kernel/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 "reiserfs.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
  55 /* gets a struct reiserfs_journal_list * from a list head */
  56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  57                                j_list))
  58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  59                                j_working_list))
  60
  61 /* must be correct to keep the desc and commit structs at 4k */
  62 #define JOURNAL_TRANS_HALF 1018
  63 #define BUFNR 64                /*read ahead */
  64
  65 /* cnode stat bits.  Move these into reiserfs_fs.h */
  66
  67 /* this block was freed, and can't be written.  */
  68 #define BLOCK_FREED 2
  69 /* this block was freed during this transaction, and can't be written */
  70 #define BLOCK_FREED_HOLDER 3
  71
  72 /* used in flush_journal_list */
  73 #define BLOCK_NEEDS_FLUSH 4
  74 #define BLOCK_DIRTIED 5
  75
  76 /* journal list state bits */
  77 #define LIST_TOUCHED 1
  78 #define LIST_DIRTY   2
  79 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  80
  81 /* flags for do_journal_end */
  82 #define FLUSH_ALL   1           /* flush commit and real blocks */
  83 #define COMMIT_NOW  2           /* end and commit this transaction */
  84 #define WAIT        4           /* wait for the log blocks to hit the disk */
  85
  86 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  87 static int flush_journal_list(struct super_block *s,
  88                               struct reiserfs_journal_list *jl, int flushall);
  89 static int flush_commit_list(struct super_block *s,
  90                              struct reiserfs_journal_list *jl, int flushall);
  91 static int can_dirty(struct reiserfs_journal_cnode *cn);
  92 static int journal_join(struct reiserfs_transaction_handle *th,
  93                         struct super_block *sb);
  94 static void release_journal_dev(struct super_block *super,
  95                                struct reiserfs_journal *journal);
  96 static int dirty_one_transaction(struct super_block *s,
  97                                  struct reiserfs_journal_list *jl);
  98 static void flush_async_commits(struct work_struct *work);
  99 static void queue_log_writer(struct super_block *s);
 100
 101 /* values for join in do_journal_begin_r */
 102 enum {
 103         JBEGIN_REG = 0,         /* regular journal begin */
 104         /* join the running transaction if at all possible */
 105         JBEGIN_JOIN = 1,
 106         /* called from cleanup code, ignores aborted flag */
 107         JBEGIN_ABORT = 2,
 108 };
 109
 110 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 111                               struct super_block *sb,
 112                               unsigned long nblocks, int join);
 113
 114 static void init_journal_hash(struct super_block *sb)
 115 {
 116         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 117         memset(journal->j_hash_table, 0,
 118                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 119 }
 120
 121 /*
 122  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 123  * I can't allow refile_buffer to make schedule happen after I've freed a
 124  * block.  Look at remove_from_transaction and journal_mark_freed for
 125  * more details.
 126  */
 127 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 128 {
 129         if (bh) {
 130                 clear_buffer_dirty(bh);
 131                 clear_buffer_journal_test(bh);
 132         }
 133         return 0;
 134 }
 135
 136 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 137                                                          *sb)
 138 {
 139         struct reiserfs_bitmap_node *bn;
 140         static int id;
 141
 142         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 143         if (!bn) {
 144                 return NULL;
 145         }
 146         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 147         if (!bn->data) {
 148                 kfree(bn);
 149                 return NULL;
 150         }
 151         bn->id = id++;
 152         INIT_LIST_HEAD(&bn->list);
 153         return bn;
 154 }
 155
 156 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 157 {
 158         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 159         struct reiserfs_bitmap_node *bn = NULL;
 160         struct list_head *entry = journal->j_bitmap_nodes.next;
 161
 162         journal->j_used_bitmap_nodes++;
 163 repeat:
 164
 165         if (entry != &journal->j_bitmap_nodes) {
 166                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 167                 list_del(entry);
 168                 memset(bn->data, 0, sb->s_blocksize);
 169                 journal->j_free_bitmap_nodes--;
 170                 return bn;
 171         }
 172         bn = allocate_bitmap_node(sb);
 173         if (!bn) {
 174                 yield();
 175                 goto repeat;
 176         }
 177         return bn;
 178 }
 179 static inline void free_bitmap_node(struct super_block *sb,
 180                                     struct reiserfs_bitmap_node *bn)
 181 {
 182         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 183         journal->j_used_bitmap_nodes--;
 184         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 185                 kfree(bn->data);
 186                 kfree(bn);
 187         } else {
 188                 list_add(&bn->list, &journal->j_bitmap_nodes);
 189                 journal->j_free_bitmap_nodes++;
 190         }
 191 }
 192
 193 static void allocate_bitmap_nodes(struct super_block *sb)
 194 {
 195         int i;
 196         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 197         struct reiserfs_bitmap_node *bn = NULL;
 198         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 199                 bn = allocate_bitmap_node(sb);
 200                 if (bn) {
 201                         list_add(&bn->list, &journal->j_bitmap_nodes);
 202                         journal->j_free_bitmap_nodes++;
 203                 } else {
 204                         /* this is ok, we'll try again when more are needed */
 205                         break;
 206                 }
 207         }
 208 }
 209
 210 static int set_bit_in_list_bitmap(struct super_block *sb,
 211                                   b_blocknr_t block,
 212                                   struct reiserfs_list_bitmap *jb)
 213 {
 214         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 215         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 216
 217         if (!jb->bitmaps[bmap_nr]) {
 218                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 219         }
 220         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 221         return 0;
 222 }
 223
 224 static void cleanup_bitmap_list(struct super_block *sb,
 225                                 struct reiserfs_list_bitmap *jb)
 226 {
 227         int i;
 228         if (jb->bitmaps == NULL)
 229                 return;
 230
 231         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 232                 if (jb->bitmaps[i]) {
 233                         free_bitmap_node(sb, jb->bitmaps[i]);
 234                         jb->bitmaps[i] = NULL;
 235                 }
 236         }
 237 }
 238
 239 /*
 240  * only call this on FS unmount.
 241  */
 242 static int free_list_bitmaps(struct super_block *sb,
 243                              struct reiserfs_list_bitmap *jb_array)
 244 {
 245         int i;
 246         struct reiserfs_list_bitmap *jb;
 247         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 248                 jb = jb_array + i;
 249                 jb->journal_list = NULL;
 250                 cleanup_bitmap_list(sb, jb);
 251                 vfree(jb->bitmaps);
 252                 jb->bitmaps = NULL;
 253         }
 254         return 0;
 255 }
 256
 257 static int free_bitmap_nodes(struct super_block *sb)
 258 {
 259         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 260         struct list_head *next = journal->j_bitmap_nodes.next;
 261         struct reiserfs_bitmap_node *bn;
 262
 263         while (next != &journal->j_bitmap_nodes) {
 264                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 265                 list_del(next);
 266                 kfree(bn->data);
 267                 kfree(bn);
 268                 next = journal->j_bitmap_nodes.next;
 269                 journal->j_free_bitmap_nodes--;
 270         }
 271
 272         return 0;
 273 }
 274
 275 /*
 276  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 277  * jb_array is the array to be filled in.
 278  */
 279 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 280                                    struct reiserfs_list_bitmap *jb_array,
 281                                    unsigned int bmap_nr)
 282 {
 283         int i;
 284         int failed = 0;
 285         struct reiserfs_list_bitmap *jb;
 286         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 287
 288         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 289                 jb = jb_array + i;
 290                 jb->journal_list = NULL;
 291                 jb->bitmaps = vzalloc(mem);
 292                 if (!jb->bitmaps) {
 293                         reiserfs_warning(sb, "clm-2000", "unable to "
 294                                          "allocate bitmaps for journal lists");
 295                         failed = 1;
 296                         break;
 297                 }
 298         }
 299         if (failed) {
 300                 free_list_bitmaps(sb, jb_array);
 301                 return -1;
 302         }
 303         return 0;
 304 }
 305
 306 /*
 307  * find an available list bitmap.  If you can't find one, flush a commit list
 308  * and try again
 309  */
 310 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 311                                                     struct reiserfs_journal_list
 312                                                     *jl)
 313 {
 314         int i, j;
 315         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 316         struct reiserfs_list_bitmap *jb = NULL;
 317
 318         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 319                 i = journal->j_list_bitmap_index;
 320                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 321                 jb = journal->j_list_bitmap + i;
 322                 if (journal->j_list_bitmap[i].journal_list) {
 323                         flush_commit_list(sb,
 324                                           journal->j_list_bitmap[i].
 325                                           journal_list, 1);
 326                         if (!journal->j_list_bitmap[i].journal_list) {
 327                                 break;
 328                         }
 329                 } else {
 330                         break;
 331                 }
 332         }
 333         /* double check to make sure if flushed correctly */
 334         if (jb->journal_list)
 335                 return NULL;
 336         jb->journal_list = jl;
 337         return jb;
 338 }
 339
 340 /*
 341  * allocates a new chunk of X nodes, and links them all together as a list.
 342  * Uses the cnode->next and cnode->prev pointers
 343  * returns NULL on failure
 344  */
 345 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 346 {
 347         struct reiserfs_journal_cnode *head;
 348         int i;
 349         if (num_cnodes <= 0) {
 350                 return NULL;
 351         }
 352         head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 353         if (!head) {
 354                 return NULL;
 355         }
 356         head[0].prev = NULL;
 357         head[0].next = head + 1;
 358         for (i = 1; i < num_cnodes; i++) {
 359                 head[i].prev = head + (i - 1);
 360                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 361         }
 362         head[num_cnodes - 1].next = NULL;
 363         return head;
 364 }
 365
 366 /* pulls a cnode off the free list, or returns NULL on failure */
 367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 368 {
 369         struct reiserfs_journal_cnode *cn;
 370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 371
 372         reiserfs_check_lock_depth(sb, "get_cnode");
 373
 374         if (journal->j_cnode_free <= 0) {
 375                 return NULL;
 376         }
 377         journal->j_cnode_used++;
 378         journal->j_cnode_free--;
 379         cn = journal->j_cnode_free_list;
 380         if (!cn) {
 381                 return cn;
 382         }
 383         if (cn->next) {
 384                 cn->next->prev = NULL;
 385         }
 386         journal->j_cnode_free_list = cn->next;
 387         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 388         return cn;
 389 }
 390
 391 /*
 392  * returns a cnode to the free list
 393  */
 394 static void free_cnode(struct super_block *sb,
 395                        struct reiserfs_journal_cnode *cn)
 396 {
 397         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 398
 399         reiserfs_check_lock_depth(sb, "free_cnode");
 400
 401         journal->j_cnode_used--;
 402         journal->j_cnode_free++;
 403         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 404         cn->next = journal->j_cnode_free_list;
 405         if (journal->j_cnode_free_list) {
 406                 journal->j_cnode_free_list->prev = cn;
 407         }
 408         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 409         journal->j_cnode_free_list = cn;
 410 }
 411
 412 static void clear_prepared_bits(struct buffer_head *bh)
 413 {
 414         clear_buffer_journal_prepared(bh);
 415         clear_buffer_journal_restore_dirty(bh);
 416 }
 417
 418 /*
 419  * return a cnode with same dev, block number and size in table,
 420  * or null if not found
 421  */
 422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 423                                                                   super_block
 424                                                                   *sb,
 425                                                                   struct
 426                                                                   reiserfs_journal_cnode
 427                                                                   **table,
 428                                                                   long bl)
 429 {
 430         struct reiserfs_journal_cnode *cn;
 431         cn = journal_hash(table, sb, bl);
 432         while (cn) {
 433                 if (cn->blocknr == bl && cn->sb == sb)
 434                         return cn;
 435                 cn = cn->hnext;
 436         }
 437         return (struct reiserfs_journal_cnode *)0;
 438 }
 439
 440 /*
 441  * this actually means 'can this block be reallocated yet?'.  If you set
 442  * search_all, a block can only be allocated if it is not in the current
 443  * transaction, was not freed by the current transaction, and has no chance
 444  * of ever being overwritten by a replay after crashing.
 445  *
 446  * If you don't set search_all, a block can only be allocated if it is not
 447  * in the current transaction.  Since deleting a block removes it from the
 448  * current transaction, this case should never happen.  If you don't set
 449  * search_all, make sure you never write the block without logging it.
 450  *
 451  * next_zero_bit is a suggestion about the next block to try for find_forward.
 452  * when bl is rejected because it is set in a journal list bitmap, we search
 453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
 454  * that through next_zero_bit for find_forward to try.
 455  *
 456  * Just because we return something in next_zero_bit does not mean we won't
 457  * reject it on the next call to reiserfs_in_journal
 458  */
 459 int reiserfs_in_journal(struct super_block *sb,
 460                         unsigned int bmap_nr, int bit_nr, int search_all,
 461                         b_blocknr_t * next_zero_bit)
 462 {
 463         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 464         struct reiserfs_journal_cnode *cn;
 465         struct reiserfs_list_bitmap *jb;
 466         int i;
 467         unsigned long bl;
 468
 469         *next_zero_bit = 0;     /* always start this at zero. */
 470
 471         PROC_INFO_INC(sb, journal.in_journal);
 472         /*
 473          * If we aren't doing a search_all, this is a metablock, and it
 474          * will be logged before use.  if we crash before the transaction
 475          * that freed it commits,  this transaction won't have committed
 476          * either, and the block will never be written
 477          */
 478         if (search_all) {
 479                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 480                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 481                         jb = journal->j_list_bitmap + i;
 482                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 483                             test_bit(bit_nr,
 484                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 485                                      data)) {
 486                                 *next_zero_bit =
 487                                     find_next_zero_bit((unsigned long *)
 488                                                        (jb->bitmaps[bmap_nr]->
 489                                                         data),
 490                                                        sb->s_blocksize << 3,
 491                                                        bit_nr + 1);
 492                                 return 1;
 493                         }
 494                 }
 495         }
 496
 497         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 498         /* is it in any old transactions? */
 499         if (search_all
 500             && (cn =
 501                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 502                 return 1;
 503         }
 504
 505         /* is it in the current transaction.  This should never happen */
 506         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 507                 BUG();
 508                 return 1;
 509         }
 510
 511         PROC_INFO_INC(sb, journal.in_journal_reusable);
 512         /* safe for reuse */
 513         return 0;
 514 }
 515
 516 /* insert cn into table */
 517 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 518                                        struct reiserfs_journal_cnode *cn)
 519 {
 520         struct reiserfs_journal_cnode *cn_orig;
 521
 522         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 523         cn->hnext = cn_orig;
 524         cn->hprev = NULL;
 525         if (cn_orig) {
 526                 cn_orig->hprev = cn;
 527         }
 528         journal_hash(table, cn->sb, cn->blocknr) = cn;
 529 }
 530
 531 /* lock the current transaction */
 532 static inline void lock_journal(struct super_block *sb)
 533 {
 534         PROC_INFO_INC(sb, journal.lock_journal);
 535
 536         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 537 }
 538
 539 /* unlock the current transaction */
 540 static inline void unlock_journal(struct super_block *sb)
 541 {
 542         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 543 }
 544
 545 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 546 {
 547         jl->j_refcount++;
 548 }
 549
 550 static inline void put_journal_list(struct super_block *s,
 551                                     struct reiserfs_journal_list *jl)
 552 {
 553         if (jl->j_refcount < 1) {
 554                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 555                                jl->j_trans_id, jl->j_refcount);
 556         }
 557         if (--jl->j_refcount == 0)
 558                 kfree(jl);
 559 }
 560
 561 /*
 562  * this used to be much more involved, and I'm keeping it just in case
 563  * things get ugly again.  it gets called by flush_commit_list, and
 564  * cleans up any data stored about blocks freed during a transaction.
 565  */
 566 static void cleanup_freed_for_journal_list(struct super_block *sb,
 567                                            struct reiserfs_journal_list *jl)
 568 {
 569
 570         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 571         if (jb) {
 572                 cleanup_bitmap_list(sb, jb);
 573         }
 574         jl->j_list_bitmap->journal_list = NULL;
 575         jl->j_list_bitmap = NULL;
 576 }
 577
 578 static int journal_list_still_alive(struct super_block *s,
 579                                     unsigned int trans_id)
 580 {
 581         struct reiserfs_journal *journal = SB_JOURNAL(s);
 582         struct list_head *entry = &journal->j_journal_list;
 583         struct reiserfs_journal_list *jl;
 584
 585         if (!list_empty(entry)) {
 586                 jl = JOURNAL_LIST_ENTRY(entry->next);
 587                 if (jl->j_trans_id <= trans_id) {
 588                         return 1;
 589                 }
 590         }
 591         return 0;
 592 }
 593
 594 /*
 595  * If page->mapping was null, we failed to truncate this page for
 596  * some reason.  Most likely because it was truncated after being
 597  * logged via data=journal.
 598  *
 599  * This does a check to see if the buffer belongs to one of these
 600  * lost pages before doing the final put_bh.  If page->mapping was
 601  * null, it tries to free buffers on the page, which should make the
 602  * final page_cache_release drop the page from the lru.
