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" |
6ca1c906 | 20 | #include "xfs_format.h" |
1da177e4 LT |
21 | #include "xfs_log.h" |
22 | #include "xfs_trans.h" | |
23 | #include "xfs_sb.h" | |
24 | #include "xfs_ag.h" | |
1da177e4 | 25 | #include "xfs_alloc.h" |
1da177e4 LT |
26 | #include "xfs_quota.h" |
27 | #include "xfs_mount.h" | |
1da177e4 | 28 | #include "xfs_bmap_btree.h" |
a844f451 | 29 | #include "xfs_alloc_btree.h" |
1da177e4 | 30 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
31 | #include "xfs_dinode.h" |
32 | #include "xfs_inode.h" | |
84803fb7 | 33 | #include "xfs_inode_item.h" |
a844f451 | 34 | #include "xfs_btree.h" |
1da177e4 | 35 | #include "xfs_bmap.h" |
68988114 | 36 | #include "xfs_bmap_util.h" |
1da177e4 LT |
37 | #include "xfs_rtalloc.h" |
38 | #include "xfs_error.h" | |
39 | #include "xfs_itable.h" | |
1da177e4 LT |
40 | #include "xfs_attr.h" |
41 | #include "xfs_buf_item.h" | |
42 | #include "xfs_trans_space.h" | |
1da177e4 | 43 | #include "xfs_iomap.h" |
0b1b213f | 44 | #include "xfs_trace.h" |
27b52867 | 45 | #include "xfs_icache.h" |
76a4202a BF |
46 | #include "xfs_dquot_item.h" |
47 | #include "xfs_dquot.h" | |
1da177e4 | 48 | |
1da177e4 LT |
49 | |
50 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
51 | << mp->m_writeio_log) | |
1da177e4 LT |
52 | #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP |
53 | ||
dd9f438e NS |
54 | STATIC int |
55 | xfs_iomap_eof_align_last_fsb( | |
56 | xfs_mount_t *mp, | |
541d7d3c | 57 | xfs_inode_t *ip, |
dd9f438e NS |
58 | xfs_extlen_t extsize, |
59 | xfs_fileoff_t *last_fsb) | |
60 | { | |
61 | xfs_fileoff_t new_last_fsb = 0; | |
bf322d98 | 62 | xfs_extlen_t align = 0; |
dd9f438e NS |
63 | int eof, error; |
64 | ||
bf322d98 CH |
65 | if (!XFS_IS_REALTIME_INODE(ip)) { |
66 | /* | |
67 | * Round up the allocation request to a stripe unit | |
68 | * (m_dalign) boundary if the file size is >= stripe unit | |
69 | * size, and we are allocating past the allocation eof. | |
70 | * | |
71 | * If mounted with the "-o swalloc" option the alignment is | |
72 | * increased from the strip unit size to the stripe width. | |
73 | */ | |
74 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
75 | align = mp->m_swidth; | |
76 | else if (mp->m_dalign) | |
77 | align = mp->m_dalign; | |
78 | ||
ce7ae151 | 79 | if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align)) |
bf322d98 CH |
80 | new_last_fsb = roundup_64(*last_fsb, align); |
81 | } | |
dd9f438e NS |
82 | |
83 | /* | |
84 | * Always round up the allocation request to an extent boundary | |
85 | * (when file on a real-time subvolume or has di_extsize hint). | |
86 | */ | |
87 | if (extsize) { | |
88 | if (new_last_fsb) | |
89 | align = roundup_64(new_last_fsb, extsize); | |
90 | else | |
91 | align = extsize; | |
92 | new_last_fsb = roundup_64(*last_fsb, align); | |
93 | } | |
94 | ||
95 | if (new_last_fsb) { | |
541d7d3c | 96 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
97 | if (error) |
98 | return error; | |
99 | if (eof) | |
100 | *last_fsb = new_last_fsb; | |
101 | } | |
102 | return 0; | |
103 | } | |
104 | ||
572d95f4 | 105 | STATIC int |
6d4a8ecb | 106 | xfs_alert_fsblock_zero( |
572d95f4 NS |
107 | xfs_inode_t *ip, |
108 | xfs_bmbt_irec_t *imap) | |
109 | { | |
6a19d939 | 110 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
111 | "Access to block zero in inode %llu " |
112 | "start_block: %llx start_off: %llx " | |
113 | "blkcnt: %llx extent-state: %x\n", | |
114 | (unsigned long long)ip->i_ino, | |
115 | (unsigned long long)imap->br_startblock, | |
116 | (unsigned long long)imap->br_startoff, | |
117 | (unsigned long long)imap->br_blockcount, | |
118 | imap->br_state); | |
119 | return EFSCORRUPTED; | |
120 | } | |
121 | ||
a206c817 | 122 | int |
1da177e4 LT |
123 | xfs_iomap_write_direct( |
124 | xfs_inode_t *ip, | |
f403b7f4 | 125 | xfs_off_t offset, |
1da177e4 | 126 | size_t count, |
3070451e | 127 | xfs_bmbt_irec_t *imap, |
405f8042 | 128 | int nmaps) |
1da177e4 LT |
129 | { |
130 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
131 | xfs_fileoff_t offset_fsb; |
132 | xfs_fileoff_t last_fsb; | |
dd9f438e | 133 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 134 | xfs_fsblock_t firstfsb; |
dd9f438e | 135 | xfs_extlen_t extsz, temp; |
0116d935 | 136 | int nimaps; |
1da177e4 | 137 | int bmapi_flag; |
06d10dd9 | 138 | int quota_flag; |
