]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/xfs/xfs_iomap.c
xfs: zero allocation_args on the kernel stack
[mirror_ubuntu-bionic-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"
1da177e4
LT
20#include "xfs_log.h"
21#include "xfs_trans.h"
22#include "xfs_sb.h"
23#include "xfs_ag.h"
1da177e4 24#include "xfs_alloc.h"
1da177e4
LT
25#include "xfs_quota.h"
26#include "xfs_mount.h"
1da177e4 27#include "xfs_bmap_btree.h"
a844f451 28#include "xfs_alloc_btree.h"
1da177e4 29#include "xfs_ialloc_btree.h"
1da177e4
LT
30#include "xfs_dinode.h"
31#include "xfs_inode.h"
84803fb7 32#include "xfs_inode_item.h"
a844f451 33#include "xfs_btree.h"
1da177e4 34#include "xfs_bmap.h"
1da177e4
LT
35#include "xfs_rtalloc.h"
36#include "xfs_error.h"
37#include "xfs_itable.h"
1da177e4
LT
38#include "xfs_attr.h"
39#include "xfs_buf_item.h"
40#include "xfs_trans_space.h"
41#include "xfs_utils.h"
42#include "xfs_iomap.h"
0b1b213f 43#include "xfs_trace.h"
1da177e4 44
1da177e4
LT
45
46#define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
47 << mp->m_writeio_log)
1da177e4
LT
48#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
49
dd9f438e
NS
50STATIC int
51xfs_iomap_eof_align_last_fsb(
52 xfs_mount_t *mp,
541d7d3c 53 xfs_inode_t *ip,
dd9f438e
NS
54 xfs_extlen_t extsize,
55 xfs_fileoff_t *last_fsb)
56{
57 xfs_fileoff_t new_last_fsb = 0;
bf322d98 58 xfs_extlen_t align = 0;
dd9f438e
NS
59 int eof, error;
60
bf322d98
CH
61 if (!XFS_IS_REALTIME_INODE(ip)) {
62 /*
63 * Round up the allocation request to a stripe unit
64 * (m_dalign) boundary if the file size is >= stripe unit
65 * size, and we are allocating past the allocation eof.
66 *
67 * If mounted with the "-o swalloc" option the alignment is
68 * increased from the strip unit size to the stripe width.
69 */
70 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
71 align = mp->m_swidth;
72 else if (mp->m_dalign)
73 align = mp->m_dalign;
74
ce7ae151 75 if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
bf322d98
CH
76 new_last_fsb = roundup_64(*last_fsb, align);
77 }
dd9f438e
NS
78
79 /*
80 * Always round up the allocation request to an extent boundary
81 * (when file on a real-time subvolume or has di_extsize hint).
82 */
83 if (extsize) {
84 if (new_last_fsb)
85 align = roundup_64(new_last_fsb, extsize);
86 else
87 align = extsize;
88 new_last_fsb = roundup_64(*last_fsb, align);
89 }
90
91 if (new_last_fsb) {
541d7d3c 92 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
dd9f438e
NS
93 if (error)
94 return error;
95 if (eof)
96 *last_fsb = new_last_fsb;
97 }
98 return 0;
99}
100
572d95f4 101STATIC int
6d4a8ecb 102xfs_alert_fsblock_zero(
572d95f4
NS
103 xfs_inode_t *ip,
104 xfs_bmbt_irec_t *imap)
105{
6a19d939 106 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
572d95f4
NS
107 "Access to block zero in inode %llu "
108 "start_block: %llx start_off: %llx "
109 "blkcnt: %llx extent-state: %x\n",
110 (unsigned long long)ip->i_ino,
111 (unsigned long long)imap->br_startblock,
112 (unsigned long long)imap->br_startoff,
113 (unsigned long long)imap->br_blockcount,
114 imap->br_state);
115 return EFSCORRUPTED;
116}
117
a206c817 118int
1da177e4
LT
