Commit | Line | Data |
---|---|---|
64b5a32e RK |
1 | /* |
2 | * segbuf.c - NILFS segment buffer | |
3 | * | |
4 | * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, | |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
19 | * | |
20 | * Written by Ryusuke Konishi <ryusuke@osrg.net> | |
21 | * | |
22 | */ | |
23 | ||
24 | #include <linux/buffer_head.h> | |
25 | #include <linux/writeback.h> | |
26 | #include <linux/crc32.h> | |
9c965bac | 27 | #include <linux/backing-dev.h> |
64b5a32e RK |
28 | #include "page.h" |
29 | #include "segbuf.h" | |
64b5a32e RK |
30 | |
31 | ||
9c965bac RK |
32 | struct nilfs_write_info { |
33 | struct the_nilfs *nilfs; | |
34 | struct bio *bio; | |
35 | int start, end; /* The region to be submitted */ | |
36 | int rest_blocks; | |
37 | int max_pages; | |
38 | int nr_vecs; | |
39 | sector_t blocknr; | |
40 | }; | |
41 | ||
42 | ||
64b5a32e RK |
43 | static struct kmem_cache *nilfs_segbuf_cachep; |
44 | ||
45 | static void nilfs_segbuf_init_once(void *obj) | |
46 | { | |
47 | memset(obj, 0, sizeof(struct nilfs_segment_buffer)); | |
48 | } | |
49 | ||
50 | int __init nilfs_init_segbuf_cache(void) | |
51 | { | |
52 | nilfs_segbuf_cachep = | |
53 | kmem_cache_create("nilfs2_segbuf_cache", | |
54 | sizeof(struct nilfs_segment_buffer), | |
55 | 0, SLAB_RECLAIM_ACCOUNT, | |
56 | nilfs_segbuf_init_once); | |
57 | ||
58 | return (nilfs_segbuf_cachep == NULL) ? -ENOMEM : 0; | |
59 | } | |
60 | ||
61 | void nilfs_destroy_segbuf_cache(void) | |
62 | { | |
63 | kmem_cache_destroy(nilfs_segbuf_cachep); | |
64 | } | |
65 | ||
66 | struct nilfs_segment_buffer *nilfs_segbuf_new(struct super_block *sb) | |
67 | { | |
68 | struct nilfs_segment_buffer *segbuf; | |
69 | ||
70 | segbuf = kmem_cache_alloc(nilfs_segbuf_cachep, GFP_NOFS); | |
71 | if (unlikely(!segbuf)) | |
72 | return NULL; | |
73 | ||
74 | segbuf->sb_super = sb; | |
75 | INIT_LIST_HEAD(&segbuf->sb_list); | |
76 | INIT_LIST_HEAD(&segbuf->sb_segsum_buffers); | |
77 | INIT_LIST_HEAD(&segbuf->sb_payload_buffers); | |
9284ad2a RK |
78 | |
79 | init_completion(&segbuf->sb_bio_event); | |
80 | atomic_set(&segbuf->sb_err, 0); | |
81 | segbuf->sb_nbio = 0; | |
82 | ||
64b5a32e RK |
83 | return segbuf; |
84 | } | |
85 | ||
86 | void nilfs_segbuf_free(struct nilfs_segment_buffer *segbuf) | |
87 | { | |
64b5a32e RK |
88 | kmem_cache_free(nilfs_segbuf_cachep, segbuf); |
89 | } | |
90 | ||
cece5520 | 91 | void nilfs_segbuf_map(struct nilfs_segment_buffer *segbuf, __u64 segnum, |
64b5a32e RK |
92 | unsigned long offset, struct the_nilfs *nilfs) |
93 | { | |
64b5a32e RK |
94 | segbuf->sb_segnum = segnum; |
95 | nilfs_get_segment_range(nilfs, segnum, &segbuf->sb_fseg_start, | |
96 | &segbuf->sb_fseg_end); | |
97 | ||
98 | segbuf->sb_pseg_start = segbuf->sb_fseg_start + offset; | |
99 | segbuf->sb_rest_blocks = | |
100 | segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1; | |
64b5a32e RK |
101 | } |
102 | ||
103 | void nilfs_segbuf_set_next_segnum(struct nilfs_segment_buffer *segbuf, | |
104 | __u64 nextnum, struct the_nilfs *nilfs) | |
105 | { | |
106 | segbuf->sb_nextnum = nextnum; | |
107 | segbuf->sb_sum.next = nilfs_get_segment_start_blocknr(nilfs, nextnum); | |
108 | } | |
109 | ||
110 | int nilfs_segbuf_extend_segsum(struct nilfs_segment_buffer *segbuf) | |
111 | { | |
112 | struct buffer_head *bh; | |
113 | ||
114 | bh = sb_getblk(segbuf->sb_super, | |
