fs/reiserfs/journal.c

   1 /*
   2 ** Write ahead logging implementation copyright Chris Mason 2000
   3 **
   4 ** The background commits make this code very interelated, 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 <asm/uaccess.h>
  38 #include <asm/system.h>
  39
  40 #include <linux/time.h>
  41 #include <asm/semaphore.h>
  42
  43 #include <linux/vmalloc.h>
  44 #include <linux/reiserfs_fs.h>
  45
  46 #include <linux/kernel.h>
  47 #include <linux/errno.h>
  48 #include <linux/fcntl.h>
  49 #include <linux/stat.h>
  50 #include <linux/string.h>
  51 #include <linux/smp_lock.h>
  52 #include <linux/buffer_head.h>
  53 #include <linux/workqueue.h>
  54 #include <linux/writeback.h>
  55 #include <linux/blkdev.h>
  56 #include <linux/backing-dev.h>
  57
  58 /* gets a struct reiserfs_journal_list * from a list head */
  59 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_list))
  61 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  62                                j_working_list))
  63
  64 /* the number of mounted filesystems.  This is used to decide when to
  65 ** start and kill the commit workqueue
  66 */
  67 static int reiserfs_mounted_fs_count;
  68
  69 static struct workqueue_struct *commit_wq;
  70
  71 #define JOURNAL_TRANS_HALF 1018 /* must be correct to keep the desc and commit
  72                                    structs at 4k */
  73 #define BUFNR 64                /*read ahead */
  74
  75 /* cnode stat bits.  Move these into reiserfs_fs.h */
  76
  77 #define BLOCK_FREED 2           /* this block was freed, and can't be written.  */
  78 #define BLOCK_FREED_HOLDER 3    /* this block was freed during this transaction, and can't be written */
  79
  80 #define BLOCK_NEEDS_FLUSH 4     /* used in flush_journal_list */
  81 #define BLOCK_DIRTIED 5
  82
  83 /* journal list state bits */
  84 #define LIST_TOUCHED 1
  85 #define LIST_DIRTY   2
  86 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  87
  88 /* flags for do_journal_end */
  89 #define FLUSH_ALL   1           /* flush commit and real blocks */
  90 #define COMMIT_NOW  2           /* end and commit this transaction */
  91 #define WAIT        4           /* wait for the log blocks to hit the disk */
  92
  93 static int do_journal_end(struct reiserfs_transaction_handle *,
  94                           struct super_block *, unsigned long nblocks,
  95                           int flags);
  96 static int flush_journal_list(struct super_block *s,
  97                               struct reiserfs_journal_list *jl, int flushall);
  98 static int flush_commit_list(struct super_block *s,
  99                              struct reiserfs_journal_list *jl, int flushall);
 100 static int can_dirty(struct reiserfs_journal_cnode *cn);
 101 static int journal_join(struct reiserfs_transaction_handle *th,
 102                         struct super_block *p_s_sb, unsigned long nblocks);
 103 static int release_journal_dev(struct super_block *super,
 104                                struct reiserfs_journal *journal);
 105 static int dirty_one_transaction(struct super_block *s,
 106                                  struct reiserfs_journal_list *jl);
 107 static void flush_async_commits(struct work_struct *work);
 108 static void queue_log_writer(struct super_block *s);
 109
 110 /* values for join in do_journal_begin_r */
 111 enum {
 112         JBEGIN_REG = 0,         /* regular journal begin */
 113         JBEGIN_JOIN = 1,        /* join the running transaction if at all possible */
 114         JBEGIN_ABORT = 2,       /* called from cleanup code, ignores aborted flag */
 115 };
 116
 117 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 118                               struct super_block *p_s_sb,
 119                               unsigned long nblocks, int join);
 120
 121 static void init_journal_hash(struct super_block *p_s_sb)
 122 {
 123         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 124         memset(journal->j_hash_table, 0,
 125                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 126 }
 127
 128 /*
 129 ** clears BH_Dirty and sticks the buffer on the clean list.  Called because I can't allow refile_buffer to
 130 ** make schedule happen after I've freed a block.  Look at remove_from_transaction and journal_mark_freed for
 131 ** more details.
 132 */
 133 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 134 {
 135         if (bh) {
 136                 clear_buffer_dirty(bh);
 137                 clear_buffer_journal_test(bh);
 138         }
 139         return 0;
 140 }
 141
 142 static void disable_barrier(struct super_block *s)
 143 {
 144         REISERFS_SB(s)->s_mount_opt &= ~(1 << REISERFS_BARRIER_FLUSH);
 145         printk("reiserfs: disabling flush barriers on %s\n",
 146                reiserfs_bdevname(s));
 147 }
 148
 149 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 150                                                          *p_s_sb)
 151 {
 152         struct reiserfs_bitmap_node *bn;
 153         static int id;
 154
 155         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 156         if (!bn) {
 157                 return NULL;
 158         }
 159         bn->data = kzalloc(p_s_sb->s_blocksize, GFP_NOFS);
 160         if (!bn->data) {
 161                 kfree(bn);
 162                 return NULL;
 163         }
 164         bn->id = id++;
 165         INIT_LIST_HEAD(&bn->list);
 166         return bn;
 167 }
 168
 169 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *p_s_sb)
 170 {
 171         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 172         struct reiserfs_bitmap_node *bn = NULL;
 173         struct list_head *entry = journal->j_bitmap_nodes.next;
 174
 175         journal->j_used_bitmap_nodes++;
 176       repeat:
 177
 178         if (entry != &journal->j_bitmap_nodes) {
 179                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 180                 list_del(entry);
 181                 memset(bn->data, 0, p_s_sb->s_blocksize);
 182                 journal->j_free_bitmap_nodes--;
 183                 return bn;
 184         }
 185         bn = allocate_bitmap_node(p_s_sb);
 186         if (!bn) {
 187                 yield();
 188                 goto repeat;
 189         }
 190         return bn;
 191 }
 192 static inline void free_bitmap_node(struct super_block *p_s_sb,
 193                                     struct reiserfs_bitmap_node *bn)
 194 {
 195         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 196         journal->j_used_bitmap_nodes--;
 197         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 198                 kfree(bn->data);
 199                 kfree(bn);
 200         } else {
 201                 list_add(&bn->list, &journal->j_bitmap_nodes);
 202                 journal->j_free_bitmap_nodes++;
 203         }
 204 }
 205
 206 static void allocate_bitmap_nodes(struct super_block *p_s_sb)
 207 {
 208         int i;
 209         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 210         struct reiserfs_bitmap_node *bn = NULL;
 211         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 212                 bn = allocate_bitmap_node(p_s_sb);
 213                 if (bn) {
 214                         list_add(&bn->list, &journal->j_bitmap_nodes);
 215                         journal->j_free_bitmap_nodes++;
 216                 } else {
 217                         break;  // this is ok, we'll try again when more are needed
 218                 }
 219         }
 220 }
 221
 222 static int set_bit_in_list_bitmap(struct super_block *p_s_sb, int block,
 223                                   struct reiserfs_list_bitmap *jb)
 224 {
 225         int bmap_nr = block / (p_s_sb->s_blocksize << 3);
 226         int bit_nr = block % (p_s_sb->s_blocksize << 3);
 227
 228         if (!jb->bitmaps[bmap_nr]) {
 229                 jb->bitmaps[bmap_nr] = get_bitmap_node(p_s_sb);
 230         }
 231         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 232         return 0;
 233 }
 234
 235 static void cleanup_bitmap_list(struct super_block *p_s_sb,
 236                                 struct reiserfs_list_bitmap *jb)
 237 {
 238         int i;
 239         if (jb->bitmaps == NULL)
 240                 return;
 241
 242         for (i = 0; i < SB_BMAP_NR(p_s_sb); i++) {
 243                 if (jb->bitmaps[i]) {
 244                         free_bitmap_node(p_s_sb, jb->bitmaps[i]);
 245                         jb->bitmaps[i] = NULL;
 246                 }
 247         }
 248 }
 249
 250 /*
 251 ** only call this on FS unmount.
 252 */
 253 static int free_list_bitmaps(struct super_block *p_s_sb,
 254                              struct reiserfs_list_bitmap *jb_array)
 255 {
 256         int i;
 257         struct reiserfs_list_bitmap *jb;
 258         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 259                 jb = jb_array + i;
 260                 jb->journal_list = NULL;
 261                 cleanup_bitmap_list(p_s_sb, jb);
 262                 vfree(jb->bitmaps);
 263                 jb->bitmaps = NULL;
 264         }
 265         return 0;
 266 }
 267
 268 static int free_bitmap_nodes(struct super_block *p_s_sb)
 269 {
 270         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 271         struct list_head *next = journal->j_bitmap_nodes.next;
 272         struct reiserfs_bitmap_node *bn;
 273
 274         while (next != &journal->j_bitmap_nodes) {
 275                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 276                 list_del(next);
 277                 kfree(bn->data);
 278                 kfree(bn);
 279                 next = journal->j_bitmap_nodes.next;
 280                 journal->j_free_bitmap_nodes--;
 281         }
 282
 283         return 0;
 284 }
 285
 286 /*
 287 ** get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 288 ** jb_array is the array to be filled in.
 289 */
 290 int reiserfs_allocate_list_bitmaps(struct super_block *p_s_sb,
 291                                    struct reiserfs_list_bitmap *jb_array,
 292                                    int bmap_nr)
 293 {
 294         int i;
 295         int failed = 0;
 296         struct reiserfs_list_bitmap *jb;
 297         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 298
 299         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 300                 jb = jb_array + i;
 301                 jb->journal_list = NULL;
 302                 jb->bitmaps = vmalloc(mem);
 303                 if (!jb->bitmaps) {
 304                         reiserfs_warning(p_s_sb,
 305                                          "clm-2000, unable to allocate bitmaps for journal lists");
 306                         failed = 1;
 307                         break;
 308                 }
 309                 memset(jb->bitmaps, 0, mem);
 310         }
 311         if (failed) {
 312                 free_list_bitmaps(p_s_sb, jb_array);
 313                 return -1;
 314         }
 315         return 0;
 316 }
 317
 318 /*
 319 ** find an available list bitmap.  If you can't find one, flush a commit list
 320 ** and try again
 321 */
 322 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *p_s_sb,
 323                                                     struct reiserfs_journal_list
 324                                                     *jl)
 325 {
 326         int i, j;
 327         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 328         struct reiserfs_list_bitmap *jb = NULL;
 329
 330         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 331                 i = journal->j_list_bitmap_index;
 332                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 333                 jb = journal->j_list_bitmap + i;
 334                 if (journal->j_list_bitmap[i].journal_list) {
 335                         flush_commit_list(p_s_sb,
 336                                           journal->j_list_bitmap[i].
 337                                           journal_list, 1);
 338                         if (!journal->j_list_bitmap[i].journal_list) {
 339                                 break;
 340                         }
 341                 } else {
 342                         break;
 343                 }
 344         }
 345         if (jb->journal_list) { /* double check to make sure if flushed correctly */
 346                 return NULL;
 347         }
 348         jb->journal_list = jl;
 349         return jb;
 350 }
 351
 352 /*
 353 ** allocates a new chunk of X nodes, and links them all together as a list.
 354 ** Uses the cnode->next and cnode->prev pointers
 355 ** returns NULL on failure
 356 */
 357 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 358 {
 359         struct reiserfs_journal_cnode *head;
 360         int i;
 361         if (num_cnodes <= 0) {
 362                 return NULL;
 363         }
 364         head = vmalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 365         if (!head) {
 366                 return NULL;
 367         }
 368         memset(head, 0, num_cnodes * sizeof(struct reiserfs_journal_cnode));
 369         head[0].prev = NULL;
 370         head[0].next = head + 1;
 371         for (i = 1; i < num_cnodes; i++) {
 372                 head[i].prev = head + (i - 1);
 373                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 374         }
 375         head[num_cnodes - 1].next = NULL;
 376         return head;
 377 }
 378
 379 /*
 380 ** pulls a cnode off the free list, or returns NULL on failure
 381 */
 382 static struct reiserfs_journal_cnode *get_cnode(struct super_block *p_s_sb)
 383 {
 384         struct reiserfs_journal_cnode *cn;
 385         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 386
 387         reiserfs_check_lock_depth(p_s_sb, "get_cnode");
 388
 389         if (journal->j_cnode_free <= 0) {
 390                 return NULL;
 391         }
 392         journal->j_cnode_used++;
 393         journal->j_cnode_free--;
 394         cn = journal->j_cnode_free_list;
 395         if (!cn) {
 396                 return cn;
 397         }
 398         if (cn->next) {
 399                 cn->next->prev = NULL;
 400         }
 401         journal->j_cnode_free_list = cn->next;
 402         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 403         return cn;
 404 }
 405
 406 /*
 407 ** returns a cnode to the free list
 408 */
 409 static void free_cnode(struct super_block *p_s_sb,
 410                        struct reiserfs_journal_cnode *cn)
 411 {
 412         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 413
 414         reiserfs_check_lock_depth(p_s_sb, "free_cnode");
 415
 416         journal->j_cnode_used--;
 417         journal->j_cnode_free++;
 418         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 419         cn->next = journal->j_cnode_free_list;
 420         if (journal->j_cnode_free_list) {
 421                 journal->j_cnode_free_list->prev = cn;
 422         }
 423         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 424         journal->j_cnode_free_list = cn;
 425 }
 426
 427 static void clear_prepared_bits(struct buffer_head *bh)
 428 {
 429         clear_buffer_journal_prepared(bh);
 430         clear_buffer_journal_restore_dirty(bh);
 431 }
 432
 433 /* utility function to force a BUG if it is called without the big
 434 ** kernel lock held.  caller is the string printed just before calling BUG()
 435 */
 436 void reiserfs_check_lock_depth(struct super_block *sb, char *caller)
 437 {
 438 #ifdef CONFIG_SMP
 439         if (current->lock_depth < 0) {
 440                 reiserfs_panic(sb, "%s called without kernel lock held",
 441                                caller);
 442         }
 443 #else
 444         ;
 445 #endif
 446 }
 447
 448 /* return a cnode with same dev, block number and size in table, or null if not found */
 449 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 450                                                                   super_block
 451                                                                   *sb,
 452                                                                   struct
 453                                                                   reiserfs_journal_cnode
 454                                                                   **table,
 455                                                                   long bl)
 456 {
 457         struct reiserfs_journal_cnode *cn;
 458         cn = journal_hash(table, sb, bl);
 459         while (cn) {
 460                 if (cn->blocknr == bl && cn->sb == sb)
 461                         return cn;
 462                 cn = cn->hnext;
 463         }
 464         return (struct reiserfs_journal_cnode *)0;
 465 }
 466
 467 /*
 468 ** this actually means 'can this block be reallocated yet?'.  If you set search_all, a block can only be allocated
 469 ** if it is not in the current transaction, was not freed by the current transaction, and has no chance of ever
 470 ** being overwritten by a replay after crashing.
 471 **
 472 ** If you don't set search_all, a block can only be allocated if it is not in the current transaction.  Since deleting
 473 ** a block removes it from the current transaction, this case should never happen.  If you don't set search_all, make
 474 ** sure you never write the block without logging it.
 475 **
 476 ** next_zero_bit is a suggestion about the next block to try for find_forward.
 477 ** when bl is rejected because it is set in a journal list bitmap, we search
 478 ** for the next zero bit in the bitmap that rejected bl.  Then, we return that
 479 ** through next_zero_bit for find_forward to try.
 480 **
 481 ** Just because we return something in next_zero_bit does not mean we won't
 482 ** reject it on the next call to reiserfs_in_journal
 483 **
 484 */
 485 int reiserfs_in_journal(struct super_block *p_s_sb,
 486                         int bmap_nr, int bit_nr, int search_all,
 487                         b_blocknr_t * next_zero_bit)
 488 {
 489         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
 490         struct reiserfs_journal_cnode *cn;
 491         struct reiserfs_list_bitmap *jb;
 492         int i;
 493         unsigned long bl;
 494
 495         *next_zero_bit = 0;     /* always start this at zero. */
 496
 497         PROC_INFO_INC(p_s_sb, journal.in_journal);
 498         /* If we aren't doing a search_all, this is a metablock, and it will be logged before use.
 499          ** if we crash before the transaction that freed it commits,  this transaction won't
 500          ** have committed either, and the block will never be written
 501          */
 502         if (search_all) {
 503                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 504                         PROC_INFO_INC(p_s_sb, journal.in_journal_bitmap);
 505                         jb = journal->j_list_bitmap + i;
 506                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 507                             test_bit(bit_nr,
 508                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 509                                      data)) {
 510                                 *next_zero_bit =
 511                                     find_next_zero_bit((unsigned long *)
 512                                                        (jb->bitmaps[bmap_nr]->
 513                                                         data),
 514                                                        p_s_sb->s_blocksize << 3,
 515                                                        bit_nr + 1);
 516                                 return 1;
 517                         }
 518                 }
 519         }
 520
 521         bl = bmap_nr * (p_s_sb->s_blocksize << 3) + bit_nr;
 522         /* is it in any old transactions? */
 523         if (search_all
 524             && (cn =
 525                 get_journal_hash_dev(p_s_sb, journal->j_list_hash_table, bl))) {
 526                 return 1;
 527         }
 528
 529         /* is it in the current transaction.  This should never happen */
 530         if ((cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, bl))) {
 531                 BUG();
 532                 return 1;
 533         }
 534
 535         PROC_INFO_INC(p_s_sb, journal.in_journal_reusable);
 536         /* safe for reuse */
 537         return 0;
 538 }
 539
 540 /* insert cn into table
 541 */
 542 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 543                                        struct reiserfs_journal_cnode *cn)
 544 {
 545         struct reiserfs_journal_cnode *cn_orig;
 546
 547         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 548         cn->hnext = cn_orig;
 549         cn->hprev = NULL;
 550         if (cn_orig) {
 551                 cn_orig->hprev = cn;
 552         }
 553         journal_hash(table, cn->sb, cn->blocknr) = cn;
 554 }
 555
 556 /* lock the current transaction */
 557 static inline void lock_journal(struct super_block *p_s_sb)
 558 {
 559         PROC_INFO_INC(p_s_sb, journal.lock_journal);
 560         down(&SB_JOURNAL(p_s_sb)->j_lock);
 561 }
 562
 563 /* unlock the current transaction */
 564 static inline void unlock_journal(struct super_block *p_s_sb)
 565 {
 566         up(&SB_JOURNAL(p_s_sb)->j_lock);
 567 }
 568
 569 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 570 {
 571         jl->j_refcount++;
 572 }
 573
 574 static inline void put_journal_list(struct super_block *s,
 575                                     struct reiserfs_journal_list *jl)
 576 {
 577         if (jl->j_refcount < 1) {
 578                 reiserfs_panic(s, "trans id %lu, refcount at %d",
 579                                jl->j_trans_id, jl->j_refcount);
 580         }
 581         if (--jl->j_refcount == 0)
 582                 kfree(jl);
 583 }
 584
 585 /*
 586 ** this used to be much more involved, and I'm keeping it just in case things get ugly again.
