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" |
70a9883c | 21 | #include "xfs_shared.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
24 | #include "xfs_sb.h" | |
25 | #include "xfs_ag.h" | |
1da177e4 | 26 | #include "xfs_alloc.h" |
1da177e4 LT |
27 | #include "xfs_quota.h" |
28 | #include "xfs_mount.h" | |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
a844f451 | 30 | #include "xfs_alloc_btree.h" |
1da177e4 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
32 | #include "xfs_dinode.h" |
33 | #include "xfs_inode.h" | |
84803fb7 | 34 | #include "xfs_inode_item.h" |
a844f451 | 35 | #include "xfs_btree.h" |
1da177e4 | 36 | #include "xfs_bmap.h" |
68988114 | 37 | #include "xfs_bmap_util.h" |
1da177e4 LT |
38 | #include "xfs_rtalloc.h" |
39 | #include "xfs_error.h" | |
40 | #include "xfs_itable.h" | |
1da177e4 LT |
41 | #include "xfs_attr.h" |
42 | #include "xfs_buf_item.h" | |
43 | #include "xfs_trans_space.h" | |
1da177e4 | 44 | #include "xfs_iomap.h" |
0b1b213f | 45 | #include "xfs_trace.h" |
27b52867 | 46 | #include "xfs_icache.h" |
76a4202a BF |
47 | #include "xfs_dquot_item.h" |
48 | #include "xfs_dquot.h" | |
1da177e4 | 49 | |
1da177e4 LT |
50 | |
51 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
52 | << mp->m_writeio_log) | |
1da177e4 LT |
53 | #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP |
54 | ||
dd9f438e NS |
55 | STATIC int |
56 | xfs_iomap_eof_align_last_fsb( | |
57 | xfs_mount_t *mp, | |
541d7d3c | 58 | xfs_inode_t *ip, |
dd9f438e NS |
59 | xfs_extlen_t extsize, |
60 | xfs_fileoff_t *last_fsb) | |
61 | { | |
62 | xfs_fileoff_t new_last_fsb = 0; | |
bf322d98 | 63 | xfs_extlen_t align = 0; |
dd9f438e NS |
64 | int eof, error; |
65 | ||
bf322d98 CH |
66 | if (!XFS_IS_REALTIME_INODE(ip)) { |
67 | /* | |
68 | * Round up the allocation request to a stripe unit | |
69 | * (m_dalign) boundary if the file size is >= stripe unit | |
70 | * size, and we are allocating past the allocation eof. | |
71 | * | |
72 | * If mounted with the "-o swalloc" option the alignment is | |
73 | * increased from the strip unit size to the stripe width. | |
74 | */ | |
75 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
76 | align = mp->m_swidth; | |
77 | else if (mp->m_dalign) | |
78 | align = mp->m_dalign; | |
79 | ||
ce7ae151 | 80 | if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align)) |
bf322d98 CH |
81 | new_last_fsb = roundup_64(*last_fsb, align); |
82 | } | |
dd9f438e NS |
83 | |
84 | /* | |
85 | * Always round up the allocation request to an extent boundary | |
86 | * (when file on a real-time subvolume or has di_extsize hint). | |
87 | */ | |
88 | if (extsize) { | |
89 | if (new_last_fsb) | |
90 | align = roundup_64(new_last_fsb, extsize); | |
91 | else | |
92 | align = extsize; | |
93 | new_last_fsb = roundup_64(*last_fsb, align); | |
94 | } | |
95 | ||
96 | if (new_last_fsb) { | |
541d7d3c | 97 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
98 | if (error) |
99 | return error; | |
100 | if (eof) | |
101 | *last_fsb = new_last_fsb; | |
102 | } | |
103 | return 0; | |
104 | } | |
105 | ||
572d95f4 | 106 | STATIC int |
6d4a8ecb | 107 | xfs_alert_fsblock_zero( |
572d95f4 NS |
108 | xfs_inode_t *ip, |
109 | xfs_bmbt_irec_t *imap) | |
110 | { | |
6a19d939 | 111 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
112 | "Access to block zero in inode %llu " |
113 | "start_block: %llx start_off: %llx " | |
08e96e1a | 114 | "blkcnt: %llx extent-state: %x", |
572d95f4 NS |
115 | (unsigned long long)ip->i_ino, |
116 | (unsigned long long)imap->br_startblock, | |
117 | (unsigned long long)imap->br_startoff, | |
118 | (unsigned long long)imap->br_blockcount, | |
119 | imap->br_state); | |
120 | return EFSCORRUPTED; | |
121 | } | |
122 | ||
a206c817 | 123 | int |
1da177e4 LT |
124 | xfs_iomap_write_direct( |
125 | xfs_inode_t *ip, | |
f403b7f4 | 126 | xfs_off_t offset, |
1da177e4 | 127 | size_t count, |
3070451e | 128 | xfs_bmbt_irec_t *imap, |
405f8042 | 129 | int nmaps) |
1da177e4 LT |
130 | { |
131 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
132 | xfs_fileoff_t offset_fsb; |
133 | xfs_fileoff_t last_fsb; | |
dd9f438e | 134 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 135 | xfs_fsblock_t firstfsb; |
dd9f438e | 136 | xfs_extlen_t extsz, temp; |
0116d935 | 137 | int nimaps; |
1da177e4 | 138 | int bmapi_flag; |
06d10dd9 | 139 | int quota_flag; |