1da177e4 LT |
139 | int rt; |
140 | xfs_trans_t *tp; | |
1da177e4 | 141 | xfs_bmap_free_t free_list; |
dd9f438e | 142 | uint qblocks, resblks, resrtextents; |
1da177e4 | 143 | int committed; |
dd9f438e | 144 | int error; |
1da177e4 | 145 | |
507630b2 | 146 | error = xfs_qm_dqattach(ip, 0); |
1da177e4 LT |
147 | if (error) |
148 | return XFS_ERROR(error); | |
149 | ||
dd9f438e | 150 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 151 | extsz = xfs_get_extsz_hint(ip); |
1da177e4 | 152 | |
957d0ebe DC |
153 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
154 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 155 | if ((offset + count) > XFS_ISIZE(ip)) { |
9f6c92b9 | 156 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 157 | if (error) |
507630b2 | 158 | return XFS_ERROR(error); |
1da177e4 | 159 | } else { |
405f8042 | 160 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 161 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
162 | imap->br_blockcount + |
163 | imap->br_startoff); | |
1da177e4 | 164 | } |
dd9f438e NS |
165 | count_fsb = last_fsb - offset_fsb; |
166 | ASSERT(count_fsb > 0); | |
167 | ||
168 | resaligned = count_fsb; | |
169 | if (unlikely(extsz)) { | |
170 | if ((temp = do_mod(offset_fsb, extsz))) | |
171 | resaligned += temp; | |
172 | if ((temp = do_mod(resaligned, extsz))) | |
173 | resaligned += extsz - temp; | |
174 | } | |
175 | ||
176 | if (unlikely(rt)) { | |
177 | resrtextents = qblocks = resaligned; | |
178 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
179 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
180 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
181 | } else { | |
182 | resrtextents = 0; | |
dd9f438e | 183 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
184 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
185 | } | |
1da177e4 LT |
186 | |
187 | /* | |
06d10dd9 | 188 | * Allocate and setup the transaction |
1da177e4 | 189 | */ |
1da177e4 | 190 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); |
1da177e4 | 191 | error = xfs_trans_reserve(tp, resblks, |
d52b44d0 | 192 | XFS_WRITE_LOG_RES(mp), resrtextents, |
1da177e4 LT |
193 | XFS_TRANS_PERM_LOG_RES, |
194 | XFS_WRITE_LOG_COUNT); | |
1da177e4 | 195 | /* |
06d10dd9 | 196 | * Check for running out of space, note: need lock to return |
1da177e4 | 197 | */ |
507630b2 | 198 | if (error) { |
1da177e4 | 199 | xfs_trans_cancel(tp, 0); |
507630b2 DC |
200 | return XFS_ERROR(error); |
201 | } | |
202 | ||
1da177e4 | 203 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 204 | |
7d095257 | 205 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 206 | if (error) |
507630b2 | 207 | goto out_trans_cancel; |
1da177e4 | 208 | |
ddc3415a | 209 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 210 | |
c0dc7828 | 211 | bmapi_flag = 0; |
ce7ae151 | 212 | if (offset < XFS_ISIZE(ip) || extsz) |
1da177e4 LT |
213 | bmapi_flag |= XFS_BMAPI_PREALLOC; |
214 | ||
215 | /* | |
3070451e CH |
216 | * From this point onwards we overwrite the imap pointer that the |
217 | * caller gave to us. | |
1da177e4 | 218 | */ |
9d87c319 | 219 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 220 | nimaps = 1; |
c0dc7828 DC |
221 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag, |
222 | &firstfsb, 0, imap, &nimaps, &free_list); | |
06d10dd9 | 223 | if (error) |
507630b2 | 224 | goto out_bmap_cancel; |
1da177e4 LT |
225 | |
226 | /* | |
06d10dd9 | 227 | * Complete the transaction |
1da177e4 | 228 | */ |
f7c99b6f | 229 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
06d10dd9 | 230 | if (error) |
507630b2 | 231 | goto out_bmap_cancel; |
1c72bf90 | 232 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
06d10dd9 | 233 | if (error) |
507630b2 | 234 | goto out_unlock; |
1da177e4 | 235 | |
06d10dd9 NS |
236 | /* |
237 | * Copy any maps to caller's array and return any error. | |
238 | */ | |
1da177e4 | 239 | if (nimaps == 0) { |
507630b2 DC |
240 | error = XFS_ERROR(ENOSPC); |
241 | goto out_unlock; | |
572d95f4 NS |
242 | } |
243 | ||
507630b2 | 244 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 245 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 246 | |
507630b2 DC |
247 | out_unlock: |
248 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
249 | return error; | |
1da177e4 | 250 | |
507630b2 | 251 | out_bmap_cancel: |
