Pileus Git - ~andy/linux/blob - fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interrelated, and
   5 ** overly complex.  I need to rethink things a bit....The major players:
   6 **
   7 ** journal_begin -- call with the number of blocks you expect to log.
   8 **                  If the current transaction is too
   9 **                  old, it will block until the current transaction is
  10 **                  finished, and then start a new one.
  11 **                  Usually, your transaction will get joined in with
  12 **                  previous ones for speed.
  13 **
  14 ** journal_join  -- same as journal_begin, but won't block on the current
  15 **                  transaction regardless of age.  Don't ever call
  16 **                  this.  Ever.  There are only two places it should be
  17 **                  called from, and they are both inside this file.
  18 **
  19 ** journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20 **                       that might make them get sent to disk
  21 **                       and then marks them BH_JDirty.  Puts the buffer head
  22 **                       into the current transaction hash.
  23 **
  24 ** journal_end -- if the current transaction is batchable, it does nothing
  25 **                   otherwise, it could do an async/synchronous commit, or
  26 **                   a full flush of all log and real blocks in the
  27 **                   transaction.
  28 **
  29 ** flush_old_commits -- if the current transaction is too old, it is ended and
  30 **                      commit blocks are sent to disk.  Forces commit blocks
  31 **                      to disk for all backgrounded commits that have been
  32 **                      around too long.
  33 **                   -- Note, if you call this as an immediate flush from
  34 **                      from within kupdate, it will ignore the immediate flag
  35 */
  36
  37 #include <linux/time.h>
  38 #include <linux/semaphore.h>
  39 #include <linux/vmalloc.h>
  40 #include <linux/reiserfs_fs.h>
  41 #include <linux/kernel.h>
  42 #include <linux/errno.h>
  43 #include <linux/fcntl.h>
  44 #include <linux/stat.h>
  45 #include <linux/string.h>
  46 #include <linux/buffer_head.h>
  47 #include <linux/workqueue.h>
  48 #include <linux/writeback.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/backing-dev.h>
  51 #include <linux/uaccess.h>
  52 #include <linux/slab.h>
  53
  54 #include <asm/system.h>
  55
  56 /* gets a struct reiserfs_journal_list * from a list head */
  57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                                j_list))
  59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_working_list))
  61
  62 /* the number of mounted filesystems.  This is used to decide when to
  63 ** start and kill the commit workqueue
  64 */
  65 static int reiserfs_mounted_fs_count;
  66
  67 static struct workqueue_struct *commit_wq;
  68
  69 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  70                                    structs at 4k */
  71 #define BUFNR 64                /*read ahead */
  72
  73 /* cnode stat bits.  Move these into reiserfs_fs.h */
  74
  75 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  76 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  77
  78 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  79 #define BLOCK_DIRTIED 5
  80
  81 /* journal list state bits */
  82 #define LIST_TOUCHED 1
  83 #define LIST_DIRTY   2
  84 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  85
  86 /* flags for do_journal_end */
  87 #define FLUSH_ALL   1           /* flush commit and real blocks */
  88 #define COMMIT_NOW  2           /* end and commit this transaction */
  89 #define WAIT        4           /* wait for the log blocks to hit the disk */
  90
  91 static int do_journal_end(struct reiserfs_transaction_handle *,
  92                           struct super_block *, unsigned long nblocks,
  93                           int flags);
  94 static int flush_journal_list(struct super_block *s,
  95                               struct reiserfs_journal_list *jl, int flushall);
  96 static int flush_commit_list(struct super_block *s,
  97                              struct reiserfs_journal_list *jl, int flushall);
  98 static int can_dirty(struct reiserfs_journal_cnode *cn);
  99 static int journal_join(struct reiserfs_transaction_handle *th,
 100                         struct super_block *sb, unsigned long nblocks);
 101 static int release_journal_dev(struct super_block *super,
 102                                struct reiserfs_journal *journal);
 103 static int dirty_one_transaction(struct super_block *s,
 104                                  struct reiserfs_journal_list *jl);
 105 static void flush_async_commits(struct work_struct *work);
 106 static void queue_log_writer(struct super_block *s);
 107
 108 /* values for join in do_journal_begin_r */
 109 enum {
 110         JBEGIN_REG = 0,         /* regular journal begin */
 111         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 112         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 113 };
 114
 115 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 116                               struct super_block *sb,
 117                               unsigned long nblocks, int join);
 118
 119 static void init_journal_hash(struct super_block *sb)
 120 {
 121         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 122         memset(journal->j_hash_table, 0,
 123                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 124 }
 125
 126 /*
 127 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 128 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 129 ** more details.
 130 */
 131 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 132 {
 133         if (bh) {
 134                 clear_buffer_dirty(bh);
 135                 clear_buffer_journal_test(bh);
 136         }
 137         return 0;
 138 }
 139
 140 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 141                                                          *sb)
 142 {
 143         struct reiserfs_bitmap_node *bn;
 144         static int id;
 145
 146         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 147         if (!bn) {
 148                 return NULL;
 149         }
 150         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 151         if (!bn->data) {
 152                 kfree(bn);
 153                 return NULL;
 154         }
 155         bn->id = id++;
 156         INIT_LIST_HEAD(&bn->list);
 157         return bn;
 158 }
 159
 160 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 161 {
 162         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 163         struct reiserfs_bitmap_node *bn = NULL;
 164         struct list_head *entry = journal->j_bitmap_nodes.next;
 165
 166         journal->j_used_bitmap_nodes++;
 167       repeat:
 168
 169         if (entry != &journal->j_bitmap_nodes) {
 170                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 171                 list_del(entry);
 172                 memset(bn->data, 0, sb->s_blocksize);
 173                 journal->j_free_bitmap_nodes--;
 174                 return bn;
 175         }
 176         bn = allocate_bitmap_node(sb);
 177         if (!bn) {
 178                 yield();
 179                 goto repeat;
 180         }
 181         return bn;
 182 }
 183 static inline void free_bitmap_node(struct super_block *sb,
 184                                     struct reiserfs_bitmap_node *bn)
 185 {
 186         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 187         journal->j_used_bitmap_nodes--;
 188         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 189                 kfree(bn->data);
 190                 kfree(bn);
 191         } else {
 192                 list_add(&bn->list, &journal->j_bitmap_nodes);
 193                 journal->j_free_bitmap_nodes++;
 194         }
 195 }
 196
 197 static void allocate_bitmap_nodes(struct super_block *sb)
 198 {
 199         int i;
 200         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 201         struct reiserfs_bitmap_node *bn = NULL;
 202         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 203                 bn = allocate_bitmap_node(sb);
 204                 if (bn) {
 205                         list_add(&bn->list, &journal->j_bitmap_nodes);
 206                         journal->j_free_bitmap_nodes++;
 207                 } else {
 208                         break;  /* this is ok, we'll try again when more are needed */
 209                 }
 210         }
 211 }
 212
 213 static int set_bit_in_list_bitmap(struct super_block *sb,
 214                                   b_blocknr_t block,
 215                                   struct reiserfs_list_bitmap *jb)
 216 {
 217         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 218         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 219
 220         if (!jb->bitmaps[bmap_nr]) {
 221                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 222         }
 223         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 224         return 0;
 225 }
 226
 227 static void cleanup_bitmap_list(struct super_block *sb,
 228                                 struct reiserfs_list_bitmap *jb)
 229 {
 230         int i;
 231         if (jb->bitmaps == NULL)
 232                 return;
 233
 234         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 235                 if (jb->bitmaps[i]) {
 236                         free_bitmap_node(sb, jb->bitmaps[i]);
 237                         jb->bitmaps[i] = NULL;
 238                 }
 239         }
 240 }
 241
 242 /*
 243 ** only call this on FS unmount.
 244 */
 245 static int free_list_bitmaps(struct super_block *sb,
 246                              struct reiserfs_list_bitmap *jb_array)
 247 {
 248         int i;
 249         struct reiserfs_list_bitmap *jb;
 250         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 251                 jb = jb_array + i;
 252                 jb->journal_list = NULL;
 253                 cleanup_bitmap_list(sb, jb);
 254                 vfree(jb->bitmaps);
 255                 jb->bitmaps = NULL;
 256         }
 257         return 0;
 258 }
 259
 260 static int free_bitmap_nodes(struct super_block *sb)
 261 {
 262         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 263         struct list_head *next = journal->j_bitmap_nodes.next;
 264         struct reiserfs_bitmap_node *bn;
 265
 266         while (next != &journal->j_bitmap_nodes) {
 267                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 268                 list_del(next);
 269                 kfree(bn->data);
 270                 kfree(bn);
 271                 next = journal->j_bitmap_nodes.next;
 272                 journal->j_free_bitmap_nodes--;
 273         }
 274
 275         return 0;
 276 }
 277
 278 /*
 279 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 280 ** jb_array is the array to be filled in.
 281 */
 282 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 283                                    struct reiserfs_list_bitmap *jb_array,
 284                                    unsigned int bmap_nr)
 285 {
 286         int i;
 287         int failed = 0;
 288         struct reiserfs_list_bitmap *jb;
 289         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 290
 291         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 292                 jb = jb_array + i;
 293                 jb->journal_list = NULL;
 294                 jb->bitmaps = vmalloc(mem);
 295                 if (!jb->bitmaps) {
 296                         reiserfs_warning(sb, "clm-2000", "unable to "
 297                                          "allocate bitmaps for journal lists");
 298                         failed = 1;
 299                         break;
 300                 }
 301                 memset(jb->bitmaps, 0, mem);
 302         }
 303         if (failed) {
 304                 free_list_bitmaps(sb, jb_array);
 305                 return -1;
 306         }
 307         return 0;
 308 }
 309
 310 /*
 311 ** find an available list bitmap.  If you can't find one, flush a commit list
 312 ** and try again
 313 */
 314 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 315                                                     struct reiserfs_journal_list
 316                                                     *jl)
 317 {
 318         int i, j;
 319         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 320         struct reiserfs_list_bitmap *jb = NULL;
 321
 322         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 323                 i = journal->j_list_bitmap_index;
 324                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 325                 jb = journal->j_list_bitmap + i;
 326                 if (journal->j_list_bitmap[i].journal_list) {
 327                         flush_commit_list(sb,
 328                                           journal->j_list_bitmap[i].
 329                                           journal_list, 1);
 330                         if (!journal->j_list_bitmap[i].journal_list) {
 331                                 break;
 332                         }
 333                 } else {
 334                         break;
 335                 }
 336         }
 337         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 338                 return NULL;
 339         }
 340         jb->journal_list = jl;
 341         return jb;
 342 }
 343
 344 /*
 345 ** allocates a new chunk of X nodes, and links them all together as a list.
 346 ** Uses the cnode->next and cnode->prev pointers
 347 ** returns NULL on failure
 348 */
 349 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 350 {
 351         struct reiserfs_journal_cnode *head;
 352         int i;
 353         if (num_cnodes <= 0) {
 354                 return NULL;
 355         }
 356         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 357         if (!head) {
 358                 return NULL;
 359         }
 360         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 361         head[0].prev = NULL;
 362         head[0].next = head + 1;
 363         for (i = 1; i < num_cnodes; i++) {
 364                 head[i].prev = head + (i - 1);
 365                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 366         }
 367         head[num_cnodes - 1].next = NULL;
 368         return head;
 369 }
 370
 371 /*
 372 ** pulls a cnode off the free list, or returns NULL on failure
 373 */
 374 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 375 {
 376         struct reiserfs_journal_cnode *cn;
 377         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 378
 379         reiserfs_check_lock_depth(sb, "get_cnode");
 380
 381         if (journal->j_cnode_free <= 0) {
 382                 return NULL;
 383         }
 384         journal->j_cnode_used++;
 385         journal->j_cnode_free--;
 386         cn = journal->j_cnode_free_list;
 387         if (!cn) {
 388                 return cn;
 389         }
 390         if (cn->next) {
 391                 cn->next->prev = NULL;
 392         }
 393         journal->j_cnode_free_list = cn->next;
 394         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 395         return cn;
 396 }
 397
 398 /*
 399 ** returns a cnode to the free list
 400 */
 401 static void free_cnode(struct super_block *sb,
 402                        struct reiserfs_journal_cnode *cn)
 403 {
 404         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 405
 406         reiserfs_check_lock_depth(sb, "free_cnode");
 407
 408         journal->j_cnode_used--;
 409         journal->j_cnode_free++;
 410         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 411         cn->next = journal->j_cnode_free_list;
 412         if (journal->j_cnode_free_list) {
 413                 journal->j_cnode_free_list->prev = cn;
 414         }
 415         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 416         journal->j_cnode_free_list = cn;
 417 }
 418
 419 static void clear_prepared_bits(struct buffer_head *bh)
 420 {
 421         clear_buffer_journal_prepared(bh);
 422         clear_buffer_journal_restore_dirty(bh);
 423 }
 424
 425 /* return a cnode with same dev, block number and size in table, or null if not found */
 426 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 427                                                                   super_block
 428                                                                   *sb,
 429                                                                   struct
 430                                                                   reiserfs_journal_cnode
 431                                                                   **table,
 432                                                                   long bl)
 433 {
 434         struct reiserfs_journal_cnode *cn;
 435         cn = journal_hash(table, sb, bl);
 436         while (cn) {
 437                 if (cn->blocknr == bl && cn->sb == sb)
 438                         return cn;
 439                 cn = cn->hnext;
 440         }
 441         return (struct reiserfs_journal_cnode *)0;
 442 }
 443
 444 /*
 445 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 446 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 447 ** being overwritten by a replay after crashing.
 448 **
 449 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 450 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 451 ** sure you never write the block without logging it.
 452 **
 453 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 454 ** when bl is rejected because it is set in a journal list bitmap, we search
 455 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 456 ** through next_zero_bit for find_forward to try.
 457 **
 458 ** Just because we return something in next_zero_bit does not mean we won't
 459 ** reject it on the next call to reiserfs_in_journal
 460 **
 461 */
 462 int reiserfs_in_journal(struct super_block *sb,
 463                         unsigned int bmap_nr, int bit_nr, int search_all,
 464                         b_blocknr_t * next_zero_bit)
 465 {
 466         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 467         struct reiserfs_journal_cnode *cn;
 468         struct reiserfs_list_bitmap *jb;
 469         int i;
 470         unsigned long bl;
 471
 472         *next_zero_bit = 0;     /* always start this at zero. */
 473
 474         PROC_INFO_INC(sb, journal.in_journal);
 475         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 476          ** if we crash before the transaction that freed it commits,  this transaction won't
 477          ** have committed either, and the block will never be written
 478          */
 479         if (search_all) {
 480                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 481                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 482                         jb = journal->j_list_bitmap + i;
 483                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 484                             test_bit(bit_nr,
 485                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 486                                      data)) {
 487                                 *next_zero_bit =
 488                                     find_next_zero_bit((unsigned long *)
 489                                                        (jb->bitmaps[bmap_nr]->
 490                                                         data),
 491                                                        sb->s_blocksize << 3,
 492                                                        bit_nr + 1);
 493                                 return 1;
 494                         }
 495                 }
 496         }
 497
 498         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 499         /* is it in any old transactions? */
 500         if (search_all
 501             && (cn =
 502                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 503                 return 1;
 504         }
 505
 506         /* is it in the current transaction.  This should never happen */
 507         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 508                 BUG();
 509                 return 1;
 510         }
 511
 512         PROC_INFO_INC(sb, journal.in_journal_reusable);
 513         /* safe for reuse */
 514         return 0;
 515 }
 516
 517 /* insert cn into table
 518 */
 519 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 520                                        struct reiserfs_journal_cnode *cn)
 521 {
 522         struct reiserfs_journal_cnode *cn_orig;
 523
 524         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 525         cn->hnext = cn_orig;
 526         cn->hprev = NULL;
 527         if (cn_orig) {
 528                 cn_orig->hprev = cn;
 529         }
 530         journal_hash(table, cn->sb, cn->blocknr) = cn;
 531 }
 532
 533 /* lock the current transaction */
 534 static inline void lock_journal(struct super_block *sb)
 535 {
 536         PROC_INFO_INC(sb, journal.lock_journal);
 537
 538         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 539 }
 540
 541 /* unlock the current transaction */
 542 static inline void unlock_journal(struct super_block *sb)
 543 {
 544         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 545 }
 546
 547 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 548 {
 549         jl->j_refcount++;
 550 }
 551
 552 static inline void put_journal_list(struct super_block *s,
 553                                     struct reiserfs_journal_list *jl)
 554 {
 555         if (jl->j_refcount < 1) {
 556                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 557                                jl->j_trans_id, jl->j_refcount);
 558         }
 559         if (--jl->j_refcount == 0)
 560                 kfree(jl);
 561 }
 562
 563 /*
 564 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 565 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 566 ** transaction.
 567 */
 568 static void cleanup_freed_for_journal_list(struct super_block *sb,
 569                                            struct reiserfs_journal_list *jl)
 570 {
 571
 572         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 573         if (jb) {
 574                 cleanup_bitmap_list(sb, jb);
 575         }
 576         jl->j_list_bitmap->journal_list = NULL;
 577         jl->j_list_bitmap = NULL;
 578 }
 579
 580 static int journal_list_still_alive(struct super_block *s,
 581                                     unsigned int trans_id)
 582 {
 583         struct reiserfs_journal *journal = SB_JOURNAL(s);
 584         struct list_head *entry = &journal->j_journal_list;
 585         struct reiserfs_journal_list *jl;
 586
 587         if (!list_empty(entry)) {
 588                 jl = JOURNAL_LIST_ENTRY(entry->next);
 589                 if (jl->j_trans_id <= trans_id) {
 590                         return 1;
 591                 }
 592         }
 593         return 0;
 594 }
 595
 596 /*
 597  * If page->mapping was null, we failed to truncate this page for
 598  * some reason.  Most likely because it was truncated after being
 599  * logged via data=journal.
