]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
3e57ecf6 | 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
51446f5b | 3 | * Copyright (c) 2016 Christoph Hellwig. |
7b718769 | 4 | * All Rights Reserved. |
1da177e4 | 5 | * |
7b718769 NS |
6 | * This program is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
8 | * published by the Free Software Foundation. |
9 | * | |
7b718769 NS |
10 | * This program is distributed in the hope that it would be useful, |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
1da177e4 | 14 | * |
7b718769 NS |
15 | * You should have received a copy of the GNU General Public License |
16 | * along with this program; if not, write the Free Software Foundation, | |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 18 | */ |
3b3dce05 | 19 | #include <linux/iomap.h> |
1da177e4 | 20 | #include "xfs.h" |
1da177e4 | 21 | #include "xfs_fs.h" |
70a9883c | 22 | #include "xfs_shared.h" |
239880ef DC |
23 | #include "xfs_format.h" |
24 | #include "xfs_log_format.h" | |
25 | #include "xfs_trans_resv.h" | |
1da177e4 | 26 | #include "xfs_mount.h" |
3ab78df2 | 27 | #include "xfs_defer.h" |
1da177e4 | 28 | #include "xfs_inode.h" |
a844f451 | 29 | #include "xfs_btree.h" |
a4fbe6ab | 30 | #include "xfs_bmap_btree.h" |
1da177e4 | 31 | #include "xfs_bmap.h" |
68988114 | 32 | #include "xfs_bmap_util.h" |
1da177e4 | 33 | #include "xfs_error.h" |
a4fbe6ab | 34 | #include "xfs_trans.h" |
1da177e4 | 35 | #include "xfs_trans_space.h" |
1da177e4 | 36 | #include "xfs_iomap.h" |
0b1b213f | 37 | #include "xfs_trace.h" |
27b52867 | 38 | #include "xfs_icache.h" |
a4fbe6ab | 39 | #include "xfs_quota.h" |
76a4202a BF |
40 | #include "xfs_dquot_item.h" |
41 | #include "xfs_dquot.h" | |
2a06705c | 42 | #include "xfs_reflink.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | |
45 | #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \ | |
46 | << mp->m_writeio_log) | |
1da177e4 | 47 | |
e9c49736 CH |
48 | void |
49 | xfs_bmbt_to_iomap( | |
50 | struct xfs_inode *ip, | |
51 | struct iomap *iomap, | |
52 | struct xfs_bmbt_irec *imap) | |
53 | { | |
54 | struct xfs_mount *mp = ip->i_mount; | |
55 | ||
56 | if (imap->br_startblock == HOLESTARTBLOCK) { | |
57 | iomap->blkno = IOMAP_NULL_BLOCK; | |
58 | iomap->type = IOMAP_HOLE; | |
59 | } else if (imap->br_startblock == DELAYSTARTBLOCK) { | |
60 | iomap->blkno = IOMAP_NULL_BLOCK; | |
61 | iomap->type = IOMAP_DELALLOC; | |
62 | } else { | |
63 | iomap->blkno = xfs_fsb_to_db(ip, imap->br_startblock); | |
64 | if (imap->br_state == XFS_EXT_UNWRITTEN) | |
65 | iomap->type = IOMAP_UNWRITTEN; | |
66 | else | |
67 | iomap->type = IOMAP_MAPPED; | |
68 | } | |
69 | iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff); | |
70 | iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount); | |
71 | iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip)); | |
72 | } | |
73 | ||
f8e3a825 CH |
74 | static xfs_extlen_t |
75 | xfs_eof_alignment( | |
76 | struct xfs_inode *ip, | |
77 | xfs_extlen_t extsize) | |
dd9f438e | 78 | { |
f8e3a825 CH |
79 | struct xfs_mount *mp = ip->i_mount; |
80 | xfs_extlen_t align = 0; | |
dd9f438e | 81 | |
bf322d98 CH |
82 | if (!XFS_IS_REALTIME_INODE(ip)) { |
83 | /* | |
84 | * Round up the allocation request to a stripe unit | |
85 | * (m_dalign) boundary if the file size is >= stripe unit | |
86 | * size, and we are allocating past the allocation eof. | |
87 | * | |
88 | * If mounted with the "-o swalloc" option the alignment is | |
89 | * increased from the strip unit size to the stripe width. | |
90 | */ | |
91 | if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC)) | |
92 | align = mp->m_swidth; | |
93 | else if (mp->m_dalign) | |
94 | align = mp->m_dalign; | |
95 | ||
76b57302 PW |
96 | if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align)) |
97 | align = 0; | |
bf322d98 | 98 | } |
dd9f438e NS |
99 | |
100 | /* | |
101 | * Always round up the allocation request to an extent boundary | |
102 | * (when file on a real-time subvolume or has di_extsize hint). | |
103 | */ | |
104 | if (extsize) { | |
76b57302 PW |
105 | if (align) |
106 | align = roundup_64(align, extsize); | |
dd9f438e NS |
107 | else |
108 | align = extsize; | |
dd9f438e NS |
109 | } |
110 | ||
f8e3a825 CH |
111 | return align; |
112 | } | |
113 | ||
114 | STATIC int | |
115 | xfs_iomap_eof_align_last_fsb( | |
116 | struct xfs_inode *ip, | |
117 | xfs_extlen_t extsize, | |
118 | xfs_fileoff_t *last_fsb) | |
119 | { | |
120 | xfs_extlen_t align = xfs_eof_alignment(ip, extsize); | |
121 | ||
76b57302 PW |
122 | if (align) { |
123 | xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align); | |
f8e3a825 CH |
124 | int eof, error; |
125 | ||
541d7d3c | 126 | error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof); |
dd9f438e NS |
127 | if (error) |
128 | return error; | |
129 | if (eof) | |
130 | *last_fsb = new_last_fsb; | |
131 | } | |
132 | return 0; | |
133 | } | |
134 | ||
572d95f4 | 135 | STATIC int |
6d4a8ecb | 136 | xfs_alert_fsblock_zero( |
572d95f4 NS |
137 | xfs_inode_t *ip, |
138 | xfs_bmbt_irec_t *imap) | |
139 | { | |
6a19d939 | 140 | xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO, |
572d95f4 NS |
141 | "Access to block zero in inode %llu " |
142 | "start_block: %llx start_off: %llx " | |
08e96e1a | 143 | "blkcnt: %llx extent-state: %x", |
572d95f4 NS |
144 | (unsigned long long)ip->i_ino, |
145 | (unsigned long long)imap->br_startblock, | |
146 | (unsigned long long)imap->br_startoff, | |
147 | (unsigned long long)imap->br_blockcount, | |
148 | imap->br_state); | |
2451337d | 149 | return -EFSCORRUPTED; |
572d95f4 NS |
150 | } |
151 | ||
a206c817 | 152 | int |