119xfs_iomap_write_direct(
120 xfs_inode_t *ip,
f403b7f4 121 xfs_off_t offset,
1da177e4 122 size_t count,
3070451e 123 xfs_bmbt_irec_t *imap,
405f8042 124 int nmaps)
1da177e4
LT
125{
126 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
127 xfs_fileoff_t offset_fsb;
128 xfs_fileoff_t last_fsb;
dd9f438e 129 xfs_filblks_t count_fsb, resaligned;
1da177e4 130 xfs_fsblock_t firstfsb;
dd9f438e 131 xfs_extlen_t extsz, temp;
0116d935 132 int nimaps;
1da177e4 133 int bmapi_flag;
06d10dd9 134 int quota_flag;
1da177e4
LT
135 int rt;
136 xfs_trans_t *tp;
1da177e4 137 xfs_bmap_free_t free_list;
dd9f438e 138 uint qblocks, resblks, resrtextents;
1da177e4 139 int committed;
dd9f438e 140 int error;
1da177e4 141
507630b2 142 error = xfs_qm_dqattach(ip, 0);
1da177e4
LT
143 if (error)
144 return XFS_ERROR(error);
145
dd9f438e 146 rt = XFS_IS_REALTIME_INODE(ip);
957d0ebe 147 extsz = xfs_get_extsz_hint(ip);
1da177e4 148
957d0ebe
DC
149 offset_fsb = XFS_B_TO_FSBT(mp, offset);
150 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
ce7ae151 151 if ((offset + count) > XFS_ISIZE(ip)) {
9f6c92b9 152 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
dd9f438e 153 if (error)
507630b2 154 return XFS_ERROR(error);
1da177e4 155 } else {
405f8042 156 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
dd9f438e 157 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
3070451e
CH
158 imap->br_blockcount +
159 imap->br_startoff);
1da177e4 160 }
dd9f438e
NS
161 count_fsb = last_fsb - offset_fsb;
162 ASSERT(count_fsb > 0);
163
164 resaligned = count_fsb;
165 if (unlikely(extsz)) {
166 if ((temp = do_mod(offset_fsb, extsz)))
167 resaligned += temp;
168 if ((temp = do_mod(resaligned, extsz)))
169 resaligned += extsz - temp;
170 }
171
172 if (unlikely(rt)) {
173 resrtextents = qblocks = resaligned;
174 resrtextents /= mp->m_sb.sb_rextsize;
84e1e99f
DC
175 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
176 quota_flag = XFS_QMOPT_RES_RTBLKS;
177 } else {
178 resrtextents = 0;
dd9f438e 179 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
84e1e99f
DC
180 quota_flag = XFS_QMOPT_RES_REGBLKS;
181 }
1da177e4
LT
182
183 /*
06d10dd9 184 * Allocate and setup the transaction
1da177e4 185 */
1da177e4 186 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1da177e4 187 error = xfs_trans_reserve(tp, resblks,
d52b44d0 188 XFS_WRITE_LOG_RES(mp), resrtextents,
1da177e4
LT
189 XFS_TRANS_PERM_LOG_RES,
190 XFS_WRITE_LOG_COUNT);
1da177e4 191 /*
06d10dd9 192 * Check for running out of space, note: need lock to return
1da177e4 193 */
507630b2 194 if (error) {
1da177e4 195 xfs_trans_cancel(tp, 0);
507630b2
DC
196 return XFS_ERROR(error);
197 }
198
1da177e4 199 xfs_ilock(ip, XFS_ILOCK_EXCL);
1da177e4 200
7d095257 201 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
dd9f438e 202 if (error)
507630b2 203 goto out_trans_cancel;
1da177e4 204
ddc3415a 205 xfs_trans_ijoin(tp, ip, 0);
1da177e4 206
c0dc7828 207 bmapi_flag = 0;
ce7ae151 208 if (offset < XFS_ISIZE(ip) || extsz)
1da177e4
LT
209 bmapi_flag |= XFS_BMAPI_PREALLOC;
210
211 /*
3070451e
CH
212 * From this point onwards we overwrite the imap pointer that the
213 * caller gave to us.
