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         segbuf->sb_super_root = NULL;
  85
  86         init_completion(&segbuf->sb_bio_event);
  87         atomic_set(&segbuf->sb_err, 0);
  88         segbuf->sb_nbio = 0;
  89
  90         return segbuf;
  91 }
  92
  93 void nilfs_segbuf_free(struct nilfs_segment_buffer *segbuf)
  94 {
  95         kmem_cache_free(nilfs_segbuf_cachep, segbuf);
  96 }
  97
  98 void nilfs_segbuf_map(struct nilfs_segment_buffer *segbuf, __u64 segnum,
  99                      unsigned long offset, struct the_nilfs *nilfs)
 100 {
 101         segbuf->sb_segnum = segnum;
 102         nilfs_get_segment_range(nilfs, segnum, &segbuf->sb_fseg_start,
 103                                 &segbuf->sb_fseg_end);
 104
 105         segbuf->sb_pseg_start = segbuf->sb_fseg_start + offset;
 106         segbuf->sb_rest_blocks =
 107                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
 108 }
 109
 110 /**
 111  * nilfs_segbuf_map_cont - map a new log behind a given log
 112  * @segbuf: new segment buffer
 113  * @prev: segment buffer containing a log to be continued
 114  */
 115 void nilfs_segbuf_map_cont(struct nilfs_segment_buffer *segbuf,
 116                            struct nilfs_segment_buffer *prev)
 117 {
 118         segbuf->sb_segnum = prev->sb_segnum;
 119         segbuf->sb_fseg_start = prev->sb_fseg_start;
 120         segbuf->sb_fseg_end = prev->sb_fseg_end;
 121         segbuf->sb_pseg_start = prev->sb_pseg_start + prev->sb_sum.nblocks;
 122         segbuf->sb_rest_blocks =
 123                 segbuf->sb_fseg_end - segbuf->sb_pseg_start + 1;
 124 }
 125
 126 void nilfs_segbuf_set_next_segnum(struct nilfs_segment_buffer *segbuf,
 127                                   __u64 nextnum, struct the_nilfs *nilfs)
 128 {
 129         segbuf->sb_nextnum = nextnum;
 130         segbuf->sb_sum.next = nilfs_get_segment_start_blocknr(nilfs, nextnum);
 131 }
 132
 133 int nilfs_segbuf_extend_segsum(struct nilfs_segment_buffer *segbuf)
 134 {
 135         struct buffer_head *bh;
 136
 137         bh = sb_getblk(segbuf->sb_super,
 138                        segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk);
 139         if (unlikely(!bh))
 140                 return -ENOMEM;
 141
 142         nilfs_segbuf_add_segsum_buffer(segbuf, bh);
 143         return 0;
 144 }
 145
 146 int nilfs_segbuf_extend_payload(struct nilfs_segment_buffer *segbuf,
 147                                 struct buffer_head **bhp)
 148 {
 149         struct buffer_head *bh;
 150
 151         bh = sb_getblk(segbuf->sb_super,
 152                        segbuf->sb_pseg_start + segbuf->sb_sum.nblocks);
 153         if (unlikely(!bh))
 154                 return -ENOMEM;
 155
 156         nilfs_segbuf_add_payload_buffer(segbuf, bh);
 157         *bhp = bh;
 158         return 0;
 159 }
 160
 161 int nilfs_segbuf_reset(struct nilfs_segment_buffer *segbuf, unsigned flags,
 162                        time_t ctime)
 163 {
 164         int err;
 165
 166         segbuf->sb_sum.nblocks = segbuf->sb_sum.nsumblk = 0;
 167         err = nilfs_segbuf_extend_segsum(segbuf);
 168         if (unlikely(err))
 169                 return err;
 170
 171         segbuf->sb_sum.flags = flags;
 172         segbuf->sb_sum.sumbytes = sizeof(struct nilfs_segment_summary);
 173         segbuf->sb_sum.nfinfo = segbuf->sb_sum.nfileblk = 0;
 174         segbuf->sb_sum.ctime = ctime;
 175         return 0;
 176 }
 177
 178 /*
 179  * Setup segment summary
 180  */
 181 void nilfs_segbuf_fill_in_segsum(struct nilfs_segment_buffer *segbuf)
 182 {
 183         struct nilfs_segment_summary *raw_sum;
 184         struct buffer_head *bh_sum;
 185
 186         bh_sum = list_entry(segbuf->sb_segsum_buffers.next,
 187                             struct buffer_head, b_assoc_buffers);
 188         raw_sum = (struct nilfs_segment_summary *)bh_sum->b_data;
 189
 190         raw_sum->ss_magic    = cpu_to_le32(NILFS_SEGSUM_MAGIC);
 191         raw_sum->ss_bytes    = cpu_to_le16(sizeof(*raw_sum));
