Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_bmap_btree.h" | |
1da177e4 LT |
27 | #include "xfs_dinode.h" |
28 | #include "xfs_inode.h" | |
a844f451 | 29 | #include "xfs_alloc.h" |
1da177e4 LT |
30 | #include "xfs_error.h" |
31 | #include "xfs_rw.h" | |
32 | #include "xfs_iomap.h" | |
739bfb2a | 33 | #include "xfs_vnodeops.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
3ed3a434 | 35 | #include "xfs_bmap.h" |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
1da177e4 | 37 | #include <linux/mpage.h> |
10ce4444 | 38 | #include <linux/pagevec.h> |
1da177e4 LT |
39 | #include <linux/writeback.h> |
40 | ||
0b1b213f | 41 | void |
f51623b2 NS |
42 | xfs_count_page_state( |
43 | struct page *page, | |
44 | int *delalloc, | |
f51623b2 NS |
45 | int *unwritten) |
46 | { | |
47 | struct buffer_head *bh, *head; | |
48 | ||
20cb52eb | 49 | *delalloc = *unwritten = 0; |
f51623b2 NS |
50 | |
51 | bh = head = page_buffers(page); | |
52 | do { | |
20cb52eb | 53 | if (buffer_unwritten(bh)) |
f51623b2 NS |
54 | (*unwritten) = 1; |
55 | else if (buffer_delay(bh)) | |
56 | (*delalloc) = 1; | |
57 | } while ((bh = bh->b_this_page) != head); | |
58 | } | |
59 | ||
6214ed44 CH |
60 | STATIC struct block_device * |
61 | xfs_find_bdev_for_inode( | |
046f1685 | 62 | struct inode *inode) |
6214ed44 | 63 | { |
046f1685 | 64 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
65 | struct xfs_mount *mp = ip->i_mount; |
66 | ||
71ddabb9 | 67 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
68 | return mp->m_rtdev_targp->bt_bdev; |
69 | else | |
70 | return mp->m_ddev_targp->bt_bdev; | |
71 | } | |
72 | ||
f6d6d4fc CH |
73 | /* |
74 | * We're now finished for good with this ioend structure. | |
75 | * Update the page state via the associated buffer_heads, | |
76 | * release holds on the inode and bio, and finally free | |
77 | * up memory. Do not use the ioend after this. | |
78 | */ | |
0829c360 CH |
79 | STATIC void |
80 | xfs_destroy_ioend( | |
81 | xfs_ioend_t *ioend) | |
82 | { | |
f6d6d4fc CH |
83 | struct buffer_head *bh, *next; |
84 | ||
85 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
86 | next = bh->b_private; | |
7d04a335 | 87 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 88 | } |
583fa586 | 89 | |
c859cdd1 | 90 | if (ioend->io_iocb) { |
04f658ee CH |
91 | if (ioend->io_isasync) { |
92 | aio_complete(ioend->io_iocb, ioend->io_error ? | |
93 | ioend->io_error : ioend->io_result, 0); | |
94 | } | |
c859cdd1 CH |
95 | inode_dio_done(ioend->io_inode); |
96 | } | |
4a06fd26 | 97 | |
0829c360 CH |
98 | mempool_free(ioend, xfs_ioend_pool); |
99 | } | |
100 | ||
fc0063c4 CH |
101 | /* |
102 | * Fast and loose check if this write could update the on-disk inode size. | |
103 | */ | |
104 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
105 | { | |
106 | return ioend->io_offset + ioend->io_size > | |
107 | XFS_I(ioend->io_inode)->i_d.di_size; | |
108 | } | |
109 | ||
ba87ea69 | 110 | /* |
2813d682 | 111 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 112 | */ |
aa6bf01d | 113 | STATIC void |
ba87ea69 | 114 | xfs_setfilesize( |
aa6bf01d | 115 | struct xfs_ioend *ioend) |
ba87ea69 | 116 | { |
aa6bf01d | 117 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 118 | xfs_fsize_t isize; |
ba87ea69 | 119 | |
aa6bf01d | 120 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
6923e686 | 121 | isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size); |
932640e8 | 122 | if (isize) { |
55fb25d5 | 123 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
ba87ea69 | 124 | ip->i_d.di_size = isize; |
66d834ea | 125 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
126 | } |
127 | ||
128 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
129 | } |
130 | ||
131 | /* | |
209fb87a | 132 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
133 | * |
134 | * If there is no work to do we might as well call it a day and free the | |
135 | * ioend right now. | |
77d7a0c2 DC |
136 | */ |
137 | STATIC void | |
138 | xfs_finish_ioend( | |
209fb87a | 139 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
140 | { |
141 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
142 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
143 | ||
209fb87a | 144 | if (ioend->io_type == IO_UNWRITTEN) |
aa6bf01d | 145 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
fc0063c4 | 146 | else if (xfs_ioend_is_append(ioend)) |
aa6bf01d | 147 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
148 | else |
149 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 150 | } |
ba87ea69 LM |
151 | } |
152 | ||
0829c360 | 153 | /* |
5ec4fabb | 154 | * IO write completion. |
f6d6d4fc CH |
155 | */ |
156 | STATIC void | |
5ec4fabb | 157 | xfs_end_io( |
77d7a0c2 | 158 | struct work_struct *work) |
0829c360 | 159 | { |
77d7a0c2 DC |
160 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
161 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 162 | int error = 0; |
ba87ea69 | 163 | |
04f658ee | 164 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 165 | ioend->io_error = -EIO; |
04f658ee CH |
166 | goto done; |
167 | } | |
168 | if (ioend->io_error) | |
169 | goto done; | |
170 | ||
5ec4fabb CH |
171 | /* |
172 | * For unwritten extents we need to issue transactions to convert a | |
173 | * range to normal written extens after the data I/O has finished. | |
174 | */ | |
04f658ee | 175 | if (ioend->io_type == IO_UNWRITTEN) { |
5ec4fabb CH |
176 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
177 | ioend->io_size); | |
04f658ee CH |
178 | if (error) { |
179 | ioend->io_error = -error; | |
180 | goto done; | |
181 | } | |
84803fb7 CH |
182 | } else { |
183 | /* | |
184 | * We might have to update the on-disk file size after | |
185 | * extending writes. | |
186 | */ | |
187 | xfs_setfilesize(ioend); | |
5ec4fabb | 188 | } |
ba87ea69 | 189 | |
04f658ee | 190 | done: |
aa6bf01d | 191 | xfs_destroy_ioend(ioend); |
c626d174 DC |
192 | } |
193 | ||
209fb87a CH |
194 | /* |
195 | * Call IO completion handling in caller context on the final put of an ioend. | |
196 | */ | |
197 | STATIC void | |
198 | xfs_finish_ioend_sync( | |
199 | struct xfs_ioend *ioend) | |
200 | { | |
201 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
202 | xfs_end_io(&ioend->io_work); | |
203 | } | |
204 | ||
0829c360 CH |
205 | /* |
206 | * Allocate and initialise an IO completion structure. | |