1da177e4 214 */
9d87c319 215 xfs_bmap_init(&free_list, &firstfsb);
06d10dd9 216 nimaps = 1;
c0dc7828
DC
217 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
218 &firstfsb, 0, imap, &nimaps, &free_list);
06d10dd9 219 if (error)
507630b2 220 goto out_bmap_cancel;
1da177e4
LT
221
222 /*
06d10dd9 223 * Complete the transaction
1da177e4 224 */
f7c99b6f 225 error = xfs_bmap_finish(&tp, &free_list, &committed);
06d10dd9 226 if (error)
507630b2 227 goto out_bmap_cancel;
1c72bf90 228 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
06d10dd9 229 if (error)
507630b2 230 goto out_unlock;
1da177e4 231
06d10dd9
NS
232 /*
233 * Copy any maps to caller's array and return any error.
234 */
1da177e4 235 if (nimaps == 0) {
507630b2
DC
236 error = XFS_ERROR(ENOSPC);
237 goto out_unlock;
572d95f4
NS
238 }
239
507630b2 240 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 241 error = xfs_alert_fsblock_zero(ip, imap);
1da177e4 242
507630b2
DC
243out_unlock:
244 xfs_iunlock(ip, XFS_ILOCK_EXCL);
245 return error;
1da177e4 246
507630b2 247out_bmap_cancel:
1da177e4 248 xfs_bmap_cancel(&free_list);
ea562ed6 249 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
507630b2 250out_trans_cancel:
1da177e4 251 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
507630b2 252 goto out_unlock;
1da177e4
LT
253}
254
dd9f438e 255/*
8de2bf93
DC
256 * If the caller is doing a write at the end of the file, then extend the
257 * allocation out to the file system's write iosize. We clean up any extra
258 * space left over when the file is closed in xfs_inactive().
055388a3
DC
259 *
260 * If we find we already have delalloc preallocation beyond EOF, don't do more
261 * preallocation as it it not needed.
dd9f438e
NS
262 */
263STATIC int
264xfs_iomap_eof_want_preallocate(
265 xfs_mount_t *mp,
541d7d3c 266 xfs_inode_t *ip,
dd9f438e
NS
267 xfs_off_t offset,
268 size_t count,
dd9f438e
NS
269 xfs_bmbt_irec_t *imap,
270 int nimaps,
271 int *prealloc)
272{
273 xfs_fileoff_t start_fsb;
274 xfs_filblks_t count_fsb;
275 xfs_fsblock_t firstblock;
276 int n, error, imaps;
055388a3 277 int found_delalloc = 0;
dd9f438e
NS
278
279 *prealloc = 0;
ce7ae151 280 if (offset + count <= XFS_ISIZE(ip))
dd9f438e
NS
281 return 0;
282
283 /*
284 * If there are any real blocks past eof, then don't
285 * do any speculative allocation.
286 */
287 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
32972383 288 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
dd9f438e
NS
289 while (count_fsb > 0) {
290 imaps = nimaps;
3ddb8fa9 291 firstblock = NULLFSBLOCK;
5c8ed202
DC
292 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
293 0);
dd9f438e
NS
294 if (error)
295 return error;
296 for (n = 0; n < imaps; n++) {
297 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
298 (imap[n].br_startblock != DELAYSTARTBLOCK))
299 return 0;
300 start_fsb += imap[n].br_blockcount;
301 count_fsb -= imap[n].br_blockcount;
055388a3
DC
302
303 if (imap[n].br_startblock == DELAYSTARTBLOCK)
304 found_delalloc = 1;
dd9f438e
NS
305 }
306 }
055388a3
DC
307 if (!found_delalloc)
308 *prealloc = 1;
dd9f438e
NS
309 return 0;
310}
311
055388a3
DC
312/*
313 * If we don't have a user specified preallocation size, dynamically increase
314 * the preallocation size as the size of the file grows. Cap the maximum size
315 * at a single extent or less if the filesystem is near full. The closer the
316 * filesystem is to full, the smaller the maximum prealocation.