 587 ** it gets called by flush_commit_list, and cleans up any data stored about blocks freed during a
 588 ** transaction.
 589 */
 590 static void cleanup_freed_for_journal_list(struct super_block *p_s_sb,
 591                                            struct reiserfs_journal_list *jl)
 592 {
 593
 594         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 595         if (jb) {
 596                 cleanup_bitmap_list(p_s_sb, jb);
 597         }
 598         jl->j_list_bitmap->journal_list = NULL;
 599         jl->j_list_bitmap = NULL;
 600 }
 601
 602 static int journal_list_still_alive(struct super_block *s,
 603                                     unsigned long trans_id)
 604 {
 605         struct reiserfs_journal *journal = SB_JOURNAL(s);
 606         struct list_head *entry = &journal->j_journal_list;
 607         struct reiserfs_journal_list *jl;
 608
 609         if (!list_empty(entry)) {
 610                 jl = JOURNAL_LIST_ENTRY(entry->next);
 611                 if (jl->j_trans_id <= trans_id) {
 612                         return 1;
 613                 }
 614         }
 615         return 0;
 616 }
 617
 618 /*
 619  * If page->mapping was null, we failed to truncate this page for
 620  * some reason.  Most likely because it was truncated after being
 621  * logged via data=journal.
 622  *
 623  * This does a check to see if the buffer belongs to one of these
 624  * lost pages before doing the final put_bh.  If page->mapping was
 625  * null, it tries to free buffers on the page, which should make the
 626  * final page_cache_release drop the page from the lru.
 627  */
 628 static void release_buffer_page(struct buffer_head *bh)
 629 {
 630         struct page *page = bh->b_page;
 631         if (!page->mapping && !TestSetPageLocked(page)) {
 632                 page_cache_get(page);
 633                 put_bh(bh);
 634                 if (!page->mapping)
 635                         try_to_free_buffers(page);
 636                 unlock_page(page);
 637                 page_cache_release(page);
 638         } else {
 639                 put_bh(bh);
 640         }
 641 }
 642
 643 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 644 {
 645         char b[BDEVNAME_SIZE];
 646
 647         if (buffer_journaled(bh)) {
 648                 reiserfs_warning(NULL,
 649                                  "clm-2084: pinned buffer %lu:%s sent to disk",
 650                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 651         }
 652         if (uptodate)
 653                 set_buffer_uptodate(bh);
 654         else
 655                 clear_buffer_uptodate(bh);
 656
 657         unlock_buffer(bh);
 658         release_buffer_page(bh);
 659 }
 660
 661 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 662 {
 663         if (uptodate)
 664                 set_buffer_uptodate(bh);
 665         else
 666                 clear_buffer_uptodate(bh);
 667         unlock_buffer(bh);
 668         put_bh(bh);
 669 }
 670
 671 static void submit_logged_buffer(struct buffer_head *bh)
 672 {
 673         get_bh(bh);
 674         bh->b_end_io = reiserfs_end_buffer_io_sync;
 675         clear_buffer_journal_new(bh);
 676         clear_buffer_dirty(bh);
 677         if (!test_clear_buffer_journal_test(bh))
 678                 BUG();
 679         if (!buffer_uptodate(bh))
 680                 BUG();
 681         submit_bh(WRITE, bh);
 682 }
 683
 684 static void submit_ordered_buffer(struct buffer_head *bh)
 685 {
 686         get_bh(bh);
 687         bh->b_end_io = reiserfs_end_ordered_io;
 688         clear_buffer_dirty(bh);
 689         if (!buffer_uptodate(bh))
 690                 BUG();
 691         submit_bh(WRITE, bh);
 692 }
 693
 694 static int submit_barrier_buffer(struct buffer_head *bh)
 695 {
 696         get_bh(bh);
 697         bh->b_end_io = reiserfs_end_ordered_io;
 698         clear_buffer_dirty(bh);
 699         if (!buffer_uptodate(bh))
 700                 BUG();
 701         return submit_bh(WRITE_BARRIER, bh);
 702 }
 703
 704 static void check_barrier_completion(struct super_block *s,
 705                                      struct buffer_head *bh)
 706 {
 707         if (buffer_eopnotsupp(bh)) {
 708                 clear_buffer_eopnotsupp(bh);
 709                 disable_barrier(s);
 710                 set_buffer_uptodate(bh);
 711                 set_buffer_dirty(bh);
 712                 sync_dirty_buffer(bh);
 713         }
 714 }
 715
 716 #define CHUNK_SIZE 32
 717 struct buffer_chunk {
 718         struct buffer_head *bh[CHUNK_SIZE];
 719         int nr;
 720 };
 721
 722 static void write_chunk(struct buffer_chunk *chunk)
 723 {
 724         int i;
 725         get_fs_excl();
 726         for (i = 0; i < chunk->nr; i++) {
 727                 submit_logged_buffer(chunk->bh[i]);
 728         }
 729         chunk->nr = 0;
 730         put_fs_excl();
 731 }
 732
 733 static void write_ordered_chunk(struct buffer_chunk *chunk)
 734 {
 735         int i;
 736         get_fs_excl();
 737         for (i = 0; i < chunk->nr; i++) {
 738                 submit_ordered_buffer(chunk->bh[i]);
 739         }
 740         chunk->nr = 0;
 741         put_fs_excl();
 742 }
 743
 744 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 745                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 746 {
 747         int ret = 0;
 748         BUG_ON(chunk->nr >= CHUNK_SIZE);
 749         chunk->bh[chunk->nr++] = bh;
 750         if (chunk->nr >= CHUNK_SIZE) {
 751                 ret = 1;
 752                 if (lock)
 753                         spin_unlock(lock);
 754                 fn(chunk);
 755                 if (lock)
 756                         spin_lock(lock);
 757         }
 758         return ret;
 759 }
 760
 761 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 762 static struct reiserfs_jh *alloc_jh(void)
 763 {
 764         struct reiserfs_jh *jh;
 765         while (1) {
 766                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 767                 if (jh) {
 768                         atomic_inc(&nr_reiserfs_jh);
 769                         return jh;
 770                 }
 771                 yield();
 772         }
 773 }
 774
 775 /*
 776  * we want to free the jh when the buffer has been written
 777  * and waited on
 778  */
 779 void reiserfs_free_jh(struct buffer_head *bh)
 780 {
 781         struct reiserfs_jh *jh;
 782
 783         jh = bh->b_private;
 784         if (jh) {
 785                 bh->b_private = NULL;
 786                 jh->bh = NULL;
 787                 list_del_init(&jh->list);
 788                 kfree(jh);
 789                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 790                         BUG();
 791                 atomic_dec(&nr_reiserfs_jh);
 792                 put_bh(bh);
 793         }
 794 }
 795
 796 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 797                            int tail)
 798 {
 799         struct reiserfs_jh *jh;
 800
 801         if (bh->b_private) {
 802                 spin_lock(&j->j_dirty_buffers_lock);
 803                 if (!bh->b_private) {
 804                         spin_unlock(&j->j_dirty_buffers_lock);
 805                         goto no_jh;
 806                 }
 807                 jh = bh->b_private;
 808                 list_del_init(&jh->list);
 809         } else {
 810               no_jh:
 811                 get_bh(bh);
 812                 jh = alloc_jh();
 813                 spin_lock(&j->j_dirty_buffers_lock);
 814                 /* buffer must be locked for __add_jh, should be able to have
 815                  * two adds at the same time
 816                  */
 817                 BUG_ON(bh->b_private);
 818                 jh->bh = bh;
 819                 bh->b_private = jh;
 820         }
 821         jh->jl = j->j_current_jl;
 822         if (tail)
 823                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 824         else {
 825                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 826         }
 827         spin_unlock(&j->j_dirty_buffers_lock);
 828         return 0;
 829 }
 830
 831 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 832 {
 833         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 834 }
 835 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 836 {
 837         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 838 }
 839
 840 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 841 static int write_ordered_buffers(spinlock_t * lock,
 842                                  struct reiserfs_journal *j,
 843                                  struct reiserfs_journal_list *jl,
 844                                  struct list_head *list)
 845 {
 846         struct buffer_head *bh;
 847         struct reiserfs_jh *jh;
 848         int ret = j->j_errno;
 849         struct buffer_chunk chunk;
 850         struct list_head tmp;
 851         INIT_LIST_HEAD(&tmp);
 852
 853         chunk.nr = 0;
 854         spin_lock(lock);
 855         while (!list_empty(list)) {
 856                 jh = JH_ENTRY(list->next);
 857                 bh = jh->bh;
 858                 get_bh(bh);
 859                 if (test_set_buffer_locked(bh)) {
 860                         if (!buffer_dirty(bh)) {
 861                                 list_move(&jh->list, &tmp);
 862                                 goto loop_next;
 863                         }
 864                         spin_unlock(lock);
 865                         if (chunk.nr)
 866                                 write_ordered_chunk(&chunk);
 867                         wait_on_buffer(bh);
 868                         cond_resched();
 869                         spin_lock(lock);
 870                         goto loop_next;
 871                 }
 872                 /* in theory, dirty non-uptodate buffers should never get here,
 873                  * but the upper layer io error paths still have a few quirks.
 874                  * Handle them here as gracefully as we can
 875                  */
 876                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 877                         clear_buffer_dirty(bh);
 878                         ret = -EIO;
 879                 }
 880                 if (buffer_dirty(bh)) {
 881                         list_move(&jh->list, &tmp);
 882                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 883                 } else {
 884                         reiserfs_free_jh(bh);
 885                         unlock_buffer(bh);
 886                 }
 887               loop_next:
 888                 put_bh(bh);
 889                 cond_resched_lock(lock);
 890         }
 891         if (chunk.nr) {
 892                 spin_unlock(lock);
 893                 write_ordered_chunk(&chunk);
 894                 spin_lock(lock);
 895         }
 896         while (!list_empty(&tmp)) {
 897                 jh = JH_ENTRY(tmp.prev);
 898                 bh = jh->bh;
 899                 get_bh(bh);
 900                 reiserfs_free_jh(bh);
 901
 902                 if (buffer_locked(bh)) {
 903                         spin_unlock(lock);
 904                         wait_on_buffer(bh);
 905                         spin_lock(lock);
 906                 }
 907                 if (!buffer_uptodate(bh)) {
 908                         ret = -EIO;
 909                 }
 910                 /* ugly interaction with invalidatepage here.
 911                  * reiserfs_invalidate_page will pin any buffer that has a valid
 912                  * journal head from an older transaction.  If someone else sets
 913                  * our buffer dirty after we write it in the first loop, and
 914                  * then someone truncates the page away, nobody will ever write
 915                  * the buffer. We're safe if we write the page one last time
 916                  * after freeing the journal header.
 917                  */
 918                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 919                         spin_unlock(lock);
 920                         ll_rw_block(WRITE, 1, &bh);
 921                         spin_lock(lock);
 922                 }
 923                 put_bh(bh);
 924                 cond_resched_lock(lock);
 925         }
 926         spin_unlock(lock);
 927         return ret;
 928 }
 929
 930 static int flush_older_commits(struct super_block *s,
 931                                struct reiserfs_journal_list *jl)
 932 {
 933         struct reiserfs_journal *journal = SB_JOURNAL(s);
 934         struct reiserfs_journal_list *other_jl;
 935         struct reiserfs_journal_list *first_jl;
 936         struct list_head *entry;
 937         unsigned long trans_id = jl->j_trans_id;
 938         unsigned long other_trans_id;
 939         unsigned long first_trans_id;
 940
 941       find_first:
 942         /*
 943          * first we walk backwards to find the oldest uncommitted transation
 944          */
 945         first_jl = jl;
 946         entry = jl->j_list.prev;
 947         while (1) {
 948                 other_jl = JOURNAL_LIST_ENTRY(entry);
 949                 if (entry == &journal->j_journal_list ||
 950                     atomic_read(&other_jl->j_older_commits_done))
 951                         break;
 952
 953                 first_jl = other_jl;
 954                 entry = other_jl->j_list.prev;
 955         }
 956
 957         /* if we didn't find any older uncommitted transactions, return now */
 958         if (first_jl == jl) {
 959                 return 0;
 960         }
 961
 962         first_trans_id = first_jl->j_trans_id;
 963
 964         entry = &first_jl->j_list;
 965         while (1) {
 966                 other_jl = JOURNAL_LIST_ENTRY(entry);
 967                 other_trans_id = other_jl->j_trans_id;
 968
 969                 if (other_trans_id < trans_id) {
 970                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 971                                 flush_commit_list(s, other_jl, 0);
 972
 973                                 /* list we were called with is gone, return */
 974                                 if (!journal_list_still_alive(s, trans_id))
 975                                         return 1;
 976
 977                                 /* the one we just flushed is gone, this means all
 978                                  * older lists are also gone, so first_jl is no longer
 979                                  * valid either.  Go back to the beginning.
 980                                  */
 981                                 if (!journal_list_still_alive
 982                                     (s, other_trans_id)) {
 983                                         goto find_first;
 984                                 }
 985                         }
 986                         entry = entry->next;
 987                         if (entry == &journal->j_journal_list)
 988                                 return 0;
 989                 } else {
 990                         return 0;
 991                 }
 992         }
 993         return 0;
 994 }
 995
 996 static int reiserfs_async_progress_wait(struct super_block *s)
 997 {
 998         DEFINE_WAIT(wait);
 999         struct reiserfs_journal *j = SB_JOURNAL(s);
1000         if (atomic_read(&j->j_async_throttle))
1001                 congestion_wait(WRITE, HZ / 10);
1002         return 0;
1003 }
1004
1005 /*
1006 ** if this journal list still has commit blocks unflushed, send them to disk.
1007 **
1008 ** log areas must be flushed in order (transaction 2 can't commit before transaction 1)
1009 ** Before the commit block can by written, every other log block must be safely on disk
1010 **
1011 */
1012 static int flush_commit_list(struct super_block *s,
1013                              struct reiserfs_journal_list *jl, int flushall)
1014 {
1015         int i;
1016         int bn;
1017         struct buffer_head *tbh = NULL;
1018         unsigned long trans_id = jl->j_trans_id;
1019         struct reiserfs_journal *journal = SB_JOURNAL(s);
1020         int barrier = 0;
1021         int retval = 0;
1022         int write_len;
1023
1024         reiserfs_check_lock_depth(s, "flush_commit_list");
1025
1026         if (atomic_read(&jl->j_older_commits_done)) {
1027                 return 0;
1028         }
1029
1030         get_fs_excl();
1031
1032         /* before we can put our commit blocks on disk, we have to make sure everyone older than
1033          ** us is on disk too
1034          */
1035         BUG_ON(jl->j_len <= 0);
1036         BUG_ON(trans_id == journal->j_trans_id);
1037
1038         get_journal_list(jl);
1039         if (flushall) {
1040                 if (flush_older_commits(s, jl) == 1) {
1041                         /* list disappeared during flush_older_commits.  return */
1042                         goto put_jl;
1043                 }
1044         }
1045
1046         /* make sure nobody is trying to flush this one at the same time */
1047         down(&jl->j_commit_lock);
1048         if (!journal_list_still_alive(s, trans_id)) {
1049                 up(&jl->j_commit_lock);
1050                 goto put_jl;
1051         }
1052         BUG_ON(jl->j_trans_id == 0);
1053
1054         /* this commit is done, exit */
1055         if (atomic_read(&(jl->j_commit_left)) <= 0) {
1056                 if (flushall) {
1057                         atomic_set(&(jl->j_older_commits_done), 1);
1058                 }
1059                 up(&jl->j_commit_lock);
1060                 goto put_jl;
1061         }
1062
1063         if (!list_empty(&jl->j_bh_list)) {
1064                 int ret;
1065                 unlock_kernel();
1066                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1067                                             journal, jl, &jl->j_bh_list);
1068                 if (ret < 0 && retval == 0)
1069                         retval = ret;
1070                 lock_kernel();
1071         }
1072         BUG_ON(!list_empty(&jl->j_bh_list));
1073         /*
1074          * for the description block and all the log blocks, submit any buffers
1075          * that haven't already reached the disk.  Try to write at least 256
1076          * log blocks. later on, we will only wait on blocks that correspond
1077          * to this transaction, but while we're unplugging we might as well
1078          * get a chunk of data on there.