207 | * We need to track unwritten extent write completion here initially. | |
208 | * We'll need to extend this for updating the ondisk inode size later | |
209 | * (vs. incore size). | |
210 | */ | |
211 | STATIC xfs_ioend_t * | |
212 | xfs_alloc_ioend( | |
f6d6d4fc CH |
213 | struct inode *inode, |
214 | unsigned int type) | |
0829c360 CH |
215 | { |
216 | xfs_ioend_t *ioend; | |
217 | ||
218 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
219 | ||
220 | /* | |
221 | * Set the count to 1 initially, which will prevent an I/O | |
222 | * completion callback from happening before we have started | |
223 | * all the I/O from calling the completion routine too early. | |
224 | */ | |
225 | atomic_set(&ioend->io_remaining, 1); | |
c859cdd1 | 226 | ioend->io_isasync = 0; |
7d04a335 | 227 | ioend->io_error = 0; |
f6d6d4fc CH |
228 | ioend->io_list = NULL; |
229 | ioend->io_type = type; | |
b677c210 | 230 | ioend->io_inode = inode; |
c1a073bd | 231 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 232 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
233 | ioend->io_offset = 0; |
234 | ioend->io_size = 0; | |
fb511f21 CH |
235 | ioend->io_iocb = NULL; |
236 | ioend->io_result = 0; | |
0829c360 | 237 | |
5ec4fabb | 238 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
239 | return ioend; |
240 | } | |
241 | ||
1da177e4 LT |
242 | STATIC int |
243 | xfs_map_blocks( | |
244 | struct inode *inode, | |
245 | loff_t offset, | |
207d0416 | 246 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
247 | int type, |
248 | int nonblocking) | |
1da177e4 | 249 | { |
a206c817 CH |
250 | struct xfs_inode *ip = XFS_I(inode); |
251 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 252 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
253 | xfs_fileoff_t offset_fsb, end_fsb; |
254 | int error = 0; | |
a206c817 CH |
255 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
256 | int nimaps = 1; | |
257 | ||
258 | if (XFS_FORCED_SHUTDOWN(mp)) | |
259 | return -XFS_ERROR(EIO); | |
260 | ||
8ff2957d | 261 | if (type == IO_UNWRITTEN) |
a206c817 | 262 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
263 | |
264 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
265 | if (nonblocking) | |
266 | return -XFS_ERROR(EAGAIN); | |
267 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
268 | } |
269 | ||
8ff2957d CH |
270 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
271 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
a206c817 | 272 | ASSERT(offset <= mp->m_maxioffset); |
8ff2957d | 273 | |
a206c817 CH |
274 | if (offset + count > mp->m_maxioffset) |
275 | count = mp->m_maxioffset - offset; | |
276 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
277 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
278 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
279 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 280 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 281 | |
8ff2957d CH |
282 | if (error) |
283 | return -XFS_ERROR(error); | |
a206c817 | 284 | |
8ff2957d CH |
285 | if (type == IO_DELALLOC && |
286 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
a206c817 CH |
287 | error = xfs_iomap_write_allocate(ip, offset, count, imap); |
288 | if (!error) | |
289 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 290 | return -XFS_ERROR(error); |
a206c817 CH |
291 | } |
292 | ||
8ff2957d CH |
293 | #ifdef DEBUG |
294 | if (type == IO_UNWRITTEN) { | |
295 | ASSERT(nimaps); | |
296 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
297 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
298 | } | |
299 | #endif | |
300 | if (nimaps) | |
301 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
302 | return 0; | |
1da177e4 LT |
303 | } |
304 | ||
b8f82a4a | 305 | STATIC int |
558e6891 | 306 | xfs_imap_valid( |
8699bb0a | 307 | struct inode *inode, |
207d0416 | 308 | struct xfs_bmbt_irec *imap, |
558e6891 | 309 | xfs_off_t offset) |
1da177e4 | 310 | { |
558e6891 | 311 | offset >>= inode->i_blkbits; |
8699bb0a | 312 | |
558e6891 CH |
313 | return offset >= imap->br_startoff && |
314 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
315 | } |
316 | ||
f6d6d4fc CH |
317 | /* |
318 | * BIO completion handler for buffered IO. | |
319 | */ | |
782e3b3b | 320 | STATIC void |
f6d6d4fc CH |
321 | xfs_end_bio( |
322 | struct bio *bio, | |
f6d6d4fc CH |
323 | int error) |
324 | { | |
325 | xfs_ioend_t *ioend = bio->bi_private; | |
326 | ||
f6d6d4fc | 327 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 328 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
329 | |
330 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
331 | bio->bi_private = NULL; |
332 | bio->bi_end_io = NULL; | |
f6d6d4fc | 333 | bio_put(bio); |
7d04a335 | 334 | |
209fb87a | 335 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
336 | } |
337 | ||
338 | STATIC void | |
339 | xfs_submit_ioend_bio( | |
06342cf8 CH |
340 | struct writeback_control *wbc, |
341 | xfs_ioend_t *ioend, | |
342 | struct bio *bio) | |
f6d6d4fc | 343 | { |
6923e686 | 344 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc | 345 | atomic_inc(&ioend->io_remaining); |
f6d6d4fc CH |
346 | bio->bi_private = ioend; |
347 | bio->bi_end_io = xfs_end_bio; | |
348 | ||
932640e8 DC |
349 | /* |
350 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
351 | * but don't update the inode size until I/O completion. | |
352 | */ | |
6923e686 CH |
353 | if (xfs_new_eof(ip, ioend->io_offset + ioend->io_size)) |
354 | xfs_mark_inode_dirty(ip); | |
932640e8 | 355 | |
721a9602 | 356 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
357 | } |
358 | ||
359 | STATIC struct bio * | |
360 | xfs_alloc_ioend_bio( | |
361 | struct buffer_head *bh) | |
362 | { | |
f6d6d4fc | 363 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 364 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
365 | |
366 | ASSERT(bio->bi_private == NULL); | |
367 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
368 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
369 | return bio; |
370 | } | |
371 | ||
372 | STATIC void | |
373 | xfs_start_buffer_writeback( | |
374 | struct buffer_head *bh) | |
375 | { | |
376 | ASSERT(buffer_mapped(bh)); | |
377 | ASSERT(buffer_locked(bh)); | |
378 | ASSERT(!buffer_delay(bh)); | |
379 | ASSERT(!buffer_unwritten(bh)); | |
380 | ||
381 | mark_buffer_async_write(bh); | |