1da177e4 LT |
140 | int rt; |
141 | xfs_trans_t *tp; | |
1da177e4 | 142 | xfs_bmap_free_t free_list; |
dd9f438e | 143 | uint qblocks, resblks, resrtextents; |
1da177e4 | 144 | int committed; |
dd9f438e | 145 | int error; |
1da177e4 | 146 | |
507630b2 | 147 | error = xfs_qm_dqattach(ip, 0); |
1da177e4 LT |
148 | if (error) |
149 | return XFS_ERROR(error); | |
150 | ||
dd9f438e | 151 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 152 | extsz = xfs_get_extsz_hint(ip); |
1da177e4 | 153 | |
957d0ebe DC |
154 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
155 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 156 | if ((offset + count) > XFS_ISIZE(ip)) { |
9f6c92b9 | 157 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 158 | if (error) |
507630b2 | 159 | return XFS_ERROR(error); |
1da177e4 | 160 | } else { |
405f8042 | 161 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 162 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
163 | imap->br_blockcount + |
164 | imap->br_startoff); | |
1da177e4 | 165 | } |
dd9f438e NS |
166 | count_fsb = last_fsb - offset_fsb; |
167 | ASSERT(count_fsb > 0); | |
168 | ||
169 | resaligned = count_fsb; | |
170 | if (unlikely(extsz)) { | |
171 | if ((temp = do_mod(offset_fsb, extsz))) | |
172 | resaligned += temp; | |
173 | if ((temp = do_mod(resaligned, extsz))) | |
174 | resaligned += extsz - temp; | |
175 | } | |
176 | ||
177 | if (unlikely(rt)) { | |
178 | resrtextents = qblocks = resaligned; | |
179 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
180 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
181 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
182 | } else { | |
183 | resrtextents = 0; | |
dd9f438e | 184 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
185 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
186 | } | |
1da177e4 LT |
187 | |
188 | /* | |
06d10dd9 | 189 | * Allocate and setup the transaction |
1da177e4 | 190 | */ |
1da177e4 | 191 | tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT); |
3d3c8b52 JL |
192 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
193 | resblks, resrtextents); | |
1da177e4 | 194 | /* |
06d10dd9 | 195 | * Check for running out of space, note: need lock to return |
1da177e4 | 196 | */ |
507630b2 | 197 | if (error) { |
1da177e4 | 198 | xfs_trans_cancel(tp, 0); |
507630b2 DC |
199 | return XFS_ERROR(error); |
200 | } | |
201 | ||
1da177e4 | 202 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 203 | |
7d095257 | 204 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 205 | if (error) |
507630b2 | 206 | goto out_trans_cancel; |
1da177e4 | 207 | |
ddc3415a | 208 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 209 | |
c0dc7828 | 210 | bmapi_flag = 0; |
ce7ae151 | 211 | if (offset < XFS_ISIZE(ip) || extsz) |
1da177e4 LT |
212 | bmapi_flag |= XFS_BMAPI_PREALLOC; |
213 | ||
214 | /* | |
3070451e CH |
215 | * From this point onwards we overwrite the imap pointer that the |
216 | * caller gave to us. | |
1da177e4 | 217 | */ |
9d87c319 | 218 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 219 | nimaps = 1; |
c0dc7828 DC |
220 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag, |
221 | &firstfsb, 0, imap, &nimaps, &free_list); | |
06d10dd9 | 222 | if (error) |
507630b2 | 223 | goto out_bmap_cancel; |
1da177e4 LT |
224 | |
225 | /* | |
06d10dd9 | 226 | * Complete the transaction |
1da177e4 | 227 | */ |
f7c99b6f | 228 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
06d10dd9 | 229 | if (error) |
507630b2 | 230 | goto out_bmap_cancel; |
1c72bf90 | 231 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
06d10dd9 | 232 | if (error) |
507630b2 | 233 | goto out_unlock; |
1da177e4 | 234 | |
06d10dd9 NS |
235 | /* |
236 | * Copy any maps to caller's array and return any error. | |
237 | */ | |
1da177e4 | 238 | if (nimaps == 0) { |
507630b2 DC |
239 | error = XFS_ERROR(ENOSPC); |
240 | goto out_unlock; | |
572d95f4 NS |
241 | } |
242 | ||
507630b2 | 243 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 244 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 245 | |
507630b2 DC |
246 | out_unlock: |
247 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
248 | return error; | |
1da177e4 | 249 | |
507630b2 | 250 | out_bmap_cancel: |
1da177e4 | 251 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 252 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 253 | out_trans_cancel: |
