blob: 3d3ea8351f6d13a1037fa24f14a4e2b8617a679a [file] [log] [blame]
Ryusuke Konishi64b5a322009-04-06 19:01:36 -07001/*
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>
27#include "page.h"
28#include "segbuf.h"
29#include "seglist.h"
30
31
32static struct kmem_cache *nilfs_segbuf_cachep;
33
34static void nilfs_segbuf_init_once(void *obj)
35{
36 memset(obj, 0, sizeof(struct nilfs_segment_buffer));
37}
38
39int __init nilfs_init_segbuf_cache(void)
40{
41 nilfs_segbuf_cachep =
42 kmem_cache_create("nilfs2_segbuf_cache",
43 sizeof(struct nilfs_segment_buffer),
44 0, SLAB_RECLAIM_ACCOUNT,
45 nilfs_segbuf_init_once);
46
47 return (nilfs_segbuf_cachep == NULL) ? -ENOMEM : 0;
48}
49
50void nilfs_destroy_segbuf_cache(void)
51{
52 kmem_cache_destroy(nilfs_segbuf_cachep);
53}
54
55struct nilfs_segment_buffer *nilfs_segbuf_new(struct super_block *sb)
56{
57 struct nilfs_segment_buffer *segbuf;
58
59 segbuf = kmem_cache_alloc(nilfs_segbuf_cachep, GFP_NOFS);
60 if (unlikely(!segbuf))
61 return NULL;
62
63 segbuf->sb_super = sb;
64 INIT_LIST_HEAD(&segbuf->sb_list);
65 INIT_LIST_HEAD(&segbuf->sb_segsum_buffers);
66 INIT_LIST_HEAD(&segbuf->sb_payload_buffers);
67 segbuf->sb_segent = NULL;
68 return segbuf;
69}
70
71void nilfs_segbuf_free(struct nilfs_segment_buffer *segbuf)
72{
73 struct nilfs_segment_entry *ent = segbuf->sb_segent;
74
75 if (ent != NULL && list_empty(&ent->list)) {
76 /* free isolated segment list head */
77 nilfs_free_segment_entry(segbuf->sb_segent);
78 segbuf->sb_segent = NULL;
79 }
80 kmem_cache_free(nilfs_segbuf_cachep, segbuf);
81}
82
83int nilfs_segbuf_map(struct nilfs_segment_buffer *segbuf, __u64 segnum,
84 unsigned long offset, struct the_nilfs *nilfs)
85{
86 struct nilfs_segment_entry *ent;
87
88 segbuf->sb_segnum = segnum;
89 nilfs_get_segment_range(nilfs, segnum, &segbuf->sb_fseg_start,
90 &segbuf->sb_fseg_end);
91
92 segbuf->sb_pseg_start = segbuf->sb_fseg_start + offset;
93 segbuf->sb_rest_blocks =
94 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
95
96 /* Attach a segment list head */
97 ent = segbuf->sb_segent;
98 if (ent == NULL) {
99 segbuf->sb_segent = nilfs_alloc_segment_entry(segnum);
100 if (unlikely(!segbuf->sb_segent))
101 return -ENOMEM;
102 } else {
103 BUG_ON(ent->bh_su || !list_empty(&ent->list));
104 ent->segnum = segnum;
105 }
106 return 0;
107}
108
109void nilfs_segbuf_set_next_segnum(struct nilfs_segment_buffer *segbuf,
110 __u64 nextnum, struct the_nilfs *nilfs)
111{
112 segbuf->sb_nextnum = nextnum;
113 segbuf->sb_sum.next = nilfs_get_segment_start_blocknr(nilfs, nextnum);
114}
115
116int nilfs_segbuf_extend_segsum(struct nilfs_segment_buffer *segbuf)
117{
118 struct buffer_head *bh;
119
120 bh = sb_getblk(segbuf->sb_super,
121 segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk);
122 if (unlikely(!bh))
123 return -ENOMEM;
124
125 nilfs_segbuf_add_segsum_buffer(segbuf, bh);
126 return 0;
127}
128
