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" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
1da177e4 LT |
24 | #include "xfs_sb.h" |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_mount.h" |
1da177e4 | 27 | #include "xfs_inode.h" |
a844f451 | 28 | #include "xfs_btree.h" |
a4fbe6ab | 29 | #include "xfs_bmap_btree.h" |
1da177e4 | 30 | #include "xfs_bmap.h" |
68988114 | 31 | #include "xfs_bmap_util.h" |
1da177e4 | 32 | #include "xfs_error.h" |
a4fbe6ab | 33 | #include "xfs_trans.h" |
1da177e4 | 34 | #include "xfs_trans_space.h" |
1da177e4 | 35 | #include "xfs_iomap.h" |
0b1b213f | 36 | #include "xfs_trace.h" |
27b52867 | 37 | #include "xfs_icache.h" |
a4fbe6ab | 38 | #include "xfs_quota.h" |
76a4202a BF |
39 | #include "xfs_dquot_item.h" |
40 | #include "xfs_dquot.h" | |
a4fbe6ab | 41 | #include "xfs_dinode.h" |
1da177e4 | 42 | |
1da177e4 LT |
43 | |
44 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
45 | << mp->m_writeio_log) | |
1da177e4 LT |
46 | #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP |
47 | ||
dd9f438e NS |
48 | STATIC int |
49 | xfs_iomap_eof_align_last_fsb( | |
50 | xfs_mount_t *mp, | |
541d7d3c | 51 | xfs_inode_t *ip, |
dd9f438e NS |
52 | xfs_extlen_t extsize, |
53 | xfs_fileoff_t *last_fsb) | |
54 | { | |
55 | xfs_fileoff_t new_last_fsb = 0; | |
bf322d98 | 56 | xfs_extlen_t align = 0; |
dd9f438e NS |
57 | int eof, error; |
58 | ||
bf322d98 CH |
59 | if (!XFS_IS_REALTIME_INODE(ip)) { |
60 | /* | |
61 | * Round up the allocation request to a stripe unit | |
62 | * (m_dalign) boundary if the file size is >= stripe unit | |
63 | * size, and we are allocating past the allocation eof. | |
64 | * | |
65 | * If mounted with the "-o swalloc" option the alignment is | |
66 | * increased from the strip unit size to the stripe width. | |
67 | */ | |
68 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
69 | align = mp->m_swidth; | |
70 | else if (mp->m_dalign) | |
71 | align = mp->m_dalign; | |
72 | ||
ce7ae151 | 73 | if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align)) |
bf322d98 CH |
74 | new_last_fsb = roundup_64(*last_fsb, align); |
75 | } | |
dd9f438e NS |
76 | |
77 | /* | |
78 | * Always round up the allocation request to an extent boundary | |
79 | * (when file on a real-time subvolume or has di_extsize hint). | |
80 | */ | |
81 | if (extsize) { | |
82 | if (new_last_fsb) | |
83 | align = roundup_64(new_last_fsb, extsize); | |
84 | else | |
85 | align = extsize; | |
86 | new_last_fsb = roundup_64(*last_fsb, align); | |
87 | } | |
88 | ||
89 | if (new_last_fsb) { | |
541d7d3c | 90 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
91 | if (error) |
92 | return error; | |
93 | if (eof) | |
94 | *last_fsb = new_last_fsb; | |
95 | } | |
96 | return 0; | |
97 | } | |
98 | ||
572d95f4 | 99 | STATIC int |
6d4a8ecb | 100 | xfs_alert_fsblock_zero( |
572d95f4 NS |
101 | xfs_inode_t *ip, |
102 | xfs_bmbt_irec_t *imap) | |
103 | { | |
6a19d939 | 104 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
105 | "Access to block zero in inode %llu " |
106 | "start_block: %llx start_off: %llx " | |
08e96e1a | 107 | "blkcnt: %llx extent-state: %x", |
572d95f4 NS |
108 | (unsigned long long)ip->i_ino, |
109 | (unsigned long long)imap->br_startblock, | |
110 | (unsigned long long)imap->br_startoff, | |
111 | (unsigned long long)imap->br_blockcount, | |
112 | imap->br_state); | |
2451337d | 113 | return -EFSCORRUPTED; |
572d95f4 NS |
114 | } |
115 | ||
a206c817 | 116 | int |
1da177e4 LT |
117 | xfs_iomap_write_direct( |
118 | xfs_inode_t *ip, | |
f403b7f4 | 119 | xfs_off_t offset, |
1da177e4 | 120 | size_t count, |
3070451e | 121 | xfs_bmbt_irec_t *imap, |
405f8042 | 122 | int nmaps) |
1da177e4 LT |
123 | { |
124 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
125 | xfs_fileoff_t offset_fsb; |
126 | xfs_fileoff_t last_fsb; | |
dd9f438e | 127 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 128 | xfs_fsblock_t firstfsb; |
dd9f438e | 129 | xfs_extlen_t extsz, temp; |
0116d935 | 130 | int nimaps; |
