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 <linux/slab.h>
  29 #include "page.h"
  30 #include "segbuf.h"
  31
  32
  33 struct nilfs_write_info {
  34         struct the_nilfs       *nilfs;
  35         struct bio             *bio;
  36         int                     start, end; /* The region to be submitted */
  37         int                     rest_blocks;
  38         int                     max_pages;
  39         int                     nr_vecs;
  40         sector_t                blocknr;
  41 };
  42
  43
  44 static int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
  45                               struct the_nilfs *nilfs);
  46 static int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf);
  47
  48
  49 static struct kmem_cache *nilfs_segbuf_cachep;
  50
  51 static void nilfs_segbuf_init_once(void *obj)
  52 {
  53         memset(obj, 0, sizeof(struct nilfs_segment_buffer));
  54 }
  55
  56 int __init nilfs_init_segbuf_cache(void)
  57 {
  58         nilfs_segbuf_cachep =
  59                 kmem_cache_create("nilfs2_segbuf_cache",
  60                                   sizeof(struct nilfs_segment_buffer),
  61                                   0, SLAB_RECLAIM_ACCOUNT,
  62                                   nilfs_segbuf_init_once);
  63
  64         return (nilfs_segbuf_cachep == NULL) ? -ENOMEM : 0;
  65 }
  66
  67 void nilfs_destroy_segbuf_cache(void)
  68 {
  69         kmem_cache_destroy(nilfs_segbuf_cachep);
  70 }
  71
  72 struct nilfs_segment_buffer *nilfs_segbuf_new(struct super_block *sb)
  73 {
  74         struct nilfs_segment_buffer *segbuf;
  75
  76         segbuf = kmem_cache_alloc(nilfs_segbuf_cachep, GFP_NOFS);
  77         if (unlikely(!segbuf))
  78                 return NULL;
  79
  80         segbuf->sb_super = sb;
  81         INIT_LIST_HEAD(&segbuf->sb_list);
  82         INIT_LIST_HEAD(&segbuf->sb_segsum_buffers);
  83         INIT_LIST_HEAD(&segbuf->sb_payload_buffers);
  84
  85         init_completion(&segbuf->sb_bio_event);
  86         atomic_set(&segbuf->sb_err, 0);
  87         segbuf->sb_nbio = 0;
  88
  89         return segbuf;
  90 }
  91
  92 void nilfs_segbuf_free(struct nilfs_segment_buffer *segbuf)
  93 {
  94         kmem_cache_free(nilfs_segbuf_cachep, segbuf);
  95 }
  96
  97 void nilfs_segbuf_map(struct nilfs_segment_buffer *segbuf, __u64 segnum,
  98                      unsigned long offset, struct the_nilfs *nilfs)
  99 {
 100         segbuf->sb_segnum = segnum;
 101         nilfs_get_segment_range(nilfs, segnum, &segbuf->sb_fseg_start,
 102                                 &segbuf->sb_fseg_end);
 103
 104         segbuf->sb_pseg_start = segbuf->sb_fseg_start + offset;
 105         segbuf->sb_rest_blocks =
 106                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
 107 }
 108
 109 /**
 110  * nilfs_segbuf_map_cont - map a new log behind a given log
 111  * @segbuf: new segment buffer
 112  * @prev: segment buffer containing a log to be continued
 113  */
 114 void nilfs_segbuf_map_cont(struct nilfs_segment_buffer *segbuf,
 115                            struct nilfs_segment_buffer *prev)
 116 {
 117         segbuf->sb_segnum = prev->sb_segnum;
 118         segbuf->sb_fseg_start = prev->sb_fseg_start;
 119         segbuf->sb_fseg_end = prev->sb_fseg_end;
 120         segbuf->sb_pseg_start = prev->sb_pseg_start + prev->sb_sum.nblocks;
 121         segbuf->sb_rest_blocks =
 122                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