115 | segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk); | |
116 | if (unlikely(!bh)) | |
117 | return -ENOMEM; | |
118 | ||
119 | nilfs_segbuf_add_segsum_buffer(segbuf, bh); | |
120 | return 0; | |
121 | } | |
122 | ||
123 | int nilfs_segbuf_extend_payload(struct nilfs_segment_buffer *segbuf, | |
124 | struct buffer_head **bhp) | |
125 | { | |
126 | struct buffer_head *bh; | |
127 | ||
128 | bh = sb_getblk(segbuf->sb_super, | |
129 | segbuf->sb_pseg_start + segbuf->sb_sum.nblocks); | |
130 | if (unlikely(!bh)) | |
131 | return -ENOMEM; | |
132 | ||
133 | nilfs_segbuf_add_payload_buffer(segbuf, bh); | |
134 | *bhp = bh; | |
135 | return 0; | |
136 | } | |
137 | ||
138 | int nilfs_segbuf_reset(struct nilfs_segment_buffer *segbuf, unsigned flags, | |
139 | time_t ctime) | |
140 | { | |
141 | int err; | |
142 | ||
143 | segbuf->sb_sum.nblocks = segbuf->sb_sum.nsumblk = 0; | |
144 | err = nilfs_segbuf_extend_segsum(segbuf); | |
145 | if (unlikely(err)) | |
146 | return err; | |
147 | ||
148 | segbuf->sb_sum.flags = flags; | |
149 | segbuf->sb_sum.sumbytes = sizeof(struct nilfs_segment_summary); | |
150 | segbuf->sb_sum.nfinfo = segbuf->sb_sum.nfileblk = 0; | |
151 | segbuf->sb_sum.ctime = ctime; | |
64b5a32e RK |
152 | return 0; |
153 | } | |
154 | ||
155 | /* | |
156 | * Setup segument summary | |
157 | */ | |
158 | void nilfs_segbuf_fill_in_segsum(struct nilfs_segment_buffer *segbuf) | |
159 | { | |
160 | struct nilfs_segment_summary *raw_sum; | |
161 | struct buffer_head *bh_sum; | |
162 | ||
163 | bh_sum = list_entry(segbuf->sb_segsum_buffers.next, | |
164 | struct buffer_head, b_assoc_buffers); | |
165 | raw_sum = (struct nilfs_segment_summary *)bh_sum->b_data; | |
166 | ||
167 | raw_sum->ss_magic = cpu_to_le32(NILFS_SEGSUM_MAGIC); | |
168 | raw_sum->ss_bytes = cpu_to_le16(sizeof(*raw_sum)); | |
169 | raw_sum->ss_flags = cpu_to_le16(segbuf->sb_sum.flags); | |
170 | raw_sum->ss_seq = cpu_to_le64(segbuf->sb_sum.seg_seq); | |
171 | raw_sum->ss_create = cpu_to_le64(segbuf->sb_sum.ctime); | |
172 | raw_sum->ss_next = cpu_to_le64(segbuf->sb_sum.next); | |
173 | raw_sum->ss_nblocks = cpu_to_le32(segbuf->sb_sum.nblocks); | |
174 | raw_sum->ss_nfinfo = cpu_to_le32(segbuf->sb_sum.nfinfo); | |
175 | raw_sum->ss_sumbytes = cpu_to_le32(segbuf->sb_sum.sumbytes); | |
176 | raw_sum->ss_pad = 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * CRC calculation routines | |
181 | */ | |
182 | void nilfs_segbuf_fill_in_segsum_crc(struct nilfs_segment_buffer *segbuf, | |
183 | u32 seed) | |
184 | { | |
185 | struct buffer_head *bh; | |
186 | struct nilfs_segment_summary *raw_sum; | |
187 | unsigned long size, bytes = segbuf->sb_sum.sumbytes; | |
188 | u32 crc; | |
189 | ||
190 | bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head, | |
191 | b_assoc_buffers); | |
192 | ||
193 | raw_sum = (struct nilfs_segment_summary *)bh->b_data; | |
194 | size = min_t(unsigned long, bytes, bh->b_size); | |
195 | crc = crc32_le(seed, | |
196 | (unsigned char *)raw_sum + | |
197 | sizeof(raw_sum->ss_datasum) + sizeof(raw_sum->ss_sumsum), | |
198 | size - (sizeof(raw_sum->ss_datasum) + | |
199 | sizeof(raw_sum->ss_sumsum))); | |
200 | ||
201 | list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers, | |
202 | b_assoc_buffers) { | |
203 | bytes -= size; | |
204 | size = min_t(unsigned long, bytes, bh->b_size); | |
205 | crc = crc32_le(crc, bh->b_data, size); | |
206 | } | |