06d10dd9 | 131 | int quota_flag; |
1da177e4 LT |
132 | int rt; |
133 | xfs_trans_t *tp; | |
1da177e4 | 134 | xfs_bmap_free_t free_list; |
dd9f438e | 135 | uint qblocks, resblks, resrtextents; |
1da177e4 | 136 | int committed; |
dd9f438e | 137 | int error; |
1da177e4 | 138 | |
507630b2 | 139 | error = xfs_qm_dqattach(ip, 0); |
1da177e4 | 140 | if (error) |
b474c7ae | 141 | return error; |
1da177e4 | 142 | |
dd9f438e | 143 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 144 | extsz = xfs_get_extsz_hint(ip); |
1da177e4 | 145 | |
957d0ebe DC |
146 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
147 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 148 | if ((offset + count) > XFS_ISIZE(ip)) { |
9f6c92b9 | 149 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 150 | if (error) |
b474c7ae | 151 | return error; |
1da177e4 | 152 | } else { |
405f8042 | 153 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 154 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
155 | imap->br_blockcount + |
156 | imap->br_startoff); | |
1da177e4 | 157 | } |
dd9f438e NS |
158 | count_fsb = last_fsb - offset_fsb; |
159 | ASSERT(count_fsb > 0); | |
160 | ||
161 | resaligned = count_fsb; | |
162 | if (unlikely(extsz)) { | |
163 | if ((temp = do_mod(offset_fsb, extsz))) | |
164 | resaligned += temp; | |
165 | if ((temp = do_mod(resaligned, extsz))) | |
166 | resaligned += extsz - temp; | |
167 | } | |
168 | ||
169 | if (unlikely(rt)) { | |
170 | resrtextents = qblocks = resaligned; | |
171 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
172 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
173 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
174 | } else { | |
175 | resrtextents = 0; | |
dd9f438e | 176 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
177 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
178 | } | |
1da177e4 LT |
179 | |
180 | /* | |
06d10dd9 | 181 | * Allocate and setup the transaction |
1da177e4 | 182 | */ |
1da177e4 | 183 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); |
3d3c8b52 JL |
184 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
185 | resblks, resrtextents); | |
1da177e4 | 186 | /* |
06d10dd9 | 187 | * Check for running out of space, note: need lock to return |
1da177e4 | 188 | */ |
507630b2 | 189 | if (error) { |
1da177e4 | 190 | xfs_trans_cancel(tp, 0); |
b474c7ae | 191 | return error; |
507630b2 DC |
192 | } |
193 | ||
1da177e4 | 194 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 195 | |
7d095257 | 196 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 197 | if (error) |
507630b2 | 198 | goto out_trans_cancel; |
1da177e4 | 199 | |
ddc3415a | 200 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 201 | |
1da177e4 | 202 | /* |
3070451e CH |
203 | * From this point onwards we overwrite the imap pointer that the |
204 | * caller gave to us. | |
1da177e4 | 205 | */ |
9d87c319 | 206 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 207 | nimaps = 1; |
d531d91d CH |
208 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
209 | XFS_BMAPI_PREALLOC, &firstfsb, 0, | |
210 | imap, &nimaps, &free_list); | |
06d10dd9 | 211 | if (error) |
507630b2 | 212 | goto out_bmap_cancel; |
1da177e4 LT |
213 | |
214 | /* | |
06d10dd9 | 215 | * Complete the transaction |
1da177e4 | 216 | */ |
f7c99b6f | 217 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
06d10dd9 | 218 | if (error) |
507630b2 | 219 | goto out_bmap_cancel; |
1c72bf90 | 220 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
06d10dd9 | 221 | if (error) |
507630b2 | 222 | goto out_unlock; |
1da177e4 | 223 | |
06d10dd9 NS |
224 | /* |
225 | * Copy any maps to caller's array and return any error. | |
226 | */ | |
1da177e4 | 227 | if (nimaps == 0) { |
2451337d | 228 | error = -ENOSPC; |
507630b2 | 229 | goto out_unlock; |
572d95f4 NS |
230 | } |
231 | ||
507630b2 | 232 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 233 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 234 | |
507630b2 DC |
235 | out_unlock: |