1da177e4 LT |
153 | xfs_iomap_write_direct( |
154 | xfs_inode_t *ip, | |
f403b7f4 | 155 | xfs_off_t offset, |
1da177e4 | 156 | size_t count, |
3070451e | 157 | xfs_bmbt_irec_t *imap, |
405f8042 | 158 | int nmaps) |
1da177e4 LT |
159 | { |
160 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
161 | xfs_fileoff_t offset_fsb; |
162 | xfs_fileoff_t last_fsb; | |
dd9f438e | 163 | xfs_filblks_t count_fsb, resaligned; |
1da177e4 | 164 | xfs_fsblock_t firstfsb; |
dd9f438e | 165 | xfs_extlen_t extsz, temp; |
0116d935 | 166 | int nimaps; |
06d10dd9 | 167 | int quota_flag; |
1da177e4 LT |
168 | int rt; |
169 | xfs_trans_t *tp; | |
2c3234d1 | 170 | struct xfs_defer_ops dfops; |
dd9f438e | 171 | uint qblocks, resblks, resrtextents; |
dd9f438e | 172 | int error; |
009c6e87 | 173 | int lockmode; |
1ca19157 | 174 | int bmapi_flags = XFS_BMAPI_PREALLOC; |
253f4911 | 175 | uint tflags = 0; |
1da177e4 | 176 | |
dd9f438e | 177 | rt = XFS_IS_REALTIME_INODE(ip); |
957d0ebe | 178 | extsz = xfs_get_extsz_hint(ip); |
009c6e87 BF |
179 | lockmode = XFS_ILOCK_SHARED; /* locked by caller */ |
180 | ||
181 | ASSERT(xfs_isilocked(ip, lockmode)); | |
1da177e4 | 182 | |
957d0ebe DC |
183 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
184 | last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))); | |
ce7ae151 | 185 | if ((offset + count) > XFS_ISIZE(ip)) { |
009c6e87 BF |
186 | /* |
187 | * Assert that the in-core extent list is present since this can | |
188 | * call xfs_iread_extents() and we only have the ilock shared. | |
189 | * This should be safe because the lock was held around a bmapi | |
190 | * call in the caller and we only need it to access the in-core | |
191 | * list. | |
192 | */ | |
193 | ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags & | |
194 | XFS_IFEXTENTS); | |
f8e3a825 | 195 | error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb); |
dd9f438e | 196 | if (error) |
009c6e87 | 197 | goto out_unlock; |
1da177e4 | 198 | } else { |
405f8042 | 199 | if (nmaps && (imap->br_startblock == HOLESTARTBLOCK)) |
dd9f438e | 200 | last_fsb = MIN(last_fsb, (xfs_fileoff_t) |
3070451e CH |
201 | imap->br_blockcount + |
202 | imap->br_startoff); | |
1da177e4 | 203 | } |
dd9f438e NS |
204 | count_fsb = last_fsb - offset_fsb; |
205 | ASSERT(count_fsb > 0); | |
206 | ||
207 | resaligned = count_fsb; | |
208 | if (unlikely(extsz)) { | |
209 | if ((temp = do_mod(offset_fsb, extsz))) | |
210 | resaligned += temp; | |
211 | if ((temp = do_mod(resaligned, extsz))) | |
212 | resaligned += extsz - temp; | |
213 | } | |
214 | ||
215 | if (unlikely(rt)) { | |
216 | resrtextents = qblocks = resaligned; | |
217 | resrtextents /= mp->m_sb.sb_rextsize; | |
84e1e99f DC |
218 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); |
219 | quota_flag = XFS_QMOPT_RES_RTBLKS; | |
220 | } else { | |
221 | resrtextents = 0; | |
dd9f438e | 222 | resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned); |
84e1e99f DC |
223 | quota_flag = XFS_QMOPT_RES_REGBLKS; |
224 | } | |
1da177e4 | 225 | |
009c6e87 BF |
226 | /* |
227 | * Drop the shared lock acquired by the caller, attach the dquot if | |
228 | * necessary and move on to transaction setup. | |
229 | */ | |
230 | xfs_iunlock(ip, lockmode); | |
231 | error = xfs_qm_dqattach(ip, 0); | |
232 | if (error) | |
233 | return error; | |
234 | ||
1ca19157 DC |
235 | /* |
236 | * For DAX, we do not allocate unwritten extents, but instead we zero | |
237 | * the block before we commit the transaction. Ideally we'd like to do | |
238 | * this outside the transaction context, but if we commit and then crash | |
239 | * we may not have zeroed the blocks and this will be exposed on | |
240 | * recovery of the allocation. Hence we must zero before commit. | |
3b0fe478 | 241 | * |
1ca19157 DC |
242 | * Further, if we are mapping unwritten extents here, we need to zero |
243 | * and convert them to written so that we don't need an unwritten extent | |
244 | * callback for DAX. This also means that we need to be able to dip into | |
3b0fe478 DC |
245 | * the reserve block pool for bmbt block allocation if there is no space |
246 | * left but we need to do unwritten extent conversion. | |
1ca19157 DC |
247 | */ |
248 | if (IS_DAX(VFS_I(ip))) { | |
249 | bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO; | |
3b0fe478 | 250 | if (ISUNWRITTEN(imap)) { |
253f4911 | 251 | tflags |= XFS_TRANS_RESERVE; |
3b0fe478 DC |
252 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
253 | } | |
1ca19157 | 254 | } |
253f4911 CH |
255 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents, |
256 | tflags, &tp); | |
257 | if (error) | |
b474c7ae | 258 | return error; |
507630b2 | 259 | |
009c6e87 BF |
260 | lockmode = XFS_ILOCK_EXCL; |
261 | xfs_ilock(ip, lockmode); | |
1da177e4 | 262 | |
7d095257 | 263 | error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag); |
dd9f438e | 264 | if (error) |
507630b2 | 265 | goto out_trans_cancel; |
1da177e4 | 266 | |
ddc3415a | 267 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 268 | |
1da177e4 | 269 | /* |
3070451e CH |
270 | * From this point onwards we overwrite the imap pointer that the |
271 | * caller gave to us. | |
1da177e4 | 272 | */ |
2c3234d1 | 273 | xfs_defer_init(&dfops, &firstfsb); |
06d10dd9 | 274 | nimaps = 1; |
d531d91d | 275 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
264e89ad | 276 | bmapi_flags, &firstfsb, resblks, imap, |
2c3234d1 | 277 | &nimaps, &dfops); |
06d10dd9 | 278 | if (error) |
507630b2 | 279 | goto out_bmap_cancel; |
1da177e4 LT |
280 | |
281 | /* | |
06d10dd9 | 282 | * Complete the transaction |
1da177e4 | 283 | */ |