129int nilfs_segbuf_extend_payload(struct nilfs_segment_buffer *segbuf,
130 struct buffer_head **bhp)
131{
132 struct buffer_head *bh;
133
134 bh = sb_getblk(segbuf->sb_super,
135 segbuf->sb_pseg_start + segbuf->sb_sum.nblocks);
136 if (unlikely(!bh))
137 return -ENOMEM;
138
139 nilfs_segbuf_add_payload_buffer(segbuf, bh);
140 *bhp = bh;
141 return 0;
142}
143
144int nilfs_segbuf_reset(struct nilfs_segment_buffer *segbuf, unsigned flags,
145 time_t ctime)
146{
147 int err;
148
149 segbuf->sb_sum.nblocks = segbuf->sb_sum.nsumblk = 0;
150 err = nilfs_segbuf_extend_segsum(segbuf);
151 if (unlikely(err))
152 return err;
153
154 segbuf->sb_sum.flags = flags;
155 segbuf->sb_sum.sumbytes = sizeof(struct nilfs_segment_summary);
156 segbuf->sb_sum.nfinfo = segbuf->sb_sum.nfileblk = 0;
157 segbuf->sb_sum.ctime = ctime;
158
159 segbuf->sb_io_error = 0;
160 return 0;
161}
162
163/*
164 * Setup segument summary
165 */
166void nilfs_segbuf_fill_in_segsum(struct nilfs_segment_buffer *segbuf)
167{
168 struct nilfs_segment_summary *raw_sum;
169 struct buffer_head *bh_sum;
170
171 bh_sum = list_entry(segbuf->sb_segsum_buffers.next,
172 struct buffer_head, b_assoc_buffers);
173 raw_sum = (struct nilfs_segment_summary *)bh_sum->b_data;
174
175 raw_sum->ss_magic = cpu_to_le32(NILFS_SEGSUM_MAGIC);
176 raw_sum->ss_bytes = cpu_to_le16(sizeof(*raw_sum));
177 raw_sum->ss_flags = cpu_to_le16(segbuf->sb_sum.flags);
178 raw_sum->ss_seq = cpu_to_le64(segbuf->sb_sum.seg_seq);
179 raw_sum->ss_create = cpu_to_le64(segbuf->sb_sum.ctime);
180 raw_sum->ss_next = cpu_to_le64(segbuf->sb_sum.next);
181 raw_sum->ss_nblocks = cpu_to_le32(segbuf->sb_sum.nblocks);
182 raw_sum->ss_nfinfo = cpu_to_le32(segbuf->sb_sum.nfinfo);
183 raw_sum->ss_sumbytes = cpu_to_le32(segbuf->sb_sum.sumbytes);
184 raw_sum->ss_pad = 0;
185}
186
187/*
188 * CRC calculation routines
189 */
190void nilfs_segbuf_fill_in_segsum_crc(struct nilfs_segment_buffer *segbuf,
191 u32 seed)
192{
193 struct buffer_head *bh;
194 struct nilfs_segment_summary *raw_sum;
195 unsigned long size, bytes = segbuf->sb_sum.sumbytes;
196 u32 crc;
197
198 bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
199 b_assoc_buffers);
200
201 raw_sum = (struct nilfs_segment_summary *)bh->b_data;
202 size = min_t(unsigned long, bytes, bh->b_size);
203 crc = crc32_le(seed,
204 (unsigned char *)raw_sum +
205 sizeof(raw_sum->ss_datasum) + sizeof(raw_sum->ss_sumsum),
206 size - (sizeof(raw_sum->ss_datasum) +
207 sizeof(raw_sum->ss_sumsum)));
208
209 list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
210 b_assoc_buffers) {
211 bytes -= size;
212 size = min_t(unsigned long, bytes, bh->b_size);
213 crc = crc32_le(crc, bh->b_data, size);
214 }
215 raw_sum->ss_sumsum = cpu_to_le32(crc);
216}
217
218void nilfs_segbuf_fill_in_data_crc(struct nilfs_segment_buffer *segbuf,
219 u32 seed)
220{
221 struct buffer_head *bh;
222 struct nilfs_segment_summary *raw_sum;
223 void *kaddr;
224 u32 crc;
225