1da177e4 | 254 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); |
507630b2 | 255 | goto out_unlock; |
1da177e4 LT |
256 | } |
257 | ||
dd9f438e | 258 | /* |
8de2bf93 DC |
259 | * If the caller is doing a write at the end of the file, then extend the |
260 | * allocation out to the file system's write iosize. We clean up any extra | |
261 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
262 | * |
263 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
264 | * preallocation as it it not needed. | |
dd9f438e NS |
265 | */ |
266 | STATIC int | |
267 | xfs_iomap_eof_want_preallocate( | |
268 | xfs_mount_t *mp, | |
541d7d3c | 269 | xfs_inode_t *ip, |
dd9f438e NS |
270 | xfs_off_t offset, |
271 | size_t count, | |
dd9f438e NS |
272 | xfs_bmbt_irec_t *imap, |
273 | int nimaps, | |
274 | int *prealloc) | |
275 | { | |
276 | xfs_fileoff_t start_fsb; | |
277 | xfs_filblks_t count_fsb; | |
278 | xfs_fsblock_t firstblock; | |
279 | int n, error, imaps; | |
055388a3 | 280 | int found_delalloc = 0; |
dd9f438e NS |
281 | |
282 | *prealloc = 0; | |
ce7ae151 | 283 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
284 | return 0; |
285 | ||
133eeb17 DC |
286 | /* |
287 | * If the file is smaller than the minimum prealloc and we are using | |
288 | * dynamic preallocation, don't do any preallocation at all as it is | |
289 | * likely this is the only write to the file that is going to be done. | |
290 | */ | |
291 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
292 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)) | |
293 | return 0; | |
294 | ||
dd9f438e NS |
295 | /* |
296 | * If there are any real blocks past eof, then don't | |
297 | * do any speculative allocation. | |
298 | */ | |
299 | start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1))); | |
32972383 | 300 | count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
dd9f438e NS |
301 | while (count_fsb > 0) { |
302 | imaps = nimaps; | |
3ddb8fa9 | 303 | firstblock = NULLFSBLOCK; |
5c8ed202 DC |
304 | error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps, |
305 | 0); | |
dd9f438e NS |
306 | if (error) |
307 | return error; | |
308 | for (n = 0; n < imaps; n++) { | |
309 | if ((imap[n].br_startblock != HOLESTARTBLOCK) && | |
310 | (imap[n].br_startblock != DELAYSTARTBLOCK)) | |
311 | return 0; | |
312 | start_fsb += imap[n].br_blockcount; | |
313 | count_fsb -= imap[n].br_blockcount; | |
055388a3 DC |
314 | |
315 | if (imap[n].br_startblock == DELAYSTARTBLOCK) | |
316 | found_delalloc = 1; | |
dd9f438e NS |
317 | } |
318 | } | |
055388a3 DC |
319 | if (!found_delalloc) |
320 | *prealloc = 1; | |
dd9f438e NS |
321 | return 0; |
322 | } | |
323 | ||
a1e16c26 DC |
324 | /* |
325 | * Determine the initial size of the preallocation. We are beyond the current | |
326 | * EOF here, but we need to take into account whether this is a sparse write or | |
327 | * an extending write when determining the preallocation size. Hence we need to | |
328 | * look up the extent that ends at the current write offset and use the result | |
329 | * to determine the preallocation size. | |
330 | * | |
331 | * If the extent is a hole, then preallocation is essentially disabled. | |
332 | * Otherwise we take the size of the preceeding data extent as the basis for the | |
333 | * preallocation size. If the size of the extent is greater than half the | |
334 | * maximum extent length, then use the current offset as the basis. This ensures | |
335 | * that for large files the preallocation size always extends to MAXEXTLEN | |
336 | * rather than falling short due to things like stripe unit/width alignment of | |
337 | * real extents. | |
338 | */ | |
e8108ced | 339 | STATIC xfs_fsblock_t |
a1e16c26 DC |
340 | xfs_iomap_eof_prealloc_initial_size( |
341 | struct xfs_mount *mp, | |
342 | struct xfs_inode *ip, | |
343 | xfs_off_t offset, | |
344 | xfs_bmbt_irec_t *imap, | |
345 | int nimaps) | |
346 | { | |
347 | xfs_fileoff_t start_fsb; | |
348 | int imaps = 1; | |
349 | int error; | |
350 | ||
351 | ASSERT(nimaps >= imaps); | |
352 | ||
353 | /* if we are using a specific prealloc size, return now */ | |
354 | if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) | |
355 | return 0; | |
356 | ||
133eeb17 DC |
357 | /* If the file is small, then use the minimum prealloc */ |