1da177e4 | 252 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 253 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 254 | out_trans_cancel: |
1da177e4 | 255 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
507630b2 | 256 | goto out_unlock; |
1da177e4 LT |
257 | } |
258 | ||
dd9f438e | 259 | /* |
8de2bf93 DC |
260 | * If the caller is doing a write at the end of the file, then extend the |
261 | * allocation out to the file system's write iosize. We clean up any extra | |
262 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
263 | * |
264 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
265 | * preallocation as it it not needed. | |
dd9f438e NS |
266 | */ |
267 | STATIC int | |
268 | xfs_iomap_eof_want_preallocate( | |
269 | xfs_mount_t *mp, | |
541d7d3c | 270 | xfs_inode_t *ip, |
dd9f438e NS |
271 | xfs_off_t offset, |
272 | size_t count, | |
dd9f438e NS |
273 | xfs_bmbt_irec_t *imap, |
274 | int nimaps, | |
275 | int *prealloc) | |
276 | { | |
277 | xfs_fileoff_t start_fsb; | |
278 | xfs_filblks_t count_fsb; | |
279 | xfs_fsblock_t firstblock; | |
280 | int n, error, imaps; | |
055388a3 | 281 | int found_delalloc = 0; |
dd9f438e NS |
282 | |
283 | *prealloc = 0; | |
ce7ae151 | 284 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
285 | return 0; |
286 | ||
133eeb17 DC |
287 | /* |
288 | * If the file is smaller than the minimum prealloc and we are using | |
289 | * dynamic preallocation, don't do any preallocation at all as it is | |
290 | * likely this is the only write to the file that is going to be done. | |
291 | */ | |
292 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
293 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)) | |
294 | return 0; | |
295 | ||
dd9f438e NS |
296 | /* |
297 | * If there are any real blocks past eof, then don't | |
298 | * do any speculative allocation. | |
299 | */ | |
300 | start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1))); | |
32972383 | 301 | count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
dd9f438e NS |
302 | while (count_fsb > 0) { |
303 | imaps = nimaps; | |
3ddb8fa9 | 304 | firstblock = NULLFSBLOCK; |
5c8ed202 DC |
305 | error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps, |
306 | 0); | |
dd9f438e NS |
307 | if (error) |
308 | return error; | |
309 | for (n = 0; n < imaps; n++) { | |
310 | if ((imap[n].br_startblock != HOLESTARTBLOCK) && | |
311 | (imap[n].br_startblock != DELAYSTARTBLOCK)) | |
312 | return 0; | |
313 | start_fsb += imap[n].br_blockcount; | |
314 | count_fsb -= imap[n].br_blockcount; | |
055388a3 DC |
315 | |
316 | if (imap[n].br_startblock == DELAYSTARTBLOCK) | |
317 | found_delalloc = 1; | |
dd9f438e NS |
318 | } |
319 | } | |
055388a3 DC |
320 | if (!found_delalloc) |
321 | *prealloc = 1; | |
dd9f438e NS |
322 | return 0; |
323 | } | |
324 | ||
a1e16c26 DC |
325 | /* |
326 | * Determine the initial size of the preallocation. We are beyond the current | |
327 | * EOF here, but we need to take into account whether this is a sparse write or | |
328 | * an extending write when determining the preallocation size. Hence we need to | |
329 | * look up the extent that ends at the current write offset and use the result | |
330 | * to determine the preallocation size. | |
331 | * | |
332 | * If the extent is a hole, then preallocation is essentially disabled. | |
333 | * Otherwise we take the size of the preceeding data extent as the basis for the | |
334 | * preallocation size. If the size of the extent is greater than half the | |
335 | * maximum extent length, then use the current offset as the basis. This ensures | |
336 | * that for large files the preallocation size always extends to MAXEXTLEN | |
337 | * rather than falling short due to things like stripe unit/width alignment of | |
338 | * real extents. | |
339 | */ | |
e8108ced | 340 | STATIC xfs_fsblock_t |
a1e16c26 DC |
341 | xfs_iomap_eof_prealloc_initial_size( |
342 | struct xfs_mount *mp, | |
343 | struct xfs_inode *ip, | |
344 | xfs_off_t offset, | |
345 | xfs_bmbt_irec_t *imap, | |
346 | int nimaps) | |
347 | { | |
348 | xfs_fileoff_t start_fsb; | |
349 | int imaps = 1; | |
350 | int error; | |
351 | ||
352 | ASSERT(nimaps >= imaps); | |
353 | ||
354 | /* if we are using a specific prealloc size, return now */ | |
355 | if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) | |
356 | return 0; | |
357 | ||
133eeb17 DC |
358 | /* If the file is small, then use the minimum prealloc */ |
