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); | |
113 | return EFSCORRUPTED; | |
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; |
1da177e4 | 131 | int bmapi_flag; |
06d10dd9 | 132 | int quota_flag; |
1da177e4 LT |
133 | int rt; |
134 | xfs_trans_t *tp; | |
1da177e4 | 135 | xfs_bmap_free_t free_list; |
dd9f438e | 136 | uint qblocks, resblks, resrtextents; |
1da177e4 | 137 | int committed; |
dd9f438e | 138 | int error; |
1da177e4 | 139 | |
507630b2 | 140 | error = xfs_qm_dqattach(ip, 0); |
1da177e4 LT |
141 | if (error) |
142 | return XFS_ERROR(error); | |
143 | ||
dd9f438e | 144 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 145 | extsz = xfs_get_extsz_hint(ip); |
1da177e4 | 146 | |
957d0ebe DC |
147 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
148 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 149 | if ((offset + count) > XFS_ISIZE(ip)) { |
9f6c92b9 | 150 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 151 | if (error) |
507630b2 | 152 | return XFS_ERROR(error); |
1da177e4 | 153 | } else { |
405f8042 | 154 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 155 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
156 | imap->br_blockcount + |
157 | imap->br_startoff); | |
1da177e4 | 158 | } |
dd9f438e NS |
159 | count_fsb = last_fsb - offset_fsb; |
160 | ASSERT(count_fsb > 0); | |
161 | ||
162 | resaligned = count_fsb; | |
163 | if (unlikely(extsz)) { | |
164 | if ((temp = do_mod(offset_fsb, extsz))) | |
165 | resaligned += temp; | |
166 | if ((temp = do_mod(resaligned, extsz))) | |
167 | resaligned += extsz - temp; | |
168 | } | |
169 | ||
170 | if (unlikely(rt)) { | |
171 | resrtextents = qblocks = resaligned; | |
172 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
173 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
174 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
175 | } else { | |
176 | resrtextents = 0; | |
dd9f438e | 177 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
178 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
179 | } | |
1da177e4 LT |
180 | |
181 | /* | |
06d10dd9 | 182 | * Allocate and setup the transaction |
1da177e4 | 183 | */ |
1da177e4 | 184 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); |
3d3c8b52 JL |
185 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
186 | resblks, resrtextents); | |
1da177e4 | 187 | /* |
06d10dd9 | 188 | * Check for running out of space, note: need lock to return |
1da177e4 | 189 | */ |
507630b2 | 190 | if (error) { |
1da177e4 | 191 | xfs_trans_cancel(tp, 0); |
507630b2 DC |
192 | return XFS_ERROR(error); |
193 | } | |
194 | ||
1da177e4 | 195 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 196 | |
7d095257 | 197 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 198 | if (error) |
507630b2 | 199 | goto out_trans_cancel; |
1da177e4 | 200 | |
ddc3415a | 201 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 202 | |
c0dc7828 | 203 | bmapi_flag = 0; |
ce7ae151 | 204 | if (offset < XFS_ISIZE(ip) || extsz) |
1da177e4 LT |
205 | bmapi_flag |= XFS_BMAPI_PREALLOC; |
206 | ||
207 | /* | |
3070451e CH |
208 | * From this point onwards we overwrite the imap pointer that the |
209 | * caller gave to us. | |
1da177e4 | 210 | */ |
9d87c319 | 211 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 212 | nimaps = 1; |
c0dc7828 DC |
213 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag, |
214 | &firstfsb, 0, imap, &nimaps, &free_list); | |
06d10dd9 | 215 | if (error) |
507630b2 | 216 | goto out_bmap_cancel; |
1da177e4 LT |
217 | |
218 | /* | |
06d10dd9 | 219 | * Complete the transaction |
1da177e4 | 220 | */ |
f7c99b6f | 221 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
06d10dd9 | 222 | if (error) |
507630b2 | 223 | goto out_bmap_cancel; |
1c72bf90 | 224 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
06d10dd9 | 225 | if (error) |
507630b2 | 226 | goto out_unlock; |
1da177e4 | 227 | |
06d10dd9 NS |
228 | /* |
229 | * Copy any maps to caller's array and return any error. | |
230 | */ | |
1da177e4 | 231 | if (nimaps == 0) { |
507630b2 DC |
232 | error = XFS_ERROR(ENOSPC); |
233 | goto out_unlock; | |
572d95f4 NS |
234 | } |
235 | ||
507630b2 | 236 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 237 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 238 | |
507630b2 DC |
239 | out_unlock: |
240 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