 603  */
 604 static void release_buffer_page(struct buffer_head *bh)
 605 {
 606         struct page *page = bh->b_page;
 607         if (!page->mapping && trylock_page(page)) {
 608                 page_cache_get(page);
 609                 put_bh(bh);
 610                 if (!page->mapping)
 611                         try_to_free_buffers(page);
 612                 unlock_page(page);
 613                 page_cache_release(page);
 614         } else {
 615                 put_bh(bh);
 616         }
 617 }
 618
 619 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 620 {
 621         char b[BDEVNAME_SIZE];
 622
 623         if (buffer_journaled(bh)) {
 624                 reiserfs_warning(NULL, "clm-2084",
 625                                  "pinned buffer %lu:%s sent to disk",
 626                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 627         }
 628         if (uptodate)
 629                 set_buffer_uptodate(bh);
 630         else
 631                 clear_buffer_uptodate(bh);
 632
 633         unlock_buffer(bh);
 634         release_buffer_page(bh);
 635 }
 636
 637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 638 {
 639         if (uptodate)
 640                 set_buffer_uptodate(bh);
 641         else
 642                 clear_buffer_uptodate(bh);
 643         unlock_buffer(bh);
 644         put_bh(bh);
 645 }
 646
 647 static void submit_logged_buffer(struct buffer_head *bh)
 648 {
 649         get_bh(bh);
 650         bh->b_end_io = reiserfs_end_buffer_io_sync;
 651         clear_buffer_journal_new(bh);
 652         clear_buffer_dirty(bh);
 653         if (!test_clear_buffer_journal_test(bh))
 654                 BUG();
 655         if (!buffer_uptodate(bh))
 656                 BUG();
 657         submit_bh(WRITE, bh);
 658 }
 659
 660 static void submit_ordered_buffer(struct buffer_head *bh)
 661 {
 662         get_bh(bh);
 663         bh->b_end_io = reiserfs_end_ordered_io;
 664         clear_buffer_dirty(bh);
 665         if (!buffer_uptodate(bh))
 666                 BUG();
 667         submit_bh(WRITE, bh);
 668 }
 669
 670 #define CHUNK_SIZE 32
 671 struct buffer_chunk {
 672         struct buffer_head *bh[CHUNK_SIZE];
 673         int nr;
 674 };
 675
 676 static void write_chunk(struct buffer_chunk *chunk)
 677 {
 678         int i;
 679         for (i = 0; i < chunk->nr; i++) {
 680                 submit_logged_buffer(chunk->bh[i]);
 681         }
 682         chunk->nr = 0;
 683 }
 684
 685 static void write_ordered_chunk(struct buffer_chunk *chunk)
 686 {
 687         int i;
 688         for (i = 0; i < chunk->nr; i++) {
 689                 submit_ordered_buffer(chunk->bh[i]);
 690         }
 691         chunk->nr = 0;
 692 }
 693
 694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 695                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 696 {
 697         int ret = 0;
 698         BUG_ON(chunk->nr >= CHUNK_SIZE);
 699         chunk->bh[chunk->nr++] = bh;
 700         if (chunk->nr >= CHUNK_SIZE) {
 701                 ret = 1;
 702                 if (lock) {
 703                         spin_unlock(lock);
 704                         fn(chunk);
 705                         spin_lock(lock);
 706                 } else {
 707                         fn(chunk);
 708                 }
 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                 /*
 767                  * buffer must be locked for __add_jh, should be able to have
 768                  * two adds at the same time
 769                  */
 770                 BUG_ON(bh->b_private);
 771                 jh->bh = bh;
 772                 bh->b_private = jh;
 773         }
 774         jh->jl = j->j_current_jl;
 775         if (tail)
 776                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 777         else {
 778                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 779         }
 780         spin_unlock(&j->j_dirty_buffers_lock);
 781         return 0;
 782 }
 783
 784 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 785 {
 786         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 787 }
 788 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 789 {
 790         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 791 }
 792
 793 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 794 static int write_ordered_buffers(spinlock_t * lock,
 795                                  struct reiserfs_journal *j,
 796                                  struct reiserfs_journal_list *jl,
 797                                  struct list_head *list)
 798 {
 799         struct buffer_head *bh;
 800         struct reiserfs_jh *jh;
 801         int ret = j->j_errno;
 802         struct buffer_chunk chunk;
 803         struct list_head tmp;
 804         INIT_LIST_HEAD(&tmp);
 805
 806         chunk.nr = 0;
 807         spin_lock(lock);
 808         while (!list_empty(list)) {
 809                 jh = JH_ENTRY(list->next);
 810                 bh = jh->bh;
 811                 get_bh(bh);
 812                 if (!trylock_buffer(bh)) {
 813                         if (!buffer_dirty(bh)) {
 814                                 list_move(&jh->list, &tmp);
 815                                 goto loop_next;
 816                         }
 817                         spin_unlock(lock);
 818                         if (chunk.nr)
 819                                 write_ordered_chunk(&chunk);
 820                         wait_on_buffer(bh);
 821                         cond_resched();
 822                         spin_lock(lock);
 823                         goto loop_next;
 824                 }
 825                 /*
 826                  * in theory, dirty non-uptodate buffers should never get here,
 827                  * but the upper layer io error paths still have a few quirks.
 828                  * Handle them here as gracefully as we can
 829                  */
 830                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 831                         clear_buffer_dirty(bh);
 832                         ret = -EIO;
 833                 }
 834                 if (buffer_dirty(bh)) {
 835                         list_move(&jh->list, &tmp);
 836                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 837                 } else {
 838                         reiserfs_free_jh(bh);
 839                         unlock_buffer(bh);
 840                 }
 841 loop_next:
 842                 put_bh(bh);
 843                 cond_resched_lock(lock);
 844         }
 845         if (chunk.nr) {
 846                 spin_unlock(lock);
 847                 write_ordered_chunk(&chunk);
 848                 spin_lock(lock);
 849         }
 850         while (!list_empty(&tmp)) {
 851                 jh = JH_ENTRY(tmp.prev);
 852                 bh = jh->bh;
 853                 get_bh(bh);
 854                 reiserfs_free_jh(bh);
 855
 856                 if (buffer_locked(bh)) {
 857                         spin_unlock(lock);
 858                         wait_on_buffer(bh);
 859                         spin_lock(lock);
 860                 }
 861                 if (!buffer_uptodate(bh)) {
 862                         ret = -EIO;
 863                 }
 864                 /*
 865                  * ugly interaction with invalidatepage here.
 866                  * reiserfs_invalidate_page will pin any buffer that has a
 867                  * valid journal head from an older transaction.  If someone
 868                  * else sets our buffer dirty after we write it in the first
 869                  * loop, and then someone truncates the page away, nobody
 870                  * will ever write the buffer. We're safe if we write the
 871                  * page one last time after freeing the journal header.
 872                  */
 873                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 874                         spin_unlock(lock);
 875                         ll_rw_block(WRITE, 1, &bh);
 876                         spin_lock(lock);
 877                 }
 878                 put_bh(bh);
 879                 cond_resched_lock(lock);
 880         }
 881         spin_unlock(lock);
 882         return ret;
 883 }
 884
 885 static int flush_older_commits(struct super_block *s,
 886                                struct reiserfs_journal_list *jl)
 887 {
 888         struct reiserfs_journal *journal = SB_JOURNAL(s);
 889         struct reiserfs_journal_list *other_jl;
 890         struct reiserfs_journal_list *first_jl;
 891         struct list_head *entry;
 892         unsigned int trans_id = jl->j_trans_id;
 893         unsigned int other_trans_id;
 894         unsigned int first_trans_id;
 895
 896 find_first:
 897         /*
 898          * first we walk backwards to find the oldest uncommitted transation
 899          */
 900         first_jl = jl;
 901         entry = jl->j_list.prev;
 902         while (1) {
 903                 other_jl = JOURNAL_LIST_ENTRY(entry);
 904                 if (entry == &journal->j_journal_list ||
 905                     atomic_read(&other_jl->j_older_commits_done))
 906                         break;
 907
 908                 first_jl = other_jl;
 909                 entry = other_jl->j_list.prev;
 910         }
 911
 912         /* if we didn't find any older uncommitted transactions, return now */
 913         if (first_jl == jl) {
 914                 return 0;
 915         }
 916
 917         first_trans_id = first_jl->j_trans_id;
 918
 919         entry = &first_jl->j_list;
 920         while (1) {
 921                 other_jl = JOURNAL_LIST_ENTRY(entry);
 922                 other_trans_id = other_jl->j_trans_id;
 923
 924                 if (other_trans_id < trans_id) {
 925                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 926                                 flush_commit_list(s, other_jl, 0);
 927
 928                                 /* list we were called with is gone, return */
 929                                 if (!journal_list_still_alive(s, trans_id))
 930                                         return 1;
 931
 932                                 /*
 933                                  * the one we just flushed is gone, this means
 934                                  * all older lists are also gone, so first_jl
 935                                  * is no longer valid either.  Go back to the
 936                                  * beginning.
 937                                  */
 938                                 if (!journal_list_still_alive
 939                                     (s, other_trans_id)) {
 940                                         goto find_first;
 941                                 }
 942                         }
 943                         entry = entry->next;
 944                         if (entry == &journal->j_journal_list)
 945                                 return 0;
 946                 } else {
 947                         return 0;
 948                 }
 949         }
 950         return 0;
 951 }
 952
 953 static int reiserfs_async_progress_wait(struct super_block *s)
 954 {
 955         struct reiserfs_journal *j = SB_JOURNAL(s);
 956
 957         if (atomic_read(&j->j_async_throttle)) {
 958                 int depth;
 959
 960                 depth = reiserfs_write_unlock_nested(s);
 961                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 962                 reiserfs_write_lock_nested(s, depth);
 963         }
 964
 965         return 0;
 966 }
 967
 968 /*
 969  * if this journal list still has commit blocks unflushed, send them to disk.
 970  *
 971  * log areas must be flushed in order (transaction 2 can't commit before
 972  * transaction 1) Before the commit block can by written, every other log
 973  * block must be safely on disk
 974  */
 975 static int flush_commit_list(struct super_block *s,
 976                              struct reiserfs_journal_list *jl, int flushall)
 977 {
 978         int i;
 979         b_blocknr_t bn;
 980         struct buffer_head *tbh = NULL;
 981         unsigned int trans_id = jl->j_trans_id;
 982         struct reiserfs_journal *journal = SB_JOURNAL(s);
 983         int retval = 0;
 984         int write_len;
 985         int depth;
 986
 987         reiserfs_check_lock_depth(s, "flush_commit_list");
 988
 989         if (atomic_read(&jl->j_older_commits_done)) {
 990                 return 0;
 991         }
 992
 993         /*
 994          * before we can put our commit blocks on disk, we have to make
 995          * sure everyone older than us is on disk too
 996          */
 997         BUG_ON(jl->j_len <= 0);
 998         BUG_ON(trans_id == journal->j_trans_id);
 999
1000         get_journal_list(jl);
1001         if (flushall) {
1002                 if (flush_older_commits(s, jl) == 1) {
1003                         /*
1004                          * list disappeared during flush_older_commits.
1005                          * return
1006                          */
1007                         goto put_jl;
1008                 }
1009         }
1010
1011         /* make sure nobody is trying to flush this one at the same time */
1012         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1013
1014         if (!journal_list_still_alive(s, trans_id)) {
1015                 mutex_unlock(&jl->j_commit_mutex);
1016                 goto put_jl;
1017         }
1018         BUG_ON(jl->j_trans_id == 0);
1019
1020         /* this commit is done, exit */
1021         if (atomic_read(&jl->j_commit_left) <= 0) {
1022                 if (flushall) {
1023                         atomic_set(&jl->j_older_commits_done, 1);
1024                 }
1025                 mutex_unlock(&jl->j_commit_mutex);
1026                 goto put_jl;
1027         }
1028
1029         if (!list_empty(&jl->j_bh_list)) {
1030                 int ret;
1031
1032                 /*
1033                  * We might sleep in numerous places inside
1034                  * write_ordered_buffers. Relax the write lock.
1035                  */
1036                 depth = reiserfs_write_unlock_nested(s);
1037                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1038                                             journal, jl, &jl->j_bh_list);
1039                 if (ret < 0 && retval == 0)
1040                         retval = ret;
1041                 reiserfs_write_lock_nested(s, depth);
1042         }
1043         BUG_ON(!list_empty(&jl->j_bh_list));
1044         /*
1045          * for the description block and all the log blocks, submit any buffers
1046          * that haven't already reached the disk.  Try to write at least 256
1047          * log blocks. later on, we will only wait on blocks that correspond
1048          * to this transaction, but while we're unplugging we might as well
1049          * get a chunk of data on there.
1050          */
1051         atomic_inc(&journal->j_async_throttle);
1052         write_len = jl->j_len + 1;
1053         if (write_len < 256)
1054                 write_len = 256;
1055         for (i = 0 ; i < write_len ; i++) {
1056                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1057                     SB_ONDISK_JOURNAL_SIZE(s);
1058                 tbh = journal_find_get_block(s, bn);
1059                 if (tbh) {
1060                         if (buffer_dirty(tbh)) {
1061                             depth = reiserfs_write_unlock_nested(s);
1062                             ll_rw_block(WRITE, 1, &tbh);
1063                             reiserfs_write_lock_nested(s, depth);
1064                         }
1065                         put_bh(tbh) ;
1066                 }
1067         }
1068         atomic_dec(&journal->j_async_throttle);
1069
1070         for (i = 0; i < (jl->j_len + 1); i++) {
1071                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1072                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1073                 tbh = journal_find_get_block(s, bn);
1074
1075                 depth = reiserfs_write_unlock_nested(s);
1076                 __wait_on_buffer(tbh);
1077                 reiserfs_write_lock_nested(s, depth);
1078                 /*
1079                  * since we're using ll_rw_blk above, it might have skipped
1080                  * over a locked buffer.  Double check here
1081                  */
1082                 /* redundant, sync_dirty_buffer() checks */
1083                 if (buffer_dirty(tbh)) {
1084                         depth = reiserfs_write_unlock_nested(s);
1085                         sync_dirty_buffer(tbh);
1086                         reiserfs_write_lock_nested(s, depth);
1087                 }
1088                 if (unlikely(!buffer_uptodate(tbh))) {
1089 #ifdef CONFIG_REISERFS_CHECK
1090                         reiserfs_warning(s, "journal-601",
1091                                          "buffer write failed");
1092 #endif
1093                         retval = -EIO;
1094                 }
1095                 /* once for journal_find_get_block */
1096                 put_bh(tbh);
1097                 /* once due to original getblk in do_journal_end */
1098                 put_bh(tbh);
1099                 atomic_dec(&jl->j_commit_left);
1100         }
1101
1102         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1103
1104         /*
1105          * If there was a write error in the journal - we can't commit
1106          * this transaction - it will be invalid and, if successful,
1107          * will just end up propagating the write error out to
1108          * the file system.
1109          */
1110         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1111                 if (buffer_dirty(jl->j_commit_bh))
1112                         BUG();
1113                 mark_buffer_dirty(jl->j_commit_bh) ;
1114                 depth = reiserfs_write_unlock_nested(s);
1115                 if (reiserfs_barrier_flush(s))
1116                         __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1117                 else
1118                         sync_dirty_buffer(jl->j_commit_bh);
1119                 reiserfs_write_lock_nested(s, depth);
1120         }
1121
1122         /*
1123          * If there was a write error in the journal - we can't commit this
1124          * transaction - it will be invalid and, if successful, will just end
1125          * up propagating the write error out to the filesystem.
1126          */
1127         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1128 #ifdef CONFIG_REISERFS_CHECK
1129                 reiserfs_warning(s, "journal-615", "buffer write failed");
1130 #endif
1131                 retval = -EIO;
1132         }
1133         bforget(jl->j_commit_bh);
1134         if (journal->j_last_commit_id != 0 &&
1135             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1136                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1137                                  journal->j_last_commit_id, jl->j_trans_id);
1138         }
1139         journal->j_last_commit_id = jl->j_trans_id;
1140
1141         /*
1142          * now, every commit block is on the disk.  It is safe to allow
1143          * blocks freed during this transaction to be reallocated
1144          */
1145         cleanup_freed_for_journal_list(s, jl);
1146
1147         retval = retval ? retval : journal->j_errno;
1148
1149         /* mark the metadata dirty */
1150         if (!retval)
1151                 dirty_one_transaction(s, jl);
1152         atomic_dec(&jl->j_commit_left);
1153
1154         if (flushall) {
1155                 atomic_set(&jl->j_older_commits_done, 1);
1156         }
1157         mutex_unlock(&jl->j_commit_mutex);
1158 put_jl:
1159         put_journal_list(s, jl);
1160
1161         if (retval)
1162                 reiserfs_abort(s, retval, "Journal write error in %s",
1163                                __func__);
1164         return retval;
1165 }
1166
1167 /*
1168  * flush_journal_list frequently needs to find a newer transaction for a
1169  * given block.  This does that, or returns NULL if it can't find anything
1170  */
1171 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1172                                                           reiserfs_journal_cnode
1173                                                           *cn)
1174 {
1175         struct super_block *sb = cn->sb;
1176         b_blocknr_t blocknr = cn->blocknr;
1177
1178         cn = cn->hprev;
1179         while (cn) {
1180                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1181                         return cn->jlist;
1182                 }
1183                 cn = cn->hprev;
1184         }
1185         return NULL;
1186 }
1187
1188 static void remove_journal_hash(struct super_block *,
1189                                 struct reiserfs_journal_cnode **,
1190                                 struct reiserfs_journal_list *, unsigned long,
1191                                 int);
1192
1193 /*
1194  * once all the real blocks have been flushed, it is safe to remove them
1195  * from the journal list for this transaction.  Aside from freeing the
1196  * cnode, this also allows the block to be reallocated for data blocks
1197  * if it had been deleted.
1198  */
1199 static void remove_all_from_journal_list(struct super_block *sb,
1200                                          struct reiserfs_journal_list *jl,
1201                                          int debug)
1202 {
1203         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1204         struct reiserfs_journal_cnode *cn, *last;
1205         cn = jl->j_realblock;
1206
1207         /*
1208          * which is better, to lock once around the whole loop, or
1209          * to lock for each call to remove_journal_hash?