207 | raw_sum->ss_sumsum = cpu_to_le32(crc); | |
208 | } | |
209 | ||
210 | void nilfs_segbuf_fill_in_data_crc(struct nilfs_segment_buffer *segbuf, | |
211 | u32 seed) | |
212 | { | |
213 | struct buffer_head *bh; | |
214 | struct nilfs_segment_summary *raw_sum; | |
215 | void *kaddr; | |
216 | u32 crc; | |
217 | ||
218 | bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head, | |
219 | b_assoc_buffers); | |
220 | raw_sum = (struct nilfs_segment_summary *)bh->b_data; | |
221 | crc = crc32_le(seed, | |
222 | (unsigned char *)raw_sum + sizeof(raw_sum->ss_datasum), | |
223 | bh->b_size - sizeof(raw_sum->ss_datasum)); | |
224 | ||
225 | list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers, | |
226 | b_assoc_buffers) { | |
227 | crc = crc32_le(crc, bh->b_data, bh->b_size); | |
228 | } | |
229 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) { | |
230 | kaddr = kmap_atomic(bh->b_page, KM_USER0); | |
231 | crc = crc32_le(crc, kaddr + bh_offset(bh), bh->b_size); | |
232 | kunmap_atomic(kaddr, KM_USER0); | |
233 | } | |
234 | raw_sum->ss_datasum = cpu_to_le32(crc); | |
235 | } | |
236 | ||
e29df395 | 237 | static void nilfs_release_buffers(struct list_head *list) |
64b5a32e RK |
238 | { |
239 | struct buffer_head *bh, *n; | |
240 | ||
241 | list_for_each_entry_safe(bh, n, list, b_assoc_buffers) { | |
242 | list_del_init(&bh->b_assoc_buffers); | |
243 | if (buffer_nilfs_allocated(bh)) { | |
244 | struct page *clone_page = bh->b_page; | |
245 | ||
246 | /* remove clone page */ | |
247 | brelse(bh); | |
248 | page_cache_release(clone_page); /* for each bh */ | |
249 | if (page_count(clone_page) <= 2) { | |
250 | lock_page(clone_page); | |
251 | nilfs_free_private_page(clone_page); | |
252 | } | |
253 | continue; | |
254 | } | |
255 | brelse(bh); | |
256 | } | |
257 | } | |
258 | ||
e29df395 RK |
259 | static void nilfs_segbuf_clear(struct nilfs_segment_buffer *segbuf) |
260 | { | |
261 | nilfs_release_buffers(&segbuf->sb_segsum_buffers); | |
262 | nilfs_release_buffers(&segbuf->sb_payload_buffers); | |
263 | } | |
264 | ||
265 | /* | |
266 | * Iterators for segment buffers | |
267 | */ | |
268 | void nilfs_clear_logs(struct list_head *logs) | |
269 | { | |
270 | struct nilfs_segment_buffer *segbuf; | |
271 | ||
272 | list_for_each_entry(segbuf, logs, sb_list) | |
273 | nilfs_segbuf_clear(segbuf); | |
274 | } | |
275 | ||
276 | void nilfs_truncate_logs(struct list_head *logs, | |
277 | struct nilfs_segment_buffer *last) | |
278 | { | |
279 | struct nilfs_segment_buffer *n, *segbuf; | |
280 | ||
281 | segbuf = list_prepare_entry(last, logs, sb_list); | |
282 | list_for_each_entry_safe_continue(segbuf, n, logs, sb_list) { | |
283 | list_del_init(&segbuf->sb_list); | |
284 | nilfs_segbuf_clear(segbuf); | |
285 | nilfs_segbuf_free(segbuf); | |
286 | } | |
287 | } | |
288 | ||
289 | int nilfs_wait_on_logs(struct list_head *logs) | |
290 | { | |
291 | struct nilfs_segment_buffer *segbuf; | |
292 | int err; | |
293 | ||
294 | list_for_each_entry(segbuf, logs, sb_list) { | |
295 | err = nilfs_segbuf_wait(segbuf); | |
296 | if (err) | |
297 | return err; | |
298 | } | |
299 | return 0; | |
300 | } | |
301 | ||
64b5a32e RK |
302 | /* |
303 | * BIO operations | |
304 | */ | |
305 | static void nilfs_end_bio_write(struct bio *bio, int err) | |
306 | { | |
307 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
9284ad2a | 308 | struct nilfs_segment_buffer *segbuf = bio->bi_private; |
