]>
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 LT |
24 | #include "xfs_dir2.h" |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_dmapi.h" | |
27 | #include "xfs_mount.h" | |
28 | #include "xfs_bmap_btree.h" | |
29 | #include "xfs_alloc_btree.h" | |
30 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 31 | #include "xfs_dir2_sf.h" |
a844f451 | 32 | #include "xfs_attr_sf.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_alloc.h" |
36 | #include "xfs_btree.h" | |
1da177e4 LT |
37 | #include "xfs_error.h" |
38 | #include "xfs_rw.h" | |
39 | #include "xfs_iomap.h" | |
40 | #include <linux/mpage.h> | |
10ce4444 | 41 | #include <linux/pagevec.h> |
1da177e4 LT |
42 | #include <linux/writeback.h> |
43 | ||
f51623b2 NS |
44 | STATIC void |
45 | xfs_count_page_state( | |
46 | struct page *page, | |
47 | int *delalloc, | |
48 | int *unmapped, | |
49 | int *unwritten) | |
50 | { | |
51 | struct buffer_head *bh, *head; | |
52 | ||
53 | *delalloc = *unmapped = *unwritten = 0; | |
54 | ||
55 | bh = head = page_buffers(page); | |
56 | do { | |
57 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
58 | (*unmapped) = 1; | |
59 | else if (buffer_unwritten(bh) && !buffer_delay(bh)) | |
60 | clear_buffer_unwritten(bh); | |
61 | else if (buffer_unwritten(bh)) | |
62 | (*unwritten) = 1; | |
63 | else if (buffer_delay(bh)) | |
64 | (*delalloc) = 1; | |
65 | } while ((bh = bh->b_this_page) != head); | |
66 | } | |
67 | ||
1da177e4 LT |
68 | #if defined(XFS_RW_TRACE) |
69 | void | |
70 | xfs_page_trace( | |
71 | int tag, | |
72 | struct inode *inode, | |
73 | struct page *page, | |
ed9d88f7 | 74 | unsigned long pgoff) |
1da177e4 LT |
75 | { |
76 | xfs_inode_t *ip; | |
67fcaa73 | 77 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 78 | loff_t isize = i_size_read(inode); |
f6d6d4fc | 79 | loff_t offset = page_offset(page); |
1da177e4 LT |
80 | int delalloc = -1, unmapped = -1, unwritten = -1; |
81 | ||
82 | if (page_has_buffers(page)) | |
83 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
84 | ||
75e17b3c | 85 | ip = xfs_vtoi(vp); |
1da177e4 LT |
86 | if (!ip->i_rwtrace) |
87 | return; | |
88 | ||
89 | ktrace_enter(ip->i_rwtrace, | |
90 | (void *)((unsigned long)tag), | |
91 | (void *)ip, | |
92 | (void *)inode, | |
93 | (void *)page, | |
ed9d88f7 | 94 | (void *)pgoff, |
1da177e4 LT |
95 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), |
96 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
97 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
98 | (void *)((unsigned long)(isize & 0xffffffff)), | |
99 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
100 | (void *)((unsigned long)(offset & 0xffffffff)), | |
101 | (void *)((unsigned long)delalloc), | |
102 | (void *)((unsigned long)unmapped), | |
103 | (void *)((unsigned long)unwritten), | |
f1fdc848 | 104 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
105 | (void *)NULL); |
106 | } | |
107 | #else | |
ed9d88f7 | 108 | #define xfs_page_trace(tag, inode, page, pgoff) |
1da177e4 LT |
109 | #endif |
110 | ||
0829c360 CH |
111 | /* |
112 | * Schedule IO completion handling on a xfsdatad if this was | |
113 | * the final hold on this ioend. | |
114 | */ | |
115 | STATIC void | |
116 | xfs_finish_ioend( | |
117 | xfs_ioend_t *ioend) | |
118 | { | |
119 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
120 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
121 | } | |
122 | ||
f6d6d4fc CH |
123 | /* |
124 | * We're now finished for good with this ioend structure. | |
125 | * Update the page state via the associated buffer_heads, | |
126 | * release holds on the inode and bio, and finally free | |
127 | * up memory. Do not use the ioend after this. | |
128 | */ | |
0829c360 CH |
129 | STATIC void |
130 | xfs_destroy_ioend( | |
131 | xfs_ioend_t *ioend) | |
132 | { | |
f6d6d4fc CH |
133 | struct buffer_head *bh, *next; |
134 | ||
135 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
136 | next = bh->b_private; | |
7d04a335 | 137 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 138 | } |
7d04a335 NS |
139 | if (unlikely(ioend->io_error)) |
140 | vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__); | |
0829c360 CH |
141 | vn_iowake(ioend->io_vnode); |
142 | mempool_free(ioend, xfs_ioend_pool); | |
143 | } | |
144 | ||
145 | /* | |
f6d6d4fc CH |
146 | * Buffered IO write completion for delayed allocate extents. |
147 | * TODO: Update ondisk isize now that we know the file data | |
148 | * has been flushed (i.e. the notorious "NULL file" problem). | |
149 | */ | |
150 | STATIC void | |
151 | xfs_end_bio_delalloc( | |
152 | void *data) | |
153 | { | |
154 | xfs_ioend_t *ioend = data; | |
155 | ||
156 | xfs_destroy_ioend(ioend); | |
157 | } | |
158 | ||
159 | /* | |
160 | * Buffered IO write completion for regular, written extents. | |
161 | */ | |
162 | STATIC void | |
163 | xfs_end_bio_written( | |
164 | void *data) | |
165 | { | |
166 | xfs_ioend_t *ioend = data; | |
167 | ||
168 | xfs_destroy_ioend(ioend); | |
169 | } | |
170 | ||
171 | /* | |
172 | * IO write completion for unwritten extents. | |
173 | * | |
0829c360 | 174 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 175 | * to written extents. |
0829c360 CH |
176 | */ |
177 | STATIC void | |
178 | xfs_end_bio_unwritten( | |
179 | void *data) | |
180 | { | |
181 | xfs_ioend_t *ioend = data; | |
67fcaa73 | 182 | bhv_vnode_t *vp = ioend->io_vnode; |
0829c360 CH |
183 | xfs_off_t offset = ioend->io_offset; |
184 | size_t size = ioend->io_size; | |
0829c360 | 185 | |
7d04a335 | 186 | if (likely(!ioend->io_error)) |
67fcaa73 | 187 | bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL); |
0829c360 CH |
188 | xfs_destroy_ioend(ioend); |
189 | } | |
190 | ||
191 | /* | |
192 | * Allocate and initialise an IO completion structure. | |
193 | * We need to track unwritten extent write completion here initially. | |
194 | * We'll need to extend this for updating the ondisk inode size later | |
195 | * (vs. incore size). | |
196 | */ | |
197 | STATIC xfs_ioend_t * | |
198 | xfs_alloc_ioend( | |
f6d6d4fc CH |
199 | struct inode *inode, |
200 | unsigned int type) | |
0829c360 CH |
201 | { |
202 | xfs_ioend_t *ioend; | |
203 | ||
204 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
205 | ||
206 | /* | |
207 | * Set the count to 1 initially, which will prevent an I/O | |
208 | * completion callback from happening before we have started | |
