fs/nilfs2/mdt.c

   1 /*
   2  * mdt.c - meta data file for NILFS
   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 #include <linux/buffer_head.h>
  24 #include <linux/mpage.h>
  25 #include <linux/mm.h>
  26 #include <linux/writeback.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/swap.h>
  29 #include "nilfs.h"
  30 #include "segment.h"
  31 #include "page.h"
  32 #include "mdt.h"
  33
  34
  35 #define NILFS_MDT_MAX_RA_BLOCKS         (16 - 1)
  36
  37 #define INIT_UNUSED_INODE_FIELDS
  38
  39 static int
  40 nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
  41                            struct buffer_head *bh,
  42                            void (*init_block)(struct inode *,
  43                                               struct buffer_head *, void *))
  44 {
  45         struct nilfs_inode_info *ii = NILFS_I(inode);
  46         void *kaddr;
  47         int ret;
  48
  49         /* Caller exclude read accesses using page lock */
  50
  51         /* set_buffer_new(bh); */
  52         bh->b_blocknr = 0;
  53
  54         ret = nilfs_bmap_insert(ii->i_bmap, block, (unsigned long)bh);
  55         if (unlikely(ret))
  56                 return ret;
  57
  58         set_buffer_mapped(bh);
  59
  60         kaddr = kmap_atomic(bh->b_page, KM_USER0);
  61         memset(kaddr + bh_offset(bh), 0, 1 << inode->i_blkbits);
  62         if (init_block)
  63                 init_block(inode, bh, kaddr);
  64         flush_dcache_page(bh->b_page);
  65         kunmap_atomic(kaddr, KM_USER0);
  66
  67         set_buffer_uptodate(bh);
  68         nilfs_mark_buffer_dirty(bh);
  69         nilfs_mdt_mark_dirty(inode);
  70         return 0;
  71 }
  72
  73 static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
  74                                   struct buffer_head **out_bh,
  75                                   void (*init_block)(struct inode *,
  76                                                      struct buffer_head *,
  77                                                      void *))
  78 {
  79         struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
  80         struct super_block *sb = inode->i_sb;
  81         struct nilfs_transaction_info ti;
  82         struct buffer_head *bh;
  83         int err;
  84
  85         if (!sb) {
  86                 /*
  87                  * Make sure this function is not called from any
  88                  * read-only context.
  89                  */
  90                 if (!nilfs->ns_writer) {
  91                         WARN_ON(1);
  92                         err = -EROFS;
  93                         goto out;
  94                 }
  95                 sb = nilfs->ns_writer->s_super;
  96         }
  97
  98         nilfs_transaction_begin(sb, &ti, 0);
  99
 100         err = -ENOMEM;
 101         bh = nilfs_grab_buffer(inode, inode->i_mapping, block, 0);
 102         if (unlikely(!bh))
 103                 goto failed_unlock;
 104
 105         err = -EEXIST;
 106         if (buffer_uptodate(bh) || buffer_mapped(bh))
 107                 goto failed_bh;
 108 #if 0
 109         /* The uptodate flag is not protected by the page lock, but
 110            the mapped flag is.  Thus, we don't have to wait the buffer. */
 111         wait_on_buffer(bh);
 112         if (buffer_uptodate(bh))
 113                 goto failed_bh;
 114 #endif
 115
 116         bh->b_bdev = nilfs->ns_bdev;
 117         err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
 118         if (likely(!err)) {
 119                 get_bh(bh);
 120                 *out_bh = bh;
 121         }
 122
 123  failed_bh:
 124         unlock_page(bh->b_page);
 125         page_cache_release(bh->b_page);
 126         brelse(bh);
 127
 128  failed_unlock:
 129         if (likely(!err))
 130                 err = nilfs_transaction_commit(sb);
 131         else
