Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes...
[linux-2.6.git] / fs / nilfs2 / segbuf.c
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>
27 #include <linux/backing-dev.h>
28 #include "page.h"
29 #include "segbuf.h"
30
31
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
43 static int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
44                               struct the_nilfs *nilfs);
45 static int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf);
46
47
48 static struct kmem_cache *nilfs_segbuf_cachep;
49
50 static void nilfs_segbuf_init_once(void *obj)
51 {
52         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
53 }
54
55 int __init nilfs_init_segbuf_cache(void)
56 {
57         nilfs_segbuf_cachep =
58                 kmem_cache_create("nilfs2_segbuf_cache",
59                                   sizeof(struct nilfs_segment_buffer),
60                                   0, SLAB_RECLAIM_ACCOUNT,
61                                   nilfs_segbuf_init_once);
62
63         return (nilfs_segbuf_cachep == NULL) ? -ENOMEM : 0;
64 }
65
66 void nilfs_destroy_segbuf_cache(void)
67 {
68         kmem_cache_destroy(nilfs_segbuf_cachep);
69 }
70
71 struct nilfs_segment_buffer *nilfs_segbuf_new(struct super_block *sb)
72 {
73         struct nilfs_segment_buffer *segbuf;
74
75         segbuf = kmem_cache_alloc(nilfs_segbuf_cachep, GFP_NOFS);
76         if (unlikely(!segbuf))
77                 return NULL;
78
79         segbuf->sb_super = sb;
80         INIT_LIST_HEAD(&segbuf->sb_list);
81         INIT_LIST_HEAD(&segbuf->sb_segsum_buffers);
82         INIT_LIST_HEAD(&segbuf->sb_payload_buffers);
83
84         init_completion(&segbuf->sb_bio_event);
85         atomic_set(&segbuf->sb_err, 0);
86         segbuf->sb_nbio = 0;
87
88         return segbuf;
89 }
90
91 void nilfs_segbuf_free(struct nilfs_segment_buffer *segbuf)
92 {
93         kmem_cache_free(nilfs_segbuf_cachep, segbuf);
94 }
95
96 void nilfs_segbuf_map(struct nilfs_segment_buffer *segbuf, __u64 segnum,
97                      unsigned long offset, struct the_nilfs *nilfs)
98 {
99         segbuf->sb_segnum = segnum;
100         nilfs_get_segment_range(nilfs, segnum, &segbuf->sb_fseg_start,
101                                 &segbuf->sb_fseg_end);
102
103         segbuf->sb_pseg_start = segbuf->sb_fseg_start + offset;
104         segbuf->sb_rest_blocks =
105                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
106 }
107
108 /**
109  * nilfs_segbuf_map_cont - map a new log behind a given log
110  * @segbuf: new segment buffer
111  * @prev: segment buffer containing a log to be continued
112  */
113 void nilfs_segbuf_map_cont(struct nilfs_segment_buffer *segbuf,
114                            struct nilfs_segment_buffer *prev)
115 {
116         segbuf->sb_segnum = prev->sb_segnum;
117         segbuf->sb_fseg_start = prev->sb_fseg_start;
118         segbuf->sb_fseg_end = prev->sb_fseg_end;
119         segbuf->sb_pseg_start = prev->sb_pseg_start + prev->sb_sum.nblocks;
120         segbuf->sb_rest_blocks =
121                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
122 }
123
124 void nilfs_segbuf_set_next_segnum(struct nilfs_segment_buffer *segbuf,
125                                   __u64 nextnum, struct the_nilfs *nilfs)
126 {
127         segbuf->sb_nextnum = nextnum;
128         segbuf->sb_sum.next = nilfs_get_segment_start_blocknr(nilfs, nextnum);
129 }
130
131 int nilfs_segbuf_extend_segsum(struct nilfs_segment_buffer *segbuf)
132 {
133         struct buffer_head *bh;
134
135         bh = sb_getblk(segbuf->sb_super,
136                        segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk);
137         if (unlikely(!bh))
138                 return -ENOMEM;
139
140         nilfs_segbuf_add_segsum_buffer(segbuf, bh);
141         return 0;
142 }
143
144 int nilfs_segbuf_extend_payload(struct nilfs_segment_buffer *segbuf,
145                                 struct buffer_head **bhp)
146 {
147         struct buffer_head *bh;
148
149         bh = sb_getblk(segbuf->sb_super,
150                        segbuf->sb_pseg_start + segbuf->sb_sum.nblocks);
151         if (unlikely(!bh))
152                 return -ENOMEM;
153
154         nilfs_segbuf_add_payload_buffer(segbuf, bh);
155         *bhp = bh;
156         return 0;
157 }
158
159 int nilfs_segbuf_reset(struct nilfs_segment_buffer *segbuf, unsigned flags,