2c3234d1 | 284 | error = xfs_defer_finish(&tp, &dfops, NULL); |
06d10dd9 | 285 | if (error) |
507630b2 | 286 | goto out_bmap_cancel; |
1ca19157 | 287 | |
70393313 | 288 | error = xfs_trans_commit(tp); |
06d10dd9 | 289 | if (error) |
507630b2 | 290 | goto out_unlock; |
1da177e4 | 291 | |
06d10dd9 NS |
292 | /* |
293 | * Copy any maps to caller's array and return any error. | |
294 | */ | |
1da177e4 | 295 | if (nimaps == 0) { |
2451337d | 296 | error = -ENOSPC; |
507630b2 | 297 | goto out_unlock; |
572d95f4 NS |
298 | } |
299 | ||
507630b2 | 300 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 301 | error = xfs_alert_fsblock_zero(ip, imap); |
1da177e4 | 302 | |
507630b2 | 303 | out_unlock: |
009c6e87 | 304 | xfs_iunlock(ip, lockmode); |
507630b2 | 305 | return error; |
1da177e4 | 306 | |
507630b2 | 307 | out_bmap_cancel: |
2c3234d1 | 308 | xfs_defer_cancel(&dfops); |
ea562ed6 | 309 | xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag); |
507630b2 | 310 | out_trans_cancel: |
4906e215 | 311 | xfs_trans_cancel(tp); |
507630b2 | 312 | goto out_unlock; |
1da177e4 LT |
313 | } |
314 | ||
76a4202a BF |
315 | STATIC bool |
316 | xfs_quota_need_throttle( | |
317 | struct xfs_inode *ip, | |
318 | int type, | |
319 | xfs_fsblock_t alloc_blocks) | |
320 | { | |
321 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
322 | ||
323 | if (!dq || !xfs_this_quota_on(ip->i_mount, type)) | |
324 | return false; | |
325 | ||
326 | /* no hi watermark, no throttle */ | |
327 | if (!dq->q_prealloc_hi_wmark) | |
328 | return false; | |
329 | ||
330 | /* under the lo watermark, no throttle */ | |
331 | if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark) | |
332 | return false; | |
333 | ||
334 | return true; | |
335 | } | |
336 | ||
337 | STATIC void | |
338 | xfs_quota_calc_throttle( | |
339 | struct xfs_inode *ip, | |
340 | int type, | |
341 | xfs_fsblock_t *qblocks, | |
f074051f BF |
342 | int *qshift, |
343 | int64_t *qfreesp) | |
76a4202a BF |
344 | { |
345 | int64_t freesp; | |
346 | int shift = 0; | |
347 | struct xfs_dquot *dq = xfs_inode_dquot(ip, type); | |
348 | ||
5cca3f61 ES |
349 | /* no dq, or over hi wmark, squash the prealloc completely */ |
350 | if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) { | |
76a4202a | 351 | *qblocks = 0; |
f074051f | 352 | *qfreesp = 0; |
76a4202a BF |
353 | return; |
354 | } | |
355 | ||
356 | freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount; | |
357 | if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) { | |
358 | shift = 2; | |
359 | if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT]) | |
360 | shift += 2; | |
361 | if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT]) | |
362 | shift += 2; | |
363 | } | |
364 | ||
f074051f BF |
365 | if (freesp < *qfreesp) |
366 | *qfreesp = freesp; | |
367 | ||
76a4202a BF |
368 | /* only overwrite the throttle values if we are more aggressive */ |
369 | if ((freesp >> shift) < (*qblocks >> *qshift)) { | |
370 | *qblocks = freesp; | |
371 | *qshift = shift; | |
372 | } | |
373 | } | |
374 | ||
055388a3 | 375 | /* |
51446f5b CH |
376 | * If we are doing a write at the end of the file and there are no allocations |
377 | * past this one, then extend the allocation out to the file system's write | |
378 | * iosize. | |
379 | * | |
055388a3 | 380 | * If we don't have a user specified preallocation size, dynamically increase |
51446f5b | 381 | * the preallocation size as the size of the file grows. Cap the maximum size |
055388a3 DC |
382 | * at a single extent or less if the filesystem is near full. The closer the |
383 | * filesystem is to full, the smaller the maximum prealocation. | |
51446f5b CH |
384 | * |
385 | * As an exception we don't do any preallocation at all if the file is smaller | |
386 | * than the minimum preallocation and we are using the default dynamic | |
387 | * preallocation scheme, as it is likely this is the only write to the file that | |
388 | * is going to be done. | |
389 | * | |
390 | * We clean up any extra space left over when the file is closed in | |
391 | * xfs_inactive(). | |
055388a3 DC |
392 | */ |
393 | STATIC xfs_fsblock_t | |
394 | xfs_iomap_prealloc_size( | |
a1e16c26 | 395 | struct xfs_inode *ip, |
51446f5b CH |
396 | loff_t offset, |
397 | loff_t count, | |
398 | xfs_extnum_t idx, | |
399 | struct xfs_bmbt_irec *prev) | |
055388a3 | 400 | { |
51446f5b CH |
401 | struct xfs_mount *mp = ip->i_mount; |
402 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
3c58b5f8 BF |
403 | int shift = 0; |
404 | int64_t freesp; | |
76a4202a BF |
405 | xfs_fsblock_t qblocks; |
406 | int qshift = 0; | |
51446f5b CH |
407 | xfs_fsblock_t alloc_blocks = 0; |
408 | ||
409 | if (offset + count <= XFS_ISIZE(ip)) | |
410 | return 0; | |
411 | ||
412 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) && | |
413 | (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))) | |
414 | return 0; | |
415 | ||
416 | /* | |
417 | * If an explicit allocsize is set, the file is small, or we | |
418 | * are writing behind a hole, then use the minimum prealloc: | |
419 | */ | |
420 | if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) || | |
421 | XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) || | |
422 | idx == 0 || | |
423 | prev->br_startoff + prev->br_blockcount < offset_fsb) | |
424 | return mp->m_writeio_blocks; | |
055388a3 | 425 | |
51446f5b CH |
426 | /* |
427 | * Determine the initial size of the preallocation. We are beyond the | |
428 | * current EOF here, but we need to take into account whether this is | |
429 | * a sparse write or an extending write when determining the | |
430 | * preallocation size. Hence we need to look up the extent that ends | |
