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 | */ |
3b3dce05 | 18 | #include <linux/iomap.h> |
1da177e4 | 19 | #include "xfs.h" |
1da177e4 | 20 | #include "xfs_fs.h" |
70a9883c | 21 | #include "xfs_shared.h" |
239880ef DC |
22 | #include "xfs_format.h" |
23 | #include "xfs_log_format.h" | |
24 | #include "xfs_trans_resv.h" | |
1da177e4 | 25 | #include "xfs_mount.h" |
3ab78df2 | 26 | #include "xfs_defer.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" | |
1da177e4 | 41 | |
1da177e4 LT |
42 | |
43 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
44 | << mp->m_writeio_log) | |
1da177e4 LT |
45 | #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP |
46 | ||
dd9f438e NS |
47 | STATIC int |
48 | xfs_iomap_eof_align_last_fsb( | |
49 | xfs_mount_t *mp, | |
541d7d3c | 50 | xfs_inode_t *ip, |
dd9f438e NS |
51 | xfs_extlen_t extsize, |
52 | xfs_fileoff_t *last_fsb) | |
53 | { | |
bf322d98 | 54 | xfs_extlen_t align = 0; |
dd9f438e NS |
55 | int eof, error; |
56 | ||
bf322d98 CH |
57 | if (!XFS_IS_REALTIME_INODE(ip)) { |
58 | /* | |
59 | * Round up the allocation request to a stripe unit | |
60 | * (m_dalign) boundary if the file size is >= stripe unit | |
61 | * size, and we are allocating past the allocation eof. | |
62 | * | |
63 | * If mounted with the "-o swalloc" option the alignment is | |
64 | * increased from the strip unit size to the stripe width. | |
65 | */ | |
66 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
67 | align = mp->m_swidth; | |
68 | else if (mp->m_dalign) | |
69 | align = mp->m_dalign; | |
70 | ||
76b57302 PW |
71 | if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align)) |
72 | align = 0; | |
bf322d98 | 73 | } |
dd9f438e NS |
74 | |
75 | /* | |
76 | * Always round up the allocation request to an extent boundary | |
77 | * (when file on a real-time subvolume or has di_extsize hint). | |
78 | */ | |
79 | if (extsize) { | |
76b57302 PW |
80 | if (align) |
81 | align = roundup_64(align, extsize); | |
dd9f438e NS |
82 | else |
83 | align = extsize; | |
dd9f438e NS |
84 | } |
85 | ||
76b57302 PW |
86 | if (align) { |
87 | xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align); | |
541d7d3c | 88 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
89 | if (error) |
90 | return error; | |
91 | if (eof) | |
92 | *last_fsb = new_last_fsb; | |
93 | } | |
94 | return 0; | |
95 | } | |
96 | ||
572d95f4 | 97 | STATIC int |
6d4a8ecb | 98 | xfs_alert_fsblock_zero( |
572d95f4 NS |
99 | xfs_inode_t *ip, |
100 | xfs_bmbt_irec_t *imap) | |
101 | { | |
6a19d939 | 102 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
103 | "Access to block zero in inode %llu " |
104 | "start_block: %llx start_off: %llx " | |
08e96e1a | 105 | "blkcnt: %llx extent-state: %x", |
572d95f4 NS |
106 | (unsigned long long)ip->i_ino, |
107 | (unsigned long long)imap->br_startblock, | |
108 | (unsigned long long)imap->br_startoff, | |
109 | (unsigned long long)imap->br_blockcount, | |
110 | imap->br_state); | |
2451337d | 111 | return -EFSCORRUPTED; |
572d95f4 NS |
112 | } |
113 | ||
a206c817 | 114 | int |
1da177e4 LT |
115 | xfs_iomap_write_direct( |
116 | xfs_inode_t *ip, | |
f403b7f4 | 117 | xfs_off_t offset, |
1da177e4 | 118 | size_t count, |
3070451e | 119 | xfs_bmbt_irec_t *imap, |
405f8042 | 120 | int nmaps) |
1da177e4 LT |
121 | { |
122 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
123 | xfs_fileoff_t offset_fsb; |
124 | xfs_fileoff_t last_fsb; | |
dd9f438e | 125 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 126 | xfs_fsblock_t firstfsb; |
dd9f438e | 127 | xfs_extlen_t extsz, temp; |
0116d935 | 128 | int nimaps; |
06d10dd9 | 129 | int quota_flag; |
1da177e4 LT |
130 | int rt; |
131 | xfs_trans_t *tp; | |
1da177e4 | 132 | xfs_bmap_free_t free_list; |
dd9f438e | 133 | uint qblocks, resblks, resrtextents; |
dd9f438e | 134 | int error; |
009c6e87 | 135 | int lockmode; |
1ca19157 | 136 | int bmapi_flags = XFS_BMAPI_PREALLOC; |
253f4911 | 137 | uint tflags = 0; |
1da177e4 | 138 | |
dd9f438e | 139 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 140 | extsz = xfs_get_extsz_hint(ip); |
009c6e87 BF |
141 | lockmode = XFS_ILOCK_SHARED; /* locked by caller */ |
142 | ||
143 | ASSERT(xfs_isilocked(ip, lockmode)); | |
1da177e4 | 144 | |
957d0ebe DC |
145 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
146 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 147 | if ((offset + count) > XFS_ISIZE(ip)) { |
009c6e87 BF |
148 | /* |
149 | * Assert that the in-core extent list is present since this can | |
150 | * call xfs_iread_extents() and we only have the ilock shared. | |
151 | * This should be safe because the lock was held around a bmapi | |
152 | * call in the caller and we only need it to access the in-core | |
153 | * list. | |
154 | */ | |
155 | ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags & | |
156 | XFS_IFEXTENTS); | |
9f6c92b9 | 157 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 158 | if (error) |
009c6e87 | 159 | goto out_unlock; |
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 | 187 | |
009c6e87 BF |
188 | /* |
189 | * Drop the shared lock acquired by the caller, attach the dquot if | |
