Commit | Line | Data |
---|---|---|
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" |
27b52867 | 44 | #include "xfs_icache.h" |
1da177e4 | 45 | |
1da177e4 LT |
46 | |
47 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
48 | << mp->m_writeio_log) | |
1da177e4 LT |
49 | #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP |
50 | ||
dd9f438e NS |
51 | STATIC int |
52 | xfs_iomap_eof_align_last_fsb( | |
53 | xfs_mount_t *mp, | |
541d7d3c | 54 | xfs_inode_t *ip, |
dd9f438e NS |
55 | xfs_extlen_t extsize, |
56 | xfs_fileoff_t *last_fsb) | |
57 | { | |
58 | xfs_fileoff_t new_last_fsb = 0; | |
bf322d98 | 59 | xfs_extlen_t align = 0; |
dd9f438e NS |
60 | int eof, error; |
61 | ||
bf322d98 CH |
62 | if (!XFS_IS_REALTIME_INODE(ip)) { |
63 | /* | |
64 | * Round up the allocation request to a stripe unit | |
65 | * (m_dalign) boundary if the file size is >= stripe unit | |
66 | * size, and we are allocating past the allocation eof. | |
67 | * | |
68 | * If mounted with the "-o swalloc" option the alignment is | |
69 | * increased from the strip unit size to the stripe width. | |
70 | */ | |
71 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
72 | align = mp->m_swidth; | |
73 | else if (mp->m_dalign) | |
74 | align = mp->m_dalign; | |
75 | ||
ce7ae151 | 76 | if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align)) |
bf322d98 CH |
77 | new_last_fsb = roundup_64(*last_fsb, align); |
78 | } | |
dd9f438e NS |
79 | |
80 | /* | |
81 | * Always round up the allocation request to an extent boundary | |
82 | * (when file on a real-time subvolume or has di_extsize hint). | |
83 | */ | |
84 | if (extsize) { | |
85 | if (new_last_fsb) | |
86 | align = roundup_64(new_last_fsb, extsize); | |
87 | else | |
88 | align = extsize; | |
89 | new_last_fsb = roundup_64(*last_fsb, align); | |
90 | } | |
91 | ||
92 | if (new_last_fsb) { | |
541d7d3c | 93 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
94 | if (error) |
95 | return error; | |
96 | if (eof) | |
97 | *last_fsb = new_last_fsb; | |
98 | } | |
99 | return 0; | |
100 | } | |
101 | ||
572d95f4 | 102 | STATIC int |
6d4a8ecb | 103 | xfs_alert_fsblock_zero( |
572d95f4 NS |
104 | xfs_inode_t *ip, |
105 | xfs_bmbt_irec_t *imap) | |
106 | { | |
6a19d939 | 107 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
108 | "Access to block zero in inode %llu " |
109 | "start_block: %llx start_off: %llx " | |
110 | "blkcnt: %llx extent-state: %x\n", | |
111 | (unsigned long long)ip->i_ino, | |
112 | (unsigned long long)imap->br_startblock, | |
113 | (unsigned long long)imap->br_startoff, | |
114 | (unsigned long long)imap->br_blockcount, | |
115 | imap->br_state); | |
116 | return EFSCORRUPTED; | |
117 | } | |
118 | ||
a206c817 | 119 | int |
1da177e4 LT |
120 | xfs_iomap_write_direct( |
121 | xfs_inode_t *ip, | |
f403b7f4 | 122 | xfs_off_t offset, |
1da177e4 | 123 | size_t count, |
3070451e | 124 | xfs_bmbt_irec_t *imap, |
405f8042 | 125 | int nmaps) |
1da177e4 LT |
126 | { |
127 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
128 | xfs_fileoff_t offset_fsb; |
129 | xfs_fileoff_t last_fsb; | |
dd9f438e | 130 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 131 | xfs_fsblock_t firstfsb; |
dd9f438e | 132 | xfs_extlen_t extsz, temp; |
0116d935 | 133 | int nimaps; |
1da177e4 | 134 | int bmapi_flag; |
06d10dd9 | 135 | int quota_flag; |
1da177e4 LT |
136 | int rt; |
137 | xfs_trans_t *tp; | |
1da177e4 | 138 | xfs_bmap_free_t free_list; |
dd9f438e | 139 | uint qblocks, resblks, resrtextents; |
1da177e4 | 140 | int committed; |
dd9f438e | 141 | int error; |
1da177e4 | 142 | |
507630b2 | 143 | error = xfs_qm_dqattach(ip, 0); |
1da177e4 LT |
144 | if (error) |
145 | return XFS_ERROR(error); | |
146 | ||
dd9f438e | 147 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 148 | extsz = xfs_get_extsz_hint(ip); |