 123 }
 124
 125 void nilfs_segbuf_set_next_segnum(struct nilfs_segment_buffer *segbuf,
 126                                   __u64 nextnum, struct the_nilfs *nilfs)
 127 {
 128         segbuf->sb_nextnum = nextnum;
 129         segbuf->sb_sum.next = nilfs_get_segment_start_blocknr(nilfs, nextnum);
 130 }
 131
 132 int nilfs_segbuf_extend_segsum(struct nilfs_segment_buffer *segbuf)
 133 {
 134         struct buffer_head *bh;
 135
 136         bh = sb_getblk(segbuf->sb_super,
 137                        segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk);
 138         if (unlikely(!bh))
 139                 return -ENOMEM;
 140
 141         nilfs_segbuf_add_segsum_buffer(segbuf, bh);
 142         return 0;
 143 }
 144
 145 int nilfs_segbuf_extend_payload(struct nilfs_segment_buffer *segbuf,
 146                                 struct buffer_head **bhp)
 147 {
 148         struct buffer_head *bh;
 149
 150         bh = sb_getblk(segbuf->sb_super,
 151                        segbuf->sb_pseg_start + segbuf->sb_sum.nblocks);
 152         if (unlikely(!bh))
 153                 return -ENOMEM;
 154
 155         nilfs_segbuf_add_payload_buffer(segbuf, bh);
 156         *bhp = bh;
 157         return 0;
 158 }
 159
 160 int nilfs_segbuf_reset(struct nilfs_segment_buffer *segbuf, unsigned flags,
 161                        time_t ctime)
 162 {
 163         int err;
 164
 165         segbuf->sb_sum.nblocks = segbuf->sb_sum.nsumblk = 0;
 166         err = nilfs_segbuf_extend_segsum(segbuf);
 167         if (unlikely(err))
 168                 return err;
 169
 170         segbuf->sb_sum.flags = flags;
 171         segbuf->sb_sum.sumbytes = sizeof(struct nilfs_segment_summary);
 172         segbuf->sb_sum.nfinfo = segbuf->sb_sum.nfileblk = 0;
 173         segbuf->sb_sum.ctime = ctime;
 174         return 0;
 175 }
 176
 177 /*
 178  * Setup segment summary
 179  */
 180 void nilfs_segbuf_fill_in_segsum(struct nilfs_segment_buffer *segbuf)
 181 {
 182         struct nilfs_segment_summary *raw_sum;
 183         struct buffer_head *bh_sum;
 184
 185         bh_sum = list_entry(segbuf->sb_segsum_buffers.next,
 186                             struct buffer_head, b_assoc_buffers);
 187         raw_sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 188
 189         raw_sum->ss_magic    = cpu_to_le32(NILFS_SEGSUM_MAGIC);
 190         raw_sum->ss_bytes    = cpu_to_le16(sizeof(*raw_sum));
 191         raw_sum->ss_flags    = cpu_to_le16(segbuf->sb_sum.flags);
 192         raw_sum->ss_seq      = cpu_to_le64(segbuf->sb_sum.seg_seq);
 193         raw_sum->ss_create   = cpu_to_le64(segbuf->sb_sum.ctime);
 194         raw_sum->ss_next     = cpu_to_le64(segbuf->sb_sum.next);
 195         raw_sum->ss_nblocks  = cpu_to_le32(segbuf->sb_sum.nblocks);
 196         raw_sum->ss_nfinfo   = cpu_to_le32(segbuf->sb_sum.nfinfo);
 197         raw_sum->ss_sumbytes = cpu_to_le32(segbuf->sb_sum.sumbytes);
 198         raw_sum->ss_pad      = 0;
 199 }
 200
 201 /*
 202  * CRC calculation routines
 203  */
 204 void nilfs_segbuf_fill_in_segsum_crc(struct nilfs_segment_buffer *segbuf,
 205                                      u32 seed)
 206 {
 207         struct buffer_head *bh;
 208         struct nilfs_segment_summary *raw_sum;
 209         unsigned long size, bytes = segbuf->sb_sum.sumbytes;
 210         u32 crc;
 211
 212         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
 213                         b_assoc_buffers);
 214
