Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 | 3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
98c1a7c0 | 4 | * Copyright (c) 2016-2018 Christoph Hellwig. |
7b718769 | 5 | * All Rights Reserved. |
1da177e4 | 6 | */ |
1da177e4 | 7 | #include "xfs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
1da177e4 | 12 | #include "xfs_mount.h" |
1da177e4 | 13 | #include "xfs_inode.h" |
239880ef | 14 | #include "xfs_trans.h" |
1da177e4 | 15 | #include "xfs_iomap.h" |
0b1b213f | 16 | #include "xfs_trace.h" |
3ed3a434 | 17 | #include "xfs_bmap.h" |
68988114 | 18 | #include "xfs_bmap_util.h" |
ef473667 | 19 | #include "xfs_reflink.h" |
1da177e4 | 20 | |
fbcc0256 | 21 | struct xfs_writepage_ctx { |
598ecfba | 22 | struct iomap_writepage_ctx ctx; |
d9252d52 | 23 | unsigned int data_seq; |
e666aa37 | 24 | unsigned int cow_seq; |
fbcc0256 DC |
25 | }; |
26 | ||
598ecfba CH |
27 | static inline struct xfs_writepage_ctx * |
28 | XFS_WPC(struct iomap_writepage_ctx *ctx) | |
29 | { | |
30 | return container_of(ctx, struct xfs_writepage_ctx, ctx); | |
31 | } | |
32 | ||
fc0063c4 CH |
33 | /* |
34 | * Fast and loose check if this write could update the on-disk inode size. | |
35 | */ | |
598ecfba | 36 | static inline bool xfs_ioend_is_append(struct iomap_ioend *ioend) |
fc0063c4 CH |
37 | { |
38 | return ioend->io_offset + ioend->io_size > | |
13d2c10b | 39 | XFS_I(ioend->io_inode)->i_disk_size; |
fc0063c4 CH |
40 | } |
41 | ||
ba87ea69 | 42 | /* |
2813d682 | 43 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 44 | */ |
e7a3d7e7 BF |
45 | int |
46 | xfs_setfilesize( | |
2ba66237 | 47 | struct xfs_inode *ip, |
2ba66237 CH |
48 | xfs_off_t offset, |
49 | size_t size) | |
ba87ea69 | 50 | { |
e7a3d7e7 BF |
51 | struct xfs_mount *mp = ip->i_mount; |
52 | struct xfs_trans *tp; | |
ba87ea69 | 53 | xfs_fsize_t isize; |
e7a3d7e7 BF |
54 | int error; |
55 | ||
56 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
57 | if (error) | |
58 | return error; | |
ba87ea69 | 59 | |
aa6bf01d | 60 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 61 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
62 | if (!isize) { |
63 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 64 | xfs_trans_cancel(tp); |
281627df | 65 | return 0; |
ba87ea69 LM |
66 | } |
67 | ||
2ba66237 | 68 | trace_xfs_setfilesize(ip, offset, size); |
281627df | 69 | |
13d2c10b | 70 | ip->i_disk_size = isize; |
281627df CH |
71 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); |
72 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
73 | ||
70393313 | 74 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
75 | } |
76 | ||
0829c360 | 77 | /* |
5ec4fabb | 78 | * IO write completion. |
f6d6d4fc CH |
79 | */ |
80 | STATIC void | |
cb357bf3 | 81 | xfs_end_ioend( |
598ecfba | 82 | struct iomap_ioend *ioend) |
0829c360 | 83 | { |
0e51a8e1 | 84 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
5ca5916b | 85 | struct xfs_mount *mp = ip->i_mount; |
787eb485 CH |
86 | xfs_off_t offset = ioend->io_offset; |
87 | size_t size = ioend->io_size; | |
73d30d48 | 88 | unsigned int nofs_flag; |
4e4cbee9 | 89 | int error; |
ba87ea69 | 90 | |