209 | * all the I/O from calling the completion routine too early. | |
210 | */ | |
211 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 212 | ioend->io_error = 0; |
f6d6d4fc CH |
213 | ioend->io_list = NULL; |
214 | ioend->io_type = type; | |
ec86dc02 | 215 | ioend->io_vnode = vn_from_inode(inode); |
c1a073bd | 216 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 217 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
218 | atomic_inc(&ioend->io_vnode->v_iocount); |
219 | ioend->io_offset = 0; | |
220 | ioend->io_size = 0; | |
221 | ||
f6d6d4fc CH |
222 | if (type == IOMAP_UNWRITTEN) |
223 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend); | |
224 | else if (type == IOMAP_DELAY) | |
225 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend); | |
226 | else | |
227 | INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend); | |
0829c360 CH |
228 | |
229 | return ioend; | |
230 | } | |
231 | ||
1da177e4 LT |
232 | STATIC int |
233 | xfs_map_blocks( | |
234 | struct inode *inode, | |
235 | loff_t offset, | |
236 | ssize_t count, | |
237 | xfs_iomap_t *mapp, | |
238 | int flags) | |
239 | { | |
67fcaa73 | 240 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
241 | int error, nmaps = 1; |
242 | ||
67fcaa73 | 243 | error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps); |
1da177e4 LT |
244 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) |
245 | VMODIFY(vp); | |
246 | return -error; | |
247 | } | |
248 | ||
1defeac9 CH |
249 | STATIC inline int |
250 | xfs_iomap_valid( | |
1da177e4 | 251 | xfs_iomap_t *iomapp, |
1defeac9 | 252 | loff_t offset) |
1da177e4 | 253 | { |
1defeac9 CH |
254 | return offset >= iomapp->iomap_offset && |
255 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
256 | } |
257 | ||
f6d6d4fc CH |
258 | /* |
259 | * BIO completion handler for buffered IO. | |
260 | */ | |
261 | STATIC int | |
262 | xfs_end_bio( | |
263 | struct bio *bio, | |
264 | unsigned int bytes_done, | |
265 | int error) | |
266 | { | |
267 | xfs_ioend_t *ioend = bio->bi_private; | |
268 | ||
269 | if (bio->bi_size) | |
270 | return 1; | |
271 | ||
f6d6d4fc | 272 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 273 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
274 | |
275 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
276 | bio->bi_private = NULL; |
277 | bio->bi_end_io = NULL; | |
f6d6d4fc | 278 | bio_put(bio); |
7d04a335 | 279 | |
f6d6d4fc CH |
280 | xfs_finish_ioend(ioend); |
281 | return 0; | |
282 | } | |
283 | ||
284 | STATIC void | |
285 | xfs_submit_ioend_bio( | |
286 | xfs_ioend_t *ioend, | |
287 | struct bio *bio) | |
288 | { | |
289 | atomic_inc(&ioend->io_remaining); | |
290 | ||
291 | bio->bi_private = ioend; | |
292 | bio->bi_end_io = xfs_end_bio; | |
293 | ||
294 | submit_bio(WRITE, bio); | |
295 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
296 | bio_put(bio); | |
297 | } | |
298 | ||
299 | STATIC struct bio * | |
300 | xfs_alloc_ioend_bio( | |
301 | struct buffer_head *bh) | |
302 | { | |
303 | struct bio *bio; | |
304 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
305 | ||
306 | do { | |
307 | bio = bio_alloc(GFP_NOIO, nvecs); | |
308 | nvecs >>= 1; | |
309 | } while (!bio); | |
310 | ||
311 | ASSERT(bio->bi_private == NULL); | |
312 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
313 | bio->bi_bdev = bh->b_bdev; | |
314 | bio_get(bio); | |
315 | return bio; | |
316 | } | |
317 | ||
318 | STATIC void | |
319 | xfs_start_buffer_writeback( | |
320 | struct buffer_head *bh) | |
321 | { | |
322 | ASSERT(buffer_mapped(bh)); | |
323 | ASSERT(buffer_locked(bh)); | |
324 | ASSERT(!buffer_delay(bh)); | |
325 | ASSERT(!buffer_unwritten(bh)); | |
326 | ||
327 | mark_buffer_async_write(bh); | |
328 | set_buffer_uptodate(bh); | |
329 | clear_buffer_dirty(bh); | |
330 | } | |
331 | ||
332 | STATIC void | |
333 | xfs_start_page_writeback( | |
334 | struct page *page, | |
335 | struct writeback_control *wbc, | |
336 | int clear_dirty, | |
337 | int buffers) | |
338 | { | |
339 | ASSERT(PageLocked(page)); | |
340 | ASSERT(!PageWriteback(page)); | |
341 | set_page_writeback(page); | |
342 | if (clear_dirty) | |
343 | clear_page_dirty(page); | |
344 | unlock_page(page); | |
345 | if (!buffers) { | |
346 | end_page_writeback(page); | |
347 | wbc->pages_skipped++; /* We didn't write this page */ | |
348 | } | |
349 | } | |
350 | ||
351 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
352 | { | |
353 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
354 | } | |
355 | ||
356 | /* | |
d88992f6 DC |
357 | * Submit all of the bios for all of the ioends we have saved up, covering the |
358 | * initial writepage page and also any probed pages. | |
359 | * | |
360 | * Because we may have multiple ioends spanning a page, we need to start | |
361 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
362 | * buffers as we got, then we can end up with a page that only has buffers | |
363 | * marked async write and I/O complete on can occur before we mark the other | |
364 | * buffers async write. | |
365 | * | |
366 | * The end result of this is that we trip a bug in end_page_writeback() because | |
367 | * we call it twice for the one page as the code in end_buffer_async_write() | |
368 | * assumes that all buffers on the page are started at the same time. | |
369 | * | |
370 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 371 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
372 | */ |
373 | STATIC void | |
374 | xfs_submit_ioend( | |
375 | xfs_ioend_t *ioend) | |
376 | { | |
d88992f6 | 377 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
378 | xfs_ioend_t *next; |
379 | struct buffer_head *bh; | |
380 | struct bio *bio; | |
381 | sector_t lastblock = 0; | |
382 | ||
d88992f6 DC |
383 | /* Pass 1 - start writeback */ |
384 | do { | |
385 | next = ioend->io_list; | |
386 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
387 | xfs_start_buffer_writeback(bh); | |
388 | } | |
389 | } while ((ioend = next) != NULL); | |
390 | ||
391 | /* Pass 2 - submit I/O */ | |
392 | ioend = head; | |
f6d6d4fc CH |
393 | do { |
394 | next = ioend->io_list; | |
395 | bio = NULL; | |
396 | ||
397 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
398 | |
399 | if (!bio) { | |
400 | retry: | |
401 | bio = xfs_alloc_ioend_bio(bh); | |
402 | } else if (bh->b_blocknr != lastblock + 1) { | |