241 | return error; | |
1da177e4 | 242 | |
507630b2 | 243 | out_bmap_cancel: |
1da177e4 | 244 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 245 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 246 | out_trans_cancel: |
1da177e4 | 247 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
507630b2 | 248 | goto out_unlock; |
1da177e4 LT |
249 | } |
250 | ||
dd9f438e | 251 | /* |
8de2bf93 DC |
252 | * If the caller is doing a write at the end of the file, then extend the |
253 | * allocation out to the file system's write iosize. We clean up any extra | |
254 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
255 | * |
256 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
257 | * preallocation as it it not needed. | |
dd9f438e NS |
258 | */ |
259 | STATIC int | |
260 | xfs_iomap_eof_want_preallocate( | |
261 | xfs_mount_t *mp, | |
541d7d3c | 262 | xfs_inode_t *ip, |
dd9f438e NS |
263 | xfs_off_t offset, |
264 | size_t count, | |
dd9f438e NS |
265 | xfs_bmbt_irec_t *imap, |
266 | int nimaps, | |
267 | int *prealloc) | |
268 | { | |
269 | xfs_fileoff_t start_fsb; | |
270 | xfs_filblks_t count_fsb; | |
271 | xfs_fsblock_t firstblock; | |
272 | int n, error, imaps; | |
055388a3 | 273 | int found_delalloc = 0; |
dd9f438e NS |
274 | |
275 | *prealloc = 0; | |
ce7ae151 | 276 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
277 | return 0; |
278 | ||
133eeb17 DC |
279 | /* |
280 | * If the file is smaller than the minimum prealloc and we are using | |
281 | * dynamic preallocation, don't do any preallocation at all as it is | |
282 | * likely this is the only write to the file that is going to be done. | |
283 | */ | |
284 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
285 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)) | |
286 | return 0; | |
287 | ||
dd9f438e NS |
288 | /* |
289 | * If there are any real blocks past eof, then don't | |
290 | * do any speculative allocation. | |
291 | */ | |
292 | start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1))); | |
32972383 | 293 | count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
dd9f438e NS |
294 | while (count_fsb > 0) { |
295 | imaps = nimaps; | |
3ddb8fa9 | 296 | firstblock = NULLFSBLOCK; |
5c8ed202 DC |
297 | error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps, |
298 | 0); | |
dd9f438e NS |
299 | if (error) |
300 | return error; | |
301 | for (n = 0; n < imaps; n++) { | |
302 | if ((imap[n].br_startblock != HOLESTARTBLOCK) && | |
303 | (imap[n].br_startblock != DELAYSTARTBLOCK)) | |
304 | return 0; | |
305 | start_fsb += imap[n].br_blockcount; | |
306 | count_fsb -= imap[n].br_blockcount; | |
055388a3 DC |
307 | |
308 | if (imap[n].br_startblock == DELAYSTARTBLOCK) | |
309 | found_delalloc = 1; | |
dd9f438e NS |
310 | } |
311 | } | |
055388a3 DC |
312 | if (!found_delalloc) |
313 | *prealloc = 1; | |
dd9f438e NS |
314 | return 0; |
315 | } | |
316 | ||
a1e16c26 DC |
317 | /* |
318 | * Determine the initial size of the preallocation. We are beyond the current | |
319 | * EOF here, but we need to take into account whether this is a sparse write or | |
320 | * an extending write when determining the preallocation size. Hence we need to | |
321 | * look up the extent that ends at the current write offset and use the result | |
322 | * to determine the preallocation size. | |
323 | * | |
324 | * If the extent is a hole, then preallocation is essentially disabled. | |
325 | * Otherwise we take the size of the preceeding data extent as the basis for the | |
326 | * preallocation size. If the size of the extent is greater than half the | |
327 | * maximum extent length, then use the current offset as the basis. This ensures | |
328 | * that for large files the preallocation size always extends to MAXEXTLEN | |
329 | * rather than falling short due to things like stripe unit/width alignment of | |
330 | * real extents. | |
331 | */ | |
e8108ced | 332 | STATIC xfs_fsblock_t |
a1e16c26 DC |
333 | xfs_iomap_eof_prealloc_initial_size( |
334 | struct xfs_mount *mp, | |
335 | struct xfs_inode *ip, | |
336 | xfs_off_t offset, | |
337 | xfs_bmbt_irec_t *imap, | |
338 | int nimaps) | |
339 | { | |
340 | xfs_fileoff_t start_fsb; | |
341 | int imaps = 1; | |
342 | int error; | |
343 | ||
344 | ASSERT(nimaps >= imaps); | |
345 | ||
346 | /* if we are using a specific prealloc size, return now */ | |