431 | * at the current write offset and use the result to determine the | |
432 | * preallocation size. | |
433 | * | |
434 | * If the extent is a hole, then preallocation is essentially disabled. | |
435 | * Otherwise we take the size of the preceding data extent as the basis | |
436 | * for the preallocation size. If the size of the extent is greater than | |
437 | * half the maximum extent length, then use the current offset as the | |
438 | * basis. This ensures that for large files the preallocation size | |
439 | * always extends to MAXEXTLEN rather than falling short due to things | |
440 | * like stripe unit/width alignment of real extents. | |
441 | */ | |
442 | if (prev->br_blockcount <= (MAXEXTLEN >> 1)) | |
443 | alloc_blocks = prev->br_blockcount << 1; | |
444 | else | |
445 | alloc_blocks = XFS_B_TO_FSB(mp, offset); | |
3c58b5f8 BF |
446 | if (!alloc_blocks) |
447 | goto check_writeio; | |
76a4202a | 448 | qblocks = alloc_blocks; |
3c58b5f8 | 449 | |
c9bdbdc0 BF |
450 | /* |
451 | * MAXEXTLEN is not a power of two value but we round the prealloc down | |
452 | * to the nearest power of two value after throttling. To prevent the | |
453 | * round down from unconditionally reducing the maximum supported prealloc | |
454 | * size, we round up first, apply appropriate throttling, round down and | |
455 | * cap the value to MAXEXTLEN. | |
456 | */ | |
457 | alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN), | |
458 | alloc_blocks); | |
3c58b5f8 | 459 | |
0d485ada | 460 | freesp = percpu_counter_read_positive(&mp->m_fdblocks); |
3c58b5f8 BF |
461 | if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) { |
462 | shift = 2; | |
463 | if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT]) | |
464 | shift++; | |
465 | if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT]) | |
466 | shift++; | |
467 | if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT]) | |
468 | shift++; | |
469 | if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT]) | |
470 | shift++; | |
055388a3 | 471 | } |
76a4202a BF |
472 | |
473 | /* | |
f074051f BF |
474 | * Check each quota to cap the prealloc size, provide a shift value to |
475 | * throttle with and adjust amount of available space. | |
76a4202a BF |
476 | */ |
477 | if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks)) | |
f074051f BF |
478 | xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift, |
479 | &freesp); | |
76a4202a | 480 | if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks)) |
f074051f BF |
481 | xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift, |
482 | &freesp); | |
76a4202a | 483 | if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks)) |
f074051f BF |
484 | xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift, |
485 | &freesp); | |
76a4202a BF |
486 | |
487 | /* | |
488 | * The final prealloc size is set to the minimum of free space available | |
489 | * in each of the quotas and the overall filesystem. | |
490 | * | |
491 | * The shift throttle value is set to the maximum value as determined by | |
492 | * the global low free space values and per-quota low free space values. | |
493 | */ | |
494 | alloc_blocks = MIN(alloc_blocks, qblocks); | |
495 | shift = MAX(shift, qshift); | |
496 | ||
3c58b5f8 BF |
497 | if (shift) |
498 | alloc_blocks >>= shift; | |
c9bdbdc0 BF |
499 | /* |
500 | * rounddown_pow_of_two() returns an undefined result if we pass in | |
501 | * alloc_blocks = 0. | |
502 | */ | |
503 | if (alloc_blocks) | |
504 | alloc_blocks = rounddown_pow_of_two(alloc_blocks); | |
505 | if (alloc_blocks > MAXEXTLEN) | |
506 | alloc_blocks = MAXEXTLEN; | |
3c58b5f8 BF |
507 | |
508 | /* | |
509 | * If we are still trying to allocate more space than is | |
510 | * available, squash the prealloc hard. This can happen if we | |
511 | * have a large file on a small filesystem and the above | |
512 | * lowspace thresholds are smaller than MAXEXTLEN. | |
513 | */ | |
514 | while (alloc_blocks && alloc_blocks >= freesp) | |
515 | alloc_blocks >>= 4; | |
3c58b5f8 | 516 | check_writeio: |
055388a3 DC |
517 | if (alloc_blocks < mp->m_writeio_blocks) |
518 | alloc_blocks = mp->m_writeio_blocks; | |
19cb7e38 BF |
519 | trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift, |
520 | mp->m_writeio_blocks); | |
055388a3 DC |
521 | return alloc_blocks; |
522 | } | |
523 | ||
51446f5b CH |
524 | static int |
525 | xfs_file_iomap_begin_delay( | |
526 | struct inode *inode, | |
527 | loff_t offset, | |
528 | loff_t count, | |
529 | unsigned flags, | |
530 | struct iomap *iomap) | |
1da177e4 | 531 | { |
51446f5b CH |
532 | struct xfs_inode *ip = XFS_I(inode); |
533 | struct xfs_mount *mp = ip->i_mount; | |
534 | struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK); | |
535 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
536 | xfs_fileoff_t maxbytes_fsb = | |
537 | XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes); | |
538 | xfs_fileoff_t end_fsb, orig_end_fsb; | |
539 | int error = 0, eof = 0; | |
540 | struct xfs_bmbt_irec got; | |
541 | struct xfs_bmbt_irec prev; | |
542 | xfs_extnum_t idx; | |
543 | ||
544 | ASSERT(!XFS_IS_REALTIME_INODE(ip)); | |
545 | ASSERT(!xfs_get_extsz_hint(ip)); | |
dd9f438e | 546 | |
51446f5b | 547 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
1da177e4 | 548 | |
51446f5b CH |
549 | if (unlikely(XFS_TEST_ERROR( |
550 | (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS && | |
551 | XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE), | |
552 | mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) { | |
553 | XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp); | |