382 | set_buffer_uptodate(bh); | |
383 | clear_buffer_dirty(bh); | |
384 | } | |
385 | ||
386 | STATIC void | |
387 | xfs_start_page_writeback( | |
388 | struct page *page, | |
f6d6d4fc CH |
389 | int clear_dirty, |
390 | int buffers) | |
391 | { | |
392 | ASSERT(PageLocked(page)); | |
393 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 394 | if (clear_dirty) |
92132021 DC |
395 | clear_page_dirty_for_io(page); |
396 | set_page_writeback(page); | |
f6d6d4fc | 397 | unlock_page(page); |
1f7decf6 FW |
398 | /* If no buffers on the page are to be written, finish it here */ |
399 | if (!buffers) | |
f6d6d4fc | 400 | end_page_writeback(page); |
f6d6d4fc CH |
401 | } |
402 | ||
403 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
404 | { | |
405 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
406 | } | |
407 | ||
408 | /* | |
d88992f6 DC |
409 | * Submit all of the bios for all of the ioends we have saved up, covering the |
410 | * initial writepage page and also any probed pages. | |
411 | * | |
412 | * Because we may have multiple ioends spanning a page, we need to start | |
413 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
414 | * buffers as we got, then we can end up with a page that only has buffers | |
415 | * marked async write and I/O complete on can occur before we mark the other | |
416 | * buffers async write. | |
417 | * | |
418 | * The end result of this is that we trip a bug in end_page_writeback() because | |
419 | * we call it twice for the one page as the code in end_buffer_async_write() | |
420 | * assumes that all buffers on the page are started at the same time. | |
421 | * | |
422 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 423 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
424 | */ |
425 | STATIC void | |
426 | xfs_submit_ioend( | |
06342cf8 | 427 | struct writeback_control *wbc, |
f6d6d4fc CH |
428 | xfs_ioend_t *ioend) |
429 | { | |
d88992f6 | 430 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
431 | xfs_ioend_t *next; |
432 | struct buffer_head *bh; | |
433 | struct bio *bio; | |
434 | sector_t lastblock = 0; | |
435 | ||
d88992f6 DC |
436 | /* Pass 1 - start writeback */ |
437 | do { | |
438 | next = ioend->io_list; | |
221cb251 | 439 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 440 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
441 | } while ((ioend = next) != NULL); |
442 | ||
443 | /* Pass 2 - submit I/O */ | |
444 | ioend = head; | |
f6d6d4fc CH |
445 | do { |
446 | next = ioend->io_list; | |
447 | bio = NULL; | |
448 | ||
449 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
450 | |
451 | if (!bio) { | |
452 | retry: | |
453 | bio = xfs_alloc_ioend_bio(bh); | |
454 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 455 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
456 | goto retry; |
457 | } | |
458 | ||
459 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 460 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
461 | goto retry; |
462 | } | |
463 | ||
464 | lastblock = bh->b_blocknr; | |
465 | } | |
466 | if (bio) | |
06342cf8 | 467 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 468 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
469 | } while ((ioend = next) != NULL); |
470 | } | |
471 | ||
472 | /* | |
473 | * Cancel submission of all buffer_heads so far in this endio. | |
474 | * Toss the endio too. Only ever called for the initial page | |
475 | * in a writepage request, so only ever one page. | |
476 | */ | |
477 | STATIC void | |
478 | xfs_cancel_ioend( | |
479 | xfs_ioend_t *ioend) | |
480 | { | |
481 | xfs_ioend_t *next; | |
482 | struct buffer_head *bh, *next_bh; | |
483 | ||
484 | do { | |
485 | next = ioend->io_list; | |
486 | bh = ioend->io_buffer_head; | |
487 | do { | |
488 | next_bh = bh->b_private; | |
489 | clear_buffer_async_write(bh); | |
490 | unlock_buffer(bh); | |
491 | } while ((bh = next_bh) != NULL); | |
492 | ||
f6d6d4fc CH |
493 | mempool_free(ioend, xfs_ioend_pool); |
494 | } while ((ioend = next) != NULL); | |
495 | } | |
496 | ||
497 | /* | |
498 | * Test to see if we've been building up a completion structure for | |
499 | * earlier buffers -- if so, we try to append to this ioend if we | |
500 | * can, otherwise we finish off any current ioend and start another. | |
501 | * Return true if we've finished the given ioend. | |
502 | */ | |
503 | STATIC void | |
504 | xfs_add_to_ioend( | |
505 | struct inode *inode, | |
506 | struct buffer_head *bh, | |
7336cea8 | 507 | xfs_off_t offset, |
f6d6d4fc CH |
508 | unsigned int type, |
509 | xfs_ioend_t **result, | |
510 | int need_ioend) | |
511 | { | |
512 | xfs_ioend_t *ioend = *result; | |
513 | ||
514 | if (!ioend || need_ioend || type != ioend->io_type) { | |
515 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 516 | |
f6d6d4fc CH |
517 | ioend = xfs_alloc_ioend(inode, type); |
518 | ioend->io_offset = offset; | |
519 | ioend->io_buffer_head = bh; | |
520 | ioend->io_buffer_tail = bh; | |
521 | if (previous) | |
522 | previous->io_list = ioend; | |
523 | *result = ioend; | |
524 | } else { | |
525 | ioend->io_buffer_tail->b_private = bh; | |
526 | ioend->io_buffer_tail = bh; | |
527 | } | |
528 | ||
529 | bh->b_private = NULL; | |
530 | ioend->io_size += bh->b_size; | |
531 | } | |
532 | ||
87cbc49c NS |
533 | STATIC void |
534 | xfs_map_buffer( | |
046f1685 | 535 | struct inode *inode, |
87cbc49c | 536 | struct buffer_head *bh, |
207d0416 | 537 | struct xfs_bmbt_irec *imap, |
046f1685 | 538 | xfs_off_t offset) |
87cbc49c NS |
539 | { |
540 | sector_t bn; | |
8699bb0a | 541 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
542 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
543 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 544 | |
207d0416 CH |
545 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
546 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 547 | |
e513182d | 548 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 549 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 550 | |
046f1685 | 551 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
552 | |
553 | bh->b_blocknr = bn; | |
554 | set_buffer_mapped(bh); | |
555 | } | |
556 | ||
1da177e4 LT |
557 | STATIC void |
558 | xfs_map_at_offset( | |
046f1685 | 559 | struct inode *inode, |
1da177e4 | 560 | struct buffer_head *bh, |
207d0416 | 561 | struct xfs_bmbt_irec *imap, |