 600  *
 601  * This does a check to see if the buffer belongs to one of these
 602  * lost pages before doing the final put_bh.  If page->mapping was
 603  * null, it tries to free buffers on the page, which should make the
 604  * final page_cache_release drop the page from the lru.
 605  */
 606 static void release_buffer_page(struct buffer_head *bh)
 607 {
 608         struct page *page = bh->b_page;
 609         if (!page->mapping && trylock_page(page)) {
 610                 page_cache_get(page);
 611                 put_bh(bh);
 612                 if (!page->mapping)
 613                         try_to_free_buffers(page);
 614                 unlock_page(page);
 615                 page_cache_release(page);
 616         } else {
 617                 put_bh(bh);
 618         }
 619 }
 620
 621 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 622 {
 623         char b[BDEVNAME_SIZE];
 624
 625         if (buffer_journaled(bh)) {
 626                 reiserfs_warning(NULL, "clm-2084",
 627                                  "pinned buffer %lu:%s sent to disk",
 628                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 629         }
 630         if (uptodate)
 631                 set_buffer_uptodate(bh);
 632         else
 633                 clear_buffer_uptodate(bh);
 634
 635         unlock_buffer(bh);
 636         release_buffer_page(bh);
 637 }
 638
 639 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 640 {
 641         if (uptodate)
 642                 set_buffer_uptodate(bh);
 643         else
 644                 clear_buffer_uptodate(bh);
 645         unlock_buffer(bh);
 646         put_bh(bh);
 647 }
 648
 649 static void submit_logged_buffer(struct buffer_head *bh)
 650 {
 651         get_bh(bh);
 652         bh->b_end_io = reiserfs_end_buffer_io_sync;
 653         clear_buffer_journal_new(bh);
 654         clear_buffer_dirty(bh);
 655         if (!test_clear_buffer_journal_test(bh))
 656                 BUG();
 657         if (!buffer_uptodate(bh))
 658                 BUG();
 659         submit_bh(WRITE, bh);
 660 }
 661
 662 static void submit_ordered_buffer(struct buffer_head *bh)
 663 {
 664         get_bh(bh);
 665         bh->b_end_io = reiserfs_end_ordered_io;
 666         clear_buffer_dirty(bh);
 667         if (!buffer_uptodate(bh))
 668                 BUG();
 669         submit_bh(WRITE, bh);
 670 }
 671
 672 #define CHUNK_SIZE 32
 673 struct buffer_chunk {
 674         struct buffer_head *bh[CHUNK_SIZE];
 675         int nr;
 676 };
 677
 678 static void write_chunk(struct buffer_chunk *chunk)
 679 {
 680         int i;
 681         for (i = 0; i < chunk->nr; i++) {
 682                 submit_logged_buffer(chunk->bh[i]);
 683         }
 684         chunk->nr = 0;
 685 }
 686
 687 static void write_ordered_chunk(struct buffer_chunk *chunk)
 688 {
 689         int i;
 690         for (i = 0; i < chunk->nr; i++) {
 691                 submit_ordered_buffer(chunk->bh[i]);
 692         }
 693         chunk->nr = 0;
 694 }
 695
 696 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 697                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 698 {
 699         int ret = 0;
 700         BUG_ON(chunk->nr >= CHUNK_SIZE);
 701         chunk->bh[chunk->nr++] = bh;
 702         if (chunk->nr >= CHUNK_SIZE) {
 703                 ret = 1;
 704                 if (lock)
 705                         spin_unlock(lock);
 706                 fn(chunk);
 707                 if (lock)
 708                         spin_lock(lock);
 709         }
 710         return ret;
 711 }
 712
 713 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 714 static struct reiserfs_jh *alloc_jh(void)
 715 {
 716         struct reiserfs_jh *jh;
 717         while (1) {
 718                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 719                 if (jh) {
 720                         atomic_inc(&nr_reiserfs_jh);
 721                         return jh;
 722                 }
 723                 yield();
 724         }
 725 }
 726
 727 /*
 728  * we want to free the jh when the buffer has been written
 729  * and waited on
 730  */
 731 void reiserfs_free_jh(struct buffer_head *bh)
 732 {
 733         struct reiserfs_jh *jh;
 734
 735         jh = bh->b_private;
 736         if (jh) {
 737                 bh->b_private = NULL;
 738                 jh->bh = NULL;
 739                 list_del_init(&jh->list);
 740                 kfree(jh);
 741                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 742                         BUG();
 743                 atomic_dec(&nr_reiserfs_jh);
 744                 put_bh(bh);
 745         }
 746 }
 747
 748 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 749                            int tail)
 750 {
 751         struct reiserfs_jh *jh;
 752
 753         if (bh->b_private) {
 754                 spin_lock(&j->j_dirty_buffers_lock);
 755                 if (!bh->b_private) {
 756                         spin_unlock(&j->j_dirty_buffers_lock);
 757                         goto no_jh;
 758                 }
 759                 jh = bh->b_private;
 760                 list_del_init(&jh->list);
 761         } else {
 762               no_jh:
 763                 get_bh(bh);
 764                 jh = alloc_jh();
 765                 spin_lock(&j->j_dirty_buffers_lock);
 766                 /* buffer must be locked for __add_jh, should be able to have
 767                  * two adds at the same time
 768                  */
 769                 BUG_ON(bh->b_private);
 770                 jh->bh = bh;
 771                 bh->b_private = jh;
 772         }
 773         jh->jl = j->j_current_jl;
 774         if (tail)
 775                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 776         else {
 777                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 778         }
 779         spin_unlock(&j->j_dirty_buffers_lock);
 780         return 0;
 781 }
 782
 783 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 784 {
 785         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 786 }
 787 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 788 {
 789         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 790 }
 791
 792 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 793 static int write_ordered_buffers(spinlock_t * lock,
 794                                  struct reiserfs_journal *j,
 795                                  struct reiserfs_journal_list *jl,
 796                                  struct list_head *list)
 797 {
 798         struct buffer_head *bh;
 799         struct reiserfs_jh *jh;
 800         int ret = j->j_errno;
 801         struct buffer_chunk chunk;
 802         struct list_head tmp;
 803         INIT_LIST_HEAD(&tmp);
 804
 805         chunk.nr = 0;
 806         spin_lock(lock);
 807         while (!list_empty(list)) {
 808                 jh = JH_ENTRY(list->next);
 809                 bh = jh->bh;
 810                 get_bh(bh);
 811                 if (!trylock_buffer(bh)) {
 812                         if (!buffer_dirty(bh)) {
 813                                 list_move(&jh->list, &tmp);
 814                                 goto loop_next;
 815                         }
 816                         spin_unlock(lock);
 817                         if (chunk.nr)
 818                                 write_ordered_chunk(&chunk);
 819                         wait_on_buffer(bh);
 820                         cond_resched();
 821                         spin_lock(lock);
 822                         goto loop_next;
 823                 }
 824                 /* in theory, dirty non-uptodate buffers should never get here,
 825                  * but the upper layer io error paths still have a few quirks.
 826                  * Handle them here as gracefully as we can
 827                  */
 828                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 829                         clear_buffer_dirty(bh);
 830                         ret = -EIO;
 831                 }
 832                 if (buffer_dirty(bh)) {
 833                         list_move(&jh->list, &tmp);
 834                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 835                 } else {
 836                         reiserfs_free_jh(bh);
 837                         unlock_buffer(bh);
 838                 }
 839               loop_next:
 840                 put_bh(bh);
 841                 cond_resched_lock(lock);
 842         }
 843         if (chunk.nr) {
 844                 spin_unlock(lock);
 845                 write_ordered_chunk(&chunk);
 846                 spin_lock(lock);
 847         }
 848         while (!list_empty(&tmp)) {
 849                 jh = JH_ENTRY(tmp.prev);
 850                 bh = jh->bh;
 851                 get_bh(bh);
 852                 reiserfs_free_jh(bh);
 853
 854                 if (buffer_locked(bh)) {
 855                         spin_unlock(lock);
 856                         wait_on_buffer(bh);
 857                         spin_lock(lock);
 858                 }
 859                 if (!buffer_uptodate(bh)) {
 860                         ret = -EIO;
 861                 }
 862                 /* ugly interaction with invalidatepage here.
 863                  * reiserfs_invalidate_page will pin any buffer that has a valid
 864                  * journal head from an older transaction.  If someone else sets
 865                  * our buffer dirty after we write it in the first loop, and
 866                  * then someone truncates the page away, nobody will ever write
 867                  * the buffer. We're safe if we write the page one last time
 868                  * after freeing the journal header.
 869                  */
 870                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 871                         spin_unlock(lock);
 872                         ll_rw_block(WRITE, 1, &bh);
 873                         spin_lock(lock);
 874                 }
 875                 put_bh(bh);
 876                 cond_resched_lock(lock);
 877         }
 878         spin_unlock(lock);
 879         return ret;
 880 }
 881
 882 static int flush_older_commits(struct super_block *s,
 883                                struct reiserfs_journal_list *jl)
 884 {
 885         struct reiserfs_journal *journal = SB_JOURNAL(s);
 886         struct reiserfs_journal_list *other_jl;
 887         struct reiserfs_journal_list *first_jl;
 888         struct list_head *entry;
 889         unsigned int trans_id = jl->j_trans_id;
 890         unsigned int other_trans_id;
 891         unsigned int first_trans_id;
 892
 893       find_first:
 894         /*
 895          * first we walk backwards to find the oldest uncommitted transation
 896          */
 897         first_jl = jl;
 898         entry = jl->j_list.prev;
 899         while (1) {
 900                 other_jl = JOURNAL_LIST_ENTRY(entry);
 901                 if (entry == &journal->j_journal_list ||
 902                     atomic_read(&other_jl->j_older_commits_done))
 903                         break;
 904
 905                 first_jl = other_jl;
 906                 entry = other_jl->j_list.prev;
 907         }
 908
 909         /* if we didn't find any older uncommitted transactions, return now */
 910         if (first_jl == jl) {
 911                 return 0;
 912         }
 913
 914         first_trans_id = first_jl->j_trans_id;
 915
 916         entry = &first_jl->j_list;
 917         while (1) {
 918                 other_jl = JOURNAL_LIST_ENTRY(entry);
 919                 other_trans_id = other_jl->j_trans_id;
 920
 921                 if (other_trans_id < trans_id) {
 922                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 923                                 flush_commit_list(s, other_jl, 0);
 924
 925                                 /* list we were called with is gone, return */
 926                                 if (!journal_list_still_alive(s, trans_id))
 927                                         return 1;
 928
 929                                 /* the one we just flushed is gone, this means all
 930                                  * older lists are also gone, so first_jl is no longer
 931                                  * valid either.  Go back to the beginning.
 932                                  */
 933                                 if (!journal_list_still_alive
 934                                     (s, other_trans_id)) {
 935                                         goto find_first;
 936                                 }
 937                         }
 938                         entry = entry->next;
 939                         if (entry == &journal->j_journal_list)
 940                                 return 0;
 941                 } else {
 942                         return 0;
 943                 }
 944         }
 945         return 0;
 946 }
 947
 948 static int reiserfs_async_progress_wait(struct super_block *s)
 949 {
 950         struct reiserfs_journal *j = SB_JOURNAL(s);
 951
 952         if (atomic_read(&j->j_async_throttle)) {
 953                 reiserfs_write_unlock(s);
 954                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 955                 reiserfs_write_lock(s);
 956         }
 957
 958         return 0;
 959 }
 960
 961 /*
 962 ** if this journal list still has commit blocks unflushed, send them to disk.
 963 **
 964 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
 965 ** Before the commit block can by written, every other log block must be safely on disk
 966 **
 967 */
 968 static int flush_commit_list(struct super_block *s,
 969                              struct reiserfs_journal_list *jl, int flushall)
 970 {
 971         int i;
 972         b_blocknr_t bn;
 973         struct buffer_head *tbh = NULL;
 974         unsigned int trans_id = jl->j_trans_id;
 975         struct reiserfs_journal *journal = SB_JOURNAL(s);
 976         int retval = 0;
 977         int write_len;
 978
 979         reiserfs_check_lock_depth(s, "flush_commit_list");
 980
 981         if (atomic_read(&jl->j_older_commits_done)) {
 982                 return 0;
 983         }
 984
 985         /* before we can put our commit blocks on disk, we have to make sure everyone older than
 986          ** us is on disk too
 987          */
 988         BUG_ON(jl->j_len <= 0);
 989         BUG_ON(trans_id == journal->j_trans_id);
 990
 991         get_journal_list(jl);
 992         if (flushall) {
 993                 if (flush_older_commits(s, jl) == 1) {
 994                         /* list disappeared during flush_older_commits.  return */
 995                         goto put_jl;
 996                 }
 997         }
 998
 999         /* make sure nobody is trying to flush this one at the same time */
1000         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1001
1002         if (!journal_list_still_alive(s, trans_id)) {
1003                 mutex_unlock(&jl->j_commit_mutex);
1004                 goto put_jl;
1005         }
1006         BUG_ON(jl->j_trans_id == 0);
1007
1008         /* this commit is done, exit */
1009         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1010                 if (flushall) {
1011                         atomic_set(&(jl->j_older_commits_done), 1);
1012                 }
1013                 mutex_unlock(&jl->j_commit_mutex);
1014                 goto put_jl;
1015         }
1016
1017         if (!list_empty(&jl->j_bh_list)) {
1018                 int ret;
1019
1020                 /*
1021                  * We might sleep in numerous places inside
1022                  * write_ordered_buffers. Relax the write lock.
1023                  */
1024                 reiserfs_write_unlock(s);
1025                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1026                                             journal, jl, &jl->j_bh_list);
1027                 if (ret < 0 && retval == 0)
1028                         retval = ret;
1029                 reiserfs_write_lock(s);
1030         }
1031         BUG_ON(!list_empty(&jl->j_bh_list));
1032         /*
1033          * for the description block and all the log blocks, submit any buffers
1034          * that haven't already reached the disk.  Try to write at least 256
1035          * log blocks. later on, we will only wait on blocks that correspond
1036          * to this transaction, but while we're unplugging we might as well
1037          * get a chunk of data on there.
1038          */
1039         atomic_inc(&journal->j_async_throttle);
1040         write_len = jl->j_len + 1;
1041         if (write_len < 256)
1042                 write_len = 256;
1043         for (i = 0 ; i < write_len ; i++) {
1044                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1045                     SB_ONDISK_JOURNAL_SIZE(s);
1046                 tbh = journal_find_get_block(s, bn);
1047                 if (tbh) {
1048                         if (buffer_dirty(tbh)) {
1049                             reiserfs_write_unlock(s);
1050                             ll_rw_block(WRITE, 1, &tbh);
1051                             reiserfs_write_lock(s);
1052                         }
1053                         put_bh(tbh) ;
1054                 }
1055         }
1056         atomic_dec(&journal->j_async_throttle);
1057
1058         for (i = 0; i < (jl->j_len + 1); i++) {
1059                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1060                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1061                 tbh = journal_find_get_block(s, bn);
1062
1063                 reiserfs_write_unlock(s);
1064                 wait_on_buffer(tbh);
1065                 reiserfs_write_lock(s);
1066                 // since we're using ll_rw_blk above, it might have skipped over
1067                 // a locked buffer.  Double check here
1068                 //
1069                 /* redundant, sync_dirty_buffer() checks */
1070                 if (buffer_dirty(tbh)) {
1071                         reiserfs_write_unlock(s);
1072                         sync_dirty_buffer(tbh);
1073                         reiserfs_write_lock(s);
1074                 }
1075                 if (unlikely(!buffer_uptodate(tbh))) {
1076 #ifdef CONFIG_REISERFS_CHECK
1077                         reiserfs_warning(s, "journal-601",
1078                                          "buffer write failed");
1079 #endif
1080                         retval = -EIO;
1081                 }
1082                 put_bh(tbh);    /* once for journal_find_get_block */
1083                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1084                 atomic_dec(&(jl->j_commit_left));
1085         }
1086
1087         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1088
1089         /* If there was a write error in the journal - we can't commit
1090          * this transaction - it will be invalid and, if successful,
1091          * will just end up propagating the write error out to
1092          * the file system. */
1093         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1094                 if (buffer_dirty(jl->j_commit_bh))
1095                         BUG();
1096                 mark_buffer_dirty(jl->j_commit_bh) ;
1097                 reiserfs_write_unlock(s);
1098                 if (reiserfs_barrier_flush(s))
1099                         __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1100                 else
1101                         sync_dirty_buffer(jl->j_commit_bh);
1102                 reiserfs_write_lock(s);
1103         }
1104
1105         /* If there was a write error in the journal - we can't commit this
1106          * transaction - it will be invalid and, if successful, will just end
1107          * up propagating the write error out to the filesystem. */
1108         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1109 #ifdef CONFIG_REISERFS_CHECK
1110                 reiserfs_warning(s, "journal-615", "buffer write failed");
1111 #endif
1112                 retval = -EIO;
1113         }
1114         bforget(jl->j_commit_bh);
1115         if (journal->j_last_commit_id != 0 &&
1116             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1117                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1118                                  journal->j_last_commit_id, jl->j_trans_id);
1119         }
1120         journal->j_last_commit_id = jl->j_trans_id;
1121
1122         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1123         cleanup_freed_for_journal_list(s, jl);
1124
1125         retval = retval ? retval : journal->j_errno;
1126
1127         /* mark the metadata dirty */
1128         if (!retval)
1129                 dirty_one_transaction(s, jl);
1130         atomic_dec(&(jl->j_commit_left));
1131
1132         if (flushall) {
1133                 atomic_set(&(jl->j_older_commits_done), 1);
1134         }
1135         mutex_unlock(&jl->j_commit_mutex);
1136       put_jl:
1137         put_journal_list(s, jl);
1138
1139         if (retval)
1140                 reiserfs_abort(s, retval, "Journal write error in %s",
1141                                __func__);
1142         return retval;
1143 }
1144
1145 /*
1146 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1147 ** returns NULL if it can't find anything
1148 */
1149 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1150                                                           reiserfs_journal_cnode
1151                                                           *cn)
1152 {
1153         struct super_block *sb = cn->sb;
1154         b_blocknr_t blocknr = cn->blocknr;
1155
1156         cn = cn->hprev;
1157         while (cn) {
1158                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1159                         return cn->jlist;
1160                 }
1161                 cn = cn->hprev;
1162         }
1163         return NULL;
1164 }
1165
1166 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1167 {
1168         struct super_block *sb = cn->sb;
1169         b_blocknr_t blocknr = cn->blocknr;
1170
1171         cn = cn->hprev;
1172         while (cn) {
1173                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1174                     atomic_read(&cn->jlist->j_commit_left) != 0)
1175                                     return 0;
1176                 cn = cn->hprev;
1177         }
1178         return 1;
1179 }
1180
1181 static void remove_journal_hash(struct super_block *,
1182                                 struct reiserfs_journal_cnode **,
1183                                 struct reiserfs_journal_list *, unsigned long,
1184                                 int);
1185
1186 /*
1187 ** once all the real blocks have been flushed, it is safe to remove them from the
1188 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1189 ** block to be reallocated for data blocks if it had been deleted.