359 | if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign)) | |
360 | return 0; | |
361 | ||
a1e16c26 DC |
362 | /* |
363 | * As we write multiple pages, the offset will always align to the | |
364 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
365 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
366 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
367 | * point to the block prior to the EOF block and not the hole that maps | |
368 | * directly at @offset. | |
369 | */ | |
370 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
371 | if (start_fsb) | |
372 | start_fsb--; | |
373 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
374 | if (error) | |
375 | return 0; | |
376 | ||
377 | ASSERT(imaps == 1); | |
378 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
379 | return 0; | |
380 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 381 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
382 | return XFS_B_TO_FSB(mp, offset); |
383 | } | |
384 | ||
76a4202a BF |
385 | STATIC bool |
386 | xfs_quota_need_throttle( | |
387 | struct xfs_inode *ip, | |
388 | int type, | |
389 | xfs_fsblock_t alloc_blocks) | |
390 | { | |
391 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
392 | ||
393 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
394 | return false; | |
395 | ||
396 | /* no hi watermark, no throttle */ | |
397 | if (!dq->q_prealloc_hi_wmark) | |
398 | return false; | |
399 | ||
400 | /* under the lo watermark, no throttle */ | |
401 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
402 | return false; | |
403 | ||
404 | return true; | |
405 | } | |
406 | ||
407 | STATIC void | |
408 | xfs_quota_calc_throttle( | |
409 | struct xfs_inode *ip, | |
410 | int type, | |
411 | xfs_fsblock_t *qblocks, | |
412 | int *qshift) | |
413 | { | |
414 | int64_t freesp; | |
415 | int shift = 0; | |
416 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
417 | ||
418 | /* over hi wmark, squash the prealloc completely */ | |
419 | if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
420 | *qblocks = 0; | |
421 | return; | |
422 | } | |
423 | ||
424 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
425 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
426 | shift = 2; | |
427 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
428 | shift += 2; | |
429 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
430 | shift += 2; | |
431 | } | |
432 | ||
433 | /* only overwrite the throttle values if we are more aggressive */ | |
434 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
435 | *qblocks = freesp; | |
436 | *qshift = shift; | |
437 | } | |
438 | } | |
439 | ||
055388a3 DC |
440 | /* |
441 | * If we don't have a user specified preallocation size, dynamically increase | |
442 | * the preallocation size as the size of the file grows. Cap the maximum size | |
443 | * at a single extent or less if the filesystem is near full. The closer the | |
444 | * filesystem is to full, the smaller the maximum prealocation. | |
445 | */ | |
446 | STATIC xfs_fsblock_t | |
447 | xfs_iomap_prealloc_size( | |
448 | struct xfs_mount *mp, | |
a1e16c26 DC |
449 | struct xfs_inode *ip, |
450 | xfs_off_t offset, | |
451 | struct xfs_bmbt_irec *imap, | |
452 | int nimaps) | |
055388a3 DC |
453 | { |
454 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
455 | int shift = 0; |
456 | int64_t freesp; | |
76a4202a BF |
457 | xfs_fsblock_t qblocks; |
458 | int qshift = 0; | |
055388a3 | 459 | |
a1e16c26 DC |
460 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
461 | imap, nimaps); | |
3c58b5f8 BF |
462 | if (!alloc_blocks) |
463 | goto check_writeio; | |
76a4202a | 464 | qblocks = alloc_blocks; |
3c58b5f8 | 465 | |
c9bdbdc0 BF |
466 | /* |
467 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
468 | * to the nearest power of two value after throttling. To prevent the | |
469 | * round down from unconditionally reducing the maximum supported prealloc | |
470 | * size, we round up first, apply appropriate throttling, round down and | |
471 | * cap the value to MAXEXTLEN. | |
472 | */ | |
473 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
474 | alloc_blocks); | |
3c58b5f8 BF |
475 | |
476 | xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); | |
477 | freesp = mp->m_sb.sb_fdblocks; | |
478 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { | |
479 | shift = 2; | |