1da177e4 | 149 | |
957d0ebe DC |
150 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
151 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 152 | if ((offset + count) > XFS_ISIZE(ip)) { |
9f6c92b9 | 153 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 154 | if (error) |
507630b2 | 155 | return XFS_ERROR(error); |
1da177e4 | 156 | } else { |
405f8042 | 157 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 158 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
159 | imap->br_blockcount + |
160 | imap->br_startoff); | |
1da177e4 | 161 | } |
dd9f438e NS |
162 | count_fsb = last_fsb - offset_fsb; |
163 | ASSERT(count_fsb > 0); | |
164 | ||
165 | resaligned = count_fsb; | |
166 | if (unlikely(extsz)) { | |
167 | if ((temp = do_mod(offset_fsb, extsz))) | |
168 | resaligned += temp; | |
169 | if ((temp = do_mod(resaligned, extsz))) | |
170 | resaligned += extsz - temp; | |
171 | } | |
172 | ||
173 | if (unlikely(rt)) { | |
174 | resrtextents = qblocks = resaligned; | |
175 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
176 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
177 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
178 | } else { | |
179 | resrtextents = 0; | |
dd9f438e | 180 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
181 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
182 | } | |
1da177e4 LT |
183 | |
184 | /* | |
06d10dd9 | 185 | * Allocate and setup the transaction |
1da177e4 | 186 | */ |
1da177e4 | 187 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); |
1da177e4 | 188 | error = xfs_trans_reserve(tp, resblks, |
d52b44d0 | 189 | XFS_WRITE_LOG_RES(mp), resrtextents, |
1da177e4 LT |
190 | XFS_TRANS_PERM_LOG_RES, |
191 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 192 | /* |
06d10dd9 | 193 | * Check for running out of space, note: need lock to return |
1da177e4 | 194 | */ |
507630b2 | 195 | if (error) { |
1da177e4 | 196 | xfs_trans_cancel(tp, 0); |
507630b2 DC |
197 | return XFS_ERROR(error); |
198 | } | |
199 | ||
1da177e4 | 200 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 201 | |
7d095257 | 202 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 203 | if (error) |
507630b2 | 204 | goto out_trans_cancel; |
1da177e4 | 205 | |
ddc3415a | 206 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 207 | |
c0dc7828 | 208 | bmapi_flag = 0; |
ce7ae151 | 209 | if (offset < XFS_ISIZE(ip) || extsz) |
1da177e4 LT |
210 | bmapi_flag |= XFS_BMAPI_PREALLOC; |
211 | ||
212 | /* | |
3070451e CH |
213 | * From this point onwards we overwrite the imap pointer that the |
214 | * caller gave to us. | |
1da177e4 | 215 | */ |
9d87c319 | 216 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 217 | nimaps = 1; |
c0dc7828 DC |
218 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag, |
219 | &firstfsb, 0, imap, &nimaps, &free_list); | |
06d10dd9 | 220 | if (error) |
507630b2 | 221 | goto out_bmap_cancel; |
1da177e4 LT |
222 | |
223 | /* | |
06d10dd9 | 224 | * Complete the transaction |
1da177e4 | 225 | */ |
f7c99b6f | 226 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
06d10dd9 | 227 | if (error) |
507630b2 | 228 | goto out_bmap_cancel; |
1c72bf90 | 229 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
06d10dd9 | 230 | if (error) |
507630b2 | 231 | goto out_unlock; |
1da177e4 | 232 | |
06d10dd9 NS |
233 | /* |
234 | * Copy any maps to caller's array and return any error. | |
235 | */ | |
1da177e4 | 236 | if (nimaps == 0) { |
507630b2 DC |
237 | error = XFS_ERROR(ENOSPC); |
238 | goto out_unlock; | |
572d95f4 NS |
239 | } |
240 | ||
507630b2 | 241 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 242 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 243 | |
507630b2 DC |
244 | out_unlock: |
245 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
246 | return error; | |