1079          */
1080         atomic_inc(&journal->j_async_throttle);
1081         write_len = jl->j_len + 1;
1082         if (write_len < 256)
1083                 write_len = 256;
1084         for (i = 0 ; i < write_len ; i++) {
1085                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1086                     SB_ONDISK_JOURNAL_SIZE(s);
1087                 tbh = journal_find_get_block(s, bn);
1088                 if (tbh) {
1089                         if (buffer_dirty(tbh))
1090                             ll_rw_block(WRITE, 1, &tbh) ;
1091                         put_bh(tbh) ;
1092                 }
1093         }
1094         atomic_dec(&journal->j_async_throttle);
1095
1096         /* We're skipping the commit if there's an error */
1097         if (retval || reiserfs_is_journal_aborted(journal))
1098                 barrier = 0;
1099
1100         /* wait on everything written so far before writing the commit
1101          * if we are in barrier mode, send the commit down now
1102          */
1103         barrier = reiserfs_barrier_flush(s);
1104         if (barrier) {
1105                 int ret;
1106                 lock_buffer(jl->j_commit_bh);
1107                 ret = submit_barrier_buffer(jl->j_commit_bh);
1108                 if (ret == -EOPNOTSUPP) {
1109                         set_buffer_uptodate(jl->j_commit_bh);
1110                         disable_barrier(s);
1111                         barrier = 0;
1112                 }
1113         }
1114         for (i = 0; i < (jl->j_len + 1); i++) {
1115                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1116                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1117                 tbh = journal_find_get_block(s, bn);
1118                 wait_on_buffer(tbh);
1119                 // since we're using ll_rw_blk above, it might have skipped over
1120                 // a locked buffer.  Double check here
1121                 //
1122                 if (buffer_dirty(tbh))  /* redundant, sync_dirty_buffer() checks */
1123                         sync_dirty_buffer(tbh);
1124                 if (unlikely(!buffer_uptodate(tbh))) {
1125 #ifdef CONFIG_REISERFS_CHECK
1126                         reiserfs_warning(s, "journal-601, buffer write failed");
1127 #endif
1128                         retval = -EIO;
1129                 }
1130                 put_bh(tbh);    /* once for journal_find_get_block */
1131                 put_bh(tbh);    /* once due to original getblk in do_journal_end */
1132                 atomic_dec(&(jl->j_commit_left));
1133         }
1134
1135         BUG_ON(atomic_read(&(jl->j_commit_left)) != 1);
1136
1137         if (!barrier) {
1138                 /* If there was a write error in the journal - we can't commit
1139                  * this transaction - it will be invalid and, if successful,
1140                  * will just end up propagating the write error out to
1141                  * the file system. */
1142                 if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1143                         if (buffer_dirty(jl->j_commit_bh))
1144                                 BUG();
1145                         mark_buffer_dirty(jl->j_commit_bh) ;
1146                         sync_dirty_buffer(jl->j_commit_bh) ;
1147                 }
1148         } else
1149                 wait_on_buffer(jl->j_commit_bh);
1150
1151         check_barrier_completion(s, jl->j_commit_bh);
1152
1153         /* If there was a write error in the journal - we can't commit this
1154          * transaction - it will be invalid and, if successful, will just end
1155          * up propagating the write error out to the filesystem. */
1156         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1157 #ifdef CONFIG_REISERFS_CHECK
1158                 reiserfs_warning(s, "journal-615: buffer write failed");
1159 #endif
1160                 retval = -EIO;
1161         }
1162         bforget(jl->j_commit_bh);
1163         if (journal->j_last_commit_id != 0 &&
1164             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1165                 reiserfs_warning(s, "clm-2200: last commit %lu, current %lu",
1166                                  journal->j_last_commit_id, jl->j_trans_id);
1167         }
1168         journal->j_last_commit_id = jl->j_trans_id;
1169
1170         /* now, every commit block is on the disk.  It is safe to allow blocks freed during this transaction to be reallocated */
1171         cleanup_freed_for_journal_list(s, jl);
1172
1173         retval = retval ? retval : journal->j_errno;
1174
1175         /* mark the metadata dirty */
1176         if (!retval)
1177                 dirty_one_transaction(s, jl);
1178         atomic_dec(&(jl->j_commit_left));
1179
1180         if (flushall) {
1181                 atomic_set(&(jl->j_older_commits_done), 1);
1182         }
1183         up(&jl->j_commit_lock);
1184       put_jl:
1185         put_journal_list(s, jl);
1186
1187         if (retval)
1188                 reiserfs_abort(s, retval, "Journal write error in %s",
1189                                __FUNCTION__);
1190         put_fs_excl();
1191         return retval;
1192 }
1193
1194 /*
1195 ** flush_journal_list frequently needs to find a newer transaction for a given block.  This does that, or
1196 ** returns NULL if it can't find anything
1197 */
1198 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1199                                                           reiserfs_journal_cnode
1200                                                           *cn)
1201 {
1202         struct super_block *sb = cn->sb;
1203         b_blocknr_t blocknr = cn->blocknr;
1204
1205         cn = cn->hprev;
1206         while (cn) {
1207                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1208                         return cn->jlist;
1209                 }
1210                 cn = cn->hprev;
1211         }
1212         return NULL;
1213 }
1214
1215 static int newer_jl_done(struct reiserfs_journal_cnode *cn)
1216 {
1217         struct super_block *sb = cn->sb;
1218         b_blocknr_t blocknr = cn->blocknr;
1219
1220         cn = cn->hprev;
1221         while (cn) {
1222                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist &&
1223                     atomic_read(&cn->jlist->j_commit_left) != 0)
1224                                     return 0;
1225                 cn = cn->hprev;
1226         }
1227         return 1;
1228 }
1229
1230 static void remove_journal_hash(struct super_block *,
1231                                 struct reiserfs_journal_cnode **,
1232                                 struct reiserfs_journal_list *, unsigned long,
1233                                 int);
1234
1235 /*
1236 ** once all the real blocks have been flushed, it is safe to remove them from the
1237 ** journal list for this transaction.  Aside from freeing the cnode, this also allows the
1238 ** block to be reallocated for data blocks if it had been deleted.
1239 */
1240 static void remove_all_from_journal_list(struct super_block *p_s_sb,
1241                                          struct reiserfs_journal_list *jl,
1242                                          int debug)
1243 {
1244         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1245         struct reiserfs_journal_cnode *cn, *last;
1246         cn = jl->j_realblock;
1247
1248         /* which is better, to lock once around the whole loop, or
1249          ** to lock for each call to remove_journal_hash?
1250          */
1251         while (cn) {
1252                 if (cn->blocknr != 0) {
1253                         if (debug) {
1254                                 reiserfs_warning(p_s_sb,
1255                                                  "block %u, bh is %d, state %ld",
1256                                                  cn->blocknr, cn->bh ? 1 : 0,
1257                                                  cn->state);
1258                         }
1259                         cn->state = 0;
1260                         remove_journal_hash(p_s_sb, journal->j_list_hash_table,
1261                                             jl, cn->blocknr, 1);
1262                 }
1263                 last = cn;
1264                 cn = cn->next;
1265                 free_cnode(p_s_sb, last);
1266         }
1267         jl->j_realblock = NULL;
1268 }
1269
1270 /*
1271 ** if this timestamp is greater than the timestamp we wrote last to the header block, write it to the header block.
1272 ** once this is done, I can safely say the log area for this transaction won't ever be replayed, and I can start
1273 ** releasing blocks in this transaction for reuse as data blocks.
1274 ** called by flush_journal_list, before it calls remove_all_from_journal_list
1275 **
1276 */
1277 static int _update_journal_header_block(struct super_block *p_s_sb,
1278                                         unsigned long offset,
1279                                         unsigned long trans_id)
1280 {
1281         struct reiserfs_journal_header *jh;
1282         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1283
1284         if (reiserfs_is_journal_aborted(journal))
1285                 return -EIO;
1286
1287         if (trans_id >= journal->j_last_flush_trans_id) {
1288                 if (buffer_locked((journal->j_header_bh))) {
1289                         wait_on_buffer((journal->j_header_bh));
1290                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1291 #ifdef CONFIG_REISERFS_CHECK
1292                                 reiserfs_warning(p_s_sb,
1293                                                  "journal-699: buffer write failed");
1294 #endif
1295                                 return -EIO;
1296                         }
1297                 }
1298                 journal->j_last_flush_trans_id = trans_id;
1299                 journal->j_first_unflushed_offset = offset;
1300                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1301                                                         b_data);
1302                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1303                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1304                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1305
1306                 if (reiserfs_barrier_flush(p_s_sb)) {
1307                         int ret;
1308                         lock_buffer(journal->j_header_bh);
1309                         ret = submit_barrier_buffer(journal->j_header_bh);
1310                         if (ret == -EOPNOTSUPP) {
1311                                 set_buffer_uptodate(journal->j_header_bh);
1312                                 disable_barrier(p_s_sb);
1313                                 goto sync;
1314                         }
1315                         wait_on_buffer(journal->j_header_bh);
1316                         check_barrier_completion(p_s_sb, journal->j_header_bh);
1317                 } else {
1318                       sync:
1319                         set_buffer_dirty(journal->j_header_bh);
1320                         sync_dirty_buffer(journal->j_header_bh);
1321                 }
1322                 if (!buffer_uptodate(journal->j_header_bh)) {
1323                         reiserfs_warning(p_s_sb,
1324                                          "journal-837: IO error during journal replay");
1325                         return -EIO;
1326                 }
1327         }
1328         return 0;
1329 }
1330
1331 static int update_journal_header_block(struct super_block *p_s_sb,
1332                                        unsigned long offset,
1333                                        unsigned long trans_id)
1334 {
1335         return _update_journal_header_block(p_s_sb, offset, trans_id);
1336 }
1337
1338 /*
1339 ** flush any and all journal lists older than you are
1340 ** can only be called from flush_journal_list
1341 */
1342 static int flush_older_journal_lists(struct super_block *p_s_sb,
1343                                      struct reiserfs_journal_list *jl)
1344 {
1345         struct list_head *entry;
1346         struct reiserfs_journal_list *other_jl;
1347         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1348         unsigned long trans_id = jl->j_trans_id;
1349
1350         /* we know we are the only ones flushing things, no extra race
1351          * protection is required.
1352          */
1353       restart:
1354         entry = journal->j_journal_list.next;
1355         /* Did we wrap? */
1356         if (entry == &journal->j_journal_list)
1357                 return 0;
1358         other_jl = JOURNAL_LIST_ENTRY(entry);
1359         if (other_jl->j_trans_id < trans_id) {
1360                 BUG_ON(other_jl->j_refcount <= 0);
1361                 /* do not flush all */
1362                 flush_journal_list(p_s_sb, other_jl, 0);
1363
1364                 /* other_jl is now deleted from the list */
1365                 goto restart;
1366         }
1367         return 0;
1368 }
1369
1370 static void del_from_work_list(struct super_block *s,
1371                                struct reiserfs_journal_list *jl)
1372 {
1373         struct reiserfs_journal *journal = SB_JOURNAL(s);
1374         if (!list_empty(&jl->j_working_list)) {
1375                 list_del_init(&jl->j_working_list);
1376                 journal->j_num_work_lists--;
1377         }
1378 }
1379
1380 /* flush a journal list, both commit and real blocks
1381 **
1382 ** always set flushall to 1, unless you are calling from inside
1383 ** flush_journal_list
1384 **
1385 ** IMPORTANT.  This can only be called while there are no journal writers,
1386 ** and the journal is locked.  That means it can only be called from
1387 ** do_journal_end, or by journal_release
1388 */
1389 static int flush_journal_list(struct super_block *s,
1390                               struct reiserfs_journal_list *jl, int flushall)
1391 {
1392         struct reiserfs_journal_list *pjl;
1393         struct reiserfs_journal_cnode *cn, *last;
1394         int count;
1395         int was_jwait = 0;
1396         int was_dirty = 0;
1397         struct buffer_head *saved_bh;
1398         unsigned long j_len_saved = jl->j_len;
1399         struct reiserfs_journal *journal = SB_JOURNAL(s);
1400         int err = 0;
1401
1402         BUG_ON(j_len_saved <= 0);
1403
1404         if (atomic_read(&journal->j_wcount) != 0) {
1405                 reiserfs_warning(s,
1406                                  "clm-2048: flush_journal_list called with wcount %d",
1407                                  atomic_read(&journal->j_wcount));
1408         }
1409         BUG_ON(jl->j_trans_id == 0);
1410
1411         /* if flushall == 0, the lock is already held */
1412         if (flushall) {
1413                 down(&journal->j_flush_sem);
1414         } else if (!down_trylock(&journal->j_flush_sem)) {
1415                 BUG();
1416         }
1417
1418         count = 0;
1419         if (j_len_saved > journal->j_trans_max) {
1420                 reiserfs_panic(s,
1421                                "journal-715: flush_journal_list, length is %lu, trans id %lu\n",
1422                                j_len_saved, jl->j_trans_id);
1423                 return 0;
1424         }
1425
1426         get_fs_excl();
1427
1428         /* if all the work is already done, get out of here */
1429         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1430             atomic_read(&(jl->j_commit_left)) <= 0) {
1431                 goto flush_older_and_return;
1432         }
1433
1434         /* start by putting the commit list on disk.  This will also flush
1435          ** the commit lists of any olders transactions
1436          */
1437         flush_commit_list(s, jl, 1);
1438
1439         if (!(jl->j_state & LIST_DIRTY)
1440             && !reiserfs_is_journal_aborted(journal))
1441                 BUG();
1442
1443         /* are we done now? */
1444         if (atomic_read(&(jl->j_nonzerolen)) <= 0 &&
1445             atomic_read(&(jl->j_commit_left)) <= 0) {
1446                 goto flush_older_and_return;
1447         }
1448
1449         /* loop through each cnode, see if we need to write it,
1450          ** or wait on a more recent transaction, or just ignore it
1451          */
1452         if (atomic_read(&(journal->j_wcount)) != 0) {
1453                 reiserfs_panic(s,
1454                                "journal-844: panic journal list is flushing, wcount is not 0\n");
1455         }
1456         cn = jl->j_realblock;
1457         while (cn) {
1458                 was_jwait = 0;
1459                 was_dirty = 0;
1460                 saved_bh = NULL;
1461                 /* blocknr of 0 is no longer in the hash, ignore it */
1462                 if (cn->blocknr == 0) {
1463                         goto free_cnode;
1464                 }
1465
1466                 /* This transaction failed commit. Don't write out to the disk */
1467                 if (!(jl->j_state & LIST_DIRTY))
1468                         goto free_cnode;
1469
1470                 pjl = find_newer_jl_for_cn(cn);
1471                 /* the order is important here.  We check pjl to make sure we
1472                  ** don't clear BH_JDirty_wait if we aren't the one writing this
1473                  ** block to disk
1474                  */
1475                 if (!pjl && cn->bh) {
1476                         saved_bh = cn->bh;
1477
1478                         /* we do this to make sure nobody releases the buffer while
1479                          ** we are working with it
1480                          */
1481                         get_bh(saved_bh);
1482
1483                         if (buffer_journal_dirty(saved_bh)) {
1484                                 BUG_ON(!can_dirty(cn));
1485                                 was_jwait = 1;
1486                                 was_dirty = 1;
1487                         } else if (can_dirty(cn)) {
1488                                 /* everything with !pjl && jwait should be writable */
1489                                 BUG();
1490                         }
1491                 }
1492
1493                 /* if someone has this block in a newer transaction, just make
1494                  ** sure they are committed, and don't try writing it to disk
1495                  */
1496                 if (pjl) {
1497                         if (atomic_read(&pjl->j_commit_left))
1498                                 flush_commit_list(s, pjl, 1);
1499                         goto free_cnode;
1500                 }
1501
1502                 /* bh == NULL when the block got to disk on its own, OR,
1503                  ** the block got freed in a future transaction
1504                  */
1505                 if (saved_bh == NULL) {
1506                         goto free_cnode;
1507                 }
1508
1509                 /* this should never happen.  kupdate_one_transaction has this list
1510                  ** locked while it works, so we should never see a buffer here that
1511                  ** is not marked JDirty_wait
1512                  */
1513                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1514                         reiserfs_warning(s,
1515                                          "journal-813: BAD! buffer %llu %cdirty %cjwait, "
1516                                          "not in a newer tranasction",
1517                                          (unsigned long long)saved_bh->
1518                                          b_blocknr, was_dirty ? ' ' : '!',
1519                                          was_jwait ? ' ' : '!');
1520                 }
1521                 if (was_dirty) {
1522                         /* we inc again because saved_bh gets decremented at free_cnode */
1523                         get_bh(saved_bh);
1524                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1525                         lock_buffer(saved_bh);
1526                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1527                         if (buffer_dirty(saved_bh))
1528                                 submit_logged_buffer(saved_bh);
1529                         else
1530                                 unlock_buffer(saved_bh);
1531                         count++;
1532                 } else {
1533                         reiserfs_warning(s,
1534                                          "clm-2082: Unable to flush buffer %llu in %s",
1535                                          (unsigned long long)saved_bh->
1536                                          b_blocknr, __FUNCTION__);
1537                 }
1538               free_cnode:
1539                 last = cn;
1540                 cn = cn->next;
1541                 if (saved_bh) {
1542                         /* we incremented this to keep others from taking the buffer head away */
1543                         put_bh(saved_bh);
1544                         if (atomic_read(&(saved_bh->b_count)) < 0) {
1545                                 reiserfs_warning(s,
1546                                                  "journal-945: saved_bh->b_count < 0");
1547                         }
1548                 }
1549         }
1550         if (count > 0) {
1551                 cn = jl->j_realblock;
1552                 while (cn) {
1553                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1554                                 if (!cn->bh) {
1555                                         reiserfs_panic(s,
1556                                                        "journal-1011: cn->bh is NULL\n");
1557                                 }
1558                                 wait_on_buffer(cn->bh);
1559                                 if (!cn->bh) {
1560                                         reiserfs_panic(s,
1561                                                        "journal-1012: cn->bh is NULL\n");
1562                                 }
1563                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1564 #ifdef CONFIG_REISERFS_CHECK
1565                                         reiserfs_warning(s,
1566                                                          "journal-949: buffer write failed\n");
1567 #endif
1568                                         err = -EIO;
1569                                 }
1570                                 /* note, we must clear the JDirty_wait bit after the up to date
1571                                  ** check, otherwise we race against our flushpage routine
1572                                  */
1573                                 BUG_ON(!test_clear_buffer_journal_dirty
1574                                        (cn->bh));
1575
1576                                 /* drop one ref for us */
1577                                 put_bh(cn->bh);
1578                                 /* drop one ref for journal_mark_dirty */
1579                                 release_buffer_page(cn->bh);
1580                         }
1581                         cn = cn->next;
1582                 }
1583         }
1584
1585         if (err)
1586                 reiserfs_abort(s, -EIO,
1587                                "Write error while pushing transaction to disk in %s",
1588                                __FUNCTION__);
1589       flush_older_and_return:
1590
1591         /* before we can update the journal header block, we _must_ flush all
1592          ** real blocks from all older transactions to disk.  This is because
1593          ** once the header block is updated, this transaction will not be
1594          ** replayed after a crash