160                        time_t ctime)
161 {
162         int err;
163
164         segbuf->sb_sum.nblocks = segbuf->sb_sum.nsumblk = 0;
165         err = nilfs_segbuf_extend_segsum(segbuf);
166         if (unlikely(err))
167                 return err;
168
169         segbuf->sb_sum.flags = flags;
170         segbuf->sb_sum.sumbytes = sizeof(struct nilfs_segment_summary);
171         segbuf->sb_sum.nfinfo = segbuf->sb_sum.nfileblk = 0;
172         segbuf->sb_sum.ctime = ctime;
173         return 0;
174 }
175
176 /*
177  * Setup segment summary
178  */
179 void nilfs_segbuf_fill_in_segsum(struct nilfs_segment_buffer *segbuf)
180 {
181         struct nilfs_segment_summary *raw_sum;
182         struct buffer_head *bh_sum;
183
184         bh_sum = list_entry(segbuf->sb_segsum_buffers.next,
185                             struct buffer_head, b_assoc_buffers);
186         raw_sum = (struct nilfs_segment_summary *)bh_sum->b_data;
187
188         raw_sum->ss_magic    = cpu_to_le32(NILFS_SEGSUM_MAGIC);
189         raw_sum->ss_bytes    = cpu_to_le16(sizeof(*raw_sum));
190         raw_sum->ss_flags    = cpu_to_le16(segbuf->sb_sum.flags);
191         raw_sum->ss_seq      = cpu_to_le64(segbuf->sb_sum.seg_seq);
192         raw_sum->ss_create   = cpu_to_le64(segbuf->sb_sum.ctime);
193         raw_sum->ss_next     = cpu_to_le64(segbuf->sb_sum.next);
194         raw_sum->ss_nblocks  = cpu_to_le32(segbuf->sb_sum.nblocks);
195         raw_sum->ss_nfinfo   = cpu_to_le32(segbuf->sb_sum.nfinfo);
196         raw_sum->ss_sumbytes = cpu_to_le32(segbuf->sb_sum.sumbytes);
197         raw_sum->ss_pad      = 0;
198 }
199
200 /*
201  * CRC calculation routines
202  */
203 void nilfs_segbuf_fill_in_segsum_crc(struct nilfs_segment_buffer *segbuf,
204                                      u32 seed)
205 {
206         struct buffer_head *bh;
207         struct nilfs_segment_summary *raw_sum;
208         unsigned long size, bytes = segbuf->sb_sum.sumbytes;
209         u32 crc;
210
211         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
212                         b_assoc_buffers);
213
214         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
215         size = min_t(unsigned long, bytes, bh->b_size);
216         crc = crc32_le(seed,
217                        (unsigned char *)raw_sum +
218                        sizeof(raw_sum->ss_datasum) + sizeof(raw_sum->ss_sumsum),
219                        size - (sizeof(raw_sum->ss_datasum) +
220                                sizeof(raw_sum->ss_sumsum)));
221
222         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
223                                      b_assoc_buffers) {
224                 bytes -= size;
225                 size = min_t(unsigned long, bytes, bh->b_size);
226                 crc = crc32_le(crc, bh->b_data, size);
227         }
228         raw_sum->ss_sumsum = cpu_to_le32(crc);
229 }
230
231 void nilfs_segbuf_fill_in_data_crc(struct nilfs_segment_buffer *segbuf,
232                                    u32 seed)
233 {
234         struct buffer_head *bh;
235         struct nilfs_segment_summary *raw_sum;
236         void *kaddr;
237         u32 crc;
238
239         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
240                         b_assoc_buffers);
241         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
242         crc = crc32_le(seed,
243                        (unsigned char *)raw_sum + sizeof(raw_sum->ss_datasum),
244                        bh->b_size - sizeof(raw_sum->ss_datasum));
245
246         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
247                                      b_assoc_buffers) {
248                 crc = crc32_le(crc, bh->b_data, bh->b_size);
249         }
250         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
251                 kaddr = kmap_atomic(bh->b_page, KM_USER0);
252                 crc = crc32_le(crc, kaddr + bh_offset(bh), bh->b_size);
253                 kunmap_atomic(kaddr, KM_USER0);
254         }
255         raw_sum->ss_datasum = cpu_to_le32(crc);
256 }
257
258 static void nilfs_release_buffers(struct list_head *list)
259 {
260         struct buffer_head *bh, *n;
261
262         list_for_each_entry_safe(bh, n, list, b_assoc_buffers) {
263                 list_del_init(&bh->b_assoc_buffers);
264                 if (buffer_nilfs_allocated(bh)) {
265                         struct page *clone_page = bh->b_page;