 215         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
 216         size = min_t(unsigned long, bytes, bh->b_size);
 217         crc = crc32_le(seed,
 218                        (unsigned char *)raw_sum +
 219                        sizeof(raw_sum->ss_datasum) + sizeof(raw_sum->ss_sumsum),
 220                        size - (sizeof(raw_sum->ss_datasum) +
 221                                sizeof(raw_sum->ss_sumsum)));
 222
 223         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
 224                                      b_assoc_buffers) {
 225                 bytes -= size;
 226                 size = min_t(unsigned long, bytes, bh->b_size);
 227                 crc = crc32_le(crc, bh->b_data, size);
 228         }
 229         raw_sum->ss_sumsum = cpu_to_le32(crc);
 230 }
 231
 232 void nilfs_segbuf_fill_in_data_crc(struct nilfs_segment_buffer *segbuf,
 233                                    u32 seed)
 234 {
 235         struct buffer_head *bh;
 236         struct nilfs_segment_summary *raw_sum;
 237         void *kaddr;
 238         u32 crc;
 239
 240         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
 241                         b_assoc_buffers);
 242         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
 243         crc = crc32_le(seed,
 244                        (unsigned char *)raw_sum + sizeof(raw_sum->ss_datasum),
 245                        bh->b_size - sizeof(raw_sum->ss_datasum));
 246
 247         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
 248                                      b_assoc_buffers) {
 249                 crc = crc32_le(crc, bh->b_data, bh->b_size);
 250         }
 251         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
 252                 kaddr = kmap_atomic(bh->b_page, KM_USER0);
 253                 crc = crc32_le(crc, kaddr + bh_offset(bh), bh->b_size);
 254                 kunmap_atomic(kaddr, KM_USER0);
 255         }
 256         raw_sum->ss_datasum = cpu_to_le32(crc);
 257 }
 258
 259 static void nilfs_release_buffers(struct list_head *list)
 260 {
 261         struct buffer_head *bh, *n;
 262
 263         list_for_each_entry_safe(bh, n, list, b_assoc_buffers) {
 264                 list_del_init(&bh->b_assoc_buffers);
 265                 if (buffer_nilfs_allocated(bh)) {
 266                         struct page *clone_page = bh->b_page;
 267
 268                         /* remove clone page */
 269                         brelse(bh);
 270                         page_cache_release(clone_page); /* for each bh */
 271                         if (page_count(clone_page) <= 2) {
 272                                 lock_page(clone_page);
 273                                 nilfs_free_private_page(clone_page);
 274                         }
 275                         continue;
 276                 }
 277                 brelse(bh);
 278         }
 279 }
 280
 281 static void nilfs_segbuf_clear(struct nilfs_segment_buffer *segbuf)
 282 {
 283         nilfs_release_buffers(&segbuf->sb_segsum_buffers);
 284         nilfs_release_buffers(&segbuf->sb_payload_buffers);
 285 }
 286
 287 /*
 288  * Iterators for segment buffers
 289  */
 290 void nilfs_clear_logs(struct list_head *logs)
 291 {
 292         struct nilfs_segment_buffer *segbuf;
 293
 294         list_for_each_entry(segbuf, logs, sb_list)
 295                 nilfs_segbuf_clear(segbuf);
 296 }
 297
 298 void nilfs_truncate_logs(struct list_head *logs,
 299                          struct nilfs_segment_buffer *last)