1210          */
1211         while (cn) {
1212                 if (cn->blocknr != 0) {
1213                         if (debug) {
1214                                 reiserfs_warning(sb, "reiserfs-2201",
1215                                                  "block %u, bh is %d, state %ld",
1216                                                  cn->blocknr, cn->bh ? 1 : 0,
1217                                                  cn->state);
1218                         }
1219                         cn->state = 0;
1220                         remove_journal_hash(sb, journal->j_list_hash_table,
1221                                             jl, cn->blocknr, 1);
1222                 }
1223                 last = cn;
1224                 cn = cn->next;
1225                 free_cnode(sb, last);
1226         }
1227         jl->j_realblock = NULL;
1228 }
1229
1230 /*
1231  * if this timestamp is greater than the timestamp we wrote last to the
1232  * header block, write it to the header block.  once this is done, I can
1233  * safely say the log area for this transaction won't ever be replayed,
1234  * and I can start releasing blocks in this transaction for reuse as data
1235  * blocks.  called by flush_journal_list, before it calls
1236  * remove_all_from_journal_list
1237  */
1238 static int _update_journal_header_block(struct super_block *sb,
1239                                         unsigned long offset,
1240                                         unsigned int trans_id)
1241 {
1242         struct reiserfs_journal_header *jh;
1243         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1244         int depth;
1245
1246         if (reiserfs_is_journal_aborted(journal))
1247                 return -EIO;
1248
1249         if (trans_id >= journal->j_last_flush_trans_id) {
1250                 if (buffer_locked((journal->j_header_bh))) {
1251                         depth = reiserfs_write_unlock_nested(sb);
1252                         __wait_on_buffer(journal->j_header_bh);
1253                         reiserfs_write_lock_nested(sb, depth);
1254                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1255 #ifdef CONFIG_REISERFS_CHECK
1256                                 reiserfs_warning(sb, "journal-699",
1257                                                  "buffer write failed");
1258 #endif
1259                                 return -EIO;
1260                         }
1261                 }
1262                 journal->j_last_flush_trans_id = trans_id;
1263                 journal->j_first_unflushed_offset = offset;
1264                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1265                                                         b_data);
1266                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1267                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1268                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1269
1270                 set_buffer_dirty(journal->j_header_bh);
1271                 depth = reiserfs_write_unlock_nested(sb);
1272
1273                 if (reiserfs_barrier_flush(sb))
1274                         __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1275                 else
1276                         sync_dirty_buffer(journal->j_header_bh);
1277
1278                 reiserfs_write_lock_nested(sb, depth);
1279                 if (!buffer_uptodate(journal->j_header_bh)) {
1280                         reiserfs_warning(sb, "journal-837",
1281                                          "IO error during journal replay");
1282                         return -EIO;
1283                 }
1284         }
1285         return 0;
1286 }
1287
1288 static int update_journal_header_block(struct super_block *sb,
1289                                        unsigned long offset,
1290                                        unsigned int trans_id)
1291 {
1292         return _update_journal_header_block(sb, offset, trans_id);
1293 }
1294
1295 /*
1296 ** flush any and all journal lists older than you are
1297 ** can only be called from flush_journal_list
1298 */
1299 static int flush_older_journal_lists(struct super_block *sb,
1300                                      struct reiserfs_journal_list *jl)
1301 {
1302         struct list_head *entry;
1303         struct reiserfs_journal_list *other_jl;
1304         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1305         unsigned int trans_id = jl->j_trans_id;
1306
1307         /*
1308          * we know we are the only ones flushing things, no extra race
1309          * protection is required.
1310          */
1311 restart:
1312         entry = journal->j_journal_list.next;
1313         /* Did we wrap? */
1314         if (entry == &journal->j_journal_list)
1315                 return 0;
1316         other_jl = JOURNAL_LIST_ENTRY(entry);
1317         if (other_jl->j_trans_id < trans_id) {
1318                 BUG_ON(other_jl->j_refcount <= 0);
1319                 /* do not flush all */
1320                 flush_journal_list(sb, other_jl, 0);
1321
1322                 /* other_jl is now deleted from the list */
1323                 goto restart;
1324         }
1325         return 0;
1326 }
1327
1328 static void del_from_work_list(struct super_block *s,
1329                                struct reiserfs_journal_list *jl)
1330 {
1331         struct reiserfs_journal *journal = SB_JOURNAL(s);
1332         if (!list_empty(&jl->j_working_list)) {
1333                 list_del_init(&jl->j_working_list);
1334                 journal->j_num_work_lists--;
1335         }
1336 }
1337
1338 /*
1339  * flush a journal list, both commit and real blocks
1340  *
1341  * always set flushall to 1, unless you are calling from inside
1342  * flush_journal_list
1343  *
1344  * IMPORTANT.  This can only be called while there are no journal writers,
1345  * and the journal is locked.  That means it can only be called from
1346  * do_journal_end, or by journal_release
1347  */
1348 static int flush_journal_list(struct super_block *s,
1349                               struct reiserfs_journal_list *jl, int flushall)
1350 {
1351         struct reiserfs_journal_list *pjl;
1352         struct reiserfs_journal_cnode *cn, *last;
1353         int count;
1354         int was_jwait = 0;
1355         int was_dirty = 0;
1356         struct buffer_head *saved_bh;
1357         unsigned long j_len_saved = jl->j_len;
1358         struct reiserfs_journal *journal = SB_JOURNAL(s);
1359         int err = 0;
1360         int depth;
1361
1362         BUG_ON(j_len_saved <= 0);
1363
1364         if (atomic_read(&journal->j_wcount) != 0) {
1365                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1366                                  atomic_read(&journal->j_wcount));
1367         }
1368
1369         /* if flushall == 0, the lock is already held */
1370         if (flushall) {
1371                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1372         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1373                 BUG();
1374         }
1375
1376         count = 0;
1377         if (j_len_saved > journal->j_trans_max) {
1378                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1379                                j_len_saved, jl->j_trans_id);
1380                 return 0;
1381         }
1382
1383         /* if all the work is already done, get out of here */
1384         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1385             atomic_read(&jl->j_commit_left) <= 0) {
1386                 goto flush_older_and_return;
1387         }
1388
1389         /*
1390          * start by putting the commit list on disk.  This will also flush
1391          * the commit lists of any olders transactions
1392          */
1393         flush_commit_list(s, jl, 1);
1394
1395         if (!(jl->j_state & LIST_DIRTY)
1396             && !reiserfs_is_journal_aborted(journal))
1397                 BUG();
1398
1399         /* are we done now? */
1400         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1401             atomic_read(&jl->j_commit_left) <= 0) {
1402                 goto flush_older_and_return;
1403         }
1404
1405         /*
1406          * loop through each cnode, see if we need to write it,
1407          * or wait on a more recent transaction, or just ignore it
1408          */
1409         if (atomic_read(&journal->j_wcount) != 0) {
1410                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1411                                "wcount is not 0");
1412         }
1413         cn = jl->j_realblock;
1414         while (cn) {
1415                 was_jwait = 0;
1416                 was_dirty = 0;
1417                 saved_bh = NULL;
1418                 /* blocknr of 0 is no longer in the hash, ignore it */
1419                 if (cn->blocknr == 0) {
1420                         goto free_cnode;
1421                 }
1422
1423                 /*
1424                  * This transaction failed commit.
1425                  * Don't write out to the disk
1426                  */
1427                 if (!(jl->j_state & LIST_DIRTY))
1428                         goto free_cnode;
1429
1430                 pjl = find_newer_jl_for_cn(cn);
1431                 /*
1432                  * the order is important here.  We check pjl to make sure we
1433                  * don't clear BH_JDirty_wait if we aren't the one writing this
1434                  * block to disk
1435                  */
1436                 if (!pjl && cn->bh) {
1437                         saved_bh = cn->bh;
1438
1439                         /*
1440                          * we do this to make sure nobody releases the
1441                          * buffer while we are working with it
1442                          */
1443                         get_bh(saved_bh);
1444
1445                         if (buffer_journal_dirty(saved_bh)) {
1446                                 BUG_ON(!can_dirty(cn));
1447                                 was_jwait = 1;
1448                                 was_dirty = 1;
1449                         } else if (can_dirty(cn)) {
1450                                 /*
1451                                  * everything with !pjl && jwait
1452                                  * should be writable
1453                                  */
1454                                 BUG();
1455                         }
1456                 }
1457
1458                 /*
1459                  * if someone has this block in a newer transaction, just make
1460                  * sure they are committed, and don't try writing it to disk
1461                  */
1462                 if (pjl) {
1463                         if (atomic_read(&pjl->j_commit_left))
1464                                 flush_commit_list(s, pjl, 1);
1465                         goto free_cnode;
1466                 }
1467
1468                 /*
1469                  * bh == NULL when the block got to disk on its own, OR,
1470                  * the block got freed in a future transaction
1471                  */
1472                 if (saved_bh == NULL) {
1473                         goto free_cnode;
1474                 }
1475
1476                 /*
1477                  * this should never happen.  kupdate_one_transaction has
1478                  * this list locked while it works, so we should never see a
1479                  * buffer here that is not marked JDirty_wait
1480                  */
1481                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1482                         reiserfs_warning(s, "journal-813",
1483                                          "BAD! buffer %llu %cdirty %cjwait, "
1484                                          "not in a newer tranasction",
1485                                          (unsigned long long)saved_bh->
1486                                          b_blocknr, was_dirty ? ' ' : '!',
1487                                          was_jwait ? ' ' : '!');
1488                 }
1489                 if (was_dirty) {
1490                         /*
1491                          * we inc again because saved_bh gets decremented
1492                          * at free_cnode
1493                          */
1494                         get_bh(saved_bh);
1495                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1496                         lock_buffer(saved_bh);
1497                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1498                         if (buffer_dirty(saved_bh))
1499                                 submit_logged_buffer(saved_bh);
1500                         else
1501                                 unlock_buffer(saved_bh);
1502                         count++;
1503                 } else {
1504                         reiserfs_warning(s, "clm-2082",
1505                                          "Unable to flush buffer %llu in %s",
1506                                          (unsigned long long)saved_bh->
1507                                          b_blocknr, __func__);
1508                 }
1509 free_cnode:
1510                 last = cn;
1511                 cn = cn->next;
1512                 if (saved_bh) {
1513                         /*
1514                          * we incremented this to keep others from
1515                          * taking the buffer head away
1516                          */
1517                         put_bh(saved_bh);
1518                         if (atomic_read(&saved_bh->b_count) < 0) {
1519                                 reiserfs_warning(s, "journal-945",
1520                                                  "saved_bh->b_count < 0");
1521                         }
1522                 }
1523         }
1524         if (count > 0) {
1525                 cn = jl->j_realblock;
1526                 while (cn) {
1527                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1528                                 if (!cn->bh) {
1529                                         reiserfs_panic(s, "journal-1011",
1530                                                        "cn->bh is NULL");
1531                                 }
1532
1533                                 depth = reiserfs_write_unlock_nested(s);
1534                                 __wait_on_buffer(cn->bh);
1535                                 reiserfs_write_lock_nested(s, depth);
1536
1537                                 if (!cn->bh) {
1538                                         reiserfs_panic(s, "journal-1012",
1539                                                        "cn->bh is NULL");
1540                                 }
1541                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1542 #ifdef CONFIG_REISERFS_CHECK
1543                                         reiserfs_warning(s, "journal-949",
1544                                                          "buffer write failed");
1545 #endif
1546                                         err = -EIO;
1547                                 }
1548                                 /*
1549                                  * note, we must clear the JDirty_wait bit
1550                                  * after the up to date check, otherwise we
1551                                  * race against our flushpage routine
1552                                  */
1553                                 BUG_ON(!test_clear_buffer_journal_dirty
1554                                        (cn->bh));
1555
1556                                 /* drop one ref for us */
1557                                 put_bh(cn->bh);
1558                                 /* drop one ref for journal_mark_dirty */
1559                                 release_buffer_page(cn->bh);
1560                         }
1561                         cn = cn->next;
1562                 }
1563         }
1564
1565         if (err)
1566                 reiserfs_abort(s, -EIO,
1567                                "Write error while pushing transaction to disk in %s",
1568                                __func__);
1569 flush_older_and_return:
1570
1571         /*
1572          * before we can update the journal header block, we _must_ flush all
1573          * real blocks from all older transactions to disk.  This is because
1574          * once the header block is updated, this transaction will not be
1575          * replayed after a crash
1576          */
1577         if (flushall) {
1578                 flush_older_journal_lists(s, jl);
1579         }
1580
1581         err = journal->j_errno;
1582         /*
1583          * before we can remove everything from the hash tables for this
1584          * transaction, we must make sure it can never be replayed
1585          *
1586          * since we are only called from do_journal_end, we know for sure there
1587          * are no allocations going on while we are flushing journal lists.  So,
1588          * we only need to update the journal header block for the last list
1589          * being flushed
1590          */
1591         if (!err && flushall) {
1592                 err =
1593                     update_journal_header_block(s,
1594                                                 (jl->j_start + jl->j_len +
1595                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1596                                                 jl->j_trans_id);
1597                 if (err)
1598                         reiserfs_abort(s, -EIO,
1599                                        "Write error while updating journal header in %s",
1600                                        __func__);
1601         }
1602         remove_all_from_journal_list(s, jl, 0);
1603         list_del_init(&jl->j_list);
1604         journal->j_num_lists--;
1605         del_from_work_list(s, jl);
1606
1607         if (journal->j_last_flush_id != 0 &&
1608             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1609                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1610                                  journal->j_last_flush_id, jl->j_trans_id);
1611         }
1612         journal->j_last_flush_id = jl->j_trans_id;
1613
1614         /*
1615          * not strictly required since we are freeing the list, but it should
1616          * help find code using dead lists later on
1617          */
1618         jl->j_len = 0;
1619         atomic_set(&jl->j_nonzerolen, 0);
1620         jl->j_start = 0;
1621         jl->j_realblock = NULL;
1622         jl->j_commit_bh = NULL;
1623         jl->j_trans_id = 0;
1624         jl->j_state = 0;
1625         put_journal_list(s, jl);
1626         if (flushall)
1627                 mutex_unlock(&journal->j_flush_mutex);
1628         return err;
1629 }
1630
1631 static int write_one_transaction(struct super_block *s,
1632                                  struct reiserfs_journal_list *jl,
1633                                  struct buffer_chunk *chunk)
1634 {
1635         struct reiserfs_journal_cnode *cn;
1636         int ret = 0;
1637
1638         jl->j_state |= LIST_TOUCHED;
1639         del_from_work_list(s, jl);
1640         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1641                 return 0;
1642         }
1643
1644         cn = jl->j_realblock;
1645         while (cn) {
1646                 /*
1647                  * if the blocknr == 0, this has been cleared from the hash,
1648                  * skip it
1649                  */
1650                 if (cn->blocknr == 0) {
1651                         goto next;
1652                 }
1653                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1654                         struct buffer_head *tmp_bh;
1655                         /*
1656                          * we can race against journal_mark_freed when we try
1657                          * to lock_buffer(cn->bh), so we have to inc the buffer
1658                          * count, and recheck things after locking
1659                          */
1660                         tmp_bh = cn->bh;
1661                         get_bh(tmp_bh);
1662                         lock_buffer(tmp_bh);
1663                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1664                                 if (!buffer_journal_dirty(tmp_bh) ||
1665                                     buffer_journal_prepared(tmp_bh))
1666                                         BUG();
1667                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1668                                 ret++;
1669                         } else {
1670                                 /* note, cn->bh might be null now */
1671                                 unlock_buffer(tmp_bh);
1672                         }
1673                         put_bh(tmp_bh);
1674                 }
1675 next:
1676                 cn = cn->next;
1677                 cond_resched();
1678         }
1679         return ret;
1680 }
1681
1682 /* used by flush_commit_list */
1683 static int dirty_one_transaction(struct super_block *s,
1684                                  struct reiserfs_journal_list *jl)
1685 {
1686         struct reiserfs_journal_cnode *cn;
1687         struct reiserfs_journal_list *pjl;
1688         int ret = 0;
1689
1690         jl->j_state |= LIST_DIRTY;
1691         cn = jl->j_realblock;
1692         while (cn) {
1693                 /*
1694                  * look for a more recent transaction that logged this
1695                  * buffer.  Only the most recent transaction with a buffer in
1696                  * it is allowed to send that buffer to disk
1697                  */
1698                 pjl = find_newer_jl_for_cn(cn);
1699                 if (!pjl && cn->blocknr && cn->bh
1700                     && buffer_journal_dirty(cn->bh)) {
1701                         BUG_ON(!can_dirty(cn));
1702                         /*
1703                          * if the buffer is prepared, it will either be logged
1704                          * or restored.  If restored, we need to make sure
1705                          * it actually gets marked dirty
1706                          */
1707                         clear_buffer_journal_new(cn->bh);
1708                         if (buffer_journal_prepared(cn->bh)) {
1709                                 set_buffer_journal_restore_dirty(cn->bh);
1710                         } else {
1711                                 set_buffer_journal_test(cn->bh);
1712                                 mark_buffer_dirty(cn->bh);
1713                         }
1714                 }
1715                 cn = cn->next;
1716         }
1717         return ret;
1718 }
1719
1720 static int kupdate_transactions(struct super_block *s,
1721                                 struct reiserfs_journal_list *jl,
1722                                 struct reiserfs_journal_list **next_jl,
1723                                 unsigned int *next_trans_id,
1724                                 int num_blocks, int num_trans)
1725 {
1726         int ret = 0;
1727         int written = 0;
1728         int transactions_flushed = 0;
1729         unsigned int orig_trans_id = jl->j_trans_id;
1730         struct buffer_chunk chunk;
1731         struct list_head *entry;
1732         struct reiserfs_journal *journal = SB_JOURNAL(s);
1733         chunk.nr = 0;
1734
1735         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1736         if (!journal_list_still_alive(s, orig_trans_id)) {
1737                 goto done;
1738         }
1739
1740         /*
1741          * we've got j_flush_mutex held, nobody is going to delete any
1742          * of these lists out from underneath us
1743          */
1744         while ((num_trans && transactions_flushed < num_trans) ||
1745                (!num_trans && written < num_blocks)) {
1746
1747                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1748                     atomic_read(&jl->j_commit_left)
1749                     || !(jl->j_state & LIST_DIRTY)) {
1750                         del_from_work_list(s, jl);
1751                         break;
1752                 }
1753                 ret = write_one_transaction(s, jl, &chunk);
1754
1755                 if (ret < 0)
1756                         goto done;
1757                 transactions_flushed++;
1758                 written += ret;
1759                 entry = jl->j_list.next;
1760
1761                 /* did we wrap? */
1762                 if (entry == &journal->j_journal_list) {
1763                         break;
1764                 }
1765                 jl = JOURNAL_LIST_ENTRY(entry);
1766
1767                 /* don't bother with older transactions */
1768                 if (jl->j_trans_id <= orig_trans_id)
1769                         break;
1770         }
1771         if (chunk.nr) {
1772                 write_chunk(&chunk);
1773         }
1774
1775 done:
1776         mutex_unlock(&journal->j_flush_mutex);
1777         return ret;
1778 }
1779
1780 /*
1781  * for o_sync and fsync heavy applications, they tend to use
1782  * all the journa list slots with tiny transactions.  These
1783  * trigger lots and lots of calls to update the header block, which
1784  * adds seeks and slows things down.