 132                 nilfs_transaction_abort(sb);
 133  out:
 134         return err;
 135 }
 136
 137 static int
 138 nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
 139                        int mode, struct buffer_head **out_bh)
 140 {
 141         struct buffer_head *bh;
 142         unsigned long blknum = 0;
 143         int ret = -ENOMEM;
 144
 145         bh = nilfs_grab_buffer(inode, inode->i_mapping, blkoff, 0);
 146         if (unlikely(!bh))
 147                 goto failed;
 148
 149         ret = -EEXIST; /* internal code */
 150         if (buffer_uptodate(bh))
 151                 goto out;
 152
 153         if (mode == READA) {
 154                 if (!trylock_buffer(bh)) {
 155                         ret = -EBUSY;
 156                         goto failed_bh;
 157                 }
 158         } else /* mode == READ */
 159                 lock_buffer(bh);
 160
 161         if (buffer_uptodate(bh)) {
 162                 unlock_buffer(bh);
 163                 goto out;
 164         }
 165         if (!buffer_mapped(bh)) { /* unused buffer */
 166                 ret = nilfs_bmap_lookup(NILFS_I(inode)->i_bmap, blkoff,
 167                                         &blknum);
 168                 if (unlikely(ret)) {
 169                         unlock_buffer(bh);
 170                         goto failed_bh;
 171                 }
 172                 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev;
 173                 bh->b_blocknr = blknum;
 174                 set_buffer_mapped(bh);
 175         }
 176
 177         bh->b_end_io = end_buffer_read_sync;
 178         get_bh(bh);
 179         submit_bh(mode, bh);
 180         ret = 0;
 181  out:
 182         get_bh(bh);
 183         *out_bh = bh;
 184
 185  failed_bh:
 186         unlock_page(bh->b_page);
 187         page_cache_release(bh->b_page);
 188         brelse(bh);
 189  failed:
 190         return ret;
 191 }
 192
 193 static int nilfs_mdt_read_block(struct inode *inode, unsigned long block,
 194                                 struct buffer_head **out_bh)
 195 {
 196         struct buffer_head *first_bh, *bh;
 197         unsigned long blkoff;
 198         int i, nr_ra_blocks = NILFS_MDT_MAX_RA_BLOCKS;
 199         int err;
 200
 201         err = nilfs_mdt_submit_block(inode, block, READ, &first_bh);
 202         if (err == -EEXIST) /* internal code */
 203                 goto out;
 204
 205         if (unlikely(err))
 206                 goto failed;
 207
 208         blkoff = block + 1;
 209         for (i = 0; i < nr_ra_blocks; i++, blkoff++) {
 210                 err = nilfs_mdt_submit_block(inode, blkoff, READA, &bh);
 211                 if (likely(!err || err == -EEXIST))
 212                         brelse(bh);
 213                 else if (err != -EBUSY)
 214                         break; /* abort readahead if bmap lookup failed */
 215
 216                 if (!buffer_locked(first_bh))
 217                         goto out_no_wait;
 218         }
 219
 220         wait_on_buffer(first_bh);
 221
 222  out_no_wait:
 223         err = -EIO;
 224         if (!buffer_uptodate(first_bh))
 225                 goto failed_bh;
 226  out:
 227         *out_bh = first_bh;
 228         return 0;
 229
 230  failed_bh:
 231         brelse(first_bh);
 232  failed:
 233         return err;
 234 }
 235
 236 /**
 237  * nilfs_mdt_get_block - read or create a buffer on meta data file.
 238  * @inode: inode of the meta data file
 239  * @blkoff: block offset
 240  * @create: create flag
 241  * @init_block: initializer used for newly allocated block
 242  * @out_bh: output of a pointer to the buffer_head
 243  *
 244  * nilfs_mdt_get_block() looks up the specified buffer and tries to create
 245  * a new buffer if @create is not zero.  On success, the returned buffer is
 246  * assured to be either existing or formatted using a buffer lock on success.