 192         raw_sum->ss_flags    = cpu_to_le16(segbuf->sb_sum.flags);
 193         raw_sum->ss_seq      = cpu_to_le64(segbuf->sb_sum.seg_seq);
 194         raw_sum->ss_create   = cpu_to_le64(segbuf->sb_sum.ctime);
 195         raw_sum->ss_next     = cpu_to_le64(segbuf->sb_sum.next);
 196         raw_sum->ss_nblocks  = cpu_to_le32(segbuf->sb_sum.nblocks);
 197         raw_sum->ss_nfinfo   = cpu_to_le32(segbuf->sb_sum.nfinfo);
 198         raw_sum->ss_sumbytes = cpu_to_le32(segbuf->sb_sum.sumbytes);
 199         raw_sum->ss_pad      = 0;
 200 }
 201
 202 /*
 203  * CRC calculation routines
 204  */
 205 void nilfs_segbuf_fill_in_segsum_crc(struct nilfs_segment_buffer *segbuf,
 206                                      u32 seed)
 207 {
 208         struct buffer_head *bh;
 209         struct nilfs_segment_summary *raw_sum;
 210         unsigned long size, bytes = segbuf->sb_sum.sumbytes;
 211         u32 crc;
 212
 213         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
 214                         b_assoc_buffers);
 215
 216         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
 217         size = min_t(unsigned long, bytes, bh->b_size);
 218         crc = crc32_le(seed,
 219                        (unsigned char *)raw_sum +
 220                        sizeof(raw_sum->ss_datasum) + sizeof(raw_sum->ss_sumsum),
 221                        size - (sizeof(raw_sum->ss_datasum) +
 222                                sizeof(raw_sum->ss_sumsum)));
 223
 224         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
 225                                      b_assoc_buffers) {
 226                 bytes -= size;
 227                 size = min_t(unsigned long, bytes, bh->b_size);
 228                 crc = crc32_le(crc, bh->b_data, size);
 229         }
 230         raw_sum->ss_sumsum = cpu_to_le32(crc);
 231 }
 232
 233 void nilfs_segbuf_fill_in_data_crc(struct nilfs_segment_buffer *segbuf,
 234                                    u32 seed)
 235 {
 236         struct buffer_head *bh;
 237         struct nilfs_segment_summary *raw_sum;
 238         void *kaddr;
 239         u32 crc;
 240
 241         bh = list_entry(segbuf->sb_segsum_buffers.next, struct buffer_head,
 242                         b_assoc_buffers);
 243         raw_sum = (struct nilfs_segment_summary *)bh->b_data;
 244         crc = crc32_le(seed,
 245                        (unsigned char *)raw_sum + sizeof(raw_sum->ss_datasum),
 246                        bh->b_size - sizeof(raw_sum->ss_datasum));
 247
 248         list_for_each_entry_continue(bh, &segbuf->sb_segsum_buffers,
 249                                      b_assoc_buffers) {
 250                 crc = crc32_le(crc, bh->b_data, bh->b_size);
 251         }
 252         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
 253                 kaddr = kmap_atomic(bh->b_page, KM_USER0);
 254                 crc = crc32_le(crc, kaddr + bh_offset(bh), bh->b_size);
 255                 kunmap_atomic(kaddr, KM_USER0);
 256         }
 257         raw_sum->ss_datasum = cpu_to_le32(crc);
 258 }
 259
 260 static void nilfs_release_buffers(struct list_head *list)
 261 {
 262         struct buffer_head *bh, *n;
 263
 264         list_for_each_entry_safe(bh, n, list, b_assoc_buffers) {
 265                 list_del_init(&bh->b_assoc_buffers);
 266                 if (buffer_nilfs_allocated(bh)) {
 267                         struct page *clone_page = bh->b_page;
 268
 269                         /* remove clone page */
 270                         brelse(bh);
 271                         page_cache_release(clone_page); /* for each bh */
 272                         if (page_count(clone_page) <= 2) {
 273                                 lock_page(clone_page);
 274                                 nilfs_free_private_page(clone_page);
 275                         }
 276                         continue;
 277                 }
 278                 brelse(bh);
 279         }
 280 }
 281
 282 static void nilfs_segbuf_clear(struct nilfs_segment_buffer *segbuf)
 283 {