236 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
237 | return error; | |
1da177e4 | 238 | |
507630b2 | 239 | out_bmap_cancel: |
1da177e4 | 240 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 241 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 242 | out_trans_cancel: |
1da177e4 | 243 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
507630b2 | 244 | goto out_unlock; |
1da177e4 LT |
245 | } |
246 | ||
dd9f438e | 247 | /* |
8de2bf93 DC |
248 | * If the caller is doing a write at the end of the file, then extend the |
249 | * allocation out to the file system's write iosize. We clean up any extra | |
250 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
251 | * |
252 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
253 | * preallocation as it it not needed. | |
dd9f438e NS |
254 | */ |
255 | STATIC int | |
256 | xfs_iomap_eof_want_preallocate( | |
257 | xfs_mount_t *mp, | |
541d7d3c | 258 | xfs_inode_t *ip, |
dd9f438e NS |
259 | xfs_off_t offset, |
260 | size_t count, | |
dd9f438e NS |
261 | xfs_bmbt_irec_t *imap, |
262 | int nimaps, | |
263 | int *prealloc) | |
264 | { | |
265 | xfs_fileoff_t start_fsb; | |
266 | xfs_filblks_t count_fsb; | |
267 | xfs_fsblock_t firstblock; | |
268 | int n, error, imaps; | |
055388a3 | 269 | int found_delalloc = 0; |
dd9f438e NS |
270 | |
271 | *prealloc = 0; | |
ce7ae151 | 272 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
273 | return 0; |
274 | ||
133eeb17 DC |
275 | /* |
276 | * If the file is smaller than the minimum prealloc and we are using | |
277 | * dynamic preallocation, don't do any preallocation at all as it is | |
278 | * likely this is the only write to the file that is going to be done. | |
279 | */ | |
280 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
281 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)) | |
282 | return 0; | |
283 | ||
dd9f438e NS |
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 | ||
133eeb17 DC |
346 | /* If the file is small, then use the minimum prealloc */ |
347 | if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign)) | |
348 | return 0; | |
349 | ||
a1e16c26 DC |
350 | /* |
351 | * As we write multiple pages, the offset will always align to the | |
352 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
353 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
354 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
355 | * point to the block prior to the EOF block and not the hole that maps | |
356 | * directly at @offset. | |
357 | */ | |
358 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
359 | if (start_fsb) | |
360 | start_fsb--; | |
361 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
362 | if (error) | |
363 | return 0; | |
364 | ||
365 | ASSERT(imaps == 1); | |
366 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
367 | return 0; | |
368 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 369 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
370 | return XFS_B_TO_FSB(mp, offset); |
371 | } | |
372 | ||
76a4202a BF |
373 | STATIC bool |
374 | xfs_quota_need_throttle( | |
375 | struct xfs_inode *ip, | |
376 | int type, | |
377 | xfs_fsblock_t alloc_blocks) | |
378 | { | |
379 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
380 | ||
381 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
382 | return false; | |
383 | ||
384 | /* no hi watermark, no throttle */ | |
385 | if (!dq->q_prealloc_hi_wmark) | |
386 | return false; | |
387 | ||
388 | /* under the lo watermark, no throttle */ | |
389 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
390 | return false; | |
391 | ||
392 | return true; | |
393 | } | |
394 | ||
395 | STATIC void | |
396 | xfs_quota_calc_throttle( | |
397 | struct xfs_inode *ip, | |
398 | int type, | |
399 | xfs_fsblock_t *qblocks, | |
f074051f BF |
400 | int *qshift, |
401 | int64_t *qfreesp) | |
76a4202a BF |
402 | { |
403 | int64_t freesp; | |
404 | int shift = 0; | |
405 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
406 | ||
407 | /* over hi wmark, squash the prealloc completely */ | |