1da177e4 | 247 | |
507630b2 | 248 | out_bmap_cancel: |
1da177e4 | 249 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 250 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 251 | out_trans_cancel: |
1da177e4 | 252 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
507630b2 | 253 | goto out_unlock; |
1da177e4 LT |
254 | } |
255 | ||
dd9f438e | 256 | /* |
8de2bf93 DC |
257 | * If the caller is doing a write at the end of the file, then extend the |
258 | * allocation out to the file system's write iosize. We clean up any extra | |
259 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
260 | * |
261 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
262 | * preallocation as it it not needed. | |
dd9f438e NS |
263 | */ |
264 | STATIC int | |
265 | xfs_iomap_eof_want_preallocate( | |
266 | xfs_mount_t *mp, | |
541d7d3c | 267 | xfs_inode_t *ip, |
dd9f438e NS |
268 | xfs_off_t offset, |
269 | size_t count, | |
dd9f438e NS |
270 | xfs_bmbt_irec_t *imap, |
271 | int nimaps, | |
272 | int *prealloc) | |
273 | { | |
274 | xfs_fileoff_t start_fsb; | |
275 | xfs_filblks_t count_fsb; | |
276 | xfs_fsblock_t firstblock; | |
277 | int n, error, imaps; | |
055388a3 | 278 | int found_delalloc = 0; |
dd9f438e NS |
279 | |
280 | *prealloc = 0; | |
ce7ae151 | 281 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
282 | return 0; |
283 | ||
284 | /* | |
285 | * If there are any real blocks past eof, then don't | |
286 | * do any speculative allocation. | |
287 | */ | |
288 | start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1))); | |
32972383 | 289 | count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
dd9f438e NS |
290 | while (count_fsb > 0) { |
291 | imaps = nimaps; | |
3ddb8fa9 | 292 | firstblock = NULLFSBLOCK; |
5c8ed202 DC |
293 | error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps, |
294 | 0); | |
dd9f438e NS |
295 | if (error) |
296 | return error; | |
297 | for (n = 0; n < imaps; n++) { | |
298 | if ((imap[n].br_startblock != HOLESTARTBLOCK) && | |
299 | (imap[n].br_startblock != DELAYSTARTBLOCK)) | |
300 | return 0; | |
301 | start_fsb += imap[n].br_blockcount; | |
302 | count_fsb -= imap[n].br_blockcount; | |
055388a3 DC |
303 | |
304 | if (imap[n].br_startblock == DELAYSTARTBLOCK) | |
305 | found_delalloc = 1; | |
dd9f438e NS |
306 | } |
307 | } | |
055388a3 DC |
308 | if (!found_delalloc) |
309 | *prealloc = 1; | |
dd9f438e NS |
310 | return 0; |
311 | } | |
312 | ||
a1e16c26 DC |
313 | /* |
314 | * Determine the initial size of the preallocation. We are beyond the current | |
315 | * EOF here, but we need to take into account whether this is a sparse write or | |
316 | * an extending write when determining the preallocation size. Hence we need to | |
317 | * look up the extent that ends at the current write offset and use the result | |
318 | * to determine the preallocation size. | |
319 | * | |
320 | * If the extent is a hole, then preallocation is essentially disabled. | |
321 | * Otherwise we take the size of the preceeding data extent as the basis for the | |
322 | * preallocation size. If the size of the extent is greater than half the | |
323 | * maximum extent length, then use the current offset as the basis. This ensures | |
324 | * that for large files the preallocation size always extends to MAXEXTLEN | |
325 | * rather than falling short due to things like stripe unit/width alignment of | |
326 | * real extents. | |
327 | */ | |
e8108ced | 328 | STATIC xfs_fsblock_t |
a1e16c26 DC |
329 | xfs_iomap_eof_prealloc_initial_size( |
330 | struct xfs_mount *mp, | |
331 | struct xfs_inode *ip, | |
332 | xfs_off_t offset, | |
333 | xfs_bmbt_irec_t *imap, | |
334 | int nimaps) | |
335 | { | |
336 | xfs_fileoff_t start_fsb; | |
337 | int imaps = 1; | |
338 | int error; | |
339 | ||
340 | ASSERT(nimaps >= imaps); | |