190 | * necessary and move on to transaction setup. | |
191 | */ | |
192 | xfs_iunlock(ip, lockmode); | |
193 | error = xfs_qm_dqattach(ip, 0); | |
194 | if (error) | |
195 | return error; | |
196 | ||
1ca19157 DC |
197 | /* |
198 | * For DAX, we do not allocate unwritten extents, but instead we zero | |
199 | * the block before we commit the transaction. Ideally we'd like to do | |
200 | * this outside the transaction context, but if we commit and then crash | |
201 | * we may not have zeroed the blocks and this will be exposed on | |
202 | * recovery of the allocation. Hence we must zero before commit. | |
3b0fe478 | 203 | * |
1ca19157 DC |
204 | * Further, if we are mapping unwritten extents here, we need to zero |
205 | * and convert them to written so that we don't need an unwritten extent | |
206 | * callback for DAX. This also means that we need to be able to dip into | |
3b0fe478 DC |
207 | * the reserve block pool for bmbt block allocation if there is no space |
208 | * left but we need to do unwritten extent conversion. | |
1ca19157 DC |
209 | */ |
210 | if (IS_DAX(VFS_I(ip))) { | |
211 | bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO; | |
3b0fe478 | 212 | if (ISUNWRITTEN(imap)) { |
253f4911 | 213 | tflags |= XFS_TRANS_RESERVE; |
3b0fe478 DC |
214 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
215 | } | |
1ca19157 | 216 | } |
253f4911 CH |
217 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents, |
218 | tflags, &tp); | |
219 | if (error) | |
b474c7ae | 220 | return error; |
507630b2 | 221 | |
009c6e87 BF |
222 | lockmode = XFS_ILOCK_EXCL; |
223 | xfs_ilock(ip, lockmode); | |
1da177e4 | 224 | |
7d095257 | 225 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 226 | if (error) |
507630b2 | 227 | goto out_trans_cancel; |
1da177e4 | 228 | |
ddc3415a | 229 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 230 | |
1da177e4 | 231 | /* |
3070451e CH |
232 | * From this point onwards we overwrite the imap pointer that the |
233 | * caller gave to us. | |
1da177e4 | 234 | */ |
9d87c319 | 235 | xfs_bmap_init(&free_list, &firstfsb); |
06d10dd9 | 236 | nimaps = 1; |
d531d91d | 237 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
264e89ad | 238 | bmapi_flags, &firstfsb, resblks, imap, |
dbd5c8c9 | 239 | &nimaps, &free_list); |
06d10dd9 | 240 | if (error) |
507630b2 | 241 | goto out_bmap_cancel; |
1da177e4 LT |
242 | |
243 | /* | |
06d10dd9 | 244 | * Complete the transaction |
1da177e4 | 245 | */ |
f6106efa | 246 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
06d10dd9 | 247 | if (error) |
507630b2 | 248 | goto out_bmap_cancel; |
1ca19157 | 249 | |
70393313 | 250 | error = xfs_trans_commit(tp); |
06d10dd9 | 251 | if (error) |
507630b2 | 252 | goto out_unlock; |
1da177e4 | 253 | |
06d10dd9 NS |
254 | /* |
255 | * Copy any maps to caller's array and return any error. | |
256 | */ | |
1da177e4 | 257 | if (nimaps == 0) { |
2451337d | 258 | error = -ENOSPC; |
507630b2 | 259 | goto out_unlock; |
572d95f4 NS |
260 | } |
261 | ||
507630b2 | 262 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 263 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 264 | |
507630b2 | 265 | out_unlock: |
009c6e87 | 266 | xfs_iunlock(ip, lockmode); |
507630b2 | 267 | return error; |
1da177e4 | 268 | |
507630b2 | 269 | out_bmap_cancel: |
1da177e4 | 270 | xfs_bmap_cancel(&free_list); |
ea562ed6 | 271 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 272 | out_trans_cancel: |
4906e215 | 273 | xfs_trans_cancel(tp); |
507630b2 | 274 | goto out_unlock; |
1da177e4 LT |
275 | } |
276 | ||
dd9f438e | 277 | /* |
8de2bf93 DC |
278 | * If the caller is doing a write at the end of the file, then extend the |
279 | * allocation out to the file system's write iosize. We clean up any extra | |
280 | * space left over when the file is closed in xfs_inactive(). | |
055388a3 DC |
281 | * |
282 | * If we find we already have delalloc preallocation beyond EOF, don't do more | |
283 | * preallocation as it it not needed. | |
dd9f438e NS |
284 | */ |
285 | STATIC int | |
286 | xfs_iomap_eof_want_preallocate( | |
287 | xfs_mount_t *mp, | |
541d7d3c | 288 | xfs_inode_t *ip, |
dd9f438e NS |
289 | xfs_off_t offset, |
290 | size_t count, | |
dd9f438e NS |
291 | xfs_bmbt_irec_t *imap, |
292 | int nimaps, | |
293 | int *prealloc) | |
294 | { | |
295 | xfs_fileoff_t start_fsb; | |
296 | xfs_filblks_t count_fsb; | |
dd9f438e | 297 | int n, error, imaps; |
055388a3 | 298 | int found_delalloc = 0; |
dd9f438e NS |
299 | |
300 | *prealloc = 0; | |
ce7ae151 | 301 | if (offset + count <= XFS_ISIZE(ip)) |
dd9f438e NS |
302 | return 0; |
303 | ||
133eeb17 DC |
304 | /* |
305 | * If the file is smaller than the minimum prealloc and we are using | |
306 | * dynamic preallocation, don't do any preallocation at all as it is | |
307 | * likely this is the only write to the file that is going to be done. | |
308 | */ | |
309 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
310 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)) | |
311 | return 0; | |
312 | ||
dd9f438e NS |
313 | /* |