73d30d48 CH |
91 | /* |
92 | * We can allocate memory here while doing writeback on behalf of | |
93 | * memory reclaim. To avoid memory allocation deadlocks set the | |
94 | * task-wide nofs context for the following operations. | |
95 | */ | |
96 | nofs_flag = memalloc_nofs_save(); | |
97 | ||
af055e37 | 98 | /* |
f9dd7ba4 | 99 | * Just clean up the in-memory structures if the fs has been shut down. |
af055e37 | 100 | */ |
5ca5916b | 101 | if (xfs_is_shutdown(mp)) { |
0e51a8e1 | 102 | error = -EIO; |
787eb485 CH |
103 | goto done; |
104 | } | |
04f658ee | 105 | |
43caeb18 | 106 | /* |
5ca5916b BF |
107 | * Clean up all COW blocks and underlying data fork delalloc blocks on |
108 | * I/O error. The delalloc punch is required because this ioend was | |
109 | * mapped to blocks in the COW fork and the associated pages are no | |
110 | * longer dirty. If we don't remove delalloc blocks here, they become | |
111 | * stale and can corrupt free space accounting on unmount. | |
43caeb18 | 112 | */ |
4e4cbee9 | 113 | error = blk_status_to_errno(ioend->io_bio->bi_status); |
787eb485 | 114 | if (unlikely(error)) { |
5ca5916b | 115 | if (ioend->io_flags & IOMAP_F_SHARED) { |
787eb485 | 116 | xfs_reflink_cancel_cow_range(ip, offset, size, true); |
5ca5916b BF |
117 | xfs_bmap_punch_delalloc_range(ip, |
118 | XFS_B_TO_FSBT(mp, offset), | |
119 | XFS_B_TO_FSB(mp, size)); | |
120 | } | |
787eb485 | 121 | goto done; |
43caeb18 DW |
122 | } |
123 | ||
5ec4fabb | 124 | /* |
be225fec | 125 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 126 | */ |
760fea8b | 127 | if (ioend->io_flags & IOMAP_F_SHARED) |
787eb485 | 128 | error = xfs_reflink_end_cow(ip, offset, size); |
4e087a3b | 129 | else if (ioend->io_type == IOMAP_UNWRITTEN) |
ee70daab | 130 | error = xfs_iomap_write_unwritten(ip, offset, size, false); |
ba87ea69 | 131 | |
7cd3099f BF |
132 | if (!error && xfs_ioend_is_append(ioend)) |
133 | error = xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); | |
04f658ee | 134 | done: |
598ecfba | 135 | iomap_finish_ioends(ioend, error); |
73d30d48 | 136 | memalloc_nofs_restore(nofs_flag); |
3994fc48 DW |
137 | } |
138 | ||
ebb7fb15 DC |
139 | /* |
140 | * Finish all pending IO completions that require transactional modifications. | |
141 | * | |
142 | * We try to merge physical and logically contiguous ioends before completion to | |
143 | * minimise the number of transactions we need to perform during IO completion. | |
144 | * Both unwritten extent conversion and COW remapping need to iterate and modify | |
145 | * one physical extent at a time, so we gain nothing by merging physically | |
146 | * discontiguous extents here. | |
147 | * | |
148 | * The ioend chain length that we can be processing here is largely unbound in | |
149 | * length and we may have to perform significant amounts of work on each ioend | |
150 | * to complete it. Hence we have to be careful about holding the CPU for too | |
151 | * long in this loop. | |
152 | */ | |
cb357bf3 DW |
153 | void |
154 | xfs_end_io( | |
155 | struct work_struct *work) | |
156 | { | |
433dad94 CH |
157 | struct xfs_inode *ip = |