266
267                         /* remove clone page */
268                         brelse(bh);
269                         page_cache_release(clone_page); /* for each bh */
270                         if (page_count(clone_page) <= 2) {
271                                 lock_page(clone_page);
272                                 nilfs_free_private_page(clone_page);
273                         }
274                         continue;
275                 }
276                 brelse(bh);
277         }
278 }
279
280 static void nilfs_segbuf_clear(struct nilfs_segment_buffer *segbuf)
281 {
282         nilfs_release_buffers(&segbuf->sb_segsum_buffers);
283         nilfs_release_buffers(&segbuf->sb_payload_buffers);
284 }
285
286 /*
287  * Iterators for segment buffers
288  */
289 void nilfs_clear_logs(struct list_head *logs)
290 {
291         struct nilfs_segment_buffer *segbuf;
292
293         list_for_each_entry(segbuf, logs, sb_list)
294                 nilfs_segbuf_clear(segbuf);
295 }
296
297 void nilfs_truncate_logs(struct list_head *logs,
298                          struct nilfs_segment_buffer *last)
299 {
300         struct nilfs_segment_buffer *n, *segbuf;
301
302         segbuf = list_prepare_entry(last, logs, sb_list);
303         list_for_each_entry_safe_continue(segbuf, n, logs, sb_list) {
304                 list_del_init(&segbuf->sb_list);
305                 nilfs_segbuf_clear(segbuf);
306                 nilfs_segbuf_free(segbuf);
307         }
308 }
309
310 int nilfs_write_logs(struct list_head *logs, struct the_nilfs *nilfs)
311 {
312         struct nilfs_segment_buffer *segbuf;
313         int ret = 0;
314
315         list_for_each_entry(segbuf, logs, sb_list) {
316                 ret = nilfs_segbuf_write(segbuf, nilfs);
317                 if (ret)
318                         break;
319         }
320         return ret;
321 }
322
323 int nilfs_wait_on_logs(struct list_head *logs)
324 {
325         struct nilfs_segment_buffer *segbuf;
326         int err, ret = 0;
327
328         list_for_each_entry(segbuf, logs, sb_list) {
329                 err = nilfs_segbuf_wait(segbuf);
330                 if (err && !ret)
331                         ret = err;
332         }
333         return ret;
334 }
335
336 /*
337  * BIO operations
338  */
339 static void nilfs_end_bio_write(struct bio *bio, int err)
340 {
341         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
342         struct nilfs_segment_buffer *segbuf = bio->bi_private;
343
344         if (err == -EOPNOTSUPP) {
345                 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
346                 bio_put(bio);
347                 /* to be detected by submit_seg_bio() */
348         }
349
350         if (!uptodate)
351                 atomic_inc(&segbuf->sb_err);
352
353         bio_put(bio);
354         complete(&segbuf->sb_bio_event);
355 }
356
357 static int nilfs_segbuf_submit_bio(struct nilfs_segment_buffer *segbuf,
358                                    struct nilfs_write_info *wi, int mode)
359 {
360         struct bio *bio = wi->bio;
361         int err;
362
363         if (segbuf->sb_nbio > 0 && bdi_write_congested(wi->nilfs->ns_bdi)) {
364                 wait_for_completion(&segbuf->sb_bio_event);
365                 segbuf->sb_nbio--;
366                 if (unlikely(atomic_read(&segbuf->sb_err))) {
367                         bio_put(bio);
368                         err = -EIO;
369                         goto failed;
370                 }
371         }
372
373         bio->bi_end_io = nilfs_end_bio_write;
374         bio->bi_private = segbuf;
375         bio_get(bio);
376         submit_bio(mode, bio);
377         if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
378                 bio_put(bio);
379                 err = -EOPNOTSUPP;
380                 goto failed;
381         }
382         segbuf->sb_nbio++;
383         bio_put(bio);
384
385         wi->bio = NULL;
386         wi->rest_blocks -= wi->end - wi->start;
387         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
388         wi->start = wi->end;
389         return 0;
390
391  failed:
392         wi->bio = NULL;
393         return err;
394 }
395
396 /**
397  * nilfs_alloc_seg_bio - allocate a new bio for writing log
398  * @nilfs: nilfs object
399  * @start: start block number of the bio
400  * @nr_vecs: request size of page vector.