1785  *
1786  * This function tries to clear out a large chunk of the journal lists
1787  * at once, which makes everything faster since only the newest journal
1788  * list updates the header block
1789  */
1790 static int flush_used_journal_lists(struct super_block *s,
1791                                     struct reiserfs_journal_list *jl)
1792 {
1793         unsigned long len = 0;
1794         unsigned long cur_len;
1795         int ret;
1796         int i;
1797         int limit = 256;
1798         struct reiserfs_journal_list *tjl;
1799         struct reiserfs_journal_list *flush_jl;
1800         unsigned int trans_id;
1801         struct reiserfs_journal *journal = SB_JOURNAL(s);
1802
1803         flush_jl = tjl = jl;
1804
1805         /* in data logging mode, try harder to flush a lot of blocks */
1806         if (reiserfs_data_log(s))
1807                 limit = 1024;
1808         /* flush for 256 transactions or limit blocks, whichever comes first */
1809         for (i = 0; i < 256 && len < limit; i++) {
1810                 if (atomic_read(&tjl->j_commit_left) ||
1811                     tjl->j_trans_id < jl->j_trans_id) {
1812                         break;
1813                 }
1814                 cur_len = atomic_read(&tjl->j_nonzerolen);
1815                 if (cur_len > 0) {
1816                         tjl->j_state &= ~LIST_TOUCHED;
1817                 }
1818                 len += cur_len;
1819                 flush_jl = tjl;
1820                 if (tjl->j_list.next == &journal->j_journal_list)
1821                         break;
1822                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1823         }
1824         get_journal_list(jl);
1825         get_journal_list(flush_jl);
1826
1827         /*
1828          * try to find a group of blocks we can flush across all the
1829          * transactions, but only bother if we've actually spanned
1830          * across multiple lists
1831          */
1832         if (flush_jl != jl) {
1833                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1834         }
1835         flush_journal_list(s, flush_jl, 1);
1836         put_journal_list(s, flush_jl);
1837         put_journal_list(s, jl);
1838         return 0;
1839 }
1840
1841 /*
1842  * removes any nodes in table with name block and dev as bh.
1843  * only touchs the hnext and hprev pointers.
1844  */
1845 void remove_journal_hash(struct super_block *sb,
1846                          struct reiserfs_journal_cnode **table,
1847                          struct reiserfs_journal_list *jl,
1848                          unsigned long block, int remove_freed)
1849 {
1850         struct reiserfs_journal_cnode *cur;
1851         struct reiserfs_journal_cnode **head;
1852
1853         head = &(journal_hash(table, sb, block));
1854         if (!head) {
1855                 return;
1856         }
1857         cur = *head;
1858         while (cur) {
1859                 if (cur->blocknr == block && cur->sb == sb
1860                     && (jl == NULL || jl == cur->jlist)
1861                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1862                         if (cur->hnext) {
1863                                 cur->hnext->hprev = cur->hprev;
1864                         }
1865                         if (cur->hprev) {
1866                                 cur->hprev->hnext = cur->hnext;
1867                         } else {
1868                                 *head = cur->hnext;
1869                         }
1870                         cur->blocknr = 0;
1871                         cur->sb = NULL;
1872                         cur->state = 0;
1873                         /*
1874                          * anybody who clears the cur->bh will also
1875                          * dec the nonzerolen
1876                          */
1877                         if (cur->bh && cur->jlist)
1878                                 atomic_dec(&cur->jlist->j_nonzerolen);
1879                         cur->bh = NULL;
1880                         cur->jlist = NULL;
1881                 }
1882                 cur = cur->hnext;
1883         }
1884 }
1885
1886 static void free_journal_ram(struct super_block *sb)
1887 {
1888         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1889         kfree(journal->j_current_jl);
1890         journal->j_num_lists--;
1891
1892         vfree(journal->j_cnode_free_orig);
1893         free_list_bitmaps(sb, journal->j_list_bitmap);
1894         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1895         if (journal->j_header_bh) {
1896                 brelse(journal->j_header_bh);
1897         }
1898         /*
1899          * j_header_bh is on the journal dev, make sure
1900          * not to release the journal dev until we brelse j_header_bh
1901          */
1902         release_journal_dev(sb, journal);
1903         vfree(journal);
1904 }
1905
1906 /*
1907  * call on unmount.  Only set error to 1 if you haven't made your way out
1908  * of read_super() yet.  Any other caller must keep error at 0.
1909  */
1910 static int do_journal_release(struct reiserfs_transaction_handle *th,
1911                               struct super_block *sb, int error)
1912 {
1913         struct reiserfs_transaction_handle myth;
1914         int flushed = 0;
1915         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1916
1917         /*
1918          * we only want to flush out transactions if we were
1919          * called with error == 0
1920          */
1921         if (!error && !(sb->s_flags & MS_RDONLY)) {
1922                 /* end the current trans */
1923                 BUG_ON(!th->t_trans_id);
1924                 do_journal_end(th, FLUSH_ALL);
1925
1926                 /*
1927                  * make sure something gets logged to force
1928                  * our way into the flush code
1929                  */
1930                 if (!journal_join(&myth, sb)) {
1931                         reiserfs_prepare_for_journal(sb,
1932                                                      SB_BUFFER_WITH_SB(sb),
1933                                                      1);
1934                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1935                         do_journal_end(&myth, FLUSH_ALL);
1936                         flushed = 1;
1937                 }
1938         }
1939
1940         /* this also catches errors during the do_journal_end above */
1941         if (!error && reiserfs_is_journal_aborted(journal)) {
1942                 memset(&myth, 0, sizeof(myth));
1943                 if (!journal_join_abort(&myth, sb)) {
1944                         reiserfs_prepare_for_journal(sb,
1945                                                      SB_BUFFER_WITH_SB(sb),
1946                                                      1);
1947                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1948                         do_journal_end(&myth, FLUSH_ALL);
1949                 }
1950         }
1951
1952
1953         /*
1954          * We must release the write lock here because
1955          * the workqueue job (flush_async_commit) needs this lock
1956          */
1957         reiserfs_write_unlock(sb);
1958
1959         /*
1960          * Cancel flushing of old commits. Note that neither of these works
1961          * will be requeued because superblock is being shutdown and doesn't
1962          * have MS_ACTIVE set.
1963          */
1964         cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
1965         /* wait for all commits to finish */
1966         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1967
1968         free_journal_ram(sb);
1969
1970         reiserfs_write_lock(sb);
1971
1972         return 0;
1973 }
1974
1975 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1976 int journal_release(struct reiserfs_transaction_handle *th,
1977                     struct super_block *sb)
1978 {
1979         return do_journal_release(th, sb, 0);
1980 }
1981
1982 /* only call from an error condition inside reiserfs_read_super!  */
1983 int journal_release_error(struct reiserfs_transaction_handle *th,
1984                           struct super_block *sb)
1985 {
1986         return do_journal_release(th, sb, 1);
1987 }
1988
1989 /*
1990  * compares description block with commit block.
1991  * returns 1 if they differ, 0 if they are the same
1992  */
1993 static int journal_compare_desc_commit(struct super_block *sb,
1994                                        struct reiserfs_journal_desc *desc,
1995                                        struct reiserfs_journal_commit *commit)
1996 {
1997         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1998             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1999             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2000             get_commit_trans_len(commit) <= 0) {
2001                 return 1;
2002         }
2003         return 0;
2004 }
2005
2006 /*
2007  * returns 0 if it did not find a description block
2008  * returns -1 if it found a corrupt commit block
2009  * returns 1 if both desc and commit were valid
2010  * NOTE: only called during fs mount
2011  */
2012 static int journal_transaction_is_valid(struct super_block *sb,
2013                                         struct buffer_head *d_bh,
2014                                         unsigned int *oldest_invalid_trans_id,
2015                                         unsigned long *newest_mount_id)
2016 {
2017         struct reiserfs_journal_desc *desc;
2018         struct reiserfs_journal_commit *commit;
2019         struct buffer_head *c_bh;
2020         unsigned long offset;
2021
2022         if (!d_bh)
2023                 return 0;
2024
2025         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2026         if (get_desc_trans_len(desc) > 0
2027             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2028                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2029                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2030                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2031                                        "journal-986: transaction "
2032                                        "is valid returning because trans_id %d is greater than "
2033                                        "oldest_invalid %lu",
2034                                        get_desc_trans_id(desc),
2035                                        *oldest_invalid_trans_id);
2036                         return 0;
2037                 }
2038                 if (newest_mount_id
2039                     && *newest_mount_id > get_desc_mount_id(desc)) {
2040                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2041                                        "journal-1087: transaction "
2042                                        "is valid returning because mount_id %d is less than "
2043                                        "newest_mount_id %lu",
2044                                        get_desc_mount_id(desc),
2045                                        *newest_mount_id);
2046                         return -1;
2047                 }
2048                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2049                         reiserfs_warning(sb, "journal-2018",
2050                                          "Bad transaction length %d "
2051                                          "encountered, ignoring transaction",
2052                                          get_desc_trans_len(desc));
2053                         return -1;
2054                 }
2055                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2056
2057                 /*
2058                  * ok, we have a journal description block,
2059                  * let's see if the transaction was valid
2060                  */
2061                 c_bh =
2062                     journal_bread(sb,
2063                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2064                                   ((offset + get_desc_trans_len(desc) +
2065                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2066                 if (!c_bh)
2067                         return 0;
2068                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2069                 if (journal_compare_desc_commit(sb, desc, commit)) {
2070                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2071                                        "journal_transaction_is_valid, commit offset %ld had bad "
2072                                        "time %d or length %d",
2073                                        c_bh->b_blocknr -
2074                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2075                                        get_commit_trans_id(commit),
2076                                        get_commit_trans_len(commit));
2077                         brelse(c_bh);
2078                         if (oldest_invalid_trans_id) {
2079                                 *oldest_invalid_trans_id =
2080                                     get_desc_trans_id(desc);
2081                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2082                                                "journal-1004: "
2083                                                "transaction_is_valid setting oldest invalid trans_id "
2084                                                "to %d",
2085                                                get_desc_trans_id(desc));
2086                         }
2087                         return -1;
2088                 }
2089                 brelse(c_bh);
2090                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2091                                "journal-1006: found valid "
2092                                "transaction start offset %llu, len %d id %d",
2093                                d_bh->b_blocknr -
2094                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2095                                get_desc_trans_len(desc),
2096                                get_desc_trans_id(desc));
2097                 return 1;
2098         } else {
2099                 return 0;
2100         }
2101 }
2102
2103 static void brelse_array(struct buffer_head **heads, int num)
2104 {
2105         int i;
2106         for (i = 0; i < num; i++) {
2107                 brelse(heads[i]);
2108         }
2109 }
2110
2111 /*
2112  * given the start, and values for the oldest acceptable transactions,
2113  * this either reads in a replays a transaction, or returns because the
2114  * transaction is invalid, or too old.
2115  * NOTE: only called during fs mount
2116  */
2117 static int journal_read_transaction(struct super_block *sb,
2118                                     unsigned long cur_dblock,
2119                                     unsigned long oldest_start,
2120                                     unsigned int oldest_trans_id,
2121                                     unsigned long newest_mount_id)
2122 {
2123         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2124         struct reiserfs_journal_desc *desc;
2125         struct reiserfs_journal_commit *commit;
2126         unsigned int trans_id = 0;
2127         struct buffer_head *c_bh;
2128         struct buffer_head *d_bh;
2129         struct buffer_head **log_blocks = NULL;
2130         struct buffer_head **real_blocks = NULL;
2131         unsigned int trans_offset;
2132         int i;
2133         int trans_half;
2134
2135         d_bh = journal_bread(sb, cur_dblock);
2136         if (!d_bh)
2137                 return 1;
2138         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2139         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2140         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2141                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2142                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2143                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2144         if (get_desc_trans_id(desc) < oldest_trans_id) {
2145                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2146                                "journal_read_trans skipping because %lu is too old",
2147                                cur_dblock -
2148                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2149                 brelse(d_bh);
2150                 return 1;
2151         }
2152         if (get_desc_mount_id(desc) != newest_mount_id) {
2153                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2154                                "journal_read_trans skipping because %d is != "
2155                                "newest_mount_id %lu", get_desc_mount_id(desc),
2156                                newest_mount_id);
2157                 brelse(d_bh);
2158                 return 1;
2159         }
2160         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2161                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2162                               SB_ONDISK_JOURNAL_SIZE(sb)));
2163         if (!c_bh) {
2164                 brelse(d_bh);
2165                 return 1;
2166         }
2167         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2168         if (journal_compare_desc_commit(sb, desc, commit)) {
2169                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2170                                "journal_read_transaction, "
2171                                "commit offset %llu had bad time %d or length %d",
2172                                c_bh->b_blocknr -
2173                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2174                                get_commit_trans_id(commit),
2175                                get_commit_trans_len(commit));
2176                 brelse(c_bh);
2177                 brelse(d_bh);
2178                 return 1;
2179         }
2180
2181         if (bdev_read_only(sb->s_bdev)) {
2182                 reiserfs_warning(sb, "clm-2076",
2183                                  "device is readonly, unable to replay log");
2184                 brelse(c_bh);
2185                 brelse(d_bh);
2186                 return -EROFS;
2187         }
2188
2189         trans_id = get_desc_trans_id(desc);
2190         /*
2191          * now we know we've got a good transaction, and it was
2192          * inside the valid time ranges
2193          */
2194         log_blocks = kmalloc(get_desc_trans_len(desc) *
2195                              sizeof(struct buffer_head *), GFP_NOFS);
2196         real_blocks = kmalloc(get_desc_trans_len(desc) *
2197                               sizeof(struct buffer_head *), GFP_NOFS);
2198         if (!log_blocks || !real_blocks) {
2199                 brelse(c_bh);
2200                 brelse(d_bh);
2201                 kfree(log_blocks);
2202                 kfree(real_blocks);
2203                 reiserfs_warning(sb, "journal-1169",
2204                                  "kmalloc failed, unable to mount FS");
2205                 return -1;
2206         }
2207         /* get all the buffer heads */
2208         trans_half = journal_trans_half(sb->s_blocksize);
2209         for (i = 0; i < get_desc_trans_len(desc); i++) {
2210                 log_blocks[i] =
2211                     journal_getblk(sb,
2212                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2213                                    (trans_offset + 1 +
2214                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2215                 if (i < trans_half) {
2216                         real_blocks[i] =
2217                             sb_getblk(sb,
2218                                       le32_to_cpu(desc->j_realblock[i]));
2219                 } else {
2220                         real_blocks[i] =
2221                             sb_getblk(sb,
2222                                       le32_to_cpu(commit->
2223                                                   j_realblock[i - trans_half]));
2224                 }
2225                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2226                         reiserfs_warning(sb, "journal-1207",
2227                                          "REPLAY FAILURE fsck required! "
2228                                          "Block to replay is outside of "
2229                                          "filesystem");
2230                         goto abort_replay;
2231                 }
2232                 /* make sure we don't try to replay onto log or reserved area */
2233                 if (is_block_in_log_or_reserved_area
2234                     (sb, real_blocks[i]->b_blocknr)) {
2235                         reiserfs_warning(sb, "journal-1204",
2236                                          "REPLAY FAILURE fsck required! "
2237                                          "Trying to replay onto a log block");
2238 abort_replay:
2239                         brelse_array(log_blocks, i);
2240                         brelse_array(real_blocks, i);
2241                         brelse(c_bh);
2242                         brelse(d_bh);
2243                         kfree(log_blocks);
2244                         kfree(real_blocks);
2245                         return -1;
2246                 }
2247         }
2248         /* read in the log blocks, memcpy to the corresponding real block */
2249         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2250         for (i = 0; i < get_desc_trans_len(desc); i++) {
2251
2252                 wait_on_buffer(log_blocks[i]);
2253                 if (!buffer_uptodate(log_blocks[i])) {
2254                         reiserfs_warning(sb, "journal-1212",
2255                                          "REPLAY FAILURE fsck required! "
2256                                          "buffer write failed");
2257                         brelse_array(log_blocks + i,
2258                                      get_desc_trans_len(desc) - i);
2259                         brelse_array(real_blocks, get_desc_trans_len(desc));
2260                         brelse(c_bh);
2261                         brelse(d_bh);
2262                         kfree(log_blocks);
2263                         kfree(real_blocks);
2264                         return -1;
2265                 }
2266                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2267                        real_blocks[i]->b_size);
2268                 set_buffer_uptodate(real_blocks[i]);
2269                 brelse(log_blocks[i]);
2270         }
2271         /* flush out the real blocks */
2272         for (i = 0; i < get_desc_trans_len(desc); i++) {
2273                 set_buffer_dirty(real_blocks[i]);
2274                 write_dirty_buffer(real_blocks[i], WRITE);
2275         }
2276         for (i = 0; i < get_desc_trans_len(desc); i++) {
2277                 wait_on_buffer(real_blocks[i]);
2278                 if (!buffer_uptodate(real_blocks[i])) {
2279                         reiserfs_warning(sb, "journal-1226",
2280                                          "REPLAY FAILURE, fsck required! "
2281                                          "buffer write failed");
2282                         brelse_array(real_blocks + i,
2283                                      get_desc_trans_len(desc) - i);
2284                         brelse(c_bh);
2285                         brelse(d_bh);
2286                         kfree(log_blocks);
2287                         kfree(real_blocks);
2288                         return -1;
2289                 }
2290                 brelse(real_blocks[i]);
2291         }
2292         cur_dblock =
2293             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2294             ((trans_offset + get_desc_trans_len(desc) +
2295               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2296         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2297                        "journal-1095: setting journal " "start to offset %ld",
2298                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2299
2300         /*
2301          * init starting values for the first transaction, in case
2302          * this is the last transaction to be replayed.