1190 */
1191 static void remove_all_from_journal_list(struct super_block *sb,
1192                                          struct reiserfs_journal_list *jl,
1193                                          int debug)
1194 {
1195         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1196         struct reiserfs_journal_cnode *cn, *last;
1197         cn = jl->j_realblock;
1198
1199         /* which is better, to lock once around the whole loop, or
1200          ** to lock for each call to remove_journal_hash?
1201          */
1202         while (cn) {
1203                 if (cn->blocknr != 0) {
1204                         if (debug) {
1205                                 reiserfs_warning(sb, "reiserfs-2201",
1206                                                  "block %u, bh is %d, state %ld",
1207                                                  cn->blocknr, cn->bh ? 1 : 0,
1208                                                  cn->state);
1209                         }
1210                         cn->state = 0;
1211                         remove_journal_hash(sb, journal->j_list_hash_table,
1212                                             jl, cn->blocknr, 1);
1213                 }
1214                 last = cn;
1215                 cn = cn->next;
1216                 free_cnode(sb, last);
1217         }
1218         jl->j_realblock = NULL;
1219 }
1220
1221 /*
1222 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1223 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1224 ** releasing blocks in this transaction for reuse as data blocks.
1225 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1226 **
1227 */
1228 static int _update_journal_header_block(struct super_block *sb,
1229                                         unsigned long offset,
1230                                         unsigned int trans_id)
1231 {
1232         struct reiserfs_journal_header *jh;
1233         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1234
1235         if (reiserfs_is_journal_aborted(journal))
1236                 return -EIO;
1237
1238         if (trans_id >= journal->j_last_flush_trans_id) {
1239                 if (buffer_locked((journal->j_header_bh))) {
1240                         reiserfs_write_unlock(sb);
1241                         wait_on_buffer((journal->j_header_bh));
1242                         reiserfs_write_lock(sb);
1243                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1244 #ifdef CONFIG_REISERFS_CHECK
1245                                 reiserfs_warning(sb, "journal-699",
1246                                                  "buffer write failed");
1247 #endif
1248                                 return -EIO;
1249                         }
1250                 }
1251                 journal->j_last_flush_trans_id = trans_id;
1252                 journal->j_first_unflushed_offset = offset;
1253                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1254                                                         b_data);
1255                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1256                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1257                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1258
1259                 set_buffer_dirty(journal->j_header_bh);
1260                 reiserfs_write_unlock(sb);
1261
1262                 if (reiserfs_barrier_flush(sb))
1263                         __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1264                 else
1265                         sync_dirty_buffer(journal->j_header_bh);
1266
1267                 reiserfs_write_lock(sb);
1268                 if (!buffer_uptodate(journal->j_header_bh)) {
1269                         reiserfs_warning(sb, "journal-837",
1270                                          "IO error during journal replay");
1271                         return -EIO;
1272                 }
1273         }
1274         return 0;
1275 }
1276
1277 static int update_journal_header_block(struct super_block *sb,
1278                                        unsigned long offset,
1279                                        unsigned int trans_id)
1280 {
1281         return _update_journal_header_block(sb, offset, trans_id);
1282 }
1283
1284 /*
1285 ** flush any and all journal lists older than you are
1286 ** can only be called from flush_journal_list
1287 */
1288 static int flush_older_journal_lists(struct super_block *sb,
1289                                      struct reiserfs_journal_list *jl)
1290 {
1291         struct list_head *entry;
1292         struct reiserfs_journal_list *other_jl;
1293         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1294         unsigned int trans_id = jl->j_trans_id;
1295
1296         /* we know we are the only ones flushing things, no extra race
1297          * protection is required.
1298          */
1299       restart:
1300         entry = journal->j_journal_list.next;
1301         /* Did we wrap? */
1302         if (entry == &journal->j_journal_list)
1303                 return 0;
1304         other_jl = JOURNAL_LIST_ENTRY(entry);
1305         if (other_jl->j_trans_id < trans_id) {
1306                 BUG_ON(other_jl->j_refcount <= 0);
1307                 /* do not flush all */
1308                 flush_journal_list(sb, other_jl, 0);
1309
1310                 /* other_jl is now deleted from the list */
1311                 goto restart;
1312         }
1313         return 0;
1314 }
1315
1316 static void del_from_work_list(struct super_block *s,
1317                                struct reiserfs_journal_list *jl)
1318 {
1319         struct reiserfs_journal *journal = SB_JOURNAL(s);
1320         if (!list_empty(&jl->j_working_list)) {
1321                 list_del_init(&jl->j_working_list);
1322                 journal->j_num_work_lists--;
1323         }
1324 }
1325
1326 /* flush a journal list, both commit and real blocks
1327 **
1328 ** always set flushall to 1, unless you are calling from inside
1329 ** flush_journal_list
1330 **
1331 ** IMPORTANT.  This can only be called while there are no journal writers,
1332 ** and the journal is locked.  That means it can only be called from
1333 ** do_journal_end, or by journal_release
1334 */
1335 static int flush_journal_list(struct super_block *s,
1336                               struct reiserfs_journal_list *jl, int flushall)
1337 {
1338         struct reiserfs_journal_list *pjl;
1339         struct reiserfs_journal_cnode *cn, *last;
1340         int count;
1341         int was_jwait = 0;
1342         int was_dirty = 0;
1343         struct buffer_head *saved_bh;
1344         unsigned long j_len_saved = jl->j_len;
1345         struct reiserfs_journal *journal = SB_JOURNAL(s);
1346         int err = 0;
1347
1348         BUG_ON(j_len_saved <= 0);
1349
1350         if (atomic_read(&journal->j_wcount) != 0) {
1351                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1352                                  atomic_read(&journal->j_wcount));
1353         }
1354         BUG_ON(jl->j_trans_id == 0);
1355
1356         /* if flushall == 0, the lock is already held */
1357         if (flushall) {
1358                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1359         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1360                 BUG();
1361         }
1362
1363         count = 0;
1364         if (j_len_saved > journal->j_trans_max) {
1365                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1366                                j_len_saved, jl->j_trans_id);
1367                 return 0;
1368         }
1369
1370         /* if all the work is already done, get out of here */
1371         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1372             atomic_read(&(jl->j_commit_left)) <= 0) {
1373                 goto flush_older_and_return;
1374         }
1375
1376         /* start by putting the commit list on disk.  This will also flush
1377          ** the commit lists of any olders transactions
1378          */
1379         flush_commit_list(s, jl, 1);
1380
1381         if (!(jl->j_state & LIST_DIRTY)
1382             && !reiserfs_is_journal_aborted(journal))
1383                 BUG();
1384
1385         /* are we done now? */
1386         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1387             atomic_read(&(jl->j_commit_left)) <= 0) {
1388                 goto flush_older_and_return;
1389         }
1390
1391         /* loop through each cnode, see if we need to write it,
1392          ** or wait on a more recent transaction, or just ignore it
1393          */
1394         if (atomic_read(&(journal->j_wcount)) != 0) {
1395                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1396                                "wcount is not 0");
1397         }
1398         cn = jl->j_realblock;
1399         while (cn) {
1400                 was_jwait = 0;
1401                 was_dirty = 0;
1402                 saved_bh = NULL;
1403                 /* blocknr of 0 is no longer in the hash, ignore it */
1404                 if (cn->blocknr == 0) {
1405                         goto free_cnode;
1406                 }
1407
1408                 /* This transaction failed commit. Don't write out to the disk */
1409                 if (!(jl->j_state & LIST_DIRTY))
1410                         goto free_cnode;
1411
1412                 pjl = find_newer_jl_for_cn(cn);
1413                 /* the order is important here.  We check pjl to make sure we
1414                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1415                  ** block to disk
1416                  */
1417                 if (!pjl && cn->bh) {
1418                         saved_bh = cn->bh;
1419
1420                         /* we do this to make sure nobody releases the buffer while
1421                          ** we are working with it
1422                          */
1423                         get_bh(saved_bh);
1424
1425                         if (buffer_journal_dirty(saved_bh)) {
1426                                 BUG_ON(!can_dirty(cn));
1427                                 was_jwait = 1;
1428                                 was_dirty = 1;
1429                         } else if (can_dirty(cn)) {
1430                                 /* everything with !pjl && jwait should be writable */
1431                                 BUG();
1432                         }
1433                 }
1434
1435                 /* if someone has this block in a newer transaction, just make
1436                  ** sure they are committed, and don't try writing it to disk
1437                  */
1438                 if (pjl) {
1439                         if (atomic_read(&pjl->j_commit_left))
1440                                 flush_commit_list(s, pjl, 1);
1441                         goto free_cnode;
1442                 }
1443
1444                 /* bh == NULL when the block got to disk on its own, OR,
1445                  ** the block got freed in a future transaction
1446                  */
1447                 if (saved_bh == NULL) {
1448                         goto free_cnode;
1449                 }
1450
1451                 /* this should never happen.  kupdate_one_transaction has this list
1452                  ** locked while it works, so we should never see a buffer here that
1453                  ** is not marked JDirty_wait
1454                  */
1455                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1456                         reiserfs_warning(s, "journal-813",
1457                                          "BAD! buffer %llu %cdirty %cjwait, "
1458                                          "not in a newer tranasction",
1459                                          (unsigned long long)saved_bh->
1460                                          b_blocknr, was_dirty ? ' ' : '!',
1461                                          was_jwait ? ' ' : '!');
1462                 }
1463                 if (was_dirty) {
1464                         /* we inc again because saved_bh gets decremented at free_cnode */
1465                         get_bh(saved_bh);
1466                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1467                         lock_buffer(saved_bh);
1468                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1469                         if (buffer_dirty(saved_bh))
1470                                 submit_logged_buffer(saved_bh);
1471                         else
1472                                 unlock_buffer(saved_bh);
1473                         count++;
1474                 } else {
1475                         reiserfs_warning(s, "clm-2082",
1476                                          "Unable to flush buffer %llu in %s",
1477                                          (unsigned long long)saved_bh->
1478                                          b_blocknr, __func__);
1479                 }
1480               free_cnode:
1481                 last = cn;
1482                 cn = cn->next;
1483                 if (saved_bh) {
1484                         /* we incremented this to keep others from taking the buffer head away */
1485                         put_bh(saved_bh);
1486                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1487                                 reiserfs_warning(s, "journal-945",
1488                                                  "saved_bh->b_count < 0");
1489                         }
1490                 }
1491         }
1492         if (count > 0) {
1493                 cn = jl->j_realblock;
1494                 while (cn) {
1495                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1496                                 if (!cn->bh) {
1497                                         reiserfs_panic(s, "journal-1011",
1498                                                        "cn->bh is NULL");
1499                                 }
1500
1501                                 reiserfs_write_unlock(s);
1502                                 wait_on_buffer(cn->bh);
1503                                 reiserfs_write_lock(s);
1504
1505                                 if (!cn->bh) {
1506                                         reiserfs_panic(s, "journal-1012",
1507                                                        "cn->bh is NULL");
1508                                 }
1509                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1510 #ifdef CONFIG_REISERFS_CHECK
1511                                         reiserfs_warning(s, "journal-949",
1512                                                          "buffer write failed");
1513 #endif
1514                                         err = -EIO;
1515                                 }
1516                                 /* note, we must clear the JDirty_wait bit after the up to date
1517                                  ** check, otherwise we race against our flushpage routine
1518                                  */
1519                                 BUG_ON(!test_clear_buffer_journal_dirty
1520                                        (cn->bh));
1521
1522                                 /* drop one ref for us */
1523                                 put_bh(cn->bh);
1524                                 /* drop one ref for journal_mark_dirty */
1525                                 release_buffer_page(cn->bh);
1526                         }
1527                         cn = cn->next;
1528                 }
1529         }
1530
1531         if (err)
1532                 reiserfs_abort(s, -EIO,
1533                                "Write error while pushing transaction to disk in %s",
1534                                __func__);
1535       flush_older_and_return:
1536
1537         /* before we can update the journal header block, we _must_ flush all
1538          ** real blocks from all older transactions to disk.  This is because
1539          ** once the header block is updated, this transaction will not be
1540          ** replayed after a crash
1541          */
1542         if (flushall) {
1543                 flush_older_journal_lists(s, jl);
1544         }
1545
1546         err = journal->j_errno;
1547         /* before we can remove everything from the hash tables for this
1548          ** transaction, we must make sure it can never be replayed
1549          **
1550          ** since we are only called from do_journal_end, we know for sure there
1551          ** are no allocations going on while we are flushing journal lists.  So,
1552          ** we only need to update the journal header block for the last list
1553          ** being flushed
1554          */
1555         if (!err && flushall) {
1556                 err =
1557                     update_journal_header_block(s,
1558                                                 (jl->j_start + jl->j_len +
1559                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1560                                                 jl->j_trans_id);
1561                 if (err)
1562                         reiserfs_abort(s, -EIO,
1563                                        "Write error while updating journal header in %s",
1564                                        __func__);
1565         }
1566         remove_all_from_journal_list(s, jl, 0);
1567         list_del_init(&jl->j_list);
1568         journal->j_num_lists--;
1569         del_from_work_list(s, jl);
1570
1571         if (journal->j_last_flush_id != 0 &&
1572             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1573                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1574                                  journal->j_last_flush_id, jl->j_trans_id);
1575         }
1576         journal->j_last_flush_id = jl->j_trans_id;
1577
1578         /* not strictly required since we are freeing the list, but it should
1579          * help find code using dead lists later on
1580          */
1581         jl->j_len = 0;
1582         atomic_set(&(jl->j_nonzerolen), 0);
1583         jl->j_start = 0;
1584         jl->j_realblock = NULL;
1585         jl->j_commit_bh = NULL;
1586         jl->j_trans_id = 0;
1587         jl->j_state = 0;
1588         put_journal_list(s, jl);
1589         if (flushall)
1590                 mutex_unlock(&journal->j_flush_mutex);
1591         return err;
1592 }
1593
1594 static int test_transaction(struct super_block *s,
1595                             struct reiserfs_journal_list *jl)
1596 {
1597         struct reiserfs_journal_cnode *cn;
1598
1599         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1600                 return 1;
1601
1602         cn = jl->j_realblock;
1603         while (cn) {
1604                 /* if the blocknr == 0, this has been cleared from the hash,
1605                  ** skip it
1606                  */
1607                 if (cn->blocknr == 0) {
1608                         goto next;
1609                 }
1610                 if (cn->bh && !newer_jl_done(cn))
1611                         return 0;
1612               next:
1613                 cn = cn->next;
1614                 cond_resched();
1615         }
1616         return 0;
1617 }
1618
1619 static int write_one_transaction(struct super_block *s,
1620                                  struct reiserfs_journal_list *jl,
1621                                  struct buffer_chunk *chunk)
1622 {
1623         struct reiserfs_journal_cnode *cn;
1624         int ret = 0;
1625
1626         jl->j_state |= LIST_TOUCHED;
1627         del_from_work_list(s, jl);
1628         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1629                 return 0;
1630         }
1631
1632         cn = jl->j_realblock;
1633         while (cn) {
1634                 /* if the blocknr == 0, this has been cleared from the hash,
1635                  ** skip it
1636                  */
1637                 if (cn->blocknr == 0) {
1638                         goto next;
1639                 }
1640                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1641                         struct buffer_head *tmp_bh;
1642                         /* we can race against journal_mark_freed when we try
1643                          * to lock_buffer(cn->bh), so we have to inc the buffer
1644                          * count, and recheck things after locking
1645                          */
1646                         tmp_bh = cn->bh;
1647                         get_bh(tmp_bh);
1648                         lock_buffer(tmp_bh);
1649                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1650                                 if (!buffer_journal_dirty(tmp_bh) ||
1651                                     buffer_journal_prepared(tmp_bh))
1652                                         BUG();
1653                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1654                                 ret++;
1655                         } else {
1656                                 /* note, cn->bh might be null now */
1657                                 unlock_buffer(tmp_bh);
1658                         }
1659                         put_bh(tmp_bh);
1660                 }
1661               next:
1662                 cn = cn->next;
1663                 cond_resched();
1664         }
1665         return ret;
1666 }
1667
1668 /* used by flush_commit_list */
1669 static int dirty_one_transaction(struct super_block *s,
1670                                  struct reiserfs_journal_list *jl)
1671 {
1672         struct reiserfs_journal_cnode *cn;
1673         struct reiserfs_journal_list *pjl;
1674         int ret = 0;
1675
1676         jl->j_state |= LIST_DIRTY;
1677         cn = jl->j_realblock;
1678         while (cn) {
1679                 /* look for a more recent transaction that logged this
1680                  ** buffer.  Only the most recent transaction with a buffer in
1681                  ** it is allowed to send that buffer to disk
1682                  */
1683                 pjl = find_newer_jl_for_cn(cn);
1684                 if (!pjl && cn->blocknr && cn->bh
1685                     && buffer_journal_dirty(cn->bh)) {
1686                         BUG_ON(!can_dirty(cn));
1687                         /* if the buffer is prepared, it will either be logged
1688                          * or restored.  If restored, we need to make sure
1689                          * it actually gets marked dirty
1690                          */
1691                         clear_buffer_journal_new(cn->bh);
1692                         if (buffer_journal_prepared(cn->bh)) {
1693                                 set_buffer_journal_restore_dirty(cn->bh);
1694                         } else {
1695                                 set_buffer_journal_test(cn->bh);
1696                                 mark_buffer_dirty(cn->bh);
1697                         }
1698                 }
1699                 cn = cn->next;
1700         }
1701         return ret;
1702 }
1703
1704 static int kupdate_transactions(struct super_block *s,
1705                                 struct reiserfs_journal_list *jl,
1706                                 struct reiserfs_journal_list **next_jl,
1707                                 unsigned int *next_trans_id,
1708                                 int num_blocks, int num_trans)
1709 {
1710         int ret = 0;
1711         int written = 0;
1712         int transactions_flushed = 0;
1713         unsigned int orig_trans_id = jl->j_trans_id;
1714         struct buffer_chunk chunk;
1715         struct list_head *entry;
1716         struct reiserfs_journal *journal = SB_JOURNAL(s);
1717         chunk.nr = 0;
1718
1719         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1720         if (!journal_list_still_alive(s, orig_trans_id)) {
1721                 goto done;
1722         }
1723
1724         /* we've got j_flush_mutex held, nobody is going to delete any
1725          * of these lists out from underneath us
1726          */
1727         while ((num_trans && transactions_flushed < num_trans) ||
1728                (!num_trans && written < num_blocks)) {
1729
1730                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1731                     atomic_read(&jl->j_commit_left)
1732                     || !(jl->j_state & LIST_DIRTY)) {
1733                         del_from_work_list(s, jl);
1734                         break;
1735                 }
1736                 ret = write_one_transaction(s, jl, &chunk);
1737
1738                 if (ret < 0)
1739                         goto done;
1740                 transactions_flushed++;
1741                 written += ret;
1742                 entry = jl->j_list.next;
1743
1744                 /* did we wrap? */
1745                 if (entry == &journal->j_journal_list) {
1746                         break;
1747                 }
1748                 jl = JOURNAL_LIST_ENTRY(entry);
1749
1750                 /* don't bother with older transactions */
1751                 if (jl->j_trans_id <= orig_trans_id)
1752                         break;
1753         }
1754         if (chunk.nr) {
1755                 write_chunk(&chunk);
1756         }
1757
1758       done:
1759         mutex_unlock(&journal->j_flush_mutex);
1760         return ret;
1761 }
1762
1763 /* for o_sync and fsync heavy applications, they tend to use
1764 ** all the journa list slots with tiny transactions.  These
1765 ** trigger lots and lots of calls to update the header block, which
1766 ** adds seeks and slows things down.