317 */
318STATIC xfs_fsblock_t
319xfs_iomap_prealloc_size(
320 struct xfs_mount *mp,
321 struct xfs_inode *ip)
322{
323 xfs_fsblock_t alloc_blocks = 0;
324
325 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
326 int shift = 0;
327 int64_t freesp;
328
b8fc8263
DC
329 /*
330 * rounddown_pow_of_two() returns an undefined result
331 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
332 * ensure we always pass in a non-zero value.
333 */
ce7ae151 334 alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
055388a3
DC
335 alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
336 rounddown_pow_of_two(alloc_blocks));
337
338 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
339 freesp = mp->m_sb.sb_fdblocks;
340 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
341 shift = 2;
342 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
343 shift++;
344 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
345 shift++;
346 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
347 shift++;
348 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
349 shift++;
350 }
351 if (shift)
352 alloc_blocks >>= shift;
353 }
354
355 if (alloc_blocks < mp->m_writeio_blocks)
356 alloc_blocks = mp->m_writeio_blocks;
357
358 return alloc_blocks;
359}
360
a206c817 361int
1da177e4
LT
362xfs_iomap_write_delay(
363 xfs_inode_t *ip,
f403b7f4 364 xfs_off_t offset,
1da177e4 365 size_t count,
405f8042 366 xfs_bmbt_irec_t *ret_imap)
1da177e4
LT
367{
368 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
369 xfs_fileoff_t offset_fsb;
370 xfs_fileoff_t last_fsb;
dd9f438e
NS
371 xfs_off_t aligned_offset;
372 xfs_fileoff_t ioalign;
dd9f438e 373 xfs_extlen_t extsz;
1da177e4 374 int nimaps;
1da177e4 375 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
9aa05000 376 int prealloc;
dd9f438e 377 int error;
1da177e4 378
579aa9ca 379 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
1da177e4
LT
380
381 /*
382 * Make sure that the dquots are there. This doesn't hold
383 * the ilock across a disk read.
384 */
7d095257 385 error = xfs_qm_dqattach_locked(ip, 0);
1da177e4
LT
386 if (error)
387 return XFS_ERROR(error);
388
957d0ebe 389 extsz = xfs_get_extsz_hint(ip);
dd9f438e
NS
390 offset_fsb = XFS_B_TO_FSBT(mp, offset);
391
055388a3 392
9f6c92b9 393 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
405f8042 394 imap, XFS_WRITE_IMAPS, &prealloc);
dd9f438e
NS
395 if (error)
396 return error;
1da177e4 397
8de2bf93 398retry:
dd9f438e 399 if (prealloc) {
055388a3
DC
400 xfs_fsblock_t alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
401
1da177e4
LT
402 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
403 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
055388a3 404 last_fsb = ioalign + alloc_blocks;
dd9f438e
NS
405 } else {
406 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
1da177e4 407 }
1da177e4 408
dd9f438e 409 if (prealloc || extsz) {
9f6c92b9 410 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
dd9f438e 411 if (error)
1da177e4 412 return error;
1da177e4 413 }
dd9f438e 414
3ed9116e
DC
415 /*
416 * Make sure preallocation does not create extents beyond the range we
417 * actually support in this filesystem.
418 */
d2c28191
DC
419 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
420 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
3ed9116e
DC
421
422 ASSERT(last_fsb > offset_fsb);
423
dd9f438e 424 nimaps = XFS_WRITE_IMAPS;
4403280a
CH
425 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
426 imap, &nimaps, XFS_BMAPI_ENTIRE);
055388a3
DC
427 switch (error) {
428 case 0:
429 case ENOSPC:
430 case EDQUOT:
431 break;
432 default:
1da177e4 433 return XFS_ERROR(error);
055388a3 434 }
dd9f438e 435
1da177e4 436 /*
9aa05000 437 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
055388a3 438 * without EOF preallocation.