158 | container_of(work, struct xfs_inode, i_ioend_work); | |
598ecfba | 159 | struct iomap_ioend *ioend; |
433dad94 | 160 | struct list_head tmp; |
cb357bf3 DW |
161 | unsigned long flags; |
162 | ||
cb357bf3 | 163 | spin_lock_irqsave(&ip->i_ioend_lock, flags); |
433dad94 | 164 | list_replace_init(&ip->i_ioend_list, &tmp); |
cb357bf3 DW |
165 | spin_unlock_irqrestore(&ip->i_ioend_lock, flags); |
166 | ||
598ecfba CH |
167 | iomap_sort_ioends(&tmp); |
168 | while ((ioend = list_first_entry_or_null(&tmp, struct iomap_ioend, | |
433dad94 | 169 | io_list))) { |
cb357bf3 | 170 | list_del_init(&ioend->io_list); |
6e552494 | 171 | iomap_ioend_try_merge(ioend, &tmp); |
cb357bf3 | 172 | xfs_end_ioend(ioend); |
ebb7fb15 | 173 | cond_resched(); |
cb357bf3 DW |
174 | } |
175 | } | |
176 | ||
0e51a8e1 CH |
177 | STATIC void |
178 | xfs_end_bio( | |
179 | struct bio *bio) | |
0829c360 | 180 | { |
598ecfba | 181 | struct iomap_ioend *ioend = bio->bi_private; |
cb357bf3 | 182 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
cb357bf3 | 183 | unsigned long flags; |
0829c360 | 184 | |
598ecfba CH |
185 | spin_lock_irqsave(&ip->i_ioend_lock, flags); |
186 | if (list_empty(&ip->i_ioend_list)) | |
187 | WARN_ON_ONCE(!queue_work(ip->i_mount->m_unwritten_workqueue, | |
188 | &ip->i_ioend_work)); | |
189 | list_add_tail(&ioend->io_list, &ip->i_ioend_list); | |
190 | spin_unlock_irqrestore(&ip->i_ioend_lock, flags); | |
0829c360 CH |
191 | } |
192 | ||
d9252d52 BF |
193 | /* |
194 | * Fast revalidation of the cached writeback mapping. Return true if the current | |
195 | * mapping is valid, false otherwise. | |
196 | */ | |
197 | static bool | |
198 | xfs_imap_valid( | |
598ecfba | 199 | struct iomap_writepage_ctx *wpc, |
d9252d52 | 200 | struct xfs_inode *ip, |
4e087a3b | 201 | loff_t offset) |
d9252d52 | 202 | { |
4e087a3b CH |
203 | if (offset < wpc->iomap.offset || |
204 | offset >= wpc->iomap.offset + wpc->iomap.length) | |
d9252d52 BF |
205 | return false; |
206 | /* | |
207 | * If this is a COW mapping, it is sufficient to check that the mapping | |
208 | * covers the offset. Be careful to check this first because the caller | |
209 | * can revalidate a COW mapping without updating the data seqno. | |
210 | */ | |
760fea8b | 211 | if (wpc->iomap.flags & IOMAP_F_SHARED) |
d9252d52 BF |
212 | return true; |
213 | ||
214 | /* | |
215 | * This is not a COW mapping. Check the sequence number of the data fork | |
216 | * because concurrent changes could have invalidated the extent. Check | |
217 | * the COW fork because concurrent changes since the last time we | |
218 | * checked (and found nothing at this offset) could have added | |
219 | * overlapping blocks. | |
220 | */ | |
598ecfba | 221 | if (XFS_WPC(wpc)->data_seq != READ_ONCE(ip->i_df.if_seq)) |
d9252d52 BF |
222 | return false; |
223 | if (xfs_inode_has_cow_data(ip) && | |
598ecfba | 224 | XFS_WPC(wpc)->cow_seq != READ_ONCE(ip->i_cowfp->if_seq)) |
d9252d52 BF |
225 | return false; |
226 | return true; | |
227 | } | |
228 | ||
4ad765ed CH |
229 | /* |
230 | * Pass in a dellalloc extent and convert it to real extents, return the real | |
4e087a3b | 231 | * extent that maps offset_fsb in wpc->iomap. |