314 | * If there are any real blocks past eof, then don't | |
315 | * do any speculative allocation. | |
316 | */ | |
317 | start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1))); | |
32972383 | 318 | count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); |
dd9f438e NS |
319 | while (count_fsb > 0) { |
320 | imaps = nimaps; | |
5c8ed202 DC |
321 | error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps, |
322 | 0); | |
dd9f438e NS |
323 | if (error) |
324 | return error; | |
325 | for (n = 0; n < imaps; n++) { | |
326 | if ((imap[n].br_startblock != HOLESTARTBLOCK) && | |
327 | (imap[n].br_startblock != DELAYSTARTBLOCK)) | |
328 | return 0; | |
329 | start_fsb += imap[n].br_blockcount; | |
330 | count_fsb -= imap[n].br_blockcount; | |
055388a3 DC |
331 | |
332 | if (imap[n].br_startblock == DELAYSTARTBLOCK) | |
333 | found_delalloc = 1; | |
dd9f438e NS |
334 | } |
335 | } | |
055388a3 DC |
336 | if (!found_delalloc) |
337 | *prealloc = 1; | |
dd9f438e NS |
338 | return 0; |
339 | } | |
340 | ||
a1e16c26 DC |
341 | /* |
342 | * Determine the initial size of the preallocation. We are beyond the current | |
343 | * EOF here, but we need to take into account whether this is a sparse write or | |
344 | * an extending write when determining the preallocation size. Hence we need to | |
345 | * look up the extent that ends at the current write offset and use the result | |
346 | * to determine the preallocation size. | |
347 | * | |
348 | * If the extent is a hole, then preallocation is essentially disabled. | |
349 | * Otherwise we take the size of the preceeding data extent as the basis for the | |
350 | * preallocation size. If the size of the extent is greater than half the | |
351 | * maximum extent length, then use the current offset as the basis. This ensures | |
352 | * that for large files the preallocation size always extends to MAXEXTLEN | |
353 | * rather than falling short due to things like stripe unit/width alignment of | |
354 | * real extents. | |
355 | */ | |
e8108ced | 356 | STATIC xfs_fsblock_t |
a1e16c26 DC |
357 | xfs_iomap_eof_prealloc_initial_size( |
358 | struct xfs_mount *mp, | |
359 | struct xfs_inode *ip, | |
360 | xfs_off_t offset, | |
361 | xfs_bmbt_irec_t *imap, | |
362 | int nimaps) | |
363 | { | |
364 | xfs_fileoff_t start_fsb; | |
365 | int imaps = 1; | |
366 | int error; | |
367 | ||
368 | ASSERT(nimaps >= imaps); | |
369 | ||
370 | /* if we are using a specific prealloc size, return now */ | |
371 | if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) | |
372 | return 0; | |
373 | ||
133eeb17 DC |
374 | /* If the file is small, then use the minimum prealloc */ |
375 | if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign)) | |
376 | return 0; | |
377 | ||
a1e16c26 DC |
378 | /* |
379 | * As we write multiple pages, the offset will always align to the | |
380 | * start of a page and hence point to a hole at EOF. i.e. if the size is | |
381 | * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096) | |
382 | * will return FSB 1. Hence if there are blocks in the file, we want to | |
383 | * point to the block prior to the EOF block and not the hole that maps | |
384 | * directly at @offset. | |
385 | */ | |
386 | start_fsb = XFS_B_TO_FSB(mp, offset); | |
387 | if (start_fsb) | |
388 | start_fsb--; | |
389 | error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE); | |
390 | if (error) | |
391 | return 0; | |
392 | ||
393 | ASSERT(imaps == 1); | |
394 | if (imap[0].br_startblock == HOLESTARTBLOCK) | |
395 | return 0; | |
396 | if (imap[0].br_blockcount <= (MAXEXTLEN >> 1)) | |
e114b5fc | 397 | return imap[0].br_blockcount << 1; |
a1e16c26 DC |
398 | return XFS_B_TO_FSB(mp, offset); |
399 | } | |
400 | ||
76a4202a BF |
401 | STATIC bool |
402 | xfs_quota_need_throttle( | |
403 | struct xfs_inode *ip, | |
404 | int type, | |
405 | xfs_fsblock_t alloc_blocks) | |
406 | { | |
407 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
408 | ||
409 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
410 | return false; | |
411 | ||
412 | /* no hi watermark, no throttle */ | |
413 | if (!dq->q_prealloc_hi_wmark) | |
414 | return false; | |
415 | ||
416 | /* under the lo watermark, no throttle */ | |
417 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
418 | return false; | |
419 | ||
420 | return true; | |
421 | } | |
422 | ||
423 | STATIC void | |
424 | xfs_quota_calc_throttle( | |
425 | struct xfs_inode *ip, | |
426 | int type, | |
427 | xfs_fsblock_t *qblocks, | |
f074051f BF |
428 | int *qshift, |
429 | int64_t *qfreesp) | |
76a4202a BF |
430 | { |
431 | int64_t freesp; | |
432 | int shift = 0; | |
433 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
434 | ||
5cca3f61 ES |
435 | /* no dq, or over hi wmark, squash the prealloc completely */ |
436 | if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
76a4202a | 437 | *qblocks = 0; |
f074051f | 438 | *qfreesp = 0; |
76a4202a BF |
439 | return; |
440 | } | |
441 | ||
442 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
443 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
444 | shift = 2; | |
445 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