403 | xfs_submit_ioend_bio(ioend, bio); | |
404 | goto retry; | |
405 | } | |
406 | ||
407 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
408 | xfs_submit_ioend_bio(ioend, bio); | |
409 | goto retry; | |
410 | } | |
411 | ||
412 | lastblock = bh->b_blocknr; | |
413 | } | |
414 | if (bio) | |
415 | xfs_submit_ioend_bio(ioend, bio); | |
416 | xfs_finish_ioend(ioend); | |
417 | } while ((ioend = next) != NULL); | |
418 | } | |
419 | ||
420 | /* | |
421 | * Cancel submission of all buffer_heads so far in this endio. | |
422 | * Toss the endio too. Only ever called for the initial page | |
423 | * in a writepage request, so only ever one page. | |
424 | */ | |
425 | STATIC void | |
426 | xfs_cancel_ioend( | |
427 | xfs_ioend_t *ioend) | |
428 | { | |
429 | xfs_ioend_t *next; | |
430 | struct buffer_head *bh, *next_bh; | |
431 | ||
432 | do { | |
433 | next = ioend->io_list; | |
434 | bh = ioend->io_buffer_head; | |
435 | do { | |
436 | next_bh = bh->b_private; | |
437 | clear_buffer_async_write(bh); | |
438 | unlock_buffer(bh); | |
439 | } while ((bh = next_bh) != NULL); | |
440 | ||
441 | vn_iowake(ioend->io_vnode); | |
442 | mempool_free(ioend, xfs_ioend_pool); | |
443 | } while ((ioend = next) != NULL); | |
444 | } | |
445 | ||
446 | /* | |
447 | * Test to see if we've been building up a completion structure for | |
448 | * earlier buffers -- if so, we try to append to this ioend if we | |
449 | * can, otherwise we finish off any current ioend and start another. | |
450 | * Return true if we've finished the given ioend. | |
451 | */ | |
452 | STATIC void | |
453 | xfs_add_to_ioend( | |
454 | struct inode *inode, | |
455 | struct buffer_head *bh, | |
7336cea8 | 456 | xfs_off_t offset, |
f6d6d4fc CH |
457 | unsigned int type, |
458 | xfs_ioend_t **result, | |
459 | int need_ioend) | |
460 | { | |
461 | xfs_ioend_t *ioend = *result; | |
462 | ||
463 | if (!ioend || need_ioend || type != ioend->io_type) { | |
464 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 465 | |
f6d6d4fc CH |
466 | ioend = xfs_alloc_ioend(inode, type); |
467 | ioend->io_offset = offset; | |
468 | ioend->io_buffer_head = bh; | |
469 | ioend->io_buffer_tail = bh; | |
470 | if (previous) | |
471 | previous->io_list = ioend; | |
472 | *result = ioend; | |
473 | } else { | |
474 | ioend->io_buffer_tail->b_private = bh; | |
475 | ioend->io_buffer_tail = bh; | |
476 | } | |
477 | ||
478 | bh->b_private = NULL; | |
479 | ioend->io_size += bh->b_size; | |
480 | } | |
481 | ||
87cbc49c NS |
482 | STATIC void |
483 | xfs_map_buffer( | |
484 | struct buffer_head *bh, | |
485 | xfs_iomap_t *mp, | |
486 | xfs_off_t offset, | |
487 | uint block_bits) | |
488 | { | |
489 | sector_t bn; | |
490 | ||
491 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
492 | ||
493 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
494 | ((offset - mp->iomap_offset) >> block_bits); | |
495 | ||
496 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
497 | ||
498 | bh->b_blocknr = bn; | |
499 | set_buffer_mapped(bh); | |
500 | } | |
501 | ||
1da177e4 LT |
502 | STATIC void |
503 | xfs_map_at_offset( | |
1da177e4 | 504 | struct buffer_head *bh, |
1defeac9 | 505 | loff_t offset, |
1da177e4 | 506 | int block_bits, |
1defeac9 | 507 | xfs_iomap_t *iomapp) |
1da177e4 | 508 | { |
1da177e4 LT |
509 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
510 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
511 | |
512 | lock_buffer(bh); | |
87cbc49c | 513 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 514 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
515 | set_buffer_mapped(bh); |
516 | clear_buffer_delay(bh); | |
f6d6d4fc | 517 | clear_buffer_unwritten(bh); |
1da177e4 LT |
518 | } |
519 | ||
520 | /* | |
6c4fe19f | 521 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
522 | */ |
523 | STATIC unsigned int | |
6c4fe19f | 524 | xfs_probe_page( |
10ce4444 | 525 | struct page *page, |
6c4fe19f CH |
526 | unsigned int pg_offset, |
527 | int mapped) | |
1da177e4 | 528 | { |
1da177e4 LT |
529 | int ret = 0; |
530 | ||
1da177e4 | 531 | if (PageWriteback(page)) |
10ce4444 | 532 | return 0; |
1da177e4 LT |
533 | |
534 | if (page->mapping && PageDirty(page)) { | |
535 | if (page_has_buffers(page)) { | |
536 | struct buffer_head *bh, *head; | |
537 | ||
538 | bh = head = page_buffers(page); | |
539 | do { | |
6c4fe19f CH |
540 | if (!buffer_uptodate(bh)) |
541 | break; | |
542 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
543 | break; |
544 | ret += bh->b_size; | |
545 | if (ret >= pg_offset) | |
546 | break; | |
547 | } while ((bh = bh->b_this_page) != head); | |
548 | } else | |
6c4fe19f | 549 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
550 | } |
551 | ||
1da177e4 LT |
552 | return ret; |
553 | } | |
554 | ||
f6d6d4fc | 555 | STATIC size_t |
6c4fe19f | 556 | xfs_probe_cluster( |
1da177e4 LT |
557 | struct inode *inode, |
558 | struct page *startpage, | |
559 | struct buffer_head *bh, | |
6c4fe19f CH |
560 | struct buffer_head *head, |
561 | int mapped) | |
1da177e4 | 562 | { |
10ce4444 | 563 | struct pagevec pvec; |
1da177e4 | 564 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
565 | size_t total = 0; |
566 | int done = 0, i; | |
1da177e4 LT |
567 | |
568 | /* First sum forwards in this page */ | |
569 | do { | |
2353e8e9 | 570 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 571 | return total; |
1da177e4 LT |
572 | total += bh->b_size; |
573 | } while ((bh = bh->b_this_page) != head); | |
574 | ||
10ce4444 CH |
575 | /* if we reached the end of the page, sum forwards in following pages */ |
576 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
577 | tindex = startpage->index + 1; | |
578 | ||
579 | /* Prune this back to avoid pathological behavior */ | |
580 | tloff = min(tlast, startpage->index + 64); | |
581 | ||
582 | pagevec_init(&pvec, 0); | |
583 | while (!done && tindex <= tloff) { | |
584 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
585 | ||
586 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
587 | break; | |
588 | ||
589 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
590 | struct page *page = pvec.pages[i]; | |
591 | size_t pg_offset, len = 0; | |
592 | ||
593 | if (tindex == tlast) { | |
594 | pg_offset = | |
595 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
596 | if (!pg_offset) { |
597 | done = 1; | |