347 | if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) | |
348 | return 0; | |
349 | ||
133eeb17 DC |
350 | /* If the file is small, then use the minimum prealloc */ |
351 | if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign)) | |
352 | return 0; | |
353 | ||
a1e16c26 DC |
354 | /* |
355 | * As we write multiple pages, the offset will always align to the | |
356 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
357 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
358 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
359 | * point to the block prior to the EOF block and not the hole that maps | |
360 | * directly at @offset. | |
361 | */ | |
362 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
363 | if (start_fsb) | |
364 | start_fsb--; | |
365 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
366 | if (error) | |
367 | return 0; | |
368 | ||
369 | ASSERT(imaps == 1); | |
370 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
371 | return 0; | |
372 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 373 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
374 | return XFS_B_TO_FSB(mp, offset); |
375 | } | |
376 | ||
76a4202a BF |
377 | STATIC bool |
378 | xfs_quota_need_throttle( | |
379 | struct xfs_inode *ip, | |
380 | int type, | |
381 | xfs_fsblock_t alloc_blocks) | |
382 | { | |
383 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
384 | ||
385 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
386 | return false; | |
387 | ||
388 | /* no hi watermark, no throttle */ | |
389 | if (!dq->q_prealloc_hi_wmark) | |
390 | return false; | |
391 | ||
392 | /* under the lo watermark, no throttle */ | |
393 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
394 | return false; | |
395 | ||
396 | return true; | |
397 | } | |
398 | ||
399 | STATIC void | |
400 | xfs_quota_calc_throttle( | |
401 | struct xfs_inode *ip, | |
402 | int type, | |
403 | xfs_fsblock_t *qblocks, | |
404 | int *qshift) | |
405 | { | |
406 | int64_t freesp; | |
407 | int shift = 0; | |
408 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
409 | ||
410 | /* over hi wmark, squash the prealloc completely */ | |
411 | if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
412 | *qblocks = 0; | |
413 | return; | |
414 | } | |
415 | ||
416 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
417 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
418 | shift = 2; | |
419 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
420 | shift += 2; | |
421 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
422 | shift += 2; | |
423 | } | |
424 | ||
425 | /* only overwrite the throttle values if we are more aggressive */ | |
426 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
427 | *qblocks = freesp; | |
428 | *qshift = shift; | |
429 | } | |
430 | } | |
431 | ||
055388a3 DC |
432 | /* |
433 | * If we don't have a user specified preallocation size, dynamically increase | |
434 | * the preallocation size as the size of the file grows. Cap the maximum size | |
435 | * at a single extent or less if the filesystem is near full. The closer the | |
436 | * filesystem is to full, the smaller the maximum prealocation. | |
437 | */ | |
438 | STATIC xfs_fsblock_t | |
439 | xfs_iomap_prealloc_size( | |
440 | struct xfs_mount *mp, | |
a1e16c26 DC |
441 | struct xfs_inode *ip, |
442 | xfs_off_t offset, | |
443 | struct xfs_bmbt_irec *imap, | |
444 | int nimaps) | |
055388a3 DC |
445 | { |
446 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
447 | int shift = 0; |
448 | int64_t freesp; | |
76a4202a BF |
449 | xfs_fsblock_t qblocks; |
450 | int qshift = 0; | |
055388a3 | 451 | |
a1e16c26 DC |
452 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
453 | imap, nimaps); | |
3c58b5f8 BF |
454 | if (!alloc_blocks) |
455 | goto check_writeio; | |
76a4202a | 456 | qblocks = alloc_blocks; |
3c58b5f8 | 457 | |
c9bdbdc0 BF |
458 | /* |
459 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
460 | * to the nearest power of two value after throttling. To prevent the | |
461 | * round down from unconditionally reducing the maximum supported prealloc | |
462 | * size, we round up first, apply appropriate throttling, round down and | |
463 | * cap the value to MAXEXTLEN. | |
464 | */ | |
465 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
466 | alloc_blocks); | |
3c58b5f8 BF |
467 | |
468 | xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); | |