358 | if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign)) | |
359 | return 0; | |
360 | ||
a1e16c26 DC |
361 | /* |
362 | * As we write multiple pages, the offset will always align to the | |
363 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
364 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
365 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
366 | * point to the block prior to the EOF block and not the hole that maps | |
367 | * directly at @offset. | |
368 | */ | |
369 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
370 | if (start_fsb) | |
371 | start_fsb--; | |
372 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
373 | if (error) | |
374 | return 0; | |
375 | ||
376 | ASSERT(imaps == 1); | |
377 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
378 | return 0; | |
379 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 380 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
381 | return XFS_B_TO_FSB(mp, offset); |
382 | } | |
383 | ||
76a4202a BF |
384 | STATIC bool |
385 | xfs_quota_need_throttle( | |
386 | struct xfs_inode *ip, | |
387 | int type, | |
388 | xfs_fsblock_t alloc_blocks) | |
389 | { | |
390 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
391 | ||
392 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
393 | return false; | |
394 | ||
395 | /* no hi watermark, no throttle */ | |
396 | if (!dq->q_prealloc_hi_wmark) | |
397 | return false; | |
398 | ||
399 | /* under the lo watermark, no throttle */ | |
400 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
401 | return false; | |
402 | ||
403 | return true; | |
404 | } | |
405 | ||
406 | STATIC void | |
407 | xfs_quota_calc_throttle( | |
408 | struct xfs_inode *ip, | |
409 | int type, | |
410 | xfs_fsblock_t *qblocks, | |
411 | int *qshift) | |
412 | { | |
413 | int64_t freesp; | |
414 | int shift = 0; | |
415 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
416 | ||
417 | /* over hi wmark, squash the prealloc completely */ | |
418 | if (dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
419 | *qblocks = 0; | |
420 | return; | |
421 | } | |
422 | ||
423 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
424 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
425 | shift = 2; | |
426 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
427 | shift += 2; | |
428 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
429 | shift += 2; | |
430 | } | |
431 | ||
432 | /* only overwrite the throttle values if we are more aggressive */ | |
433 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
434 | *qblocks = freesp; | |
435 | *qshift = shift; | |
436 | } | |
437 | } | |
438 | ||
055388a3 DC |
439 | /* |
440 | * If we don't have a user specified preallocation size, dynamically increase | |
441 | * the preallocation size as the size of the file grows. Cap the maximum size | |
442 | * at a single extent or less if the filesystem is near full. The closer the | |
443 | * filesystem is to full, the smaller the maximum prealocation. | |
444 | */ | |
445 | STATIC xfs_fsblock_t | |
446 | xfs_iomap_prealloc_size( | |
447 | struct xfs_mount *mp, | |
a1e16c26 DC |
448 | struct xfs_inode *ip, |
449 | xfs_off_t offset, | |
450 | struct xfs_bmbt_irec *imap, | |
451 | int nimaps) | |
055388a3 DC |
452 | { |
453 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
454 | int shift = 0; |
455 | int64_t freesp; | |
76a4202a BF |
456 | xfs_fsblock_t qblocks; |
457 | int qshift = 0; | |
055388a3 | 458 | |
a1e16c26 DC |
459 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
460 | imap, nimaps); | |
3c58b5f8 BF |
461 | if (!alloc_blocks) |
462 | goto check_writeio; | |
76a4202a | 463 | qblocks = alloc_blocks; |
3c58b5f8 | 464 | |
c9bdbdc0 BF |
465 | /* |
466 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
467 | * to the nearest power of two value after throttling. To prevent the | |
468 | * round down from unconditionally reducing the maximum supported prealloc | |
469 | * size, we round up first, apply appropriate throttling, round down and | |
470 | * cap the value to MAXEXTLEN. | |
471 | */ | |
472 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
473 | alloc_blocks); | |
3c58b5f8 BF |
474 | |
475 | xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT); | |
476 | freesp = mp->m_sb.sb_fdblocks; | |
477 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { | |
478 | shift = 2; | |