446 | shift += 2; | |
447 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
448 | shift += 2; | |
449 | } | |
450 | ||
f074051f BF |
451 | if (freesp < *qfreesp) |
452 | *qfreesp = freesp; | |
453 | ||
76a4202a BF |
454 | /* only overwrite the throttle values if we are more aggressive */ |
455 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
456 | *qblocks = freesp; | |
457 | *qshift = shift; | |
458 | } | |
459 | } | |
460 | ||
055388a3 DC |
461 | /* |
462 | * If we don't have a user specified preallocation size, dynamically increase | |
463 | * the preallocation size as the size of the file grows. Cap the maximum size | |
464 | * at a single extent or less if the filesystem is near full. The closer the | |
465 | * filesystem is to full, the smaller the maximum prealocation. | |
466 | */ | |
467 | STATIC xfs_fsblock_t | |
468 | xfs_iomap_prealloc_size( | |
469 | struct xfs_mount *mp, | |
a1e16c26 DC |
470 | struct xfs_inode *ip, |
471 | xfs_off_t offset, | |
472 | struct xfs_bmbt_irec *imap, | |
473 | int nimaps) | |
055388a3 DC |
474 | { |
475 | xfs_fsblock_t alloc_blocks = 0; | |
3c58b5f8 BF |
476 | int shift = 0; |
477 | int64_t freesp; | |
76a4202a BF |
478 | xfs_fsblock_t qblocks; |
479 | int qshift = 0; | |
055388a3 | 480 | |
a1e16c26 DC |
481 | alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset, |
482 | imap, nimaps); | |
3c58b5f8 BF |
483 | if (!alloc_blocks) |
484 | goto check_writeio; | |
76a4202a | 485 | qblocks = alloc_blocks; |
3c58b5f8 | 486 | |
c9bdbdc0 BF |
487 | /* |
488 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
489 | * to the nearest power of two value after throttling. To prevent the | |
490 | * round down from unconditionally reducing the maximum supported prealloc | |
491 | * size, we round up first, apply appropriate throttling, round down and | |
492 | * cap the value to MAXEXTLEN. | |
493 | */ | |
494 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
495 | alloc_blocks); | |
3c58b5f8 | 496 | |
0d485ada | 497 | freesp = percpu_counter_read_positive(&mp->m_fdblocks); |
3c58b5f8 BF |
498 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { |
499 | shift = 2; | |
500 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
501 | shift++; | |
502 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
503 | shift++; | |
504 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
505 | shift++; | |
506 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
507 | shift++; | |
055388a3 | 508 | } |
76a4202a BF |
509 | |
510 | /* | |
f074051f BF |
511 | * Check each quota to cap the prealloc size, provide a shift value to |
512 | * throttle with and adjust amount of available space. | |
76a4202a BF |
513 | */ |
514 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
f074051f BF |
515 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift, |
516 | &freesp); | |
76a4202a | 517 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) |
f074051f BF |
518 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift, |
519 | &freesp); | |
76a4202a | 520 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) |
f074051f BF |
521 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift, |
522 | &freesp); | |
76a4202a BF |
523 | |
524 | /* | |
525 | * The final prealloc size is set to the minimum of free space available | |
526 | * in each of the quotas and the overall filesystem. | |
527 | * | |
528 | * The shift throttle value is set to the maximum value as determined by | |
529 | * the global low free space values and per-quota low free space values. | |
530 | */ | |
531 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
532 | shift = MAX(shift, qshift); | |
533 | ||
3c58b5f8 BF |
534 | if (shift) |
535 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
536 | /* |
537 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
538 | * alloc_blocks = 0. | |
539 | */ | |
540 | if (alloc_blocks) | |
541 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
542 | if (alloc_blocks > MAXEXTLEN) | |
543 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
544 | |
545 | /* | |
546 | * If we are still trying to allocate more space than is | |
547 | * available, squash the prealloc hard. This can happen if we | |
548 | * have a large file on a small filesystem and the above | |
549 | * lowspace thresholds are smaller than MAXEXTLEN. | |
550 | */ | |
551 | while (alloc_blocks && alloc_blocks >= freesp) | |
552 | alloc_blocks >>= 4; | |
055388a3 | 553 | |
3c58b5f8 | 554 | check_writeio: |
055388a3 DC |
555 | if (alloc_blocks < mp->m_writeio_blocks) |
556 | alloc_blocks = mp->m_writeio_blocks; | |
557 | ||
19cb7e38 BF |
558 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
559 | mp->m_writeio_blocks); | |
560 | ||
055388a3 DC |
561 | return alloc_blocks; |
562 | } | |
563 | ||
a206c817 | 564 | int |
1da177e4 LT |
565 | xfs_iomap_write_delay( |
566 | xfs_inode_t *ip, | |
f403b7f4 | 567 | xfs_off_t offset, |
1da177e4 | 568 | size_t count, |
405f8042 | 569 | xfs_bmbt_irec_t *ret_imap) |
1da177e4 LT |
570 | { |
571 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
572 | xfs_fileoff_t offset_fsb; |