408 | if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
409 | *qblocks = 0; | |
f074051f | 410 | *qfreesp = 0; |
76a4202a BF |
411 | return; |
412 | } | |
413 | ||
414 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
415 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
416 | shift = 2; | |
417 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
418 | shift += 2; | |
419 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
420 | shift += 2; | |
421 | } | |
422 | ||
f074051f BF |
423 | if (freesp < *qfreesp) |
424 | *qfreesp = freesp; | |
425 | ||
76a4202a BF |
426 | /* only overwrite the throttle values if we are more aggressive */ |
427 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
428 | *qblocks = freesp; | |
429 | *qshift = shift; | |
430 | } | |
431 | } | |
432 | ||
055388a3 DC |
433 | /* |
434 | * If we don't have a user specified preallocation size, dynamically increase | |
435 | * the preallocation size as the size of the file grows. Cap the maximum size | |
436 | * at a single extent or less if the filesystem is near full. The closer the | |
437 | * filesystem is to full, the smaller the maximum prealocation. | |
438 | */ | |
439 | STATIC xfs_fsblock_t | |
440 | xfs_iomap_prealloc_size( | |
441 | struct xfs_mount *mp, | |
a1e16c26 DC |
442 | struct xfs_inode *ip, |
443 | xfs_off_t offset, | |
444 | struct xfs_bmbt_irec *imap, | |
445 | int nimaps) | |
055388a3 DC |
446 | { |
447 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
448 | int shift = 0; |
449 | int64_t freesp; | |
76a4202a BF |
450 | xfs_fsblock_t qblocks; |
451 | int qshift = 0; | |
055388a3 | 452 | |
a1e16c26 DC |
453 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
454 | imap, nimaps); | |
3c58b5f8 BF |
455 | if (!alloc_blocks) |
456 | goto check_writeio; | |
76a4202a | 457 | qblocks = alloc_blocks; |
3c58b5f8 | 458 | |
c9bdbdc0 BF |
459 | /* |
460 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
461 | * to the nearest power of two value after throttling. To prevent the | |
462 | * round down from unconditionally reducing the maximum supported prealloc | |
463 | * size, we round up first, apply appropriate throttling, round down and | |
464 | * cap the value to MAXEXTLEN. | |
465 | */ | |
466 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
467 | alloc_blocks); | |
3c58b5f8 BF |
468 | |
469 | xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); | |
470 | freesp = mp->m_sb.sb_fdblocks; | |
471 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { | |
472 | shift = 2; | |
473 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
474 | shift++; | |
475 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
476 | shift++; | |
477 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
478 | shift++; | |
479 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
480 | shift++; | |
055388a3 | 481 | } |
76a4202a BF |
482 | |
483 | /* | |
f074051f BF |
484 | * Check each quota to cap the prealloc size, provide a shift value to |
485 | * throttle with and adjust amount of available space. | |
76a4202a BF |
486 | */ |
487 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
f074051f BF |
488 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift, |
489 | &freesp); | |
76a4202a | 490 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) |
f074051f BF |
491 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift, |
492 | &freesp); | |
76a4202a | 493 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) |
f074051f BF |
494 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift, |
495 | &freesp); | |
76a4202a BF |
496 | |
497 | /* | |
498 | * The final prealloc size is set to the minimum of free space available | |
499 | * in each of the quotas and the overall filesystem. | |
500 | * | |
501 | * The shift throttle value is set to the maximum value as determined by | |
502 | * the global low free space values and per-quota low free space values. | |
503 | */ | |
504 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