479 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
480 | shift++; | |
481 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
482 | shift++; | |
483 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
484 | shift++; | |
485 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
486 | shift++; | |
055388a3 | 487 | } |
76a4202a BF |
488 | |
489 | /* | |
490 | * Check each quota to cap the prealloc size and provide a shift | |
491 | * value to throttle with. | |
492 | */ | |
493 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
494 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift); | |
495 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) | |
496 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift); | |
497 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) | |
498 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift); | |
499 | ||
500 | /* | |
501 | * The final prealloc size is set to the minimum of free space available | |
502 | * in each of the quotas and the overall filesystem. | |
503 | * | |
504 | * The shift throttle value is set to the maximum value as determined by | |
505 | * the global low free space values and per-quota low free space values. | |
506 | */ | |
507 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
508 | shift = MAX(shift, qshift); | |
509 | ||
3c58b5f8 BF |
510 | if (shift) |
511 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
512 | /* |
513 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
514 | * alloc_blocks = 0. | |
515 | */ | |
516 | if (alloc_blocks) | |
517 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
518 | if (alloc_blocks > MAXEXTLEN) | |
519 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
520 | |
521 | /* | |
522 | * If we are still trying to allocate more space than is | |
523 | * available, squash the prealloc hard. This can happen if we | |
524 | * have a large file on a small filesystem and the above | |
525 | * lowspace thresholds are smaller than MAXEXTLEN. | |
526 | */ | |
527 | while (alloc_blocks && alloc_blocks >= freesp) | |
528 | alloc_blocks >>= 4; | |
055388a3 | 529 | |
3c58b5f8 | 530 | check_writeio: |
055388a3 DC |
531 | if (alloc_blocks < mp->m_writeio_blocks) |
532 | alloc_blocks = mp->m_writeio_blocks; | |
533 | ||
19cb7e38 BF |
534 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
535 | mp->m_writeio_blocks); | |
536 | ||
055388a3 DC |
537 | return alloc_blocks; |
538 | } | |
539 | ||
a206c817 | 540 | int |
1da177e4 LT |
541 | xfs_iomap_write_delay( |
542 | xfs_inode_t *ip, | |
f403b7f4 | 543 | xfs_off_t offset, |
1da177e4 | 544 | size_t count, |
405f8042 | 545 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
546 | { |
547 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
548 | xfs_fileoff_t offset_fsb; |
549 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
550 | xfs_off_t aligned_offset; |
551 | xfs_fileoff_t ioalign; | |
dd9f438e | 552 | xfs_extlen_t extsz; |
1da177e4 | 553 | int nimaps; |
1da177e4 | 554 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 555 | int prealloc; |
dd9f438e | 556 | int error; |
1da177e4 | 557 | |
579aa9ca | 558 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
559 | |
560 | /* | |
561 | * Make sure that the dquots are there. This doesn't hold | |
562 | * the ilock across a disk read. | |
563 | */ | |
7d095257 | 564 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 LT |
565 | if (error) |
566 | return XFS_ERROR(error); | |
567 | ||
957d0ebe | 568 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
569 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
570 | ||
9f6c92b9 | 571 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 572 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
573 | if (error) |
574 | return error; | |
1da177e4 | 575 | |
8de2bf93 | 576 | retry: |
dd9f438e | 577 | if (prealloc) { |
a1e16c26 DC |
578 | xfs_fsblock_t alloc_blocks; |
579 | ||
580 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
581 | XFS_WRITE_IMAPS); | |
055388a3 | 582 | |
1da177e4 LT |
583 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
584 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 585 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
586 | } else { |
587 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 588 | } |
1da177e4 | 589 | |
dd9f438e | 590 | if (prealloc || extsz) { |