1595          */
1596         if (flushall) {
1597                 flush_older_journal_lists(s, jl);
1598         }
1599
1600         err = journal->j_errno;
1601         /* before we can remove everything from the hash tables for this
1602          ** transaction, we must make sure it can never be replayed
1603          **
1604          ** since we are only called from do_journal_end, we know for sure there
1605          ** are no allocations going on while we are flushing journal lists.  So,
1606          ** we only need to update the journal header block for the last list
1607          ** being flushed
1608          */
1609         if (!err && flushall) {
1610                 err =
1611                     update_journal_header_block(s,
1612                                                 (jl->j_start + jl->j_len +
1613                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1614                                                 jl->j_trans_id);
1615                 if (err)
1616                         reiserfs_abort(s, -EIO,
1617                                        "Write error while updating journal header in %s",
1618                                        __FUNCTION__);
1619         }
1620         remove_all_from_journal_list(s, jl, 0);
1621         list_del_init(&jl->j_list);
1622         journal->j_num_lists--;
1623         del_from_work_list(s, jl);
1624
1625         if (journal->j_last_flush_id != 0 &&
1626             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1627                 reiserfs_warning(s, "clm-2201: last flush %lu, current %lu",
1628                                  journal->j_last_flush_id, jl->j_trans_id);
1629         }
1630         journal->j_last_flush_id = jl->j_trans_id;
1631
1632         /* not strictly required since we are freeing the list, but it should
1633          * help find code using dead lists later on
1634          */
1635         jl->j_len = 0;
1636         atomic_set(&(jl->j_nonzerolen), 0);
1637         jl->j_start = 0;
1638         jl->j_realblock = NULL;
1639         jl->j_commit_bh = NULL;
1640         jl->j_trans_id = 0;
1641         jl->j_state = 0;
1642         put_journal_list(s, jl);
1643         if (flushall)
1644                 up(&journal->j_flush_sem);
1645         put_fs_excl();
1646         return err;
1647 }
1648
1649 static int test_transaction(struct super_block *s,
1650                             struct reiserfs_journal_list *jl)
1651 {
1652         struct reiserfs_journal_cnode *cn;
1653
1654         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0)
1655                 return 1;
1656
1657         cn = jl->j_realblock;
1658         while (cn) {
1659                 /* if the blocknr == 0, this has been cleared from the hash,
1660                  ** skip it
1661                  */
1662                 if (cn->blocknr == 0) {
1663                         goto next;
1664                 }
1665                 if (cn->bh && !newer_jl_done(cn))
1666                         return 0;
1667               next:
1668                 cn = cn->next;
1669                 cond_resched();
1670         }
1671         return 0;
1672 }
1673
1674 static int write_one_transaction(struct super_block *s,
1675                                  struct reiserfs_journal_list *jl,
1676                                  struct buffer_chunk *chunk)
1677 {
1678         struct reiserfs_journal_cnode *cn;
1679         int ret = 0;
1680
1681         jl->j_state |= LIST_TOUCHED;
1682         del_from_work_list(s, jl);
1683         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1684                 return 0;
1685         }
1686
1687         cn = jl->j_realblock;
1688         while (cn) {
1689                 /* if the blocknr == 0, this has been cleared from the hash,
1690                  ** skip it
1691                  */
1692                 if (cn->blocknr == 0) {
1693                         goto next;
1694                 }
1695                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1696                         struct buffer_head *tmp_bh;
1697                         /* we can race against journal_mark_freed when we try
1698                          * to lock_buffer(cn->bh), so we have to inc the buffer
1699                          * count, and recheck things after locking
1700                          */
1701                         tmp_bh = cn->bh;
1702                         get_bh(tmp_bh);
1703                         lock_buffer(tmp_bh);
1704                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1705                                 if (!buffer_journal_dirty(tmp_bh) ||
1706                                     buffer_journal_prepared(tmp_bh))
1707                                         BUG();
1708                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1709                                 ret++;
1710                         } else {
1711                                 /* note, cn->bh might be null now */
1712                                 unlock_buffer(tmp_bh);
1713                         }
1714                         put_bh(tmp_bh);
1715                 }
1716               next:
1717                 cn = cn->next;
1718                 cond_resched();
1719         }
1720         return ret;
1721 }
1722
1723 /* used by flush_commit_list */
1724 static int dirty_one_transaction(struct super_block *s,
1725                                  struct reiserfs_journal_list *jl)
1726 {
1727         struct reiserfs_journal_cnode *cn;
1728         struct reiserfs_journal_list *pjl;
1729         int ret = 0;
1730
1731         jl->j_state |= LIST_DIRTY;
1732         cn = jl->j_realblock;
1733         while (cn) {
1734                 /* look for a more recent transaction that logged this
1735                  ** buffer.  Only the most recent transaction with a buffer in
1736                  ** it is allowed to send that buffer to disk
1737                  */
1738                 pjl = find_newer_jl_for_cn(cn);
1739                 if (!pjl && cn->blocknr && cn->bh
1740                     && buffer_journal_dirty(cn->bh)) {
1741                         BUG_ON(!can_dirty(cn));
1742                         /* if the buffer is prepared, it will either be logged
1743                          * or restored.  If restored, we need to make sure
1744                          * it actually gets marked dirty
1745                          */
1746                         clear_buffer_journal_new(cn->bh);
1747                         if (buffer_journal_prepared(cn->bh)) {
1748                                 set_buffer_journal_restore_dirty(cn->bh);
1749                         } else {
1750                                 set_buffer_journal_test(cn->bh);
1751                                 mark_buffer_dirty(cn->bh);
1752                         }
1753                 }
1754                 cn = cn->next;
1755         }
1756         return ret;
1757 }
1758
1759 static int kupdate_transactions(struct super_block *s,
1760                                 struct reiserfs_journal_list *jl,
1761                                 struct reiserfs_journal_list **next_jl,
1762                                 unsigned long *next_trans_id,
1763                                 int num_blocks, int num_trans)
1764 {
1765         int ret = 0;
1766         int written = 0;
1767         int transactions_flushed = 0;
1768         unsigned long orig_trans_id = jl->j_trans_id;
1769         struct buffer_chunk chunk;
1770         struct list_head *entry;
1771         struct reiserfs_journal *journal = SB_JOURNAL(s);
1772         chunk.nr = 0;
1773
1774         down(&journal->j_flush_sem);
1775         if (!journal_list_still_alive(s, orig_trans_id)) {
1776                 goto done;
1777         }
1778
1779         /* we've got j_flush_sem held, nobody is going to delete any
1780          * of these lists out from underneath us
1781          */
1782         while ((num_trans && transactions_flushed < num_trans) ||
1783                (!num_trans && written < num_blocks)) {
1784
1785                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1786                     atomic_read(&jl->j_commit_left)
1787                     || !(jl->j_state & LIST_DIRTY)) {
1788                         del_from_work_list(s, jl);
1789                         break;
1790                 }
1791                 ret = write_one_transaction(s, jl, &chunk);
1792
1793                 if (ret < 0)
1794                         goto done;
1795                 transactions_flushed++;
1796                 written += ret;
1797                 entry = jl->j_list.next;
1798
1799                 /* did we wrap? */
1800                 if (entry == &journal->j_journal_list) {
1801                         break;
1802                 }
1803                 jl = JOURNAL_LIST_ENTRY(entry);
1804
1805                 /* don't bother with older transactions */
1806                 if (jl->j_trans_id <= orig_trans_id)
1807                         break;
1808         }
1809         if (chunk.nr) {
1810                 write_chunk(&chunk);
1811         }
1812
1813       done:
1814         up(&journal->j_flush_sem);
1815         return ret;
1816 }
1817
1818 /* for o_sync and fsync heavy applications, they tend to use
1819 ** all the journa list slots with tiny transactions.  These
1820 ** trigger lots and lots of calls to update the header block, which
1821 ** adds seeks and slows things down.
1822 **
1823 ** This function tries to clear out a large chunk of the journal lists
1824 ** at once, which makes everything faster since only the newest journal
1825 ** list updates the header block
1826 */
1827 static int flush_used_journal_lists(struct super_block *s,
1828                                     struct reiserfs_journal_list *jl)
1829 {
1830         unsigned long len = 0;
1831         unsigned long cur_len;
1832         int ret;
1833         int i;
1834         int limit = 256;
1835         struct reiserfs_journal_list *tjl;
1836         struct reiserfs_journal_list *flush_jl;
1837         unsigned long trans_id;
1838         struct reiserfs_journal *journal = SB_JOURNAL(s);
1839
1840         flush_jl = tjl = jl;
1841
1842         /* in data logging mode, try harder to flush a lot of blocks */
1843         if (reiserfs_data_log(s))
1844                 limit = 1024;
1845         /* flush for 256 transactions or limit blocks, whichever comes first */
1846         for (i = 0; i < 256 && len < limit; i++) {
1847                 if (atomic_read(&tjl->j_commit_left) ||
1848                     tjl->j_trans_id < jl->j_trans_id) {
1849                         break;
1850                 }
1851                 cur_len = atomic_read(&tjl->j_nonzerolen);
1852                 if (cur_len > 0) {
1853                         tjl->j_state &= ~LIST_TOUCHED;
1854                 }
1855                 len += cur_len;
1856                 flush_jl = tjl;
1857                 if (tjl->j_list.next == &journal->j_journal_list)
1858                         break;
1859                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1860         }
1861         /* try to find a group of blocks we can flush across all the
1862          ** transactions, but only bother if we've actually spanned
1863          ** across multiple lists
1864          */
1865         if (flush_jl != jl) {
1866                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1867         }
1868         flush_journal_list(s, flush_jl, 1);
1869         return 0;
1870 }
1871
1872 /*
1873 ** removes any nodes in table with name block and dev as bh.
1874 ** only touchs the hnext and hprev pointers.
1875 */
1876 void remove_journal_hash(struct super_block *sb,
1877                          struct reiserfs_journal_cnode **table,
1878                          struct reiserfs_journal_list *jl,
1879                          unsigned long block, int remove_freed)
1880 {
1881         struct reiserfs_journal_cnode *cur;
1882         struct reiserfs_journal_cnode **head;
1883
1884         head = &(journal_hash(table, sb, block));
1885         if (!head) {
1886                 return;
1887         }
1888         cur = *head;
1889         while (cur) {
1890                 if (cur->blocknr == block && cur->sb == sb
1891                     && (jl == NULL || jl == cur->jlist)
1892                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1893                         if (cur->hnext) {
1894                                 cur->hnext->hprev = cur->hprev;
1895                         }
1896                         if (cur->hprev) {
1897                                 cur->hprev->hnext = cur->hnext;
1898                         } else {
1899                                 *head = cur->hnext;
1900                         }
1901                         cur->blocknr = 0;
1902                         cur->sb = NULL;
1903                         cur->state = 0;
1904                         if (cur->bh && cur->jlist)      /* anybody who clears the cur->bh will also dec the nonzerolen */
1905                                 atomic_dec(&(cur->jlist->j_nonzerolen));
1906                         cur->bh = NULL;
1907                         cur->jlist = NULL;
1908                 }
1909                 cur = cur->hnext;
1910         }
1911 }
1912
1913 static void free_journal_ram(struct super_block *p_s_sb)
1914 {
1915         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1916         kfree(journal->j_current_jl);
1917         journal->j_num_lists--;
1918
1919         vfree(journal->j_cnode_free_orig);
1920         free_list_bitmaps(p_s_sb, journal->j_list_bitmap);
1921         free_bitmap_nodes(p_s_sb);      /* must be after free_list_bitmaps */
1922         if (journal->j_header_bh) {
1923                 brelse(journal->j_header_bh);
1924         }
1925         /* j_header_bh is on the journal dev, make sure not to release the journal
1926          * dev until we brelse j_header_bh
1927          */
1928         release_journal_dev(p_s_sb, journal);
1929         vfree(journal);
1930 }
1931
1932 /*
1933 ** call on unmount.  Only set error to 1 if you haven't made your way out
1934 ** of read_super() yet.  Any other caller must keep error at 0.
1935 */
1936 static int do_journal_release(struct reiserfs_transaction_handle *th,
1937                               struct super_block *p_s_sb, int error)
1938 {
1939         struct reiserfs_transaction_handle myth;
1940         int flushed = 0;
1941         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
1942
1943         /* we only want to flush out transactions if we were called with error == 0
1944          */
1945         if (!error && !(p_s_sb->s_flags & MS_RDONLY)) {
1946                 /* end the current trans */
1947                 BUG_ON(!th->t_trans_id);
1948                 do_journal_end(th, p_s_sb, 10, FLUSH_ALL);
1949
1950                 /* make sure something gets logged to force our way into the flush code */
1951                 if (!journal_join(&myth, p_s_sb, 1)) {
1952                         reiserfs_prepare_for_journal(p_s_sb,
1953                                                      SB_BUFFER_WITH_SB(p_s_sb),
1954                                                      1);
1955                         journal_mark_dirty(&myth, p_s_sb,
1956                                            SB_BUFFER_WITH_SB(p_s_sb));
1957                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1958                         flushed = 1;
1959                 }
1960         }
1961
1962         /* this also catches errors during the do_journal_end above */
1963         if (!error && reiserfs_is_journal_aborted(journal)) {
1964                 memset(&myth, 0, sizeof(myth));
1965                 if (!journal_join_abort(&myth, p_s_sb, 1)) {
1966                         reiserfs_prepare_for_journal(p_s_sb,
1967                                                      SB_BUFFER_WITH_SB(p_s_sb),
1968                                                      1);
1969                         journal_mark_dirty(&myth, p_s_sb,
1970                                            SB_BUFFER_WITH_SB(p_s_sb));
1971                         do_journal_end(&myth, p_s_sb, 1, FLUSH_ALL);
1972                 }
1973         }
1974
1975         reiserfs_mounted_fs_count--;
1976         /* wait for all commits to finish */
1977         cancel_delayed_work(&SB_JOURNAL(p_s_sb)->j_work);
1978         flush_workqueue(commit_wq);
1979         if (!reiserfs_mounted_fs_count) {
1980                 destroy_workqueue(commit_wq);
1981                 commit_wq = NULL;
1982         }
1983
1984         free_journal_ram(p_s_sb);
1985
1986         return 0;
1987 }
1988
1989 /*
1990 ** call on unmount.  flush all journal trans, release all alloc'd ram
1991 */
1992 int journal_release(struct reiserfs_transaction_handle *th,
1993                     struct super_block *p_s_sb)
1994 {
1995         return do_journal_release(th, p_s_sb, 0);
1996 }
1997
1998 /*
1999 ** only call from an error condition inside reiserfs_read_super!
2000 */
2001 int journal_release_error(struct reiserfs_transaction_handle *th,
2002                           struct super_block *p_s_sb)
2003 {
2004         return do_journal_release(th, p_s_sb, 1);
2005 }
2006
2007 /* compares description block with commit block.  returns 1 if they differ, 0 if they are the same */
2008 static int journal_compare_desc_commit(struct super_block *p_s_sb,
2009                                        struct reiserfs_journal_desc *desc,
2010                                        struct reiserfs_journal_commit *commit)
2011 {
2012         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
2013             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
2014             get_commit_trans_len(commit) > SB_JOURNAL(p_s_sb)->j_trans_max ||
2015             get_commit_trans_len(commit) <= 0) {
2016                 return 1;
2017         }
2018         return 0;
2019 }
2020
2021 /* returns 0 if it did not find a description block
2022 ** returns -1 if it found a corrupt commit block
2023 ** returns 1 if both desc and commit were valid
2024 */
2025 static int journal_transaction_is_valid(struct super_block *p_s_sb,
2026                                         struct buffer_head *d_bh,
2027                                         unsigned long *oldest_invalid_trans_id,
2028                                         unsigned long *newest_mount_id)
2029 {
2030         struct reiserfs_journal_desc *desc;
2031         struct reiserfs_journal_commit *commit;
2032         struct buffer_head *c_bh;
2033         unsigned long offset;
2034
2035         if (!d_bh)
2036                 return 0;
2037
2038         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2039         if (get_desc_trans_len(desc) > 0
2040             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2041                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2042                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2043                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2044                                        "journal-986: transaction "
2045                                        "is valid returning because trans_id %d is greater than "
2046                                        "oldest_invalid %lu",
2047                                        get_desc_trans_id(desc),
2048                                        *oldest_invalid_trans_id);
2049                         return 0;
2050                 }
2051                 if (newest_mount_id
2052                     && *newest_mount_id > get_desc_mount_id(desc)) {
2053                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2054                                        "journal-1087: transaction "
2055                                        "is valid returning because mount_id %d is less than "
2056                                        "newest_mount_id %lu",
2057                                        get_desc_mount_id(desc),
2058                                        *newest_mount_id);
2059                         return -1;
2060                 }
2061                 if (get_desc_trans_len(desc) > SB_JOURNAL(p_s_sb)->j_trans_max) {
2062                         reiserfs_warning(p_s_sb,
2063                                          "journal-2018: Bad transaction length %d encountered, ignoring transaction",
2064                                          get_desc_trans_len(desc));
2065                         return -1;
2066                 }
2067                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2068
2069                 /* ok, we have a journal description block, lets see if the transaction was valid */
2070                 c_bh =
2071                     journal_bread(p_s_sb,
2072                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2073                                   ((offset + get_desc_trans_len(desc) +
2074                                     1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2075                 if (!c_bh)
2076                         return 0;
2077                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2078                 if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2079                         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2080                                        "journal_transaction_is_valid, commit offset %ld had bad "
2081                                        "time %d or length %d",
2082                                        c_bh->b_blocknr -
2083                                        SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2084                                        get_commit_trans_id(commit),
2085                                        get_commit_trans_len(commit));
2086                         brelse(c_bh);
2087                         if (oldest_invalid_trans_id) {
2088                                 *oldest_invalid_trans_id =
2089                                     get_desc_trans_id(desc);
2090                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2091                                                "journal-1004: "
2092                                                "transaction_is_valid setting oldest invalid trans_id "
2093                                                "to %d",
2094                                                get_desc_trans_id(desc));
2095                         }
2096                         return -1;
2097                 }
2098                 brelse(c_bh);
2099                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2100                                "journal-1006: found valid "
2101                                "transaction start offset %llu, len %d id %d",
2102                                d_bh->b_blocknr -
2103                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2104                                get_desc_trans_len(desc),
2105                                get_desc_trans_id(desc));
2106                 return 1;
2107         } else {
2108                 return 0;
2109         }
2110 }
2111
2112 static void brelse_array(struct buffer_head **heads, int num)
2113 {
2114         int i;
2115         for (i = 0; i < num; i++) {
2116                 brelse(heads[i]);
2117         }
2118 }
2119
2120 /*
2121 ** given the start, and values for the oldest acceptable transactions,
2122 ** this either reads in a replays a transaction, or returns because the transaction
2123 ** is invalid, or too old.