480 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
481 | shift++; | |
482 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
483 | shift++; | |
484 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
485 | shift++; | |
486 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
487 | shift++; | |
055388a3 | 488 | } |
76a4202a BF |
489 | |
490 | /* | |
491 | * Check each quota to cap the prealloc size and provide a shift | |
492 | * value to throttle with. | |
493 | */ | |
494 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
495 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift); | |
496 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) | |
497 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift); | |
498 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) | |
499 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift); | |
500 | ||
501 | /* | |
502 | * The final prealloc size is set to the minimum of free space available | |
503 | * in each of the quotas and the overall filesystem. | |
504 | * | |
505 | * The shift throttle value is set to the maximum value as determined by | |
506 | * the global low free space values and per-quota low free space values. | |
507 | */ | |
508 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
509 | shift = MAX(shift, qshift); | |
510 | ||
3c58b5f8 BF |
511 | if (shift) |
512 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
513 | /* |
514 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
515 | * alloc_blocks = 0. | |
516 | */ | |
517 | if (alloc_blocks) | |
518 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
519 | if (alloc_blocks > MAXEXTLEN) | |
520 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
521 | |
522 | /* | |
523 | * If we are still trying to allocate more space than is | |
524 | * available, squash the prealloc hard. This can happen if we | |
525 | * have a large file on a small filesystem and the above | |
526 | * lowspace thresholds are smaller than MAXEXTLEN. | |
527 | */ | |
528 | while (alloc_blocks && alloc_blocks >= freesp) | |
529 | alloc_blocks >>= 4; | |
055388a3 | 530 | |
3c58b5f8 | 531 | check_writeio: |
055388a3 DC |
532 | if (alloc_blocks < mp->m_writeio_blocks) |
533 | alloc_blocks = mp->m_writeio_blocks; | |
534 | ||
19cb7e38 BF |
535 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
536 | mp->m_writeio_blocks); | |
537 | ||
055388a3 DC |
538 | return alloc_blocks; |
539 | } | |
540 | ||
a206c817 | 541 | int |
1da177e4 LT |
542 | xfs_iomap_write_delay( |
543 | xfs_inode_t *ip, | |
f403b7f4 | 544 | xfs_off_t offset, |
1da177e4 | 545 | size_t count, |
405f8042 | 546 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
547 | { |
548 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
549 | xfs_fileoff_t offset_fsb; |
550 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
551 | xfs_off_t aligned_offset; |
552 | xfs_fileoff_t ioalign; | |
dd9f438e | 553 | xfs_extlen_t extsz; |
1da177e4 | 554 | int nimaps; |
1da177e4 | 555 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 556 | int prealloc; |
dd9f438e | 557 | int error; |
1da177e4 | 558 | |
579aa9ca | 559 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
560 | |
561 | /* | |
562 | * Make sure that the dquots are there. This doesn't hold | |
563 | * the ilock across a disk read. | |
564 | */ | |
7d095257 | 565 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 LT |
566 | if (error) |
567 | return XFS_ERROR(error); | |
568 | ||
957d0ebe | 569 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
570 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
571 | ||
9f6c92b9 | 572 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 573 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
574 | if (error) |
575 | return error; | |
1da177e4 | 576 | |
8de2bf93 | 577 | retry: |
dd9f438e | 578 | if (prealloc) { |
a1e16c26 DC |
579 | xfs_fsblock_t alloc_blocks; |
580 | ||
581 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
582 | XFS_WRITE_IMAPS); | |
055388a3 | 583 | |
1da177e4 LT |
584 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
585 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 586 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
587 | } else { |
588 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 589 | } |
1da177e4 | 590 | |
dd9f438e | 591 | if (prealloc || extsz) { |
