1da177e4 591 XFS_BMAPI_WRITE, &first_block, 1,
3070451e 592 imap, &nimaps, &free_list);
1da177e4
LT
593 if (error)
594 goto trans_cancel;
595
f7c99b6f 596 error = xfs_bmap_finish(&tp, &free_list, &committed);
1da177e4
LT
597 if (error)
598 goto trans_cancel;
599
1c72bf90 600 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1da177e4
LT
601 if (error)
602 goto error0;
603
604 xfs_iunlock(ip, XFS_ILOCK_EXCL);
605 }
606
607 /*
608 * See if we were able to allocate an extent that
609 * covers at least part of the callers request
610 */
3070451e 611 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 612 return xfs_alert_fsblock_zero(ip, imap);
86c4d623 613
3070451e
CH
614 if ((offset_fsb >= imap->br_startoff) &&
615 (offset_fsb < (imap->br_startoff +
616 imap->br_blockcount))) {
e4143a1c
DC
617 XFS_STATS_INC(xs_xstrat_quick);
618 return 0;
1da177e4
LT
619 }
620
e4143a1c
DC
621 /*
622 * So far we have not mapped the requested part of the
1da177e4
LT
623 * file, just surrounding data, try again.
624 */
3070451e
CH
625 count_fsb -= imap->br_blockcount;
626 map_start_fsb = imap->br_startoff + imap->br_blockcount;
1da177e4
LT
627 }
628
629trans_cancel:
630 xfs_bmap_cancel(&free_list);
631 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
632error0:
633 xfs_iunlock(ip, XFS_ILOCK_EXCL);
634 return XFS_ERROR(error);
635}
636
637int
638xfs_iomap_write_unwritten(
639 xfs_inode_t *ip,
f403b7f4 640 xfs_off_t offset,
1da177e4
LT
641 size_t count)
642{
643 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
644 xfs_fileoff_t offset_fsb;
645 xfs_filblks_t count_fsb;
646 xfs_filblks_t numblks_fsb;
dd9f438e
NS
647 xfs_fsblock_t firstfsb;
648 int nimaps;
649 xfs_trans_t *tp;
650 xfs_bmbt_irec_t imap;
651 xfs_bmap_free_t free_list;
652 uint resblks;
1da177e4
LT
653 int committed;
654 int error;
1da177e4 655
0b1b213f 656 trace_xfs_unwritten_convert(ip, offset, count);
1da177e4
LT
657
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);
661
4ddd8bb1
LM
662 /*
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.
667 *
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.
671 */
dd9f438e 672 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
1da177e4 673
dd9f438e 674 do {
1da177e4
LT
675 /*
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.
80641dc6
CH
679 *
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
684 * on the iolock.
1da177e4 685 */
80641dc6
CH
686 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
687 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
84e1e99f 688 tp->t_flags |= XFS_TRANS_RESERVE;
dd9f438e 689 error = xfs_trans_reserve(tp, resblks,
1da177e4
LT
690 XFS_WRITE_LOG_RES(mp), 0,
691 XFS_TRANS_PERM_LOG_RES,
692 XFS_WRITE_LOG_COUNT);
693 if (error) {
694 xfs_trans_cancel(tp, 0);
572d95f4 695 return XFS_ERROR(error);
1da177e4
LT
696 }
697
698 xfs_ilock(ip, XFS_ILOCK_EXCL);
898621d5 699 xfs_trans_ijoin(tp, ip);
1da177e4
LT
700
701 /*
702 * Modify the unwritten extent state of the buffer.
703 */
9d87c319 704 xfs_bmap_init(&free_list, &firstfsb);
1da177e4 705 nimaps = 1;
541d7d3c 706 error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
dd9f438e 707 XFS_BMAPI_WRITE|XFS_BMAPI_CONVERT, &firstfsb,
b4e9181e 708 1, &imap, &nimaps, &free_list);
1da177e4
LT
709 if (error)
710 goto error_on_bmapi_transaction;
711
f7c99b6f 712 error = xfs_bmap_finish(&(tp), &(free_list), &committed);
1da177e4
LT
713 if (error)
714 goto error_on_bmapi_transaction;
715
1c72bf90 716 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1da177e4
LT
717 xfs_iunlock(ip, XFS_ILOCK_EXCL);
718 if (error)
572d95f4
NS
719 return XFS_ERROR(error);
720
86c4d623 721 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 722 return xfs_alert_fsblock_zero(ip, &imap);
1da177e4
LT
723
724 if ((numblks_fsb = imap.br_blockcount) == 0) {
725 /*
726 * The numblks_fsb value should always get
727 * smaller, otherwise the loop is stuck.
728 */
729 ASSERT(imap.br_blockcount);
730 break;
731 }
732 offset_fsb += numblks_fsb;
733 count_fsb -= numblks_fsb;
734 } while (count_fsb > 0);
735
736 return 0;
737
738error_on_bmapi_transaction:
739 xfs_bmap_cancel(&free_list);
740 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
741 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1da177e4
LT
742 return XFS_ERROR(error);
743}