046f1685 | 562 | xfs_off_t offset) |
1da177e4 | 563 | { |
207d0416 CH |
564 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
565 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 566 | |
207d0416 | 567 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
568 | set_buffer_mapped(bh); |
569 | clear_buffer_delay(bh); | |
f6d6d4fc | 570 | clear_buffer_unwritten(bh); |
1da177e4 LT |
571 | } |
572 | ||
1da177e4 | 573 | /* |
10ce4444 CH |
574 | * Test if a given page is suitable for writing as part of an unwritten |
575 | * or delayed allocate extent. | |
1da177e4 | 576 | */ |
10ce4444 CH |
577 | STATIC int |
578 | xfs_is_delayed_page( | |
579 | struct page *page, | |
f6d6d4fc | 580 | unsigned int type) |
1da177e4 | 581 | { |
1da177e4 | 582 | if (PageWriteback(page)) |
10ce4444 | 583 | return 0; |
1da177e4 LT |
584 | |
585 | if (page->mapping && page_has_buffers(page)) { | |
586 | struct buffer_head *bh, *head; | |
587 | int acceptable = 0; | |
588 | ||
589 | bh = head = page_buffers(page); | |
590 | do { | |
f6d6d4fc | 591 | if (buffer_unwritten(bh)) |
34a52c6c | 592 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 593 | else if (buffer_delay(bh)) |
a206c817 | 594 | acceptable = (type == IO_DELALLOC); |
2ddee844 | 595 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
a206c817 | 596 | acceptable = (type == IO_OVERWRITE); |
f6d6d4fc | 597 | else |
1da177e4 | 598 | break; |
1da177e4 LT |
599 | } while ((bh = bh->b_this_page) != head); |
600 | ||
601 | if (acceptable) | |
10ce4444 | 602 | return 1; |
1da177e4 LT |
603 | } |
604 | ||
10ce4444 | 605 | return 0; |
1da177e4 LT |
606 | } |
607 | ||
1da177e4 LT |
608 | /* |
609 | * Allocate & map buffers for page given the extent map. Write it out. | |
610 | * except for the original page of a writepage, this is called on | |
611 | * delalloc/unwritten pages only, for the original page it is possible | |
612 | * that the page has no mapping at all. | |
613 | */ | |
f6d6d4fc | 614 | STATIC int |
1da177e4 LT |
615 | xfs_convert_page( |
616 | struct inode *inode, | |
617 | struct page *page, | |
10ce4444 | 618 | loff_t tindex, |
207d0416 | 619 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 620 | xfs_ioend_t **ioendp, |
2fa24f92 | 621 | struct writeback_control *wbc) |
1da177e4 | 622 | { |
f6d6d4fc | 623 | struct buffer_head *bh, *head; |
9260dc6b CH |
624 | xfs_off_t end_offset; |
625 | unsigned long p_offset; | |
f6d6d4fc | 626 | unsigned int type; |
24e17b5f | 627 | int len, page_dirty; |
f6d6d4fc | 628 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 629 | xfs_off_t offset = page_offset(page); |
1da177e4 | 630 | |
10ce4444 CH |
631 | if (page->index != tindex) |
632 | goto fail; | |
529ae9aa | 633 | if (!trylock_page(page)) |
10ce4444 CH |
634 | goto fail; |
635 | if (PageWriteback(page)) | |
636 | goto fail_unlock_page; | |
637 | if (page->mapping != inode->i_mapping) | |
638 | goto fail_unlock_page; | |
639 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
640 | goto fail_unlock_page; | |
641 | ||
24e17b5f NS |
642 | /* |
643 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 644 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
645 | * |
646 | * Derivation: | |
647 | * | |
648 | * End offset is the highest offset that this page should represent. | |
649 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
650 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
651 | * hence give us the correct page_dirty count. On any other page, | |
652 | * it will be zero and in that case we need page_dirty to be the | |
653 | * count of buffers on the page. | |
24e17b5f | 654 | */ |
9260dc6b CH |
655 | end_offset = min_t(unsigned long long, |
656 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
657 | i_size_read(inode)); | |
658 | ||
24e17b5f | 659 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
660 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
661 | PAGE_CACHE_SIZE); | |
662 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
663 | page_dirty = p_offset / len; | |
24e17b5f | 664 | |
1da177e4 LT |
665 | bh = head = page_buffers(page); |
666 | do { | |
9260dc6b | 667 | if (offset >= end_offset) |
1da177e4 | 668 | break; |
f6d6d4fc CH |
669 | if (!buffer_uptodate(bh)) |
670 | uptodate = 0; | |
671 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
672 | done = 1; | |
1da177e4 | 673 | continue; |
f6d6d4fc CH |
674 | } |
675 | ||
2fa24f92 CH |
676 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
677 | buffer_mapped(bh)) { | |
9260dc6b | 678 | if (buffer_unwritten(bh)) |
34a52c6c | 679 | type = IO_UNWRITTEN; |
2fa24f92 | 680 | else if (buffer_delay(bh)) |
a206c817 | 681 | type = IO_DELALLOC; |
2fa24f92 CH |
682 | else |
683 | type = IO_OVERWRITE; | |
9260dc6b | 684 | |
558e6891 | 685 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 686 | done = 1; |
9260dc6b CH |
687 | continue; |
688 | } | |
689 | ||
ecff71e6 CH |
690 | lock_buffer(bh); |
691 | if (type != IO_OVERWRITE) | |
2fa24f92 | 692 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
693 | xfs_add_to_ioend(inode, bh, offset, type, |
694 | ioendp, done); | |
695 | ||
9260dc6b CH |
696 | page_dirty--; |
697 | count++; | |
698 | } else { | |
2fa24f92 | 699 | done = 1; |
1da177e4 | 700 | } |
7336cea8 | 701 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 702 | |
f6d6d4fc CH |
703 | if (uptodate && bh == head) |
704 | SetPageUptodate(page); | |
705 | ||
89f3b363 | 706 | if (count) { |
efceab1d DC |
707 | if (--wbc->nr_to_write <= 0 && |
708 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 709 | done = 1; |
1da177e4 | 710 | } |
89f3b363 | 711 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
712 | |
713 | return done; | |
10ce4444 CH |
714 | fail_unlock_page: |
715 | unlock_page(page); | |
716 | fail: | |
717 | return 1; | |
1da177e4 LT |
718 | } |
719 | ||
720 | /* | |
721 | * Convert & write out a cluster of pages in the same extent as defined | |
722 | * by mp and following the start page. | |
723 | */ | |
724 | STATIC void | |
725 | xfs_cluster_write( | |
726 | struct inode *inode, | |
727 | pgoff_t tindex, | |
207d0416 | 728 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 729 | xfs_ioend_t **ioendp, |
1da177e4 | 730 | struct writeback_control *wbc, |
1da177e4 LT |
731 | pgoff_t tlast) |
732 | { | |
10ce4444 CH |
733 | struct pagevec pvec; |
734 | int done = 0, i; | |
1da177e4 | 735 | |