4ad765ed CH |
232 | * |
233 | * The current page is held locked so nothing could have removed the block | |
7588cbee CH |
234 | * backing offset_fsb, although it could have moved from the COW to the data |
235 | * fork by another thread. | |
4ad765ed CH |
236 | */ |
237 | static int | |
238 | xfs_convert_blocks( | |
598ecfba | 239 | struct iomap_writepage_ctx *wpc, |
4ad765ed | 240 | struct xfs_inode *ip, |
760fea8b | 241 | int whichfork, |
4e087a3b | 242 | loff_t offset) |
4ad765ed CH |
243 | { |
244 | int error; | |
598ecfba CH |
245 | unsigned *seq; |
246 | ||
247 | if (whichfork == XFS_COW_FORK) | |
248 | seq = &XFS_WPC(wpc)->cow_seq; | |
249 | else | |
250 | seq = &XFS_WPC(wpc)->data_seq; | |
4ad765ed CH |
251 | |
252 | /* | |
4e087a3b CH |
253 | * Attempt to allocate whatever delalloc extent currently backs offset |
254 | * and put the result into wpc->iomap. Allocate in a loop because it | |
255 | * may take several attempts to allocate real blocks for a contiguous | |
256 | * delalloc extent if free space is sufficiently fragmented. | |
4ad765ed CH |
257 | */ |
258 | do { | |
760fea8b | 259 | error = xfs_bmapi_convert_delalloc(ip, whichfork, offset, |
598ecfba | 260 | &wpc->iomap, seq); |
4ad765ed CH |
261 | if (error) |
262 | return error; | |
4e087a3b | 263 | } while (wpc->iomap.offset + wpc->iomap.length <= offset); |
4ad765ed CH |
264 | |
265 | return 0; | |
266 | } | |
267 | ||
598ecfba | 268 | static int |
1da177e4 | 269 | xfs_map_blocks( |
598ecfba | 270 | struct iomap_writepage_ctx *wpc, |
1da177e4 | 271 | struct inode *inode, |
5c665e5b | 272 | loff_t offset) |
1da177e4 | 273 | { |
a206c817 CH |
274 | struct xfs_inode *ip = XFS_I(inode); |
275 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 276 | ssize_t count = i_blocksize(inode); |
b4e29032 CH |
277 | xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset); |
278 | xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count); | |
c2f09217 DW |
279 | xfs_fileoff_t cow_fsb; |
280 | int whichfork; | |
5c665e5b | 281 | struct xfs_bmbt_irec imap; |
060d4eaa | 282 | struct xfs_iext_cursor icur; |
7588cbee | 283 | int retries = 0; |
a206c817 | 284 | int error = 0; |
a206c817 | 285 | |
75c8c50f | 286 | if (xfs_is_shutdown(mp)) |
d9252d52 BF |
287 | return -EIO; |
288 | ||
889c65b3 CH |
289 | /* |
290 | * COW fork blocks can overlap data fork blocks even if the blocks | |
291 | * aren't shared. COW I/O always takes precedent, so we must always | |
292 | * check for overlap on reflink inodes unless the mapping is already a | |
e666aa37 CH |
293 | * COW one, or the COW fork hasn't changed from the last time we looked |
294 | * at it. | |
295 | * | |
296 | * It's safe to check the COW fork if_seq here without the ILOCK because | |
297 | * we've indirectly protected against concurrent updates: writeback has | |
298 | * the page locked, which prevents concurrent invalidations by reflink | |
299 | * and directio and prevents concurrent buffered writes to the same | |
300 | * page. Changes to if_seq always happen under i_lock, which protects | |
301 | * against concurrent updates and provides a memory barrier on the way | |