1da177e4
LT
439 */
440 if (nimaps == 0) {
0b1b213f 441 trace_xfs_delalloc_enospc(ip, offset, count);
9aa05000
DC
442 if (prealloc) {
443 prealloc = 0;
444 error = 0;
445 goto retry;
055388a3 446 }
9aa05000 447 return XFS_ERROR(error ? error : ENOSPC);
1da177e4
LT
448 }
449
86c4d623 450 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 451 return xfs_alert_fsblock_zero(ip, &imap[0]);
dd9f438e
NS
452
453 *ret_imap = imap[0];
1da177e4
LT
454 return 0;
455}
456
457/*
458 * Pass in a delayed allocate extent, convert it to real extents;
459 * return to the caller the extent we create which maps on top of
460 * the originating callers request.
461 *
462 * Called without a lock on the inode.
e4143a1c
DC
463 *
464 * We no longer bother to look at the incoming map - all we have to
465 * guarantee is that whatever we allocate fills the required range.
1da177e4 466 */
a206c817 467int
1da177e4
LT
468xfs_iomap_write_allocate(
469 xfs_inode_t *ip,
f403b7f4 470 xfs_off_t offset,
24e17b5f 471 size_t count,
405f8042 472 xfs_bmbt_irec_t *imap)
1da177e4
LT
473{
474 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
475 xfs_fileoff_t offset_fsb, last_block;
476 xfs_fileoff_t end_fsb, map_start_fsb;
477 xfs_fsblock_t first_block;
478 xfs_bmap_free_t free_list;
479 xfs_filblks_t count_fsb;
1da177e4 480 xfs_trans_t *tp;
e4143a1c 481 int nimaps, committed;
1da177e4
LT
482 int error = 0;
483 int nres;
484
1da177e4
LT
485 /*
486 * Make sure that the dquots are there.
487 */
7d095257
CH
488 error = xfs_qm_dqattach(ip, 0);
489 if (error)
1da177e4
LT
490 return XFS_ERROR(error);
491
24e17b5f 492 offset_fsb = XFS_B_TO_FSBT(mp, offset);
3070451e
CH
493 count_fsb = imap->br_blockcount;
494 map_start_fsb = imap->br_startoff;
1da177e4
LT
495
496 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
497
498 while (count_fsb != 0) {
499 /*
500 * Set up a transaction with which to allocate the
501 * backing store for the file. Do allocations in a
502 * loop until we get some space in the range we are
503 * interested in. The other space that might be allocated
504 * is in the delayed allocation extent on which we sit
505 * but before our buffer starts.
506 */
507
508 nimaps = 0;
509 while (nimaps == 0) {
510 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
84e1e99f 511 tp->t_flags |= XFS_TRANS_RESERVE;
1da177e4
LT
512 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
513 error = xfs_trans_reserve(tp, nres,
514 XFS_WRITE_LOG_RES(mp),
515 0, XFS_TRANS_PERM_LOG_RES,
516 XFS_WRITE_LOG_COUNT);
1da177e4
LT
517 if (error) {
518 xfs_trans_cancel(tp, 0);
519 return XFS_ERROR(error);
520 }
521 xfs_ilock(ip, XFS_ILOCK_EXCL);
ddc3415a 522 xfs_trans_ijoin(tp, ip, 0);
1da177e4 523
9d87c319 524 xfs_bmap_init(&free_list, &first_block);
1da177e4 525
1da177e4 526 /*
e4143a1c
DC
527 * it is possible that the extents have changed since
528 * we did the read call as we dropped the ilock for a
529 * while. We have to be careful about truncates or hole
530 * punchs here - we are not allowed to allocate
531 * non-delalloc blocks here.
532 *
533 * The only protection against truncation is the pages
534 * for the range we are being asked to convert are
535 * locked and hence a truncate will block on them
536 * first.