469 | freesp = mp->m_sb.sb_fdblocks; | |
470 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { | |
471 | shift = 2; | |
472 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
473 | shift++; | |
474 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
475 | shift++; | |
476 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
477 | shift++; | |
478 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
479 | shift++; | |
055388a3 | 480 | } |
76a4202a BF |
481 | |
482 | /* | |
483 | * Check each quota to cap the prealloc size and provide a shift | |
484 | * value to throttle with. | |
485 | */ | |
486 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
487 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift); | |
488 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) | |
489 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift); | |
490 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) | |
491 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift); | |
492 | ||
493 | /* | |
494 | * The final prealloc size is set to the minimum of free space available | |
495 | * in each of the quotas and the overall filesystem. | |
496 | * | |
497 | * The shift throttle value is set to the maximum value as determined by | |
498 | * the global low free space values and per-quota low free space values. | |
499 | */ | |
500 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
501 | shift = MAX(shift, qshift); | |
502 | ||
3c58b5f8 BF |
503 | if (shift) |
504 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
505 | /* |
506 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
507 | * alloc_blocks = 0. | |
508 | */ | |
509 | if (alloc_blocks) | |
510 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
511 | if (alloc_blocks > MAXEXTLEN) | |
512 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
513 | |
514 | /* | |
515 | * If we are still trying to allocate more space than is | |
516 | * available, squash the prealloc hard. This can happen if we | |
517 | * have a large file on a small filesystem and the above | |
518 | * lowspace thresholds are smaller than MAXEXTLEN. | |
519 | */ | |
520 | while (alloc_blocks && alloc_blocks >= freesp) | |
521 | alloc_blocks >>= 4; | |
055388a3 | 522 | |
3c58b5f8 | 523 | check_writeio: |
055388a3 DC |
524 | if (alloc_blocks < mp->m_writeio_blocks) |
525 | alloc_blocks = mp->m_writeio_blocks; | |
526 | ||
19cb7e38 BF |
527 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
528 | mp->m_writeio_blocks); | |
529 | ||
055388a3 DC |
530 | return alloc_blocks; |
531 | } | |
532 | ||
a206c817 | 533 | int |
1da177e4 LT |
534 | xfs_iomap_write_delay( |
535 | xfs_inode_t *ip, | |
f403b7f4 | 536 | xfs_off_t offset, |
1da177e4 | 537 | size_t count, |
405f8042 | 538 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
539 | { |
540 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
541 | xfs_fileoff_t offset_fsb; |
542 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
543 | xfs_off_t aligned_offset; |
544 | xfs_fileoff_t ioalign; | |
dd9f438e | 545 | xfs_extlen_t extsz; |
1da177e4 | 546 | int nimaps; |
1da177e4 | 547 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 548 | int prealloc; |
dd9f438e | 549 | int error; |
1da177e4 | 550 | |
579aa9ca | 551 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
552 | |
553 | /* | |
554 | * Make sure that the dquots are there. This doesn't hold | |
555 | * the ilock across a disk read. | |
556 | */ | |
7d095257 | 557 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 LT |
558 | if (error) |
559 | return XFS_ERROR(error); | |
560 | ||
957d0ebe | 561 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
562 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
563 | ||
9f6c92b9 | 564 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 565 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
566 | if (error) |
567 | return error; | |
1da177e4 | 568 | |
8de2bf93 | 569 | retry: |
dd9f438e | 570 | if (prealloc) { |
a1e16c26 DC |
571 | xfs_fsblock_t alloc_blocks; |
572 | ||
573 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
574 | XFS_WRITE_IMAPS); | |
055388a3 | 575 | |
1da177e4 LT |
576 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
577 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 578 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
579 | } else { |