 284         nilfs_release_buffers(&segbuf->sb_segsum_buffers);
 285         nilfs_release_buffers(&segbuf->sb_payload_buffers);
 286         segbuf->sb_super_root = NULL;
 287 }
 288
 289 /*
 290  * Iterators for segment buffers
 291  */
 292 void nilfs_clear_logs(struct list_head *logs)
 293 {
 294         struct nilfs_segment_buffer *segbuf;
 295
 296         list_for_each_entry(segbuf, logs, sb_list)
 297                 nilfs_segbuf_clear(segbuf);
 298 }
 299
 300 void nilfs_truncate_logs(struct list_head *logs,
 301                          struct nilfs_segment_buffer *last)
 302 {
 303         struct nilfs_segment_buffer *n, *segbuf;
 304
 305         segbuf = list_prepare_entry(last, logs, sb_list);
 306         list_for_each_entry_safe_continue(segbuf, n, logs, sb_list) {
 307                 list_del_init(&segbuf->sb_list);
 308                 nilfs_segbuf_clear(segbuf);
 309                 nilfs_segbuf_free(segbuf);
 310         }
 311 }
 312
 313 int nilfs_write_logs(struct list_head *logs, struct the_nilfs *nilfs)
 314 {
 315         struct nilfs_segment_buffer *segbuf;
 316         int ret = 0;
 317
 318         list_for_each_entry(segbuf, logs, sb_list) {
 319                 ret = nilfs_segbuf_write(segbuf, nilfs);
 320                 if (ret)
 321                         break;
 322         }
 323         return ret;
 324 }
 325
 326 int nilfs_wait_on_logs(struct list_head *logs)
 327 {
 328         struct nilfs_segment_buffer *segbuf;
 329         int err, ret = 0;
 330
 331         list_for_each_entry(segbuf, logs, sb_list) {
 332                 err = nilfs_segbuf_wait(segbuf);
 333                 if (err && !ret)
 334                         ret = err;
 335         }
 336         return ret;
 337 }
 338
 339 /*
 340  * BIO operations
 341  */
 342 static void nilfs_end_bio_write(struct bio *bio, int err)
 343 {
 344         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
 345         struct nilfs_segment_buffer *segbuf = bio->bi_private;
 346
 347         if (err == -EOPNOTSUPP) {
 348                 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
 349                 bio_put(bio);
 350                 /* to be detected by submit_seg_bio() */
 351         }
 352
 353         if (!uptodate)
 354                 atomic_inc(&segbuf->sb_err);
 355
 356         bio_put(bio);
 357         complete(&segbuf->sb_bio_event);
 358 }
 359
 360 static int nilfs_segbuf_submit_bio(struct nilfs_segment_buffer *segbuf,
 361                                    struct nilfs_write_info *wi, int mode)
 362 {
 363         struct bio *bio = wi->bio;
 364         int err;
 365
 366         if (segbuf->sb_nbio > 0 && bdi_write_congested(wi->nilfs->ns_bdi)) {
 367                 wait_for_completion(&segbuf->sb_bio_event);
 368                 segbuf->sb_nbio--;
 369                 if (unlikely(atomic_read(&segbuf->sb_err))) {
 370                         bio_put(bio);
 371                         err = -EIO;
 372                         goto failed;
 373                 }
 374         }
 375
 376         bio->bi_end_io = nilfs_end_bio_write;
 377         bio->bi_private = segbuf;
 378         bio_get(bio);
 379         submit_bio(mode, bio);
 380         if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
 381                 bio_put(bio);
 382                 err = -EOPNOTSUPP;
 383                 goto failed;
 384         }
 385         segbuf->sb_nbio++;
 386         bio_put(bio);
 387
 388         wi->bio = NULL;
 389         wi->rest_blocks -= wi->end - wi->start;
 390         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
 391         wi->start = wi->end;
 392         return 0;
 393
 394  failed:
 395         wi->bio = NULL;
 396         return err;
 397 }
 398
 399 /**
 400  * nilfs_alloc_seg_bio - allocate a new bio for writing log
 401  * @nilfs: nilfs object
 402  * @start: start block number of the bio
 403  * @nr_vecs: request size of page vector.
 404  *
 405  * Return Value: On success, pointer to the struct bio is returned.
 406  * On error, NULL is returned.