537 *
538 * As a result, if we go beyond the range we really
539 * need and hit an delalloc extent boundary followed by
540 * a hole while we have excess blocks in the map, we
541 * will fill the hole incorrectly and overrun the
542 * transaction reservation.
543 *
544 * Using a single map prevents this as we are forced to
545 * check each map we look for overlap with the desired
546 * range and abort as soon as we find it. Also, given
547 * that we only return a single map, having one beyond
548 * what we can return is probably a bit silly.
549 *
550 * We also need to check that we don't go beyond EOF;
551 * this is a truncate optimisation as a truncate sets
552 * the new file size before block on the pages we
553 * currently have locked under writeback. Because they
554 * are about to be tossed, we don't need to write them
555 * back....
1da177e4 556 */
e4143a1c 557 nimaps = 1;
ce7ae151 558 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
7c9ef85c
DC
559 error = xfs_bmap_last_offset(NULL, ip, &last_block,
560 XFS_DATA_FORK);
561 if (error)
562 goto trans_cancel;
563
1da177e4
LT
564 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
565 if ((map_start_fsb + count_fsb) > last_block) {
566 count_fsb = last_block - map_start_fsb;
567 if (count_fsb == 0) {
568 error = EAGAIN;
569 goto trans_cancel;
570 }
571 }
572
3070451e 573 /*
3070451e
CH
574 * From this point onwards we overwrite the imap
575 * pointer that the caller gave to us.
576 */
c0dc7828
DC
577 error = xfs_bmapi_write(tp, ip, map_start_fsb,
578 count_fsb, 0, &first_block, 1,
579 imap, &nimaps, &free_list);
1da177e4
LT
580 if (error)
581 goto trans_cancel;
582
f7c99b6f 583 error = xfs_bmap_finish(&tp, &free_list, &committed);
1da177e4
LT
584 if (error)
585 goto trans_cancel;
586
1c72bf90 587 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1da177e4
LT
588 if (error)
589 goto error0;
590
591 xfs_iunlock(ip, XFS_ILOCK_EXCL);
592 }
593
594 /*
595 * See if we were able to allocate an extent that
596 * covers at least part of the callers request
597 */
3070451e 598 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 599 return xfs_alert_fsblock_zero(ip, imap);
86c4d623 600
3070451e
CH
601 if ((offset_fsb >= imap->br_startoff) &&
602 (offset_fsb < (imap->br_startoff +
603 imap->br_blockcount))) {
e4143a1c
DC
604 XFS_STATS_INC(xs_xstrat_quick);
605 return 0;
1da177e4
LT
606 }
607
e4143a1c
DC
608 /*
609 * So far we have not mapped the requested part of the
1da177e4
LT
610 * file, just surrounding data, try again.
611 */
3070451e
CH
612 count_fsb -= imap->br_blockcount;
613 map_start_fsb = imap->br_startoff + imap->br_blockcount;
1da177e4
LT
614 }
615
616trans_cancel:
617 xfs_bmap_cancel(&free_list);
618 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
619error0:
620 xfs_iunlock(ip, XFS_ILOCK_EXCL);
621 return XFS_ERROR(error);
622}
623
624int
625xfs_iomap_write_unwritten(
626 xfs_inode_t *ip,
f403b7f4 627 xfs_off_t offset,
1da177e4
LT
628 size_t count)
629{
630 xfs_mount_t *mp = ip->i_mount;
1da177e4
LT
631 xfs_fileoff_t offset_fsb;
632 xfs_filblks_t count_fsb;
633 xfs_filblks_t numblks_fsb;
dd9f438e
NS
634 xfs_fsblock_t firstfsb;
635 int nimaps;
636 xfs_trans_t *tp;
637 xfs_bmbt_irec_t imap;
638 xfs_bmap_free_t free_list;
84803fb7 639 xfs_fsize_t i_size;
dd9f438e 640 uint resblks;
1da177e4
LT
641 int committed;
642 int error;
1da177e4 643
0b1b213f 644 trace_xfs_unwritten_convert(ip, offset, count);
1da177e4
LT
645
646 offset_fsb = XFS_B_TO_FSBT(mp, offset);
647 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
648 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
649
4ddd8bb1
LM
650 /*
651 * Reserve enough blocks in this transaction for two complete extent
652 * btree splits. We may be converting the middle part of an unwritten
653 * extent and in this case we will insert two new extents in the btree
654 * each of which could cause a full split.