302 | * out that ensures that we always see the current value. | |
889c65b3 | 303 | */ |
4e087a3b | 304 | if (xfs_imap_valid(wpc, ip, offset)) |
889c65b3 CH |
305 | return 0; |
306 | ||
889c65b3 CH |
307 | /* |
308 | * If we don't have a valid map, now it's time to get a new one for this | |
309 | * offset. This will convert delayed allocations (including COW ones) | |
310 | * into real extents. If we return without a valid map, it means we | |
311 | * landed in a hole and we skip the block. | |
312 | */ | |
7588cbee | 313 | retry: |
c2f09217 DW |
314 | cow_fsb = NULLFILEOFF; |
315 | whichfork = XFS_DATA_FORK; | |
988ef927 | 316 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
b2197a36 | 317 | ASSERT(!xfs_need_iread_extents(&ip->i_df)); |
060d4eaa CH |
318 | |
319 | /* | |
320 | * Check if this is offset is covered by a COW extents, and if yes use | |
321 | * it directly instead of looking up anything in the data fork. | |
322 | */ | |
51d62690 | 323 | if (xfs_inode_has_cow_data(ip) && |
e666aa37 CH |
324 | xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap)) |
325 | cow_fsb = imap.br_startoff; | |
326 | if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) { | |
598ecfba | 327 | XFS_WPC(wpc)->cow_seq = READ_ONCE(ip->i_cowfp->if_seq); |
5c665e5b | 328 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
be225fec | 329 | |
760fea8b | 330 | whichfork = XFS_COW_FORK; |
5c665e5b CH |
331 | goto allocate_blocks; |
332 | } | |
333 | ||
334 | /* | |
d9252d52 BF |
335 | * No COW extent overlap. Revalidate now that we may have updated |
336 | * ->cow_seq. If the data mapping is still valid, we're done. | |
5c665e5b | 337 | */ |
4e087a3b | 338 | if (xfs_imap_valid(wpc, ip, offset)) { |
5c665e5b CH |
339 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
340 | return 0; | |
341 | } | |
342 | ||
343 | /* | |
344 | * If we don't have a valid map, now it's time to get a new one for this | |
345 | * offset. This will convert delayed allocations (including COW ones) | |
346 | * into real extents. | |
347 | */ | |
3345746e CH |
348 | if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) |
349 | imap.br_startoff = end_fsb; /* fake a hole past EOF */ | |
598ecfba | 350 | XFS_WPC(wpc)->data_seq = READ_ONCE(ip->i_df.if_seq); |
8ff2957d | 351 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 352 | |
12df89f2 | 353 | /* landed in a hole or beyond EOF? */ |
3345746e | 354 | if (imap.br_startoff > offset_fsb) { |
3345746e | 355 | imap.br_blockcount = imap.br_startoff - offset_fsb; |
5c665e5b | 356 | imap.br_startoff = offset_fsb; |
5c665e5b | 357 | imap.br_startblock = HOLESTARTBLOCK; |
be225fec | 358 | imap.br_state = XFS_EXT_NORM; |
8ff2957d | 359 | } |
e2f6ad46 | 360 | |
12df89f2 CH |
361 | /* |
362 | * Truncate to the next COW extent if there is one. This is the only | |
363 | * opportunity to do this because we can skip COW fork lookups for the | |
364 | * subsequent blocks in the mapping; however, the requirement to treat | |
365 | * the COW range separately remains. | |
366 | */ | |
367 | if (cow_fsb != NULLFILEOFF && | |
368 | cow_fsb < imap.br_startoff + imap.br_blockcount) | |