2303          */
2304         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2305         journal->j_last_flush_trans_id = trans_id;
2306         journal->j_trans_id = trans_id + 1;
2307         /* check for trans_id overflow */
2308         if (journal->j_trans_id == 0)
2309                 journal->j_trans_id = 10;
2310         brelse(c_bh);
2311         brelse(d_bh);
2312         kfree(log_blocks);
2313         kfree(real_blocks);
2314         return 0;
2315 }
2316
2317 /*
2318  * This function reads blocks starting from block and to max_block of bufsize
2319  * size (but no more than BUFNR blocks at a time). This proved to improve
2320  * mounting speed on self-rebuilding raid5 arrays at least.
2321  * Right now it is only used from journal code. But later we might use it
2322  * from other places.
2323  * Note: Do not use journal_getblk/sb_getblk functions here!
2324  */
2325 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2326                                            b_blocknr_t block, int bufsize,
2327                                            b_blocknr_t max_block)
2328 {
2329         struct buffer_head *bhlist[BUFNR];
2330         unsigned int blocks = BUFNR;
2331         struct buffer_head *bh;
2332         int i, j;
2333
2334         bh = __getblk(dev, block, bufsize);
2335         if (buffer_uptodate(bh))
2336                 return (bh);
2337
2338         if (block + BUFNR > max_block) {
2339                 blocks = max_block - block;
2340         }
2341         bhlist[0] = bh;
2342         j = 1;
2343         for (i = 1; i < blocks; i++) {
2344                 bh = __getblk(dev, block + i, bufsize);
2345                 if (buffer_uptodate(bh)) {
2346                         brelse(bh);
2347                         break;
2348                 } else
2349                         bhlist[j++] = bh;
2350         }
2351         ll_rw_block(READ, j, bhlist);
2352         for (i = 1; i < j; i++)
2353                 brelse(bhlist[i]);
2354         bh = bhlist[0];
2355         wait_on_buffer(bh);
2356         if (buffer_uptodate(bh))
2357                 return bh;
2358         brelse(bh);
2359         return NULL;
2360 }
2361
2362 /*
2363  * read and replay the log
2364  * on a clean unmount, the journal header's next unflushed pointer will be
2365  * to an invalid transaction.  This tests that before finding all the
2366  * transactions in the log, which makes normal mount times fast.
2367  *
2368  * After a crash, this starts with the next unflushed transaction, and
2369  * replays until it finds one too old, or invalid.
2370  *
2371  * On exit, it sets things up so the first transaction will work correctly.
2372  * NOTE: only called during fs mount
2373  */
2374 static int journal_read(struct super_block *sb)
2375 {
2376         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2377         struct reiserfs_journal_desc *desc;
2378         unsigned int oldest_trans_id = 0;
2379         unsigned int oldest_invalid_trans_id = 0;
2380         time_t start;
2381         unsigned long oldest_start = 0;
2382         unsigned long cur_dblock = 0;
2383         unsigned long newest_mount_id = 9;
2384         struct buffer_head *d_bh;
2385         struct reiserfs_journal_header *jh;
2386         int valid_journal_header = 0;
2387         int replay_count = 0;
2388         int continue_replay = 1;
2389         int ret;
2390         char b[BDEVNAME_SIZE];
2391
2392         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2393         reiserfs_info(sb, "checking transaction log (%s)\n",
2394                       bdevname(journal->j_dev_bd, b));
2395         start = get_seconds();
2396
2397         /*
2398          * step 1, read in the journal header block.  Check the transaction
2399          * it says is the first unflushed, and if that transaction is not
2400          * valid, replay is done
2401          */
2402         journal->j_header_bh = journal_bread(sb,
2403                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2404                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2405         if (!journal->j_header_bh) {
2406                 return 1;
2407         }
2408         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2409         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2410             SB_ONDISK_JOURNAL_SIZE(sb)
2411             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2412                 oldest_start =
2413                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2414                     le32_to_cpu(jh->j_first_unflushed_offset);
2415                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2416                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2417                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2418                                "journal-1153: found in "
2419                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2420                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2421                                le32_to_cpu(jh->j_last_flush_trans_id));
2422                 valid_journal_header = 1;
2423
2424                 /*
2425                  * now, we try to read the first unflushed offset.  If it
2426                  * is not valid, there is nothing more we can do, and it
2427                  * makes no sense to read through the whole log.
2428                  */
2429                 d_bh =
2430                     journal_bread(sb,
2431                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2432                                   le32_to_cpu(jh->j_first_unflushed_offset));
2433                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2434                 if (!ret) {
2435                         continue_replay = 0;
2436                 }
2437                 brelse(d_bh);
2438                 goto start_log_replay;
2439         }
2440
2441         /*
2442          * ok, there are transactions that need to be replayed.  start
2443          * with the first log block, find all the valid transactions, and
2444          * pick out the oldest.
2445          */
2446         while (continue_replay
2447                && cur_dblock <
2448                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2449                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2450                 /*
2451                  * Note that it is required for blocksize of primary fs
2452                  * device and journal device to be the same
2453                  */
2454                 d_bh =
2455                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2456                                     sb->s_blocksize,
2457                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2458                                     SB_ONDISK_JOURNAL_SIZE(sb));
2459                 ret =
2460                     journal_transaction_is_valid(sb, d_bh,
2461                                                  &oldest_invalid_trans_id,
2462                                                  &newest_mount_id);
2463                 if (ret == 1) {
2464                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2465                         if (oldest_start == 0) {        /* init all oldest_ values */
2466                                 oldest_trans_id = get_desc_trans_id(desc);
2467                                 oldest_start = d_bh->b_blocknr;
2468                                 newest_mount_id = get_desc_mount_id(desc);
2469                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2470                                                "journal-1179: Setting "
2471                                                "oldest_start to offset %llu, trans_id %lu",
2472                                                oldest_start -
2473                                                SB_ONDISK_JOURNAL_1st_BLOCK
2474                                                (sb), oldest_trans_id);
2475                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2476                                 /* one we just read was older */
2477                                 oldest_trans_id = get_desc_trans_id(desc);
2478                                 oldest_start = d_bh->b_blocknr;
2479                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2480                                                "journal-1180: Resetting "
2481                                                "oldest_start to offset %lu, trans_id %lu",
2482                                                oldest_start -
2483                                                SB_ONDISK_JOURNAL_1st_BLOCK
2484                                                (sb), oldest_trans_id);
2485                         }
2486                         if (newest_mount_id < get_desc_mount_id(desc)) {
2487                                 newest_mount_id = get_desc_mount_id(desc);
2488                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2489                                                "journal-1299: Setting "
2490                                                "newest_mount_id to %d",
2491                                                get_desc_mount_id(desc));
2492                         }
2493                         cur_dblock += get_desc_trans_len(desc) + 2;
2494                 } else {
2495                         cur_dblock++;
2496                 }
2497                 brelse(d_bh);
2498         }
2499
2500 start_log_replay:
2501         cur_dblock = oldest_start;
2502         if (oldest_trans_id) {
2503                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2504                                "journal-1206: Starting replay "
2505                                "from offset %llu, trans_id %lu",
2506                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2507                                oldest_trans_id);
2508
2509         }
2510         replay_count = 0;
2511         while (continue_replay && oldest_trans_id > 0) {
2512                 ret =
2513                     journal_read_transaction(sb, cur_dblock, oldest_start,
2514                                              oldest_trans_id, newest_mount_id);
2515                 if (ret < 0) {
2516                         return ret;
2517                 } else if (ret != 0) {
2518                         break;
2519                 }
2520                 cur_dblock =
2521                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2522                 replay_count++;
2523                 if (cur_dblock == oldest_start)
2524                         break;
2525         }
2526
2527         if (oldest_trans_id == 0) {
2528                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2529                                "journal-1225: No valid " "transactions found");
2530         }
2531         /*
2532          * j_start does not get set correctly if we don't replay any
2533          * transactions.  if we had a valid journal_header, set j_start
2534          * to the first unflushed transaction value, copy the trans_id
2535          * from the header
2536          */
2537         if (valid_journal_header && replay_count == 0) {
2538                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2539                 journal->j_trans_id =
2540                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2541                 /* check for trans_id overflow */
2542                 if (journal->j_trans_id == 0)
2543                         journal->j_trans_id = 10;
2544                 journal->j_last_flush_trans_id =
2545                     le32_to_cpu(jh->j_last_flush_trans_id);
2546                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2547         } else {
2548                 journal->j_mount_id = newest_mount_id + 1;
2549         }
2550         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2551                        "newest_mount_id to %lu", journal->j_mount_id);
2552         journal->j_first_unflushed_offset = journal->j_start;
2553         if (replay_count > 0) {
2554                 reiserfs_info(sb,
2555                               "replayed %d transactions in %lu seconds\n",
2556                               replay_count, get_seconds() - start);
2557         }
2558         /* needed to satisfy the locking in _update_journal_header_block */
2559         reiserfs_write_lock(sb);
2560         if (!bdev_read_only(sb->s_bdev) &&
2561             _update_journal_header_block(sb, journal->j_start,
2562                                          journal->j_last_flush_trans_id)) {
2563                 reiserfs_write_unlock(sb);
2564                 /*
2565                  * replay failed, caller must call free_journal_ram and abort
2566                  * the mount
2567                  */
2568                 return -1;
2569         }
2570         reiserfs_write_unlock(sb);
2571         return 0;
2572 }
2573
2574 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2575 {
2576         struct reiserfs_journal_list *jl;
2577         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2578                      GFP_NOFS | __GFP_NOFAIL);
2579         INIT_LIST_HEAD(&jl->j_list);
2580         INIT_LIST_HEAD(&jl->j_working_list);
2581         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2582         INIT_LIST_HEAD(&jl->j_bh_list);
2583         mutex_init(&jl->j_commit_mutex);
2584         SB_JOURNAL(s)->j_num_lists++;
2585         get_journal_list(jl);
2586         return jl;
2587 }
2588
2589 static void journal_list_init(struct super_block *sb)
2590 {
2591         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2592 }
2593
2594 static void release_journal_dev(struct super_block *super,
2595                                struct reiserfs_journal *journal)
2596 {
2597         if (journal->j_dev_bd != NULL) {
2598                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2599                 journal->j_dev_bd = NULL;
2600         }
2601 }
2602
2603 static int journal_init_dev(struct super_block *super,
2604                             struct reiserfs_journal *journal,
2605                             const char *jdev_name)
2606 {
2607         int result;
2608         dev_t jdev;
2609         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2610         char b[BDEVNAME_SIZE];
2611
2612         result = 0;
2613
2614         journal->j_dev_bd = NULL;
2615         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2616             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2617
2618         if (bdev_read_only(super->s_bdev))
2619                 blkdev_mode = FMODE_READ;
2620
2621         /* there is no "jdev" option and journal is on separate device */
2622         if ((!jdev_name || !jdev_name[0])) {
2623                 if (jdev == super->s_dev)
2624                         blkdev_mode &= ~FMODE_EXCL;
2625                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2626                                                       journal);
2627                 journal->j_dev_mode = blkdev_mode;
2628                 if (IS_ERR(journal->j_dev_bd)) {
2629                         result = PTR_ERR(journal->j_dev_bd);
2630                         journal->j_dev_bd = NULL;
2631                         reiserfs_warning(super, "sh-458",
2632                                          "cannot init journal device '%s': %i",
2633                                          __bdevname(jdev, b), result);
2634                         return result;
2635                 } else if (jdev != super->s_dev)
2636                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2637
2638                 return 0;
2639         }
2640
2641         journal->j_dev_mode = blkdev_mode;
2642         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2643         if (IS_ERR(journal->j_dev_bd)) {
2644                 result = PTR_ERR(journal->j_dev_bd);
2645                 journal->j_dev_bd = NULL;
2646                 reiserfs_warning(super,
2647                                  "journal_init_dev: Cannot open '%s': %i",
2648                                  jdev_name, result);
2649                 return result;
2650         }
2651
2652         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2653         reiserfs_info(super,
2654                       "journal_init_dev: journal device: %s\n",
2655                       bdevname(journal->j_dev_bd, b));
2656         return 0;
2657 }
2658
2659 /*
2660  * When creating/tuning a file system user can assign some
2661  * journal params within boundaries which depend on the ratio
2662  * blocksize/standard_blocksize.
2663  *
2664  * For blocks >= standard_blocksize transaction size should
2665  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2666  * then JOURNAL_TRANS_MAX_DEFAULT.
2667  *
2668  * For blocks < standard_blocksize these boundaries should be
2669  * decreased proportionally.
2670  */
2671 #define REISERFS_STANDARD_BLKSIZE (4096)
2672
2673 static int check_advise_trans_params(struct super_block *sb,
2674                                      struct reiserfs_journal *journal)
2675 {
2676         if (journal->j_trans_max) {
2677                 /* Non-default journal params.  Do sanity check for them. */
2678                 int ratio = 1;
2679                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2680                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2681
2682                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2683                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2684                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2685                     JOURNAL_MIN_RATIO) {
2686                         reiserfs_warning(sb, "sh-462",
2687                                          "bad transaction max size (%u). "
2688                                          "FSCK?", journal->j_trans_max);
2689                         return 1;
2690                 }
2691                 if (journal->j_max_batch != (journal->j_trans_max) *
2692                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2693                         reiserfs_warning(sb, "sh-463",
2694                                          "bad transaction max batch (%u). "
2695                                          "FSCK?", journal->j_max_batch);
2696                         return 1;
2697                 }
2698         } else {
2699                 /*
2700                  * Default journal params.
2701                  * The file system was created by old version
2702                  * of mkreiserfs, so some fields contain zeros,
2703                  * and we need to advise proper values for them
2704                  */
2705                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2706                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2707                                          sb->s_blocksize);
2708                         return 1;
2709                 }
2710                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2711                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2712                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2713         }
2714         return 0;
2715 }
2716
2717 /* must be called once on fs mount.  calls journal_read for you */
2718 int journal_init(struct super_block *sb, const char *j_dev_name,
2719                  int old_format, unsigned int commit_max_age)
2720 {
2721         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2722         struct buffer_head *bhjh;
2723         struct reiserfs_super_block *rs;
2724         struct reiserfs_journal_header *jh;
2725         struct reiserfs_journal *journal;
2726         struct reiserfs_journal_list *jl;
2727         char b[BDEVNAME_SIZE];
2728         int ret;
2729
2730         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2731         if (!journal) {
2732                 reiserfs_warning(sb, "journal-1256",
2733                                  "unable to get memory for journal structure");
2734                 return 1;
2735         }
2736         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2737         INIT_LIST_HEAD(&journal->j_prealloc_list);
2738         INIT_LIST_HEAD(&journal->j_working_list);
2739         INIT_LIST_HEAD(&journal->j_journal_list);
2740         journal->j_persistent_trans = 0;
2741         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2742                                            reiserfs_bmap_count(sb)))
2743                 goto free_and_return;
2744
2745         allocate_bitmap_nodes(sb);
2746
2747         /* reserved for journal area support */
2748         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2749                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2750                                                  / sb->s_blocksize +
2751                                                  reiserfs_bmap_count(sb) +
2752                                                  1 :
2753                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2754                                                  sb->s_blocksize + 2);
2755
2756         /*
2757          * Sanity check to see is the standard journal fitting
2758          * within first bitmap (actual for small blocksizes)
2759          */
2760         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2761             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2762              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2763                 reiserfs_warning(sb, "journal-1393",
2764                                  "journal does not fit for area addressed "
2765                                  "by first of bitmap blocks. It starts at "
2766                                  "%u and its size is %u. Block size %ld",
2767                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2768                                  SB_ONDISK_JOURNAL_SIZE(sb),
2769                                  sb->s_blocksize);
2770                 goto free_and_return;
2771         }
2772
2773         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2774                 reiserfs_warning(sb, "sh-462",
2775                                  "unable to initialize journal device");
2776                 goto free_and_return;
2777         }
2778
2779         rs = SB_DISK_SUPER_BLOCK(sb);
2780
2781         /* read journal header */
2782         bhjh = journal_bread(sb,
2783                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2784                              SB_ONDISK_JOURNAL_SIZE(sb));
2785         if (!bhjh) {
2786                 reiserfs_warning(sb, "sh-459",
2787                                  "unable to read journal header");
2788                 goto free_and_return;
2789         }
2790         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2791
2792         /* make sure that journal matches to the super block */
2793         if (is_reiserfs_jr(rs)
2794             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2795                 sb_jp_journal_magic(rs))) {
2796                 reiserfs_warning(sb, "sh-460",
2797                                  "journal header magic %x (device %s) does "
2798                                  "not match to magic found in super block %x",
2799                                  jh->jh_journal.jp_journal_magic,
2800                                  bdevname(journal->j_dev_bd, b),
2801                                  sb_jp_journal_magic(rs));
2802                 brelse(bhjh);
2803                 goto free_and_return;
2804         }
2805
2806         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2807         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2808         journal->j_max_commit_age =
2809             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2810         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2811
2812         if (check_advise_trans_params(sb, journal) != 0)
2813                 goto free_and_return;
2814         journal->j_default_max_commit_age = journal->j_max_commit_age;
2815
2816         if (commit_max_age != 0) {
2817                 journal->j_max_commit_age = commit_max_age;
2818                 journal->j_max_trans_age = commit_max_age;
2819         }
2820
2821         reiserfs_info(sb, "journal params: device %s, size %u, "
2822                       "journal first block %u, max trans len %u, max batch %u, "
2823                       "max commit age %u, max trans age %u\n",
2824                       bdevname(journal->j_dev_bd, b),
2825                       SB_ONDISK_JOURNAL_SIZE(sb),
2826                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2827                       journal->j_trans_max,
2828                       journal->j_max_batch,
2829                       journal->j_max_commit_age, journal->j_max_trans_age);
2830
2831         brelse(bhjh);
2832
2833         journal->j_list_bitmap_index = 0;
2834         journal_list_init(sb);
2835
2836         memset(journal->j_list_hash_table, 0,
2837                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2838
2839         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2840         spin_lock_init(&journal->j_dirty_buffers_lock);
2841
2842         journal->j_start = 0;
2843         journal->j_len = 0;
2844         journal->j_len_alloc = 0;
2845         atomic_set(&journal->j_wcount, 0);
2846         atomic_set(&journal->j_async_throttle, 0);
2847         journal->j_bcount = 0;
2848         journal->j_trans_start_time = 0;
2849         journal->j_last = NULL;
2850         journal->j_first = NULL;
2851         init_waitqueue_head(&journal->j_join_wait);
2852         mutex_init(&journal->j_mutex);
2853         mutex_init(&journal->j_flush_mutex);
2854
2855         journal->j_trans_id = 10;
2856         journal->j_mount_id = 10;
2857         journal->j_state = 0;
2858         atomic_set(&journal->j_jlock, 0);
2859         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2860         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2861         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2862         journal->j_cnode_used = 0;
2863         journal->j_must_wait = 0;
2864
2865         if (journal->j_cnode_free == 0) {
2866                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2867                                  "allocation failed (%ld bytes). Journal is "
2868                                  "too large for available memory. Usually "
2869                                  "this is due to a journal that is too large.",
2870                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2871                 goto free_and_return;
2872         }
2873
2874         init_journal_hash(sb);
2875         jl = journal->j_current_jl;
2876
2877         /*
2878          * get_list_bitmap() may call flush_commit_list() which
2879          * requires the lock. Calling flush_commit_list() shouldn't happen
2880          * this early but I like to be paranoid.