573 | xfs_fileoff_t last_fsb; | |
dd9f438e NS |
574 | xfs_off_t aligned_offset; |
575 | xfs_fileoff_t ioalign; | |
dd9f438e | 576 | xfs_extlen_t extsz; |
1da177e4 | 577 | int nimaps; |
1da177e4 | 578 | xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS]; |
9aa05000 | 579 | int prealloc; |
dd9f438e | 580 | int error; |
1da177e4 | 581 | |
579aa9ca | 582 | ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); |
1da177e4 LT |
583 | |
584 | /* | |
585 | * Make sure that the dquots are there. This doesn't hold | |
586 | * the ilock across a disk read. | |
587 | */ | |
7d095257 | 588 | error = xfs_qm_dqattach_locked(ip, 0); |
1da177e4 | 589 | if (error) |
b474c7ae | 590 | return error; |
1da177e4 | 591 | |
957d0ebe | 592 | extsz = xfs_get_extsz_hint(ip); |
dd9f438e NS |
593 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
594 | ||
9f6c92b9 | 595 | error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count, |
405f8042 | 596 | imap, XFS_WRITE_IMAPS, &prealloc); |
dd9f438e NS |
597 | if (error) |
598 | return error; | |
1da177e4 | 599 | |
8de2bf93 | 600 | retry: |
dd9f438e | 601 | if (prealloc) { |
a1e16c26 DC |
602 | xfs_fsblock_t alloc_blocks; |
603 | ||
604 | alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap, | |
605 | XFS_WRITE_IMAPS); | |
055388a3 | 606 | |
1da177e4 LT |
607 | aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1)); |
608 | ioalign = XFS_B_TO_FSBT(mp, aligned_offset); | |
055388a3 | 609 | last_fsb = ioalign + alloc_blocks; |
dd9f438e NS |
610 | } else { |
611 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
1da177e4 | 612 | } |
1da177e4 | 613 | |
dd9f438e | 614 | if (prealloc || extsz) { |
9f6c92b9 | 615 | error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb); |
dd9f438e | 616 | if (error) |
1da177e4 | 617 | return error; |
1da177e4 | 618 | } |
dd9f438e | 619 | |
3ed9116e DC |
620 | /* |
621 | * Make sure preallocation does not create extents beyond the range we | |
622 | * actually support in this filesystem. | |
623 | */ | |
d2c28191 DC |
624 | if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes)) |
625 | last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
3ed9116e DC |
626 | |
627 | ASSERT(last_fsb > offset_fsb); | |
628 | ||
dd9f438e | 629 | nimaps = XFS_WRITE_IMAPS; |
4403280a CH |
630 | error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb, |
631 | imap, &nimaps, XFS_BMAPI_ENTIRE); | |
055388a3 DC |
632 | switch (error) { |
633 | case 0: | |
2451337d DC |
634 | case -ENOSPC: |
635 | case -EDQUOT: | |
055388a3 DC |
636 | break; |
637 | default: | |
b474c7ae | 638 | return error; |
055388a3 | 639 | } |
dd9f438e | 640 | |
1da177e4 | 641 | /* |
9aa05000 | 642 | * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry |
055388a3 | 643 | * without EOF preallocation. |
1da177e4 LT |
644 | */ |
645 | if (nimaps == 0) { | |
0b1b213f | 646 | trace_xfs_delalloc_enospc(ip, offset, count); |
9aa05000 DC |
647 | if (prealloc) { |
648 | prealloc = 0; | |
649 | error = 0; | |
650 | goto retry; | |
055388a3 | 651 | } |
2451337d | 652 | return error ? error : -ENOSPC; |
1da177e4 LT |
653 | } |
654 | ||
86c4d623 | 655 | if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 656 | return xfs_alert_fsblock_zero(ip, &imap[0]); |
dd9f438e | 657 | |
27b52867 BF |
658 | /* |
659 | * Tag the inode as speculatively preallocated so we can reclaim this | |
660 | * space on demand, if necessary. | |
661 | */ | |
662 | if (prealloc) | |
663 | xfs_inode_set_eofblocks_tag(ip); | |
664 | ||
dd9f438e | 665 | *ret_imap = imap[0]; |
1da177e4 LT |
666 | return 0; |
667 | } | |
668 | ||
669 | /* | |
670 | * Pass in a delayed allocate extent, convert it to real extents; | |
671 | * return to the caller the extent we create which maps on top of | |
672 | * the originating callers request. | |
673 | * | |
674 | * Called without a lock on the inode. | |
e4143a1c DC |
675 | * |
676 | * We no longer bother to look at the incoming map - all we have to | |
677 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 678 | */ |
a206c817 | 679 | int |
1da177e4 LT |
680 | xfs_iomap_write_allocate( |
681 | xfs_inode_t *ip, | |
f403b7f4 | 682 | xfs_off_t offset, |
405f8042 | 683 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
684 | { |
685 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
686 | xfs_fileoff_t offset_fsb, last_block; |
687 | xfs_fileoff_t end_fsb, map_start_fsb; | |
688 | xfs_fsblock_t first_block; | |
689 | xfs_bmap_free_t free_list; | |
690 | xfs_filblks_t count_fsb; | |
1da177e4 | 691 | xfs_trans_t *tp; |
f6106efa | 692 | int nimaps; |
1da177e4 LT |
693 | int error = 0; |
694 | int nres; | |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Make sure that the dquots are there. | |
698 | */ | |
7d095257 CH |
699 | error = xfs_qm_dqattach(ip, 0); |
700 | if (error) | |
b474c7ae | 701 | return error; |
1da177e4 | 702 | |
24e17b5f | 703 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
704 | count_fsb = imap->br_blockcount; |
705 | map_start_fsb = imap->br_startoff; | |
1da177e4 | 706 | |
ff6d6af2 | 707 | XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); |