401  *
402  * Return Value: On success, pointer to the struct bio is returned.
403  * On error, NULL is returned.
404  */
405 static struct bio *nilfs_alloc_seg_bio(struct the_nilfs *nilfs, sector_t start,
406                                        int nr_vecs)
407 {
408         struct bio *bio;
409
410         bio = bio_alloc(GFP_NOIO, nr_vecs);
411         if (bio == NULL) {
412                 while (!bio && (nr_vecs >>= 1))
413                         bio = bio_alloc(GFP_NOIO, nr_vecs);
414         }
415         if (likely(bio)) {
416                 bio->bi_bdev = nilfs->ns_bdev;
417                 bio->bi_sector = start << (nilfs->ns_blocksize_bits - 9);
418         }
419         return bio;
420 }
421
422 static void nilfs_segbuf_prepare_write(struct nilfs_segment_buffer *segbuf,
423                                        struct nilfs_write_info *wi)
424 {
425         wi->bio = NULL;
426         wi->rest_blocks = segbuf->sb_sum.nblocks;
427         wi->max_pages = bio_get_nr_vecs(wi->nilfs->ns_bdev);
428         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
429         wi->start = wi->end = 0;
430         wi->blocknr = segbuf->sb_pseg_start;
431 }
432
433 static int nilfs_segbuf_submit_bh(struct nilfs_segment_buffer *segbuf,
434                                   struct nilfs_write_info *wi,
435                                   struct buffer_head *bh, int mode)
436 {
437         int len, err;
438
439         BUG_ON(wi->nr_vecs <= 0);
440  repeat:
441         if (!wi->bio) {
442                 wi->bio = nilfs_alloc_seg_bio(wi->nilfs, wi->blocknr + wi->end,
443                                               wi->nr_vecs);
444                 if (unlikely(!wi->bio))
445                         return -ENOMEM;
446         }
447
448         len = bio_add_page(wi->bio, bh->b_page, bh->b_size, bh_offset(bh));
449         if (len == bh->b_size) {
450                 wi->end++;
451                 return 0;
452         }
453         /* bio is FULL */
454         err = nilfs_segbuf_submit_bio(segbuf, wi, mode);
455         /* never submit current bh */
456         if (likely(!err))
457                 goto repeat;
458         return err;
459 }
460
461 /**
462  * nilfs_segbuf_write - submit write requests of a log
463  * @segbuf: buffer storing a log to be written
464  * @nilfs: nilfs object
465  *
466  * Return Value: On Success, 0 is returned. On Error, one of the following
467  * negative error code is returned.
468  *
469  * %-EIO - I/O error
470  *
471  * %-ENOMEM - Insufficient memory available.
472  */
473 static int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
474                               struct the_nilfs *nilfs)
475 {
476         struct nilfs_write_info wi;
477         struct buffer_head *bh;
478         int res = 0, rw = WRITE;
479
480         wi.nilfs = nilfs;
481         nilfs_segbuf_prepare_write(segbuf, &wi);
482
483         list_for_each_entry(bh, &segbuf->sb_segsum_buffers, b_assoc_buffers) {
484                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
485                 if (unlikely(res))
486                         goto failed_bio;
487         }
488
489         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
490                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
491                 if (unlikely(res))
492                         goto failed_bio;
493         }
494
495         if (wi.bio) {
496                 /*
497                  * Last BIO is always sent through the following
498                  * submission.
499                  */
500                 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
501                 res = nilfs_segbuf_submit_bio(segbuf, &wi, rw);
502         }
503
504  failed_bio:
505         return res;
506 }
507
508 /**
509  * nilfs_segbuf_wait - wait for completion of requested BIOs
510  * @segbuf: segment buffer
511  *
512  * Return Value: On Success, 0 is returned. On Error, one of the following
513  * negative error code is returned.
514  *
515  * %-EIO - I/O error
516  */
517 static int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf)
518 {
519         int err = 0;
520
521         if (!segbuf->sb_nbio)
522                 return 0;
523
524         do {
525                 wait_for_completion(&segbuf->sb_bio_event);
526         } while (--segbuf->sb_nbio > 0);
527
528         if (unlikely(atomic_read(&segbuf->sb_err) > 0)) {
529                 printk(KERN_ERR "NILFS: IO error writing segment\n");
530                 err = -EIO;
531         }
532         return err;
533 }