1767 **
1768 ** This function tries to clear out a large chunk of the journal lists
1769 ** at once, which makes everything faster since only the newest journal
1770 ** list updates the header block
1771 */
1772 static int flush_used_journal_lists(struct super_block *s,
1773                                     struct reiserfs_journal_list *jl)
1774 {
1775         unsigned long len = 0;
1776         unsigned long cur_len;
1777         int ret;
1778         int i;
1779         int limit = 256;
1780         struct reiserfs_journal_list *tjl;
1781         struct reiserfs_journal_list *flush_jl;
1782         unsigned int trans_id;
1783         struct reiserfs_journal *journal = SB_JOURNAL(s);
1784
1785         flush_jl = tjl = jl;
1786
1787         /* in data logging mode, try harder to flush a lot of blocks */
1788         if (reiserfs_data_log(s))
1789                 limit = 1024;
1790         /* flush for 256 transactions or limit blocks, whichever comes first */
1791         for (i = 0; i < 256 && len < limit; i++) {
1792                 if (atomic_read(&tjl->j_commit_left) ||
1793                     tjl->j_trans_id < jl->j_trans_id) {
1794                         break;
1795                 }
1796                 cur_len = atomic_read(&tjl->j_nonzerolen);
1797                 if (cur_len > 0) {
1798                         tjl->j_state &= ~LIST_TOUCHED;
1799                 }
1800                 len += cur_len;
1801                 flush_jl = tjl;
1802                 if (tjl->j_list.next == &journal->j_journal_list)
1803                         break;
1804                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1805         }
1806         /* try to find a group of blocks we can flush across all the
1807          ** transactions, but only bother if we've actually spanned
1808          ** across multiple lists
1809          */
1810         if (flush_jl != jl) {
1811                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1812         }
1813         flush_journal_list(s, flush_jl, 1);
1814         return 0;
1815 }
1816
1817 /*
1818 ** removes any nodes in table with name block and dev as bh.
1819 ** only touchs the hnext and hprev pointers.
1820 */
1821 void remove_journal_hash(struct super_block *sb,
1822                          struct reiserfs_journal_cnode **table,
1823                          struct reiserfs_journal_list *jl,
1824                          unsigned long block, int remove_freed)
1825 {
1826         struct reiserfs_journal_cnode *cur;
1827         struct reiserfs_journal_cnode **head;
1828
1829         head = &(journal_hash(table, sb, block));
1830         if (!head) {
1831                 return;
1832         }
1833         cur = *head;
1834         while (cur) {
1835                 if (cur->blocknr == block && cur->sb == sb
1836                     && (jl == NULL || jl == cur->jlist)
1837                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1838                         if (cur->hnext) {
1839                                 cur->hnext->hprev = cur->hprev;
1840                         }
1841                         if (cur->hprev) {
1842                                 cur->hprev->hnext = cur->hnext;
1843                         } else {
1844                                 *head = cur->hnext;
1845                         }
1846                         cur->blocknr = 0;
1847                         cur->sb = NULL;
1848                         cur->state = 0;
1849                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1850                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1851                         cur->bh = NULL;
1852                         cur->jlist = NULL;
1853                 }
1854                 cur = cur->hnext;
1855         }
1856 }
1857
1858 static void free_journal_ram(struct super_block *sb)
1859 {
1860         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1861         kfree(journal->j_current_jl);
1862         journal->j_num_lists--;
1863
1864         vfree(journal->j_cnode_free_orig);
1865         free_list_bitmaps(sb, journal->j_list_bitmap);
1866         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1867         if (journal->j_header_bh) {
1868                 brelse(journal->j_header_bh);
1869         }
1870         /* j_header_bh is on the journal dev, make sure not to release the journal
1871          * dev until we brelse j_header_bh
1872          */
1873         release_journal_dev(sb, journal);
1874         vfree(journal);
1875 }
1876
1877 /*
1878 ** call on unmount.  Only set error to 1 if you haven't made your way out
1879 ** of read_super() yet.  Any other caller must keep error at 0.
1880 */
1881 static int do_journal_release(struct reiserfs_transaction_handle *th,
1882                               struct super_block *sb, int error)
1883 {
1884         struct reiserfs_transaction_handle myth;
1885         int flushed = 0;
1886         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1887
1888         /* we only want to flush out transactions if we were called with error == 0
1889          */
1890         if (!error && !(sb->s_flags & MS_RDONLY)) {
1891                 /* end the current trans */
1892                 BUG_ON(!th->t_trans_id);
1893                 do_journal_end(th, sb, 10, FLUSH_ALL);
1894
1895                 /* make sure something gets logged to force our way into the flush code */
1896                 if (!journal_join(&myth, sb, 1)) {
1897                         reiserfs_prepare_for_journal(sb,
1898                                                      SB_BUFFER_WITH_SB(sb),
1899                                                      1);
1900                         journal_mark_dirty(&myth, sb,
1901                                            SB_BUFFER_WITH_SB(sb));
1902                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1903                         flushed = 1;
1904                 }
1905         }
1906
1907         /* this also catches errors during the do_journal_end above */
1908         if (!error && reiserfs_is_journal_aborted(journal)) {
1909                 memset(&myth, 0, sizeof(myth));
1910                 if (!journal_join_abort(&myth, sb, 1)) {
1911                         reiserfs_prepare_for_journal(sb,
1912                                                      SB_BUFFER_WITH_SB(sb),
1913                                                      1);
1914                         journal_mark_dirty(&myth, sb,
1915                                            SB_BUFFER_WITH_SB(sb));
1916                         do_journal_end(&myth, sb, 1, FLUSH_ALL);
1917                 }
1918         }
1919
1920         reiserfs_mounted_fs_count--;
1921         /* wait for all commits to finish */
1922         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
1923
1924         /*
1925          * We must release the write lock here because
1926          * the workqueue job (flush_async_commit) needs this lock
1927          */
1928         reiserfs_write_unlock(sb);
1929         flush_workqueue(commit_wq);
1930
1931         if (!reiserfs_mounted_fs_count) {
1932                 destroy_workqueue(commit_wq);
1933                 commit_wq = NULL;
1934         }
1935
1936         free_journal_ram(sb);
1937
1938         reiserfs_write_lock(sb);
1939
1940         return 0;
1941 }
1942
1943 /*
1944 ** call on unmount.  flush all journal trans, release all alloc'd ram
1945 */
1946 int journal_release(struct reiserfs_transaction_handle *th,
1947                     struct super_block *sb)
1948 {
1949         return do_journal_release(th, sb, 0);
1950 }
1951
1952 /*
1953 ** only call from an error condition inside reiserfs_read_super!
1954 */
1955 int journal_release_error(struct reiserfs_transaction_handle *th,
1956                           struct super_block *sb)
1957 {
1958         return do_journal_release(th, sb, 1);
1959 }
1960
1961 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
1962 static int journal_compare_desc_commit(struct super_block *sb,
1963                                        struct reiserfs_journal_desc *desc,
1964                                        struct reiserfs_journal_commit *commit)
1965 {
1966         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1967             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1968             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1969             get_commit_trans_len(commit) <= 0) {
1970                 return 1;
1971         }
1972         return 0;
1973 }
1974
1975 /* returns 0 if it did not find a description block
1976 ** returns -1 if it found a corrupt commit block
1977 ** returns 1 if both desc and commit were valid
1978 */
1979 static int journal_transaction_is_valid(struct super_block *sb,
1980                                         struct buffer_head *d_bh,
1981                                         unsigned int *oldest_invalid_trans_id,
1982                                         unsigned long *newest_mount_id)
1983 {
1984         struct reiserfs_journal_desc *desc;
1985         struct reiserfs_journal_commit *commit;
1986         struct buffer_head *c_bh;
1987         unsigned long offset;
1988
1989         if (!d_bh)
1990                 return 0;
1991
1992         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
1993         if (get_desc_trans_len(desc) > 0
1994             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
1995                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
1996                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
1997                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
1998                                        "journal-986: transaction "
1999                                        "is valid returning because trans_id %d is greater than "
2000                                        "oldest_invalid %lu",
2001                                        get_desc_trans_id(desc),
2002                                        *oldest_invalid_trans_id);
2003                         return 0;
2004                 }
2005                 if (newest_mount_id
2006                     && *newest_mount_id > get_desc_mount_id(desc)) {
2007                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2008                                        "journal-1087: transaction "
2009                                        "is valid returning because mount_id %d is less than "
2010                                        "newest_mount_id %lu",
2011                                        get_desc_mount_id(desc),
2012                                        *newest_mount_id);
2013                         return -1;
2014                 }
2015                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2016                         reiserfs_warning(sb, "journal-2018",
2017                                          "Bad transaction length %d "
2018                                          "encountered, ignoring transaction",
2019                                          get_desc_trans_len(desc));
2020                         return -1;
2021                 }
2022                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2023
2024                 /* ok, we have a journal description block, lets see if the transaction was valid */
2025                 c_bh =
2026                     journal_bread(sb,
2027                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2028                                   ((offset + get_desc_trans_len(desc) +
2029                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2030                 if (!c_bh)
2031                         return 0;
2032                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2033                 if (journal_compare_desc_commit(sb, desc, commit)) {
2034                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2035                                        "journal_transaction_is_valid, commit offset %ld had bad "
2036                                        "time %d or length %d",
2037                                        c_bh->b_blocknr -
2038                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2039                                        get_commit_trans_id(commit),
2040                                        get_commit_trans_len(commit));
2041                         brelse(c_bh);
2042                         if (oldest_invalid_trans_id) {
2043                                 *oldest_invalid_trans_id =
2044                                     get_desc_trans_id(desc);
2045                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2046                                                "journal-1004: "
2047                                                "transaction_is_valid setting oldest invalid trans_id "
2048                                                "to %d",
2049                                                get_desc_trans_id(desc));
2050                         }
2051                         return -1;
2052                 }
2053                 brelse(c_bh);
2054                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2055                                "journal-1006: found valid "
2056                                "transaction start offset %llu, len %d id %d",
2057                                d_bh->b_blocknr -
2058                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2059                                get_desc_trans_len(desc),
2060                                get_desc_trans_id(desc));
2061                 return 1;
2062         } else {
2063                 return 0;
2064         }
2065 }
2066
2067 static void brelse_array(struct buffer_head **heads, int num)
2068 {
2069         int i;
2070         for (i = 0; i < num; i++) {
2071                 brelse(heads[i]);
2072         }
2073 }
2074
2075 /*
2076 ** given the start, and values for the oldest acceptable transactions,
2077 ** this either reads in a replays a transaction, or returns because the transaction
2078 ** is invalid, or too old.
2079 */
2080 static int journal_read_transaction(struct super_block *sb,
2081                                     unsigned long cur_dblock,
2082                                     unsigned long oldest_start,
2083                                     unsigned int oldest_trans_id,
2084                                     unsigned long newest_mount_id)
2085 {
2086         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2087         struct reiserfs_journal_desc *desc;
2088         struct reiserfs_journal_commit *commit;
2089         unsigned int trans_id = 0;
2090         struct buffer_head *c_bh;
2091         struct buffer_head *d_bh;
2092         struct buffer_head **log_blocks = NULL;
2093         struct buffer_head **real_blocks = NULL;
2094         unsigned int trans_offset;
2095         int i;
2096         int trans_half;
2097
2098         d_bh = journal_bread(sb, cur_dblock);
2099         if (!d_bh)
2100                 return 1;
2101         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2102         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2103         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2104                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2105                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2106                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2107         if (get_desc_trans_id(desc) < oldest_trans_id) {
2108                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2109                                "journal_read_trans skipping because %lu is too old",
2110                                cur_dblock -
2111                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2112                 brelse(d_bh);
2113                 return 1;
2114         }
2115         if (get_desc_mount_id(desc) != newest_mount_id) {
2116                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2117                                "journal_read_trans skipping because %d is != "
2118                                "newest_mount_id %lu", get_desc_mount_id(desc),
2119                                newest_mount_id);
2120                 brelse(d_bh);
2121                 return 1;
2122         }
2123         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2124                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2125                               SB_ONDISK_JOURNAL_SIZE(sb)));
2126         if (!c_bh) {
2127                 brelse(d_bh);
2128                 return 1;
2129         }
2130         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2131         if (journal_compare_desc_commit(sb, desc, commit)) {
2132                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2133                                "journal_read_transaction, "
2134                                "commit offset %llu had bad time %d or length %d",
2135                                c_bh->b_blocknr -
2136                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2137                                get_commit_trans_id(commit),
2138                                get_commit_trans_len(commit));
2139                 brelse(c_bh);
2140                 brelse(d_bh);
2141                 return 1;
2142         }
2143
2144         if (bdev_read_only(sb->s_bdev)) {
2145                 reiserfs_warning(sb, "clm-2076",
2146                                  "device is readonly, unable to replay log");
2147                 brelse(c_bh);
2148                 brelse(d_bh);
2149                 return -EROFS;
2150         }
2151
2152         trans_id = get_desc_trans_id(desc);
2153         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2154         log_blocks = kmalloc(get_desc_trans_len(desc) *
2155                              sizeof(struct buffer_head *), GFP_NOFS);
2156         real_blocks = kmalloc(get_desc_trans_len(desc) *
2157                               sizeof(struct buffer_head *), GFP_NOFS);
2158         if (!log_blocks || !real_blocks) {
2159                 brelse(c_bh);
2160                 brelse(d_bh);
2161                 kfree(log_blocks);
2162                 kfree(real_blocks);
2163                 reiserfs_warning(sb, "journal-1169",
2164                                  "kmalloc failed, unable to mount FS");
2165                 return -1;
2166         }
2167         /* get all the buffer heads */
2168         trans_half = journal_trans_half(sb->s_blocksize);
2169         for (i = 0; i < get_desc_trans_len(desc); i++) {
2170                 log_blocks[i] =
2171                     journal_getblk(sb,
2172                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2173                                    (trans_offset + 1 +
2174                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2175                 if (i < trans_half) {
2176                         real_blocks[i] =
2177                             sb_getblk(sb,
2178                                       le32_to_cpu(desc->j_realblock[i]));
2179                 } else {
2180                         real_blocks[i] =
2181                             sb_getblk(sb,
2182                                       le32_to_cpu(commit->
2183                                                   j_realblock[i - trans_half]));
2184                 }
2185                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2186                         reiserfs_warning(sb, "journal-1207",
2187                                          "REPLAY FAILURE fsck required! "
2188                                          "Block to replay is outside of "
2189                                          "filesystem");
2190                         goto abort_replay;
2191                 }
2192                 /* make sure we don't try to replay onto log or reserved area */
2193                 if (is_block_in_log_or_reserved_area
2194                     (sb, real_blocks[i]->b_blocknr)) {
2195                         reiserfs_warning(sb, "journal-1204",
2196                                          "REPLAY FAILURE fsck required! "
2197                                          "Trying to replay onto a log block");
2198                       abort_replay:
2199                         brelse_array(log_blocks, i);
2200                         brelse_array(real_blocks, i);
2201                         brelse(c_bh);
2202                         brelse(d_bh);
2203                         kfree(log_blocks);
2204                         kfree(real_blocks);
2205                         return -1;
2206                 }
2207         }
2208         /* read in the log blocks, memcpy to the corresponding real block */
2209         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2210         for (i = 0; i < get_desc_trans_len(desc); i++) {
2211
2212                 reiserfs_write_unlock(sb);
2213                 wait_on_buffer(log_blocks[i]);
2214                 reiserfs_write_lock(sb);
2215
2216                 if (!buffer_uptodate(log_blocks[i])) {
2217                         reiserfs_warning(sb, "journal-1212",
2218                                          "REPLAY FAILURE fsck required! "
2219                                          "buffer write failed");
2220                         brelse_array(log_blocks + i,
2221                                      get_desc_trans_len(desc) - i);
2222                         brelse_array(real_blocks, get_desc_trans_len(desc));
2223                         brelse(c_bh);
2224                         brelse(d_bh);
2225                         kfree(log_blocks);
2226                         kfree(real_blocks);
2227                         return -1;
2228                 }
2229                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2230                        real_blocks[i]->b_size);
2231                 set_buffer_uptodate(real_blocks[i]);
2232                 brelse(log_blocks[i]);
2233         }
2234         /* flush out the real blocks */
2235         for (i = 0; i < get_desc_trans_len(desc); i++) {
2236                 set_buffer_dirty(real_blocks[i]);
2237                 write_dirty_buffer(real_blocks[i], WRITE);
2238         }
2239         for (i = 0; i < get_desc_trans_len(desc); i++) {
2240                 wait_on_buffer(real_blocks[i]);
2241                 if (!buffer_uptodate(real_blocks[i])) {
2242                         reiserfs_warning(sb, "journal-1226",
2243                                          "REPLAY FAILURE, fsck required! "
2244                                          "buffer write failed");
2245                         brelse_array(real_blocks + i,
2246                                      get_desc_trans_len(desc) - i);
2247                         brelse(c_bh);
2248                         brelse(d_bh);
2249                         kfree(log_blocks);
2250                         kfree(real_blocks);
2251                         return -1;
2252                 }
2253                 brelse(real_blocks[i]);
2254         }
2255         cur_dblock =
2256             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2257             ((trans_offset + get_desc_trans_len(desc) +
2258               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2259         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2260                        "journal-1095: setting journal " "start to offset %ld",
2261                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2262
2263         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2264         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2265         journal->j_last_flush_trans_id = trans_id;
2266         journal->j_trans_id = trans_id + 1;
2267         /* check for trans_id overflow */
2268         if (journal->j_trans_id == 0)
2269                 journal->j_trans_id = 10;
2270         brelse(c_bh);
2271         brelse(d_bh);
2272         kfree(log_blocks);
2273         kfree(real_blocks);
2274         return 0;
2275 }
2276
2277 /* This function reads blocks starting from block and to max_block of bufsize
2278    size (but no more than BUFNR blocks at a time). This proved to improve
2279    mounting speed on self-rebuilding raid5 arrays at least.