226 bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
227 b_assoc_buffers);
228 raw_sum = (struct nilfs_segment_summary *)bh->b_data;
229 crc = crc32_le(seed,
230 (unsigned char *)raw_sum + sizeof(raw_sum->ss_datasum),
231 bh->b_size - sizeof(raw_sum->ss_datasum));
232
233 list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
234 b_assoc_buffers) {
235 crc = crc32_le(crc, bh->b_data, bh->b_size);
236 }
237 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
238 kaddr = kmap_atomic(bh->b_page, KM_USER0);
239 crc = crc32_le(crc, kaddr + bh_offset(bh), bh->b_size);
240 kunmap_atomic(kaddr, KM_USER0);
241 }
242 raw_sum->ss_datasum = cpu_to_le32(crc);
243}
244
245void nilfs_release_buffers(struct list_head *list)
246{
247 struct buffer_head *bh, *n;
248
249 list_for_each_entry_safe(bh, n, list, b_assoc_buffers) {
250 list_del_init(&bh->b_assoc_buffers);
251 if (buffer_nilfs_allocated(bh)) {
252 struct page *clone_page = bh->b_page;
253
254 /* remove clone page */
255 brelse(bh);
256 page_cache_release(clone_page); /* for each bh */
257 if (page_count(clone_page) <= 2) {
258 lock_page(clone_page);
259 nilfs_free_private_page(clone_page);
260 }
261 continue;
262 }
263 brelse(bh);
264 }
265}
266
267/*
268 * BIO operations
269 */
270static void nilfs_end_bio_write(struct bio *bio, int err)
271{
272 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
273 struct nilfs_write_info *wi = bio->bi_private;
274
275 if (err == -EOPNOTSUPP) {
276 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
277 bio_put(bio);
278 /* to be detected by submit_seg_bio() */
279 }
280
281 if (!uptodate)
282 atomic_inc(&wi->err);
283
284 bio_put(bio);
285 complete(&wi->bio_event);
286}
287
288static int nilfs_submit_seg_bio(struct nilfs_write_info *wi, int mode)
289{
290 struct bio *bio = wi->bio;
291 int err;
292
293 if (wi->nbio > 0 && bdi_write_congested(wi->bdi)) {
294 wait_for_completion(&wi->bio_event);
295 wi->nbio--;
296 if (unlikely(atomic_read(&wi->err))) {
297 bio_put(bio);
298 err = -EIO;
299 goto failed;
300 }
301 }
302
303 bio->bi_end_io = nilfs_end_bio_write;
304 bio->bi_private = wi;
305 bio_get(bio);
306 submit_bio(mode, bio);
307 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
308 bio_put(bio);
309 err = -EOPNOTSUPP;
310 goto failed;
311 }
312 wi->nbio++;
313 bio_put(bio);
314
315 wi->bio = NULL;
316 wi->rest_blocks -= wi->end - wi->start;
317 wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
318 wi->start = wi->end;
319 return 0;
320
321 failed:
322 wi->bio = NULL;
323 return err;
324}
325
326/**
327 * nilfs_alloc_seg_bio - allocate a bio for writing segment.
328 * @sb: super block
329 * @start: beginning disk block number of this BIO.
330 * @nr_vecs: request size of page vector.
331 *
332 * alloc_seg_bio() allocates a new BIO structure and initialize it.
333 *
334 * Return Value: On success, pointer to the struct bio is returned.
335 * On error, NULL is returned.