9f6c92b9 | 591 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 592 | if (error) |
1da177e4 | 593 | return error; |
1da177e4 | 594 | } |
dd9f438e | 595 | |
3ed9116e DC |
596 | /* |
597 | * Make sure preallocation does not create extents beyond the range we | |
598 | * actually support in this filesystem. | |
599 | */ | |
d2c28191 DC |
600 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
601 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
602 | |
603 | ASSERT(last_fsb > offset_fsb); | |
604 | ||
dd9f438e | 605 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
606 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
607 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
608 | switch (error) { |
609 | case 0: | |
610 | case ENOSPC: | |
611 | case EDQUOT: | |
612 | break; | |
613 | default: | |
1da177e4 | 614 | return XFS_ERROR(error); |
055388a3 | 615 | } |
dd9f438e | 616 | |
1da177e4 | 617 | /* |
9aa05000 | 618 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 619 | * without EOF preallocation. |
1da177e4 LT |
620 | */ |
621 | if (nimaps == 0) { | |
0b1b213f | 622 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
623 | if (prealloc) { |
624 | prealloc = 0; | |
625 | error = 0; | |
626 | goto retry; | |
055388a3 | 627 | } |
9aa05000 | 628 | return XFS_ERROR(error ? error : ENOSPC); |
1da177e4 LT |
629 | } |
630 | ||
86c4d623 | 631 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 632 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 633 | |
27b52867 BF |
634 | /* |
635 | * Tag the inode as speculatively preallocated so we can reclaim this | |
636 | * space on demand, if necessary. | |
637 | */ | |
638 | if (prealloc) | |
639 | xfs_inode_set_eofblocks_tag(ip); | |
640 | ||
dd9f438e | 641 | *ret_imap = imap[0]; |
1da177e4 LT |
642 | return 0; |
643 | } | |
644 | ||
645 | /* | |
646 | * Pass in a delayed allocate extent, convert it to real extents; | |
647 | * return to the caller the extent we create which maps on top of | |
648 | * the originating callers request. | |
649 | * | |
650 | * Called without a lock on the inode. | |
e4143a1c DC |
651 | * |
652 | * We no longer bother to look at the incoming map - all we have to | |
653 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 654 | */ |
a206c817 | 655 | int |
1da177e4 LT |
656 | xfs_iomap_write_allocate( |
657 | xfs_inode_t *ip, | |
f403b7f4 | 658 | xfs_off_t offset, |
405f8042 | 659 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
660 | { |
661 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
662 | xfs_fileoff_t offset_fsb, last_block; |
663 | xfs_fileoff_t end_fsb, map_start_fsb; | |
664 | xfs_fsblock_t first_block; | |
665 | xfs_bmap_free_t free_list; | |
666 | xfs_filblks_t count_fsb; | |
1da177e4 | 667 | xfs_trans_t *tp; |
e4143a1c | 668 | int nimaps, committed; |
1da177e4 LT |
669 | int error = 0; |
670 | int nres; | |
671 | ||
1da177e4 LT |
672 | /* |
673 | * Make sure that the dquots are there. | |
674 | */ | |
7d095257 CH |
675 | error = xfs_qm_dqattach(ip, 0); |
676 | if (error) | |
1da177e4 LT |
677 | return XFS_ERROR(error); |
678 | ||
24e17b5f | 679 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
680 | count_fsb = imap->br_blockcount; |
681 | map_start_fsb = imap->br_startoff; | |
1da177e4 LT |
682 | |
683 | XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); | |
684 | ||
685 | while (count_fsb != 0) { | |
686 | /* | |
687 | * Set up a transaction with which to allocate the | |
688 | * backing store for the file. Do allocations in a | |
689 | * loop until we get some space in the range we are | |
690 | * interested in. The other space that might be allocated | |
691 | * is in the delayed allocation extent on which we sit | |
692 | * but before our buffer starts. | |
693 | */ | |
694 | ||
695 | nimaps = 0; | |
696 | while (nimaps == 0) { | |
697 | tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE); | |
84e1e99f | 698 | tp->t_flags |= XFS_TRANS_RESERVE; |
1da177e4 | 699 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
3d3c8b52 JL |
700 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
701 | nres, 0); | |
1da177e4 LT |
702 | if (error) { |
703 | xfs_trans_cancel(tp, 0); | |
704 | return XFS_ERROR(error); | |
705 | } | |
706 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 707 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 708 | |