2280    Right now it is only used from journal code. But later we might use it
2281    from other places.
2282    Note: Do not use journal_getblk/sb_getblk functions here! */
2283 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2284                                            b_blocknr_t block, int bufsize,
2285                                            b_blocknr_t max_block)
2286 {
2287         struct buffer_head *bhlist[BUFNR];
2288         unsigned int blocks = BUFNR;
2289         struct buffer_head *bh;
2290         int i, j;
2291
2292         bh = __getblk(dev, block, bufsize);
2293         if (buffer_uptodate(bh))
2294                 return (bh);
2295
2296         if (block + BUFNR > max_block) {
2297                 blocks = max_block - block;
2298         }
2299         bhlist[0] = bh;
2300         j = 1;
2301         for (i = 1; i < blocks; i++) {
2302                 bh = __getblk(dev, block + i, bufsize);
2303                 if (buffer_uptodate(bh)) {
2304                         brelse(bh);
2305                         break;
2306                 } else
2307                         bhlist[j++] = bh;
2308         }
2309         ll_rw_block(READ, j, bhlist);
2310         for (i = 1; i < j; i++)
2311                 brelse(bhlist[i]);
2312         bh = bhlist[0];
2313         wait_on_buffer(bh);
2314         if (buffer_uptodate(bh))
2315                 return bh;
2316         brelse(bh);
2317         return NULL;
2318 }
2319
2320 /*
2321 ** read and replay the log
2322 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2323 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2324 **
2325 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2326 **
2327 ** On exit, it sets things up so the first transaction will work correctly.
2328 */
2329 static int journal_read(struct super_block *sb)
2330 {
2331         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2332         struct reiserfs_journal_desc *desc;
2333         unsigned int oldest_trans_id = 0;
2334         unsigned int oldest_invalid_trans_id = 0;
2335         time_t start;
2336         unsigned long oldest_start = 0;
2337         unsigned long cur_dblock = 0;
2338         unsigned long newest_mount_id = 9;
2339         struct buffer_head *d_bh;
2340         struct reiserfs_journal_header *jh;
2341         int valid_journal_header = 0;
2342         int replay_count = 0;
2343         int continue_replay = 1;
2344         int ret;
2345         char b[BDEVNAME_SIZE];
2346
2347         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2348         reiserfs_info(sb, "checking transaction log (%s)\n",
2349                       bdevname(journal->j_dev_bd, b));
2350         start = get_seconds();
2351
2352         /* step 1, read in the journal header block.  Check the transaction it says
2353          ** is the first unflushed, and if that transaction is not valid,
2354          ** replay is done
2355          */
2356         journal->j_header_bh = journal_bread(sb,
2357                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2358                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2359         if (!journal->j_header_bh) {
2360                 return 1;
2361         }
2362         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2363         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2364             SB_ONDISK_JOURNAL_SIZE(sb)
2365             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2366                 oldest_start =
2367                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2368                     le32_to_cpu(jh->j_first_unflushed_offset);
2369                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2370                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2371                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2372                                "journal-1153: found in "
2373                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2374                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2375                                le32_to_cpu(jh->j_last_flush_trans_id));
2376                 valid_journal_header = 1;
2377
2378                 /* now, we try to read the first unflushed offset.  If it is not valid,
2379                  ** there is nothing more we can do, and it makes no sense to read
2380                  ** through the whole log.
2381                  */
2382                 d_bh =
2383                     journal_bread(sb,
2384                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2385                                   le32_to_cpu(jh->j_first_unflushed_offset));
2386                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2387                 if (!ret) {
2388                         continue_replay = 0;
2389                 }
2390                 brelse(d_bh);
2391                 goto start_log_replay;
2392         }
2393
2394         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2395          ** all the valid transactions, and pick out the oldest.
2396          */
2397         while (continue_replay
2398                && cur_dblock <
2399                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2400                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2401                 /* Note that it is required for blocksize of primary fs device and journal
2402                    device to be the same */
2403                 d_bh =
2404                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2405                                     sb->s_blocksize,
2406                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2407                                     SB_ONDISK_JOURNAL_SIZE(sb));
2408                 ret =
2409                     journal_transaction_is_valid(sb, d_bh,
2410                                                  &oldest_invalid_trans_id,
2411                                                  &newest_mount_id);
2412                 if (ret == 1) {
2413                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2414                         if (oldest_start == 0) {        /* init all oldest_ values */
2415                                 oldest_trans_id = get_desc_trans_id(desc);
2416                                 oldest_start = d_bh->b_blocknr;
2417                                 newest_mount_id = get_desc_mount_id(desc);
2418                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2419                                                "journal-1179: Setting "
2420                                                "oldest_start to offset %llu, trans_id %lu",
2421                                                oldest_start -
2422                                                SB_ONDISK_JOURNAL_1st_BLOCK
2423                                                (sb), oldest_trans_id);
2424                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2425                                 /* one we just read was older */
2426                                 oldest_trans_id = get_desc_trans_id(desc);
2427                                 oldest_start = d_bh->b_blocknr;
2428                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2429                                                "journal-1180: Resetting "
2430                                                "oldest_start to offset %lu, trans_id %lu",
2431                                                oldest_start -
2432                                                SB_ONDISK_JOURNAL_1st_BLOCK
2433                                                (sb), oldest_trans_id);
2434                         }
2435                         if (newest_mount_id < get_desc_mount_id(desc)) {
2436                                 newest_mount_id = get_desc_mount_id(desc);
2437                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2438                                                "journal-1299: Setting "
2439                                                "newest_mount_id to %d",
2440                                                get_desc_mount_id(desc));
2441                         }
2442                         cur_dblock += get_desc_trans_len(desc) + 2;
2443                 } else {
2444                         cur_dblock++;
2445                 }
2446                 brelse(d_bh);
2447         }
2448
2449       start_log_replay:
2450         cur_dblock = oldest_start;
2451         if (oldest_trans_id) {
2452                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2453                                "journal-1206: Starting replay "
2454                                "from offset %llu, trans_id %lu",
2455                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2456                                oldest_trans_id);
2457
2458         }
2459         replay_count = 0;
2460         while (continue_replay && oldest_trans_id > 0) {
2461                 ret =
2462                     journal_read_transaction(sb, cur_dblock, oldest_start,
2463                                              oldest_trans_id, newest_mount_id);
2464                 if (ret < 0) {
2465                         return ret;
2466                 } else if (ret != 0) {
2467                         break;
2468                 }
2469                 cur_dblock =
2470                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2471                 replay_count++;
2472                 if (cur_dblock == oldest_start)
2473                         break;
2474         }
2475
2476         if (oldest_trans_id == 0) {
2477                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2478                                "journal-1225: No valid " "transactions found");
2479         }
2480         /* j_start does not get set correctly if we don't replay any transactions.
2481          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2482          ** copy the trans_id from the header
2483          */
2484         if (valid_journal_header && replay_count == 0) {
2485                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2486                 journal->j_trans_id =
2487                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2488                 /* check for trans_id overflow */
2489                 if (journal->j_trans_id == 0)
2490                         journal->j_trans_id = 10;
2491                 journal->j_last_flush_trans_id =
2492                     le32_to_cpu(jh->j_last_flush_trans_id);
2493                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2494         } else {
2495                 journal->j_mount_id = newest_mount_id + 1;
2496         }
2497         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2498                        "newest_mount_id to %lu", journal->j_mount_id);
2499         journal->j_first_unflushed_offset = journal->j_start;
2500         if (replay_count > 0) {
2501                 reiserfs_info(sb,
2502                               "replayed %d transactions in %lu seconds\n",
2503                               replay_count, get_seconds() - start);
2504         }
2505         if (!bdev_read_only(sb->s_bdev) &&
2506             _update_journal_header_block(sb, journal->j_start,
2507                                          journal->j_last_flush_trans_id)) {
2508                 /* replay failed, caller must call free_journal_ram and abort
2509                  ** the mount
2510                  */
2511                 return -1;
2512         }
2513         return 0;
2514 }
2515
2516 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2517 {
2518         struct reiserfs_journal_list *jl;
2519         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2520                      GFP_NOFS | __GFP_NOFAIL);
2521         INIT_LIST_HEAD(&jl->j_list);
2522         INIT_LIST_HEAD(&jl->j_working_list);
2523         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2524         INIT_LIST_HEAD(&jl->j_bh_list);
2525         mutex_init(&jl->j_commit_mutex);
2526         SB_JOURNAL(s)->j_num_lists++;
2527         get_journal_list(jl);
2528         return jl;
2529 }
2530
2531 static void journal_list_init(struct super_block *sb)
2532 {
2533         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2534 }
2535
2536 static int release_journal_dev(struct super_block *super,
2537                                struct reiserfs_journal *journal)
2538 {
2539         int result;
2540
2541         result = 0;
2542
2543         if (journal->j_dev_bd != NULL) {
2544                 result = blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2545                 journal->j_dev_bd = NULL;
2546         }
2547
2548         if (result != 0) {
2549                 reiserfs_warning(super, "sh-457",
2550                                  "Cannot release journal device: %i", result);
2551         }
2552         return result;
2553 }
2554
2555 static int journal_init_dev(struct super_block *super,
2556                             struct reiserfs_journal *journal,
2557                             const char *jdev_name)
2558 {
2559         int result;
2560         dev_t jdev;
2561         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2562         char b[BDEVNAME_SIZE];
2563
2564         result = 0;
2565
2566         journal->j_dev_bd = NULL;
2567         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2568             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2569
2570         if (bdev_read_only(super->s_bdev))
2571                 blkdev_mode = FMODE_READ;
2572
2573         /* there is no "jdev" option and journal is on separate device */
2574         if ((!jdev_name || !jdev_name[0])) {
2575                 if (jdev == super->s_dev)
2576                         blkdev_mode &= ~FMODE_EXCL;
2577                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2578                                                       journal);
2579                 journal->j_dev_mode = blkdev_mode;
2580                 if (IS_ERR(journal->j_dev_bd)) {
2581                         result = PTR_ERR(journal->j_dev_bd);
2582                         journal->j_dev_bd = NULL;
2583                         reiserfs_warning(super, "sh-458",
2584                                          "cannot init journal device '%s': %i",
2585                                          __bdevname(jdev, b), result);
2586                         return result;
2587                 } else if (jdev != super->s_dev)
2588                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2589
2590                 return 0;
2591         }
2592
2593         journal->j_dev_mode = blkdev_mode;
2594         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2595         if (IS_ERR(journal->j_dev_bd)) {
2596                 result = PTR_ERR(journal->j_dev_bd);
2597                 journal->j_dev_bd = NULL;
2598                 reiserfs_warning(super,
2599                                  "journal_init_dev: Cannot open '%s': %i",
2600                                  jdev_name, result);
2601                 return result;
2602         }
2603
2604         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2605         reiserfs_info(super,
2606                       "journal_init_dev: journal device: %s\n",
2607                       bdevname(journal->j_dev_bd, b));
2608         return 0;
2609 }
2610
2611 /**
2612  * When creating/tuning a file system user can assign some
2613  * journal params within boundaries which depend on the ratio
2614  * blocksize/standard_blocksize.
2615  *
2616  * For blocks >= standard_blocksize transaction size should
2617  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2618  * then JOURNAL_TRANS_MAX_DEFAULT.
2619  *
2620  * For blocks < standard_blocksize these boundaries should be
2621  * decreased proportionally.
2622  */
2623 #define REISERFS_STANDARD_BLKSIZE (4096)
2624
2625 static int check_advise_trans_params(struct super_block *sb,
2626                                      struct reiserfs_journal *journal)
2627 {
2628         if (journal->j_trans_max) {
2629                 /* Non-default journal params.
2630                    Do sanity check for them. */
2631                 int ratio = 1;
2632                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2633                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2634
2635                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2636                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2637                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2638                     JOURNAL_MIN_RATIO) {
2639                         reiserfs_warning(sb, "sh-462",
2640                                          "bad transaction max size (%u). "
2641                                          "FSCK?", journal->j_trans_max);
2642                         return 1;
2643                 }
2644                 if (journal->j_max_batch != (journal->j_trans_max) *
2645                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2646                         reiserfs_warning(sb, "sh-463",
2647                                          "bad transaction max batch (%u). "
2648                                          "FSCK?", journal->j_max_batch);
2649                         return 1;
2650                 }
2651         } else {
2652                 /* Default journal params.
2653                    The file system was created by old version
2654                    of mkreiserfs, so some fields contain zeros,
2655                    and we need to advise proper values for them */
2656                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2657                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2658                                          sb->s_blocksize);
2659                         return 1;
2660                 }
2661                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2662                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2663                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2664         }
2665         return 0;
2666 }
2667
2668 /*
2669 ** must be called once on fs mount.  calls journal_read for you
2670 */
2671 int journal_init(struct super_block *sb, const char *j_dev_name,
2672                  int old_format, unsigned int commit_max_age)
2673 {
2674         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2675         struct buffer_head *bhjh;
2676         struct reiserfs_super_block *rs;
2677         struct reiserfs_journal_header *jh;
2678         struct reiserfs_journal *journal;
2679         struct reiserfs_journal_list *jl;
2680         char b[BDEVNAME_SIZE];
2681         int ret;
2682
2683         /*
2684          * Unlock here to avoid various RECLAIM-FS-ON <-> IN-RECLAIM-FS
2685          * dependency inversion warnings.
2686          */
2687         reiserfs_write_unlock(sb);
2688         journal = SB_JOURNAL(sb) = vmalloc(sizeof(struct reiserfs_journal));
2689         if (!journal) {
2690                 reiserfs_warning(sb, "journal-1256",
2691                                  "unable to get memory for journal structure");
2692                 reiserfs_write_lock(sb);
2693                 return 1;
2694         }
2695         memset(journal, 0, sizeof(struct reiserfs_journal));
2696         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2697         INIT_LIST_HEAD(&journal->j_prealloc_list);
2698         INIT_LIST_HEAD(&journal->j_working_list);
2699         INIT_LIST_HEAD(&journal->j_journal_list);
2700         journal->j_persistent_trans = 0;
2701         ret = reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2702                                            reiserfs_bmap_count(sb));
2703         reiserfs_write_lock(sb);
2704         if (ret)
2705                 goto free_and_return;
2706
2707         allocate_bitmap_nodes(sb);
2708
2709         /* reserved for journal area support */
2710         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2711                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2712                                                  / sb->s_blocksize +
2713                                                  reiserfs_bmap_count(sb) +
2714                                                  1 :
2715                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2716                                                  sb->s_blocksize + 2);
2717
2718         /* Sanity check to see is the standard journal fitting within first bitmap
2719            (actual for small blocksizes) */
2720         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2721             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2722              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2723                 reiserfs_warning(sb, "journal-1393",
2724                                  "journal does not fit for area addressed "
2725                                  "by first of bitmap blocks. It starts at "
2726                                  "%u and its size is %u. Block size %ld",
2727                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2728                                  SB_ONDISK_JOURNAL_SIZE(sb),
2729                                  sb->s_blocksize);
2730                 goto free_and_return;
2731         }
2732
2733         /*
2734          * We need to unlock here to avoid creating the following
2735          * dependency:
2736          * reiserfs_lock -> sysfs_mutex
2737          * Because the reiserfs mmap path creates the following dependency:
2738          * mm->mmap -> reiserfs_lock, hence we have
2739          * mm->mmap -> reiserfs_lock ->sysfs_mutex
2740          * This would ends up in a circular dependency with sysfs readdir path
2741          * which does sysfs_mutex -> mm->mmap_sem
2742          * This is fine because the reiserfs lock is useless in mount path,
2743          * at least until we call journal_begin. We keep it for paranoid
2744          * reasons.