505 | shift = MAX(shift, qshift); | |
506 | ||
3c58b5f8 BF |
507 | if (shift) |
508 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
509 | /* |
510 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
511 | * alloc_blocks = 0. | |
512 | */ | |
513 | if (alloc_blocks) | |
514 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
515 | if (alloc_blocks > MAXEXTLEN) | |
516 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
517 | |
518 | /* | |
519 | * If we are still trying to allocate more space than is | |
520 | * available, squash the prealloc hard. This can happen if we | |
521 | * have a large file on a small filesystem and the above | |
522 | * lowspace thresholds are smaller than MAXEXTLEN. | |
523 | */ | |
524 | while (alloc_blocks && alloc_blocks >= freesp) | |
525 | alloc_blocks >>= 4; | |
055388a3 | 526 | |
3c58b5f8 | 527 | check_writeio: |
055388a3 DC |
528 | if (alloc_blocks < mp->m_writeio_blocks) |
529 | alloc_blocks = mp->m_writeio_blocks; | |
530 | ||
19cb7e38 BF |
531 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
532 | mp->m_writeio_blocks); | |
533 | ||
055388a3 DC |
534 | return alloc_blocks; |
535 | } | |
536 | ||
a206c817 | 537 | int |
1da177e4 LT |
538 | xfs_iomap_write_delay( |
539 | xfs_inode_t *ip, | |
f403b7f4 | 540 | xfs_off_t offset, |
1da177e4 | 541 | size_t count, |
405f8042 | 542 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
543 | { |
544 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
545 | xfs_fileoff_t offset_fsb; |
546 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
547 | xfs_off_t aligned_offset; |
548 | xfs_fileoff_t ioalign; | |
dd9f438e | 549 | xfs_extlen_t extsz; |
1da177e4 | 550 | int nimaps; |
1da177e4 | 551 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 552 | int prealloc; |
dd9f438e | 553 | int error; |
1da177e4 | 554 | |
579aa9ca | 555 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
556 | |
557 | /* | |
558 | * Make sure that the dquots are there. This doesn't hold | |
559 | * the ilock across a disk read. | |
560 | */ | |
7d095257 | 561 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 | 562 | if (error) |
b474c7ae | 563 | return error; |
1da177e4 | 564 | |
957d0ebe | 565 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
566 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
567 | ||
9f6c92b9 | 568 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 569 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
570 | if (error) |
571 | return error; | |
1da177e4 | 572 | |
8de2bf93 | 573 | retry: |
dd9f438e | 574 | if (prealloc) { |
a1e16c26 DC |
575 | xfs_fsblock_t alloc_blocks; |
576 | ||
577 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
578 | XFS_WRITE_IMAPS); | |
055388a3 | 579 | |
1da177e4 LT |
580 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
581 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 582 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
583 | } else { |
584 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 585 | } |
1da177e4 | 586 | |
dd9f438e | 587 | if (prealloc || extsz) { |
9f6c92b9 | 588 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 589 | if (error) |
1da177e4 | 590 | return error; |
1da177e4 | 591 | } |
dd9f438e | 592 | |
3ed9116e DC |
593 | /* |
594 | * Make sure preallocation does not create extents beyond the range we | |
595 | * actually support in this filesystem. | |
596 | */ | |
d2c28191 DC |
597 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
598 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
599 | |
600 | ASSERT(last_fsb > offset_fsb); | |
601 | ||
dd9f438e | 602 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
603 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
604 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
605 | switch (error) { |
606 | case 0: | |
2451337d DC |
607 | case -ENOSPC: |
608 | case -EDQUOT: | |
055388a3 DC |
609 | break; |
610 | default: | |
b474c7ae | 611 | return error; |
055388a3 | 612 | } |
dd9f438e | 613 | |
1da177e4 | 614 | /* |