10ce4444 CH |
736 | pagevec_init(&pvec, 0); |
737 | while (!done && tindex <= tlast) { | |
738 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
739 | ||
740 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 741 | break; |
10ce4444 CH |
742 | |
743 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
744 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 745 | imap, ioendp, wbc); |
10ce4444 CH |
746 | if (done) |
747 | break; | |
748 | } | |
749 | ||
750 | pagevec_release(&pvec); | |
751 | cond_resched(); | |
1da177e4 LT |
752 | } |
753 | } | |
754 | ||
3ed3a434 DC |
755 | STATIC void |
756 | xfs_vm_invalidatepage( | |
757 | struct page *page, | |
758 | unsigned long offset) | |
759 | { | |
760 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
761 | block_invalidatepage(page, offset); | |
762 | } | |
763 | ||
764 | /* | |
765 | * If the page has delalloc buffers on it, we need to punch them out before we | |
766 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
767 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
768 | * is done on that same region - the delalloc extent is returned when none is | |
769 | * supposed to be there. | |
770 | * | |
771 | * We prevent this by truncating away the delalloc regions on the page before | |
772 | * invalidating it. Because they are delalloc, we can do this without needing a | |
773 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
774 | * truncation without a transaction as there is no space left for block | |
775 | * reservation (typically why we see a ENOSPC in writeback). | |
776 | * | |
777 | * This is not a performance critical path, so for now just do the punching a | |
778 | * buffer head at a time. | |
779 | */ | |
780 | STATIC void | |
781 | xfs_aops_discard_page( | |
782 | struct page *page) | |
783 | { | |
784 | struct inode *inode = page->mapping->host; | |
785 | struct xfs_inode *ip = XFS_I(inode); | |
786 | struct buffer_head *bh, *head; | |
787 | loff_t offset = page_offset(page); | |
3ed3a434 | 788 | |
a206c817 | 789 | if (!xfs_is_delayed_page(page, IO_DELALLOC)) |
3ed3a434 DC |
790 | goto out_invalidate; |
791 | ||
e8c3753c DC |
792 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
793 | goto out_invalidate; | |
794 | ||
4f10700a | 795 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
796 | "page discard on page %p, inode 0x%llx, offset %llu.", |
797 | page, ip->i_ino, offset); | |
798 | ||
799 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
800 | bh = head = page_buffers(page); | |
801 | do { | |
3ed3a434 | 802 | int error; |
c726de44 | 803 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
804 | |
805 | if (!buffer_delay(bh)) | |
806 | goto next_buffer; | |
807 | ||
c726de44 DC |
808 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
809 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
810 | if (error) { |
811 | /* something screwed, just bail */ | |
e8c3753c | 812 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 813 | xfs_alert(ip->i_mount, |
3ed3a434 | 814 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 815 | } |
3ed3a434 DC |
816 | break; |
817 | } | |
818 | next_buffer: | |
c726de44 | 819 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
820 | |
821 | } while ((bh = bh->b_this_page) != head); | |
822 | ||
823 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
824 | out_invalidate: | |
825 | xfs_vm_invalidatepage(page, 0); | |
826 | return; | |
827 | } | |
828 | ||
1da177e4 | 829 | /* |
89f3b363 CH |
830 | * Write out a dirty page. |
831 | * | |
832 | * For delalloc space on the page we need to allocate space and flush it. | |
833 | * For unwritten space on the page we need to start the conversion to | |
834 | * regular allocated space. | |
89f3b363 | 835 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 836 | */ |
1da177e4 | 837 | STATIC int |
89f3b363 CH |
838 | xfs_vm_writepage( |
839 | struct page *page, | |
840 | struct writeback_control *wbc) | |
1da177e4 | 841 | { |
89f3b363 | 842 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 843 | struct buffer_head *bh, *head; |
207d0416 | 844 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 845 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 846 | loff_t offset; |
f6d6d4fc | 847 | unsigned int type; |
1da177e4 | 848 | __uint64_t end_offset; |
bd1556a1 | 849 | pgoff_t end_index, last_index; |
ed1e7b7e | 850 | ssize_t len; |
a206c817 | 851 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 852 | int count = 0; |
a206c817 | 853 | int nonblocking = 0; |
89f3b363 CH |
854 | |
855 | trace_xfs_writepage(inode, page, 0); | |
856 | ||
20cb52eb CH |
857 | ASSERT(page_has_buffers(page)); |
858 | ||
89f3b363 CH |
859 | /* |
860 | * Refuse to write the page out if we are called from reclaim context. | |
861 | * | |
d4f7a5cb CH |
862 | * This avoids stack overflows when called from deeply used stacks in |
863 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
864 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 865 | * |
94054fa3 MG |
866 | * This should never happen except in the case of a VM regression so |
867 | * warn about it. | |
89f3b363 | 868 | */ |
94054fa3 MG |
869 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
870 | PF_MEMALLOC)) | |
b5420f23 | 871 | goto redirty; |
1da177e4 | 872 | |
89f3b363 | 873 | /* |
680a647b CH |
874 | * Given that we do not allow direct reclaim to call us, we should |
875 | * never be called while in a filesystem transaction. | |
89f3b363 | 876 | */ |
680a647b | 877 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 878 | goto redirty; |
89f3b363 | 879 | |
1da177e4 LT |
880 | /* Is this page beyond the end of the file? */ |
881 | offset = i_size_read(inode); | |
882 | end_index = offset >> PAGE_CACHE_SHIFT; | |
883 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
884 | if (page->index >= end_index) { | |
885 | if ((page->index >= end_index + 1) || | |
886 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 887 | unlock_page(page); |
19d5bcf3 | 888 | return 0; |
1da177e4 LT |
889 | } |
890 | } | |
891 | ||
f6d6d4fc | 892 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
893 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
894 | offset); | |
24e17b5f | 895 | len = 1 << inode->i_blkbits; |
24e17b5f | 896 | |
24e17b5f | 897 | bh = head = page_buffers(page); |
f6d6d4fc | 898 | offset = page_offset(page); |