2745          */
2746         reiserfs_write_unlock(sb);
2747         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2748                 reiserfs_write_lock(sb);
2749                 reiserfs_warning(sb, "sh-462",
2750                                  "unable to initialize jornal device");
2751                 goto free_and_return;
2752         }
2753         reiserfs_write_lock(sb);
2754
2755         rs = SB_DISK_SUPER_BLOCK(sb);
2756
2757         /* read journal header */
2758         bhjh = journal_bread(sb,
2759                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2760                              SB_ONDISK_JOURNAL_SIZE(sb));
2761         if (!bhjh) {
2762                 reiserfs_warning(sb, "sh-459",
2763                                  "unable to read journal header");
2764                 goto free_and_return;
2765         }
2766         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2767
2768         /* make sure that journal matches to the super block */
2769         if (is_reiserfs_jr(rs)
2770             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2771                 sb_jp_journal_magic(rs))) {
2772                 reiserfs_warning(sb, "sh-460",
2773                                  "journal header magic %x (device %s) does "
2774                                  "not match to magic found in super block %x",
2775                                  jh->jh_journal.jp_journal_magic,
2776                                  bdevname(journal->j_dev_bd, b),
2777                                  sb_jp_journal_magic(rs));
2778                 brelse(bhjh);
2779                 goto free_and_return;
2780         }
2781
2782         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2783         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2784         journal->j_max_commit_age =
2785             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2786         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2787
2788         if (check_advise_trans_params(sb, journal) != 0)
2789                 goto free_and_return;
2790         journal->j_default_max_commit_age = journal->j_max_commit_age;
2791
2792         if (commit_max_age != 0) {
2793                 journal->j_max_commit_age = commit_max_age;
2794                 journal->j_max_trans_age = commit_max_age;
2795         }
2796
2797         reiserfs_info(sb, "journal params: device %s, size %u, "
2798                       "journal first block %u, max trans len %u, max batch %u, "
2799                       "max commit age %u, max trans age %u\n",
2800                       bdevname(journal->j_dev_bd, b),
2801                       SB_ONDISK_JOURNAL_SIZE(sb),
2802                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2803                       journal->j_trans_max,
2804                       journal->j_max_batch,
2805                       journal->j_max_commit_age, journal->j_max_trans_age);
2806
2807         brelse(bhjh);
2808
2809         journal->j_list_bitmap_index = 0;
2810         journal_list_init(sb);
2811
2812         memset(journal->j_list_hash_table, 0,
2813                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2814
2815         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2816         spin_lock_init(&journal->j_dirty_buffers_lock);
2817
2818         journal->j_start = 0;
2819         journal->j_len = 0;
2820         journal->j_len_alloc = 0;
2821         atomic_set(&(journal->j_wcount), 0);
2822         atomic_set(&(journal->j_async_throttle), 0);
2823         journal->j_bcount = 0;
2824         journal->j_trans_start_time = 0;
2825         journal->j_last = NULL;
2826         journal->j_first = NULL;
2827         init_waitqueue_head(&(journal->j_join_wait));
2828         mutex_init(&journal->j_mutex);
2829         mutex_init(&journal->j_flush_mutex);
2830
2831         journal->j_trans_id = 10;
2832         journal->j_mount_id = 10;
2833         journal->j_state = 0;
2834         atomic_set(&(journal->j_jlock), 0);
2835         reiserfs_write_unlock(sb);
2836         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2837         reiserfs_write_lock(sb);
2838         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2839         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2840         journal->j_cnode_used = 0;
2841         journal->j_must_wait = 0;
2842
2843         if (journal->j_cnode_free == 0) {
2844                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2845                                  "allocation failed (%ld bytes). Journal is "
2846                                  "too large for available memory. Usually "
2847                                  "this is due to a journal that is too large.",
2848                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2849                 goto free_and_return;
2850         }
2851
2852         init_journal_hash(sb);
2853         jl = journal->j_current_jl;
2854         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2855         if (!jl->j_list_bitmap) {
2856                 reiserfs_warning(sb, "journal-2005",
2857                                  "get_list_bitmap failed for journal list 0");
2858                 goto free_and_return;
2859         }
2860         if (journal_read(sb) < 0) {
2861                 reiserfs_warning(sb, "reiserfs-2006",
2862                                  "Replay Failure, unable to mount");
2863                 goto free_and_return;
2864         }
2865
2866         reiserfs_mounted_fs_count++;
2867         if (reiserfs_mounted_fs_count <= 1) {
2868                 reiserfs_write_unlock(sb);
2869                 commit_wq = alloc_workqueue("reiserfs", WQ_MEM_RECLAIM, 0);
2870                 reiserfs_write_lock(sb);
2871         }
2872
2873         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2874         journal->j_work_sb = sb;
2875         return 0;
2876       free_and_return:
2877         free_journal_ram(sb);
2878         return 1;
2879 }
2880
2881 /*
2882 ** test for a polite end of the current transaction.  Used by file_write, and should
2883 ** be used by delete to make sure they don't write more than can fit inside a single
2884 ** transaction
2885 */
2886 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2887                                    int new_alloc)
2888 {
2889         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2890         time_t now = get_seconds();
2891         /* cannot restart while nested */
2892         BUG_ON(!th->t_trans_id);
2893         if (th->t_refcount > 1)
2894                 return 0;
2895         if (journal->j_must_wait > 0 ||
2896             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2897             atomic_read(&(journal->j_jlock)) ||
2898             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2899             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2900                 return 1;
2901         }
2902         /* protected by the BKL here */
2903         journal->j_len_alloc += new_alloc;
2904         th->t_blocks_allocated += new_alloc ;
2905         return 0;
2906 }
2907
2908 /* this must be called inside a transaction, and requires the
2909 ** kernel_lock to be held
2910 */
2911 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2912 {
2913         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2914         BUG_ON(!th->t_trans_id);
2915         journal->j_must_wait = 1;
2916         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2917         return;
2918 }
2919
2920 /* this must be called without a transaction started, and does not
2921 ** require BKL
2922 */
2923 void reiserfs_allow_writes(struct super_block *s)
2924 {
2925         struct reiserfs_journal *journal = SB_JOURNAL(s);
2926         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2927         wake_up(&journal->j_join_wait);
2928 }
2929
2930 /* this must be called without a transaction started, and does not
2931 ** require BKL
2932 */
2933 void reiserfs_wait_on_write_block(struct super_block *s)
2934 {
2935         struct reiserfs_journal *journal = SB_JOURNAL(s);
2936         wait_event(journal->j_join_wait,
2937                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2938 }
2939
2940 static void queue_log_writer(struct super_block *s)
2941 {
2942         wait_queue_t wait;
2943         struct reiserfs_journal *journal = SB_JOURNAL(s);
2944         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2945
2946         /*
2947          * we don't want to use wait_event here because
2948          * we only want to wait once.
2949          */
2950         init_waitqueue_entry(&wait, current);
2951         add_wait_queue(&journal->j_join_wait, &wait);
2952         set_current_state(TASK_UNINTERRUPTIBLE);
2953         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2954                 reiserfs_write_unlock(s);
2955                 schedule();
2956                 reiserfs_write_lock(s);
2957         }
2958         __set_current_state(TASK_RUNNING);
2959         remove_wait_queue(&journal->j_join_wait, &wait);
2960 }
2961
2962 static void wake_queued_writers(struct super_block *s)
2963 {
2964         struct reiserfs_journal *journal = SB_JOURNAL(s);
2965         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2966                 wake_up(&journal->j_join_wait);
2967 }
2968
2969 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2970 {
2971         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2972         unsigned long bcount = journal->j_bcount;
2973         while (1) {
2974                 reiserfs_write_unlock(sb);
2975                 schedule_timeout_uninterruptible(1);
2976                 reiserfs_write_lock(sb);
2977                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2978                 while ((atomic_read(&journal->j_wcount) > 0 ||
2979                         atomic_read(&journal->j_jlock)) &&
2980                        journal->j_trans_id == trans_id) {
2981                         queue_log_writer(sb);
2982                 }
2983                 if (journal->j_trans_id != trans_id)
2984                         break;
2985                 if (bcount == journal->j_bcount)
2986                         break;
2987                 bcount = journal->j_bcount;
2988         }
2989 }
2990
2991 /* join == true if you must join an existing transaction.
2992 ** join == false if you can deal with waiting for others to finish
2993 **
2994 ** this will block until the transaction is joinable.  send the number of blocks you
2995 ** expect to use in nblocks.
2996 */
2997 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
2998                               struct super_block *sb, unsigned long nblocks,
2999                               int join)
3000 {
3001         time_t now = get_seconds();
3002         unsigned int old_trans_id;
3003         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3004         struct reiserfs_transaction_handle myth;
3005         int sched_count = 0;
3006         int retval;
3007
3008         reiserfs_check_lock_depth(sb, "journal_begin");
3009         BUG_ON(nblocks > journal->j_trans_max);
3010
3011         PROC_INFO_INC(sb, journal.journal_being);
3012         /* set here for journal_join */
3013         th->t_refcount = 1;
3014         th->t_super = sb;
3015
3016       relock:
3017         lock_journal(sb);
3018         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3019                 unlock_journal(sb);
3020                 retval = journal->j_errno;
3021                 goto out_fail;
3022         }
3023         journal->j_bcount++;
3024
3025         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3026                 unlock_journal(sb);
3027                 reiserfs_write_unlock(sb);
3028                 reiserfs_wait_on_write_block(sb);
3029                 reiserfs_write_lock(sb);
3030                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3031                 goto relock;
3032         }
3033         now = get_seconds();
3034
3035         /* if there is no room in the journal OR
3036          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3037          ** we don't sleep if there aren't other writers
3038          */
3039
3040         if ((!join && journal->j_must_wait > 0) ||
3041             (!join
3042              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3043             || (!join && atomic_read(&journal->j_wcount) > 0
3044                 && journal->j_trans_start_time > 0
3045                 && (now - journal->j_trans_start_time) >
3046                 journal->j_max_trans_age) || (!join
3047                                               && atomic_read(&journal->j_jlock))
3048             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3049
3050                 old_trans_id = journal->j_trans_id;
3051                 unlock_journal(sb);     /* allow others to finish this transaction */
3052
3053                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3054                     journal->j_max_batch &&
3055                     ((journal->j_len + nblocks + 2) * 100) <
3056                     (journal->j_len_alloc * 75)) {
3057                         if (atomic_read(&journal->j_wcount) > 10) {
3058                                 sched_count++;
3059                                 queue_log_writer(sb);
3060                                 goto relock;
3061                         }
3062                 }
3063                 /* don't mess with joining the transaction if all we have to do is
3064                  * wait for someone else to do a commit
3065                  */
3066                 if (atomic_read(&journal->j_jlock)) {
3067                         while (journal->j_trans_id == old_trans_id &&
3068                                atomic_read(&journal->j_jlock)) {
3069                                 queue_log_writer(sb);
3070                         }
3071                         goto relock;
3072                 }
3073                 retval = journal_join(&myth, sb, 1);
3074                 if (retval)
3075                         goto out_fail;
3076
3077                 /* someone might have ended the transaction while we joined */
3078                 if (old_trans_id != journal->j_trans_id) {
3079                         retval = do_journal_end(&myth, sb, 1, 0);
3080                 } else {
3081                         retval = do_journal_end(&myth, sb, 1, COMMIT_NOW);
3082                 }
3083
3084                 if (retval)
3085                         goto out_fail;
3086
3087                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3088                 goto relock;
3089         }
3090         /* we are the first writer, set trans_id */
3091         if (journal->j_trans_start_time == 0) {
3092                 journal->j_trans_start_time = get_seconds();
3093         }
3094         atomic_inc(&(journal->j_wcount));
3095         journal->j_len_alloc += nblocks;
3096         th->t_blocks_logged = 0;
3097         th->t_blocks_allocated = nblocks;
3098         th->t_trans_id = journal->j_trans_id;
3099         unlock_journal(sb);
3100         INIT_LIST_HEAD(&th->t_list);
3101         return 0;
3102
3103       out_fail:
3104         memset(th, 0, sizeof(*th));
3105         /* Re-set th->t_super, so we can properly keep track of how many
3106          * persistent transactions there are. We need to do this so if this
3107          * call is part of a failed restart_transaction, we can free it later */
3108         th->t_super = sb;
3109         return retval;
3110 }
3111
3112 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3113                                                                     super_block
3114                                                                     *s,
3115                                                                     int nblocks)
3116 {
3117         int ret;
3118         struct reiserfs_transaction_handle *th;
3119
3120         /* if we're nesting into an existing transaction.  It will be
3121          ** persistent on its own
3122          */
3123         if (reiserfs_transaction_running(s)) {
3124                 th = current->journal_info;
3125                 th->t_refcount++;
3126                 BUG_ON(th->t_refcount < 2);
3127
3128                 return th;
3129         }
3130         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3131         if (!th)
3132                 return NULL;
3133         ret = journal_begin(th, s, nblocks);
3134         if (ret) {
3135                 kfree(th);
3136                 return NULL;
3137         }
3138
3139         SB_JOURNAL(s)->j_persistent_trans++;
3140         return th;
3141 }
3142
3143 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3144 {
3145         struct super_block *s = th->t_super;
3146         int ret = 0;
3147         if (th->t_trans_id)
3148                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3149         else
3150                 ret = -EIO;
3151         if (th->t_refcount == 0) {
3152                 SB_JOURNAL(s)->j_persistent_trans--;
3153                 kfree(th);
3154         }
3155         return ret;
3156 }
3157
3158 static int journal_join(struct reiserfs_transaction_handle *th,
3159                         struct super_block *sb, unsigned long nblocks)
3160 {
3161         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3162
3163         /* this keeps do_journal_end from NULLing out the current->journal_info
3164          ** pointer
3165          */
3166         th->t_handle_save = cur_th;
3167         BUG_ON(cur_th && cur_th->t_refcount > 1);
3168         return do_journal_begin_r(th, sb, nblocks, JBEGIN_JOIN);
3169 }
3170
3171 int journal_join_abort(struct reiserfs_transaction_handle *th,
3172                        struct super_block *sb, unsigned long nblocks)
3173 {
3174         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3175
3176         /* this keeps do_journal_end from NULLing out the current->journal_info
3177          ** pointer
3178          */
3179         th->t_handle_save = cur_th;
3180         BUG_ON(cur_th && cur_th->t_refcount > 1);
3181         return do_journal_begin_r(th, sb, nblocks, JBEGIN_ABORT);
3182 }
3183
3184 int journal_begin(struct reiserfs_transaction_handle *th,
3185                   struct super_block *sb, unsigned long nblocks)
3186 {
3187         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3188         int ret;
3189
3190         th->t_handle_save = NULL;
3191         if (cur_th) {
3192                 /* we are nesting into the current transaction */
3193                 if (cur_th->t_super == sb) {
3194                         BUG_ON(!cur_th->t_refcount);
3195                         cur_th->t_refcount++;
3196                         memcpy(th, cur_th, sizeof(*th));
3197                         if (th->t_refcount <= 1)
3198                                 reiserfs_warning(sb, "reiserfs-2005",
3199                                                  "BAD: refcount <= 1, but "
3200                                                  "journal_info != 0");
3201                         return 0;
3202                 } else {
3203                         /* we've ended up with a handle from a different filesystem.
3204                          ** save it and restore on journal_end.  This should never
3205                          ** really happen...
3206                          */
3207                         reiserfs_warning(sb, "clm-2100",
3208                                          "nesting info a different FS");
3209                         th->t_handle_save = current->journal_info;
3210                         current->journal_info = th;
3211                 }
3212         } else {
3213                 current->journal_info = th;
3214         }
3215         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3216         BUG_ON(current->journal_info != th);
3217
3218         /* I guess this boils down to being the reciprocal of clm-2100 above.
3219          * If do_journal_begin_r fails, we need to put it back, since journal_end
3220          * won't be called to do it. */
3221         if (ret)
3222                 current->journal_info = th->t_handle_save;
3223         else
3224                 BUG_ON(!th->t_refcount);
3225
3226         return ret;
3227 }
3228
3229 /*
3230 ** puts bh into the current transaction.  If it was already there, reorders removes the
3231 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3232 **
3233 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3234 ** transaction is committed.
3235 **
3236 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3237 */
3238 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3239                        struct super_block *sb, struct buffer_head *bh)
3240 {
3241         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3242         struct reiserfs_journal_cnode *cn = NULL;
3243         int count_already_incd = 0;
3244         int prepared = 0;
3245         BUG_ON(!th->t_trans_id);
3246
3247         PROC_INFO_INC(sb, journal.mark_dirty);
3248         if (th->t_trans_id != journal->j_trans_id) {
3249                 reiserfs_panic(th->t_super, "journal-1577",
3250                                "handle trans id %ld != current trans id %ld",
3251                                th->t_trans_id, journal->j_trans_id);
3252         }
3253
3254         sb->s_dirt = 1;
3255
3256         prepared = test_clear_buffer_journal_prepared(bh);
3257         clear_buffer_journal_restore_dirty(bh);
3258         /* already in this transaction, we are done */
3259         if (buffer_journaled(bh)) {
3260                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3261                 return 0;
3262         }
3263
3264         /* this must be turned into a panic instead of a warning.  We can't allow
3265          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3266          ** could get to disk too early.  NOT GOOD.