9d87c319 | 709 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 710 | |
1da177e4 | 711 | /* |
e4143a1c DC |
712 | * it is possible that the extents have changed since |
713 | * we did the read call as we dropped the ilock for a | |
714 | * while. We have to be careful about truncates or hole | |
715 | * punchs here - we are not allowed to allocate | |
716 | * non-delalloc blocks here. | |
717 | * | |
718 | * The only protection against truncation is the pages | |
719 | * for the range we are being asked to convert are | |
720 | * locked and hence a truncate will block on them | |
721 | * first. | |
722 | * | |
723 | * As a result, if we go beyond the range we really | |
724 | * need and hit an delalloc extent boundary followed by | |
725 | * a hole while we have excess blocks in the map, we | |
726 | * will fill the hole incorrectly and overrun the | |
727 | * transaction reservation. | |
728 | * | |
729 | * Using a single map prevents this as we are forced to | |
730 | * check each map we look for overlap with the desired | |
731 | * range and abort as soon as we find it. Also, given | |
732 | * that we only return a single map, having one beyond | |
733 | * what we can return is probably a bit silly. | |
734 | * | |
735 | * We also need to check that we don't go beyond EOF; | |
736 | * this is a truncate optimisation as a truncate sets | |
737 | * the new file size before block on the pages we | |
738 | * currently have locked under writeback. Because they | |
739 | * are about to be tossed, we don't need to write them | |
740 | * back.... | |
1da177e4 | 741 | */ |
e4143a1c | 742 | nimaps = 1; |
ce7ae151 | 743 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7c9ef85c DC |
744 | error = xfs_bmap_last_offset(NULL, ip, &last_block, |
745 | XFS_DATA_FORK); | |
746 | if (error) | |
747 | goto trans_cancel; | |
748 | ||
1da177e4 LT |
749 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
750 | if ((map_start_fsb + count_fsb) > last_block) { | |
751 | count_fsb = last_block - map_start_fsb; | |
752 | if (count_fsb == 0) { | |
753 | error = EAGAIN; | |
754 | goto trans_cancel; | |
755 | } | |
756 | } | |
757 | ||
3070451e | 758 | /* |
3070451e CH |
759 | * From this point onwards we overwrite the imap |
760 | * pointer that the caller gave to us. | |
761 | */ | |
c0dc7828 | 762 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
2455881c DC |
763 | count_fsb, |
764 | XFS_BMAPI_STACK_SWITCH, | |
765 | &first_block, 1, | |
c0dc7828 | 766 | imap, &nimaps, &free_list); |
1da177e4 LT |
767 | if (error) |
768 | goto trans_cancel; | |
769 | ||
f7c99b6f | 770 | error = xfs_bmap_finish(&tp, &free_list, &committed); |
1da177e4 LT |
771 | if (error) |
772 | goto trans_cancel; | |
773 | ||
1c72bf90 | 774 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
775 | if (error) |
776 | goto error0; | |
777 | ||
778 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
779 | } | |
780 | ||
781 | /* | |
782 | * See if we were able to allocate an extent that | |
783 | * covers at least part of the callers request | |
784 | */ | |
3070451e | 785 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 786 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 787 | |
3070451e CH |
788 | if ((offset_fsb >= imap->br_startoff) && |
789 | (offset_fsb < (imap->br_startoff + | |
790 | imap->br_blockcount))) { | |
e4143a1c DC |
791 | XFS_STATS_INC(xs_xstrat_quick); |
792 | return 0; | |
1da177e4 LT |
793 | } |
794 | ||
e4143a1c DC |
795 | /* |
796 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
797 | * file, just surrounding data, try again. |
798 | */ | |
3070451e CH |
799 | count_fsb -= imap->br_blockcount; |
800 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
801 | } |
802 | ||
803 | trans_cancel: | |
804 | xfs_bmap_cancel(&free_list); | |
805 | xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT); | |
806 | error0: | |
807 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
808 | return XFS_ERROR(error); | |
809 | } | |
810 | ||
811 | int | |
812 | xfs_iomap_write_unwritten( | |
813 | xfs_inode_t *ip, | |
f403b7f4 | 814 | xfs_off_t offset, |
1da177e4 LT |
815 | size_t count) |
816 | { | |
817 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
818 | xfs_fileoff_t offset_fsb; |
819 | xfs_filblks_t count_fsb; | |
820 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
821 | xfs_fsblock_t firstfsb; |