10ce4444 | 598 | break; |
1defeac9 | 599 | } |
10ce4444 CH |
600 | } else |
601 | pg_offset = PAGE_CACHE_SIZE; | |
602 | ||
603 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
6c4fe19f | 604 | len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
605 | unlock_page(page); |
606 | } | |
607 | ||
608 | if (!len) { | |
609 | done = 1; | |
610 | break; | |
611 | } | |
612 | ||
1da177e4 | 613 | total += len; |
1defeac9 | 614 | tindex++; |
1da177e4 | 615 | } |
10ce4444 CH |
616 | |
617 | pagevec_release(&pvec); | |
618 | cond_resched(); | |
1da177e4 | 619 | } |
10ce4444 | 620 | |
1da177e4 LT |
621 | return total; |
622 | } | |
623 | ||
624 | /* | |
10ce4444 CH |
625 | * Test if a given page is suitable for writing as part of an unwritten |
626 | * or delayed allocate extent. | |
1da177e4 | 627 | */ |
10ce4444 CH |
628 | STATIC int |
629 | xfs_is_delayed_page( | |
630 | struct page *page, | |
f6d6d4fc | 631 | unsigned int type) |
1da177e4 | 632 | { |
1da177e4 | 633 | if (PageWriteback(page)) |
10ce4444 | 634 | return 0; |
1da177e4 LT |
635 | |
636 | if (page->mapping && page_has_buffers(page)) { | |
637 | struct buffer_head *bh, *head; | |
638 | int acceptable = 0; | |
639 | ||
640 | bh = head = page_buffers(page); | |
641 | do { | |
f6d6d4fc CH |
642 | if (buffer_unwritten(bh)) |
643 | acceptable = (type == IOMAP_UNWRITTEN); | |
644 | else if (buffer_delay(bh)) | |
645 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 646 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
6c4fe19f | 647 | acceptable = (type == 0); |
f6d6d4fc | 648 | else |
1da177e4 | 649 | break; |
1da177e4 LT |
650 | } while ((bh = bh->b_this_page) != head); |
651 | ||
652 | if (acceptable) | |
10ce4444 | 653 | return 1; |
1da177e4 LT |
654 | } |
655 | ||
10ce4444 | 656 | return 0; |
1da177e4 LT |
657 | } |
658 | ||
1da177e4 LT |
659 | /* |
660 | * Allocate & map buffers for page given the extent map. Write it out. | |
661 | * except for the original page of a writepage, this is called on | |
662 | * delalloc/unwritten pages only, for the original page it is possible | |
663 | * that the page has no mapping at all. | |
664 | */ | |
f6d6d4fc | 665 | STATIC int |
1da177e4 LT |
666 | xfs_convert_page( |
667 | struct inode *inode, | |
668 | struct page *page, | |
10ce4444 | 669 | loff_t tindex, |
1defeac9 | 670 | xfs_iomap_t *mp, |
f6d6d4fc | 671 | xfs_ioend_t **ioendp, |
1da177e4 | 672 | struct writeback_control *wbc, |
1da177e4 LT |
673 | int startio, |
674 | int all_bh) | |
675 | { | |
f6d6d4fc | 676 | struct buffer_head *bh, *head; |
9260dc6b CH |
677 | xfs_off_t end_offset; |
678 | unsigned long p_offset; | |
f6d6d4fc | 679 | unsigned int type; |
1da177e4 | 680 | int bbits = inode->i_blkbits; |
24e17b5f | 681 | int len, page_dirty; |
f6d6d4fc | 682 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 683 | xfs_off_t offset = page_offset(page); |
1da177e4 | 684 | |
10ce4444 CH |
685 | if (page->index != tindex) |
686 | goto fail; | |
687 | if (TestSetPageLocked(page)) | |
688 | goto fail; | |
689 | if (PageWriteback(page)) | |
690 | goto fail_unlock_page; | |
691 | if (page->mapping != inode->i_mapping) | |
692 | goto fail_unlock_page; | |
693 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
694 | goto fail_unlock_page; | |
695 | ||
24e17b5f NS |
696 | /* |
697 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 698 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
699 | * |
700 | * Derivation: | |
701 | * | |
702 | * End offset is the highest offset that this page should represent. | |
703 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
704 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
705 | * hence give us the correct page_dirty count. On any other page, | |
706 | * it will be zero and in that case we need page_dirty to be the | |
707 | * count of buffers on the page. | |
24e17b5f | 708 | */ |
9260dc6b CH |
709 | end_offset = min_t(unsigned long long, |
710 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
711 | i_size_read(inode)); | |
712 | ||
24e17b5f | 713 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
714 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
715 | PAGE_CACHE_SIZE); | |
716 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
717 | page_dirty = p_offset / len; | |
24e17b5f | 718 | |
1da177e4 LT |
719 | bh = head = page_buffers(page); |
720 | do { | |
9260dc6b | 721 | if (offset >= end_offset) |
1da177e4 | 722 | break; |
f6d6d4fc CH |
723 | if (!buffer_uptodate(bh)) |
724 | uptodate = 0; | |
725 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
726 | done = 1; | |
1da177e4 | 727 | continue; |
f6d6d4fc CH |
728 | } |
729 | ||
9260dc6b CH |
730 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
731 | if (buffer_unwritten(bh)) | |
732 | type = IOMAP_UNWRITTEN; | |
733 | else | |
734 | type = IOMAP_DELAY; | |
735 | ||
736 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 737 | done = 1; |
9260dc6b CH |
738 | continue; |
739 | } | |
740 | ||
741 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
742 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
743 | ||
744 | xfs_map_at_offset(bh, offset, bbits, mp); | |
745 | if (startio) { | |
7336cea8 | 746 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
747 | type, ioendp, done); |
748 | } else { | |
749 | set_buffer_dirty(bh); | |
750 | unlock_buffer(bh); | |
751 | mark_buffer_dirty(bh); | |
752 | } | |
753 | page_dirty--; | |
754 | count++; | |
755 | } else { | |
756 | type = 0; | |
757 | if (buffer_mapped(bh) && all_bh && startio) { | |
1da177e4 | 758 | lock_buffer(bh); |
7336cea8 | 759 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
760 | type, ioendp, done); |
761 | count++; | |
24e17b5f | 762 | page_dirty--; |
9260dc6b CH |
763 | } else { |
764 | done = 1; | |
1da177e4 | 765 | } |
1da177e4 | 766 | } |
7336cea8 | 767 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 768 | |
f6d6d4fc CH |
769 | if (uptodate && bh == head) |
770 | SetPageUptodate(page); | |
771 | ||
772 | if (startio) { | |
f5e596bb CH |
773 | if (count) { |
774 | struct backing_dev_info *bdi; | |
775 | ||
776 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 777 | wbc->nr_to_write--; |
f5e596bb CH |
778 | if (bdi_write_congested(bdi)) { |
779 | wbc->encountered_congestion = 1; | |
780 | done = 1; | |
9fddaca2 | 781 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
782 | done = 1; |