554 | error = -EFSCORRUPTED; | |
555 | goto out_unlock; | |
556 | } | |
a1e16c26 | 557 | |
51446f5b | 558 | XFS_STATS_INC(mp, xs_blk_mapw); |
055388a3 | 559 | |
51446f5b CH |
560 | if (!(ifp->if_flags & XFS_IFEXTENTS)) { |
561 | error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK); | |
562 | if (error) | |
563 | goto out_unlock; | |
1da177e4 | 564 | } |
1da177e4 | 565 | |
51446f5b CH |
566 | xfs_bmap_search_extents(ip, offset_fsb, XFS_DATA_FORK, &eof, &idx, |
567 | &got, &prev); | |
568 | if (!eof && got.br_startoff <= offset_fsb) { | |
569 | trace_xfs_iomap_found(ip, offset, count, 0, &got); | |
570 | goto done; | |
1da177e4 | 571 | } |
dd9f438e | 572 | |
51446f5b CH |
573 | error = xfs_qm_dqattach_locked(ip, 0); |
574 | if (error) | |
575 | goto out_unlock; | |
576 | ||
3ed9116e | 577 | /* |
51446f5b CH |
578 | * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages |
579 | * to keep the chunks of work done where somewhat symmetric with the | |
580 | * work writeback does. This is a completely arbitrary number pulled | |
581 | * out of thin air as a best guess for initial testing. | |
582 | * | |
583 | * Note that the values needs to be less than 32-bits wide until | |
584 | * the lower level functions are updated. | |
3ed9116e | 585 | */ |
51446f5b CH |
586 | count = min_t(loff_t, count, 1024 * PAGE_SIZE); |
587 | end_fsb = orig_end_fsb = | |
588 | min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb); | |
589 | ||
590 | if (eof) { | |
591 | xfs_fsblock_t prealloc_blocks; | |
3ed9116e | 592 | |
51446f5b CH |
593 | prealloc_blocks = |
594 | xfs_iomap_prealloc_size(ip, offset, count, idx, &prev); | |
595 | if (prealloc_blocks) { | |
596 | xfs_extlen_t align; | |
597 | xfs_off_t end_offset; | |
3ed9116e | 598 | |
51446f5b CH |
599 | end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1); |
600 | end_fsb = XFS_B_TO_FSBT(mp, end_offset) + | |
601 | prealloc_blocks; | |
602 | ||
603 | align = xfs_eof_alignment(ip, 0); | |
604 | if (align) | |
605 | end_fsb = roundup_64(end_fsb, align); | |
606 | ||
607 | end_fsb = min(end_fsb, maxbytes_fsb); | |
608 | ASSERT(end_fsb > offset_fsb); | |
609 | } | |
610 | } | |
611 | ||
612 | retry: | |
be51f811 | 613 | error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb, |
51446f5b CH |
614 | end_fsb - offset_fsb, &got, |
615 | &prev, &idx, eof); | |
055388a3 DC |
616 | switch (error) { |
617 | case 0: | |
51446f5b | 618 | break; |
2451337d DC |
619 | case -ENOSPC: |
620 | case -EDQUOT: | |
51446f5b | 621 | /* retry without any preallocation */ |
0b1b213f | 622 | trace_xfs_delalloc_enospc(ip, offset, count); |
51446f5b CH |
623 | if (end_fsb != orig_end_fsb) { |
624 | end_fsb = orig_end_fsb; | |
9aa05000 | 625 | goto retry; |
055388a3 | 626 | } |
51446f5b CH |
627 | /*FALLTHRU*/ |
628 | default: | |
629 | goto out_unlock; | |
1da177e4 LT |
630 | } |
631 | ||
27b52867 BF |
632 | /* |
633 | * Tag the inode as speculatively preallocated so we can reclaim this | |
634 | * space on demand, if necessary. | |
635 | */ | |
51446f5b | 636 | if (end_fsb != orig_end_fsb) |
27b52867 BF |
637 | xfs_inode_set_eofblocks_tag(ip); |
638 | ||
51446f5b CH |
639 | trace_xfs_iomap_alloc(ip, offset, count, 0, &got); |
640 | done: | |
641 | if (isnullstartblock(got.br_startblock)) | |
642 | got.br_startblock = DELAYSTARTBLOCK; | |
643 | ||
644 | if (!got.br_startblock) { | |
645 | error = xfs_alert_fsblock_zero(ip, &got); | |
646 | if (error) | |
647 | goto out_unlock; | |
648 | } | |
649 | ||
650 | xfs_bmbt_to_iomap(ip, iomap, &got); | |
651 | ||
652 | out_unlock: | |
653 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
654 | return error; | |
1da177e4 LT |
655 | } |
656 | ||
657 | /* | |
658 | * Pass in a delayed allocate extent, convert it to real extents; | |
659 | * return to the caller the extent we create which maps on top of | |
660 | * the originating callers request. | |
661 | * | |
662 | * Called without a lock on the inode. | |
e4143a1c DC |
663 | * |
664 | * We no longer bother to look at the incoming map - all we have to | |
665 | * guarantee is that whatever we allocate fills the required range. | |
1da177e4 | 666 | */ |
a206c817 | 667 | int |
1da177e4 LT |
668 | xfs_iomap_write_allocate( |
669 | xfs_inode_t *ip, | |
60b4984f | 670 | int whichfork, |
f403b7f4 | 671 | xfs_off_t offset, |
405f8042 | 672 | xfs_bmbt_irec_t *imap) |
1da177e4 LT |
673 | { |
674 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
675 | xfs_fileoff_t offset_fsb, last_block; |
676 | xfs_fileoff_t end_fsb, map_start_fsb; | |
677 | xfs_fsblock_t first_block; | |
2c3234d1 | 678 | struct xfs_defer_ops dfops; |
1da177e4 | 679 | xfs_filblks_t count_fsb; |
1da177e4 | 680 | xfs_trans_t *tp; |
f6106efa | 681 | int nimaps; |
1da177e4 | 682 | int error = 0; |
60b4984f | 683 | int flags = 0; |
1da177e4 LT |
684 | int nres; |
685 | ||
60b4984f DW |
686 | if (whichfork == XFS_COW_FORK) |
687 | flags |= XFS_BMAPI_COWFORK; | |
688 | ||
1da177e4 LT |
689 | /* |
690 | * Make sure that the dquots are there. | |
691 | */ | |
7d095257 CH |
692 | error = xfs_qm_dqattach(ip, 0); |
693 | if (error) | |
b474c7ae | 694 | return error; |
1da177e4 | 695 | |
24e17b5f | 696 | offset_fsb = XFS_B_TO_FSBT(mp, offset); |
3070451e CH |
697 | count_fsb = imap->br_blockcount; |
698 | map_start_fsb = imap->br_startoff; | |
1da177e4 | 699 | |
ff6d6af2 | 700 | XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb)); |
1da177e4 LT |
701 | |
702 | while (count_fsb != 0) { | |
703 | /* | |
704 | * Set up a transaction with which to allocate the | |
705 | * backing store for the file. Do allocations in a | |
706 | * loop until we get some space in the range we are | |