a206c817 CH |
899 | type = IO_OVERWRITE; |
900 | ||
dbcdde3e | 901 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 902 | nonblocking = 1; |
f6d6d4fc | 903 | |
1da177e4 | 904 | do { |
6ac7248e CH |
905 | int new_ioend = 0; |
906 | ||
1da177e4 LT |
907 | if (offset >= end_offset) |
908 | break; | |
909 | if (!buffer_uptodate(bh)) | |
910 | uptodate = 0; | |
1da177e4 | 911 | |
3d9b02e3 | 912 | /* |
ece413f5 CH |
913 | * set_page_dirty dirties all buffers in a page, independent |
914 | * of their state. The dirty state however is entirely | |
915 | * meaningless for holes (!mapped && uptodate), so skip | |
916 | * buffers covering holes here. | |
3d9b02e3 ES |
917 | */ |
918 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
919 | imap_valid = 0; |
920 | continue; | |
921 | } | |
922 | ||
aeea1b1f CH |
923 | if (buffer_unwritten(bh)) { |
924 | if (type != IO_UNWRITTEN) { | |
925 | type = IO_UNWRITTEN; | |
926 | imap_valid = 0; | |
1da177e4 | 927 | } |
aeea1b1f CH |
928 | } else if (buffer_delay(bh)) { |
929 | if (type != IO_DELALLOC) { | |
930 | type = IO_DELALLOC; | |
931 | imap_valid = 0; | |
1da177e4 | 932 | } |
89f3b363 | 933 | } else if (buffer_uptodate(bh)) { |
a206c817 CH |
934 | if (type != IO_OVERWRITE) { |
935 | type = IO_OVERWRITE; | |
85da94c6 CH |
936 | imap_valid = 0; |
937 | } | |
aeea1b1f CH |
938 | } else { |
939 | if (PageUptodate(page)) { | |
940 | ASSERT(buffer_mapped(bh)); | |
941 | imap_valid = 0; | |
6c4fe19f | 942 | } |
aeea1b1f CH |
943 | continue; |
944 | } | |
d5cb48aa | 945 | |
aeea1b1f CH |
946 | if (imap_valid) |
947 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
948 | if (!imap_valid) { | |
949 | /* | |
950 | * If we didn't have a valid mapping then we need to | |
951 | * put the new mapping into a separate ioend structure. | |
952 | * This ensures non-contiguous extents always have | |
953 | * separate ioends, which is particularly important | |
954 | * for unwritten extent conversion at I/O completion | |
955 | * time. | |
956 | */ | |
957 | new_ioend = 1; | |
958 | err = xfs_map_blocks(inode, offset, &imap, type, | |
959 | nonblocking); | |
960 | if (err) | |
961 | goto error; | |
962 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
963 | } | |
964 | if (imap_valid) { | |
ecff71e6 CH |
965 | lock_buffer(bh); |
966 | if (type != IO_OVERWRITE) | |
aeea1b1f CH |
967 | xfs_map_at_offset(inode, bh, &imap, offset); |
968 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
969 | new_ioend); | |
970 | count++; | |
1da177e4 | 971 | } |
f6d6d4fc CH |
972 | |
973 | if (!iohead) | |
974 | iohead = ioend; | |
975 | ||
976 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
977 | |
978 | if (uptodate && bh == head) | |
979 | SetPageUptodate(page); | |
980 | ||
89f3b363 | 981 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 982 | |
558e6891 | 983 | if (ioend && imap_valid) { |
bd1556a1 CH |
984 | xfs_off_t end_index; |
985 | ||
986 | end_index = imap.br_startoff + imap.br_blockcount; | |
987 | ||
988 | /* to bytes */ | |
989 | end_index <<= inode->i_blkbits; | |
990 | ||
991 | /* to pages */ | |
992 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
993 | ||
994 | /* check against file size */ | |
995 | if (end_index > last_index) | |
996 | end_index = last_index; | |
8699bb0a | 997 | |
207d0416 | 998 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 999 | wbc, end_index); |
1da177e4 LT |
1000 | } |
1001 | ||
f6d6d4fc | 1002 | if (iohead) |
06342cf8 | 1003 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1004 | |
89f3b363 | 1005 | return 0; |
1da177e4 LT |
1006 | |
1007 | error: | |
f6d6d4fc CH |
1008 | if (iohead) |
1009 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1010 | |
b5420f23 CH |
1011 | if (err == -EAGAIN) |
1012 | goto redirty; | |
1013 | ||
20cb52eb | 1014 | xfs_aops_discard_page(page); |
89f3b363 CH |
1015 | ClearPageUptodate(page); |
1016 | unlock_page(page); | |
1da177e4 | 1017 | return err; |
f51623b2 | 1018 | |
b5420f23 | 1019 | redirty: |
f51623b2 NS |
1020 | redirty_page_for_writepage(wbc, page); |
1021 | unlock_page(page); | |
1022 | return 0; | |
f51623b2 NS |
1023 | } |
1024 | ||
7d4fb40a NS |
1025 | STATIC int |
1026 | xfs_vm_writepages( | |
1027 | struct address_space *mapping, | |
1028 | struct writeback_control *wbc) | |
1029 | { | |
b3aea4ed | 1030 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1031 | return generic_writepages(mapping, wbc); |
1032 | } | |
1033 | ||
f51623b2 NS |
1034 | /* |
1035 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1036 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1037 | * have buffer heads in this call. |
1038 | * | |
89f3b363 | 1039 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1040 | */ |
1041 | STATIC int | |
238f4c54 | 1042 | xfs_vm_releasepage( |
f51623b2 NS |
1043 | struct page *page, |
1044 | gfp_t gfp_mask) | |
1045 | { | |
20cb52eb | 1046 | int delalloc, unwritten; |
f51623b2 | 1047 | |
89f3b363 | 1048 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1049 | |
20cb52eb | 1050 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1051 | |
89f3b363 | 1052 | if (WARN_ON(delalloc)) |
f51623b2 | 1053 | return 0; |
89f3b363 | 1054 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1055 | return 0; |
1056 | ||
f51623b2 NS |
1057 | return try_to_free_buffers(page); |
1058 | } | |
1059 | ||
1da177e4 | 1060 | STATIC int |
c2536668 | 1061 | __xfs_get_blocks( |
1da177e4 LT |
1062 | struct inode *inode, |
1063 | sector_t iblock, | |
1da177e4 LT |
1064 | struct buffer_head *bh_result, |
1065 | int create, | |
f2bde9b8 | 1066 | int direct) |
1da177e4 | 1067 | { |
a206c817 CH |
1068 | struct xfs_inode *ip = XFS_I(inode); |
1069 | struct xfs_mount *mp = ip->i_mount; | |
1070 | xfs_fileoff_t offset_fsb, end_fsb; | |
1071 | int error = 0; | |
1072 | int lockmode = 0; | |
207d0416 | 1073 | struct xfs_bmbt_irec imap; |
a206c817 | 1074 | int nimaps = 1; |
fdc7ed75 NS |
1075 | xfs_off_t offset; |
1076 | ssize_t size; | |
207d0416 | 1077 | int new = 0; |
a206c817 CH |
1078 | |
1079 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1080 | return -XFS_ERROR(EIO); | |
1da177e4 | 1081 | |
fdc7ed75 | 1082 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1083 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1084 | size = bh_result->b_size; | |
364f358a LM |
1085 | |
1086 | if (!create && direct && offset >= i_size_read(inode)) | |