 247  * @out_bh is substituted only when zero is returned.
 248  *
 249  * Return Value: On success, it returns 0. On error, the following negative
 250  * error code is returned.
 251  *
 252  * %-ENOMEM - Insufficient memory available.
 253  *
 254  * %-EIO - I/O error
 255  *
 256  * %-ENOENT - the specified block does not exist (hole block)
 257  *
 258  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 259  *
 260  * %-EROFS - Read only filesystem (for create mode)
 261  */
 262 int nilfs_mdt_get_block(struct inode *inode, unsigned long blkoff, int create,
 263                         void (*init_block)(struct inode *,
 264                                            struct buffer_head *, void *),
 265                         struct buffer_head **out_bh)
 266 {
 267         int ret;
 268
 269         /* Should be rewritten with merging nilfs_mdt_read_block() */
 270  retry:
 271         ret = nilfs_mdt_read_block(inode, blkoff, out_bh);
 272         if (!create || ret != -ENOENT)
 273                 return ret;
 274
 275         ret = nilfs_mdt_create_block(inode, blkoff, out_bh, init_block);
 276         if (unlikely(ret == -EEXIST)) {
 277                 /* create = 0; */  /* limit read-create loop retries */
 278                 goto retry;
 279         }
 280         return ret;
 281 }
 282
 283 /**
 284  * nilfs_mdt_delete_block - make a hole on the meta data file.
 285  * @inode: inode of the meta data file
 286  * @block: block offset
 287  *
 288  * Return Value: On success, zero is returned.
 289  * On error, one of the following negative error code is returned.
 290  *
 291  * %-ENOMEM - Insufficient memory available.
 292  *
 293  * %-EIO - I/O error
 294  *
 295  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 296  */
 297 int nilfs_mdt_delete_block(struct inode *inode, unsigned long block)
 298 {
 299         struct nilfs_inode_info *ii = NILFS_I(inode);
 300         int err;
 301
 302         err = nilfs_bmap_delete(ii->i_bmap, block);
 303         if (!err || err == -ENOENT) {
 304                 nilfs_mdt_mark_dirty(inode);
 305                 nilfs_mdt_forget_block(inode, block);
 306         }
 307         return err;
 308 }
 309
 310 /**
 311  * nilfs_mdt_forget_block - discard dirty state and try to remove the page
 312  * @inode: inode of the meta data file
 313  * @block: block offset
 314  *
 315  * nilfs_mdt_forget_block() clears a dirty flag of the specified buffer, and
 316  * tries to release the page including the buffer from a page cache.
 317  *
 318  * Return Value: On success, 0 is returned. On error, one of the following
 319  * negative error code is returned.
 320  *
 321  * %-EBUSY - page has an active buffer.
 322  *
 323  * %-ENOENT - page cache has no page addressed by the offset.
 324  */
 325 int nilfs_mdt_forget_block(struct inode *inode, unsigned long block)
 326 {
 327         pgoff_t index = (pgoff_t)block >>
 328                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 329         struct page *page;
 330         unsigned long first_block;
 331         int ret = 0;
 332         int still_dirty;
 333
 334         page = find_lock_page(inode->i_mapping, index);
 335         if (!page)
 336                 return -ENOENT;
 337
 338         wait_on_page_writeback(page);
 339
 340         first_block = (unsigned long)index <<
 341                 (PAGE_CACHE_SHIFT - inode->i_blkbits);
 342         if (page_has_buffers(page)) {
 343                 struct buffer_head *bh;
 344
 345                 bh = nilfs_page_get_nth_block(page, block - first_block);
 346                 nilfs_forget_buffer(bh);
 347         }
 348         still_dirty = PageDirty(page);
 349         unlock_page(page);
 350         page_cache_release(page);
 351
 352         if (still_dirty ||
 353             invalidate_inode_pages2_range(inode->i_mapping, index, index) != 0)
 354                 ret = -EBUSY;
 355         return ret;
 356 }
 357
 358 /**
 359  * nilfs_mdt_mark_block_dirty - mark a block on the meta data file dirty.