2881          */
2882         reiserfs_write_lock(sb);
2883         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2884         reiserfs_write_unlock(sb);
2885         if (!jl->j_list_bitmap) {
2886                 reiserfs_warning(sb, "journal-2005",
2887                                  "get_list_bitmap failed for journal list 0");
2888                 goto free_and_return;
2889         }
2890
2891         ret = journal_read(sb);
2892         if (ret < 0) {
2893                 reiserfs_warning(sb, "reiserfs-2006",
2894                                  "Replay Failure, unable to mount");
2895                 goto free_and_return;
2896         }
2897
2898         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2899         journal->j_work_sb = sb;
2900         return 0;
2901 free_and_return:
2902         free_journal_ram(sb);
2903         return 1;
2904 }
2905
2906 /*
2907  * test for a polite end of the current transaction.  Used by file_write,
2908  * and should be used by delete to make sure they don't write more than
2909  * can fit inside a single transaction
2910  */
2911 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2912                                    int new_alloc)
2913 {
2914         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2915         time_t now = get_seconds();
2916         /* cannot restart while nested */
2917         BUG_ON(!th->t_trans_id);
2918         if (th->t_refcount > 1)
2919                 return 0;
2920         if (journal->j_must_wait > 0 ||
2921             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2922             atomic_read(&journal->j_jlock) ||
2923             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2924             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2925                 return 1;
2926         }
2927
2928         journal->j_len_alloc += new_alloc;
2929         th->t_blocks_allocated += new_alloc ;
2930         return 0;
2931 }
2932
2933 /* this must be called inside a transaction */
2934 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2935 {
2936         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2937         BUG_ON(!th->t_trans_id);
2938         journal->j_must_wait = 1;
2939         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2940         return;
2941 }
2942
2943 /* this must be called without a transaction started */
2944 void reiserfs_allow_writes(struct super_block *s)
2945 {
2946         struct reiserfs_journal *journal = SB_JOURNAL(s);
2947         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2948         wake_up(&journal->j_join_wait);
2949 }
2950
2951 /* this must be called without a transaction started */
2952 void reiserfs_wait_on_write_block(struct super_block *s)
2953 {
2954         struct reiserfs_journal *journal = SB_JOURNAL(s);
2955         wait_event(journal->j_join_wait,
2956                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2957 }
2958
2959 static void queue_log_writer(struct super_block *s)
2960 {
2961         wait_queue_t wait;
2962         struct reiserfs_journal *journal = SB_JOURNAL(s);
2963         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2964
2965         /*
2966          * we don't want to use wait_event here because
2967          * we only want to wait once.
2968          */
2969         init_waitqueue_entry(&wait, current);
2970         add_wait_queue(&journal->j_join_wait, &wait);
2971         set_current_state(TASK_UNINTERRUPTIBLE);
2972         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2973                 int depth = reiserfs_write_unlock_nested(s);
2974                 schedule();
2975                 reiserfs_write_lock_nested(s, depth);
2976         }
2977         __set_current_state(TASK_RUNNING);
2978         remove_wait_queue(&journal->j_join_wait, &wait);
2979 }
2980
2981 static void wake_queued_writers(struct super_block *s)
2982 {
2983         struct reiserfs_journal *journal = SB_JOURNAL(s);
2984         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2985                 wake_up(&journal->j_join_wait);
2986 }
2987
2988 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2989 {
2990         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2991         unsigned long bcount = journal->j_bcount;
2992         while (1) {
2993                 int depth;
2994
2995                 depth = reiserfs_write_unlock_nested(sb);
2996                 schedule_timeout_uninterruptible(1);
2997                 reiserfs_write_lock_nested(sb, depth);
2998
2999                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3000                 while ((atomic_read(&journal->j_wcount) > 0 ||
3001                         atomic_read(&journal->j_jlock)) &&
3002                        journal->j_trans_id == trans_id) {
3003                         queue_log_writer(sb);
3004                 }
3005                 if (journal->j_trans_id != trans_id)
3006                         break;
3007                 if (bcount == journal->j_bcount)
3008                         break;
3009                 bcount = journal->j_bcount;
3010         }
3011 }
3012
3013 /*
3014  * join == true if you must join an existing transaction.
3015  * join == false if you can deal with waiting for others to finish
3016  *
3017  * this will block until the transaction is joinable.  send the number of
3018  * blocks you expect to use in nblocks.
3019 */
3020 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3021                               struct super_block *sb, unsigned long nblocks,
3022                               int join)
3023 {
3024         time_t now = get_seconds();
3025         unsigned int old_trans_id;
3026         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3027         struct reiserfs_transaction_handle myth;
3028         int sched_count = 0;
3029         int retval;
3030         int depth;
3031
3032         reiserfs_check_lock_depth(sb, "journal_begin");
3033         BUG_ON(nblocks > journal->j_trans_max);
3034
3035         PROC_INFO_INC(sb, journal.journal_being);
3036         /* set here for journal_join */
3037         th->t_refcount = 1;
3038         th->t_super = sb;
3039
3040 relock:
3041         lock_journal(sb);
3042         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3043                 unlock_journal(sb);
3044                 retval = journal->j_errno;
3045                 goto out_fail;
3046         }
3047         journal->j_bcount++;
3048
3049         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3050                 unlock_journal(sb);
3051                 depth = reiserfs_write_unlock_nested(sb);
3052                 reiserfs_wait_on_write_block(sb);
3053                 reiserfs_write_lock_nested(sb, depth);
3054                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3055                 goto relock;
3056         }
3057         now = get_seconds();
3058
3059         /*
3060          * if there is no room in the journal OR
3061          * if this transaction is too old, and we weren't called joinable,
3062          * wait for it to finish before beginning we don't sleep if there
3063          * aren't other writers
3064          */
3065
3066         if ((!join && journal->j_must_wait > 0) ||
3067             (!join
3068              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3069             || (!join && atomic_read(&journal->j_wcount) > 0
3070                 && journal->j_trans_start_time > 0
3071                 && (now - journal->j_trans_start_time) >
3072                 journal->j_max_trans_age) || (!join
3073                                               && atomic_read(&journal->j_jlock))
3074             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3075
3076                 old_trans_id = journal->j_trans_id;
3077                 /* allow others to finish this transaction */
3078                 unlock_journal(sb);
3079
3080                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3081                     journal->j_max_batch &&
3082                     ((journal->j_len + nblocks + 2) * 100) <
3083                     (journal->j_len_alloc * 75)) {
3084                         if (atomic_read(&journal->j_wcount) > 10) {
3085                                 sched_count++;
3086                                 queue_log_writer(sb);
3087                                 goto relock;
3088                         }
3089                 }
3090                 /*
3091                  * don't mess with joining the transaction if all we
3092                  * have to do is wait for someone else to do a commit
3093                  */
3094                 if (atomic_read(&journal->j_jlock)) {
3095                         while (journal->j_trans_id == old_trans_id &&
3096                                atomic_read(&journal->j_jlock)) {
3097                                 queue_log_writer(sb);
3098                         }
3099                         goto relock;
3100                 }
3101                 retval = journal_join(&myth, sb);
3102                 if (retval)
3103                         goto out_fail;
3104
3105                 /* someone might have ended the transaction while we joined */
3106                 if (old_trans_id != journal->j_trans_id) {
3107                         retval = do_journal_end(&myth, 0);
3108                 } else {
3109                         retval = do_journal_end(&myth, COMMIT_NOW);
3110                 }
3111
3112                 if (retval)
3113                         goto out_fail;
3114
3115                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3116                 goto relock;
3117         }
3118         /* we are the first writer, set trans_id */
3119         if (journal->j_trans_start_time == 0) {
3120                 journal->j_trans_start_time = get_seconds();
3121         }
3122         atomic_inc(&journal->j_wcount);
3123         journal->j_len_alloc += nblocks;
3124         th->t_blocks_logged = 0;
3125         th->t_blocks_allocated = nblocks;
3126         th->t_trans_id = journal->j_trans_id;
3127         unlock_journal(sb);
3128         INIT_LIST_HEAD(&th->t_list);
3129         return 0;
3130
3131 out_fail:
3132         memset(th, 0, sizeof(*th));
3133         /*
3134          * Re-set th->t_super, so we can properly keep track of how many
3135          * persistent transactions there are. We need to do this so if this
3136          * call is part of a failed restart_transaction, we can free it later
3137          */
3138         th->t_super = sb;
3139         return retval;
3140 }
3141
3142 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3143                                                                     super_block
3144                                                                     *s,
3145                                                                     int nblocks)
3146 {
3147         int ret;
3148         struct reiserfs_transaction_handle *th;
3149
3150         /*
3151          * if we're nesting into an existing transaction.  It will be
3152          * persistent on its own
3153          */
3154         if (reiserfs_transaction_running(s)) {
3155                 th = current->journal_info;
3156                 th->t_refcount++;
3157                 BUG_ON(th->t_refcount < 2);
3158
3159                 return th;
3160         }
3161         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3162         if (!th)
3163                 return NULL;
3164         ret = journal_begin(th, s, nblocks);
3165         if (ret) {
3166                 kfree(th);
3167                 return NULL;
3168         }
3169
3170         SB_JOURNAL(s)->j_persistent_trans++;
3171         return th;
3172 }
3173
3174 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3175 {
3176         struct super_block *s = th->t_super;
3177         int ret = 0;
3178         if (th->t_trans_id)
3179                 ret = journal_end(th);
3180         else
3181                 ret = -EIO;
3182         if (th->t_refcount == 0) {
3183                 SB_JOURNAL(s)->j_persistent_trans--;
3184                 kfree(th);
3185         }
3186         return ret;
3187 }
3188
3189 static int journal_join(struct reiserfs_transaction_handle *th,
3190                         struct super_block *sb)
3191 {
3192         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3193
3194         /*
3195          * this keeps do_journal_end from NULLing out the
3196          * current->journal_info pointer
3197          */
3198         th->t_handle_save = cur_th;
3199         BUG_ON(cur_th && cur_th->t_refcount > 1);
3200         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3201 }
3202
3203 int journal_join_abort(struct reiserfs_transaction_handle *th,
3204                        struct super_block *sb)
3205 {
3206         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3207
3208         /*
3209          * this keeps do_journal_end from NULLing out the
3210          * current->journal_info pointer
3211          */
3212         th->t_handle_save = cur_th;
3213         BUG_ON(cur_th && cur_th->t_refcount > 1);
3214         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3215 }
3216
3217 int journal_begin(struct reiserfs_transaction_handle *th,
3218                   struct super_block *sb, unsigned long nblocks)
3219 {
3220         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3221         int ret;
3222
3223         th->t_handle_save = NULL;
3224         if (cur_th) {
3225                 /* we are nesting into the current transaction */
3226                 if (cur_th->t_super == sb) {
3227                         BUG_ON(!cur_th->t_refcount);
3228                         cur_th->t_refcount++;
3229                         memcpy(th, cur_th, sizeof(*th));
3230                         if (th->t_refcount <= 1)
3231                                 reiserfs_warning(sb, "reiserfs-2005",
3232                                                  "BAD: refcount <= 1, but "
3233                                                  "journal_info != 0");
3234                         return 0;
3235                 } else {
3236                         /*
3237                          * we've ended up with a handle from a different
3238                          * filesystem.  save it and restore on journal_end.
3239                          * This should never really happen...
3240                          */
3241                         reiserfs_warning(sb, "clm-2100",
3242                                          "nesting info a different FS");
3243                         th->t_handle_save = current->journal_info;
3244                         current->journal_info = th;
3245                 }
3246         } else {
3247                 current->journal_info = th;
3248         }
3249         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3250         BUG_ON(current->journal_info != th);
3251
3252         /*
3253          * I guess this boils down to being the reciprocal of clm-2100 above.
3254          * If do_journal_begin_r fails, we need to put it back, since
3255          * journal_end won't be called to do it. */
3256         if (ret)
3257                 current->journal_info = th->t_handle_save;
3258         else
3259                 BUG_ON(!th->t_refcount);
3260
3261         return ret;
3262 }
3263
3264 /*
3265  * puts bh into the current transaction.  If it was already there, reorders
3266  * removes the old pointers from the hash, and puts new ones in (to make
3267  * sure replay happen in the right order).
3268  *
3269  * if it was dirty, cleans and files onto the clean list.  I can't let it
3270  * be dirty again until the transaction is committed.
3271  *
3272  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3273  */
3274 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3275                        struct buffer_head *bh)
3276 {
3277         struct super_block *sb = th->t_super;
3278         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3279         struct reiserfs_journal_cnode *cn = NULL;
3280         int count_already_incd = 0;
3281         int prepared = 0;
3282         BUG_ON(!th->t_trans_id);
3283
3284         PROC_INFO_INC(sb, journal.mark_dirty);
3285         if (th->t_trans_id != journal->j_trans_id) {
3286                 reiserfs_panic(th->t_super, "journal-1577",
3287                                "handle trans id %ld != current trans id %ld",
3288                                th->t_trans_id, journal->j_trans_id);
3289         }
3290
3291         prepared = test_clear_buffer_journal_prepared(bh);
3292         clear_buffer_journal_restore_dirty(bh);
3293         /* already in this transaction, we are done */
3294         if (buffer_journaled(bh)) {
3295                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3296                 return 0;
3297         }
3298
3299         /*
3300          * this must be turned into a panic instead of a warning.  We can't
3301          * allow a dirty or journal_dirty or locked buffer to be logged, as
3302          * some changes could get to disk too early.  NOT GOOD.
3303          */
3304         if (!prepared || buffer_dirty(bh)) {
3305                 reiserfs_warning(sb, "journal-1777",
3306                                  "buffer %llu bad state "
3307                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3308                                  (unsigned long long)bh->b_blocknr,
3309                                  prepared ? ' ' : '!',
3310                                  buffer_locked(bh) ? ' ' : '!',
3311                                  buffer_dirty(bh) ? ' ' : '!',
3312                                  buffer_journal_dirty(bh) ? ' ' : '!');
3313         }
3314
3315         if (atomic_read(&journal->j_wcount) <= 0) {
3316                 reiserfs_warning(sb, "journal-1409",
3317                                  "returning because j_wcount was %d",
3318                                  atomic_read(&journal->j_wcount));
3319                 return 1;
3320         }
3321         /*
3322          * this error means I've screwed up, and we've overflowed
3323          * the transaction.  Nothing can be done here, except make the
3324          * FS readonly or panic.