655 *
656 * This reservation amount will be used in the first call to
657 * xfs_bmbt_split() to select an AG with enough space to satisfy the
658 * rest of the operation.
659 */
dd9f438e 660 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
1da177e4 661
dd9f438e 662 do {
1da177e4
LT
663 /*
664 * set up a transaction to convert the range of extents
665 * from unwritten to real. Do allocations in a loop until
666 * we have covered the range passed in.
80641dc6
CH
667 *
668 * Note that we open code the transaction allocation here
669 * to pass KM_NOFS--we can't risk to recursing back into
670 * the filesystem here as we might be asked to write out
671 * the same inode that we complete here and might deadlock
672 * on the iolock.
1da177e4 673 */
d9457dc0 674 sb_start_intwrite(mp->m_super);
80641dc6 675 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
d9457dc0 676 tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT;
dd9f438e 677 error = xfs_trans_reserve(tp, resblks,
1da177e4
LT
678 XFS_WRITE_LOG_RES(mp), 0,
679 XFS_TRANS_PERM_LOG_RES,
680 XFS_WRITE_LOG_COUNT);
681 if (error) {
682 xfs_trans_cancel(tp, 0);
572d95f4 683 return XFS_ERROR(error);
1da177e4
LT
684 }
685
686 xfs_ilock(ip, XFS_ILOCK_EXCL);
ddc3415a 687 xfs_trans_ijoin(tp, ip, 0);
1da177e4
LT
688
689 /*
690 * Modify the unwritten extent state of the buffer.
691 */
9d87c319 692 xfs_bmap_init(&free_list, &firstfsb);
1da177e4 693 nimaps = 1;
c0dc7828
DC
694 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
695 XFS_BMAPI_CONVERT, &firstfsb,
b4e9181e 696 1, &imap, &nimaps, &free_list);
1da177e4
LT
697 if (error)
698 goto error_on_bmapi_transaction;
699
84803fb7
CH
700 /*
701 * Log the updated inode size as we go. We have to be careful
702 * to only log it up to the actual write offset if it is
703 * halfway into a block.
704 */
705 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
706 if (i_size > offset + count)
707 i_size = offset + count;
708
709 i_size = xfs_new_eof(ip, i_size);
710 if (i_size) {
711 ip->i_d.di_size = i_size;
712 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
713 }
714
715 error = xfs_bmap_finish(&tp, &free_list, &committed);
1da177e4
LT
716 if (error)
717 goto error_on_bmapi_transaction;
718
1c72bf90 719 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1da177e4
LT
720 xfs_iunlock(ip, XFS_ILOCK_EXCL);
721 if (error)
572d95f4
NS
722 return XFS_ERROR(error);
723
86c4d623 724 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
6d4a8ecb 725 return xfs_alert_fsblock_zero(ip, &imap);
1da177e4
LT
726
727 if ((numblks_fsb = imap.br_blockcount) == 0) {
728 /*
729 * The numblks_fsb value should always get
730 * smaller, otherwise the loop is stuck.
731 */
732 ASSERT(imap.br_blockcount);
733 break;
734 }
735 offset_fsb += numblks_fsb;
736 count_fsb -= numblks_fsb;
737 } while (count_fsb > 0);
738
739 return 0;
740
741error_on_bmapi_transaction:
742 xfs_bmap_cancel(&free_list);
743 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
744 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1da177e4
LT
745 return XFS_ERROR(error);
746}