707 | * interested in. The other space that might be allocated | |
708 | * is in the delayed allocation extent on which we sit | |
709 | * but before our buffer starts. | |
710 | */ | |
1da177e4 LT |
711 | nimaps = 0; |
712 | while (nimaps == 0) { | |
1da177e4 | 713 | nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK); |
0af32fb4 CH |
714 | /* |
715 | * We have already reserved space for the extent and any | |
716 | * indirect blocks when creating the delalloc extent, | |
717 | * there is no need to reserve space in this transaction | |
718 | * again. | |
719 | */ | |
720 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, | |
253f4911 CH |
721 | 0, XFS_TRANS_RESERVE, &tp); |
722 | if (error) | |
b474c7ae | 723 | return error; |
253f4911 | 724 | |
1da177e4 | 725 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
ddc3415a | 726 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 | 727 | |
2c3234d1 | 728 | xfs_defer_init(&dfops, &first_block); |
1da177e4 | 729 | |
1da177e4 | 730 | /* |
e4143a1c DC |
731 | * it is possible that the extents have changed since |
732 | * we did the read call as we dropped the ilock for a | |
733 | * while. We have to be careful about truncates or hole | |
734 | * punchs here - we are not allowed to allocate | |
735 | * non-delalloc blocks here. | |
736 | * | |
737 | * The only protection against truncation is the pages | |
738 | * for the range we are being asked to convert are | |
739 | * locked and hence a truncate will block on them | |
740 | * first. | |
741 | * | |
742 | * As a result, if we go beyond the range we really | |
743 | * need and hit an delalloc extent boundary followed by | |
744 | * a hole while we have excess blocks in the map, we | |
745 | * will fill the hole incorrectly and overrun the | |
746 | * transaction reservation. | |
747 | * | |
748 | * Using a single map prevents this as we are forced to | |
749 | * check each map we look for overlap with the desired | |
750 | * range and abort as soon as we find it. Also, given | |
751 | * that we only return a single map, having one beyond | |
752 | * what we can return is probably a bit silly. | |
753 | * | |
754 | * We also need to check that we don't go beyond EOF; | |
755 | * this is a truncate optimisation as a truncate sets | |
756 | * the new file size before block on the pages we | |
757 | * currently have locked under writeback. Because they | |
758 | * are about to be tossed, we don't need to write them | |
759 | * back.... | |
1da177e4 | 760 | */ |
e4143a1c | 761 | nimaps = 1; |
ce7ae151 | 762 | end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)); |
7fb2cd4d | 763 | error = xfs_bmap_last_offset(ip, &last_block, |
7c9ef85c DC |
764 | XFS_DATA_FORK); |
765 | if (error) | |
766 | goto trans_cancel; | |
767 | ||
1da177e4 LT |
768 | last_block = XFS_FILEOFF_MAX(last_block, end_fsb); |
769 | if ((map_start_fsb + count_fsb) > last_block) { | |
770 | count_fsb = last_block - map_start_fsb; | |
771 | if (count_fsb == 0) { | |
2451337d | 772 | error = -EAGAIN; |
1da177e4 LT |
773 | goto trans_cancel; |
774 | } | |
775 | } | |
776 | ||
3070451e | 777 | /* |
3070451e CH |
778 | * From this point onwards we overwrite the imap |
779 | * pointer that the caller gave to us. | |
780 | */ | |
c0dc7828 | 781 | error = xfs_bmapi_write(tp, ip, map_start_fsb, |
60b4984f | 782 | count_fsb, flags, &first_block, |
dbd5c8c9 | 783 | nres, imap, &nimaps, |
2c3234d1 | 784 | &dfops); |
1da177e4 LT |
785 | if (error) |
786 | goto trans_cancel; | |
787 | ||
2c3234d1 | 788 | error = xfs_defer_finish(&tp, &dfops, NULL); |
1da177e4 LT |
789 | if (error) |
790 | goto trans_cancel; | |
791 | ||
70393313 | 792 | error = xfs_trans_commit(tp); |
1da177e4 LT |
793 | if (error) |
794 | goto error0; | |
795 | ||
796 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
797 | } | |
798 | ||
799 | /* | |
800 | * See if we were able to allocate an extent that | |
801 | * covers at least part of the callers request | |
802 | */ | |
3070451e | 803 | if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 804 | return xfs_alert_fsblock_zero(ip, imap); |
86c4d623 | 805 | |
3070451e CH |
806 | if ((offset_fsb >= imap->br_startoff) && |
807 | (offset_fsb < (imap->br_startoff + | |
808 | imap->br_blockcount))) { | |
ff6d6af2 | 809 | XFS_STATS_INC(mp, xs_xstrat_quick); |
e4143a1c | 810 | return 0; |
1da177e4 LT |
811 | } |
812 | ||
e4143a1c DC |
813 | /* |
814 | * So far we have not mapped the requested part of the | |
1da177e4 LT |
815 | * file, just surrounding data, try again. |
816 | */ | |
3070451e CH |
817 | count_fsb -= imap->br_blockcount; |
818 | map_start_fsb = imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
819 | } |
820 | ||
821 | trans_cancel: | |
2c3234d1 | 822 | xfs_defer_cancel(&dfops); |
4906e215 | 823 | xfs_trans_cancel(tp); |
1da177e4 LT |
824 | error0: |
825 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
b474c7ae | 826 | return error; |
1da177e4 LT |
827 | } |
828 | ||
829 | int | |
830 | xfs_iomap_write_unwritten( | |
831 | xfs_inode_t *ip, | |
f403b7f4 | 832 | xfs_off_t offset, |
d32057fc | 833 | xfs_off_t count) |
1da177e4 LT |
834 | { |
835 | xfs_mount_t *mp = ip->i_mount; | |
1da177e4 LT |
836 | xfs_fileoff_t offset_fsb; |
837 | xfs_filblks_t count_fsb; | |
838 | xfs_filblks_t numblks_fsb; | |
dd9f438e NS |
839 | xfs_fsblock_t firstfsb; |
840 | int nimaps; | |
841 | xfs_trans_t *tp; | |
842 | xfs_bmbt_irec_t imap; | |
2c3234d1 | 843 | struct xfs_defer_ops dfops; |
84803fb7 | 844 | xfs_fsize_t i_size; |
dd9f438e | 845 | uint resblks; |
1da177e4 | 846 | int error; |
1da177e4 | 847 | |