341 | ||
342 | /* if we are using a specific prealloc size, return now */ | |
343 | if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) | |
344 | return 0; | |
345 | ||
346 | /* | |
347 | * As we write multiple pages, the offset will always align to the | |
348 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
349 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
350 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
351 | * point to the block prior to the EOF block and not the hole that maps | |
352 | * directly at @offset. | |
353 | */ | |
354 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
355 | if (start_fsb) | |
356 | start_fsb--; | |
357 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
358 | if (error) | |
359 | return 0; | |
360 | ||
361 | ASSERT(imaps == 1); | |
362 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
363 | return 0; | |
364 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 365 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
366 | return XFS_B_TO_FSB(mp, offset); |
367 | } | |
368 | ||
055388a3 DC |
369 | /* |
370 | * If we don't have a user specified preallocation size, dynamically increase | |
371 | * the preallocation size as the size of the file grows. Cap the maximum size | |
372 | * at a single extent or less if the filesystem is near full. The closer the | |
373 | * filesystem is to full, the smaller the maximum prealocation. | |
374 | */ | |
375 | STATIC xfs_fsblock_t | |
376 | xfs_iomap_prealloc_size( | |
377 | struct xfs_mount *mp, | |
a1e16c26 DC |
378 | struct xfs_inode *ip, |
379 | xfs_off_t offset, | |
380 | struct xfs_bmbt_irec *imap, | |
381 | int nimaps) | |
055388a3 DC |
382 | { |
383 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
384 | int shift = 0; |
385 | int64_t freesp; | |
055388a3 | 386 | |
a1e16c26 DC |
387 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
388 | imap, nimaps); | |
3c58b5f8 BF |
389 | if (!alloc_blocks) |
390 | goto check_writeio; | |
391 | ||
c9bdbdc0 BF |
392 | /* |
393 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
394 | * to the nearest power of two value after throttling. To prevent the | |
395 | * round down from unconditionally reducing the maximum supported prealloc | |
396 | * size, we round up first, apply appropriate throttling, round down and | |
397 | * cap the value to MAXEXTLEN. | |
398 | */ | |
399 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
400 | alloc_blocks); | |
3c58b5f8 BF |
401 | |
402 | xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); | |
403 | freesp = mp->m_sb.sb_fdblocks; | |
404 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { | |
405 | shift = 2; | |
406 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
407 | shift++; | |
408 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
409 | shift++; | |
410 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
411 | shift++; | |
412 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
413 | shift++; | |
055388a3 | 414 | } |
3c58b5f8 BF |
415 | if (shift) |
416 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
417 | /* |
418 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
419 | * alloc_blocks = 0. | |
420 | */ | |
421 | if (alloc_blocks) | |
422 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
423 | if (alloc_blocks > MAXEXTLEN) | |
424 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
425 | |
426 | /* | |
427 | * If we are still trying to allocate more space than is | |
428 | * available, squash the prealloc hard. This can happen if we | |
429 | * have a large file on a small filesystem and the above | |
430 | * lowspace thresholds are smaller than MAXEXTLEN. | |
431 | */ | |
432 | while (alloc_blocks && alloc_blocks >= freesp) | |
433 | alloc_blocks >>= 4; | |
055388a3 | 434 | |
3c58b5f8 | 435 | check_writeio: |