1087 | return 0; | |
1088 | ||
a206c817 CH |
1089 | if (create) { |
1090 | lockmode = XFS_ILOCK_EXCL; | |
1091 | xfs_ilock(ip, lockmode); | |
1092 | } else { | |
1093 | lockmode = xfs_ilock_map_shared(ip); | |
1094 | } | |
f2bde9b8 | 1095 | |
a206c817 CH |
1096 | ASSERT(offset <= mp->m_maxioffset); |
1097 | if (offset + size > mp->m_maxioffset) | |
1098 | size = mp->m_maxioffset - offset; | |
1099 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1100 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1101 | ||
5c8ed202 DC |
1102 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1103 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1104 | if (error) |
a206c817 CH |
1105 | goto out_unlock; |
1106 | ||
1107 | if (create && | |
1108 | (!nimaps || | |
1109 | (imap.br_startblock == HOLESTARTBLOCK || | |
1110 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1111 | if (direct) { | |
1112 | error = xfs_iomap_write_direct(ip, offset, size, | |
1113 | &imap, nimaps); | |
1114 | } else { | |
1115 | error = xfs_iomap_write_delay(ip, offset, size, &imap); | |
1116 | } | |
1117 | if (error) | |
1118 | goto out_unlock; | |
1119 | ||
1120 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1121 | } else if (nimaps) { | |
1122 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
1123 | } else { | |
1124 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1125 | goto out_unlock; | |
1126 | } | |
1127 | xfs_iunlock(ip, lockmode); | |
1da177e4 | 1128 | |
207d0416 CH |
1129 | if (imap.br_startblock != HOLESTARTBLOCK && |
1130 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1131 | /* |
1132 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1133 | * the read case (treat as if we're reading into a hole). |
1134 | */ | |
207d0416 CH |
1135 | if (create || !ISUNWRITTEN(&imap)) |
1136 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1137 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1138 | if (direct) |
1139 | bh_result->b_private = inode; | |
1140 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1141 | } |
1142 | } | |
1143 | ||
c2536668 NS |
1144 | /* |
1145 | * If this is a realtime file, data may be on a different device. | |
1146 | * to that pointed to from the buffer_head b_bdev currently. | |
1147 | */ | |
046f1685 | 1148 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1149 | |
c2536668 | 1150 | /* |
549054af DC |
1151 | * If we previously allocated a block out beyond eof and we are now |
1152 | * coming back to use it then we will need to flag it as new even if it | |
1153 | * has a disk address. | |
1154 | * | |
1155 | * With sub-block writes into unwritten extents we also need to mark | |
1156 | * the buffer as new so that the unwritten parts of the buffer gets | |
1157 | * correctly zeroed. | |
1da177e4 LT |
1158 | */ |
1159 | if (create && | |
1160 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1161 | (offset >= i_size_read(inode)) || |
207d0416 | 1162 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1163 | set_buffer_new(bh_result); |
1da177e4 | 1164 | |
207d0416 | 1165 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1166 | BUG_ON(direct); |
1167 | if (create) { | |
1168 | set_buffer_uptodate(bh_result); | |
1169 | set_buffer_mapped(bh_result); | |
1170 | set_buffer_delay(bh_result); | |
1171 | } | |
1172 | } | |
1173 | ||
2b8f12b7 CH |
1174 | /* |
1175 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1176 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1177 | */ | |
c2536668 | 1178 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1179 | xfs_off_t mapping_size; |
1180 | ||
1181 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1182 | mapping_size <<= inode->i_blkbits; | |
1183 | ||
1184 | ASSERT(mapping_size > 0); | |
1185 | if (mapping_size > size) | |
1186 | mapping_size = size; | |
1187 | if (mapping_size > LONG_MAX) | |
1188 | mapping_size = LONG_MAX; | |
1189 | ||
1190 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | return 0; | |
a206c817 CH |
1194 | |
1195 | out_unlock: | |
1196 | xfs_iunlock(ip, lockmode); | |
1197 | return -error; | |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | int | |
c2536668 | 1201 | xfs_get_blocks( |
1da177e4 LT |
1202 | struct inode *inode, |
1203 | sector_t iblock, | |
1204 | struct buffer_head *bh_result, | |
1205 | int create) | |
1206 | { | |
f2bde9b8 | 1207 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1208 | } |
1209 | ||
1210 | STATIC int | |
e4c573bb | 1211 | xfs_get_blocks_direct( |
1da177e4 LT |
1212 | struct inode *inode, |
1213 | sector_t iblock, | |
1da177e4 LT |
1214 | struct buffer_head *bh_result, |
1215 | int create) | |
1216 | { | |
f2bde9b8 | 1217 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1218 | } |
1219 | ||
209fb87a CH |
1220 | /* |
1221 | * Complete a direct I/O write request. | |
1222 | * | |
1223 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1224 | * need to issue a transaction to convert the range from unwritten to written | |
1225 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1226 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1227 | * request this handler is called from interrupt context, from which we |
1228 | * can't start transactions. In that case offload the I/O completion to | |
1229 | * the workqueues we also use for buffered I/O completion. | |
1230 | */ | |
f0973863 | 1231 | STATIC void |
209fb87a CH |
1232 | xfs_end_io_direct_write( |
1233 | struct kiocb *iocb, | |
1234 | loff_t offset, | |
1235 | ssize_t size, | |
1236 | void *private, | |
1237 | int ret, | |
1238 | bool is_async) | |
f0973863 | 1239 | { |
209fb87a | 1240 | struct xfs_ioend *ioend = iocb->private; |
f0973863 | 1241 | |
2813d682 CH |
1242 | /* |
1243 | * While the generic direct I/O code updates the inode size, it does | |
1244 | * so only after the end_io handler is called, which means our | |
1245 | * end_io handler thinks the on-disk size is outside the in-core | |
1246 | * size. To prevent this just update it a little bit earlier here. | |
1247 | */ | |
1248 | if (offset + size > i_size_read(ioend->io_inode)) | |
1249 | i_size_write(ioend->io_inode, offset + size); | |
1250 | ||
f0973863 | 1251 | /* |
209fb87a CH |
1252 | * blockdev_direct_IO can return an error even after the I/O |
1253 | * completion handler was called. Thus we need to protect | |
1254 | * against double-freeing. | |
f0973863 | 1255 | */ |