1da177e4 LT |
708 | |
709 | while (count_fsb != 0) { | |
710 | /* | |
711 | * Set up a transaction with which to allocate the | |
712 | * backing store for the file. Do allocations in a | |
713 | * loop until we get some space in the range we are | |
714 | * interested in. The other space that might be allocated | |
715 | * is in the delayed allocation extent on which we sit | |
716 | * but before our buffer starts. | |
717 | */ | |
718 | ||
719 | nimaps = 0; | |
720 | while (nimaps == 0) { | |
1da177e4 | 721 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
253f4911 CH |
722 | |
723 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, nres, | |
724 | 0, XFS_TRANS_RESERVE, &tp); | |
725 | if (error) | |
b474c7ae | 726 | return error; |
253f4911 | 727 | |
1da177e4 | 728 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 729 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 730 | |
9d87c319 | 731 | xfs_bmap_init(&free_list, &first_block); |
1da177e4 | 732 | |
1da177e4 | 733 | /* |
e4143a1c DC |
734 | * it is possible that the extents have changed since |
735 | * we did the read call as we dropped the ilock for a | |
736 | * while. We have to be careful about truncates or hole | |
737 | * punchs here - we are not allowed to allocate | |
738 | * non-delalloc blocks here. | |
739 | * | |
740 | * The only protection against truncation is the pages | |
741 | * for the range we are being asked to convert are | |
742 | * locked and hence a truncate will block on them | |
743 | * first. | |
744 | * | |
745 | * As a result, if we go beyond the range we really | |
746 | * need and hit an delalloc extent boundary followed by | |
747 | * a hole while we have excess blocks in the map, we | |
748 | * will fill the hole incorrectly and overrun the | |
749 | * transaction reservation. | |
750 | * | |
751 | * Using a single map prevents this as we are forced to | |
752 | * check each map we look for overlap with the desired | |
753 | * range and abort as soon as we find it. Also, given | |
754 | * that we only return a single map, having one beyond | |
755 | * what we can return is probably a bit silly. | |
756 | * | |
757 | * We also need to check that we don't go beyond EOF; | |
758 | * this is a truncate optimisation as a truncate sets | |
759 | * the new file size before block on the pages we | |
760 | * currently have locked under writeback. Because they | |
761 | * are about to be tossed, we don't need to write them | |
762 | * back.... | |
1da177e4 | 763 | */ |
e4143a1c | 764 | nimaps = 1; |
ce7ae151 | 765 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7fb2cd4d | 766 | error = xfs_bmap_last_offset(ip, &last_block, |
7c9ef85c DC |
767 | XFS_DATA_FORK); |
768 | if (error) | |
769 | goto trans_cancel; | |
770 | ||
1da177e4 LT |
771 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
772 | if ((map_start_fsb + count_fsb) > last_block) { | |
773 | count_fsb = last_block - map_start_fsb; | |
774 | if (count_fsb == 0) { | |
2451337d | 775 | error = -EAGAIN; |
1da177e4 LT |
776 | goto trans_cancel; |
777 | } | |
778 | } | |
779 | ||
3070451e | 780 | /* |
3070451e CH |
781 | * From this point onwards we overwrite the imap |
782 | * pointer that the caller gave to us. | |
783 | */ | |
c0dc7828 | 784 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
dbd5c8c9 BF |
785 | count_fsb, 0, &first_block, |
786 | nres, imap, &nimaps, | |
787 | &free_list); | |
1da177e4 LT |
788 | if (error) |
789 | goto trans_cancel; | |
790 | ||
f6106efa | 791 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1da177e4 LT |
792 | if (error) |
793 | goto trans_cancel; | |
794 | ||
70393313 | 795 | error = xfs_trans_commit(tp); |
1da177e4 LT |
796 | if (error) |
797 | goto error0; | |
798 | ||
799 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
800 | } | |
801 | ||
802 | /* | |
803 | * See if we were able to allocate an extent that | |
804 | * covers at least part of the callers request | |
805 | */ | |
3070451e | 806 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 807 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 808 | |
3070451e CH |
809 | if ((offset_fsb >= imap->br_startoff) && |
810 | (offset_fsb < (imap->br_startoff + | |
811 | imap->br_blockcount))) { | |
ff6d6af2 | 812 | XFS_STATS_INC(mp, xs_xstrat_quick); |
e4143a1c | 813 | return 0; |
1da177e4 LT |
814 | } |
815 | ||
e4143a1c DC |
816 | /* |
817 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
818 | * file, just surrounding data, try again. |
819 | */ | |
3070451e CH |
820 | count_fsb -= imap->br_blockcount; |
821 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
822 | } |
823 | ||
824 | trans_cancel: | |
825 | xfs_bmap_cancel(&free_list); | |
4906e215 | 826 | xfs_trans_cancel(tp); |
1da177e4 LT |
827 | error0: |
828 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
b474c7ae | 829 | return error; |
1da177e4 LT |
830 | } |
831 | ||
832 | int | |
833 | xfs_iomap_write_unwritten( | |
834 | xfs_inode_t *ip, | |
f403b7f4 | 835 | xfs_off_t offset, |
d32057fc | 836 | xfs_off_t count) |
1da177e4 LT |
837 | { |
838 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