2124 */
2125 static int journal_read_transaction(struct super_block *p_s_sb,
2126                                     unsigned long cur_dblock,
2127                                     unsigned long oldest_start,
2128                                     unsigned long oldest_trans_id,
2129                                     unsigned long newest_mount_id)
2130 {
2131         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2132         struct reiserfs_journal_desc *desc;
2133         struct reiserfs_journal_commit *commit;
2134         unsigned long trans_id = 0;
2135         struct buffer_head *c_bh;
2136         struct buffer_head *d_bh;
2137         struct buffer_head **log_blocks = NULL;
2138         struct buffer_head **real_blocks = NULL;
2139         unsigned long trans_offset;
2140         int i;
2141         int trans_half;
2142
2143         d_bh = journal_bread(p_s_sb, cur_dblock);
2144         if (!d_bh)
2145                 return 1;
2146         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2147         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2148         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1037: "
2149                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2150                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2151                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2152         if (get_desc_trans_id(desc) < oldest_trans_id) {
2153                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1039: "
2154                                "journal_read_trans skipping because %lu is too old",
2155                                cur_dblock -
2156                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2157                 brelse(d_bh);
2158                 return 1;
2159         }
2160         if (get_desc_mount_id(desc) != newest_mount_id) {
2161                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1146: "
2162                                "journal_read_trans skipping because %d is != "
2163                                "newest_mount_id %lu", get_desc_mount_id(desc),
2164                                newest_mount_id);
2165                 brelse(d_bh);
2166                 return 1;
2167         }
2168         c_bh = journal_bread(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2169                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2170                               SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
2171         if (!c_bh) {
2172                 brelse(d_bh);
2173                 return 1;
2174         }
2175         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2176         if (journal_compare_desc_commit(p_s_sb, desc, commit)) {
2177                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2178                                "journal_read_transaction, "
2179                                "commit offset %llu had bad time %d or length %d",
2180                                c_bh->b_blocknr -
2181                                SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2182                                get_commit_trans_id(commit),
2183                                get_commit_trans_len(commit));
2184                 brelse(c_bh);
2185                 brelse(d_bh);
2186                 return 1;
2187         }
2188         trans_id = get_desc_trans_id(desc);
2189         /* now we know we've got a good transaction, and it was inside the valid time ranges */
2190         log_blocks = kmalloc(get_desc_trans_len(desc) *
2191                              sizeof(struct buffer_head *), GFP_NOFS);
2192         real_blocks = kmalloc(get_desc_trans_len(desc) *
2193                               sizeof(struct buffer_head *), GFP_NOFS);
2194         if (!log_blocks || !real_blocks) {
2195                 brelse(c_bh);
2196                 brelse(d_bh);
2197                 kfree(log_blocks);
2198                 kfree(real_blocks);
2199                 reiserfs_warning(p_s_sb,
2200                                  "journal-1169: kmalloc failed, unable to mount FS");
2201                 return -1;
2202         }
2203         /* get all the buffer heads */
2204         trans_half = journal_trans_half(p_s_sb->s_blocksize);
2205         for (i = 0; i < get_desc_trans_len(desc); i++) {
2206                 log_blocks[i] =
2207                     journal_getblk(p_s_sb,
2208                                    SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2209                                    (trans_offset + 1 +
2210                                     i) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2211                 if (i < trans_half) {
2212                         real_blocks[i] =
2213                             sb_getblk(p_s_sb,
2214                                       le32_to_cpu(desc->j_realblock[i]));
2215                 } else {
2216                         real_blocks[i] =
2217                             sb_getblk(p_s_sb,
2218                                       le32_to_cpu(commit->
2219                                                   j_realblock[i - trans_half]));
2220                 }
2221                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(p_s_sb)) {
2222                         reiserfs_warning(p_s_sb,
2223                                          "journal-1207: REPLAY FAILURE fsck required! Block to replay is outside of filesystem");
2224                         goto abort_replay;
2225                 }
2226                 /* make sure we don't try to replay onto log or reserved area */
2227                 if (is_block_in_log_or_reserved_area
2228                     (p_s_sb, real_blocks[i]->b_blocknr)) {
2229                         reiserfs_warning(p_s_sb,
2230                                          "journal-1204: REPLAY FAILURE fsck required! Trying to replay onto a log block");
2231                       abort_replay:
2232                         brelse_array(log_blocks, i);
2233                         brelse_array(real_blocks, i);
2234                         brelse(c_bh);
2235                         brelse(d_bh);
2236                         kfree(log_blocks);
2237                         kfree(real_blocks);
2238                         return -1;
2239                 }
2240         }
2241         /* read in the log blocks, memcpy to the corresponding real block */
2242         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2243         for (i = 0; i < get_desc_trans_len(desc); i++) {
2244                 wait_on_buffer(log_blocks[i]);
2245                 if (!buffer_uptodate(log_blocks[i])) {
2246                         reiserfs_warning(p_s_sb,
2247                                          "journal-1212: REPLAY FAILURE fsck required! buffer write failed");
2248                         brelse_array(log_blocks + i,
2249                                      get_desc_trans_len(desc) - i);
2250                         brelse_array(real_blocks, get_desc_trans_len(desc));
2251                         brelse(c_bh);
2252                         brelse(d_bh);
2253                         kfree(log_blocks);
2254                         kfree(real_blocks);
2255                         return -1;
2256                 }
2257                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2258                        real_blocks[i]->b_size);
2259                 set_buffer_uptodate(real_blocks[i]);
2260                 brelse(log_blocks[i]);
2261         }
2262         /* flush out the real blocks */
2263         for (i = 0; i < get_desc_trans_len(desc); i++) {
2264                 set_buffer_dirty(real_blocks[i]);
2265                 ll_rw_block(SWRITE, 1, real_blocks + i);
2266         }
2267         for (i = 0; i < get_desc_trans_len(desc); i++) {
2268                 wait_on_buffer(real_blocks[i]);
2269                 if (!buffer_uptodate(real_blocks[i])) {
2270                         reiserfs_warning(p_s_sb,
2271                                          "journal-1226: REPLAY FAILURE, fsck required! buffer write failed");
2272                         brelse_array(real_blocks + i,
2273                                      get_desc_trans_len(desc) - i);
2274                         brelse(c_bh);
2275                         brelse(d_bh);
2276                         kfree(log_blocks);
2277                         kfree(real_blocks);
2278                         return -1;
2279                 }
2280                 brelse(real_blocks[i]);
2281         }
2282         cur_dblock =
2283             SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2284             ((trans_offset + get_desc_trans_len(desc) +
2285               2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2286         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2287                        "journal-1095: setting journal " "start to offset %ld",
2288                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb));
2289
2290         /* init starting values for the first transaction, in case this is the last transaction to be replayed. */
2291         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2292         journal->j_last_flush_trans_id = trans_id;
2293         journal->j_trans_id = trans_id + 1;
2294         /* check for trans_id overflow */
2295         if (journal->j_trans_id == 0)
2296                 journal->j_trans_id = 10;
2297         brelse(c_bh);
2298         brelse(d_bh);
2299         kfree(log_blocks);
2300         kfree(real_blocks);
2301         return 0;
2302 }
2303
2304 /* This function reads blocks starting from block and to max_block of bufsize
2305    size (but no more than BUFNR blocks at a time). This proved to improve
2306    mounting speed on self-rebuilding raid5 arrays at least.
2307    Right now it is only used from journal code. But later we might use it
2308    from other places.
2309    Note: Do not use journal_getblk/sb_getblk functions here! */
2310 static struct buffer_head *reiserfs_breada(struct block_device *dev, int block,
2311                                            int bufsize, unsigned int max_block)
2312 {
2313         struct buffer_head *bhlist[BUFNR];
2314         unsigned int blocks = BUFNR;
2315         struct buffer_head *bh;
2316         int i, j;
2317
2318         bh = __getblk(dev, block, bufsize);
2319         if (buffer_uptodate(bh))
2320                 return (bh);
2321
2322         if (block + BUFNR > max_block) {
2323                 blocks = max_block - block;
2324         }
2325         bhlist[0] = bh;
2326         j = 1;
2327         for (i = 1; i < blocks; i++) {
2328                 bh = __getblk(dev, block + i, bufsize);
2329                 if (buffer_uptodate(bh)) {
2330                         brelse(bh);
2331                         break;
2332                 } else
2333                         bhlist[j++] = bh;
2334         }
2335         ll_rw_block(READ, j, bhlist);
2336         for (i = 1; i < j; i++)
2337                 brelse(bhlist[i]);
2338         bh = bhlist[0];
2339         wait_on_buffer(bh);
2340         if (buffer_uptodate(bh))
2341                 return bh;
2342         brelse(bh);
2343         return NULL;
2344 }
2345
2346 /*
2347 ** read and replay the log
2348 ** on a clean unmount, the journal header's next unflushed pointer will be to an invalid
2349 ** transaction.  This tests that before finding all the transactions in the log, which makes normal mount times fast.
2350 **
2351 ** After a crash, this starts with the next unflushed transaction, and replays until it finds one too old, or invalid.
2352 **
2353 ** On exit, it sets things up so the first transaction will work correctly.
2354 */
2355 static int journal_read(struct super_block *p_s_sb)
2356 {
2357         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
2358         struct reiserfs_journal_desc *desc;
2359         unsigned long oldest_trans_id = 0;
2360         unsigned long oldest_invalid_trans_id = 0;
2361         time_t start;
2362         unsigned long oldest_start = 0;
2363         unsigned long cur_dblock = 0;
2364         unsigned long newest_mount_id = 9;
2365         struct buffer_head *d_bh;
2366         struct reiserfs_journal_header *jh;
2367         int valid_journal_header = 0;
2368         int replay_count = 0;
2369         int continue_replay = 1;
2370         int ret;
2371         char b[BDEVNAME_SIZE];
2372
2373         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb);
2374         reiserfs_info(p_s_sb, "checking transaction log (%s)\n",
2375                       bdevname(journal->j_dev_bd, b));
2376         start = get_seconds();
2377
2378         /* step 1, read in the journal header block.  Check the transaction it says
2379          ** is the first unflushed, and if that transaction is not valid,
2380          ** replay is done
2381          */
2382         journal->j_header_bh = journal_bread(p_s_sb,
2383                                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb)
2384                                              + SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2385         if (!journal->j_header_bh) {
2386                 return 1;
2387         }
2388         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2389         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2390             SB_ONDISK_JOURNAL_SIZE(p_s_sb)
2391             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2392                 oldest_start =
2393                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2394                     le32_to_cpu(jh->j_first_unflushed_offset);
2395                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2396                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2397                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2398                                "journal-1153: found in "
2399                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2400                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2401                                le32_to_cpu(jh->j_last_flush_trans_id));
2402                 valid_journal_header = 1;
2403
2404                 /* now, we try to read the first unflushed offset.  If it is not valid,
2405                  ** there is nothing more we can do, and it makes no sense to read
2406                  ** through the whole log.
2407                  */
2408                 d_bh =
2409                     journal_bread(p_s_sb,
2410                                   SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2411                                   le32_to_cpu(jh->j_first_unflushed_offset));
2412                 ret = journal_transaction_is_valid(p_s_sb, d_bh, NULL, NULL);
2413                 if (!ret) {
2414                         continue_replay = 0;
2415                 }
2416                 brelse(d_bh);
2417                 goto start_log_replay;
2418         }
2419
2420         if (continue_replay && bdev_read_only(p_s_sb->s_bdev)) {
2421                 reiserfs_warning(p_s_sb,
2422                                  "clm-2076: device is readonly, unable to replay log");
2423                 return -1;
2424         }
2425
2426         /* ok, there are transactions that need to be replayed.  start with the first log block, find
2427          ** all the valid transactions, and pick out the oldest.
2428          */
2429         while (continue_replay
2430                && cur_dblock <
2431                (SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2432                 SB_ONDISK_JOURNAL_SIZE(p_s_sb))) {
2433                 /* Note that it is required for blocksize of primary fs device and journal
2434                    device to be the same */
2435                 d_bh =
2436                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2437                                     p_s_sb->s_blocksize,
2438                                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2439                                     SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2440                 ret =
2441                     journal_transaction_is_valid(p_s_sb, d_bh,
2442                                                  &oldest_invalid_trans_id,
2443                                                  &newest_mount_id);
2444                 if (ret == 1) {
2445                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2446                         if (oldest_start == 0) {        /* init all oldest_ values */
2447                                 oldest_trans_id = get_desc_trans_id(desc);
2448                                 oldest_start = d_bh->b_blocknr;
2449                                 newest_mount_id = get_desc_mount_id(desc);
2450                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2451                                                "journal-1179: Setting "
2452                                                "oldest_start to offset %llu, trans_id %lu",
2453                                                oldest_start -
2454                                                SB_ONDISK_JOURNAL_1st_BLOCK
2455                                                (p_s_sb), oldest_trans_id);
2456                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2457                                 /* one we just read was older */
2458                                 oldest_trans_id = get_desc_trans_id(desc);
2459                                 oldest_start = d_bh->b_blocknr;
2460                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2461                                                "journal-1180: Resetting "
2462                                                "oldest_start to offset %lu, trans_id %lu",
2463                                                oldest_start -
2464                                                SB_ONDISK_JOURNAL_1st_BLOCK
2465                                                (p_s_sb), oldest_trans_id);
2466                         }
2467                         if (newest_mount_id < get_desc_mount_id(desc)) {
2468                                 newest_mount_id = get_desc_mount_id(desc);
2469                                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2470                                                "journal-1299: Setting "
2471                                                "newest_mount_id to %d",
2472                                                get_desc_mount_id(desc));
2473                         }
2474                         cur_dblock += get_desc_trans_len(desc) + 2;
2475                 } else {
2476                         cur_dblock++;
2477                 }
2478                 brelse(d_bh);
2479         }
2480
2481       start_log_replay:
2482         cur_dblock = oldest_start;
2483         if (oldest_trans_id) {
2484                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2485                                "journal-1206: Starting replay "
2486                                "from offset %llu, trans_id %lu",
2487                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2488                                oldest_trans_id);
2489
2490         }
2491         replay_count = 0;
2492         while (continue_replay && oldest_trans_id > 0) {
2493                 ret =
2494                     journal_read_transaction(p_s_sb, cur_dblock, oldest_start,
2495                                              oldest_trans_id, newest_mount_id);
2496                 if (ret < 0) {
2497                         return ret;
2498                 } else if (ret != 0) {
2499                         break;
2500                 }
2501                 cur_dblock =
2502                     SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) + journal->j_start;
2503                 replay_count++;
2504                 if (cur_dblock == oldest_start)
2505                         break;
2506         }
2507
2508         if (oldest_trans_id == 0) {
2509                 reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE,
2510                                "journal-1225: No valid " "transactions found");
2511         }
2512         /* j_start does not get set correctly if we don't replay any transactions.