3325          */
3326         if (journal->j_len >= journal->j_trans_max) {
3327                 reiserfs_panic(th->t_super, "journal-1413",
3328                                "j_len (%lu) is too big",
3329                                journal->j_len);
3330         }
3331
3332         if (buffer_journal_dirty(bh)) {
3333                 count_already_incd = 1;
3334                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3335                 clear_buffer_journal_dirty(bh);
3336         }
3337
3338         if (journal->j_len > journal->j_len_alloc) {
3339                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3340         }
3341
3342         set_buffer_journaled(bh);
3343
3344         /* now put this guy on the end */
3345         if (!cn) {
3346                 cn = get_cnode(sb);
3347                 if (!cn) {
3348                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3349                 }
3350
3351                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3352                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3353                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3354                 }
3355                 th->t_blocks_logged++;
3356                 journal->j_len++;
3357
3358                 cn->bh = bh;
3359                 cn->blocknr = bh->b_blocknr;
3360                 cn->sb = sb;
3361                 cn->jlist = NULL;
3362                 insert_journal_hash(journal->j_hash_table, cn);
3363                 if (!count_already_incd) {
3364                         get_bh(bh);
3365                 }
3366         }
3367         cn->next = NULL;
3368         cn->prev = journal->j_last;
3369         cn->bh = bh;
3370         if (journal->j_last) {
3371                 journal->j_last->next = cn;
3372                 journal->j_last = cn;
3373         } else {
3374                 journal->j_first = cn;
3375                 journal->j_last = cn;
3376         }
3377         reiserfs_schedule_old_flush(sb);
3378         return 0;
3379 }
3380
3381 int journal_end(struct reiserfs_transaction_handle *th)
3382 {
3383         struct super_block *sb = th->t_super;
3384         if (!current->journal_info && th->t_refcount > 1)
3385                 reiserfs_warning(sb, "REISER-NESTING",
3386                                  "th NULL, refcount %d", th->t_refcount);
3387
3388         if (!th->t_trans_id) {
3389                 WARN_ON(1);
3390                 return -EIO;
3391         }
3392
3393         th->t_refcount--;
3394         if (th->t_refcount > 0) {
3395                 struct reiserfs_transaction_handle *cur_th =
3396                     current->journal_info;
3397
3398                 /*
3399                  * we aren't allowed to close a nested transaction on a
3400                  * different filesystem from the one in the task struct
3401                  */
3402                 BUG_ON(cur_th->t_super != th->t_super);
3403
3404                 if (th != cur_th) {
3405                         memcpy(current->journal_info, th, sizeof(*th));
3406                         th->t_trans_id = 0;
3407                 }
3408                 return 0;
3409         } else {
3410                 return do_journal_end(th, 0);
3411         }
3412 }
3413
3414 /*
3415  * removes from the current transaction, relsing and descrementing any counters.
3416  * also files the removed buffer directly onto the clean list
3417  *
3418  * called by journal_mark_freed when a block has been deleted
3419  *
3420  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3421  */
3422 static int remove_from_transaction(struct super_block *sb,
3423                                    b_blocknr_t blocknr, int already_cleaned)
3424 {
3425         struct buffer_head *bh;
3426         struct reiserfs_journal_cnode *cn;
3427         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3428         int ret = 0;
3429
3430         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3431         if (!cn || !cn->bh) {
3432                 return ret;
3433         }
3434         bh = cn->bh;
3435         if (cn->prev) {
3436                 cn->prev->next = cn->next;
3437         }
3438         if (cn->next) {
3439                 cn->next->prev = cn->prev;
3440         }
3441         if (cn == journal->j_first) {
3442                 journal->j_first = cn->next;
3443         }
3444         if (cn == journal->j_last) {
3445                 journal->j_last = cn->prev;
3446         }
3447         if (bh)
3448                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3449                                     bh->b_blocknr, 0);
3450         clear_buffer_journaled(bh);     /* don't log this one */
3451
3452         if (!already_cleaned) {
3453                 clear_buffer_journal_dirty(bh);
3454                 clear_buffer_dirty(bh);
3455                 clear_buffer_journal_test(bh);
3456                 put_bh(bh);
3457                 if (atomic_read(&bh->b_count) < 0) {
3458                         reiserfs_warning(sb, "journal-1752",
3459                                          "b_count < 0");
3460                 }
3461                 ret = 1;
3462         }
3463         journal->j_len--;
3464         journal->j_len_alloc--;
3465         free_cnode(sb, cn);
3466         return ret;
3467 }
3468
3469 /*
3470  * for any cnode in a journal list, it can only be dirtied of all the
3471  * transactions that include it are committed to disk.
3472  * this checks through each transaction, and returns 1 if you are allowed
3473  * to dirty, and 0 if you aren't
3474  *
3475  * it is called by dirty_journal_list, which is called after
3476  * flush_commit_list has gotten all the log blocks for a given
3477  * transaction on disk
3478  *
3479  */
3480 static int can_dirty(struct reiserfs_journal_cnode *cn)
3481 {
3482         struct super_block *sb = cn->sb;
3483         b_blocknr_t blocknr = cn->blocknr;
3484         struct reiserfs_journal_cnode *cur = cn->hprev;
3485         int can_dirty = 1;
3486
3487         /*
3488          * first test hprev.  These are all newer than cn, so any node here
3489          * with the same block number and dev means this node can't be sent
3490          * to disk right now.
3491          */
3492         while (cur && can_dirty) {
3493                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3494                     cur->blocknr == blocknr) {
3495                         can_dirty = 0;
3496                 }
3497                 cur = cur->hprev;
3498         }
3499         /*
3500          * then test hnext.  These are all older than cn.  As long as they
3501          * are committed to the log, it is safe to write cn to disk
3502          */
3503         cur = cn->hnext;
3504         while (cur && can_dirty) {
3505                 if (cur->jlist && cur->jlist->j_len > 0 &&
3506                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3507                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3508                         can_dirty = 0;
3509                 }
3510                 cur = cur->hnext;
3511         }
3512         return can_dirty;
3513 }
3514
3515 /*
3516  * syncs the commit blocks, but does not force the real buffers to disk
3517  * will wait until the current transaction is done/committed before returning
3518  */
3519 int journal_end_sync(struct reiserfs_transaction_handle *th)
3520 {
3521         struct super_block *sb = th->t_super;
3522         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3523
3524         BUG_ON(!th->t_trans_id);
3525         /* you can sync while nested, very, very bad */
3526         BUG_ON(th->t_refcount > 1);
3527         if (journal->j_len == 0) {
3528                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3529                                              1);
3530                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3531         }
3532         return do_journal_end(th, COMMIT_NOW | WAIT);
3533 }
3534
3535 /* writeback the pending async commits to disk */
3536 static void flush_async_commits(struct work_struct *work)
3537 {
3538         struct reiserfs_journal *journal =
3539                 container_of(work, struct reiserfs_journal, j_work.work);
3540         struct super_block *sb = journal->j_work_sb;
3541         struct reiserfs_journal_list *jl;
3542         struct list_head *entry;
3543
3544         reiserfs_write_lock(sb);
3545         if (!list_empty(&journal->j_journal_list)) {
3546                 /* last entry is the youngest, commit it and you get everything */
3547                 entry = journal->j_journal_list.prev;
3548                 jl = JOURNAL_LIST_ENTRY(entry);
3549                 flush_commit_list(sb, jl, 1);
3550         }
3551         reiserfs_write_unlock(sb);
3552 }
3553
3554 /*
3555  * flushes any old transactions to disk
3556  * ends the current transaction if it is too old
3557  */
3558 void reiserfs_flush_old_commits(struct super_block *sb)
3559 {
3560         time_t now;
3561         struct reiserfs_transaction_handle th;
3562         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3563
3564         now = get_seconds();
3565         /*
3566          * safety check so we don't flush while we are replaying the log during
3567          * mount
3568          */
3569         if (list_empty(&journal->j_journal_list))
3570                 return;
3571
3572         /*
3573          * check the current transaction.  If there are no writers, and it is
3574          * too old, finish it, and force the commit blocks to disk
3575          */
3576         if (atomic_read(&journal->j_wcount) <= 0 &&
3577             journal->j_trans_start_time > 0 &&
3578             journal->j_len > 0 &&
3579             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3580                 if (!journal_join(&th, sb)) {
3581                         reiserfs_prepare_for_journal(sb,
3582                                                      SB_BUFFER_WITH_SB(sb),
3583                                                      1);
3584                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3585
3586                         /*
3587                          * we're only being called from kreiserfsd, it makes
3588                          * no sense to do an async commit so that kreiserfsd
3589                          * can do it later
3590                          */
3591                         do_journal_end(&th, COMMIT_NOW | WAIT);
3592                 }
3593         }
3594 }
3595
3596 /*
3597  * returns 0 if do_journal_end should return right away, returns 1 if
3598  * do_journal_end should finish the commit
3599  *
3600  * if the current transaction is too old, but still has writers, this will
3601  * wait on j_join_wait until all the writers are done.  By the time it
3602  * wakes up, the transaction it was called has already ended, so it just
3603  * flushes the commit list and returns 0.
3604  *
3605  * Won't batch when flush or commit_now is set.  Also won't batch when
3606  * others are waiting on j_join_wait.
3607  *
3608  * Note, we can't allow the journal_end to proceed while there are still
3609  * writers in the log.
3610  */
3611 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3612 {
3613
3614         time_t now;
3615         int flush = flags & FLUSH_ALL;
3616         int commit_now = flags & COMMIT_NOW;
3617         int wait_on_commit = flags & WAIT;
3618         struct reiserfs_journal_list *jl;
3619         struct super_block *sb = th->t_super;
3620         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3621
3622         BUG_ON(!th->t_trans_id);
3623
3624         if (th->t_trans_id != journal->j_trans_id) {
3625                 reiserfs_panic(th->t_super, "journal-1577",
3626                                "handle trans id %ld != current trans id %ld",
3627                                th->t_trans_id, journal->j_trans_id);
3628         }
3629
3630         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3631         /* <= 0 is allowed.  unmounting might not call begin */
3632         if (atomic_read(&journal->j_wcount) > 0)
3633                 atomic_dec(&journal->j_wcount);
3634
3635         /*
3636          * BUG, deal with case where j_len is 0, but people previously
3637          * freed blocks need to be released will be dealt with by next
3638          * transaction that actually writes something, but should be taken
3639          * care of in this trans
3640          */
3641         BUG_ON(journal->j_len == 0);
3642
3643         /*
3644          * if wcount > 0, and we are called to with flush or commit_now,
3645          * we wait on j_join_wait.  We will wake up when the last writer has
3646          * finished the transaction, and started it on its way to the disk.
3647          * Then, we flush the commit or journal list, and just return 0
3648          * because the rest of journal end was already done for this
3649          * transaction.
3650          */
3651         if (atomic_read(&journal->j_wcount) > 0) {
3652                 if (flush || commit_now) {
3653                         unsigned trans_id;
3654
3655                         jl = journal->j_current_jl;
3656                         trans_id = jl->j_trans_id;
3657                         if (wait_on_commit)
3658                                 jl->j_state |= LIST_COMMIT_PENDING;
3659                         atomic_set(&journal->j_jlock, 1);
3660                         if (flush) {
3661                                 journal->j_next_full_flush = 1;
3662                         }
3663                         unlock_journal(sb);
3664
3665                         /*
3666                          * sleep while the current transaction is
3667                          * still j_jlocked
3668                          */
3669                         while (journal->j_trans_id == trans_id) {
3670                                 if (atomic_read(&journal->j_jlock)) {
3671                                         queue_log_writer(sb);
3672                                 } else {
3673                                         lock_journal(sb);
3674                                         if (journal->j_trans_id == trans_id) {
3675                                                 atomic_set(&journal->j_jlock,
3676                                                            1);
3677                                         }
3678                                         unlock_journal(sb);
3679                                 }
3680                         }
3681                         BUG_ON(journal->j_trans_id == trans_id);
3682
3683                         if (commit_now
3684                             && journal_list_still_alive(sb, trans_id)
3685                             && wait_on_commit) {
3686                                 flush_commit_list(sb, jl, 1);
3687                         }
3688                         return 0;
3689                 }
3690                 unlock_journal(sb);
3691                 return 0;
3692         }
3693
3694         /* deal with old transactions where we are the last writers */
3695         now = get_seconds();
3696         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3697                 commit_now = 1;
3698                 journal->j_next_async_flush = 1;
3699         }
3700         /* don't batch when someone is waiting on j_join_wait */
3701         /* don't batch when syncing the commit or flushing the whole trans */
3702         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3703             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3704             && journal->j_len_alloc < journal->j_max_batch
3705             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3706                 journal->j_bcount++;
3707                 unlock_journal(sb);
3708                 return 0;
3709         }
3710
3711         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3712                 reiserfs_panic(sb, "journal-003",
3713                                "j_start (%ld) is too high",
3714                                journal->j_start);
3715         }
3716         return 1;
3717 }
3718
3719 /*
3720  * Does all the work that makes deleting blocks safe.
3721  * when deleting a block mark BH_JNew, just remove it from the current
3722  * transaction, clean it's buffer_head and move on.
3723  *
3724  * otherwise:
3725  * set a bit for the block in the journal bitmap.  That will prevent it from
3726  * being allocated for unformatted nodes before this transaction has finished.
3727  *
3728  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3729  * That will prevent any old transactions with this block from trying to flush
3730  * to the real location.  Since we aren't removing the cnode from the
3731  * journal_list_hash, *the block can't be reallocated yet.
3732  *
3733  * Then remove it from the current transaction, decrementing any counters and
3734  * filing it on the clean list.
3735  */
3736 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3737                        struct super_block *sb, b_blocknr_t blocknr)
3738 {
3739         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3740         struct reiserfs_journal_cnode *cn = NULL;
3741         struct buffer_head *bh = NULL;
3742         struct reiserfs_list_bitmap *jb = NULL;
3743         int cleaned = 0;
3744         BUG_ON(!th->t_trans_id);
3745
3746         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3747         if (cn && cn->bh) {
3748                 bh = cn->bh;
3749                 get_bh(bh);
3750         }
3751         /* if it is journal new, we just remove it from this transaction */
3752         if (bh && buffer_journal_new(bh)) {
3753                 clear_buffer_journal_new(bh);
3754                 clear_prepared_bits(bh);
3755                 reiserfs_clean_and_file_buffer(bh);
3756                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3757         } else {
3758                 /*
3759                  * set the bit for this block in the journal bitmap
3760                  * for this transaction
3761                  */
3762                 jb = journal->j_current_jl->j_list_bitmap;
3763                 if (!jb) {
3764                         reiserfs_panic(sb, "journal-1702",
3765                                        "journal_list_bitmap is NULL");
3766                 }
3767                 set_bit_in_list_bitmap(sb, blocknr, jb);
3768
3769                 /* Note, the entire while loop is not allowed to schedule.  */
3770
3771                 if (bh) {
3772                         clear_prepared_bits(bh);
3773                         reiserfs_clean_and_file_buffer(bh);
3774                 }
3775                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3776
3777                 /*
3778                  * find all older transactions with this block,
3779                  * make sure they don't try to write it out
3780                  */
3781                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3782                                           blocknr);
3783                 while (cn) {
3784                         if (sb == cn->sb && blocknr == cn->blocknr) {
3785                                 set_bit(BLOCK_FREED, &cn->state);
3786                                 if (cn->bh) {
3787                                         /*
3788                                          * remove_from_transaction will brelse
3789                                          * the buffer if it was in the current
3790                                          * trans
3791                                          */
3792                                         if (!cleaned) {
3793                                                 clear_buffer_journal_dirty(cn->
3794                                                                            bh);
3795                                                 clear_buffer_dirty(cn->bh);
3796                                                 clear_buffer_journal_test(cn->
3797                                                                           bh);
3798                                                 cleaned = 1;
3799                                                 put_bh(cn->bh);
3800                                                 if (atomic_read
3801                                                     (&cn->bh->b_count) < 0) {
3802                                                         reiserfs_warning(sb,
3803                                                                  "journal-2138",
3804                                                                  "cn->bh->b_count < 0");
3805                                                 }
3806                                         }
3807                                         /*
3808                                          * since we are clearing the bh,
3809                                          * we MUST dec nonzerolen
3810                                          */
3811                                         if (cn->jlist) {
3812                                                 atomic_dec(&cn->jlist->
3813                                                            j_nonzerolen);
3814                                         }
3815                                         cn->bh = NULL;
3816                                 }
3817                         }
3818                         cn = cn->hnext;
3819                 }
3820         }
3821
3822         if (bh)
3823                 release_buffer_page(bh); /* get_hash grabs the buffer */
3824         return 0;
3825 }
3826
3827 void reiserfs_update_inode_transaction(struct inode *inode)
3828 {
3829         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3830         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3831         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3832 }
3833
3834 /*
3835  * returns -1 on error, 0 if no commits/barriers were done and 1
3836  * if a transaction was actually committed and the barrier was done
3837  */
3838 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3839                              struct reiserfs_journal_list *jl)
3840 {
3841         struct reiserfs_transaction_handle th;
3842         struct super_block *sb = inode->i_sb;
3843         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3844         int ret = 0;
3845
3846         /*
3847          * is it from the current transaction,
3848          * or from an unknown transaction?
3849          */
3850         if (id == journal->j_trans_id) {
3851                 jl = journal->j_current_jl;
3852                 /*
3853                  * try to let other writers come in and
3854                  * grow this transaction
3855                  */
3856                 let_transaction_grow(sb, id);
3857                 if (journal->j_trans_id != id) {
3858                         goto flush_commit_only;
3859                 }
3860
3861                 ret = journal_begin(&th, sb, 1);
3862                 if (ret)
3863                         return ret;
3864
3865                 /* someone might have ended this transaction while we joined */
3866                 if (journal->j_trans_id != id) {
3867                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3868                                                      1);
3869                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3870                         ret = journal_end(&th);
3871                         goto flush_commit_only;
3872                 }
3873
3874                 ret = journal_end_sync(&th);
3875                 if (!ret)
3876                         ret = 1;
3877
3878         } else {
3879                 /*
3880                  * this gets tricky, we have to make sure the journal list in
3881                  * the inode still exists.  We know the list is still around
3882                  * if we've got a larger transaction id than the oldest list
3883                  */
3884 flush_commit_only:
3885                 if (journal_list_still_alive(inode->i_sb, id)) {
3886                         /*
3887                          * we only set ret to 1 when we know for sure
3888                          * the barrier hasn't been started yet on the commit
3889                          * block.