9aa05000 | 615 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 616 | * without EOF preallocation. |
1da177e4 LT |
617 | */ |
618 | if (nimaps == 0) { | |
0b1b213f | 619 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
620 | if (prealloc) { |
621 | prealloc = 0; | |
622 | error = 0; | |
623 | goto retry; | |
055388a3 | 624 | } |
2451337d | 625 | return error ? error : -ENOSPC; |
1da177e4 LT |
626 | } |
627 | ||
86c4d623 | 628 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 629 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 630 | |
27b52867 BF |
631 | /* |
632 | * Tag the inode as speculatively preallocated so we can reclaim this | |
633 | * space on demand, if necessary. | |
634 | */ | |
635 | if (prealloc) | |
636 | xfs_inode_set_eofblocks_tag(ip); | |
637 | ||
dd9f438e | 638 | *ret_imap = imap[0]; |
1da177e4 LT |
639 | return 0; |
640 | } | |
641 | ||
642 | /* | |
643 | * Pass in a delayed allocate extent, convert it to real extents; | |
644 | * return to the caller the extent we create which maps on top of | |
645 | * the originating callers request. | |
646 | * | |
647 | * Called without a lock on the inode. | |
e4143a1c DC |
648 | * |
649 | * We no longer bother to look at the incoming map - all we have to | |
650 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 651 | */ |
a206c817 | 652 | int |
1da177e4 LT |
653 | xfs_iomap_write_allocate( |
654 | xfs_inode_t *ip, | |
f403b7f4 | 655 | xfs_off_t offset, |
405f8042 | 656 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
657 | { |
658 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
659 | xfs_fileoff_t offset_fsb, last_block; |
660 | xfs_fileoff_t end_fsb, map_start_fsb; | |
661 | xfs_fsblock_t first_block; | |
662 | xfs_bmap_free_t free_list; | |
663 | xfs_filblks_t count_fsb; | |
1da177e4 | 664 | xfs_trans_t *tp; |
e4143a1c | 665 | int nimaps, committed; |
1da177e4 LT |
666 | int error = 0; |
667 | int nres; | |
668 | ||
1da177e4 LT |
669 | /* |
670 | * Make sure that the dquots are there. | |
671 | */ | |
7d095257 CH |
672 | error = xfs_qm_dqattach(ip, 0); |
673 | if (error) | |
b474c7ae | 674 | return error; |
1da177e4 | 675 | |
24e17b5f | 676 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
677 | count_fsb = imap->br_blockcount; |
678 | map_start_fsb = imap->br_startoff; | |
1da177e4 LT |
679 | |
680 | XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); | |
681 | ||
682 | while (count_fsb != 0) { | |
683 | /* | |
684 | * Set up a transaction with which to allocate the | |
685 | * backing store for the file. Do allocations in a | |
686 | * loop until we get some space in the range we are | |
687 | * interested in. The other space that might be allocated | |
688 | * is in the delayed allocation extent on which we sit | |
689 | * but before our buffer starts. | |
690 | */ | |
691 | ||
692 | nimaps = 0; | |
693 | while (nimaps == 0) { | |
694 | tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE); | |
84e1e99f | 695 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 | 696 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
3d3c8b52 JL |
697 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
698 | nres, 0); | |
1da177e4 LT |
699 | if (error) { |
700 | xfs_trans_cancel(tp, 0); | |
b474c7ae | 701 | return error; |
1da177e4 LT |
702 | } |
703 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 704 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 705 | |
9d87c319 | 706 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 707 | |
1da177e4 | 708 | /* |
e4143a1c DC |
709 | * it is possible that the extents have changed since |
710 | * we did the read call as we dropped the ilock for a | |
711 | * while. We have to be careful about truncates or hole | |
712 | * punchs here - we are not allowed to allocate | |
713 | * non-delalloc blocks here. | |
714 | * | |
715 | * The only protection against truncation is the pages | |
716 | * for the range we are being asked to convert are | |
717 | * locked and hence a truncate will block on them | |
718 | * first. | |
719 | * | |