822 | int nimaps; | |
823 | xfs_trans_t *tp; | |
824 | xfs_bmbt_irec_t imap; | |
825 | xfs_bmap_free_t free_list; | |
84803fb7 | 826 | xfs_fsize_t i_size; |
dd9f438e | 827 | uint resblks; |
1da177e4 LT |
828 | int committed; |
829 | int error; | |
1da177e4 | 830 | |
0b1b213f | 831 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
832 | |
833 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
834 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
835 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
836 | ||
4ddd8bb1 LM |
837 | /* |
838 | * Reserve enough blocks in this transaction for two complete extent | |
839 | * btree splits. We may be converting the middle part of an unwritten | |
840 | * extent and in this case we will insert two new extents in the btree | |
841 | * each of which could cause a full split. | |
842 | * | |
843 | * This reservation amount will be used in the first call to | |
844 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
845 | * rest of the operation. | |
846 | */ | |
dd9f438e | 847 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 848 | |
dd9f438e | 849 | do { |
1da177e4 LT |
850 | /* |
851 | * set up a transaction to convert the range of extents | |
852 | * from unwritten to real. Do allocations in a loop until | |
853 | * we have covered the range passed in. | |
80641dc6 CH |
854 | * |
855 | * Note that we open code the transaction allocation here | |
856 | * to pass KM_NOFS--we can't risk to recursing back into | |
857 | * the filesystem here as we might be asked to write out | |
858 | * the same inode that we complete here and might deadlock | |
859 | * on the iolock. | |
1da177e4 | 860 | */ |
d9457dc0 | 861 | sb_start_intwrite(mp->m_super); |
80641dc6 | 862 | tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS); |
d9457dc0 | 863 | tp->t_flags |= XFS_TRANS_RESERVE | XFS_TRANS_FREEZE_PROT; |
3d3c8b52 JL |
864 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, |
865 | resblks, 0); | |
1da177e4 LT |
866 | if (error) { |
867 | xfs_trans_cancel(tp, 0); | |
572d95f4 | 868 | return XFS_ERROR(error); |
1da177e4 LT |
869 | } |
870 | ||
871 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 872 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
873 | |
874 | /* | |
875 | * Modify the unwritten extent state of the buffer. | |
876 | */ | |
9d87c319 | 877 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 878 | nimaps = 1; |
c0dc7828 DC |
879 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
880 | XFS_BMAPI_CONVERT, &firstfsb, | |
b4e9181e | 881 | 1, &imap, &nimaps, &free_list); |
1da177e4 LT |
882 | if (error) |
883 | goto error_on_bmapi_transaction; | |
884 | ||
84803fb7 CH |
885 | /* |
886 | * Log the updated inode size as we go. We have to be careful | |
887 | * to only log it up to the actual write offset if it is | |
888 | * halfway into a block. | |
889 | */ | |
890 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
891 | if (i_size > offset + count) | |
892 | i_size = offset + count; | |
893 | ||
894 | i_size = xfs_new_eof(ip, i_size); | |
895 | if (i_size) { | |
896 | ip->i_d.di_size = i_size; | |
897 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
898 | } | |
899 | ||
900 | error = xfs_bmap_finish(&tp, &free_list, &committed); | |
1da177e4 LT |
901 | if (error) |
902 | goto error_on_bmapi_transaction; | |
903 | ||
1c72bf90 | 904 | error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); |
1da177e4 LT |
905 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
906 | if (error) | |
572d95f4 NS |
907 | return XFS_ERROR(error); |
908 | ||
86c4d623 | 909 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 910 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
911 | |
912 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
913 | /* | |
914 | * The numblks_fsb value should always get | |
915 | * smaller, otherwise the loop is stuck. | |
916 | */ | |
917 | ASSERT(imap.br_blockcount); | |
918 | break; | |
919 | } | |
920 | offset_fsb += numblks_fsb; | |
921 | count_fsb -= numblks_fsb; | |
922 | } while (count_fsb > 0); | |
923 | ||
924 | return 0; | |
925 | ||
926 | error_on_bmapi_transaction: | |
927 | xfs_bmap_cancel(&free_list); | |
928 | xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT)); | |
929 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1da177e4 LT |
930 | return XFS_ERROR(error); |
931 | } |