3267          */
3268         if (!prepared || buffer_dirty(bh)) {
3269                 reiserfs_warning(sb, "journal-1777",
3270                                  "buffer %llu bad state "
3271                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3272                                  (unsigned long long)bh->b_blocknr,
3273                                  prepared ? ' ' : '!',
3274                                  buffer_locked(bh) ? ' ' : '!',
3275                                  buffer_dirty(bh) ? ' ' : '!',
3276                                  buffer_journal_dirty(bh) ? ' ' : '!');
3277         }
3278
3279         if (atomic_read(&(journal->j_wcount)) <= 0) {
3280                 reiserfs_warning(sb, "journal-1409",
3281                                  "returning because j_wcount was %d",
3282                                  atomic_read(&(journal->j_wcount)));
3283                 return 1;
3284         }
3285         /* this error means I've screwed up, and we've overflowed the transaction.
3286          ** Nothing can be done here, except make the FS readonly or panic.
3287          */
3288         if (journal->j_len >= journal->j_trans_max) {
3289                 reiserfs_panic(th->t_super, "journal-1413",
3290                                "j_len (%lu) is too big",
3291                                journal->j_len);
3292         }
3293
3294         if (buffer_journal_dirty(bh)) {
3295                 count_already_incd = 1;
3296                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3297                 clear_buffer_journal_dirty(bh);
3298         }
3299
3300         if (journal->j_len > journal->j_len_alloc) {
3301                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3302         }
3303
3304         set_buffer_journaled(bh);
3305
3306         /* now put this guy on the end */
3307         if (!cn) {
3308                 cn = get_cnode(sb);
3309                 if (!cn) {
3310                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3311                 }
3312
3313                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3314                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3315                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3316                 }
3317                 th->t_blocks_logged++;
3318                 journal->j_len++;
3319
3320                 cn->bh = bh;
3321                 cn->blocknr = bh->b_blocknr;
3322                 cn->sb = sb;
3323                 cn->jlist = NULL;
3324                 insert_journal_hash(journal->j_hash_table, cn);
3325                 if (!count_already_incd) {
3326                         get_bh(bh);
3327                 }
3328         }
3329         cn->next = NULL;
3330         cn->prev = journal->j_last;
3331         cn->bh = bh;
3332         if (journal->j_last) {
3333                 journal->j_last->next = cn;
3334                 journal->j_last = cn;
3335         } else {
3336                 journal->j_first = cn;
3337                 journal->j_last = cn;
3338         }
3339         return 0;
3340 }
3341
3342 int journal_end(struct reiserfs_transaction_handle *th,
3343                 struct super_block *sb, unsigned long nblocks)
3344 {
3345         if (!current->journal_info && th->t_refcount > 1)
3346                 reiserfs_warning(sb, "REISER-NESTING",
3347                                  "th NULL, refcount %d", th->t_refcount);
3348
3349         if (!th->t_trans_id) {
3350                 WARN_ON(1);
3351                 return -EIO;
3352         }
3353
3354         th->t_refcount--;
3355         if (th->t_refcount > 0) {
3356                 struct reiserfs_transaction_handle *cur_th =
3357                     current->journal_info;
3358
3359                 /* we aren't allowed to close a nested transaction on a different
3360                  ** filesystem from the one in the task struct
3361                  */
3362                 BUG_ON(cur_th->t_super != th->t_super);
3363
3364                 if (th != cur_th) {
3365                         memcpy(current->journal_info, th, sizeof(*th));
3366                         th->t_trans_id = 0;
3367                 }
3368                 return 0;
3369         } else {
3370                 return do_journal_end(th, sb, nblocks, 0);
3371         }
3372 }
3373
3374 /* removes from the current transaction, relsing and descrementing any counters.
3375 ** also files the removed buffer directly onto the clean list
3376 **
3377 ** called by journal_mark_freed when a block has been deleted
3378 **
3379 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3380 */
3381 static int remove_from_transaction(struct super_block *sb,
3382                                    b_blocknr_t blocknr, int already_cleaned)
3383 {
3384         struct buffer_head *bh;
3385         struct reiserfs_journal_cnode *cn;
3386         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3387         int ret = 0;
3388
3389         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3390         if (!cn || !cn->bh) {
3391                 return ret;
3392         }
3393         bh = cn->bh;
3394         if (cn->prev) {
3395                 cn->prev->next = cn->next;
3396         }
3397         if (cn->next) {
3398                 cn->next->prev = cn->prev;
3399         }
3400         if (cn == journal->j_first) {
3401                 journal->j_first = cn->next;
3402         }
3403         if (cn == journal->j_last) {
3404                 journal->j_last = cn->prev;
3405         }
3406         if (bh)
3407                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3408                                     bh->b_blocknr, 0);
3409         clear_buffer_journaled(bh);     /* don't log this one */
3410
3411         if (!already_cleaned) {
3412                 clear_buffer_journal_dirty(bh);
3413                 clear_buffer_dirty(bh);
3414                 clear_buffer_journal_test(bh);
3415                 put_bh(bh);
3416                 if (atomic_read(&(bh->b_count)) < 0) {
3417                         reiserfs_warning(sb, "journal-1752",
3418                                          "b_count < 0");
3419                 }
3420                 ret = 1;
3421         }
3422         journal->j_len--;
3423         journal->j_len_alloc--;
3424         free_cnode(sb, cn);
3425         return ret;
3426 }
3427
3428 /*
3429 ** for any cnode in a journal list, it can only be dirtied of all the
3430 ** transactions that include it are committed to disk.
3431 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3432 ** and 0 if you aren't
3433 **
3434 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3435 ** blocks for a given transaction on disk
3436 **
3437 */
3438 static int can_dirty(struct reiserfs_journal_cnode *cn)
3439 {
3440         struct super_block *sb = cn->sb;
3441         b_blocknr_t blocknr = cn->blocknr;
3442         struct reiserfs_journal_cnode *cur = cn->hprev;
3443         int can_dirty = 1;
3444
3445         /* first test hprev.  These are all newer than cn, so any node here
3446          ** with the same block number and dev means this node can't be sent
3447          ** to disk right now.
3448          */
3449         while (cur && can_dirty) {
3450                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3451                     cur->blocknr == blocknr) {
3452                         can_dirty = 0;
3453                 }
3454                 cur = cur->hprev;
3455         }
3456         /* then test hnext.  These are all older than cn.  As long as they
3457          ** are committed to the log, it is safe to write cn to disk
3458          */
3459         cur = cn->hnext;
3460         while (cur && can_dirty) {
3461                 if (cur->jlist && cur->jlist->j_len > 0 &&
3462                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3463                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3464                         can_dirty = 0;
3465                 }
3466                 cur = cur->hnext;
3467         }
3468         return can_dirty;
3469 }
3470
3471 /* syncs the commit blocks, but does not force the real buffers to disk
3472 ** will wait until the current transaction is done/committed before returning
3473 */
3474 int journal_end_sync(struct reiserfs_transaction_handle *th,
3475                      struct super_block *sb, unsigned long nblocks)
3476 {
3477         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3478
3479         BUG_ON(!th->t_trans_id);
3480         /* you can sync while nested, very, very bad */
3481         BUG_ON(th->t_refcount > 1);
3482         if (journal->j_len == 0) {
3483                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3484                                              1);
3485                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3486         }
3487         return do_journal_end(th, sb, nblocks, COMMIT_NOW | WAIT);
3488 }
3489
3490 /*
3491 ** writeback the pending async commits to disk
3492 */
3493 static void flush_async_commits(struct work_struct *work)
3494 {
3495         struct reiserfs_journal *journal =
3496                 container_of(work, struct reiserfs_journal, j_work.work);
3497         struct super_block *sb = journal->j_work_sb;
3498         struct reiserfs_journal_list *jl;
3499         struct list_head *entry;
3500
3501         reiserfs_write_lock(sb);
3502         if (!list_empty(&journal->j_journal_list)) {
3503                 /* last entry is the youngest, commit it and you get everything */
3504                 entry = journal->j_journal_list.prev;
3505                 jl = JOURNAL_LIST_ENTRY(entry);
3506                 flush_commit_list(sb, jl, 1);
3507         }
3508         reiserfs_write_unlock(sb);
3509 }
3510
3511 /*
3512 ** flushes any old transactions to disk
3513 ** ends the current transaction if it is too old
3514 */
3515 int reiserfs_flush_old_commits(struct super_block *sb)
3516 {
3517         time_t now;
3518         struct reiserfs_transaction_handle th;
3519         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3520
3521         now = get_seconds();
3522         /* safety check so we don't flush while we are replaying the log during
3523          * mount
3524          */
3525         if (list_empty(&journal->j_journal_list)) {
3526                 return 0;
3527         }
3528
3529         /* check the current transaction.  If there are no writers, and it is
3530          * too old, finish it, and force the commit blocks to disk
3531          */
3532         if (atomic_read(&journal->j_wcount) <= 0 &&
3533             journal->j_trans_start_time > 0 &&
3534             journal->j_len > 0 &&
3535             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3536                 if (!journal_join(&th, sb, 1)) {
3537                         reiserfs_prepare_for_journal(sb,
3538                                                      SB_BUFFER_WITH_SB(sb),
3539                                                      1);
3540                         journal_mark_dirty(&th, sb,
3541                                            SB_BUFFER_WITH_SB(sb));
3542
3543                         /* we're only being called from kreiserfsd, it makes no sense to do
3544                          ** an async commit so that kreiserfsd can do it later
3545                          */
3546                         do_journal_end(&th, sb, 1, COMMIT_NOW | WAIT);
3547                 }
3548         }
3549         return sb->s_dirt;
3550 }
3551
3552 /*
3553 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3554 **
3555 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3556 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3557 ** flushes the commit list and returns 0.
3558 **
3559 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3560 **
3561 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3562 */
3563 static int check_journal_end(struct reiserfs_transaction_handle *th,
3564                              struct super_block *sb, unsigned long nblocks,
3565                              int flags)
3566 {
3567
3568         time_t now;
3569         int flush = flags & FLUSH_ALL;
3570         int commit_now = flags & COMMIT_NOW;
3571         int wait_on_commit = flags & WAIT;
3572         struct reiserfs_journal_list *jl;
3573         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3574
3575         BUG_ON(!th->t_trans_id);
3576
3577         if (th->t_trans_id != journal->j_trans_id) {
3578                 reiserfs_panic(th->t_super, "journal-1577",
3579                                "handle trans id %ld != current trans id %ld",
3580                                th->t_trans_id, journal->j_trans_id);
3581         }
3582
3583         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3584         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3585                 atomic_dec(&(journal->j_wcount));
3586         }
3587
3588         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3589          ** will be dealt with by next transaction that actually writes something, but should be taken
3590          ** care of in this trans
3591          */
3592         BUG_ON(journal->j_len == 0);
3593
3594         /* if wcount > 0, and we are called to with flush or commit_now,
3595          ** we wait on j_join_wait.  We will wake up when the last writer has
3596          ** finished the transaction, and started it on its way to the disk.
3597          ** Then, we flush the commit or journal list, and just return 0
3598          ** because the rest of journal end was already done for this transaction.
3599          */
3600         if (atomic_read(&(journal->j_wcount)) > 0) {
3601                 if (flush || commit_now) {
3602                         unsigned trans_id;
3603
3604                         jl = journal->j_current_jl;
3605                         trans_id = jl->j_trans_id;
3606                         if (wait_on_commit)
3607                                 jl->j_state |= LIST_COMMIT_PENDING;
3608                         atomic_set(&(journal->j_jlock), 1);
3609                         if (flush) {
3610                                 journal->j_next_full_flush = 1;
3611                         }
3612                         unlock_journal(sb);
3613
3614                         /* sleep while the current transaction is still j_jlocked */
3615                         while (journal->j_trans_id == trans_id) {
3616                                 if (atomic_read(&journal->j_jlock)) {
3617                                         queue_log_writer(sb);
3618                                 } else {
3619                                         lock_journal(sb);
3620                                         if (journal->j_trans_id == trans_id) {
3621                                                 atomic_set(&(journal->j_jlock),
3622                                                            1);
3623                                         }
3624                                         unlock_journal(sb);
3625                                 }
3626                         }
3627                         BUG_ON(journal->j_trans_id == trans_id);
3628
3629                         if (commit_now
3630                             && journal_list_still_alive(sb, trans_id)
3631                             && wait_on_commit) {
3632                                 flush_commit_list(sb, jl, 1);
3633                         }
3634                         return 0;
3635                 }
3636                 unlock_journal(sb);
3637                 return 0;
3638         }
3639
3640         /* deal with old transactions where we are the last writers */
3641         now = get_seconds();
3642         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3643                 commit_now = 1;
3644                 journal->j_next_async_flush = 1;
3645         }
3646         /* don't batch when someone is waiting on j_join_wait */
3647         /* don't batch when syncing the commit or flushing the whole trans */
3648         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3649             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3650             && journal->j_len_alloc < journal->j_max_batch
3651             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3652                 journal->j_bcount++;
3653                 unlock_journal(sb);
3654                 return 0;
3655         }
3656
3657         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3658                 reiserfs_panic(sb, "journal-003",
3659                                "j_start (%ld) is too high",
3660                                journal->j_start);
3661         }
3662         return 1;
3663 }
3664
3665 /*
3666 ** Does all the work that makes deleting blocks safe.
3667 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3668 **
3669 ** otherwise:
3670 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3671 ** before this transaction has finished.
3672 **
3673 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3674 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3675 ** the block can't be reallocated yet.
3676 **
3677 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3678 */
3679 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3680                        struct super_block *sb, b_blocknr_t blocknr)
3681 {
3682         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3683         struct reiserfs_journal_cnode *cn = NULL;
3684         struct buffer_head *bh = NULL;
3685         struct reiserfs_list_bitmap *jb = NULL;
3686         int cleaned = 0;
3687         BUG_ON(!th->t_trans_id);
3688
3689         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3690         if (cn && cn->bh) {
3691                 bh = cn->bh;
3692                 get_bh(bh);
3693         }
3694         /* if it is journal new, we just remove it from this transaction */
3695         if (bh && buffer_journal_new(bh)) {
3696                 clear_buffer_journal_new(bh);
3697                 clear_prepared_bits(bh);
3698                 reiserfs_clean_and_file_buffer(bh);
3699                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3700         } else {
3701                 /* set the bit for this block in the journal bitmap for this transaction */
3702                 jb = journal->j_current_jl->j_list_bitmap;
3703                 if (!jb) {
3704                         reiserfs_panic(sb, "journal-1702",
3705                                        "journal_list_bitmap is NULL");
3706                 }
3707                 set_bit_in_list_bitmap(sb, blocknr, jb);
3708
3709                 /* Note, the entire while loop is not allowed to schedule.  */
3710
3711                 if (bh) {
3712                         clear_prepared_bits(bh);
3713                         reiserfs_clean_and_file_buffer(bh);
3714                 }
3715                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3716
3717                 /* find all older transactions with this block, make sure they don't try to write it out */
3718                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3719                                           blocknr);
3720                 while (cn) {
3721                         if (sb == cn->sb && blocknr == cn->blocknr) {
3722                                 set_bit(BLOCK_FREED, &cn->state);
3723                                 if (cn->bh) {
3724                                         if (!cleaned) {
3725                                                 /* remove_from_transaction will brelse the buffer if it was
3726                                                  ** in the current trans
3727                                                  */
3728                                                 clear_buffer_journal_dirty(cn->
3729                                                                            bh);
3730                                                 clear_buffer_dirty(cn->bh);
3731                                                 clear_buffer_journal_test(cn->
3732                                                                           bh);
3733                                                 cleaned = 1;
3734                                                 put_bh(cn->bh);
3735                                                 if (atomic_read
3736                                                     (&(cn->bh->b_count)) < 0) {
3737                                                         reiserfs_warning(sb,
3738                                                                  "journal-2138",
3739                                                                  "cn->bh->b_count < 0");
3740                                                 }
3741                                         }
3742                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3743                                                 atomic_dec(&
3744                                                            (cn->jlist->
3745                                                             j_nonzerolen));
3746                                         }
3747                                         cn->bh = NULL;
3748                                 }
3749                         }
3750                         cn = cn->hnext;
3751                 }
3752         }
3753
3754         if (bh)
3755                 release_buffer_page(bh); /* get_hash grabs the buffer */
3756         return 0;
3757 }
3758
3759 void reiserfs_update_inode_transaction(struct inode *inode)
3760 {
3761         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3762         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3763         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3764 }
3765
3766 /*
3767  * returns -1 on error, 0 if no commits/barriers were done and 1
3768  * if a transaction was actually committed and the barrier was done
3769  */
3770 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3771                              struct reiserfs_journal_list *jl)
3772 {
3773         struct reiserfs_transaction_handle th;
3774         struct super_block *sb = inode->i_sb;
3775         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3776         int ret = 0;
3777
3778         /* is it from the current transaction, or from an unknown transaction? */
3779         if (id == journal->j_trans_id) {
3780                 jl = journal->j_current_jl;
3781                 /* try to let other writers come in and grow this transaction */
3782                 let_transaction_grow(sb, id);
3783                 if (journal->j_trans_id != id) {
3784                         goto flush_commit_only;
3785                 }
3786
3787                 ret = journal_begin(&th, sb, 1);
3788                 if (ret)
3789                         return ret;
3790
3791                 /* someone might have ended this transaction while we joined */
3792                 if (journal->j_trans_id != id) {
3793                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3794                                                      1);
3795                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3796                         ret = journal_end(&th, sb, 1);
3797                         goto flush_commit_only;
3798                 }
3799
3800                 ret = journal_end_sync(&th, sb, 1);
3801                 if (!ret)
3802                         ret = 1;
3803
3804         } else {
3805                 /* this gets tricky, we have to make sure the journal list in
3806                  * the inode still exists.  We know the list is still around
3807                  * if we've got a larger transaction id than the oldest list
3808                  */
3809               flush_commit_only:
3810                 if (journal_list_still_alive(inode->i_sb, id)) {
3811                         /*
3812                          * we only set ret to 1 when we know for sure
3813                          * the barrier hasn't been started yet on the commit
3814                          * block.