839 | xfs_fileoff_t offset_fsb; |
840 | xfs_filblks_t count_fsb; | |
841 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
842 | xfs_fsblock_t firstfsb; |
843 | int nimaps; | |
844 | xfs_trans_t *tp; | |
845 | xfs_bmbt_irec_t imap; | |
846 | xfs_bmap_free_t free_list; | |
84803fb7 | 847 | xfs_fsize_t i_size; |
dd9f438e | 848 | uint resblks; |
1da177e4 | 849 | int error; |
1da177e4 | 850 | |
0b1b213f | 851 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
852 | |
853 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
854 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
855 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
856 | ||
4ddd8bb1 LM |
857 | /* |
858 | * Reserve enough blocks in this transaction for two complete extent | |
859 | * btree splits. We may be converting the middle part of an unwritten | |
860 | * extent and in this case we will insert two new extents in the btree | |
861 | * each of which could cause a full split. | |
862 | * | |
863 | * This reservation amount will be used in the first call to | |
864 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
865 | * rest of the operation. | |
866 | */ | |
dd9f438e | 867 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 868 | |
dd9f438e | 869 | do { |
1da177e4 | 870 | /* |
253f4911 | 871 | * Set up a transaction to convert the range of extents |
1da177e4 LT |
872 | * from unwritten to real. Do allocations in a loop until |
873 | * we have covered the range passed in. | |
80641dc6 | 874 | * |
253f4911 CH |
875 | * Note that we can't risk to recursing back into the filesystem |
876 | * here as we might be asked to write out the same inode that we | |
877 | * complete here and might deadlock on the iolock. | |
1da177e4 | 878 | */ |
253f4911 CH |
879 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, |
880 | XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp); | |
881 | if (error) | |
b474c7ae | 882 | return error; |
1da177e4 LT |
883 | |
884 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 885 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
886 | |
887 | /* | |
888 | * Modify the unwritten extent state of the buffer. | |
889 | */ | |
9d87c319 | 890 | xfs_bmap_init(&free_list, &firstfsb); |
1da177e4 | 891 | nimaps = 1; |
c0dc7828 | 892 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
dbd5c8c9 BF |
893 | XFS_BMAPI_CONVERT, &firstfsb, resblks, |
894 | &imap, &nimaps, &free_list); | |
1da177e4 LT |
895 | if (error) |
896 | goto error_on_bmapi_transaction; | |
897 | ||
84803fb7 CH |
898 | /* |
899 | * Log the updated inode size as we go. We have to be careful | |
900 | * to only log it up to the actual write offset if it is | |
901 | * halfway into a block. | |
902 | */ | |
903 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
904 | if (i_size > offset + count) | |
905 | i_size = offset + count; | |
906 | ||
907 | i_size = xfs_new_eof(ip, i_size); | |
908 | if (i_size) { | |
909 | ip->i_d.di_size = i_size; | |
910 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
911 | } | |
912 | ||
f6106efa | 913 | error = xfs_bmap_finish(&tp, &free_list, NULL); |
1da177e4 LT |
914 | if (error) |
915 | goto error_on_bmapi_transaction; | |
916 | ||
70393313 | 917 | error = xfs_trans_commit(tp); |
1da177e4 LT |
918 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
919 | if (error) | |
b474c7ae | 920 | return error; |
572d95f4 | 921 | |
86c4d623 | 922 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 923 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
924 | |
925 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
926 | /* | |
927 | * The numblks_fsb value should always get | |
928 | * smaller, otherwise the loop is stuck. | |
929 | */ | |
930 | ASSERT(imap.br_blockcount); | |
931 | break; | |
932 | } | |
933 | offset_fsb += numblks_fsb; | |
934 | count_fsb -= numblks_fsb; | |
935 | } while (count_fsb > 0); | |
936 | ||
937 | return 0; | |
938 | ||
939 | error_on_bmapi_transaction: | |
940 | xfs_bmap_cancel(&free_list); | |
4906e215 | 941 | xfs_trans_cancel(tp); |
1da177e4 | 942 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
b474c7ae | 943 | return error; |
1da177e4 | 944 | } |
3b3dce05 CH |
945 | |
946 | void | |
947 | xfs_bmbt_to_iomap( | |
948 | struct xfs_inode *ip, | |
949 | struct iomap *iomap, | |
950 | struct xfs_bmbt_irec *imap) | |
951 | { | |
952 | struct xfs_mount *mp = ip->i_mount; | |
953 | ||
954 | if (imap->br_startblock == HOLESTARTBLOCK) { | |
955 | iomap->blkno = IOMAP_NULL_BLOCK; | |
956 | iomap->type = IOMAP_HOLE; | |
957 | } else if (imap->br_startblock == DELAYSTARTBLOCK) { | |
958 | iomap->blkno = IOMAP_NULL_BLOCK; | |
959 | iomap->type = IOMAP_DELALLOC; | |
960 | } else { | |
961 | iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock); | |
962 | if (imap->br_state == XFS_EXT_UNWRITTEN) | |
963 | iomap->type = IOMAP_UNWRITTEN; | |
964 | else | |
965 | iomap->type = IOMAP_MAPPED; | |
966 | } | |
967 | iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff); | |
968 | iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount); | |
969 | iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip)); | |