 360  * @inode: inode of the meta data file
 361  * @block: block offset
 362  *
 363  * Return Value: On success, it returns 0. On error, the following negative
 364  * error code is returned.
 365  *
 366  * %-ENOMEM - Insufficient memory available.
 367  *
 368  * %-EIO - I/O error
 369  *
 370  * %-ENOENT - the specified block does not exist (hole block)
 371  *
 372  * %-EINVAL - bmap is broken. (the caller should call nilfs_error())
 373  */
 374 int nilfs_mdt_mark_block_dirty(struct inode *inode, unsigned long block)
 375 {
 376         struct buffer_head *bh;
 377         int err;
 378
 379         err = nilfs_mdt_read_block(inode, block, &bh);
 380         if (unlikely(err))
 381                 return err;
 382         nilfs_mark_buffer_dirty(bh);
 383         nilfs_mdt_mark_dirty(inode);
 384         brelse(bh);
 385         return 0;
 386 }
 387
 388 int nilfs_mdt_fetch_dirty(struct inode *inode)
 389 {
 390         struct nilfs_inode_info *ii = NILFS_I(inode);
 391
 392         if (nilfs_bmap_test_and_clear_dirty(ii->i_bmap)) {
 393                 set_bit(NILFS_I_DIRTY, &ii->i_state);
 394                 return 1;
 395         }
 396         return test_bit(NILFS_I_DIRTY, &ii->i_state);
 397 }
 398
 399 static int
 400 nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
 401 {
 402         struct inode *inode = container_of(page->mapping,
 403                                            struct inode, i_data);
 404         struct super_block *sb = inode->i_sb;
 405         struct nilfs_sb_info *writer = NULL;
 406         int err = 0;
 407
 408         redirty_page_for_writepage(wbc, page);
 409         unlock_page(page);
 410
 411         if (page->mapping->assoc_mapping)
 412                 return 0; /* Do not request flush for shadow page cache */
 413         if (!sb) {
 414                 writer = nilfs_get_writer(NILFS_MDT(inode)->mi_nilfs);
 415                 if (!writer)
 416                         return -EROFS;
 417                 sb = writer->s_super;
 418         }
 419
 420         if (wbc->sync_mode == WB_SYNC_ALL)
 421                 err = nilfs_construct_segment(sb);
 422         else if (wbc->for_reclaim)
 423                 nilfs_flush_segment(sb, inode->i_ino);
 424
 425         if (writer)
 426                 nilfs_put_writer(NILFS_MDT(inode)->mi_nilfs);
 427         return err;
 428 }
 429
 430
 431 static struct address_space_operations def_mdt_aops = {
 432         .writepage              = nilfs_mdt_write_page,
 433 };
 434
 435 static struct inode_operations def_mdt_iops;
 436 static struct file_operations def_mdt_fops;
 437
 438 /*
 439  * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile,
 440  * ifile, or gcinodes.  This allows the B-tree code and segment constructor
 441  * to treat them like regular files, and this helps to simplify the
 442  * implementation.
 443  *   On the other hand, some of the pseudo inodes have an irregular point:
 444  * They don't have valid inode->i_sb pointer because their lifetimes are
 445  * longer than those of the super block structs; they may continue for
 446  * several consecutive mounts/umounts.  This would need discussions.