580 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 581 | } |
1da177e4 | 582 | |
dd9f438e | 583 | if (prealloc || extsz) { |
9f6c92b9 | 584 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 585 | if (error) |
1da177e4 | 586 | return error; |
1da177e4 | 587 | } |
dd9f438e | 588 | |
3ed9116e DC |
589 | /* |
590 | * Make sure preallocation does not create extents beyond the range we | |
591 | * actually support in this filesystem. | |
592 | */ | |
d2c28191 DC |
593 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
594 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
595 | |
596 | ASSERT(last_fsb > offset_fsb); | |
597 | ||
dd9f438e | 598 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
599 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
600 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
601 | switch (error) { |
602 | case 0: | |
603 | case ENOSPC: | |
604 | case EDQUOT: | |
605 | break; | |
606 | default: | |
1da177e4 | 607 | return XFS_ERROR(error); |
055388a3 | 608 | } |
dd9f438e | 609 | |
1da177e4 | 610 | /* |
9aa05000 | 611 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 612 | * without EOF preallocation. |
1da177e4 LT |
613 | */ |
614 | if (nimaps == 0) { | |
0b1b213f | 615 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
616 | if (prealloc) { |
617 | prealloc = 0; | |
618 | error = 0; | |
619 | goto retry; | |
055388a3 | 620 | } |
9aa05000 | 621 | return XFS_ERROR(error ? error : ENOSPC); |
1da177e4 LT |
622 | } |
623 | ||
86c4d623 | 624 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 625 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 626 | |
27b52867 BF |
627 | /* |
628 | * Tag the inode as speculatively preallocated so we can reclaim this | |
629 | * space on demand, if necessary. | |
630 | */ | |
631 | if (prealloc) | |
632 | xfs_inode_set_eofblocks_tag(ip); | |
633 | ||
dd9f438e | 634 | *ret_imap = imap[0]; |
1da177e4 LT |
635 | return 0; |
636 | } | |
637 | ||
638 | /* | |
639 | * Pass in a delayed allocate extent, convert it to real extents; | |
640 | * return to the caller the extent we create which maps on top of | |
641 | * the originating callers request. | |
642 | * | |
643 | * Called without a lock on the inode. | |
e4143a1c DC |
644 | * |
645 | * We no longer bother to look at the incoming map - all we have to | |
646 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 647 | */ |
a206c817 | 648 | int |
1da177e4 LT |
649 | xfs_iomap_write_allocate( |
650 | xfs_inode_t *ip, | |
f403b7f4 | 651 | xfs_off_t offset, |
405f8042 | 652 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
653 | { |
654 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
655 | xfs_fileoff_t offset_fsb, last_block; |
656 | xfs_fileoff_t end_fsb, map_start_fsb; | |
657 | xfs_fsblock_t first_block; | |
658 | xfs_bmap_free_t free_list; | |
659 | xfs_filblks_t count_fsb; | |
1da177e4 | 660 | xfs_trans_t *tp; |
e4143a1c | 661 | int nimaps, committed; |
1da177e4 LT |
662 | int error = 0; |
663 | int nres; | |
664 | ||
1da177e4 LT |
665 | /* |
666 | * Make sure that the dquots are there. | |
667 | */ | |
7d095257 CH |
668 | error = xfs_qm_dqattach(ip, 0); |
669 | if (error) | |
1da177e4 LT |
670 | return XFS_ERROR(error); |
671 | ||
24e17b5f | 672 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
673 | count_fsb = imap->br_blockcount; |
674 | map_start_fsb = imap->br_startoff; | |
1da177e4 LT |
675 | |
676 | XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); | |
677 | ||
678 | while (count_fsb != 0) { | |
679 | /* | |
680 | * Set up a transaction with which to allocate the | |
681 | * backing store for the file. Do allocations in a | |
682 | * loop until we get some space in the range we are | |
683 | * interested in. The other space that might be allocated | |
684 | * is in the delayed allocation extent on which we sit | |
685 | * but before our buffer starts. | |
686 | */ | |
687 | ||
688 | nimaps = 0; | |
689 | while (nimaps == 0) { | |
690 | tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE); | |
84e1e99f | 691 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 | 692 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
3d3c8b52 JL |
693 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
694 | nres, 0); | |
1da177e4 LT |
695 | if (error) { |
696 | xfs_trans_cancel(tp, 0); | |
697 | return XFS_ERROR(error); | |
698 | } | |