783 | } | |
784 | } | |
f6d6d4fc | 785 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 786 | } |
f6d6d4fc CH |
787 | |
788 | return done; | |
10ce4444 CH |
789 | fail_unlock_page: |
790 | unlock_page(page); | |
791 | fail: | |
792 | return 1; | |
1da177e4 LT |
793 | } |
794 | ||
795 | /* | |
796 | * Convert & write out a cluster of pages in the same extent as defined | |
797 | * by mp and following the start page. | |
798 | */ | |
799 | STATIC void | |
800 | xfs_cluster_write( | |
801 | struct inode *inode, | |
802 | pgoff_t tindex, | |
803 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 804 | xfs_ioend_t **ioendp, |
1da177e4 LT |
805 | struct writeback_control *wbc, |
806 | int startio, | |
807 | int all_bh, | |
808 | pgoff_t tlast) | |
809 | { | |
10ce4444 CH |
810 | struct pagevec pvec; |
811 | int done = 0, i; | |
1da177e4 | 812 | |
10ce4444 CH |
813 | pagevec_init(&pvec, 0); |
814 | while (!done && tindex <= tlast) { | |
815 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
816 | ||
817 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 818 | break; |
10ce4444 CH |
819 | |
820 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
821 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
822 | iomapp, ioendp, wbc, startio, all_bh); | |
823 | if (done) | |
824 | break; | |
825 | } | |
826 | ||
827 | pagevec_release(&pvec); | |
828 | cond_resched(); | |
1da177e4 LT |
829 | } |
830 | } | |
831 | ||
832 | /* | |
833 | * Calling this without startio set means we are being asked to make a dirty | |
834 | * page ready for freeing it's buffers. When called with startio set then | |
835 | * we are coming from writepage. | |
836 | * | |
837 | * When called with startio set it is important that we write the WHOLE | |
838 | * page if possible. | |
839 | * The bh->b_state's cannot know if any of the blocks or which block for | |
840 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 841 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
842 | * may clear the page dirty flag prior to calling write page, under the |
843 | * assumption the entire page will be written out; by not writing out the | |
844 | * whole page the page can be reused before all valid dirty data is | |
845 | * written out. Note: in the case of a page that has been dirty'd by | |
846 | * mapwrite and but partially setup by block_prepare_write the | |
847 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
848 | * valid state, thus the whole page must be written out thing. | |
849 | */ | |
850 | ||
851 | STATIC int | |
852 | xfs_page_state_convert( | |
853 | struct inode *inode, | |
854 | struct page *page, | |
855 | struct writeback_control *wbc, | |
856 | int startio, | |
857 | int unmapped) /* also implies page uptodate */ | |
858 | { | |
f6d6d4fc | 859 | struct buffer_head *bh, *head; |
1defeac9 | 860 | xfs_iomap_t iomap; |
f6d6d4fc | 861 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
862 | loff_t offset; |
863 | unsigned long p_offset = 0; | |
f6d6d4fc | 864 | unsigned int type; |
1da177e4 LT |
865 | __uint64_t end_offset; |
866 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
867 | ssize_t size, len; |
868 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
869 | int page_dirty, count = 0; |
870 | int trylock = 0; | |
6c4fe19f | 871 | int all_bh = unmapped; |
1da177e4 | 872 | |
8272145c NS |
873 | if (startio) { |
874 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
875 | trylock |= BMAPI_TRYLOCK; | |
876 | } | |
3ba0815a | 877 | |
1da177e4 LT |
878 | /* Is this page beyond the end of the file? */ |
879 | offset = i_size_read(inode); | |
880 | end_index = offset >> PAGE_CACHE_SHIFT; | |
881 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
882 | if (page->index >= end_index) { | |
883 | if ((page->index >= end_index + 1) || | |
884 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
885 | if (startio) |
886 | unlock_page(page); | |
887 | return 0; | |
1da177e4 LT |
888 | } |
889 | } | |
890 | ||
1da177e4 | 891 | /* |
24e17b5f | 892 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 893 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
894 | * |
895 | * Derivation: | |
896 | * | |
897 | * End offset is the highest offset that this page should represent. | |
898 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
899 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
900 | * hence give us the correct page_dirty count. On any other page, | |
901 | * it will be zero and in that case we need page_dirty to be the | |
902 | * count of buffers on the page. | |
903 | */ | |
904 | end_offset = min_t(unsigned long long, | |
905 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 906 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
907 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
908 | PAGE_CACHE_SIZE); | |
909 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
910 | page_dirty = p_offset / len; |
911 | ||
24e17b5f | 912 | bh = head = page_buffers(page); |
f6d6d4fc | 913 | offset = page_offset(page); |
6c4fe19f CH |
914 | flags = -1; |
915 | type = 0; | |
f6d6d4fc | 916 | |
f6d6d4fc | 917 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
918 | |
919 | do { | |
920 | if (offset >= end_offset) | |
921 | break; | |
922 | if (!buffer_uptodate(bh)) | |
923 | uptodate = 0; | |
f6d6d4fc | 924 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
925 | /* |
926 | * the iomap is actually still valid, but the ioend | |
927 | * isn't. shouldn't happen too often. | |
928 | */ | |
929 | iomap_valid = 0; | |
1da177e4 | 930 | continue; |
f6d6d4fc | 931 | } |
1da177e4 | 932 | |
1defeac9 CH |
933 | if (iomap_valid) |
934 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
935 | |
936 | /* | |
937 | * First case, map an unwritten extent and prepare for | |
938 | * extent state conversion transaction on completion. | |
f6d6d4fc | 939 | * |
1da177e4 LT |
940 | * Second case, allocate space for a delalloc buffer. |
941 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
942 | * |
943 | * Third case, an unmapped buffer was found, and we are | |
944 | * in a path where we need to write the whole page out. | |
945 | */ | |
946 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
947 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
948 | !buffer_mapped(bh) && (unmapped || startio))) { | |
6c4fe19f CH |
949 | /* |