 447  */
 448 struct inode *
 449 nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
 450                      ino_t ino, gfp_t gfp_mask)
 451 {
 452         struct inode *inode = nilfs_alloc_inode(sb);
 453
 454         if (!inode)
 455                 return NULL;
 456         else {
 457                 struct address_space * const mapping = &inode->i_data;
 458                 struct nilfs_mdt_info *mi = kzalloc(sizeof(*mi), GFP_NOFS);
 459
 460                 if (!mi) {
 461                         nilfs_destroy_inode(inode);
 462                         return NULL;
 463                 }
 464                 mi->mi_nilfs = nilfs;
 465                 init_rwsem(&mi->mi_sem);
 466
 467                 inode->i_sb = sb; /* sb may be NULL for some meta data files */
 468                 inode->i_blkbits = nilfs->ns_blocksize_bits;
 469                 inode->i_flags = 0;
 470                 atomic_set(&inode->i_count, 1);
 471                 inode->i_nlink = 1;
 472                 inode->i_ino = ino;
 473                 inode->i_mode = S_IFREG;
 474                 inode->i_private = mi;
 475
 476 #ifdef INIT_UNUSED_INODE_FIELDS
 477                 atomic_set(&inode->i_writecount, 0);
 478                 inode->i_size = 0;
 479                 inode->i_blocks = 0;
 480                 inode->i_bytes = 0;
 481                 inode->i_generation = 0;
 482 #ifdef CONFIG_QUOTA
 483                 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
 484 #endif
 485                 inode->i_pipe = NULL;
 486                 inode->i_bdev = NULL;
 487                 inode->i_cdev = NULL;
 488                 inode->i_rdev = 0;
 489 #ifdef CONFIG_SECURITY
 490                 inode->i_security = NULL;
 491 #endif
 492                 inode->dirtied_when = 0;
 493
 494                 INIT_LIST_HEAD(&inode->i_list);
 495                 INIT_LIST_HEAD(&inode->i_sb_list);
 496                 inode->i_state = 0;
 497 #endif
 498
 499                 spin_lock_init(&inode->i_lock);
 500                 mutex_init(&inode->i_mutex);
 501                 init_rwsem(&inode->i_alloc_sem);
 502
 503                 mapping->host = NULL;  /* instead of inode */
 504                 mapping->flags = 0;
 505                 mapping_set_gfp_mask(mapping, gfp_mask);
 506                 mapping->assoc_mapping = NULL;
 507                 mapping->backing_dev_info = nilfs->ns_bdi;
 508
 509                 inode->i_mapping = mapping;
 510         }
 511
 512         return inode;
 513 }
 514
 515 struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb,
 516                             ino_t ino, gfp_t gfp_mask)
 517 {
 518         struct inode *inode = nilfs_mdt_new_common(nilfs, sb, ino, gfp_mask);
 519
 520         if (!inode)
 521                 return NULL;
 522
 523         inode->i_op = &def_mdt_iops;
 524         inode->i_fop = &def_mdt_fops;
 525         inode->i_mapping->a_ops = &def_mdt_aops;
 526         return inode;
 527 }
 528
 529 void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
 530                               unsigned header_size)
 531 {
 532         struct nilfs_mdt_info *mi = NILFS_MDT(inode);
 533
 534         mi->mi_entry_size = entry_size;
 535         mi->mi_entries_per_block = (1 << inode->i_blkbits) / entry_size;
 536         mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
 537 }
 538
 539 void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow)
 540 {
 541         shadow->i_mapping->assoc_mapping = orig->i_mapping;
 542         NILFS_I(shadow)->i_btnode_cache.assoc_mapping =
 543                 &NILFS_I(orig)->i_btnode_cache;
 544 }
 545
 546 void nilfs_mdt_clear(struct inode *inode)
 547 {
 548         struct nilfs_inode_info *ii = NILFS_I(inode);
 549
 550         invalidate_mapping_pages(inode->i_mapping, 0, -1);
 551         truncate_inode_pages(inode->i_mapping, 0);
 552
 553         nilfs_bmap_clear(ii->i_bmap);
 554         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 555 }
 556
 557 void nilfs_mdt_destroy(struct inode *inode)
 558 {
 559         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 560
 561         kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
 562         kfree(mdi);
 563         nilfs_destroy_inode(inode);
 564 }