 300 {
 301         struct nilfs_segment_buffer *n, *segbuf;
 302
 303         segbuf = list_prepare_entry(last, logs, sb_list);
 304         list_for_each_entry_safe_continue(segbuf, n, logs, sb_list) {
 305                 list_del_init(&segbuf->sb_list);
 306                 nilfs_segbuf_clear(segbuf);
 307                 nilfs_segbuf_free(segbuf);
 308         }
 309 }
 310
 311 int nilfs_write_logs(struct list_head *logs, struct the_nilfs *nilfs)
 312 {
 313         struct nilfs_segment_buffer *segbuf;
 314         int ret = 0;
 315
 316         list_for_each_entry(segbuf, logs, sb_list) {
 317                 ret = nilfs_segbuf_write(segbuf, nilfs);
 318                 if (ret)
 319                         break;
 320         }
 321         return ret;
 322 }
 323
 324 int nilfs_wait_on_logs(struct list_head *logs)
 325 {
 326         struct nilfs_segment_buffer *segbuf;
 327         int err, ret = 0;
 328
 329         list_for_each_entry(segbuf, logs, sb_list) {
 330                 err = nilfs_segbuf_wait(segbuf);
 331                 if (err && !ret)
 332                         ret = err;
 333         }
 334         return ret;
 335 }
 336
 337 /*
 338  * BIO operations
 339  */
 340 static void nilfs_end_bio_write(struct bio *bio, int err)
 341 {
 342         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 343         struct nilfs_segment_buffer *segbuf = bio->bi_private;
 344
 345         if (err == -EOPNOTSUPP) {
 346                 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
 347                 bio_put(bio);
 348                 /* to be detected by submit_seg_bio() */
 349         }
 350
 351         if (!uptodate)
 352                 atomic_inc(&segbuf->sb_err);
 353
 354         bio_put(bio);
 355         complete(&segbuf->sb_bio_event);
 356 }
 357
 358 static int nilfs_segbuf_submit_bio(struct nilfs_segment_buffer *segbuf,
 359                                    struct nilfs_write_info *wi, int mode)
 360 {
 361         struct bio *bio = wi->bio;
 362         int err;
 363
 364         if (segbuf->sb_nbio > 0 && bdi_write_congested(wi->nilfs->ns_bdi)) {
 365                 wait_for_completion(&segbuf->sb_bio_event);
 366                 segbuf->sb_nbio--;
 367                 if (unlikely(atomic_read(&segbuf->sb_err))) {
 368                         bio_put(bio);
 369                         err = -EIO;
 370                         goto failed;
 371                 }
 372         }
 373
 374         bio->bi_end_io = nilfs_end_bio_write;
 375         bio->bi_private = segbuf;
 376         bio_get(bio);
 377         submit_bio(mode, bio);
 378         if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
 379                 bio_put(bio);
 380                 err = -EOPNOTSUPP;
 381                 goto failed;
 382         }
 383         segbuf->sb_nbio++;
 384         bio_put(bio);
 385
 386         wi->bio = NULL;
 387         wi->rest_blocks -= wi->end - wi->start;
 388         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
 389         wi->start = wi->end;
 390         return 0;
 391
 392  failed:
 393         wi->bio = NULL;
 394         return err;
 395 }
 396
 397 /**
 398  * nilfs_alloc_seg_bio - allocate a new bio for writing log
 399  * @nilfs: nilfs object
 400  * @start: start block number of the bio
 401  * @nr_vecs: request size of page vector.
 402  *
 403  * Return Value: On success, pointer to the struct bio is returned.
 404  * On error, NULL is returned.