055388a3 DC |
436 | if (alloc_blocks < mp->m_writeio_blocks) |
437 | alloc_blocks = mp->m_writeio_blocks; | |
438 | ||
439 | return alloc_blocks; | |
440 | } | |
441 | ||
a206c817 | 442 | int |
1da177e4 LT |
443 | xfs_iomap_write_delay( |
444 | xfs_inode_t *ip, | |
f403b7f4 | 445 | xfs_off_t offset, |
1da177e4 | 446 | size_t count, |
405f8042 | 447 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
448 | { |
449 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
450 | xfs_fileoff_t offset_fsb; |
451 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
452 | xfs_off_t aligned_offset; |
453 | xfs_fileoff_t ioalign; | |
dd9f438e | 454 | xfs_extlen_t extsz; |
1da177e4 | 455 | int nimaps; |
1da177e4 | 456 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 457 | int prealloc; |
dd9f438e | 458 | int error; |
1da177e4 | 459 | |
579aa9ca | 460 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
461 | |
462 | /* | |
463 | * Make sure that the dquots are there. This doesn't hold | |
464 | * the ilock across a disk read. | |
465 | */ | |
7d095257 | 466 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 LT |
467 | if (error) |
468 | return XFS_ERROR(error); | |
469 | ||
957d0ebe | 470 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
471 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
472 | ||
9f6c92b9 | 473 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 474 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
475 | if (error) |
476 | return error; | |
1da177e4 | 477 | |
8de2bf93 | 478 | retry: |
dd9f438e | 479 | if (prealloc) { |
a1e16c26 DC |
480 | xfs_fsblock_t alloc_blocks; |
481 | ||
482 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
483 | XFS_WRITE_IMAPS); | |
055388a3 | 484 | |
1da177e4 LT |
485 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
486 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 487 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
488 | } else { |
489 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 490 | } |
1da177e4 | 491 | |
dd9f438e | 492 | if (prealloc || extsz) { |
9f6c92b9 | 493 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 494 | if (error) |
1da177e4 | 495 | return error; |
1da177e4 | 496 | } |
dd9f438e | 497 | |
3ed9116e DC |
498 | /* |
499 | * Make sure preallocation does not create extents beyond the range we | |
500 | * actually support in this filesystem. | |
501 | */ | |
d2c28191 DC |
502 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
503 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
504 | |
505 | ASSERT(last_fsb > offset_fsb); | |
506 | ||
dd9f438e | 507 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
508 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
509 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
510 | switch (error) { |
511 | case 0: | |
512 | case ENOSPC: | |
513 | case EDQUOT: | |
514 | break; | |
515 | default: | |
1da177e4 | 516 | return XFS_ERROR(error); |
055388a3 | 517 | } |
dd9f438e | 518 | |
1da177e4 | 519 | /* |
9aa05000 | 520 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 521 | * without EOF preallocation. |
1da177e4 LT |
522 | */ |
523 | if (nimaps == 0) { | |
0b1b213f | 524 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
525 | if (prealloc) { |
526 | prealloc = 0; | |
527 | error = 0; | |
528 | goto retry; | |
055388a3 | 529 | } |
9aa05000 | 530 | return XFS_ERROR(error ? error : ENOSPC); |
1da177e4 LT |
531 | } |
532 | ||
86c4d623 | 533 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 534 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 535 | |
27b52867 BF |
536 | /* |
537 | * Tag the inode as speculatively preallocated so we can reclaim this | |
538 | * space on demand, if necessary. | |
539 | */ | |
540 | if (prealloc) | |
541 | xfs_inode_set_eofblocks_tag(ip); | |
542 | ||