699 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 700 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 701 | |
9d87c319 | 702 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 703 | |
1da177e4 | 704 | /* |
e4143a1c DC |
705 | * it is possible that the extents have changed since |
706 | * we did the read call as we dropped the ilock for a | |
707 | * while. We have to be careful about truncates or hole | |
708 | * punchs here - we are not allowed to allocate | |
709 | * non-delalloc blocks here. | |
710 | * | |
711 | * The only protection against truncation is the pages | |
712 | * for the range we are being asked to convert are | |
713 | * locked and hence a truncate will block on them | |
714 | * first. | |
715 | * | |
716 | * As a result, if we go beyond the range we really | |
717 | * need and hit an delalloc extent boundary followed by | |
718 | * a hole while we have excess blocks in the map, we | |
719 | * will fill the hole incorrectly and overrun the | |
720 | * transaction reservation. | |
721 | * | |
722 | * Using a single map prevents this as we are forced to | |
723 | * check each map we look for overlap with the desired | |
724 | * range and abort as soon as we find it. Also, given | |
725 | * that we only return a single map, having one beyond | |
726 | * what we can return is probably a bit silly. | |
727 | * | |
728 | * We also need to check that we don't go beyond EOF; | |
729 | * this is a truncate optimisation as a truncate sets | |
730 | * the new file size before block on the pages we | |
731 | * currently have locked under writeback. Because they | |
732 | * are about to be tossed, we don't need to write them | |
733 | * back.... | |
1da177e4 | 734 | */ |
e4143a1c | 735 | nimaps = 1; |
ce7ae151 | 736 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7c9ef85c DC |
737 | error = xfs_bmap_last_offset(NULL, ip, &last_block, |
738 | XFS_DATA_FORK); | |
739 | if (error) | |
740 | goto trans_cancel; | |
741 | ||
1da177e4 LT |
742 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
743 | if ((map_start_fsb + count_fsb) > last_block) { | |
744 | count_fsb = last_block - map_start_fsb; | |
745 | if (count_fsb == 0) { | |
746 | error = EAGAIN; | |
747 | goto trans_cancel; | |
748 | } | |
749 | } | |
750 | ||
3070451e | 751 | /* |
3070451e CH |
752 | * From this point onwards we overwrite the imap |
753 | * pointer that the caller gave to us. | |
754 | */ | |
c0dc7828 | 755 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
2455881c DC |
756 | count_fsb, |
757 | XFS_BMAPI_STACK_SWITCH, | |
758 | &first_block, 1, | |
c0dc7828 | 759 | imap, &nimaps, &free_list); |
1da177e4 LT |
760 | if (error) |
761 | goto trans_cancel; | |
762 | ||
f7c99b6f | 763 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
764 | if (error) |
765 | goto trans_cancel; | |
766 | ||
1c72bf90 | 767 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
768 | if (error) |
769 | goto error0; | |
770 | ||
771 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
772 | } | |
773 | ||
774 | /* | |
775 | * See if we were able to allocate an extent that | |
776 | * covers at least part of the callers request | |
777 | */ | |
3070451e | 778 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 779 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 780 | |
3070451e CH |
781 | if ((offset_fsb >= imap->br_startoff) && |
782 | (offset_fsb < (imap->br_startoff + | |
783 | imap->br_blockcount))) { | |
e4143a1c DC |
784 | XFS_STATS_INC(xs_xstrat_quick); |
785 | return 0; | |
1da177e4 LT |
786 | } |
787 | ||
e4143a1c DC |
788 | /* |
789 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
790 | * file, just surrounding data, try again. |
791 | */ | |
3070451e CH |
792 | count_fsb -= imap->br_blockcount; |
793 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
794 | } |
795 | ||
796 | trans_cancel: | |
797 | xfs_bmap_cancel(&free_list); | |
798 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
799 | error0: | |
800 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
801 | return XFS_ERROR(error); | |
802 | } | |
803 | ||
804 | int | |
805 | xfs_iomap_write_unwritten( | |
806 | xfs_inode_t *ip, | |
f403b7f4 | 807 | xfs_off_t offset, |
1da177e4 LT |
808 | size_t count) |
809 | { | |
810 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
811 | xfs_fileoff_t offset_fsb; |
812 | xfs_filblks_t count_fsb; | |
813 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