2513          ** if we had a valid journal_header, set j_start to the first unflushed transaction value,
2514          ** copy the trans_id from the header
2515          */
2516         if (valid_journal_header && replay_count == 0) {
2517                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2518                 journal->j_trans_id =
2519                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2520                 /* check for trans_id overflow */
2521                 if (journal->j_trans_id == 0)
2522                         journal->j_trans_id = 10;
2523                 journal->j_last_flush_trans_id =
2524                     le32_to_cpu(jh->j_last_flush_trans_id);
2525                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2526         } else {
2527                 journal->j_mount_id = newest_mount_id + 1;
2528         }
2529         reiserfs_debug(p_s_sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2530                        "newest_mount_id to %lu", journal->j_mount_id);
2531         journal->j_first_unflushed_offset = journal->j_start;
2532         if (replay_count > 0) {
2533                 reiserfs_info(p_s_sb,
2534                               "replayed %d transactions in %lu seconds\n",
2535                               replay_count, get_seconds() - start);
2536         }
2537         if (!bdev_read_only(p_s_sb->s_bdev) &&
2538             _update_journal_header_block(p_s_sb, journal->j_start,
2539                                          journal->j_last_flush_trans_id)) {
2540                 /* replay failed, caller must call free_journal_ram and abort
2541                  ** the mount
2542                  */
2543                 return -1;
2544         }
2545         return 0;
2546 }
2547
2548 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2549 {
2550         struct reiserfs_journal_list *jl;
2551         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2552                      GFP_NOFS | __GFP_NOFAIL);
2553         INIT_LIST_HEAD(&jl->j_list);
2554         INIT_LIST_HEAD(&jl->j_working_list);
2555         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2556         INIT_LIST_HEAD(&jl->j_bh_list);
2557         sema_init(&jl->j_commit_lock, 1);
2558         SB_JOURNAL(s)->j_num_lists++;
2559         get_journal_list(jl);
2560         return jl;
2561 }
2562
2563 static void journal_list_init(struct super_block *p_s_sb)
2564 {
2565         SB_JOURNAL(p_s_sb)->j_current_jl = alloc_journal_list(p_s_sb);
2566 }
2567
2568 static int release_journal_dev(struct super_block *super,
2569                                struct reiserfs_journal *journal)
2570 {
2571         int result;
2572
2573         result = 0;
2574
2575         if (journal->j_dev_file != NULL) {
2576                 result = filp_close(journal->j_dev_file, NULL);
2577                 journal->j_dev_file = NULL;
2578                 journal->j_dev_bd = NULL;
2579         } else if (journal->j_dev_bd != NULL) {
2580                 result = blkdev_put(journal->j_dev_bd);
2581                 journal->j_dev_bd = NULL;
2582         }
2583
2584         if (result != 0) {
2585                 reiserfs_warning(super,
2586                                  "sh-457: release_journal_dev: Cannot release journal device: %i",
2587                                  result);
2588         }
2589         return result;
2590 }
2591
2592 static int journal_init_dev(struct super_block *super,
2593                             struct reiserfs_journal *journal,
2594                             const char *jdev_name)
2595 {
2596         int result;
2597         dev_t jdev;
2598         int blkdev_mode = FMODE_READ | FMODE_WRITE;
2599         char b[BDEVNAME_SIZE];
2600
2601         result = 0;
2602
2603         journal->j_dev_bd = NULL;
2604         journal->j_dev_file = NULL;
2605         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2606             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2607
2608         if (bdev_read_only(super->s_bdev))
2609                 blkdev_mode = FMODE_READ;
2610
2611         /* there is no "jdev" option and journal is on separate device */
2612         if ((!jdev_name || !jdev_name[0])) {
2613                 journal->j_dev_bd = open_by_devnum(jdev, blkdev_mode);
2614                 if (IS_ERR(journal->j_dev_bd)) {
2615                         result = PTR_ERR(journal->j_dev_bd);
2616                         journal->j_dev_bd = NULL;
2617                         reiserfs_warning(super, "sh-458: journal_init_dev: "
2618                                          "cannot init journal device '%s': %i",
2619                                          __bdevname(jdev, b), result);
2620                         return result;
2621                 } else if (jdev != super->s_dev)
2622                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2623                 return 0;
2624         }
2625
2626         journal->j_dev_file = filp_open(jdev_name, 0, 0);
2627         if (!IS_ERR(journal->j_dev_file)) {
2628                 struct inode *jdev_inode = journal->j_dev_file->f_mapping->host;
2629                 if (!S_ISBLK(jdev_inode->i_mode)) {
2630                         reiserfs_warning(super, "journal_init_dev: '%s' is "
2631                                          "not a block device", jdev_name);
2632                         result = -ENOTBLK;
2633                         release_journal_dev(super, journal);
2634                 } else {
2635                         /* ok */
2636                         journal->j_dev_bd = I_BDEV(jdev_inode);
2637                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2638                         reiserfs_info(super,
2639                                       "journal_init_dev: journal device: %s\n",
2640                                       bdevname(journal->j_dev_bd, b));
2641                 }
2642         } else {
2643                 result = PTR_ERR(journal->j_dev_file);
2644                 journal->j_dev_file = NULL;
2645                 reiserfs_warning(super,
2646                                  "journal_init_dev: Cannot open '%s': %i",
2647                                  jdev_name, result);
2648         }
2649         return result;
2650 }
2651
2652 /**
2653  * When creating/tuning a file system user can assign some
2654  * journal params within boundaries which depend on the ratio
2655  * blocksize/standard_blocksize.
2656  *
2657  * For blocks >= standard_blocksize transaction size should
2658  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2659  * then JOURNAL_TRANS_MAX_DEFAULT.
2660  *
2661  * For blocks < standard_blocksize these boundaries should be
2662  * decreased proportionally.
2663  */
2664 #define REISERFS_STANDARD_BLKSIZE (4096)
2665
2666 static int check_advise_trans_params(struct super_block *p_s_sb,
2667                                      struct reiserfs_journal *journal)
2668 {
2669         if (journal->j_trans_max) {
2670                 /* Non-default journal params.
2671                    Do sanity check for them. */
2672                 int ratio = 1;
2673                 if (p_s_sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2674                         ratio = REISERFS_STANDARD_BLKSIZE / p_s_sb->s_blocksize;
2675
2676                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2677                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2678                     SB_ONDISK_JOURNAL_SIZE(p_s_sb) / journal->j_trans_max <
2679                     JOURNAL_MIN_RATIO) {
2680                         reiserfs_warning(p_s_sb,
2681                                  "sh-462: bad transaction max size (%u). FSCK?",
2682                                  journal->j_trans_max);
2683                         return 1;
2684                 }
2685                 if (journal->j_max_batch != (journal->j_trans_max) *
2686                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2687                         reiserfs_warning(p_s_sb,
2688                                 "sh-463: bad transaction max batch (%u). FSCK?",
2689                                 journal->j_max_batch);
2690                         return 1;
2691                 }
2692         } else {
2693                 /* Default journal params.
2694                    The file system was created by old version
2695                    of mkreiserfs, so some fields contain zeros,
2696                    and we need to advise proper values for them */
2697                 if (p_s_sb->s_blocksize != REISERFS_STANDARD_BLKSIZE)
2698                         reiserfs_panic(p_s_sb, "sh-464: bad blocksize (%u)",
2699                                        p_s_sb->s_blocksize);
2700                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2701                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2702                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2703         }
2704         return 0;
2705 }
2706
2707 /*
2708 ** must be called once on fs mount.  calls journal_read for you
2709 */
2710 int journal_init(struct super_block *p_s_sb, const char *j_dev_name,
2711                  int old_format, unsigned int commit_max_age)
2712 {
2713         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(p_s_sb) * 2;
2714         struct buffer_head *bhjh;
2715         struct reiserfs_super_block *rs;
2716         struct reiserfs_journal_header *jh;
2717         struct reiserfs_journal *journal;
2718         struct reiserfs_journal_list *jl;
2719         char b[BDEVNAME_SIZE];
2720
2721         journal = SB_JOURNAL(p_s_sb) = vmalloc(sizeof(struct reiserfs_journal));
2722         if (!journal) {
2723                 reiserfs_warning(p_s_sb,
2724                                  "journal-1256: unable to get memory for journal structure");
2725                 return 1;
2726         }
2727         memset(journal, 0, sizeof(struct reiserfs_journal));
2728         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2729         INIT_LIST_HEAD(&journal->j_prealloc_list);
2730         INIT_LIST_HEAD(&journal->j_working_list);
2731         INIT_LIST_HEAD(&journal->j_journal_list);
2732         journal->j_persistent_trans = 0;
2733         if (reiserfs_allocate_list_bitmaps(p_s_sb,
2734                                            journal->j_list_bitmap,
2735                                            SB_BMAP_NR(p_s_sb)))
2736                 goto free_and_return;
2737         allocate_bitmap_nodes(p_s_sb);
2738
2739         /* reserved for journal area support */
2740         SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) = (old_format ?
2741                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2742                                                  / p_s_sb->s_blocksize +
2743                                                  SB_BMAP_NR(p_s_sb) +
2744                                                  1 :
2745                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2746                                                  p_s_sb->s_blocksize + 2);
2747
2748         /* Sanity check to see is the standard journal fitting withing first bitmap
2749            (actual for small blocksizes) */
2750         if (!SB_ONDISK_JOURNAL_DEVICE(p_s_sb) &&
2751             (SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb) +
2752              SB_ONDISK_JOURNAL_SIZE(p_s_sb) > p_s_sb->s_blocksize * 8)) {
2753                 reiserfs_warning(p_s_sb,
2754                                  "journal-1393: journal does not fit for area "
2755                                  "addressed by first of bitmap blocks. It starts at "
2756                                  "%u and its size is %u. Block size %ld",
2757                                  SB_JOURNAL_1st_RESERVED_BLOCK(p_s_sb),
2758                                  SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2759                                  p_s_sb->s_blocksize);
2760                 goto free_and_return;
2761         }
2762
2763         if (journal_init_dev(p_s_sb, journal, j_dev_name) != 0) {
2764                 reiserfs_warning(p_s_sb,
2765                                  "sh-462: unable to initialize jornal device");
2766                 goto free_and_return;
2767         }
2768
2769         rs = SB_DISK_SUPER_BLOCK(p_s_sb);
2770
2771         /* read journal header */
2772         bhjh = journal_bread(p_s_sb,
2773                              SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
2774                              SB_ONDISK_JOURNAL_SIZE(p_s_sb));
2775         if (!bhjh) {
2776                 reiserfs_warning(p_s_sb,
2777                                  "sh-459: unable to read journal header");
2778                 goto free_and_return;
2779         }
2780         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2781
2782         /* make sure that journal matches to the super block */
2783         if (is_reiserfs_jr(rs)
2784             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2785                 sb_jp_journal_magic(rs))) {
2786                 reiserfs_warning(p_s_sb,
2787                                  "sh-460: journal header magic %x "
2788                                  "(device %s) does not match to magic found in super "
2789                                  "block %x", jh->jh_journal.jp_journal_magic,
2790                                  bdevname(journal->j_dev_bd, b),
2791                                  sb_jp_journal_magic(rs));
2792                 brelse(bhjh);
2793                 goto free_and_return;
2794         }
2795
2796         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2797         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2798         journal->j_max_commit_age =
2799             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2800         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2801
2802         if (check_advise_trans_params(p_s_sb, journal) != 0)
2803                 goto free_and_return;
2804         journal->j_default_max_commit_age = journal->j_max_commit_age;
2805
2806         if (commit_max_age != 0) {
2807                 journal->j_max_commit_age = commit_max_age;
2808                 journal->j_max_trans_age = commit_max_age;
2809         }
2810
2811         reiserfs_info(p_s_sb, "journal params: device %s, size %u, "
2812                       "journal first block %u, max trans len %u, max batch %u, "
2813                       "max commit age %u, max trans age %u\n",
2814                       bdevname(journal->j_dev_bd, b),
2815                       SB_ONDISK_JOURNAL_SIZE(p_s_sb),
2816                       SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb),
2817                       journal->j_trans_max,
2818                       journal->j_max_batch,
2819                       journal->j_max_commit_age, journal->j_max_trans_age);
2820
2821         brelse(bhjh);
2822
2823         journal->j_list_bitmap_index = 0;
2824         journal_list_init(p_s_sb);
2825
2826         memset(journal->j_list_hash_table, 0,
2827                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2828
2829         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2830         spin_lock_init(&journal->j_dirty_buffers_lock);
2831
2832         journal->j_start = 0;
2833         journal->j_len = 0;
2834         journal->j_len_alloc = 0;
2835         atomic_set(&(journal->j_wcount), 0);
2836         atomic_set(&(journal->j_async_throttle), 0);
2837         journal->j_bcount = 0;
2838         journal->j_trans_start_time = 0;
2839         journal->j_last = NULL;
2840         journal->j_first = NULL;
2841         init_waitqueue_head(&(journal->j_join_wait));
2842         sema_init(&journal->j_lock, 1);
2843         sema_init(&journal->j_flush_sem, 1);
2844
2845         journal->j_trans_id = 10;
2846         journal->j_mount_id = 10;
2847         journal->j_state = 0;
2848         atomic_set(&(journal->j_jlock), 0);
2849         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2850         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2851         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2852         journal->j_cnode_used = 0;
2853         journal->j_must_wait = 0;
2854
2855         if (journal->j_cnode_free == 0) {
2856                 reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
2857                                  "allocation failed (%ld bytes). Journal is "
2858                                  "too large for available memory. Usually "
2859                                  "this is due to a journal that is too large.",
2860                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2861                 goto free_and_return;
2862         }
2863
2864         init_journal_hash(p_s_sb);
2865         jl = journal->j_current_jl;
2866         jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
2867         if (!jl->j_list_bitmap) {
2868                 reiserfs_warning(p_s_sb,
2869                                  "journal-2005, get_list_bitmap failed for journal list 0");
2870                 goto free_and_return;
2871         }
2872         if (journal_read(p_s_sb) < 0) {
2873                 reiserfs_warning(p_s_sb, "Replay Failure, unable to mount");
2874                 goto free_and_return;
2875         }
2876
2877         reiserfs_mounted_fs_count++;
2878         if (reiserfs_mounted_fs_count <= 1)
2879                 commit_wq = create_workqueue("reiserfs");
2880
2881         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2882         journal->j_work_sb = p_s_sb;
2883         return 0;
2884       free_and_return:
2885         free_journal_ram(p_s_sb);
2886         return 1;
2887 }
2888
2889 /*
2890 ** test for a polite end of the current transaction.  Used by file_write, and should
2891 ** be used by delete to make sure they don't write more than can fit inside a single
2892 ** transaction
2893 */
2894 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2895                                    int new_alloc)
2896 {
2897         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2898         time_t now = get_seconds();
2899         /* cannot restart while nested */
2900         BUG_ON(!th->t_trans_id);
2901         if (th->t_refcount > 1)
2902                 return 0;
2903         if (journal->j_must_wait > 0 ||
2904             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2905             atomic_read(&(journal->j_jlock)) ||
2906             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2907             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2908                 return 1;
2909         }
2910         /* protected by the BKL here */
2911         journal->j_len_alloc += new_alloc;
2912         th->t_blocks_allocated += new_alloc ;
2913         return 0;
2914 }
2915
2916 /* this must be called inside a transaction, and requires the
2917 ** kernel_lock to be held
2918 */
2919 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2920 {
2921         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2922         BUG_ON(!th->t_trans_id);
2923         journal->j_must_wait = 1;
2924         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2925         return;
2926 }
2927
2928 /* this must be called without a transaction started, and does not
2929 ** require BKL
2930 */
2931 void reiserfs_allow_writes(struct super_block *s)
2932 {
2933         struct reiserfs_journal *journal = SB_JOURNAL(s);
2934         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2935         wake_up(&journal->j_join_wait);
2936 }
2937
2938 /* this must be called without a transaction started, and does not
2939 ** require BKL
2940 */
2941 void reiserfs_wait_on_write_block(struct super_block *s)
2942 {
2943         struct reiserfs_journal *journal = SB_JOURNAL(s);
2944         wait_event(journal->j_join_wait,
2945                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2946 }
2947
2948 static void queue_log_writer(struct super_block *s)
2949 {
2950         wait_queue_t wait;
2951         struct reiserfs_journal *journal = SB_JOURNAL(s);
2952         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2953
2954         /*
2955          * we don't want to use wait_event here because
2956          * we only want to wait once.
2957          */
2958         init_waitqueue_entry(&wait, current);
2959         add_wait_queue(&journal->j_join_wait, &wait);
2960         set_current_state(TASK_UNINTERRUPTIBLE);
2961         if (test_bit(J_WRITERS_QUEUED, &journal->j_state))
2962                 schedule();
2963         __set_current_state(TASK_RUNNING);
2964         remove_wait_queue(&journal->j_join_wait, &wait);
2965 }
2966
2967 static void wake_queued_writers(struct super_block *s)
2968 {
2969         struct reiserfs_journal *journal = SB_JOURNAL(s);
2970         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2971                 wake_up(&journal->j_join_wait);
2972 }
2973
2974 static void let_transaction_grow(struct super_block *sb, unsigned long trans_id)
2975 {
2976         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2977         unsigned long bcount = journal->j_bcount;
2978         while (1) {
2979                 schedule_timeout_uninterruptible(1);
2980                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2981                 while ((atomic_read(&journal->j_wcount) > 0 ||
2982                         atomic_read(&journal->j_jlock)) &&
2983                        journal->j_trans_id == trans_id) {
2984                         queue_log_writer(sb);
2985                 }
2986                 if (journal->j_trans_id != trans_id)
2987                         break;
2988                 if (bcount == journal->j_bcount)
2989                         break;
2990                 bcount = journal->j_bcount;
2991         }
2992 }
2993
2994 /* join == true if you must join an existing transaction.
2995 ** join == false if you can deal with waiting for others to finish
2996 **
2997 ** this will block until the transaction is joinable.  send the number of blocks you
2998 ** expect to use in nblocks.
