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