970 | } | |
68a9f5e7 CH |
971 | |
972 | static inline bool imap_needs_alloc(struct xfs_bmbt_irec *imap, int nimaps) | |
973 | { | |
974 | return !nimaps || | |
975 | imap->br_startblock == HOLESTARTBLOCK || | |
976 | imap->br_startblock == DELAYSTARTBLOCK; | |
977 | } | |
978 | ||
979 | static int | |
980 | xfs_file_iomap_begin( | |
981 | struct inode *inode, | |
982 | loff_t offset, | |
983 | loff_t length, | |
984 | unsigned flags, | |
985 | struct iomap *iomap) | |
986 | { | |
987 | struct xfs_inode *ip = XFS_I(inode); | |
988 | struct xfs_mount *mp = ip->i_mount; | |
989 | struct xfs_bmbt_irec imap; | |
990 | xfs_fileoff_t offset_fsb, end_fsb; | |
991 | int nimaps = 1, error = 0; | |
992 | ||
993 | if (XFS_FORCED_SHUTDOWN(mp)) | |
994 | return -EIO; | |
995 | ||
996 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
997 | ||
998 | ASSERT(offset <= mp->m_super->s_maxbytes); | |
999 | if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes) | |
1000 | length = mp->m_super->s_maxbytes - offset; | |
1001 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1002 | end_fsb = XFS_B_TO_FSB(mp, offset + length); | |
1003 | ||
1004 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, | |
1005 | &nimaps, XFS_BMAPI_ENTIRE); | |
1006 | if (error) { | |
1007 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1008 | return error; | |
1009 | } | |
1010 | ||
1011 | if ((flags & IOMAP_WRITE) && imap_needs_alloc(&imap, nimaps)) { | |
1012 | /* | |
1013 | * We cap the maximum length we map here to MAX_WRITEBACK_PAGES | |
1014 | * pages to keep the chunks of work done where somewhat symmetric | |
1015 | * with the work writeback does. This is a completely arbitrary | |
1016 | * number pulled out of thin air as a best guess for initial | |
1017 | * testing. | |
1018 | * | |
1019 | * Note that the values needs to be less than 32-bits wide until | |
1020 | * the lower level functions are updated. | |
1021 | */ | |
1022 | length = min_t(loff_t, length, 1024 * PAGE_SIZE); | |
1023 | if (xfs_get_extsz_hint(ip)) { | |
1024 | /* | |
1025 | * xfs_iomap_write_direct() expects the shared lock. It | |
1026 | * is unlocked on return. | |
1027 | */ | |
1028 | xfs_ilock_demote(ip, XFS_ILOCK_EXCL); | |
1029 | error = xfs_iomap_write_direct(ip, offset, length, &imap, | |
1030 | nimaps); | |
1031 | } else { | |
1032 | error = xfs_iomap_write_delay(ip, offset, length, &imap); | |
1033 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1034 | } | |
1035 | ||
1036 | if (error) | |
1037 | return error; | |
1038 | ||
1039 | trace_xfs_iomap_alloc(ip, offset, length, 0, &imap); | |
1040 | xfs_bmbt_to_iomap(ip, iomap, &imap); | |
1041 | } else if (nimaps) { | |
1042 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1043 | trace_xfs_iomap_found(ip, offset, length, 0, &imap); | |
1044 | xfs_bmbt_to_iomap(ip, iomap, &imap); | |
1045 | } else { | |
1046 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1047 | trace_xfs_iomap_not_found(ip, offset, length, 0, &imap); | |
1048 | iomap->blkno = IOMAP_NULL_BLOCK; | |
1049 | iomap->type = IOMAP_HOLE; | |
1050 | iomap->offset = offset; | |
1051 | iomap->length = length; | |
1052 | } | |
1053 | ||
1054 | return 0; | |
1055 | } | |
1056 | ||
1057 | static int | |
1058 | xfs_file_iomap_end_delalloc( | |
1059 | struct xfs_inode *ip, | |
1060 | loff_t offset, | |
1061 | loff_t length, | |
1062 | ssize_t written) | |
1063 | { | |
1064 | struct xfs_mount *mp = ip->i_mount; | |
1065 | xfs_fileoff_t start_fsb; | |
1066 | xfs_fileoff_t end_fsb; | |
1067 | int error = 0; | |
1068 | ||
1069 | start_fsb = XFS_B_TO_FSB(mp, offset + written); | |
1070 | end_fsb = XFS_B_TO_FSB(mp, offset + length); | |
1071 | ||
1072 | /* | |
1073 | * Trim back delalloc blocks if we didn't manage to write the whole | |
1074 | * range reserved. | |
1075 | * | |
1076 | * We don't need to care about racing delalloc as we hold i_mutex | |
1077 | * across the reserve/allocate/unreserve calls. If there are delalloc | |
1078 | * blocks in the range, they are ours. | |
1079 | */ | |
1080 | if (start_fsb < end_fsb) { | |
1081 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1082 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1083 | end_fsb - start_fsb); | |
1084 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1085 | ||
1086 | if (error && !XFS_FORCED_SHUTDOWN(mp)) { | |
1087 | xfs_alert(mp, "%s: unable to clean up ino %lld", | |
1088 | __func__, ip->i_ino); | |
1089 | return error; | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | static int | |
1097 | xfs_file_iomap_end( | |
1098 | struct inode *inode, | |
1099 | loff_t offset, | |
1100 | loff_t length, | |
1101 | ssize_t written, | |
1102 | unsigned flags, | |
1103 | struct iomap *iomap) | |
1104 | { | |
1105 | if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC) | |
1106 | return xfs_file_iomap_end_delalloc(XFS_I(inode), offset, | |
1107 | length, written); | |
1108 | return 0; | |
1109 | } | |
1110 | ||
1111 | struct iomap_ops xfs_iomap_ops = { | |
1112 | .iomap_begin = xfs_file_iomap_begin, | |
1113 | .iomap_end = xfs_file_iomap_end, | |
1114 | }; |