950 | * Make sure we don't use a read-only iomap | |
951 | */ | |
952 | if (flags == BMAPI_READ) | |
953 | iomap_valid = 0; | |
954 | ||
f6d6d4fc CH |
955 | if (buffer_unwritten(bh)) { |
956 | type = IOMAP_UNWRITTEN; | |
8272145c | 957 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 958 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 959 | type = IOMAP_DELAY; |
8272145c | 960 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 961 | } else { |
6c4fe19f | 962 | type = IOMAP_NEW; |
8272145c | 963 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
964 | } |
965 | ||
1defeac9 | 966 | if (!iomap_valid) { |
6c4fe19f CH |
967 | if (type == IOMAP_NEW) { |
968 | size = xfs_probe_cluster(inode, | |
969 | page, bh, head, 0); | |
d5cb48aa CH |
970 | } else { |
971 | size = len; | |
972 | } | |
973 | ||
974 | err = xfs_map_blocks(inode, offset, size, | |
975 | &iomap, flags); | |
f6d6d4fc | 976 | if (err) |
1da177e4 | 977 | goto error; |
1defeac9 | 978 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 979 | } |
1defeac9 CH |
980 | if (iomap_valid) { |
981 | xfs_map_at_offset(bh, offset, | |
982 | inode->i_blkbits, &iomap); | |
1da177e4 | 983 | if (startio) { |
7336cea8 | 984 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 CH |
985 | type, &ioend, |
986 | !iomap_valid); | |
1da177e4 LT |
987 | } else { |
988 | set_buffer_dirty(bh); | |
989 | unlock_buffer(bh); | |
990 | mark_buffer_dirty(bh); | |
991 | } | |
992 | page_dirty--; | |
f6d6d4fc | 993 | count++; |
1da177e4 | 994 | } |
d5cb48aa | 995 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
996 | /* |
997 | * we got here because the buffer is already mapped. | |
998 | * That means it must already have extents allocated | |
999 | * underneath it. Map the extent by reading it. | |
1000 | */ | |
1001 | if (!iomap_valid || type != 0) { | |
1002 | flags = BMAPI_READ; | |
1003 | size = xfs_probe_cluster(inode, page, bh, | |
1004 | head, 1); | |
1005 | err = xfs_map_blocks(inode, offset, size, | |
1006 | &iomap, flags); | |
1007 | if (err) | |
1008 | goto error; | |
1009 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1010 | } | |
d5cb48aa | 1011 | |
6c4fe19f | 1012 | type = 0; |
d5cb48aa CH |
1013 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1014 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1015 | if (iomap_valid) |
1016 | all_bh = 1; | |
7336cea8 | 1017 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1018 | &ioend, !iomap_valid); |
1019 | page_dirty--; | |
1020 | count++; | |
f6d6d4fc | 1021 | } else { |
1defeac9 | 1022 | iomap_valid = 0; |
1da177e4 | 1023 | } |
d5cb48aa CH |
1024 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1025 | (unmapped || startio)) { | |
1026 | iomap_valid = 0; | |
1da177e4 | 1027 | } |
f6d6d4fc CH |
1028 | |
1029 | if (!iohead) | |
1030 | iohead = ioend; | |
1031 | ||
1032 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1033 | |
1034 | if (uptodate && bh == head) | |
1035 | SetPageUptodate(page); | |
1036 | ||
f6d6d4fc CH |
1037 | if (startio) |
1038 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1039 | |
1defeac9 CH |
1040 | if (ioend && iomap_valid) { |
1041 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1042 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1043 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1044 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1045 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1046 | } |
1047 | ||
f6d6d4fc CH |
1048 | if (iohead) |
1049 | xfs_submit_ioend(iohead); | |
1050 | ||
1da177e4 LT |
1051 | return page_dirty; |
1052 | ||
1053 | error: | |
f6d6d4fc CH |
1054 | if (iohead) |
1055 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1056 | |
1057 | /* | |
1058 | * If it's delalloc and we have nowhere to put it, | |
1059 | * throw it away, unless the lower layers told | |
1060 | * us to try again. | |
1061 | */ | |
1062 | if (err != -EAGAIN) { | |
f6d6d4fc | 1063 | if (!unmapped) |
1da177e4 | 1064 | block_invalidatepage(page, 0); |
1da177e4 LT |
1065 | ClearPageUptodate(page); |
1066 | } | |
1067 | return err; | |
1068 | } | |
1069 | ||
f51623b2 NS |
1070 | /* |
1071 | * writepage: Called from one of two places: | |
1072 | * | |
1073 | * 1. we are flushing a delalloc buffer head. | |
1074 | * | |
1075 | * 2. we are writing out a dirty page. Typically the page dirty | |
1076 | * state is cleared before we get here. In this case is it | |
1077 | * conceivable we have no buffer heads. | |
1078 | * | |
1079 | * For delalloc space on the page we need to allocate space and | |
1080 | * flush it. For unmapped buffer heads on the page we should | |
1081 | * allocate space if the page is uptodate. For any other dirty | |
1082 | * buffer heads on the page we should flush them. | |
1083 | * | |
1084 | * If we detect that a transaction would be required to flush | |
1085 | * the page, we have to check the process flags first, if we | |
1086 | * are already in a transaction or disk I/O during allocations | |
1087 | * is off, we need to fail the writepage and redirty the page. | |
1088 | */ | |
1089 | ||
1090 | STATIC int | |
e4c573bb | 1091 | xfs_vm_writepage( |
f51623b2 NS |
1092 | struct page *page, |
1093 | struct writeback_control *wbc) | |
1094 | { | |
1095 | int error; | |
1096 | int need_trans; | |
1097 | int delalloc, unmapped, unwritten; | |
1098 | struct inode *inode = page->mapping->host; | |
1099 | ||
1100 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1101 | ||
1102 | /* | |
1103 | * We need a transaction if: | |
1104 | * 1. There are delalloc buffers on the page | |
1105 | * 2. The page is uptodate and we have unmapped buffers | |
1106 | * 3. The page is uptodate and we have no buffers | |
1107 | * 4. There are unwritten buffers on the page | |
1108 | */ | |
1109 | ||
1110 | if (!page_has_buffers(page)) { | |
1111 | unmapped = 1; | |
1112 | need_trans = 1; | |
1113 | } else { | |
1114 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1115 | if (!PageUptodate(page)) | |
1116 | unmapped = 0; | |
1117 | need_trans = delalloc + unmapped + unwritten; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * If we need a transaction and the process flags say | |
1122 | * we are already in a transaction, or no IO is allowed | |
1123 | * then mark the page dirty again and leave the page | |
1124 | * as is. | |
1125 | */ | |
59c1b082 | 1126 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1127 | goto out_fail; |