dd9f438e | 543 | *ret_imap = imap[0]; |
1da177e4 LT |
544 | return 0; |
545 | } | |
546 | ||
547 | /* | |
548 | * Pass in a delayed allocate extent, convert it to real extents; | |
549 | * return to the caller the extent we create which maps on top of | |
550 | * the originating callers request. | |
551 | * | |
552 | * Called without a lock on the inode. | |
e4143a1c DC |
553 | * |
554 | * We no longer bother to look at the incoming map - all we have to | |
555 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 556 | */ |
a206c817 | 557 | int |
1da177e4 LT |
558 | xfs_iomap_write_allocate( |
559 | xfs_inode_t *ip, | |
f403b7f4 | 560 | xfs_off_t offset, |
24e17b5f | 561 | size_t count, |
405f8042 | 562 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
563 | { |
564 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
565 | xfs_fileoff_t offset_fsb, last_block; |
566 | xfs_fileoff_t end_fsb, map_start_fsb; | |
567 | xfs_fsblock_t first_block; | |
568 | xfs_bmap_free_t free_list; | |
569 | xfs_filblks_t count_fsb; | |
1da177e4 | 570 | xfs_trans_t *tp; |
e4143a1c | 571 | int nimaps, committed; |
1da177e4 LT |
572 | int error = 0; |
573 | int nres; | |
574 | ||
1da177e4 LT |
575 | /* |
576 | * Make sure that the dquots are there. | |
577 | */ | |
7d095257 CH |
578 | error = xfs_qm_dqattach(ip, 0); |
579 | if (error) | |
1da177e4 LT |
580 | return XFS_ERROR(error); |
581 | ||
24e17b5f | 582 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
583 | count_fsb = imap->br_blockcount; |
584 | map_start_fsb = imap->br_startoff; | |
1da177e4 LT |
585 | |
586 | XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); | |
587 | ||
588 | while (count_fsb != 0) { | |
589 | /* | |
590 | * Set up a transaction with which to allocate the | |
591 | * backing store for the file. Do allocations in a | |
592 | * loop until we get some space in the range we are | |
593 | * interested in. The other space that might be allocated | |
594 | * is in the delayed allocation extent on which we sit | |
595 | * but before our buffer starts. | |
596 | */ | |
597 | ||
598 | nimaps = 0; | |
599 | while (nimaps == 0) { | |
600 | tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE); | |
84e1e99f | 601 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
602 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
603 | error = xfs_trans_reserve(tp, nres, | |
604 | XFS_WRITE_LOG_RES(mp), | |
605 | 0, XFS_TRANS_PERM_LOG_RES, | |
606 | XFS_WRITE_LOG_COUNT); | |
1da177e4 LT |
607 | if (error) { |
608 | xfs_trans_cancel(tp, 0); | |
609 | return XFS_ERROR(error); | |
610 | } | |
611 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 612 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 613 | |
9d87c319 | 614 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 615 | |
1da177e4 | 616 | /* |
e4143a1c DC |
617 | * it is possible that the extents have changed since |
618 | * we did the read call as we dropped the ilock for a | |
619 | * while. We have to be careful about truncates or hole | |
620 | * punchs here - we are not allowed to allocate | |
621 | * non-delalloc blocks here. | |
622 | * | |
623 | * The only protection against truncation is the pages | |
624 | * for the range we are being asked to convert are | |
625 | * locked and hence a truncate will block on them | |
626 | * first. | |
627 | * | |
628 | * As a result, if we go beyond the range we really | |
629 | * need and hit an delalloc extent boundary followed by | |
630 | * a hole while we have excess blocks in the map, we | |
631 | * will fill the hole incorrectly and overrun the | |
632 | * transaction reservation. | |
633 | * | |
634 | * Using a single map prevents this as we are forced to | |
635 | * check each map we look for overlap with the desired | |
636 | * range and abort as soon as we find it. Also, given | |