0b1b213f | 848 | trace_xfs_unwritten_convert(ip, offset, count); |
1da177e4 LT |
849 | |
850 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
851 | count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
852 | count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb); | |
853 | ||
4ddd8bb1 LM |
854 | /* |
855 | * Reserve enough blocks in this transaction for two complete extent | |
856 | * btree splits. We may be converting the middle part of an unwritten | |
857 | * extent and in this case we will insert two new extents in the btree | |
858 | * each of which could cause a full split. | |
859 | * | |
860 | * This reservation amount will be used in the first call to | |
861 | * xfs_bmbt_split() to select an AG with enough space to satisfy the | |
862 | * rest of the operation. | |
863 | */ | |
dd9f438e | 864 | resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1; |
1da177e4 | 865 | |
dd9f438e | 866 | do { |
1da177e4 | 867 | /* |
253f4911 | 868 | * Set up a transaction to convert the range of extents |
1da177e4 LT |
869 | * from unwritten to real. Do allocations in a loop until |
870 | * we have covered the range passed in. | |
80641dc6 | 871 | * |
253f4911 CH |
872 | * Note that we can't risk to recursing back into the filesystem |
873 | * here as we might be asked to write out the same inode that we | |
874 | * complete here and might deadlock on the iolock. | |
1da177e4 | 875 | */ |
253f4911 CH |
876 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, |
877 | XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp); | |
878 | if (error) | |
b474c7ae | 879 | return error; |
1da177e4 LT |
880 | |
881 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
ddc3415a | 882 | xfs_trans_ijoin(tp, ip, 0); |
1da177e4 LT |
883 | |
884 | /* | |
885 | * Modify the unwritten extent state of the buffer. | |
886 | */ | |
2c3234d1 | 887 | xfs_defer_init(&dfops, &firstfsb); |
1da177e4 | 888 | nimaps = 1; |
c0dc7828 | 889 | error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, |
dbd5c8c9 | 890 | XFS_BMAPI_CONVERT, &firstfsb, resblks, |
2c3234d1 | 891 | &imap, &nimaps, &dfops); |
1da177e4 LT |
892 | if (error) |
893 | goto error_on_bmapi_transaction; | |
894 | ||
84803fb7 CH |
895 | /* |
896 | * Log the updated inode size as we go. We have to be careful | |
897 | * to only log it up to the actual write offset if it is | |
898 | * halfway into a block. | |
899 | */ | |
900 | i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb); | |
901 | if (i_size > offset + count) | |
902 | i_size = offset + count; | |
903 | ||
904 | i_size = xfs_new_eof(ip, i_size); | |
905 | if (i_size) { | |
906 | ip->i_d.di_size = i_size; | |
907 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
908 | } | |
909 | ||
2c3234d1 | 910 | error = xfs_defer_finish(&tp, &dfops, NULL); |
1da177e4 LT |
911 | if (error) |
912 | goto error_on_bmapi_transaction; | |
913 | ||
70393313 | 914 | error = xfs_trans_commit(tp); |
1da177e4 LT |
915 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
916 | if (error) | |
b474c7ae | 917 | return error; |
572d95f4 | 918 | |
86c4d623 | 919 | if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip))) |
6d4a8ecb | 920 | return xfs_alert_fsblock_zero(ip, &imap); |
1da177e4 LT |
921 | |
922 | if ((numblks_fsb = imap.br_blockcount) == 0) { | |
923 | /* | |
924 | * The numblks_fsb value should always get | |
925 | * smaller, otherwise the loop is stuck. | |
926 | */ | |
927 | ASSERT(imap.br_blockcount); | |
928 | break; | |
929 | } | |
930 | offset_fsb += numblks_fsb; | |
931 | count_fsb -= numblks_fsb; | |
932 | } while (count_fsb > 0); | |
933 | ||
934 | return 0; | |
935 | ||
936 | error_on_bmapi_transaction: | |
2c3234d1 | 937 | xfs_defer_cancel(&dfops); |
4906e215 | 938 | xfs_trans_cancel(tp); |
1da177e4 | 939 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
b474c7ae | 940 | return error; |
1da177e4 | 941 | } |
3b3dce05 | 942 | |
6c31f495 CH |
943 | static inline bool imap_needs_alloc(struct inode *inode, |
944 | struct xfs_bmbt_irec *imap, int nimaps) | |
68a9f5e7 CH |
945 | { |
946 | return !nimaps || | |
947 | imap->br_startblock == HOLESTARTBLOCK || | |
6c31f495 CH |
948 | imap->br_startblock == DELAYSTARTBLOCK || |
949 | (IS_DAX(inode) && ISUNWRITTEN(imap)); | |
68a9f5e7 CH |
950 | } |
951 | ||
952 | static int | |
953 | xfs_file_iomap_begin( | |
954 | struct inode *inode, | |
955 | loff_t offset, | |
956 | loff_t length, | |
957 | unsigned flags, | |
958 | struct iomap *iomap) | |
959 | { | |
960 | struct xfs_inode *ip = XFS_I(inode); | |
961 | struct xfs_mount *mp = ip->i_mount; | |
962 | struct xfs_bmbt_irec imap; | |
963 | xfs_fileoff_t offset_fsb, end_fsb; | |
db1327b1 | 964 | bool shared, trimmed; |
68a9f5e7 | 965 | int nimaps = 1, error = 0; |
66642c5c | 966 | unsigned lockmode; |
68a9f5e7 CH |
967 | |
968 | if (XFS_FORCED_SHUTDOWN(mp)) | |
969 | return -EIO; | |
970 | ||
2a06705c DW |
971 | if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && xfs_is_reflink_inode(ip)) { |
972 | error = xfs_reflink_reserve_cow_range(ip, offset, length); | |
973 | if (error < 0) | |
974 | return error; | |
975 | } | |
976 | ||
977 | if ((flags & IOMAP_WRITE) && !IS_DAX(inode) && | |
978 | !xfs_get_extsz_hint(ip)) { | |
979 | /* Reserve delalloc blocks for regular writeback. */ | |
51446f5b CH |
980 | return xfs_file_iomap_begin_delay(inode, offset, length, flags, |
981 | iomap); | |
982 | } | |
983 | ||
66642c5c | 984 | lockmode = xfs_ilock_data_map_shared(ip); |
68a9f5e7 CH |
985 | |
986 | ASSERT(offset <= mp->m_super->s_maxbytes); | |
987 | if ((xfs_fsize_t)offset + length > mp->m_super->s_maxbytes) | |
988 | length = mp->m_super->s_maxbytes - offset; | |