3815                          */
3816                         if (atomic_read(&jl->j_commit_left) > 1)
3817                                 ret = 1;
3818                         flush_commit_list(sb, jl, 1);
3819                         if (journal->j_errno)
3820                                 ret = journal->j_errno;
3821                 }
3822         }
3823         /* otherwise the list is gone, and long since committed */
3824         return ret;
3825 }
3826
3827 int reiserfs_commit_for_inode(struct inode *inode)
3828 {
3829         unsigned int id = REISERFS_I(inode)->i_trans_id;
3830         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3831
3832         /* for the whole inode, assume unset id means it was
3833          * changed in the current transaction.  More conservative
3834          */
3835         if (!id || !jl) {
3836                 reiserfs_update_inode_transaction(inode);
3837                 id = REISERFS_I(inode)->i_trans_id;
3838                 /* jl will be updated in __commit_trans_jl */
3839         }
3840
3841         return __commit_trans_jl(inode, id, jl);
3842 }
3843
3844 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3845                                       struct buffer_head *bh)
3846 {
3847         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3848         PROC_INFO_INC(sb, journal.restore_prepared);
3849         if (!bh) {
3850                 return;
3851         }
3852         if (test_clear_buffer_journal_restore_dirty(bh) &&
3853             buffer_journal_dirty(bh)) {
3854                 struct reiserfs_journal_cnode *cn;
3855                 cn = get_journal_hash_dev(sb,
3856                                           journal->j_list_hash_table,
3857                                           bh->b_blocknr);
3858                 if (cn && can_dirty(cn)) {
3859                         set_buffer_journal_test(bh);
3860                         mark_buffer_dirty(bh);
3861                 }
3862         }
3863         clear_buffer_journal_prepared(bh);
3864 }
3865
3866 extern struct tree_balance *cur_tb;
3867 /*
3868 ** before we can change a metadata block, we have to make sure it won't
3869 ** be written to disk while we are altering it.  So, we must:
3870 ** clean it
3871 ** wait on it.
3872 **
3873 */
3874 int reiserfs_prepare_for_journal(struct super_block *sb,
3875                                  struct buffer_head *bh, int wait)
3876 {
3877         PROC_INFO_INC(sb, journal.prepare);
3878
3879         if (!trylock_buffer(bh)) {
3880                 if (!wait)
3881                         return 0;
3882                 lock_buffer(bh);
3883         }
3884         set_buffer_journal_prepared(bh);
3885         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3886                 clear_buffer_journal_test(bh);
3887                 set_buffer_journal_restore_dirty(bh);
3888         }
3889         unlock_buffer(bh);
3890         return 1;
3891 }
3892
3893 static void flush_old_journal_lists(struct super_block *s)
3894 {
3895         struct reiserfs_journal *journal = SB_JOURNAL(s);
3896         struct reiserfs_journal_list *jl;
3897         struct list_head *entry;
3898         time_t now = get_seconds();
3899
3900         while (!list_empty(&journal->j_journal_list)) {
3901                 entry = journal->j_journal_list.next;
3902                 jl = JOURNAL_LIST_ENTRY(entry);
3903                 /* this check should always be run, to send old lists to disk */
3904                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3905                     atomic_read(&jl->j_commit_left) == 0 &&
3906                     test_transaction(s, jl)) {
3907                         flush_used_journal_lists(s, jl);
3908                 } else {
3909                         break;
3910                 }
3911         }
3912 }
3913
3914 /*
3915 ** long and ugly.  If flush, will not return until all commit
3916 ** blocks and all real buffers in the trans are on disk.
3917 ** If no_async, won't return until all commit blocks are on disk.
3918 **
3919 ** keep reading, there are comments as you go along
3920 **
3921 ** If the journal is aborted, we just clean up. Things like flushing
3922 ** journal lists, etc just won't happen.
3923 */
3924 static int do_journal_end(struct reiserfs_transaction_handle *th,
3925                           struct super_block *sb, unsigned long nblocks,
3926                           int flags)
3927 {
3928         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3929         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3930         struct reiserfs_journal_cnode *last_cn = NULL;
3931         struct reiserfs_journal_desc *desc;
3932         struct reiserfs_journal_commit *commit;
3933         struct buffer_head *c_bh;       /* commit bh */
3934         struct buffer_head *d_bh;       /* desc bh */
3935         int cur_write_start = 0;        /* start index of current log write */
3936         int old_start;
3937         int i;
3938         int flush;
3939         int wait_on_commit;
3940         struct reiserfs_journal_list *jl, *temp_jl;
3941         struct list_head *entry, *safe;
3942         unsigned long jindex;
3943         unsigned int commit_trans_id;
3944         int trans_half;
3945
3946         BUG_ON(th->t_refcount > 1);
3947         BUG_ON(!th->t_trans_id);
3948
3949         /* protect flush_older_commits from doing mistakes if the
3950            transaction ID counter gets overflowed.  */
3951         if (th->t_trans_id == ~0U)
3952                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3953         flush = flags & FLUSH_ALL;
3954         wait_on_commit = flags & WAIT;
3955
3956         current->journal_info = th->t_handle_save;
3957         reiserfs_check_lock_depth(sb, "journal end");
3958         if (journal->j_len == 0) {
3959                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3960                                              1);
3961                 journal_mark_dirty(th, sb, SB_BUFFER_WITH_SB(sb));
3962         }
3963
3964         lock_journal(sb);
3965         if (journal->j_next_full_flush) {
3966                 flags |= FLUSH_ALL;
3967                 flush = 1;
3968         }
3969         if (journal->j_next_async_flush) {
3970                 flags |= COMMIT_NOW | WAIT;
3971                 wait_on_commit = 1;
3972         }
3973
3974         /* check_journal_end locks the journal, and unlocks if it does not return 1
3975          ** it tells us if we should continue with the journal_end, or just return
3976          */
3977         if (!check_journal_end(th, sb, nblocks, flags)) {
3978                 sb->s_dirt = 1;
3979                 wake_queued_writers(sb);
3980                 reiserfs_async_progress_wait(sb);
3981                 goto out;
3982         }
3983
3984         /* check_journal_end might set these, check again */
3985         if (journal->j_next_full_flush) {
3986                 flush = 1;
3987         }
3988
3989         /*
3990          ** j must wait means we have to flush the log blocks, and the real blocks for
3991          ** this transaction
3992          */
3993         if (journal->j_must_wait > 0) {
3994                 flush = 1;
3995         }
3996 #ifdef REISERFS_PREALLOCATE
3997         /* quota ops might need to nest, setup the journal_info pointer for them
3998          * and raise the refcount so that it is > 0. */
3999         current->journal_info = th;
4000         th->t_refcount++;
4001         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4002                                                  * the transaction */
4003         th->t_refcount--;
4004         current->journal_info = th->t_handle_save;
4005 #endif
4006
4007         /* setup description block */
4008         d_bh =
4009             journal_getblk(sb,
4010                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4011                            journal->j_start);
4012         set_buffer_uptodate(d_bh);
4013         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4014         memset(d_bh->b_data, 0, d_bh->b_size);
4015         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4016         set_desc_trans_id(desc, journal->j_trans_id);
4017
4018         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4019         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4020                               ((journal->j_start + journal->j_len +
4021                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4022         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4023         memset(c_bh->b_data, 0, c_bh->b_size);
4024         set_commit_trans_id(commit, journal->j_trans_id);
4025         set_buffer_uptodate(c_bh);
4026
4027         /* init this journal list */
4028         jl = journal->j_current_jl;
4029
4030         /* we lock the commit before doing anything because
4031          * we want to make sure nobody tries to run flush_commit_list until
4032          * the new transaction is fully setup, and we've already flushed the
4033          * ordered bh list
4034          */
4035         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4036
4037         /* save the transaction id in case we need to commit it later */
4038         commit_trans_id = jl->j_trans_id;
4039
4040         atomic_set(&jl->j_older_commits_done, 0);
4041         jl->j_trans_id = journal->j_trans_id;
4042         jl->j_timestamp = journal->j_trans_start_time;
4043         jl->j_commit_bh = c_bh;
4044         jl->j_start = journal->j_start;
4045         jl->j_len = journal->j_len;
4046         atomic_set(&jl->j_nonzerolen, journal->j_len);
4047         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4048         jl->j_realblock = NULL;
4049
4050         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4051          **  for each real block, add it to the journal list hash,
4052          ** copy into real block index array in the commit or desc block
4053          */
4054         trans_half = journal_trans_half(sb->s_blocksize);
4055         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4056                 if (buffer_journaled(cn->bh)) {
4057                         jl_cn = get_cnode(sb);
4058                         if (!jl_cn) {
4059                                 reiserfs_panic(sb, "journal-1676",
4060                                                "get_cnode returned NULL");
4061                         }
4062                         if (i == 0) {
4063                                 jl->j_realblock = jl_cn;
4064                         }
4065                         jl_cn->prev = last_cn;
4066                         jl_cn->next = NULL;
4067                         if (last_cn) {
4068                                 last_cn->next = jl_cn;
4069                         }
4070                         last_cn = jl_cn;
4071                         /* make sure the block we are trying to log is not a block
4072                            of journal or reserved area */
4073
4074                         if (is_block_in_log_or_reserved_area
4075                             (sb, cn->bh->b_blocknr)) {
4076                                 reiserfs_panic(sb, "journal-2332",
4077                                                "Trying to log block %lu, "
4078                                                "which is a log block",
4079                                                cn->bh->b_blocknr);
4080                         }
4081                         jl_cn->blocknr = cn->bh->b_blocknr;
4082                         jl_cn->state = 0;
4083                         jl_cn->sb = sb;
4084                         jl_cn->bh = cn->bh;
4085                         jl_cn->jlist = jl;
4086                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4087                         if (i < trans_half) {
4088                                 desc->j_realblock[i] =
4089                                     cpu_to_le32(cn->bh->b_blocknr);
4090                         } else {
4091                                 commit->j_realblock[i - trans_half] =
4092                                     cpu_to_le32(cn->bh->b_blocknr);
4093                         }
4094                 } else {
4095                         i--;
4096                 }
4097         }
4098         set_desc_trans_len(desc, journal->j_len);
4099         set_desc_mount_id(desc, journal->j_mount_id);
4100         set_desc_trans_id(desc, journal->j_trans_id);
4101         set_commit_trans_len(commit, journal->j_len);
4102
4103         /* special check in case all buffers in the journal were marked for not logging */
4104         BUG_ON(journal->j_len == 0);
4105
4106         /* we're about to dirty all the log blocks, mark the description block
4107          * dirty now too.  Don't mark the commit block dirty until all the
4108          * others are on disk
4109          */
4110         mark_buffer_dirty(d_bh);
4111
4112         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4113         cur_write_start = journal->j_start;
4114         cn = journal->j_first;
4115         jindex = 1;             /* start at one so we don't get the desc again */
4116         while (cn) {
4117                 clear_buffer_journal_new(cn->bh);
4118                 /* copy all the real blocks into log area.  dirty log blocks */
4119                 if (buffer_journaled(cn->bh)) {
4120                         struct buffer_head *tmp_bh;
4121                         char *addr;
4122                         struct page *page;
4123                         tmp_bh =
4124                             journal_getblk(sb,
4125                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4126                                            ((cur_write_start +
4127                                              jindex) %
4128                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4129                         set_buffer_uptodate(tmp_bh);
4130                         page = cn->bh->b_page;
4131                         addr = kmap(page);
4132                         memcpy(tmp_bh->b_data,
4133                                addr + offset_in_page(cn->bh->b_data),
4134                                cn->bh->b_size);
4135                         kunmap(page);
4136                         mark_buffer_dirty(tmp_bh);
4137                         jindex++;
4138                         set_buffer_journal_dirty(cn->bh);
4139                         clear_buffer_journaled(cn->bh);
4140                 } else {
4141                         /* JDirty cleared sometime during transaction.  don't log this one */
4142                         reiserfs_warning(sb, "journal-2048",
4143                                          "BAD, buffer in journal hash, "
4144                                          "but not JDirty!");
4145                         brelse(cn->bh);
4146                 }
4147                 next = cn->next;
4148                 free_cnode(sb, cn);
4149                 cn = next;
4150                 reiserfs_write_unlock(sb);
4151                 cond_resched();
4152                 reiserfs_write_lock(sb);
4153         }
4154
4155         /* we are done  with both the c_bh and d_bh, but
4156          ** c_bh must be written after all other commit blocks,
4157          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4158          */
4159
4160         journal->j_current_jl = alloc_journal_list(sb);
4161
4162         /* now it is safe to insert this transaction on the main list */
4163         list_add_tail(&jl->j_list, &journal->j_journal_list);
4164         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4165         journal->j_num_work_lists++;
4166
4167         /* reset journal values for the next transaction */
4168         old_start = journal->j_start;
4169         journal->j_start =
4170             (journal->j_start + journal->j_len +
4171              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4172         atomic_set(&(journal->j_wcount), 0);
4173         journal->j_bcount = 0;
4174         journal->j_last = NULL;
4175         journal->j_first = NULL;
4176         journal->j_len = 0;
4177         journal->j_trans_start_time = 0;
4178         /* check for trans_id overflow */
4179         if (++journal->j_trans_id == 0)
4180                 journal->j_trans_id = 10;
4181         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4182         journal->j_must_wait = 0;
4183         journal->j_len_alloc = 0;
4184         journal->j_next_full_flush = 0;
4185         journal->j_next_async_flush = 0;
4186         init_journal_hash(sb);
4187
4188         // make sure reiserfs_add_jh sees the new current_jl before we
4189         // write out the tails
4190         smp_mb();
4191
4192         /* tail conversion targets have to hit the disk before we end the
4193          * transaction.  Otherwise a later transaction might repack the tail
4194          * before this transaction commits, leaving the data block unflushed and
4195          * clean, if we crash before the later transaction commits, the data block
4196          * is lost.
4197          */
4198         if (!list_empty(&jl->j_tail_bh_list)) {
4199                 reiserfs_write_unlock(sb);
4200                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4201                                       journal, jl, &jl->j_tail_bh_list);
4202                 reiserfs_write_lock(sb);
4203         }
4204         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4205         mutex_unlock(&jl->j_commit_mutex);
4206
4207         /* honor the flush wishes from the caller, simple commits can
4208          ** be done outside the journal lock, they are done below
4209          **
4210          ** if we don't flush the commit list right now, we put it into
4211          ** the work queue so the people waiting on the async progress work
4212          ** queue don't wait for this proc to flush journal lists and such.
4213          */
4214         if (flush) {
4215                 flush_commit_list(sb, jl, 1);
4216                 flush_journal_list(sb, jl, 1);
4217         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4218                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4219
4220         /* if the next transaction has any chance of wrapping, flush
4221          ** transactions that might get overwritten.  If any journal lists are very
4222          ** old flush them as well.
4223          */
4224       first_jl:
4225         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4226                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4227                 if (journal->j_start <= temp_jl->j_start) {
4228                         if ((journal->j_start + journal->j_trans_max + 1) >=
4229                             temp_jl->j_start) {
4230                                 flush_used_journal_lists(sb, temp_jl);
4231                                 goto first_jl;
4232                         } else if ((journal->j_start +
4233                                     journal->j_trans_max + 1) <
4234                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4235                                 /* if we don't cross into the next transaction and we don't
4236                                  * wrap, there is no way we can overlap any later transactions
4237                                  * break now
4238                                  */
4239                                 break;
4240                         }
4241                 } else if ((journal->j_start +
4242                             journal->j_trans_max + 1) >
4243                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4244                         if (((journal->j_start + journal->j_trans_max + 1) %
4245                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4246                             temp_jl->j_start) {
4247                                 flush_used_journal_lists(sb, temp_jl);
4248                                 goto first_jl;
4249                         } else {
4250                                 /* we don't overlap anything from out start to the end of the
4251                                  * log, and our wrapped portion doesn't overlap anything at
4252                                  * the start of the log.  We can break
4253                                  */
4254                                 break;
4255                         }
4256                 }
4257         }
4258         flush_old_journal_lists(sb);
4259
4260         journal->j_current_jl->j_list_bitmap =
4261             get_list_bitmap(sb, journal->j_current_jl);
4262
4263         if (!(journal->j_current_jl->j_list_bitmap)) {
4264                 reiserfs_panic(sb, "journal-1996",
4265                                "could not get a list bitmap");
4266         }
4267
4268         atomic_set(&(journal->j_jlock), 0);
4269         unlock_journal(sb);
4270         /* wake up any body waiting to join. */
4271         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4272         wake_up(&(journal->j_join_wait));
4273
4274         if (!flush && wait_on_commit &&
4275             journal_list_still_alive(sb, commit_trans_id)) {
4276                 flush_commit_list(sb, jl, 1);
4277         }
4278       out:
4279         reiserfs_check_lock_depth(sb, "journal end2");
4280
4281         memset(th, 0, sizeof(*th));
4282         /* Re-set th->t_super, so we can properly keep track of how many
4283          * persistent transactions there are. We need to do this so if this
4284          * call is part of a failed restart_transaction, we can free it later */
4285         th->t_super = sb;
4286
4287         return journal->j_errno;
4288 }
4289
4290 /* Send the file system read only and refuse new transactions */
4291 void reiserfs_abort_journal(struct super_block *sb, int errno)
4292 {
4293         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4294         if (test_bit(J_ABORTED, &journal->j_state))
4295                 return;
4296
4297         if (!journal->j_errno)
4298                 journal->j_errno = errno;
4299
4300         sb->s_flags |= MS_RDONLY;
4301         set_bit(J_ABORTED, &journal->j_state);
4302
4303 #ifdef CONFIG_REISERFS_CHECK
4304         dump_stack();
4305 #endif
4306 }