637 | * that we only return a single map, having one beyond | |
638 | * what we can return is probably a bit silly. | |
639 | * | |
640 | * We also need to check that we don't go beyond EOF; | |
641 | * this is a truncate optimisation as a truncate sets | |
642 | * the new file size before block on the pages we | |
643 | * currently have locked under writeback. Because they | |
644 | * are about to be tossed, we don't need to write them | |
645 | * back.... | |
1da177e4 | 646 | */ |
e4143a1c | 647 | nimaps = 1; |
ce7ae151 | 648 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7c9ef85c DC |
649 | error = xfs_bmap_last_offset(NULL, ip, &last_block, |
650 | XFS_DATA_FORK); | |
651 | if (error) | |
652 | goto trans_cancel; | |
653 | ||
1da177e4 LT |
654 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
655 | if ((map_start_fsb + count_fsb) > last_block) { | |
656 | count_fsb = last_block - map_start_fsb; | |
657 | if (count_fsb == 0) { | |
658 | error = EAGAIN; | |
659 | goto trans_cancel; | |
660 | } | |
661 | } | |
662 | ||
3070451e | 663 | /* |
3070451e CH |
664 | * From this point onwards we overwrite the imap |
665 | * pointer that the caller gave to us. | |
666 | */ | |
c0dc7828 | 667 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
2455881c DC |
668 | count_fsb, |
669 | XFS_BMAPI_STACK_SWITCH, | |
670 | &first_block, 1, | |
c0dc7828 | 671 | imap, &nimaps, &free_list); |
1da177e4 LT |
672 | if (error) |
673 | goto trans_cancel; | |
674 | ||
f7c99b6f | 675 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
676 | if (error) |
677 | goto trans_cancel; | |
678 | ||
1c72bf90 | 679 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
680 | if (error) |
681 | goto error0; | |
682 | ||
683 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
684 | } | |
685 | ||
686 | /* | |
687 | * See if we were able to allocate an extent that | |
688 | * covers at least part of the callers request | |
689 | */ | |
3070451e | 690 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 691 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 692 | |
3070451e CH |
693 | if ((offset_fsb >= imap->br_startoff) && |
694 | (offset_fsb < (imap->br_startoff + | |
695 | imap->br_blockcount))) { | |
e4143a1c DC |
696 | XFS_STATS_INC(xs_xstrat_quick); |
697 | return 0; | |
1da177e4 LT |
698 | } |
699 | ||
e4143a1c DC |
700 | /* |
701 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
702 | * file, just surrounding data, try again. |
703 | */ | |
3070451e CH |
704 | count_fsb -= imap->br_blockcount; |
705 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
706 | } |
707 | ||
708 | trans_cancel: | |
709 | xfs_bmap_cancel(&free_list); | |
710 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
711 | error0: | |
712 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
713 | return XFS_ERROR(error); | |
714 | } | |
715 | ||
716 | int | |
717 | xfs_iomap_write_unwritten( | |
718 | xfs_inode_t *ip, | |
f403b7f4 | 719 | xfs_off_t offset, |
1da177e4 LT |
720 | size_t count) |
721 | { | |
722 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
723 | xfs_fileoff_t offset_fsb; |
724 | xfs_filblks_t count_fsb; | |
725 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
726 | xfs_fsblock_t firstfsb; |
727 | int nimaps; | |
728 | xfs_trans_t *tp; | |
729 | xfs_bmbt_irec_t imap; | |
730 | xfs_bmap_free_t free_list; | |
84803fb7 | 731 | xfs_fsize_t i_size; |
dd9f438e | 732 | uint resblks; |
1da177e4 LT |
733 | int committed; |
734 | int error; | |
1da177e4 | 735 | |
0b1b213f | 736 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
737 | |
738 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
739 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
740 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
741 | ||
4ddd8bb1 LM |
742 | /* |