3890                          */
3891                         if (atomic_read(&jl->j_commit_left) > 1)
3892                                 ret = 1;
3893                         flush_commit_list(sb, jl, 1);
3894                         if (journal->j_errno)
3895                                 ret = journal->j_errno;
3896                 }
3897         }
3898         /* otherwise the list is gone, and long since committed */
3899         return ret;
3900 }
3901
3902 int reiserfs_commit_for_inode(struct inode *inode)
3903 {
3904         unsigned int id = REISERFS_I(inode)->i_trans_id;
3905         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3906
3907         /*
3908          * for the whole inode, assume unset id means it was
3909          * changed in the current transaction.  More conservative
3910          */
3911         if (!id || !jl) {
3912                 reiserfs_update_inode_transaction(inode);
3913                 id = REISERFS_I(inode)->i_trans_id;
3914                 /* jl will be updated in __commit_trans_jl */
3915         }
3916
3917         return __commit_trans_jl(inode, id, jl);
3918 }
3919
3920 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3921                                       struct buffer_head *bh)
3922 {
3923         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3924         PROC_INFO_INC(sb, journal.restore_prepared);
3925         if (!bh) {
3926                 return;
3927         }
3928         if (test_clear_buffer_journal_restore_dirty(bh) &&
3929             buffer_journal_dirty(bh)) {
3930                 struct reiserfs_journal_cnode *cn;
3931                 reiserfs_write_lock(sb);
3932                 cn = get_journal_hash_dev(sb,
3933                                           journal->j_list_hash_table,
3934                                           bh->b_blocknr);
3935                 if (cn && can_dirty(cn)) {
3936                         set_buffer_journal_test(bh);
3937                         mark_buffer_dirty(bh);
3938                 }
3939                 reiserfs_write_unlock(sb);
3940         }
3941         clear_buffer_journal_prepared(bh);
3942 }
3943
3944 extern struct tree_balance *cur_tb;
3945 /*
3946  * before we can change a metadata block, we have to make sure it won't
3947  * be written to disk while we are altering it.  So, we must:
3948  * clean it
3949  * wait on it.
3950  */
3951 int reiserfs_prepare_for_journal(struct super_block *sb,
3952                                  struct buffer_head *bh, int wait)
3953 {
3954         PROC_INFO_INC(sb, journal.prepare);
3955
3956         if (!trylock_buffer(bh)) {
3957                 if (!wait)
3958                         return 0;
3959                 lock_buffer(bh);
3960         }
3961         set_buffer_journal_prepared(bh);
3962         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3963                 clear_buffer_journal_test(bh);
3964                 set_buffer_journal_restore_dirty(bh);
3965         }
3966         unlock_buffer(bh);
3967         return 1;
3968 }
3969
3970 /*
3971  * long and ugly.  If flush, will not return until all commit
3972  * blocks and all real buffers in the trans are on disk.
3973  * If no_async, won't return until all commit blocks are on disk.
3974  *
3975  * keep reading, there are comments as you go along
3976  *
3977  * If the journal is aborted, we just clean up. Things like flushing
3978  * journal lists, etc just won't happen.
3979  */
3980 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3981 {
3982         struct super_block *sb = th->t_super;
3983         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3984         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3985         struct reiserfs_journal_cnode *last_cn = NULL;
3986         struct reiserfs_journal_desc *desc;
3987         struct reiserfs_journal_commit *commit;
3988         struct buffer_head *c_bh;       /* commit bh */
3989         struct buffer_head *d_bh;       /* desc bh */
3990         int cur_write_start = 0;        /* start index of current log write */
3991         int old_start;
3992         int i;
3993         int flush;
3994         int wait_on_commit;
3995         struct reiserfs_journal_list *jl, *temp_jl;
3996         struct list_head *entry, *safe;
3997         unsigned long jindex;
3998         unsigned int commit_trans_id;
3999         int trans_half;
4000         int depth;
4001
4002         BUG_ON(th->t_refcount > 1);
4003         BUG_ON(!th->t_trans_id);
4004         BUG_ON(!th->t_super);
4005
4006         /*
4007          * protect flush_older_commits from doing mistakes if the
4008          * transaction ID counter gets overflowed.
4009          */
4010         if (th->t_trans_id == ~0U)
4011                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4012         flush = flags & FLUSH_ALL;
4013         wait_on_commit = flags & WAIT;
4014
4015         current->journal_info = th->t_handle_save;
4016         reiserfs_check_lock_depth(sb, "journal end");
4017         if (journal->j_len == 0) {
4018                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4019                                              1);
4020                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4021         }
4022
4023         lock_journal(sb);
4024         if (journal->j_next_full_flush) {
4025                 flags |= FLUSH_ALL;
4026                 flush = 1;
4027         }
4028         if (journal->j_next_async_flush) {
4029                 flags |= COMMIT_NOW | WAIT;
4030                 wait_on_commit = 1;
4031         }
4032
4033         /*
4034          * check_journal_end locks the journal, and unlocks if it does
4035          * not return 1 it tells us if we should continue with the
4036          * journal_end, or just return
4037          */
4038         if (!check_journal_end(th, flags)) {
4039                 reiserfs_schedule_old_flush(sb);
4040                 wake_queued_writers(sb);
4041                 reiserfs_async_progress_wait(sb);
4042                 goto out;
4043         }
4044
4045         /* check_journal_end might set these, check again */
4046         if (journal->j_next_full_flush) {
4047                 flush = 1;
4048         }
4049
4050         /*
4051          * j must wait means we have to flush the log blocks, and the
4052          * real blocks for this transaction
4053          */
4054         if (journal->j_must_wait > 0) {
4055                 flush = 1;
4056         }
4057 #ifdef REISERFS_PREALLOCATE
4058         /*
4059          * quota ops might need to nest, setup the journal_info pointer
4060          * for them and raise the refcount so that it is > 0.
4061          */
4062         current->journal_info = th;
4063         th->t_refcount++;
4064
4065         /* it should not involve new blocks into the transaction */
4066         reiserfs_discard_all_prealloc(th);
4067
4068         th->t_refcount--;
4069         current->journal_info = th->t_handle_save;
4070 #endif
4071
4072         /* setup description block */
4073         d_bh =
4074             journal_getblk(sb,
4075                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4076                            journal->j_start);
4077         set_buffer_uptodate(d_bh);
4078         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4079         memset(d_bh->b_data, 0, d_bh->b_size);
4080         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4081         set_desc_trans_id(desc, journal->j_trans_id);
4082
4083         /*
4084          * setup commit block.  Don't write (keep it clean too) this one
4085          * until after everyone else is written
4086          */
4087         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4088                               ((journal->j_start + journal->j_len +
4089                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4090         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4091         memset(c_bh->b_data, 0, c_bh->b_size);
4092         set_commit_trans_id(commit, journal->j_trans_id);
4093         set_buffer_uptodate(c_bh);
4094
4095         /* init this journal list */
4096         jl = journal->j_current_jl;
4097
4098         /*
4099          * we lock the commit before doing anything because
4100          * we want to make sure nobody tries to run flush_commit_list until
4101          * the new transaction is fully setup, and we've already flushed the
4102          * ordered bh list
4103          */
4104         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4105
4106         /* save the transaction id in case we need to commit it later */
4107         commit_trans_id = jl->j_trans_id;
4108
4109         atomic_set(&jl->j_older_commits_done, 0);
4110         jl->j_trans_id = journal->j_trans_id;
4111         jl->j_timestamp = journal->j_trans_start_time;
4112         jl->j_commit_bh = c_bh;
4113         jl->j_start = journal->j_start;
4114         jl->j_len = journal->j_len;
4115         atomic_set(&jl->j_nonzerolen, journal->j_len);
4116         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4117         jl->j_realblock = NULL;
4118
4119         /*
4120          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4121          * for each real block, add it to the journal list hash,
4122          * copy into real block index array in the commit or desc block
4123          */
4124         trans_half = journal_trans_half(sb->s_blocksize);
4125         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4126                 if (buffer_journaled(cn->bh)) {
4127                         jl_cn = get_cnode(sb);
4128                         if (!jl_cn) {
4129                                 reiserfs_panic(sb, "journal-1676",
4130                                                "get_cnode returned NULL");
4131                         }
4132                         if (i == 0) {
4133                                 jl->j_realblock = jl_cn;
4134                         }
4135                         jl_cn->prev = last_cn;
4136                         jl_cn->next = NULL;
4137                         if (last_cn) {
4138                                 last_cn->next = jl_cn;
4139                         }
4140                         last_cn = jl_cn;
4141                         /*
4142                          * make sure the block we are trying to log
4143                          * is not a block of journal or reserved area
4144                          */
4145                         if (is_block_in_log_or_reserved_area
4146                             (sb, cn->bh->b_blocknr)) {
4147                                 reiserfs_panic(sb, "journal-2332",
4148                                                "Trying to log block %lu, "
4149                                                "which is a log block",
4150                                                cn->bh->b_blocknr);
4151                         }
4152                         jl_cn->blocknr = cn->bh->b_blocknr;
4153                         jl_cn->state = 0;
4154                         jl_cn->sb = sb;
4155                         jl_cn->bh = cn->bh;
4156                         jl_cn->jlist = jl;
4157                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4158                         if (i < trans_half) {
4159                                 desc->j_realblock[i] =
4160                                     cpu_to_le32(cn->bh->b_blocknr);
4161                         } else {
4162                                 commit->j_realblock[i - trans_half] =
4163                                     cpu_to_le32(cn->bh->b_blocknr);
4164                         }
4165                 } else {
4166                         i--;
4167                 }
4168         }
4169         set_desc_trans_len(desc, journal->j_len);
4170         set_desc_mount_id(desc, journal->j_mount_id);
4171         set_desc_trans_id(desc, journal->j_trans_id);
4172         set_commit_trans_len(commit, journal->j_len);
4173
4174         /*
4175          * special check in case all buffers in the journal
4176          * were marked for not logging
4177          */
4178         BUG_ON(journal->j_len == 0);
4179
4180         /*
4181          * we're about to dirty all the log blocks, mark the description block
4182          * dirty now too.  Don't mark the commit block dirty until all the
4183          * others are on disk
4184          */
4185         mark_buffer_dirty(d_bh);
4186
4187         /*
4188          * first data block is j_start + 1, so add one to
4189          * cur_write_start wherever you use it
4190          */
4191         cur_write_start = journal->j_start;
4192         cn = journal->j_first;
4193         jindex = 1;     /* start at one so we don't get the desc again */
4194         while (cn) {
4195                 clear_buffer_journal_new(cn->bh);
4196                 /* copy all the real blocks into log area.  dirty log blocks */
4197                 if (buffer_journaled(cn->bh)) {
4198                         struct buffer_head *tmp_bh;
4199                         char *addr;
4200                         struct page *page;
4201                         tmp_bh =
4202                             journal_getblk(sb,
4203                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4204                                            ((cur_write_start +
4205                                              jindex) %
4206                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4207                         set_buffer_uptodate(tmp_bh);
4208                         page = cn->bh->b_page;
4209                         addr = kmap(page);
4210                         memcpy(tmp_bh->b_data,
4211                                addr + offset_in_page(cn->bh->b_data),
4212                                cn->bh->b_size);
4213                         kunmap(page);
4214                         mark_buffer_dirty(tmp_bh);
4215                         jindex++;
4216                         set_buffer_journal_dirty(cn->bh);
4217                         clear_buffer_journaled(cn->bh);
4218                 } else {
4219                         /*
4220                          * JDirty cleared sometime during transaction.
4221                          * don't log this one
4222                          */
4223                         reiserfs_warning(sb, "journal-2048",
4224                                          "BAD, buffer in journal hash, "
4225                                          "but not JDirty!");
4226                         brelse(cn->bh);
4227                 }
4228                 next = cn->next;
4229                 free_cnode(sb, cn);
4230                 cn = next;
4231                 reiserfs_cond_resched(sb);
4232         }
4233
4234         /*
4235          * we are done with both the c_bh and d_bh, but
4236          * c_bh must be written after all other commit blocks,
4237          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4238          */
4239
4240         journal->j_current_jl = alloc_journal_list(sb);
4241
4242         /* now it is safe to insert this transaction on the main list */
4243         list_add_tail(&jl->j_list, &journal->j_journal_list);
4244         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4245         journal->j_num_work_lists++;
4246
4247         /* reset journal values for the next transaction */
4248         old_start = journal->j_start;
4249         journal->j_start =
4250             (journal->j_start + journal->j_len +
4251              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4252         atomic_set(&journal->j_wcount, 0);
4253         journal->j_bcount = 0;
4254         journal->j_last = NULL;
4255         journal->j_first = NULL;
4256         journal->j_len = 0;
4257         journal->j_trans_start_time = 0;
4258         /* check for trans_id overflow */
4259         if (++journal->j_trans_id == 0)
4260                 journal->j_trans_id = 10;
4261         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4262         journal->j_must_wait = 0;
4263         journal->j_len_alloc = 0;
4264         journal->j_next_full_flush = 0;
4265         journal->j_next_async_flush = 0;
4266         init_journal_hash(sb);
4267
4268         /*
4269          * make sure reiserfs_add_jh sees the new current_jl before we
4270          * write out the tails
4271          */
4272         smp_mb();
4273
4274         /*
4275          * tail conversion targets have to hit the disk before we end the
4276          * transaction.  Otherwise a later transaction might repack the tail
4277          * before this transaction commits, leaving the data block unflushed
4278          * and clean, if we crash before the later transaction commits, the
4279          * data block is lost.
4280          */
4281         if (!list_empty(&jl->j_tail_bh_list)) {
4282                 depth = reiserfs_write_unlock_nested(sb);
4283                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4284                                       journal, jl, &jl->j_tail_bh_list);
4285                 reiserfs_write_lock_nested(sb, depth);
4286         }
4287         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4288         mutex_unlock(&jl->j_commit_mutex);
4289
4290         /*
4291          * honor the flush wishes from the caller, simple commits can
4292          * be done outside the journal lock, they are done below
4293          *
4294          * if we don't flush the commit list right now, we put it into
4295          * the work queue so the people waiting on the async progress work
4296          * queue don't wait for this proc to flush journal lists and such.
4297          */
4298         if (flush) {
4299                 flush_commit_list(sb, jl, 1);
4300                 flush_journal_list(sb, jl, 1);
4301         } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4302                 /*
4303                  * Avoid queueing work when sb is being shut down. Transaction
4304                  * will be flushed on journal shutdown.
4305                  */
4306                 if (sb->s_flags & MS_ACTIVE)
4307                         queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4308                                            &journal->j_work, HZ / 10);
4309         }
4310
4311         /*
4312          * if the next transaction has any chance of wrapping, flush
4313          * transactions that might get overwritten.  If any journal lists
4314          * are very old flush them as well.
4315          */
4316 first_jl:
4317         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4318                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4319                 if (journal->j_start <= temp_jl->j_start) {
4320                         if ((journal->j_start + journal->j_trans_max + 1) >=
4321                             temp_jl->j_start) {
4322                                 flush_used_journal_lists(sb, temp_jl);
4323                                 goto first_jl;
4324                         } else if ((journal->j_start +
4325                                     journal->j_trans_max + 1) <
4326                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4327                                 /*
4328                                  * if we don't cross into the next
4329                                  * transaction and we don't wrap, there is
4330                                  * no way we can overlap any later transactions
4331                                  * break now
4332                                  */
4333                                 break;
4334                         }
4335                 } else if ((journal->j_start +
4336                             journal->j_trans_max + 1) >
4337                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4338                         if (((journal->j_start + journal->j_trans_max + 1) %
4339                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4340                             temp_jl->j_start) {
4341                                 flush_used_journal_lists(sb, temp_jl);
4342                                 goto first_jl;
4343                         } else {
4344                                 /*
4345                                 * we don't overlap anything from out start
4346                                 * to the end of the log, and our wrapped
4347                                 * portion doesn't overlap anything at
4348                                 * the start of the log.  We can break
4349                                 */
4350                                 break;
4351                         }
4352                 }
4353         }
4354
4355         journal->j_current_jl->j_list_bitmap =
4356             get_list_bitmap(sb, journal->j_current_jl);
4357
4358         if (!(journal->j_current_jl->j_list_bitmap)) {
4359                 reiserfs_panic(sb, "journal-1996",
4360                                "could not get a list bitmap");
4361         }
4362
4363         atomic_set(&journal->j_jlock, 0);
4364         unlock_journal(sb);
4365         /* wake up any body waiting to join. */
4366         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4367         wake_up(&journal->j_join_wait);
4368
4369         if (!flush && wait_on_commit &&
4370             journal_list_still_alive(sb, commit_trans_id)) {
4371                 flush_commit_list(sb, jl, 1);
4372         }
4373 out:
4374         reiserfs_check_lock_depth(sb, "journal end2");
4375
4376         memset(th, 0, sizeof(*th));
4377         /*
4378          * Re-set th->t_super, so we can properly keep track of how many
4379          * persistent transactions there are. We need to do this so if this
4380          * call is part of a failed restart_transaction, we can free it later
4381          */
4382         th->t_super = sb;
4383
4384         return journal->j_errno;
4385 }
4386
4387 /* Send the file system read only and refuse new transactions */
4388 void reiserfs_abort_journal(struct super_block *sb, int errno)
4389 {
4390         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4391         if (test_bit(J_ABORTED, &journal->j_state))
4392                 return;
4393
4394         if (!journal->j_errno)
4395                 journal->j_errno = errno;
4396
4397         sb->s_flags |= MS_RDONLY;
4398         set_bit(J_ABORTED, &journal->j_state);
4399
4400 #ifdef CONFIG_REISERFS_CHECK
4401         dump_stack();
4402 #endif
4403 }