9f6c92b9 | 592 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 593 | if (error) |
1da177e4 | 594 | return error; |
1da177e4 | 595 | } |
dd9f438e | 596 | |
3ed9116e DC |
597 | /* |
598 | * Make sure preallocation does not create extents beyond the range we | |
599 | * actually support in this filesystem. | |
600 | */ | |
d2c28191 DC |
601 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
602 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
603 | |
604 | ASSERT(last_fsb > offset_fsb); | |
605 | ||
dd9f438e | 606 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
607 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
608 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
609 | switch (error) { |
610 | case 0: | |
611 | case ENOSPC: | |
612 | case EDQUOT: | |
613 | break; | |
614 | default: | |
1da177e4 | 615 | return XFS_ERROR(error); |
055388a3 | 616 | } |
dd9f438e | 617 | |
1da177e4 | 618 | /* |
9aa05000 | 619 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 620 | * without EOF preallocation. |
1da177e4 LT |
621 | */ |
622 | if (nimaps == 0) { | |
0b1b213f | 623 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
624 | if (prealloc) { |
625 | prealloc = 0; | |
626 | error = 0; | |
627 | goto retry; | |
055388a3 | 628 | } |
9aa05000 | 629 | return XFS_ERROR(error ? error : ENOSPC); |
1da177e4 LT |
630 | } |
631 | ||
86c4d623 | 632 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 633 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 634 | |
27b52867 BF |
635 | /* |
636 | * Tag the inode as speculatively preallocated so we can reclaim this | |
637 | * space on demand, if necessary. | |
638 | */ | |
639 | if (prealloc) | |
640 | xfs_inode_set_eofblocks_tag(ip); | |
641 | ||
dd9f438e | 642 | *ret_imap = imap[0]; |
1da177e4 LT |
643 | return 0; |
644 | } | |
645 | ||
646 | /* | |
647 | * Pass in a delayed allocate extent, convert it to real extents; | |
648 | * return to the caller the extent we create which maps on top of | |
649 | * the originating callers request. | |
650 | * | |
651 | * Called without a lock on the inode. | |
e4143a1c DC |
652 | * |
653 | * We no longer bother to look at the incoming map - all we have to | |
654 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 655 | */ |
a206c817 | 656 | int |
1da177e4 LT |
657 | xfs_iomap_write_allocate( |
658 | xfs_inode_t *ip, | |
f403b7f4 | 659 | xfs_off_t offset, |
24e17b5f | 660 | size_t count, |
405f8042 | 661 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
662 | { |
663 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
664 | xfs_fileoff_t offset_fsb, last_block; |
665 | xfs_fileoff_t end_fsb, map_start_fsb; | |
666 | xfs_fsblock_t first_block; | |
667 | xfs_bmap_free_t free_list; | |
668 | xfs_filblks_t count_fsb; | |
1da177e4 | 669 | xfs_trans_t *tp; |
e4143a1c | 670 | int nimaps, committed; |
1da177e4 LT |
671 | int error = 0; |
672 | int nres; | |
673 | ||
1da177e4 LT |
674 | /* |
675 | * Make sure that the dquots are there. | |
676 | */ | |
7d095257 CH |
677 | error = xfs_qm_dqattach(ip, 0); |
678 | if (error) | |
1da177e4 LT |
679 | return XFS_ERROR(error); |
680 | ||
24e17b5f | 681 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
682 | count_fsb = imap->br_blockcount; |
683 | map_start_fsb = imap->br_startoff; | |
1da177e4 LT |
684 | |
685 | XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); | |
686 | ||
687 | while (count_fsb != 0) { | |
688 | /* | |
689 | * Set up a transaction with which to allocate the | |
690 | * backing store for the file. Do allocations in a | |
691 | * loop until we get some space in the range we are | |
692 | * interested in. The other space that might be allocated | |
693 | * is in the delayed allocation extent on which we sit | |
694 | * but before our buffer starts. | |
695 | */ | |
696 | ||
697 | nimaps = 0; | |
698 | while (nimaps == 0) { | |
699 | tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE); | |
84e1e99f | 700 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 LT |
701 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
702 | error = xfs_trans_reserve(tp, nres, | |
703 | XFS_WRITE_LOG_RES(mp), | |
704 | 0, XFS_TRANS_PERM_LOG_RES, | |
705 | XFS_WRITE_LOG_COUNT); | |
1da177e4 LT |
706 | if (error) { |
707 | xfs_trans_cancel(tp, 0); | |
708 | return XFS_ERROR(error); | |
709 | } | |
710 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 711 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 712 | |