209fb87a CH |
1256 | iocb->private = NULL; |
1257 | ||
ba87ea69 LM |
1258 | ioend->io_offset = offset; |
1259 | ioend->io_size = size; | |
c859cdd1 CH |
1260 | ioend->io_iocb = iocb; |
1261 | ioend->io_result = ret; | |
209fb87a CH |
1262 | if (private && size > 0) |
1263 | ioend->io_type = IO_UNWRITTEN; | |
1264 | ||
1265 | if (is_async) { | |
c859cdd1 | 1266 | ioend->io_isasync = 1; |
209fb87a | 1267 | xfs_finish_ioend(ioend); |
f0973863 | 1268 | } else { |
209fb87a | 1269 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1270 | } |
f0973863 CH |
1271 | } |
1272 | ||
1da177e4 | 1273 | STATIC ssize_t |
e4c573bb | 1274 | xfs_vm_direct_IO( |
1da177e4 LT |
1275 | int rw, |
1276 | struct kiocb *iocb, | |
1277 | const struct iovec *iov, | |
1278 | loff_t offset, | |
1279 | unsigned long nr_segs) | |
1280 | { | |
209fb87a CH |
1281 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1282 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1283 | ssize_t ret; | |
1284 | ||
1285 | if (rw & WRITE) { | |
a206c817 | 1286 | iocb->private = xfs_alloc_ioend(inode, IO_DIRECT); |
209fb87a | 1287 | |
eafdc7d1 CH |
1288 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1289 | offset, nr_segs, | |
1290 | xfs_get_blocks_direct, | |
1291 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1292 | if (ret != -EIOCBQUEUED && iocb->private) |
1293 | xfs_destroy_ioend(iocb->private); | |
1294 | } else { | |
eafdc7d1 CH |
1295 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1296 | offset, nr_segs, | |
1297 | xfs_get_blocks_direct, | |
1298 | NULL, NULL, 0); | |
209fb87a | 1299 | } |
f0973863 | 1300 | |
f0973863 | 1301 | return ret; |
1da177e4 LT |
1302 | } |
1303 | ||
fa9b227e CH |
1304 | STATIC void |
1305 | xfs_vm_write_failed( | |
1306 | struct address_space *mapping, | |
1307 | loff_t to) | |
1308 | { | |
1309 | struct inode *inode = mapping->host; | |
1310 | ||
1311 | if (to > inode->i_size) { | |
c726de44 | 1312 | /* |
2813d682 CH |
1313 | * Punch out the delalloc blocks we have already allocated. |
1314 | * | |
1315 | * Don't bother with xfs_setattr given that nothing can have | |
1316 | * made it to disk yet as the page is still locked at this | |
1317 | * point. | |
c726de44 DC |
1318 | */ |
1319 | struct xfs_inode *ip = XFS_I(inode); | |
1320 | xfs_fileoff_t start_fsb; | |
1321 | xfs_fileoff_t end_fsb; | |
1322 | int error; | |
1323 | ||
1324 | truncate_pagecache(inode, to, inode->i_size); | |
1325 | ||
1326 | /* | |
1327 | * Check if there are any blocks that are outside of i_size | |
1328 | * that need to be trimmed back. | |
1329 | */ | |
1330 | start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1; | |
1331 | end_fsb = XFS_B_TO_FSB(ip->i_mount, to); | |
1332 | if (end_fsb <= start_fsb) | |
1333 | return; | |
1334 | ||
1335 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1336 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1337 | end_fsb - start_fsb); | |
1338 | if (error) { | |
1339 | /* something screwed, just bail */ | |
1340 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
4f10700a | 1341 | xfs_alert(ip->i_mount, |
c726de44 DC |
1342 | "xfs_vm_write_failed: unable to clean up ino %lld", |
1343 | ip->i_ino); | |
1344 | } | |
1345 | } | |
1346 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fa9b227e CH |
1347 | } |
1348 | } | |
1349 | ||
f51623b2 | 1350 | STATIC int |
d79689c7 | 1351 | xfs_vm_write_begin( |
f51623b2 | 1352 | struct file *file, |
d79689c7 NP |
1353 | struct address_space *mapping, |
1354 | loff_t pos, | |
1355 | unsigned len, | |
1356 | unsigned flags, | |
1357 | struct page **pagep, | |
1358 | void **fsdata) | |
f51623b2 | 1359 | { |
155130a4 CH |
1360 | int ret; |
1361 | ||
1362 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1363 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1364 | if (unlikely(ret)) |
1365 | xfs_vm_write_failed(mapping, pos + len); | |
1366 | return ret; | |
1367 | } | |
1368 | ||
1369 | STATIC int | |
1370 | xfs_vm_write_end( | |
1371 | struct file *file, | |
1372 | struct address_space *mapping, | |
1373 | loff_t pos, | |
1374 | unsigned len, | |
1375 | unsigned copied, | |
1376 | struct page *page, | |
1377 | void *fsdata) | |
1378 | { | |
1379 | int ret; | |
155130a4 | 1380 | |
fa9b227e CH |
1381 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1382 | if (unlikely(ret < len)) | |
1383 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1384 | return ret; |
f51623b2 | 1385 | } |
1da177e4 LT |
1386 | |
1387 | STATIC sector_t | |
e4c573bb | 1388 | xfs_vm_bmap( |
1da177e4 LT |
1389 | struct address_space *mapping, |
1390 | sector_t block) | |
1391 | { | |
1392 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1393 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1394 | |
cca28fb8 | 1395 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1396 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1397 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1398 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1399 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1400 | } |
1401 | ||
1402 | STATIC int | |
e4c573bb | 1403 | xfs_vm_readpage( |
1da177e4 LT |
1404 | struct file *unused, |
1405 | struct page *page) | |
1406 | { | |
c2536668 | 1407 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1408 | } |
1409 | ||
1410 | STATIC int | |
e4c573bb | 1411 | xfs_vm_readpages( |
1da177e4 LT |
1412 | struct file *unused, |
1413 | struct address_space *mapping, | |
1414 | struct list_head *pages, | |
1415 | unsigned nr_pages) | |
1416 | { | |
c2536668 | 1417 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1418 | } |
1419 | ||
f5e54d6e | 1420 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1421 | .readpage = xfs_vm_readpage, |
1422 | .readpages = xfs_vm_readpages, | |
1423 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1424 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1425 | .releasepage = xfs_vm_releasepage, |
1426 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1427 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1428 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1429 | .bmap = xfs_vm_bmap, |
1430 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1431 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1432 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1433 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1434 | }; |