1128 | ||
1129 | /* | |
1130 | * Delay hooking up buffer heads until we have | |
1131 | * made our go/no-go decision. | |
1132 | */ | |
1133 | if (!page_has_buffers(page)) | |
1134 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1135 | ||
1136 | /* | |
1137 | * Convert delayed allocate, unwritten or unmapped space | |
1138 | * to real space and flush out to disk. | |
1139 | */ | |
1140 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1141 | if (error == -EAGAIN) | |
1142 | goto out_fail; | |
1143 | if (unlikely(error < 0)) | |
1144 | goto out_unlock; | |
1145 | ||
1146 | return 0; | |
1147 | ||
1148 | out_fail: | |
1149 | redirty_page_for_writepage(wbc, page); | |
1150 | unlock_page(page); | |
1151 | return 0; | |
1152 | out_unlock: | |
1153 | unlock_page(page); | |
1154 | return error; | |
1155 | } | |
1156 | ||
7d4fb40a NS |
1157 | STATIC int |
1158 | xfs_vm_writepages( | |
1159 | struct address_space *mapping, | |
1160 | struct writeback_control *wbc) | |
1161 | { | |
67fcaa73 | 1162 | struct bhv_vnode *vp = vn_from_inode(mapping->host); |
7d4fb40a NS |
1163 | |
1164 | if (VN_TRUNC(vp)) | |
1165 | VUNTRUNCATE(vp); | |
1166 | return generic_writepages(mapping, wbc); | |
1167 | } | |
1168 | ||
f51623b2 NS |
1169 | /* |
1170 | * Called to move a page into cleanable state - and from there | |
1171 | * to be released. Possibly the page is already clean. We always | |
1172 | * have buffer heads in this call. | |
1173 | * | |
1174 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1175 | * | |
1176 | * Possible scenarios are: | |
1177 | * | |
1178 | * 1. We are being called to release a page which has been written | |
1179 | * to via regular I/O. buffer heads will be dirty and possibly | |
1180 | * delalloc. If no delalloc buffer heads in this case then we | |
1181 | * can just return zero. | |
1182 | * | |
1183 | * 2. We are called to release a page which has been written via | |
1184 | * mmap, all we need to do is ensure there is no delalloc | |
1185 | * state in the buffer heads, if not we can let the caller | |
1186 | * free them and we should come back later via writepage. | |
1187 | */ | |
1188 | STATIC int | |
238f4c54 | 1189 | xfs_vm_releasepage( |
f51623b2 NS |
1190 | struct page *page, |
1191 | gfp_t gfp_mask) | |
1192 | { | |
1193 | struct inode *inode = page->mapping->host; | |
1194 | int dirty, delalloc, unmapped, unwritten; | |
1195 | struct writeback_control wbc = { | |
1196 | .sync_mode = WB_SYNC_ALL, | |
1197 | .nr_to_write = 1, | |
1198 | }; | |
1199 | ||
ed9d88f7 | 1200 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1201 | |
238f4c54 NS |
1202 | if (!page_has_buffers(page)) |
1203 | return 0; | |
1204 | ||
f51623b2 NS |
1205 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1206 | if (!delalloc && !unwritten) | |
1207 | goto free_buffers; | |
1208 | ||
1209 | if (!(gfp_mask & __GFP_FS)) | |
1210 | return 0; | |
1211 | ||
1212 | /* If we are already inside a transaction or the thread cannot | |
1213 | * do I/O, we cannot release this page. | |
1214 | */ | |
59c1b082 | 1215 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1216 | return 0; |
1217 | ||
1218 | /* | |
1219 | * Convert delalloc space to real space, do not flush the | |
1220 | * data out to disk, that will be done by the caller. | |
1221 | * Never need to allocate space here - we will always | |
1222 | * come back to writepage in that case. | |
1223 | */ | |
1224 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1225 | if (dirty == 0 && !unwritten) | |
1226 | goto free_buffers; | |
1227 | return 0; | |
1228 | ||
1229 | free_buffers: | |
1230 | return try_to_free_buffers(page); | |
1231 | } | |
1232 | ||
1da177e4 | 1233 | STATIC int |
c2536668 | 1234 | __xfs_get_blocks( |
1da177e4 LT |
1235 | struct inode *inode, |
1236 | sector_t iblock, | |
1da177e4 LT |
1237 | struct buffer_head *bh_result, |
1238 | int create, | |
1239 | int direct, | |
1240 | bmapi_flags_t flags) | |
1241 | { | |
67fcaa73 | 1242 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1243 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1244 | xfs_off_t offset; |
1245 | ssize_t size; | |
c2536668 | 1246 | int niomap = 1; |
1da177e4 | 1247 | int error; |
1da177e4 | 1248 | |
fdc7ed75 | 1249 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1250 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1251 | size = bh_result->b_size; | |
67fcaa73 NS |
1252 | error = bhv_vop_bmap(vp, offset, size, |
1253 | create ? flags : BMAPI_READ, &iomap, &niomap); | |
1da177e4 LT |
1254 | if (error) |
1255 | return -error; | |
c2536668 | 1256 | if (niomap == 0) |
1da177e4 LT |
1257 | return 0; |
1258 | ||
1259 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1260 | /* |
1261 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1262 | * the read case (treat as if we're reading into a hole). |
1263 | */ | |
1264 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1265 | xfs_map_buffer(bh_result, &iomap, offset, |
1266 | inode->i_blkbits); | |
1da177e4 LT |
1267 | } |
1268 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1269 | if (direct) | |
1270 | bh_result->b_private = inode; | |
1271 | set_buffer_unwritten(bh_result); | |
1272 | set_buffer_delay(bh_result); | |
1273 | } | |
1274 | } | |
1275 | ||
c2536668 NS |
1276 | /* |
1277 | * If this is a realtime file, data may be on a different device. | |
1278 | * to that pointed to from the buffer_head b_bdev currently. | |
1279 | */ | |
ce8e922c | 1280 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1281 | |
c2536668 NS |
1282 | /* |
1283 | * If we previously allocated a block out beyond eof and we are | |
1284 | * now coming back to use it then we will need to flag it as new | |
1285 | * even if it has a disk address. | |
1da177e4 LT |
1286 | */ |
1287 | if (create && | |
1288 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
fdc7ed75 | 1289 | (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) |
1da177e4 | 1290 | set_buffer_new(bh_result); |
1da177e4 LT |
1291 | |
1292 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1293 | BUG_ON(direct); | |
1294 | if (create) { | |
1295 | set_buffer_uptodate(bh_result); | |
1296 | set_buffer_mapped(bh_result); | |
1297 | set_buffer_delay(bh_result); | |
1298 | } | |
1299 | } | |
1300 | ||