9d87c319 | 713 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 714 | |
1da177e4 | 715 | /* |
e4143a1c DC |
716 | * it is possible that the extents have changed since |
717 | * we did the read call as we dropped the ilock for a | |
718 | * while. We have to be careful about truncates or hole | |
719 | * punchs here - we are not allowed to allocate | |
720 | * non-delalloc blocks here. | |
721 | * | |
722 | * The only protection against truncation is the pages | |
723 | * for the range we are being asked to convert are | |
724 | * locked and hence a truncate will block on them | |
725 | * first. | |
726 | * | |
727 | * As a result, if we go beyond the range we really | |
728 | * need and hit an delalloc extent boundary followed by | |
729 | * a hole while we have excess blocks in the map, we | |
730 | * will fill the hole incorrectly and overrun the | |
731 | * transaction reservation. | |
732 | * | |
733 | * Using a single map prevents this as we are forced to | |
734 | * check each map we look for overlap with the desired | |
735 | * range and abort as soon as we find it. Also, given | |
736 | * that we only return a single map, having one beyond | |
737 | * what we can return is probably a bit silly. | |
738 | * | |
739 | * We also need to check that we don't go beyond EOF; | |
740 | * this is a truncate optimisation as a truncate sets | |
741 | * the new file size before block on the pages we | |
742 | * currently have locked under writeback. Because they | |
743 | * are about to be tossed, we don't need to write them | |
744 | * back.... | |
1da177e4 | 745 | */ |
e4143a1c | 746 | nimaps = 1; |
ce7ae151 | 747 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7c9ef85c DC |
748 | error = xfs_bmap_last_offset(NULL, ip, &last_block, |
749 | XFS_DATA_FORK); | |
750 | if (error) | |
751 | goto trans_cancel; | |
752 | ||
1da177e4 LT |
753 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
754 | if ((map_start_fsb + count_fsb) > last_block) { | |
755 | count_fsb = last_block - map_start_fsb; | |
756 | if (count_fsb == 0) { | |
757 | error = EAGAIN; | |
758 | goto trans_cancel; | |
759 | } | |
760 | } | |
761 | ||
3070451e | 762 | /* |
3070451e CH |
763 | * From this point onwards we overwrite the imap |
764 | * pointer that the caller gave to us. | |
765 | */ | |
c0dc7828 | 766 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
2455881c DC |
767 | count_fsb, |
768 | XFS_BMAPI_STACK_SWITCH, | |
769 | &first_block, 1, | |
c0dc7828 | 770 | imap, &nimaps, &free_list); |
1da177e4 LT |
771 | if (error) |
772 | goto trans_cancel; | |
773 | ||
f7c99b6f | 774 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
775 | if (error) |
776 | goto trans_cancel; | |
777 | ||
1c72bf90 | 778 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
779 | if (error) |
780 | goto error0; | |
781 | ||
782 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
783 | } | |
784 | ||
785 | /* | |
786 | * See if we were able to allocate an extent that | |
787 | * covers at least part of the callers request | |
788 | */ | |
3070451e | 789 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 790 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 791 | |
3070451e CH |
792 | if ((offset_fsb >= imap->br_startoff) && |
793 | (offset_fsb < (imap->br_startoff + | |
794 | imap->br_blockcount))) { | |
e4143a1c DC |
795 | XFS_STATS_INC(xs_xstrat_quick); |
796 | return 0; | |
1da177e4 LT |
797 | } |
798 | ||
e4143a1c DC |
799 | /* |
800 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
801 | * file, just surrounding data, try again. |
802 | */ | |
3070451e CH |
803 | count_fsb -= imap->br_blockcount; |
804 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
805 | } |
806 | ||
807 | trans_cancel: | |
808 | xfs_bmap_cancel(&free_list); | |
809 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
810 | error0: | |
811 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
812 | return XFS_ERROR(error); | |
813 | } | |
814 | ||
815 | int | |
816 | xfs_iomap_write_unwritten( | |
817 | xfs_inode_t *ip, | |
f403b7f4 | 818 | xfs_off_t offset, |
1da177e4 LT |
819 | size_t count) |
820 | { | |
821 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
822 | xfs_fileoff_t offset_fsb; |
823 | xfs_filblks_t count_fsb; | |
824 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
825 | xfs_fsblock_t firstfsb; |