369 | imap.br_blockcount = cow_fsb - imap.br_startoff; | |
370 | ||
371 | /* got a delalloc extent? */ | |
372 | if (imap.br_startblock != HOLESTARTBLOCK && | |
373 | isnullstartblock(imap.br_startblock)) | |
374 | goto allocate_blocks; | |
375 | ||
740fd671 | 376 | xfs_bmbt_to_iomap(ip, &wpc->iomap, &imap, 0, 0); |
760fea8b | 377 | trace_xfs_map_blocks_found(ip, offset, count, whichfork, &imap); |
5c665e5b CH |
378 | return 0; |
379 | allocate_blocks: | |
760fea8b | 380 | error = xfs_convert_blocks(wpc, ip, whichfork, offset); |
7588cbee CH |
381 | if (error) { |
382 | /* | |
383 | * If we failed to find the extent in the COW fork we might have | |
384 | * raced with a COW to data fork conversion or truncate. | |
385 | * Restart the lookup to catch the extent in the data fork for | |
386 | * the former case, but prevent additional retries to avoid | |
387 | * looping forever for the latter case. | |
388 | */ | |
760fea8b | 389 | if (error == -EAGAIN && whichfork == XFS_COW_FORK && !retries++) |
7588cbee CH |
390 | goto retry; |
391 | ASSERT(error != -EAGAIN); | |
5c665e5b | 392 | return error; |
7588cbee | 393 | } |
4ad765ed CH |
394 | |
395 | /* | |
396 | * Due to merging the return real extent might be larger than the | |
397 | * original delalloc one. Trim the return extent to the next COW | |
398 | * boundary again to force a re-lookup. | |
399 | */ | |
760fea8b | 400 | if (whichfork != XFS_COW_FORK && cow_fsb != NULLFILEOFF) { |
4e087a3b CH |
401 | loff_t cow_offset = XFS_FSB_TO_B(mp, cow_fsb); |
402 | ||
403 | if (cow_offset < wpc->iomap.offset + wpc->iomap.length) | |
404 | wpc->iomap.length = cow_offset - wpc->iomap.offset; | |
405 | } | |
4ad765ed | 406 | |
4e087a3b CH |
407 | ASSERT(wpc->iomap.offset <= offset); |
408 | ASSERT(wpc->iomap.offset + wpc->iomap.length > offset); | |
760fea8b | 409 | trace_xfs_map_blocks_alloc(ip, offset, count, whichfork, &imap); |
8ff2957d | 410 | return 0; |
1da177e4 LT |
411 | } |
412 | ||
598ecfba CH |
413 | static int |
414 | xfs_prepare_ioend( | |
415 | struct iomap_ioend *ioend, | |
e10de372 | 416 | int status) |
f6d6d4fc | 417 | { |
73d30d48 CH |
418 | unsigned int nofs_flag; |
419 | ||
420 | /* | |
421 | * We can allocate memory here while doing writeback on behalf of | |
422 | * memory reclaim. To avoid memory allocation deadlocks set the | |
423 | * task-wide nofs context for the following operations. | |
424 | */ | |
425 | nofs_flag = memalloc_nofs_save(); | |
426 | ||
5eda4300 | 427 | /* Convert CoW extents to regular */ |
760fea8b | 428 | if (!status && (ioend->io_flags & IOMAP_F_SHARED)) { |
5eda4300 DW |
429 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), |
430 | ioend->io_offset, ioend->io_size); | |
431 | } | |
432 | ||
73d30d48 CH |
433 | memalloc_nofs_restore(nofs_flag); |
434 | ||
7adb8f14 BF |
435 | /* send ioends that might require a transaction to the completion wq */ |
436 | if (xfs_ioend_is_append(ioend) || ioend->io_type == IOMAP_UNWRITTEN || | |
437 | (ioend->io_flags & IOMAP_F_SHARED)) | |
598ecfba CH |
438 | ioend->io_bio->bi_end_io = xfs_end_bio; |
439 | return status; | |
f6d6d4fc CH |
440 | } |
441 | ||