c2536668 | 1301 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1302 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1303 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1304 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1305 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1306 | } |
1307 | ||
1308 | return 0; | |
1309 | } | |
1310 | ||
1311 | int | |
c2536668 | 1312 | xfs_get_blocks( |
1da177e4 LT |
1313 | struct inode *inode, |
1314 | sector_t iblock, | |
1315 | struct buffer_head *bh_result, | |
1316 | int create) | |
1317 | { | |
c2536668 | 1318 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1319 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1320 | } |
1321 | ||
1322 | STATIC int | |
e4c573bb | 1323 | xfs_get_blocks_direct( |
1da177e4 LT |
1324 | struct inode *inode, |
1325 | sector_t iblock, | |
1da177e4 LT |
1326 | struct buffer_head *bh_result, |
1327 | int create) | |
1328 | { | |
c2536668 | 1329 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1330 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1331 | } |
1332 | ||
f0973863 | 1333 | STATIC void |
e4c573bb | 1334 | xfs_end_io_direct( |
f0973863 CH |
1335 | struct kiocb *iocb, |
1336 | loff_t offset, | |
1337 | ssize_t size, | |
1338 | void *private) | |
1339 | { | |
1340 | xfs_ioend_t *ioend = iocb->private; | |
1341 | ||
1342 | /* | |
1343 | * Non-NULL private data means we need to issue a transaction to | |
1344 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1345 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1346 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1347 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1348 | * it anyway to keep the code uniform and simpler. |
1349 | * | |
1350 | * The core direct I/O code might be changed to always call the | |
1351 | * completion handler in the future, in which case all this can | |
1352 | * go away. | |
1353 | */ | |
1354 | if (private && size > 0) { | |
1355 | ioend->io_offset = offset; | |
1356 | ioend->io_size = size; | |
1357 | xfs_finish_ioend(ioend); | |
1358 | } else { | |
f0973863 CH |
1359 | xfs_destroy_ioend(ioend); |
1360 | } | |
1361 | ||
1362 | /* | |
c41564b5 | 1363 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1364 | * completion handler was called. Thus we need to protect |
1365 | * against double-freeing. | |
1366 | */ | |
1367 | iocb->private = NULL; | |
1368 | } | |
1369 | ||
1da177e4 | 1370 | STATIC ssize_t |
e4c573bb | 1371 | xfs_vm_direct_IO( |
1da177e4 LT |
1372 | int rw, |
1373 | struct kiocb *iocb, | |
1374 | const struct iovec *iov, | |
1375 | loff_t offset, | |
1376 | unsigned long nr_segs) | |
1377 | { | |
1378 | struct file *file = iocb->ki_filp; | |
1379 | struct inode *inode = file->f_mapping->host; | |
67fcaa73 | 1380 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
1381 | xfs_iomap_t iomap; |
1382 | int maps = 1; | |
1383 | int error; | |
f0973863 | 1384 | ssize_t ret; |
1da177e4 | 1385 | |
67fcaa73 | 1386 | error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps); |
1da177e4 LT |
1387 | if (error) |
1388 | return -error; | |
1389 | ||
f6d6d4fc | 1390 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
f0973863 | 1391 | |
721259bc LM |
1392 | if (rw == WRITE) { |
1393 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, | |
1394 | iomap.iomap_target->bt_bdev, | |
1395 | iov, offset, nr_segs, | |
1396 | xfs_get_blocks_direct, | |
1397 | xfs_end_io_direct); | |
1398 | } else { | |
1399 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, | |
1400 | iomap.iomap_target->bt_bdev, | |
1401 | iov, offset, nr_segs, | |
1402 | xfs_get_blocks_direct, | |
1403 | xfs_end_io_direct); | |
1404 | } | |
f0973863 CH |
1405 | |
1406 | if (unlikely(ret <= 0 && iocb->private)) | |
1407 | xfs_destroy_ioend(iocb->private); | |
1408 | return ret; | |
1da177e4 LT |
1409 | } |
1410 | ||
f51623b2 | 1411 | STATIC int |
e4c573bb | 1412 | xfs_vm_prepare_write( |
f51623b2 NS |
1413 | struct file *file, |
1414 | struct page *page, | |
1415 | unsigned int from, | |
1416 | unsigned int to) | |
1417 | { | |
c2536668 | 1418 | return block_prepare_write(page, from, to, xfs_get_blocks); |
f51623b2 | 1419 | } |
1da177e4 LT |
1420 | |
1421 | STATIC sector_t | |
e4c573bb | 1422 | xfs_vm_bmap( |
1da177e4 LT |
1423 | struct address_space *mapping, |
1424 | sector_t block) | |
1425 | { | |
1426 | struct inode *inode = (struct inode *)mapping->host; | |
67fcaa73 | 1427 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1428 | |
e4c573bb | 1429 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
67fcaa73 NS |
1430 | bhv_vop_rwlock(vp, VRWLOCK_READ); |
1431 | bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF); | |
1432 | bhv_vop_rwunlock(vp, VRWLOCK_READ); | |
c2536668 | 1433 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1434 | } |
1435 | ||
1436 | STATIC int | |
e4c573bb | 1437 | xfs_vm_readpage( |
1da177e4 LT |
1438 | struct file *unused, |
1439 | struct page *page) | |
1440 | { | |
c2536668 | 1441 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1442 | } |
1443 | ||
1444 | STATIC int | |
e4c573bb | 1445 | xfs_vm_readpages( |
1da177e4 LT |
1446 | struct file *unused, |
1447 | struct address_space *mapping, | |
1448 | struct list_head *pages, | |
1449 | unsigned nr_pages) | |
1450 | { | |
c2536668 | 1451 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1452 | } |
1453 | ||
2ff28e22 | 1454 | STATIC void |
238f4c54 | 1455 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1456 | struct page *page, |
1457 | unsigned long offset) | |
1458 | { | |
1459 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1460 | page->mapping->host, page, offset); | |
2ff28e22 | 1461 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1462 | } |
1463 | ||
f5e54d6e | 1464 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1465 | .readpage = xfs_vm_readpage, |
1466 | .readpages = xfs_vm_readpages, | |
1467 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1468 | .writepages = xfs_vm_writepages, |
1da177e4 | 1469 | .sync_page = block_sync_page, |
238f4c54 NS |
1470 | .releasepage = xfs_vm_releasepage, |
1471 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1472 | .prepare_write = xfs_vm_prepare_write, |
1da177e4 | 1473 | .commit_write = generic_commit_write, |
e4c573bb NS |
1474 | .bmap = xfs_vm_bmap, |
1475 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1476 | .migratepage = buffer_migrate_page, |
1da177e4 | 1477 | }; |