814 | xfs_fsblock_t firstfsb; |
815 | int nimaps; | |
816 | xfs_trans_t *tp; | |
817 | xfs_bmbt_irec_t imap; | |
818 | xfs_bmap_free_t free_list; | |
84803fb7 | 819 | xfs_fsize_t i_size; |
dd9f438e | 820 | uint resblks; |
1da177e4 LT |
821 | int committed; |
822 | int error; | |
1da177e4 | 823 | |
0b1b213f | 824 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
825 | |
826 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
827 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
828 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
829 | ||
4ddd8bb1 LM |
830 | /* |
831 | * Reserve enough blocks in this transaction for two complete extent | |
832 | * btree splits. We may be converting the middle part of an unwritten | |
833 | * extent and in this case we will insert two new extents in the btree | |
834 | * each of which could cause a full split. | |
835 | * | |
836 | * This reservation amount will be used in the first call to | |
837 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
838 | * rest of the operation. | |
839 | */ | |
dd9f438e | 840 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 841 | |
dd9f438e | 842 | do { |
1da177e4 LT |
843 | /* |
844 | * set up a transaction to convert the range of extents | |
845 | * from unwritten to real. Do allocations in a loop until | |
846 | * we have covered the range passed in. | |
80641dc6 CH |
847 | * |
848 | * Note that we open code the transaction allocation here | |
849 | * to pass KM_NOFS--we can't risk to recursing back into | |
850 | * the filesystem here as we might be asked to write out | |
851 | * the same inode that we complete here and might deadlock | |
852 | * on the iolock. | |
1da177e4 | 853 | */ |
d9457dc0 | 854 | sb_start_intwrite(mp->m_super); |
80641dc6 | 855 | tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS); |
d9457dc0 | 856 | tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT; |
3d3c8b52 JL |
857 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
858 | resblks, 0); | |
1da177e4 LT |
859 | if (error) { |
860 | xfs_trans_cancel(tp, 0); | |
572d95f4 | 861 | return XFS_ERROR(error); |
1da177e4 LT |
862 | } |
863 | ||
864 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 865 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
866 | |
867 | /* | |
868 | * Modify the unwritten extent state of the buffer. | |
869 | */ | |
9d87c319 | 870 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 871 | nimaps = 1; |
c0dc7828 DC |
872 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
873 | XFS_BMAPI_CONVERT, &firstfsb, | |
b4e9181e | 874 | 1, &imap, &nimaps, &free_list); |
1da177e4 LT |
875 | if (error) |
876 | goto error_on_bmapi_transaction; | |
877 | ||
84803fb7 CH |
878 | /* |
879 | * Log the updated inode size as we go. We have to be careful | |
880 | * to only log it up to the actual write offset if it is | |
881 | * halfway into a block. | |
882 | */ | |
883 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
884 | if (i_size > offset + count) | |
885 | i_size = offset + count; | |
886 | ||
887 | i_size = xfs_new_eof(ip, i_size); | |
888 | if (i_size) { | |
889 | ip->i_d.di_size = i_size; | |
890 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
891 | } | |
892 | ||
893 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1da177e4 LT |
894 | if (error) |
895 | goto error_on_bmapi_transaction; | |
896 | ||
1c72bf90 | 897 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
898 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
899 | if (error) | |
572d95f4 NS |
900 | return XFS_ERROR(error); |
901 | ||
86c4d623 | 902 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 903 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
904 | |
905 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
906 | /* | |
907 | * The numblks_fsb value should always get | |
908 | * smaller, otherwise the loop is stuck. | |
909 | */ | |
910 | ASSERT(imap.br_blockcount); | |
911 | break; | |
912 | } | |
913 | offset_fsb += numblks_fsb; | |
914 | count_fsb -= numblks_fsb; | |
915 | } while (count_fsb > 0); | |
916 | ||
917 | return 0; | |
918 | ||
919 | error_on_bmapi_transaction: | |
920 | xfs_bmap_cancel(&free_list); | |
921 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT)); | |
922 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 LT |
923 | return XFS_ERROR(error); |
924 | } |