2999 */
3000 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3001                               struct super_block *p_s_sb, unsigned long nblocks,
3002                               int join)
3003 {
3004         time_t now = get_seconds();
3005         int old_trans_id;
3006         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3007         struct reiserfs_transaction_handle myth;
3008         int sched_count = 0;
3009         int retval;
3010
3011         reiserfs_check_lock_depth(p_s_sb, "journal_begin");
3012         BUG_ON(nblocks > journal->j_trans_max);
3013
3014         PROC_INFO_INC(p_s_sb, journal.journal_being);
3015         /* set here for journal_join */
3016         th->t_refcount = 1;
3017         th->t_super = p_s_sb;
3018
3019       relock:
3020         lock_journal(p_s_sb);
3021         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3022                 unlock_journal(p_s_sb);
3023                 retval = journal->j_errno;
3024                 goto out_fail;
3025         }
3026         journal->j_bcount++;
3027
3028         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3029                 unlock_journal(p_s_sb);
3030                 reiserfs_wait_on_write_block(p_s_sb);
3031                 PROC_INFO_INC(p_s_sb, journal.journal_relock_writers);
3032                 goto relock;
3033         }
3034         now = get_seconds();
3035
3036         /* if there is no room in the journal OR
3037          ** if this transaction is too old, and we weren't called joinable, wait for it to finish before beginning
3038          ** we don't sleep if there aren't other writers
3039          */
3040
3041         if ((!join && journal->j_must_wait > 0) ||
3042             (!join
3043              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3044             || (!join && atomic_read(&journal->j_wcount) > 0
3045                 && journal->j_trans_start_time > 0
3046                 && (now - journal->j_trans_start_time) >
3047                 journal->j_max_trans_age) || (!join
3048                                               && atomic_read(&journal->j_jlock))
3049             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3050
3051                 old_trans_id = journal->j_trans_id;
3052                 unlock_journal(p_s_sb); /* allow others to finish this transaction */
3053
3054                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3055                     journal->j_max_batch &&
3056                     ((journal->j_len + nblocks + 2) * 100) <
3057                     (journal->j_len_alloc * 75)) {
3058                         if (atomic_read(&journal->j_wcount) > 10) {
3059                                 sched_count++;
3060                                 queue_log_writer(p_s_sb);
3061                                 goto relock;
3062                         }
3063                 }
3064                 /* don't mess with joining the transaction if all we have to do is
3065                  * wait for someone else to do a commit
3066                  */
3067                 if (atomic_read(&journal->j_jlock)) {
3068                         while (journal->j_trans_id == old_trans_id &&
3069                                atomic_read(&journal->j_jlock)) {
3070                                 queue_log_writer(p_s_sb);
3071                         }
3072                         goto relock;
3073                 }
3074                 retval = journal_join(&myth, p_s_sb, 1);
3075                 if (retval)
3076                         goto out_fail;
3077
3078                 /* someone might have ended the transaction while we joined */
3079                 if (old_trans_id != journal->j_trans_id) {
3080                         retval = do_journal_end(&myth, p_s_sb, 1, 0);
3081                 } else {
3082                         retval = do_journal_end(&myth, p_s_sb, 1, COMMIT_NOW);
3083                 }
3084
3085                 if (retval)
3086                         goto out_fail;
3087
3088                 PROC_INFO_INC(p_s_sb, journal.journal_relock_wcount);
3089                 goto relock;
3090         }
3091         /* we are the first writer, set trans_id */
3092         if (journal->j_trans_start_time == 0) {
3093                 journal->j_trans_start_time = get_seconds();
3094         }
3095         atomic_inc(&(journal->j_wcount));
3096         journal->j_len_alloc += nblocks;
3097         th->t_blocks_logged = 0;
3098         th->t_blocks_allocated = nblocks;
3099         th->t_trans_id = journal->j_trans_id;
3100         unlock_journal(p_s_sb);
3101         INIT_LIST_HEAD(&th->t_list);
3102         get_fs_excl();
3103         return 0;
3104
3105       out_fail:
3106         memset(th, 0, sizeof(*th));
3107         /* Re-set th->t_super, so we can properly keep track of how many
3108          * persistent transactions there are. We need to do this so if this
3109          * call is part of a failed restart_transaction, we can free it later */
3110         th->t_super = p_s_sb;
3111         return retval;
3112 }
3113
3114 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3115                                                                     super_block
3116                                                                     *s,
3117                                                                     int nblocks)
3118 {
3119         int ret;
3120         struct reiserfs_transaction_handle *th;
3121
3122         /* if we're nesting into an existing transaction.  It will be
3123          ** persistent on its own
3124          */
3125         if (reiserfs_transaction_running(s)) {
3126                 th = current->journal_info;
3127                 th->t_refcount++;
3128                 BUG_ON(th->t_refcount < 2);
3129
3130                 return th;
3131         }
3132         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3133         if (!th)
3134                 return NULL;
3135         ret = journal_begin(th, s, nblocks);
3136         if (ret) {
3137                 kfree(th);
3138                 return NULL;
3139         }
3140
3141         SB_JOURNAL(s)->j_persistent_trans++;
3142         return th;
3143 }
3144
3145 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3146 {
3147         struct super_block *s = th->t_super;
3148         int ret = 0;
3149         if (th->t_trans_id)
3150                 ret = journal_end(th, th->t_super, th->t_blocks_allocated);
3151         else
3152                 ret = -EIO;
3153         if (th->t_refcount == 0) {
3154                 SB_JOURNAL(s)->j_persistent_trans--;
3155                 kfree(th);
3156         }
3157         return ret;
3158 }
3159
3160 static int journal_join(struct reiserfs_transaction_handle *th,
3161                         struct super_block *p_s_sb, unsigned long nblocks)
3162 {
3163         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3164
3165         /* this keeps do_journal_end from NULLing out the current->journal_info
3166          ** pointer
3167          */
3168         th->t_handle_save = cur_th;
3169         BUG_ON(cur_th && cur_th->t_refcount > 1);
3170         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_JOIN);
3171 }
3172
3173 int journal_join_abort(struct reiserfs_transaction_handle *th,
3174                        struct super_block *p_s_sb, unsigned long nblocks)
3175 {
3176         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3177
3178         /* this keeps do_journal_end from NULLing out the current->journal_info
3179          ** pointer
3180          */
3181         th->t_handle_save = cur_th;
3182         BUG_ON(cur_th && cur_th->t_refcount > 1);
3183         return do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_ABORT);
3184 }
3185
3186 int journal_begin(struct reiserfs_transaction_handle *th,
3187                   struct super_block *p_s_sb, unsigned long nblocks)
3188 {
3189         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3190         int ret;
3191
3192         th->t_handle_save = NULL;
3193         if (cur_th) {
3194                 /* we are nesting into the current transaction */
3195                 if (cur_th->t_super == p_s_sb) {
3196                         BUG_ON(!cur_th->t_refcount);
3197                         cur_th->t_refcount++;
3198                         memcpy(th, cur_th, sizeof(*th));
3199                         if (th->t_refcount <= 1)
3200                                 reiserfs_warning(p_s_sb,
3201                                                  "BAD: refcount <= 1, but journal_info != 0");
3202                         return 0;
3203                 } else {
3204                         /* we've ended up with a handle from a different filesystem.
3205                          ** save it and restore on journal_end.  This should never
3206                          ** really happen...
3207                          */
3208                         reiserfs_warning(p_s_sb,
3209                                          "clm-2100: nesting info a different FS");
3210                         th->t_handle_save = current->journal_info;
3211                         current->journal_info = th;
3212                 }
3213         } else {
3214                 current->journal_info = th;
3215         }
3216         ret = do_journal_begin_r(th, p_s_sb, nblocks, JBEGIN_REG);
3217         BUG_ON(current->journal_info != th);
3218
3219         /* I guess this boils down to being the reciprocal of clm-2100 above.
3220          * If do_journal_begin_r fails, we need to put it back, since journal_end
3221          * won't be called to do it. */
3222         if (ret)
3223                 current->journal_info = th->t_handle_save;
3224         else
3225                 BUG_ON(!th->t_refcount);
3226
3227         return ret;
3228 }
3229
3230 /*
3231 ** puts bh into the current transaction.  If it was already there, reorders removes the
3232 ** old pointers from the hash, and puts new ones in (to make sure replay happen in the right order).
3233 **
3234 ** if it was dirty, cleans and files onto the clean list.  I can't let it be dirty again until the
3235 ** transaction is committed.
3236 **
3237 ** if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3238 */
3239 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3240                        struct super_block *p_s_sb, struct buffer_head *bh)
3241 {
3242         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3243         struct reiserfs_journal_cnode *cn = NULL;
3244         int count_already_incd = 0;
3245         int prepared = 0;
3246         BUG_ON(!th->t_trans_id);
3247
3248         PROC_INFO_INC(p_s_sb, journal.mark_dirty);
3249         if (th->t_trans_id != journal->j_trans_id) {
3250                 reiserfs_panic(th->t_super,
3251                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3252                                th->t_trans_id, journal->j_trans_id);
3253         }
3254
3255         p_s_sb->s_dirt = 1;
3256
3257         prepared = test_clear_buffer_journal_prepared(bh);
3258         clear_buffer_journal_restore_dirty(bh);
3259         /* already in this transaction, we are done */
3260         if (buffer_journaled(bh)) {
3261                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_already);
3262                 return 0;
3263         }
3264
3265         /* this must be turned into a panic instead of a warning.  We can't allow
3266          ** a dirty or journal_dirty or locked buffer to be logged, as some changes
3267          ** could get to disk too early.  NOT GOOD.
3268          */
3269         if (!prepared || buffer_dirty(bh)) {
3270                 reiserfs_warning(p_s_sb, "journal-1777: buffer %llu bad state "
3271                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3272                                  (unsigned long long)bh->b_blocknr,
3273                                  prepared ? ' ' : '!',
3274                                  buffer_locked(bh) ? ' ' : '!',
3275                                  buffer_dirty(bh) ? ' ' : '!',
3276                                  buffer_journal_dirty(bh) ? ' ' : '!');
3277         }
3278
3279         if (atomic_read(&(journal->j_wcount)) <= 0) {
3280                 reiserfs_warning(p_s_sb,
3281                                  "journal-1409: journal_mark_dirty returning because j_wcount was %d",
3282                                  atomic_read(&(journal->j_wcount)));
3283                 return 1;
3284         }
3285         /* this error means I've screwed up, and we've overflowed the transaction.
3286          ** Nothing can be done here, except make the FS readonly or panic.
3287          */
3288         if (journal->j_len >= journal->j_trans_max) {
3289                 reiserfs_panic(th->t_super,
3290                                "journal-1413: journal_mark_dirty: j_len (%lu) is too big\n",
3291                                journal->j_len);
3292         }
3293
3294         if (buffer_journal_dirty(bh)) {
3295                 count_already_incd = 1;
3296                 PROC_INFO_INC(p_s_sb, journal.mark_dirty_notjournal);
3297                 clear_buffer_journal_dirty(bh);
3298         }
3299
3300         if (journal->j_len > journal->j_len_alloc) {
3301                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3302         }
3303
3304         set_buffer_journaled(bh);
3305
3306         /* now put this guy on the end */
3307         if (!cn) {
3308                 cn = get_cnode(p_s_sb);
3309                 if (!cn) {
3310                         reiserfs_panic(p_s_sb, "get_cnode failed!\n");
3311                 }
3312
3313                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3314                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3315                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3316                 }
3317                 th->t_blocks_logged++;
3318                 journal->j_len++;
3319
3320                 cn->bh = bh;
3321                 cn->blocknr = bh->b_blocknr;
3322                 cn->sb = p_s_sb;
3323                 cn->jlist = NULL;
3324                 insert_journal_hash(journal->j_hash_table, cn);
3325                 if (!count_already_incd) {
3326                         get_bh(bh);
3327                 }
3328         }
3329         cn->next = NULL;
3330         cn->prev = journal->j_last;
3331         cn->bh = bh;
3332         if (journal->j_last) {
3333                 journal->j_last->next = cn;
3334                 journal->j_last = cn;
3335         } else {
3336                 journal->j_first = cn;
3337                 journal->j_last = cn;
3338         }
3339         return 0;
3340 }
3341
3342 int journal_end(struct reiserfs_transaction_handle *th,
3343                 struct super_block *p_s_sb, unsigned long nblocks)
3344 {
3345         if (!current->journal_info && th->t_refcount > 1)
3346                 reiserfs_warning(p_s_sb, "REISER-NESTING: th NULL, refcount %d",
3347                                  th->t_refcount);
3348
3349         if (!th->t_trans_id) {
3350                 WARN_ON(1);
3351                 return -EIO;
3352         }
3353
3354         th->t_refcount--;
3355         if (th->t_refcount > 0) {
3356                 struct reiserfs_transaction_handle *cur_th =
3357                     current->journal_info;
3358
3359                 /* we aren't allowed to close a nested transaction on a different
3360                  ** filesystem from the one in the task struct
3361                  */
3362                 BUG_ON(cur_th->t_super != th->t_super);
3363
3364                 if (th != cur_th) {
3365                         memcpy(current->journal_info, th, sizeof(*th));
3366                         th->t_trans_id = 0;
3367                 }
3368                 return 0;
3369         } else {
3370                 return do_journal_end(th, p_s_sb, nblocks, 0);
3371         }
3372 }
3373
3374 /* removes from the current transaction, relsing and descrementing any counters.
3375 ** also files the removed buffer directly onto the clean list
3376 **
3377 ** called by journal_mark_freed when a block has been deleted
3378 **
3379 ** returns 1 if it cleaned and relsed the buffer. 0 otherwise
3380 */
3381 static int remove_from_transaction(struct super_block *p_s_sb,
3382                                    b_blocknr_t blocknr, int already_cleaned)
3383 {
3384         struct buffer_head *bh;
3385         struct reiserfs_journal_cnode *cn;
3386         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3387         int ret = 0;
3388
3389         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3390         if (!cn || !cn->bh) {
3391                 return ret;
3392         }
3393         bh = cn->bh;
3394         if (cn->prev) {
3395                 cn->prev->next = cn->next;
3396         }
3397         if (cn->next) {
3398                 cn->next->prev = cn->prev;
3399         }
3400         if (cn == journal->j_first) {
3401                 journal->j_first = cn->next;
3402         }
3403         if (cn == journal->j_last) {
3404                 journal->j_last = cn->prev;
3405         }
3406         if (bh)
3407                 remove_journal_hash(p_s_sb, journal->j_hash_table, NULL,
3408                                     bh->b_blocknr, 0);
3409         clear_buffer_journaled(bh);     /* don't log this one */
3410
3411         if (!already_cleaned) {
3412                 clear_buffer_journal_dirty(bh);
3413                 clear_buffer_dirty(bh);
3414                 clear_buffer_journal_test(bh);
3415                 put_bh(bh);
3416                 if (atomic_read(&(bh->b_count)) < 0) {
3417                         reiserfs_warning(p_s_sb,
3418                                          "journal-1752: remove from trans, b_count < 0");
3419                 }
3420                 ret = 1;
3421         }
3422         journal->j_len--;
3423         journal->j_len_alloc--;
3424         free_cnode(p_s_sb, cn);
3425         return ret;
3426 }
3427
3428 /*
3429 ** for any cnode in a journal list, it can only be dirtied of all the
3430 ** transactions that include it are committed to disk.
3431 ** this checks through each transaction, and returns 1 if you are allowed to dirty,
3432 ** and 0 if you aren't
3433 **
3434 ** it is called by dirty_journal_list, which is called after flush_commit_list has gotten all the log
3435 ** blocks for a given transaction on disk
3436 **
3437 */
3438 static int can_dirty(struct reiserfs_journal_cnode *cn)
3439 {
3440         struct super_block *sb = cn->sb;
3441         b_blocknr_t blocknr = cn->blocknr;
3442         struct reiserfs_journal_cnode *cur = cn->hprev;
3443         int can_dirty = 1;
3444
3445         /* first test hprev.  These are all newer than cn, so any node here
3446          ** with the same block number and dev means this node can't be sent
3447          ** to disk right now.
3448          */
3449         while (cur && can_dirty) {
3450                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3451                     cur->blocknr == blocknr) {
3452                         can_dirty = 0;
3453                 }
3454                 cur = cur->hprev;
3455         }
3456         /* then test hnext.  These are all older than cn.  As long as they
3457          ** are committed to the log, it is safe to write cn to disk
3458          */
3459         cur = cn->hnext;
3460         while (cur && can_dirty) {
3461                 if (cur->jlist && cur->jlist->j_len > 0 &&
3462                     atomic_read(&(cur->jlist->j_commit_left)) > 0 && cur->bh &&
3463                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3464                         can_dirty = 0;
3465                 }
3466                 cur = cur->hnext;
3467         }
3468         return can_dirty;
3469 }
3470
3471 /* syncs the commit blocks, but does not force the real buffers to disk
3472 ** will wait until the current transaction is done/committed before returning
3473 */
3474 int journal_end_sync(struct reiserfs_transaction_handle *th,
3475                      struct super_block *p_s_sb, unsigned long nblocks)
3476 {
3477         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3478
3479         BUG_ON(!th->t_trans_id);
3480         /* you can sync while nested, very, very bad */
3481         BUG_ON(th->t_refcount > 1);
3482         if (journal->j_len == 0) {
3483                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3484                                              1);
3485                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3486         }
3487         return do_journal_end(th, p_s_sb, nblocks, COMMIT_NOW | WAIT);
3488 }
3489
3490 /*
3491 ** writeback the pending async commits to disk
3492 */
3493 static void flush_async_commits(struct work_struct *work)
3494 {
3495         struct reiserfs_journal *journal =
3496                 container_of(work, struct reiserfs_journal, j_work.work);
3497         struct super_block *p_s_sb = journal->j_work_sb;
3498         struct reiserfs_journal_list *jl;
3499         struct list_head *entry;
3500
3501         lock_kernel();
3502         if (!list_empty(&journal->j_journal_list)) {
3503                 /* last entry is the youngest, commit it and you get everything */
3504                 entry = journal->j_journal_list.prev;
3505                 jl = JOURNAL_LIST_ENTRY(entry);
3506                 flush_commit_list(p_s_sb, jl, 1);
3507         }
3508         unlock_kernel();
3509 }
3510
3511 /*
3512 ** flushes any old transactions to disk
3513 ** ends the current transaction if it is too old
3514 */
3515 int reiserfs_flush_old_commits(struct super_block *p_s_sb)
3516 {
3517         time_t now;
3518         struct reiserfs_transaction_handle th;
3519         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3520
3521         now = get_seconds();
3522         /* safety check so we don't flush while we are replaying the log during
3523          * mount
3524          */
3525         if (list_empty(&journal->j_journal_list)) {
3526                 return 0;
3527         }
3528
3529         /* check the current transaction.  If there are no writers, and it is
3530          * too old, finish it, and force the commit blocks to disk
3531          */
3532         if (atomic_read(&journal->j_wcount) <= 0 &&
3533             journal->j_trans_start_time > 0 &&
3534             journal->j_len > 0 &&
3535             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3536                 if (!journal_join(&th, p_s_sb, 1)) {
3537                         reiserfs_prepare_for_journal(p_s_sb,
3538                                                      SB_BUFFER_WITH_SB(p_s_sb),
3539                                                      1);
3540                         journal_mark_dirty(&th, p_s_sb,
3541                                            SB_BUFFER_WITH_SB(p_s_sb));
3542
3543                         /* we're only being called from kreiserfsd, it makes no sense to do
3544                          ** an async commit so that kreiserfsd can do it later
3545                          */
3546                         do_journal_end(&th, p_s_sb, 1, COMMIT_NOW | WAIT);
3547                 }
3548         }
3549         return p_s_sb->s_dirt;
3550 }
3551
3552 /*
3553 ** returns 0 if do_journal_end should return right away, returns 1 if do_journal_end should finish the commit
3554 **
3555 ** if the current transaction is too old, but still has writers, this will wait on j_join_wait until all
3556 ** the writers are done.  By the time it wakes up, the transaction it was called has already ended, so it just
3557 ** flushes the commit list and returns 0.