826 | int nimaps; | |
827 | xfs_trans_t *tp; | |
828 | xfs_bmbt_irec_t imap; | |
829 | xfs_bmap_free_t free_list; | |
84803fb7 | 830 | xfs_fsize_t i_size; |
dd9f438e | 831 | uint resblks; |
1da177e4 LT |
832 | int committed; |
833 | int error; | |
1da177e4 | 834 | |
0b1b213f | 835 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
836 | |
837 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
838 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
839 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
840 | ||
4ddd8bb1 LM |
841 | /* |
842 | * Reserve enough blocks in this transaction for two complete extent | |
843 | * btree splits. We may be converting the middle part of an unwritten | |
844 | * extent and in this case we will insert two new extents in the btree | |
845 | * each of which could cause a full split. | |
846 | * | |
847 | * This reservation amount will be used in the first call to | |
848 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
849 | * rest of the operation. | |
850 | */ | |
dd9f438e | 851 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 852 | |
dd9f438e | 853 | do { |
1da177e4 LT |
854 | /* |
855 | * set up a transaction to convert the range of extents | |
856 | * from unwritten to real. Do allocations in a loop until | |
857 | * we have covered the range passed in. | |
80641dc6 CH |
858 | * |
859 | * Note that we open code the transaction allocation here | |
860 | * to pass KM_NOFS--we can't risk to recursing back into | |
861 | * the filesystem here as we might be asked to write out | |
862 | * the same inode that we complete here and might deadlock | |
863 | * on the iolock. | |
1da177e4 | 864 | */ |
d9457dc0 | 865 | sb_start_intwrite(mp->m_super); |
80641dc6 | 866 | tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS); |
d9457dc0 | 867 | tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT; |
dd9f438e | 868 | error = xfs_trans_reserve(tp, resblks, |
1da177e4 LT |
869 | XFS_WRITE_LOG_RES(mp), 0, |
870 | XFS_TRANS_PERM_LOG_RES, | |
871 | XFS_WRITE_LOG_COUNT); | |
872 | if (error) { | |
873 | xfs_trans_cancel(tp, 0); | |
572d95f4 | 874 | return XFS_ERROR(error); |
1da177e4 LT |
875 | } |
876 | ||
877 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 878 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
879 | |
880 | /* | |
881 | * Modify the unwritten extent state of the buffer. | |
882 | */ | |
9d87c319 | 883 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 884 | nimaps = 1; |
c0dc7828 DC |
885 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
886 | XFS_BMAPI_CONVERT, &firstfsb, | |
b4e9181e | 887 | 1, &imap, &nimaps, &free_list); |
1da177e4 LT |
888 | if (error) |
889 | goto error_on_bmapi_transaction; | |
890 | ||
84803fb7 CH |
891 | /* |
892 | * Log the updated inode size as we go. We have to be careful | |
893 | * to only log it up to the actual write offset if it is | |
894 | * halfway into a block. | |
895 | */ | |
896 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
897 | if (i_size > offset + count) | |
898 | i_size = offset + count; | |
899 | ||
900 | i_size = xfs_new_eof(ip, i_size); | |
901 | if (i_size) { | |
902 | ip->i_d.di_size = i_size; | |
903 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
904 | } | |
905 | ||
906 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1da177e4 LT |
907 | if (error) |
908 | goto error_on_bmapi_transaction; | |
909 | ||
1c72bf90 | 910 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
911 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
912 | if (error) | |
572d95f4 NS |
913 | return XFS_ERROR(error); |
914 | ||
86c4d623 | 915 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 916 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
917 | |
918 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
919 | /* | |
920 | * The numblks_fsb value should always get | |
921 | * smaller, otherwise the loop is stuck. | |
922 | */ | |
923 | ASSERT(imap.br_blockcount); | |
924 | break; | |
925 | } | |
926 | offset_fsb += numblks_fsb; | |
927 | count_fsb -= numblks_fsb; | |
928 | } while (count_fsb > 0); | |
929 | ||
930 | return 0; | |
931 | ||
932 | error_on_bmapi_transaction: | |
933 | xfs_bmap_cancel(&free_list); | |
934 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT)); | |
935 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 LT |
936 | return XFS_ERROR(error); |
937 | } |