3ed3a434 | 442 | /* |
82cb1417 CH |
443 | * If the page has delalloc blocks on it, we need to punch them out before we |
444 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
445 | * inode that can trip up a later direct I/O read operation on the same region. | |
3ed3a434 | 446 | * |
82cb1417 CH |
447 | * We prevent this by truncating away the delalloc regions on the page. Because |
448 | * they are delalloc, we can do this without needing a transaction. Indeed - if | |
449 | * we get ENOSPC errors, we have to be able to do this truncation without a | |
450 | * transaction as there is no space left for block reservation (typically why we | |
451 | * see a ENOSPC in writeback). | |
3ed3a434 | 452 | */ |
598ecfba | 453 | static void |
6e478521 MWO |
454 | xfs_discard_folio( |
455 | struct folio *folio, | |
456 | loff_t pos) | |
3ed3a434 | 457 | { |
6e478521 | 458 | struct inode *inode = folio->mapping->host; |
3ed3a434 | 459 | struct xfs_inode *ip = XFS_I(inode); |
03625721 | 460 | struct xfs_mount *mp = ip->i_mount; |
6e478521 MWO |
461 | size_t offset = offset_in_folio(folio, pos); |
462 | xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, pos); | |
463 | xfs_fileoff_t pageoff_fsb = XFS_B_TO_FSBT(mp, offset); | |
03625721 | 464 | int error; |
3ed3a434 | 465 | |
75c8c50f | 466 | if (xfs_is_shutdown(mp)) |
e8c3753c DC |
467 | goto out_invalidate; |
468 | ||
4ab45e25 | 469 | xfs_alert_ratelimited(mp, |
6e478521 MWO |
470 | "page discard on page "PTR_FMT", inode 0x%llx, pos %llu.", |
471 | folio, ip->i_ino, pos); | |
3ed3a434 | 472 | |
03625721 | 473 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, |
6e478521 | 474 | i_blocks_per_folio(inode, folio) - pageoff_fsb); |
75c8c50f | 475 | if (error && !xfs_is_shutdown(mp)) |
03625721 | 476 | xfs_alert(mp, "page discard unable to remove delalloc mapping."); |
3ed3a434 | 477 | out_invalidate: |
6e478521 | 478 | iomap_invalidate_folio(folio, offset, folio_size(folio) - offset); |
3ed3a434 DC |
479 | } |
480 | ||
598ecfba CH |
481 | static const struct iomap_writeback_ops xfs_writeback_ops = { |
482 | .map_blocks = xfs_map_blocks, | |
483 | .prepare_ioend = xfs_prepare_ioend, | |
6e478521 | 484 | .discard_folio = xfs_discard_folio, |
598ecfba | 485 | }; |
f51623b2 | 486 | |
7d4fb40a NS |
487 | STATIC int |
488 | xfs_vm_writepages( | |
489 | struct address_space *mapping, | |
490 | struct writeback_control *wbc) | |
491 | { | |
be225fec | 492 | struct xfs_writepage_ctx wpc = { }; |
fbcc0256 | 493 | |
756b1c34 DC |
494 | /* |
495 | * Writing back data in a transaction context can result in recursive | |
496 | * transactions. This is bad, so issue a warning and get out of here. | |
497 | */ | |
498 | if (WARN_ON_ONCE(current->journal_info)) | |
499 | return 0; | |
500 | ||
b3aea4ed | 501 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
598ecfba | 502 | return iomap_writepages(mapping, wbc, &wpc.ctx, &xfs_writeback_ops); |
7d4fb40a NS |
503 | } |
504 | ||
6e2608df DW |
505 | STATIC int |
506 | xfs_dax_writepages( | |
507 | struct address_space *mapping, | |
508 | struct writeback_control *wbc) | |
509 | { | |
30fa529e CH |