720 | * As a result, if we go beyond the range we really | |
721 | * need and hit an delalloc extent boundary followed by | |
722 | * a hole while we have excess blocks in the map, we | |
723 | * will fill the hole incorrectly and overrun the | |
724 | * transaction reservation. | |
725 | * | |
726 | * Using a single map prevents this as we are forced to | |
727 | * check each map we look for overlap with the desired | |
728 | * range and abort as soon as we find it. Also, given | |
729 | * that we only return a single map, having one beyond | |
730 | * what we can return is probably a bit silly. | |
731 | * | |
732 | * We also need to check that we don't go beyond EOF; | |
733 | * this is a truncate optimisation as a truncate sets | |
734 | * the new file size before block on the pages we | |
735 | * currently have locked under writeback. Because they | |
736 | * are about to be tossed, we don't need to write them | |
737 | * back.... | |
1da177e4 | 738 | */ |
e4143a1c | 739 | nimaps = 1; |
ce7ae151 | 740 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7fb2cd4d | 741 | error = xfs_bmap_last_offset(ip, &last_block, |
7c9ef85c DC |
742 | XFS_DATA_FORK); |
743 | if (error) | |
744 | goto trans_cancel; | |
745 | ||
1da177e4 LT |
746 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
747 | if ((map_start_fsb + count_fsb) > last_block) { | |
748 | count_fsb = last_block - map_start_fsb; | |
749 | if (count_fsb == 0) { | |
2451337d | 750 | error = -EAGAIN; |
1da177e4 LT |
751 | goto trans_cancel; |
752 | } | |
753 | } | |
754 | ||
3070451e | 755 | /* |
3070451e CH |
756 | * From this point onwards we overwrite the imap |
757 | * pointer that the caller gave to us. | |
758 | */ | |
c0dc7828 | 759 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
cf11da9c | 760 | count_fsb, 0, |
2455881c | 761 | &first_block, 1, |
c0dc7828 | 762 | imap, &nimaps, &free_list); |
1da177e4 LT |
763 | if (error) |
764 | goto trans_cancel; | |
765 | ||
f7c99b6f | 766 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
767 | if (error) |
768 | goto trans_cancel; | |
769 | ||
1c72bf90 | 770 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
771 | if (error) |
772 | goto error0; | |
773 | ||
774 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
775 | } | |
776 | ||
777 | /* | |
778 | * See if we were able to allocate an extent that | |
779 | * covers at least part of the callers request | |
780 | */ | |
3070451e | 781 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 782 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 783 | |
3070451e CH |
784 | if ((offset_fsb >= imap->br_startoff) && |
785 | (offset_fsb < (imap->br_startoff + | |
786 | imap->br_blockcount))) { | |
e4143a1c DC |
787 | XFS_STATS_INC(xs_xstrat_quick); |
788 | return 0; | |
1da177e4 LT |
789 | } |
790 | ||
e4143a1c DC |
791 | /* |
792 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
793 | * file, just surrounding data, try again. |
794 | */ | |
3070451e CH |
795 | count_fsb -= imap->br_blockcount; |
796 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
797 | } |
798 | ||
799 | trans_cancel: | |
800 | xfs_bmap_cancel(&free_list); | |
801 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
802 | error0: | |
803 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
b474c7ae | 804 | return error; |
1da177e4 LT |
805 | } |
806 | ||
807 | int | |
808 | xfs_iomap_write_unwritten( | |
809 | xfs_inode_t *ip, | |
f403b7f4 | 810 | xfs_off_t offset, |
1da177e4 LT |
811 | size_t count) |
812 | { | |
813 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
814 | xfs_fileoff_t offset_fsb; |
815 | xfs_filblks_t count_fsb; | |
816 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
817 | xfs_fsblock_t firstfsb; |
818 | int nimaps; | |
819 | xfs_trans_t *tp; | |
820 | xfs_bmbt_irec_t imap; | |
821 | xfs_bmap_free_t free_list; | |
84803fb7 | 822 | xfs_fsize_t i_size; |
dd9f438e | 823 | uint resblks; |
1da177e4 LT |
824 | int committed; |
825 | int error; | |
1da177e4 | 826 | |