64b5a32e RK |
309 | |
310 | if (err == -EOPNOTSUPP) { | |
311 | set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); | |
312 | bio_put(bio); | |
313 | /* to be detected by submit_seg_bio() */ | |
314 | } | |
315 | ||
316 | if (!uptodate) | |
9284ad2a | 317 | atomic_inc(&segbuf->sb_err); |
64b5a32e RK |
318 | |
319 | bio_put(bio); | |
9284ad2a | 320 | complete(&segbuf->sb_bio_event); |
64b5a32e RK |
321 | } |
322 | ||
9284ad2a RK |
323 | static int nilfs_segbuf_submit_bio(struct nilfs_segment_buffer *segbuf, |
324 | struct nilfs_write_info *wi, int mode) | |
64b5a32e RK |
325 | { |
326 | struct bio *bio = wi->bio; | |
327 | int err; | |
328 | ||
9c965bac | 329 | if (segbuf->sb_nbio > 0 && bdi_write_congested(wi->nilfs->ns_bdi)) { |
9284ad2a RK |
330 | wait_for_completion(&segbuf->sb_bio_event); |
331 | segbuf->sb_nbio--; | |
332 | if (unlikely(atomic_read(&segbuf->sb_err))) { | |
64b5a32e RK |
333 | bio_put(bio); |
334 | err = -EIO; | |
335 | goto failed; | |
336 | } | |
337 | } | |
338 | ||
339 | bio->bi_end_io = nilfs_end_bio_write; | |
9284ad2a | 340 | bio->bi_private = segbuf; |
64b5a32e RK |
341 | bio_get(bio); |
342 | submit_bio(mode, bio); | |
343 | if (bio_flagged(bio, BIO_EOPNOTSUPP)) { | |
344 | bio_put(bio); | |
345 | err = -EOPNOTSUPP; | |
346 | goto failed; | |
347 | } | |
9284ad2a | 348 | segbuf->sb_nbio++; |
64b5a32e RK |
349 | bio_put(bio); |
350 | ||
351 | wi->bio = NULL; | |
352 | wi->rest_blocks -= wi->end - wi->start; | |
353 | wi->nr_vecs = min(wi->max_pages, wi->rest_blocks); | |
354 | wi->start = wi->end; | |
355 | return 0; | |
356 | ||
357 | failed: | |
358 | wi->bio = NULL; | |
359 | return err; | |
360 | } | |
361 | ||
362 | /** | |
9c965bac RK |
363 | * nilfs_alloc_seg_bio - allocate a new bio for writing log |
364 | * @nilfs: nilfs object | |
365 | * @start: start block number of the bio | |
64b5a32e RK |
366 | * @nr_vecs: request size of page vector. |
367 | * | |
64b5a32e RK |
368 | * Return Value: On success, pointer to the struct bio is returned. |
369 | * On error, NULL is returned. | |
370 | */ | |
9c965bac | 371 | static struct bio *nilfs_alloc_seg_bio(struct the_nilfs *nilfs, sector_t start, |
64b5a32e RK |
372 | int nr_vecs) |
373 | { | |
374 | struct bio *bio; | |
375 | ||
c1b353f0 | 376 | bio = bio_alloc(GFP_NOIO, nr_vecs); |
64b5a32e RK |
377 | if (bio == NULL) { |
378 | while (!bio && (nr_vecs >>= 1)) | |
c1b353f0 | 379 | bio = bio_alloc(GFP_NOIO, nr_vecs); |
64b5a32e RK |
380 | } |
381 | if (likely(bio)) { | |
9c965bac RK |
382 | bio->bi_bdev = nilfs->ns_bdev; |
383 | bio->bi_sector = start << (nilfs->ns_blocksize_bits - 9); | |
64b5a32e RK |
384 | } |
385 | return bio; | |
386 | } | |
387 | ||
9c965bac RK |
388 | static void nilfs_segbuf_prepare_write(struct nilfs_segment_buffer *segbuf, |
389 | struct nilfs_write_info *wi) | |
64b5a32e RK |
390 | { |
391 | wi->bio = NULL; | |
392 | wi->rest_blocks = segbuf->sb_sum.nblocks; | |
9c965bac | 393 | wi->max_pages = bio_get_nr_vecs(wi->nilfs->ns_bdev); |
64b5a32e RK |
394 | wi->nr_vecs = min(wi->max_pages, wi->rest_blocks); |
395 | wi->start = wi->end = 0; | |
64b5a32e | 396 | wi->blocknr = segbuf->sb_pseg_start; |
64b5a32e RK |
397 | } |
398 | ||
9284ad2a RK |
399 | static int nilfs_segbuf_submit_bh(struct nilfs_segment_buffer *segbuf, |
400 | struct nilfs_write_info *wi, | |
401 | struct buffer_head *bh, int mode) | |
64b5a32e RK |