510 | struct xfs_inode *ip = XFS_I(mapping->host); |
511 | ||
512 | xfs_iflags_clear(ip, XFS_ITRUNCATED); | |
6e2608df | 513 | return dax_writeback_mapping_range(mapping, |
3f666c56 | 514 | xfs_inode_buftarg(ip)->bt_daxdev, wbc); |
6e2608df DW |
515 | } |
516 | ||
1da177e4 | 517 | STATIC sector_t |
e4c573bb | 518 | xfs_vm_bmap( |
1da177e4 LT |
519 | struct address_space *mapping, |
520 | sector_t block) | |
521 | { | |
b84e7722 | 522 | struct xfs_inode *ip = XFS_I(mapping->host); |
1da177e4 | 523 | |
b84e7722 | 524 | trace_xfs_vm_bmap(ip); |
db1327b1 DW |
525 | |
526 | /* | |
527 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
793057e1 | 528 | * bypasses the file system for actual I/O. We really can't allow |
db1327b1 | 529 | * that on reflinks inodes, so we have to skip out here. And yes, |
eb5e248d DW |
530 | * 0 is the magic code for a bmap error. |
531 | * | |
532 | * Since we don't pass back blockdev info, we can't return bmap | |
533 | * information for rt files either. | |
db1327b1 | 534 | */ |
66ae56a5 | 535 | if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip)) |
db1327b1 | 536 | return 0; |
690c2a38 | 537 | return iomap_bmap(mapping, block, &xfs_read_iomap_ops); |
1da177e4 LT |
538 | } |
539 | ||
540 | STATIC int | |
e4c573bb | 541 | xfs_vm_readpage( |
1da177e4 LT |
542 | struct file *unused, |
543 | struct page *page) | |
544 | { | |
690c2a38 | 545 | return iomap_readpage(page, &xfs_read_iomap_ops); |
1da177e4 LT |
546 | } |
547 | ||
9d24a13a MWO |
548 | STATIC void |
549 | xfs_vm_readahead( | |
550 | struct readahead_control *rac) | |
1da177e4 | 551 | { |
9d24a13a | 552 | iomap_readahead(rac, &xfs_read_iomap_ops); |
22e757a4 DC |
553 | } |
554 | ||
67482129 DW |
555 | static int |
556 | xfs_iomap_swapfile_activate( | |
557 | struct swap_info_struct *sis, | |
558 | struct file *swap_file, | |
559 | sector_t *span) | |
560 | { | |
30fa529e | 561 | sis->bdev = xfs_inode_buftarg(XFS_I(file_inode(swap_file)))->bt_bdev; |
690c2a38 CH |
562 | return iomap_swapfile_activate(sis, swap_file, span, |
563 | &xfs_read_iomap_ops); | |
67482129 DW |
564 | } |
565 | ||
f5e54d6e | 566 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb | 567 | .readpage = xfs_vm_readpage, |
9d24a13a | 568 | .readahead = xfs_vm_readahead, |
7d4fb40a | 569 | .writepages = xfs_vm_writepages, |
187c82cb | 570 | .dirty_folio = filemap_dirty_folio, |
9e91c572 | 571 | .releasepage = iomap_releasepage, |
d82354f6 | 572 | .invalidate_folio = iomap_invalidate_folio, |
e4c573bb | 573 | .bmap = xfs_vm_bmap, |
6e2608df | 574 | .direct_IO = noop_direct_IO, |
82cb1417 CH |
575 | .migratepage = iomap_migrate_page, |
576 | .is_partially_uptodate = iomap_is_partially_uptodate, | |
aa261f54 | 577 | .error_remove_page = generic_error_remove_page, |
67482129 | 578 | .swap_activate = xfs_iomap_swapfile_activate, |
1da177e4 | 579 | }; |
6e2608df DW |
580 | |
581 | const struct address_space_operations xfs_dax_aops = { | |
582 | .writepages = xfs_dax_writepages, | |
583 | .direct_IO = noop_direct_IO, | |
46de8b97 | 584 | .dirty_folio = noop_dirty_folio, |
67482129 | 585 | .swap_activate = xfs_iomap_swapfile_activate, |
6e2608df | 586 | }; |