336 */
337static struct bio *nilfs_alloc_seg_bio(struct super_block *sb, sector_t start,
338 int nr_vecs)
339{
340 struct bio *bio;
341
342 bio = bio_alloc(GFP_NOWAIT, nr_vecs);
343 if (bio == NULL) {
344 while (!bio && (nr_vecs >>= 1))
345 bio = bio_alloc(GFP_NOWAIT, nr_vecs);
346 }
347 if (likely(bio)) {
348 bio->bi_bdev = sb->s_bdev;
349 bio->bi_sector = (sector_t)start << (sb->s_blocksize_bits - 9);
350 }
351 return bio;
352}
353
354void nilfs_segbuf_prepare_write(struct nilfs_segment_buffer *segbuf,
355 struct nilfs_write_info *wi)
356{
357 wi->bio = NULL;
358 wi->rest_blocks = segbuf->sb_sum.nblocks;
359 wi->max_pages = bio_get_nr_vecs(wi->sb->s_bdev);
360 wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
361 wi->start = wi->end = 0;
362 wi->nbio = 0;
363 wi->blocknr = segbuf->sb_pseg_start;
364
365 atomic_set(&wi->err, 0);
366 init_completion(&wi->bio_event);
367}
368
369static int nilfs_submit_bh(struct nilfs_write_info *wi, struct buffer_head *bh,
370 int mode)
371{
372 int len, err;
373
374 BUG_ON(wi->nr_vecs <= 0);
375 repeat:
376 if (!wi->bio) {
377 wi->bio = nilfs_alloc_seg_bio(wi->sb, wi->blocknr + wi->end,
378 wi->nr_vecs);
379 if (unlikely(!wi->bio))
380 return -ENOMEM;
381 }
382
383 len = bio_add_page(wi->bio, bh->b_page, bh->b_size, bh_offset(bh));
384 if (len == bh->b_size) {
385 wi->end++;
386 return 0;
387 }
388 /* bio is FULL */
389 err = nilfs_submit_seg_bio(wi, mode);
390 /* never submit current bh */
391 if (likely(!err))
392 goto repeat;
393 return err;
394}
395
396int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
397 struct nilfs_write_info *wi)
398{
399 struct buffer_head *bh;
400 int res, rw = WRITE;
401
402 list_for_each_entry(bh, &segbuf->sb_segsum_buffers, b_assoc_buffers) {
403 res = nilfs_submit_bh(wi, bh, rw);
404 if (unlikely(res))
405 goto failed_bio;
406 }
407
408 list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
409 res = nilfs_submit_bh(wi, bh, rw);
410 if (unlikely(res))
411 goto failed_bio;
412 }
413
414 if (wi->bio) {
415 /*
416 * Last BIO is always sent through the following
417 * submission.
418 */
419 rw |= (1 << BIO_RW_SYNCIO);
420 res = nilfs_submit_seg_bio(wi, rw);
421 if (unlikely(res))
422 goto failed_bio;
423 }
424
425 res = 0;
426 out:
427 return res;
428
429 failed_bio:
430 atomic_inc(&wi->err);
431 goto out;
432}
433
434/**
435 * nilfs_segbuf_wait - wait for completion of requested BIOs
436 * @wi: nilfs_write_info
437 *
438 * Return Value: On Success, 0 is returned. On Error, one of the following
439 * negative error code is returned.
440 *
441 * %-EIO - I/O error
442 */
443int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf,
444 struct nilfs_write_info *wi)
445{
446 int err = 0;
447
448 if (!wi->nbio)
449 return 0;
450
451 do {
452 wait_for_completion(&wi->bio_event);
453 } while (--wi->nbio > 0);
454
455 if (unlikely(atomic_read(&wi->err) > 0)) {
456 printk(KERN_ERR "NILFS: IO error writing segment\n");
457 err = -EIO;
458 segbuf->sb_io_error = 1;
459 }
460 return err;
461}