743 | * Reserve enough blocks in this transaction for two complete extent | |
744 | * btree splits. We may be converting the middle part of an unwritten | |
745 | * extent and in this case we will insert two new extents in the btree | |
746 | * each of which could cause a full split. | |
747 | * | |
748 | * This reservation amount will be used in the first call to | |
749 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
750 | * rest of the operation. | |
751 | */ | |
dd9f438e | 752 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 753 | |
dd9f438e | 754 | do { |
1da177e4 LT |
755 | /* |
756 | * set up a transaction to convert the range of extents | |
757 | * from unwritten to real. Do allocations in a loop until | |
758 | * we have covered the range passed in. | |
80641dc6 CH |
759 | * |
760 | * Note that we open code the transaction allocation here | |
761 | * to pass KM_NOFS--we can't risk to recursing back into | |
762 | * the filesystem here as we might be asked to write out | |
763 | * the same inode that we complete here and might deadlock | |
764 | * on the iolock. | |
1da177e4 | 765 | */ |
d9457dc0 | 766 | sb_start_intwrite(mp->m_super); |
80641dc6 | 767 | tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS); |
d9457dc0 | 768 | tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT; |
dd9f438e | 769 | error = xfs_trans_reserve(tp, resblks, |
1da177e4 LT |
770 | XFS_WRITE_LOG_RES(mp), 0, |
771 | XFS_TRANS_PERM_LOG_RES, | |
772 | XFS_WRITE_LOG_COUNT); | |
773 | if (error) { | |
774 | xfs_trans_cancel(tp, 0); | |
572d95f4 | 775 | return XFS_ERROR(error); |
1da177e4 LT |
776 | } |
777 | ||
778 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 779 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
780 | |
781 | /* | |
782 | * Modify the unwritten extent state of the buffer. | |
783 | */ | |
9d87c319 | 784 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 785 | nimaps = 1; |
c0dc7828 DC |
786 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
787 | XFS_BMAPI_CONVERT, &firstfsb, | |
b4e9181e | 788 | 1, &imap, &nimaps, &free_list); |
1da177e4 LT |
789 | if (error) |
790 | goto error_on_bmapi_transaction; | |
791 | ||
84803fb7 CH |
792 | /* |
793 | * Log the updated inode size as we go. We have to be careful | |
794 | * to only log it up to the actual write offset if it is | |
795 | * halfway into a block. | |
796 | */ | |
797 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
798 | if (i_size > offset + count) | |
799 | i_size = offset + count; | |
800 | ||
801 | i_size = xfs_new_eof(ip, i_size); | |
802 | if (i_size) { | |
803 | ip->i_d.di_size = i_size; | |
804 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
805 | } | |
806 | ||
807 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1da177e4 LT |
808 | if (error) |
809 | goto error_on_bmapi_transaction; | |
810 | ||
1c72bf90 | 811 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
812 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
813 | if (error) | |
572d95f4 NS |
814 | return XFS_ERROR(error); |
815 | ||
86c4d623 | 816 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 817 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
818 | |
819 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
820 | /* | |
821 | * The numblks_fsb value should always get | |
822 | * smaller, otherwise the loop is stuck. | |
823 | */ | |
824 | ASSERT(imap.br_blockcount); | |
825 | break; | |
826 | } | |
827 | offset_fsb += numblks_fsb; | |
828 | count_fsb -= numblks_fsb; | |
829 | } while (count_fsb > 0); | |
830 | ||
831 | return 0; | |
832 | ||
833 | error_on_bmapi_transaction: | |
834 | xfs_bmap_cancel(&free_list); | |
835 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT)); | |
836 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 LT |
837 | return XFS_ERROR(error); |
838 | } |