0b1b213f | 827 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
828 | |
829 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
830 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
831 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
832 | ||
4ddd8bb1 LM |
833 | /* |
834 | * Reserve enough blocks in this transaction for two complete extent | |
835 | * btree splits. We may be converting the middle part of an unwritten | |
836 | * extent and in this case we will insert two new extents in the btree | |
837 | * each of which could cause a full split. | |
838 | * | |
839 | * This reservation amount will be used in the first call to | |
840 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
841 | * rest of the operation. | |
842 | */ | |
dd9f438e | 843 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 844 | |
dd9f438e | 845 | do { |
1da177e4 LT |
846 | /* |
847 | * set up a transaction to convert the range of extents | |
848 | * from unwritten to real. Do allocations in a loop until | |
849 | * we have covered the range passed in. | |
80641dc6 CH |
850 | * |
851 | * Note that we open code the transaction allocation here | |
852 | * to pass KM_NOFS--we can't risk to recursing back into | |
853 | * the filesystem here as we might be asked to write out | |
854 | * the same inode that we complete here and might deadlock | |
855 | * on the iolock. | |
1da177e4 | 856 | */ |
d9457dc0 | 857 | sb_start_intwrite(mp->m_super); |
80641dc6 | 858 | tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS); |
d9457dc0 | 859 | tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT; |
3d3c8b52 JL |
860 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
861 | resblks, 0); | |
1da177e4 LT |
862 | if (error) { |
863 | xfs_trans_cancel(tp, 0); | |
b474c7ae | 864 | return error; |
1da177e4 LT |
865 | } |
866 | ||
867 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 868 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
869 | |
870 | /* | |
871 | * Modify the unwritten extent state of the buffer. | |
872 | */ | |
9d87c319 | 873 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 874 | nimaps = 1; |
c0dc7828 DC |
875 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
876 | XFS_BMAPI_CONVERT, &firstfsb, | |
b4e9181e | 877 | 1, &imap, &nimaps, &free_list); |
1da177e4 LT |
878 | if (error) |
879 | goto error_on_bmapi_transaction; | |
880 | ||
84803fb7 CH |
881 | /* |
882 | * Log the updated inode size as we go. We have to be careful | |
883 | * to only log it up to the actual write offset if it is | |
884 | * halfway into a block. | |
885 | */ | |
886 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
887 | if (i_size > offset + count) | |
888 | i_size = offset + count; | |
889 | ||
890 | i_size = xfs_new_eof(ip, i_size); | |
891 | if (i_size) { | |
892 | ip->i_d.di_size = i_size; | |
893 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
894 | } | |
895 | ||
896 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1da177e4 LT |
897 | if (error) |
898 | goto error_on_bmapi_transaction; | |
899 | ||
1c72bf90 | 900 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
901 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
902 | if (error) | |
b474c7ae | 903 | return error; |
572d95f4 | 904 | |
86c4d623 | 905 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 906 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
907 | |
908 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
909 | /* | |
910 | * The numblks_fsb value should always get | |
911 | * smaller, otherwise the loop is stuck. | |
912 | */ | |
913 | ASSERT(imap.br_blockcount); | |
914 | break; | |
915 | } | |
916 | offset_fsb += numblks_fsb; | |
917 | count_fsb -= numblks_fsb; | |
918 | } while (count_fsb > 0); | |
919 | ||
920 | return 0; | |
921 | ||
922 | error_on_bmapi_transaction: | |
923 | xfs_bmap_cancel(&free_list); | |
924 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT)); | |
925 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
b474c7ae | 926 | return error; |
1da177e4 | 927 | } |