402 | { |
403 | int len, err; | |
404 | ||
405 | BUG_ON(wi->nr_vecs <= 0); | |
406 | repeat: | |
407 | if (!wi->bio) { | |
9c965bac | 408 | wi->bio = nilfs_alloc_seg_bio(wi->nilfs, wi->blocknr + wi->end, |
64b5a32e RK |
409 | wi->nr_vecs); |
410 | if (unlikely(!wi->bio)) | |
411 | return -ENOMEM; | |
412 | } | |
413 | ||
414 | len = bio_add_page(wi->bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
415 | if (len == bh->b_size) { | |
416 | wi->end++; | |
417 | return 0; | |
418 | } | |
419 | /* bio is FULL */ | |
9284ad2a | 420 | err = nilfs_segbuf_submit_bio(segbuf, wi, mode); |
64b5a32e RK |
421 | /* never submit current bh */ |
422 | if (likely(!err)) | |
423 | goto repeat; | |
424 | return err; | |
425 | } | |
426 | ||
9c965bac RK |
427 | /** |
428 | * nilfs_segbuf_write - submit write requests of a log | |
429 | * @segbuf: buffer storing a log to be written | |
430 | * @nilfs: nilfs object | |
431 | * | |
432 | * Return Value: On Success, 0 is returned. On Error, one of the following | |
433 | * negative error code is returned. | |
434 | * | |
435 | * %-EIO - I/O error | |
436 | * | |
437 | * %-ENOMEM - Insufficient memory available. | |
438 | */ | |
64b5a32e | 439 | int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf, |
9c965bac | 440 | struct the_nilfs *nilfs) |
64b5a32e | 441 | { |
9c965bac | 442 | struct nilfs_write_info wi; |
64b5a32e | 443 | struct buffer_head *bh; |
5f1586d0 | 444 | int res = 0, rw = WRITE; |
64b5a32e | 445 | |
9c965bac RK |
446 | wi.nilfs = nilfs; |
447 | nilfs_segbuf_prepare_write(segbuf, &wi); | |
448 | ||
64b5a32e | 449 | list_for_each_entry(bh, &segbuf->sb_segsum_buffers, b_assoc_buffers) { |
9c965bac | 450 | res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw); |
64b5a32e RK |
451 | if (unlikely(res)) |
452 | goto failed_bio; | |
453 | } | |
454 | ||
455 | list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) { | |
9c965bac | 456 | res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw); |
64b5a32e RK |
457 | if (unlikely(res)) |
458 | goto failed_bio; | |
459 | } | |
460 | ||
9c965bac | 461 | if (wi.bio) { |
64b5a32e RK |
462 | /* |
463 | * Last BIO is always sent through the following | |
464 | * submission. | |
465 | */ | |
30bda0b8 | 466 | rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG); |
9c965bac | 467 | res = nilfs_segbuf_submit_bio(segbuf, &wi, rw); |
64b5a32e RK |
468 | } |
469 | ||
64b5a32e | 470 | failed_bio: |
5f1586d0 | 471 | return res; |
64b5a32e RK |
472 | } |
473 | ||
474 | /** | |
475 | * nilfs_segbuf_wait - wait for completion of requested BIOs | |
9284ad2a | 476 | * @segbuf: segment buffer |
64b5a32e RK |
477 | * |
478 | * Return Value: On Success, 0 is returned. On Error, one of the following | |
479 | * negative error code is returned. | |
480 | * | |
481 | * %-EIO - I/O error | |
482 | */ | |
9284ad2a | 483 | int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf) |
64b5a32e RK |
484 | { |
485 | int err = 0; | |
486 | ||
9284ad2a | 487 | if (!segbuf->sb_nbio) |
64b5a32e RK |
488 | return 0; |
489 | ||
490 | do { | |
9284ad2a RK |
491 | wait_for_completion(&segbuf->sb_bio_event); |
492 | } while (--segbuf->sb_nbio > 0); | |
64b5a32e | 493 | |
9284ad2a | 494 | if (unlikely(atomic_read(&segbuf->sb_err) > 0)) { |
64b5a32e RK |
495 | printk(KERN_ERR "NILFS: IO error writing segment\n"); |
496 | err = -EIO; | |
64b5a32e RK |
497 | } |
498 | return err; | |
499 | } |