989 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
990 | end_fsb = XFS_B_TO_FSB(mp, offset + length); | |
991 | ||
992 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, | |
db1327b1 DW |
993 | &nimaps, 0); |
994 | if (error) { | |
995 | xfs_iunlock(ip, lockmode); | |
996 | return error; | |
997 | } | |
998 | ||
999 | /* Trim the mapping to the nearest shared extent boundary. */ | |
1000 | error = xfs_reflink_trim_around_shared(ip, &imap, &shared, &trimmed); | |
68a9f5e7 | 1001 | if (error) { |
66642c5c | 1002 | xfs_iunlock(ip, lockmode); |
68a9f5e7 CH |
1003 | return error; |
1004 | } | |
1005 | ||
6c31f495 | 1006 | if ((flags & IOMAP_WRITE) && imap_needs_alloc(inode, &imap, nimaps)) { |
68a9f5e7 CH |
1007 | /* |
1008 | * We cap the maximum length we map here to MAX_WRITEBACK_PAGES | |
1009 | * pages to keep the chunks of work done where somewhat symmetric | |
1010 | * with the work writeback does. This is a completely arbitrary | |
1011 | * number pulled out of thin air as a best guess for initial | |
1012 | * testing. | |
1013 | * | |
1014 | * Note that the values needs to be less than 32-bits wide until | |
1015 | * the lower level functions are updated. | |
1016 | */ | |
1017 | length = min_t(loff_t, length, 1024 * PAGE_SIZE); | |
51446f5b CH |
1018 | /* |
1019 | * xfs_iomap_write_direct() expects the shared lock. It | |
1020 | * is unlocked on return. | |
1021 | */ | |
66642c5c CH |
1022 | if (lockmode == XFS_ILOCK_EXCL) |
1023 | xfs_ilock_demote(ip, lockmode); | |
51446f5b CH |
1024 | error = xfs_iomap_write_direct(ip, offset, length, &imap, |
1025 | nimaps); | |
68a9f5e7 CH |
1026 | if (error) |
1027 | return error; | |
1028 | ||
ecd50729 | 1029 | iomap->flags = IOMAP_F_NEW; |
68a9f5e7 | 1030 | trace_xfs_iomap_alloc(ip, offset, length, 0, &imap); |
68a9f5e7 | 1031 | } else { |
b95a2127 CH |
1032 | ASSERT(nimaps); |
1033 | ||
66642c5c | 1034 | xfs_iunlock(ip, lockmode); |
b95a2127 | 1035 | trace_xfs_iomap_found(ip, offset, length, 0, &imap); |
68a9f5e7 CH |
1036 | } |
1037 | ||
b95a2127 | 1038 | xfs_bmbt_to_iomap(ip, iomap, &imap); |
db1327b1 DW |
1039 | if (shared) |
1040 | iomap->flags |= IOMAP_F_SHARED; | |
68a9f5e7 CH |
1041 | return 0; |
1042 | } | |
1043 | ||
1044 | static int | |
1045 | xfs_file_iomap_end_delalloc( | |
1046 | struct xfs_inode *ip, | |
1047 | loff_t offset, | |
1048 | loff_t length, | |
1049 | ssize_t written) | |
1050 | { | |
1051 | struct xfs_mount *mp = ip->i_mount; | |
1052 | xfs_fileoff_t start_fsb; | |
1053 | xfs_fileoff_t end_fsb; | |
1054 | int error = 0; | |
1055 | ||
1056 | start_fsb = XFS_B_TO_FSB(mp, offset + written); | |
1057 | end_fsb = XFS_B_TO_FSB(mp, offset + length); | |
1058 | ||
1059 | /* | |
1060 | * Trim back delalloc blocks if we didn't manage to write the whole | |
1061 | * range reserved. | |
1062 | * | |
1063 | * We don't need to care about racing delalloc as we hold i_mutex | |
1064 | * across the reserve/allocate/unreserve calls. If there are delalloc | |
1065 | * blocks in the range, they are ours. | |
1066 | */ | |
1067 | if (start_fsb < end_fsb) { | |
1068 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1069 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1070 | end_fsb - start_fsb); | |
1071 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1072 | ||
1073 | if (error && !XFS_FORCED_SHUTDOWN(mp)) { | |
1074 | xfs_alert(mp, "%s: unable to clean up ino %lld", | |
1075 | __func__, ip->i_ino); | |
1076 | return error; | |
1077 | } | |
1078 | } | |
1079 | ||
1080 | return 0; | |
1081 | } | |
1082 | ||
1083 | static int | |
1084 | xfs_file_iomap_end( | |
1085 | struct inode *inode, | |
1086 | loff_t offset, | |
1087 | loff_t length, | |
1088 | ssize_t written, | |
1089 | unsigned flags, | |
1090 | struct iomap *iomap) | |
1091 | { | |
1092 | if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC) | |
1093 | return xfs_file_iomap_end_delalloc(XFS_I(inode), offset, | |
1094 | length, written); | |
1095 | return 0; | |
1096 | } | |
1097 | ||
1098 | struct iomap_ops xfs_iomap_ops = { | |
1099 | .iomap_begin = xfs_file_iomap_begin, | |
1100 | .iomap_end = xfs_file_iomap_end, | |
1101 | }; | |
1d4795e7 CH |
1102 | |
1103 | static int | |
1104 | xfs_xattr_iomap_begin( | |
1105 | struct inode *inode, | |
1106 | loff_t offset, | |
1107 | loff_t length, | |
1108 | unsigned flags, | |
1109 | struct iomap *iomap) | |
1110 | { | |
1111 | struct xfs_inode *ip = XFS_I(inode); | |
1112 | struct xfs_mount *mp = ip->i_mount; | |
1113 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1114 | xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length); | |
1115 | struct xfs_bmbt_irec imap; | |
1116 | int nimaps = 1, error = 0; | |
1117 | unsigned lockmode; | |
1118 | ||
1119 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1120 | return -EIO; | |
1121 | ||
1122 | lockmode = xfs_ilock_data_map_shared(ip); | |
1123 | ||
1124 | /* if there are no attribute fork or extents, return ENOENT */ | |
1125 | if (XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) { | |
1126 | error = -ENOENT; | |
1127 | goto out_unlock; | |
1128 | } | |
1129 | ||
1130 | ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL); | |
1131 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, | |
1132 | &nimaps, XFS_BMAPI_ENTIRE | XFS_BMAPI_ATTRFORK); | |
1133 | out_unlock: | |
1134 | xfs_iunlock(ip, lockmode); | |
1135 | ||
1136 | if (!error) { | |
1137 | ASSERT(nimaps); | |
1138 | xfs_bmbt_to_iomap(ip, iomap, &imap); | |
1139 | } | |
1140 | ||
1141 | return error; | |
1142 | } | |
1143 | ||
1144 | struct iomap_ops xfs_xattr_iomap_ops = { | |
1145 | .iomap_begin = xfs_xattr_iomap_begin, | |
1146 | }; |