 405  */
 406 static struct bio *nilfs_alloc_seg_bio(struct the_nilfs *nilfs, sector_t start,
 407                                        int nr_vecs)
 408 {
 409         struct bio *bio;
 410
 411         bio = bio_alloc(GFP_NOIO, nr_vecs);
 412         if (bio == NULL) {
 413                 while (!bio && (nr_vecs >>= 1))
 414                         bio = bio_alloc(GFP_NOIO, nr_vecs);
 415         }
 416         if (likely(bio)) {
 417                 bio->bi_bdev = nilfs->ns_bdev;
 418                 bio->bi_sector = start << (nilfs->ns_blocksize_bits - 9);
 419         }
 420         return bio;
 421 }
 422
 423 static void nilfs_segbuf_prepare_write(struct nilfs_segment_buffer *segbuf,
 424                                        struct nilfs_write_info *wi)
 425 {
 426         wi->bio = NULL;
 427         wi->rest_blocks = segbuf->sb_sum.nblocks;
 428         wi->max_pages = bio_get_nr_vecs(wi->nilfs->ns_bdev);
 429         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
 430         wi->start = wi->end = 0;
 431         wi->blocknr = segbuf->sb_pseg_start;
 432 }
 433
 434 static int nilfs_segbuf_submit_bh(struct nilfs_segment_buffer *segbuf,
 435                                   struct nilfs_write_info *wi,
 436                                   struct buffer_head *bh, int mode)
 437 {
 438         int len, err;
 439
 440         BUG_ON(wi->nr_vecs <= 0);
 441  repeat:
 442         if (!wi->bio) {
 443                 wi->bio = nilfs_alloc_seg_bio(wi->nilfs, wi->blocknr + wi->end,
 444                                               wi->nr_vecs);
 445                 if (unlikely(!wi->bio))
 446                         return -ENOMEM;
 447         }
 448
 449         len = bio_add_page(wi->bio, bh->b_page, bh->b_size, bh_offset(bh));
 450         if (len == bh->b_size) {
 451                 wi->end++;
 452                 return 0;
 453         }
 454         /* bio is FULL */
 455         err = nilfs_segbuf_submit_bio(segbuf, wi, mode);
 456         /* never submit current bh */
 457         if (likely(!err))
 458                 goto repeat;
 459         return err;
 460 }
 461
 462 /**
 463  * nilfs_segbuf_write - submit write requests of a log
 464  * @segbuf: buffer storing a log to be written
 465  * @nilfs: nilfs object
 466  *
 467  * Return Value: On Success, 0 is returned. On Error, one of the following
 468  * negative error code is returned.
 469  *
 470  * %-EIO - I/O error
 471  *
 472  * %-ENOMEM - Insufficient memory available.
 473  */
 474 static int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
 475                               struct the_nilfs *nilfs)
 476 {
 477         struct nilfs_write_info wi;
 478         struct buffer_head *bh;
 479         int res = 0, rw = WRITE;
 480
 481         wi.nilfs = nilfs;
 482         nilfs_segbuf_prepare_write(segbuf, &wi);
 483
 484         list_for_each_entry(bh, &segbuf->sb_segsum_buffers, b_assoc_buffers) {
 485                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
 486                 if (unlikely(res))
 487                         goto failed_bio;
 488         }
 489
 490         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
 491                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
 492                 if (unlikely(res))
 493                         goto failed_bio;
 494         }
 495
 496         if (wi.bio) {
 497                 /*
 498                  * Last BIO is always sent through the following
 499                  * submission.
 500                  */
 501                 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
 502                 res = nilfs_segbuf_submit_bio(segbuf, &wi, rw);
 503         }
 504
 505  failed_bio:
 506         return res;
 507 }
 508
 509 /**
 510  * nilfs_segbuf_wait - wait for completion of requested BIOs
 511  * @segbuf: segment buffer
 512  *
 513  * Return Value: On Success, 0 is returned. On Error, one of the following
 514  * negative error code is returned.
 515  *
 516  * %-EIO - I/O error
 517  */
 518 static int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf)
 519 {
 520         int err = 0;
 521
 522         if (!segbuf->sb_nbio)
 523                 return 0;
 524
 525         do {
 526                 wait_for_completion(&segbuf->sb_bio_event);
 527         } while (--segbuf->sb_nbio > 0);
 528
 529         if (unlikely(atomic_read(&segbuf->sb_err) > 0)) {
 530                 printk(KERN_ERR "NILFS: IO error writing segment\n");
 531                 err = -EIO;
 532         }
 533         return err;
 534 }