 407  */
 408 static struct bio *nilfs_alloc_seg_bio(struct the_nilfs *nilfs, sector_t start,
 409                                        int nr_vecs)
 410 {
 411         struct bio *bio;
 412
 413         bio = bio_alloc(GFP_NOIO, nr_vecs);
 414         if (bio == NULL) {
 415                 while (!bio && (nr_vecs >>= 1))
 416                         bio = bio_alloc(GFP_NOIO, nr_vecs);
 417         }
 418         if (likely(bio)) {
 419                 bio->bi_bdev = nilfs->ns_bdev;
 420                 bio->bi_sector = start << (nilfs->ns_blocksize_bits - 9);
 421         }
 422         return bio;
 423 }
 424
 425 static void nilfs_segbuf_prepare_write(struct nilfs_segment_buffer *segbuf,
 426                                        struct nilfs_write_info *wi)
 427 {
 428         wi->bio = NULL;
 429         wi->rest_blocks = segbuf->sb_sum.nblocks;
 430         wi->max_pages = bio_get_nr_vecs(wi->nilfs->ns_bdev);
 431         wi->nr_vecs = min(wi->max_pages, wi->rest_blocks);
 432         wi->start = wi->end = 0;
 433         wi->blocknr = segbuf->sb_pseg_start;
 434 }
 435
 436 static int nilfs_segbuf_submit_bh(struct nilfs_segment_buffer *segbuf,
 437                                   struct nilfs_write_info *wi,
 438                                   struct buffer_head *bh, int mode)
 439 {
 440         int len, err;
 441
 442         BUG_ON(wi->nr_vecs <= 0);
 443  repeat:
 444         if (!wi->bio) {
 445                 wi->bio = nilfs_alloc_seg_bio(wi->nilfs, wi->blocknr + wi->end,
 446                                               wi->nr_vecs);
 447                 if (unlikely(!wi->bio))
 448                         return -ENOMEM;
 449         }
 450
 451         len = bio_add_page(wi->bio, bh->b_page, bh->b_size, bh_offset(bh));
 452         if (len == bh->b_size) {
 453                 wi->end++;
 454                 return 0;
 455         }
 456         /* bio is FULL */
 457         err = nilfs_segbuf_submit_bio(segbuf, wi, mode);
 458         /* never submit current bh */
 459         if (likely(!err))
 460                 goto repeat;
 461         return err;
 462 }
 463
 464 /**
 465  * nilfs_segbuf_write - submit write requests of a log
 466  * @segbuf: buffer storing a log to be written
 467  * @nilfs: nilfs object
 468  *
 469  * Return Value: On Success, 0 is returned. On Error, one of the following
 470  * negative error code is returned.
 471  *
 472  * %-EIO - I/O error
 473  *
 474  * %-ENOMEM - Insufficient memory available.
 475  */
 476 static int nilfs_segbuf_write(struct nilfs_segment_buffer *segbuf,
 477                               struct the_nilfs *nilfs)
 478 {
 479         struct nilfs_write_info wi;
 480         struct buffer_head *bh;
 481         int res = 0, rw = WRITE;
 482
 483         wi.nilfs = nilfs;
 484         nilfs_segbuf_prepare_write(segbuf, &wi);
 485
 486         list_for_each_entry(bh, &segbuf->sb_segsum_buffers, b_assoc_buffers) {
 487                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
 488                 if (unlikely(res))
 489                         goto failed_bio;
 490         }
 491
 492         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
 493                 res = nilfs_segbuf_submit_bh(segbuf, &wi, bh, rw);
 494                 if (unlikely(res))
 495                         goto failed_bio;
 496         }
 497
 498         if (wi.bio) {
 499                 /*
 500                  * Last BIO is always sent through the following
 501                  * submission.
 502                  */
 503                 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
 504                 res = nilfs_segbuf_submit_bio(segbuf, &wi, rw);
 505         }
 506
 507  failed_bio:
 508         return res;
 509 }
 510
 511 /**
 512  * nilfs_segbuf_wait - wait for completion of requested BIOs
 513  * @segbuf: segment buffer
 514  *
 515  * Return Value: On Success, 0 is returned. On Error, one of the following
 516  * negative error code is returned.
 517  *
 518  * %-EIO - I/O error
 519  */
 520 static int nilfs_segbuf_wait(struct nilfs_segment_buffer *segbuf)
 521 {
 522         int err = 0;
 523
 524         if (!segbuf->sb_nbio)
 525                 return 0;
 526
 527         do {
 528                 wait_for_completion(&segbuf->sb_bio_event);
 529         } while (--segbuf->sb_nbio > 0);
 530
 531         if (unlikely(atomic_read(&segbuf->sb_err) > 0)) {
 532                 printk(KERN_ERR "NILFS: IO error writing segment\n");
 533                 err = -EIO;
 534         }
 535         return err;
 536 }