3558 **
3559 ** Won't batch when flush or commit_now is set.  Also won't batch when others are waiting on j_join_wait.
3560 **
3561 ** Note, we can't allow the journal_end to proceed while there are still writers in the log.
3562 */
3563 static int check_journal_end(struct reiserfs_transaction_handle *th,
3564                              struct super_block *p_s_sb, unsigned long nblocks,
3565                              int flags)
3566 {
3567
3568         time_t now;
3569         int flush = flags & FLUSH_ALL;
3570         int commit_now = flags & COMMIT_NOW;
3571         int wait_on_commit = flags & WAIT;
3572         struct reiserfs_journal_list *jl;
3573         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3574
3575         BUG_ON(!th->t_trans_id);
3576
3577         if (th->t_trans_id != journal->j_trans_id) {
3578                 reiserfs_panic(th->t_super,
3579                                "journal-1577: handle trans id %ld != current trans id %ld\n",
3580                                th->t_trans_id, journal->j_trans_id);
3581         }
3582
3583         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3584         if (atomic_read(&(journal->j_wcount)) > 0) {    /* <= 0 is allowed.  unmounting might not call begin */
3585                 atomic_dec(&(journal->j_wcount));
3586         }
3587
3588         /* BUG, deal with case where j_len is 0, but people previously freed blocks need to be released
3589          ** will be dealt with by next transaction that actually writes something, but should be taken
3590          ** care of in this trans
3591          */
3592         BUG_ON(journal->j_len == 0);
3593
3594         /* if wcount > 0, and we are called to with flush or commit_now,
3595          ** we wait on j_join_wait.  We will wake up when the last writer has
3596          ** finished the transaction, and started it on its way to the disk.
3597          ** Then, we flush the commit or journal list, and just return 0
3598          ** because the rest of journal end was already done for this transaction.
3599          */
3600         if (atomic_read(&(journal->j_wcount)) > 0) {
3601                 if (flush || commit_now) {
3602                         unsigned trans_id;
3603
3604                         jl = journal->j_current_jl;
3605                         trans_id = jl->j_trans_id;
3606                         if (wait_on_commit)
3607                                 jl->j_state |= LIST_COMMIT_PENDING;
3608                         atomic_set(&(journal->j_jlock), 1);
3609                         if (flush) {
3610                                 journal->j_next_full_flush = 1;
3611                         }
3612                         unlock_journal(p_s_sb);
3613
3614                         /* sleep while the current transaction is still j_jlocked */
3615                         while (journal->j_trans_id == trans_id) {
3616                                 if (atomic_read(&journal->j_jlock)) {
3617                                         queue_log_writer(p_s_sb);
3618                                 } else {
3619                                         lock_journal(p_s_sb);
3620                                         if (journal->j_trans_id == trans_id) {
3621                                                 atomic_set(&(journal->j_jlock),
3622                                                            1);
3623                                         }
3624                                         unlock_journal(p_s_sb);
3625                                 }
3626                         }
3627                         BUG_ON(journal->j_trans_id == trans_id);
3628
3629                         if (commit_now
3630                             && journal_list_still_alive(p_s_sb, trans_id)
3631                             && wait_on_commit) {
3632                                 flush_commit_list(p_s_sb, jl, 1);
3633                         }
3634                         return 0;
3635                 }
3636                 unlock_journal(p_s_sb);
3637                 return 0;
3638         }
3639
3640         /* deal with old transactions where we are the last writers */
3641         now = get_seconds();
3642         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3643                 commit_now = 1;
3644                 journal->j_next_async_flush = 1;
3645         }
3646         /* don't batch when someone is waiting on j_join_wait */
3647         /* don't batch when syncing the commit or flushing the whole trans */
3648         if (!(journal->j_must_wait > 0) && !(atomic_read(&(journal->j_jlock)))
3649             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3650             && journal->j_len_alloc < journal->j_max_batch
3651             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3652                 journal->j_bcount++;
3653                 unlock_journal(p_s_sb);
3654                 return 0;
3655         }
3656
3657         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
3658                 reiserfs_panic(p_s_sb,
3659                                "journal-003: journal_end: j_start (%ld) is too high\n",
3660                                journal->j_start);
3661         }
3662         return 1;
3663 }
3664
3665 /*
3666 ** Does all the work that makes deleting blocks safe.
3667 ** when deleting a block mark BH_JNew, just remove it from the current transaction, clean it's buffer_head and move on.
3668 **
3669 ** otherwise:
3670 ** set a bit for the block in the journal bitmap.  That will prevent it from being allocated for unformatted nodes
3671 ** before this transaction has finished.
3672 **
3673 ** mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.  That will prevent any old transactions with
3674 ** this block from trying to flush to the real location.  Since we aren't removing the cnode from the journal_list_hash,
3675 ** the block can't be reallocated yet.
3676 **
3677 ** Then remove it from the current transaction, decrementing any counters and filing it on the clean list.
3678 */
3679 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3680                        struct super_block *p_s_sb, b_blocknr_t blocknr)
3681 {
3682         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3683         struct reiserfs_journal_cnode *cn = NULL;
3684         struct buffer_head *bh = NULL;
3685         struct reiserfs_list_bitmap *jb = NULL;
3686         int cleaned = 0;
3687         BUG_ON(!th->t_trans_id);
3688
3689         cn = get_journal_hash_dev(p_s_sb, journal->j_hash_table, blocknr);
3690         if (cn && cn->bh) {
3691                 bh = cn->bh;
3692                 get_bh(bh);
3693         }
3694         /* if it is journal new, we just remove it from this transaction */
3695         if (bh && buffer_journal_new(bh)) {
3696                 clear_buffer_journal_new(bh);
3697                 clear_prepared_bits(bh);
3698                 reiserfs_clean_and_file_buffer(bh);
3699                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3700         } else {
3701                 /* set the bit for this block in the journal bitmap for this transaction */
3702                 jb = journal->j_current_jl->j_list_bitmap;
3703                 if (!jb) {
3704                         reiserfs_panic(p_s_sb,
3705                                        "journal-1702: journal_mark_freed, journal_list_bitmap is NULL\n");
3706                 }
3707                 set_bit_in_list_bitmap(p_s_sb, blocknr, jb);
3708
3709                 /* Note, the entire while loop is not allowed to schedule.  */
3710
3711                 if (bh) {
3712                         clear_prepared_bits(bh);
3713                         reiserfs_clean_and_file_buffer(bh);
3714                 }
3715                 cleaned = remove_from_transaction(p_s_sb, blocknr, cleaned);
3716
3717                 /* find all older transactions with this block, make sure they don't try to write it out */
3718                 cn = get_journal_hash_dev(p_s_sb, journal->j_list_hash_table,
3719                                           blocknr);
3720                 while (cn) {
3721                         if (p_s_sb == cn->sb && blocknr == cn->blocknr) {
3722                                 set_bit(BLOCK_FREED, &cn->state);
3723                                 if (cn->bh) {
3724                                         if (!cleaned) {
3725                                                 /* remove_from_transaction will brelse the buffer if it was
3726                                                  ** in the current trans
3727                                                  */
3728                                                 clear_buffer_journal_dirty(cn->
3729                                                                            bh);
3730                                                 clear_buffer_dirty(cn->bh);
3731                                                 clear_buffer_journal_test(cn->
3732                                                                           bh);
3733                                                 cleaned = 1;
3734                                                 put_bh(cn->bh);
3735                                                 if (atomic_read
3736                                                     (&(cn->bh->b_count)) < 0) {
3737                                                         reiserfs_warning(p_s_sb,
3738                                                                          "journal-2138: cn->bh->b_count < 0");
3739                                                 }
3740                                         }
3741                                         if (cn->jlist) {        /* since we are clearing the bh, we MUST dec nonzerolen */
3742                                                 atomic_dec(&
3743                                                            (cn->jlist->
3744                                                             j_nonzerolen));
3745                                         }
3746                                         cn->bh = NULL;
3747                                 }
3748                         }
3749                         cn = cn->hnext;
3750                 }
3751         }
3752
3753         if (bh)
3754                 release_buffer_page(bh); /* get_hash grabs the buffer */
3755         return 0;
3756 }
3757
3758 void reiserfs_update_inode_transaction(struct inode *inode)
3759 {
3760         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3761         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3762         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3763 }
3764
3765 /*
3766  * returns -1 on error, 0 if no commits/barriers were done and 1
3767  * if a transaction was actually committed and the barrier was done
3768  */
3769 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3770                              struct reiserfs_journal_list *jl)
3771 {
3772         struct reiserfs_transaction_handle th;
3773         struct super_block *sb = inode->i_sb;
3774         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3775         int ret = 0;
3776
3777         /* is it from the current transaction, or from an unknown transaction? */
3778         if (id == journal->j_trans_id) {
3779                 jl = journal->j_current_jl;
3780                 /* try to let other writers come in and grow this transaction */
3781                 let_transaction_grow(sb, id);
3782                 if (journal->j_trans_id != id) {
3783                         goto flush_commit_only;
3784                 }
3785
3786                 ret = journal_begin(&th, sb, 1);
3787                 if (ret)
3788                         return ret;
3789
3790                 /* someone might have ended this transaction while we joined */
3791                 if (journal->j_trans_id != id) {
3792                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3793                                                      1);
3794                         journal_mark_dirty(&th, sb, SB_BUFFER_WITH_SB(sb));
3795                         ret = journal_end(&th, sb, 1);
3796                         goto flush_commit_only;
3797                 }
3798
3799                 ret = journal_end_sync(&th, sb, 1);
3800                 if (!ret)
3801                         ret = 1;
3802
3803         } else {
3804                 /* this gets tricky, we have to make sure the journal list in
3805                  * the inode still exists.  We know the list is still around
3806                  * if we've got a larger transaction id than the oldest list
3807                  */
3808               flush_commit_only:
3809                 if (journal_list_still_alive(inode->i_sb, id)) {
3810                         /*
3811                          * we only set ret to 1 when we know for sure
3812                          * the barrier hasn't been started yet on the commit
3813                          * block.
3814                          */
3815                         if (atomic_read(&jl->j_commit_left) > 1)
3816                                 ret = 1;
3817                         flush_commit_list(sb, jl, 1);
3818                         if (journal->j_errno)
3819                                 ret = journal->j_errno;
3820                 }
3821         }
3822         /* otherwise the list is gone, and long since committed */
3823         return ret;
3824 }
3825
3826 int reiserfs_commit_for_inode(struct inode *inode)
3827 {
3828         unsigned long id = REISERFS_I(inode)->i_trans_id;
3829         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3830
3831         /* for the whole inode, assume unset id means it was
3832          * changed in the current transaction.  More conservative
3833          */
3834         if (!id || !jl) {
3835                 reiserfs_update_inode_transaction(inode);
3836                 id = REISERFS_I(inode)->i_trans_id;
3837                 /* jl will be updated in __commit_trans_jl */
3838         }
3839
3840         return __commit_trans_jl(inode, id, jl);
3841 }
3842
3843 void reiserfs_restore_prepared_buffer(struct super_block *p_s_sb,
3844                                       struct buffer_head *bh)
3845 {
3846         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3847         PROC_INFO_INC(p_s_sb, journal.restore_prepared);
3848         if (!bh) {
3849                 return;
3850         }
3851         if (test_clear_buffer_journal_restore_dirty(bh) &&
3852             buffer_journal_dirty(bh)) {
3853                 struct reiserfs_journal_cnode *cn;
3854                 cn = get_journal_hash_dev(p_s_sb,
3855                                           journal->j_list_hash_table,
3856                                           bh->b_blocknr);
3857                 if (cn && can_dirty(cn)) {
3858                         set_buffer_journal_test(bh);
3859                         mark_buffer_dirty(bh);
3860                 }
3861         }
3862         clear_buffer_journal_prepared(bh);
3863 }
3864
3865 extern struct tree_balance *cur_tb;
3866 /*
3867 ** before we can change a metadata block, we have to make sure it won't
3868 ** be written to disk while we are altering it.  So, we must:
3869 ** clean it
3870 ** wait on it.
3871 **
3872 */
3873 int reiserfs_prepare_for_journal(struct super_block *p_s_sb,
3874                                  struct buffer_head *bh, int wait)
3875 {
3876         PROC_INFO_INC(p_s_sb, journal.prepare);
3877
3878         if (test_set_buffer_locked(bh)) {
3879                 if (!wait)
3880                         return 0;
3881                 lock_buffer(bh);
3882         }
3883         set_buffer_journal_prepared(bh);
3884         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3885                 clear_buffer_journal_test(bh);
3886                 set_buffer_journal_restore_dirty(bh);
3887         }
3888         unlock_buffer(bh);
3889         return 1;
3890 }
3891
3892 static void flush_old_journal_lists(struct super_block *s)
3893 {
3894         struct reiserfs_journal *journal = SB_JOURNAL(s);
3895         struct reiserfs_journal_list *jl;
3896         struct list_head *entry;
3897         time_t now = get_seconds();
3898
3899         while (!list_empty(&journal->j_journal_list)) {
3900                 entry = journal->j_journal_list.next;
3901                 jl = JOURNAL_LIST_ENTRY(entry);
3902                 /* this check should always be run, to send old lists to disk */
3903                 if (jl->j_timestamp < (now - (JOURNAL_MAX_TRANS_AGE * 4)) &&
3904                     atomic_read(&jl->j_commit_left) == 0 &&
3905                     test_transaction(s, jl)) {
3906                         flush_used_journal_lists(s, jl);
3907                 } else {
3908                         break;
3909                 }
3910         }
3911 }
3912
3913 /*
3914 ** long and ugly.  If flush, will not return until all commit
3915 ** blocks and all real buffers in the trans are on disk.
3916 ** If no_async, won't return until all commit blocks are on disk.
3917 **
3918 ** keep reading, there are comments as you go along
3919 **
3920 ** If the journal is aborted, we just clean up. Things like flushing
3921 ** journal lists, etc just won't happen.
3922 */
3923 static int do_journal_end(struct reiserfs_transaction_handle *th,
3924                           struct super_block *p_s_sb, unsigned long nblocks,
3925                           int flags)
3926 {
3927         struct reiserfs_journal *journal = SB_JOURNAL(p_s_sb);
3928         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3929         struct reiserfs_journal_cnode *last_cn = NULL;
3930         struct reiserfs_journal_desc *desc;
3931         struct reiserfs_journal_commit *commit;
3932         struct buffer_head *c_bh;       /* commit bh */
3933         struct buffer_head *d_bh;       /* desc bh */
3934         int cur_write_start = 0;        /* start index of current log write */
3935         int old_start;
3936         int i;
3937         int flush;
3938         int wait_on_commit;
3939         struct reiserfs_journal_list *jl, *temp_jl;
3940         struct list_head *entry, *safe;
3941         unsigned long jindex;
3942         unsigned long commit_trans_id;
3943         int trans_half;
3944
3945         BUG_ON(th->t_refcount > 1);
3946         BUG_ON(!th->t_trans_id);
3947
3948         /* protect flush_older_commits from doing mistakes if the
3949            transaction ID counter gets overflowed.  */
3950         if (th->t_trans_id == ~0UL)
3951                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3952         flush = flags & FLUSH_ALL;
3953         wait_on_commit = flags & WAIT;
3954
3955         put_fs_excl();
3956         current->journal_info = th->t_handle_save;
3957         reiserfs_check_lock_depth(p_s_sb, "journal end");
3958         if (journal->j_len == 0) {
3959                 reiserfs_prepare_for_journal(p_s_sb, SB_BUFFER_WITH_SB(p_s_sb),
3960                                              1);
3961                 journal_mark_dirty(th, p_s_sb, SB_BUFFER_WITH_SB(p_s_sb));
3962         }
3963
3964         lock_journal(p_s_sb);
3965         if (journal->j_next_full_flush) {
3966                 flags |= FLUSH_ALL;
3967                 flush = 1;
3968         }
3969         if (journal->j_next_async_flush) {
3970                 flags |= COMMIT_NOW | WAIT;
3971                 wait_on_commit = 1;
3972         }
3973
3974         /* check_journal_end locks the journal, and unlocks if it does not return 1
3975          ** it tells us if we should continue with the journal_end, or just return
3976          */
3977         if (!check_journal_end(th, p_s_sb, nblocks, flags)) {
3978                 p_s_sb->s_dirt = 1;
3979                 wake_queued_writers(p_s_sb);
3980                 reiserfs_async_progress_wait(p_s_sb);
3981                 goto out;
3982         }
3983
3984         /* check_journal_end might set these, check again */
3985         if (journal->j_next_full_flush) {
3986                 flush = 1;
3987         }
3988
3989         /*
3990          ** j must wait means we have to flush the log blocks, and the real blocks for
3991          ** this transaction
3992          */
3993         if (journal->j_must_wait > 0) {
3994                 flush = 1;
3995         }
3996 #ifdef REISERFS_PREALLOCATE
3997         /* quota ops might need to nest, setup the journal_info pointer for them
3998          * and raise the refcount so that it is > 0. */
3999         current->journal_info = th;
4000         th->t_refcount++;
4001         reiserfs_discard_all_prealloc(th);      /* it should not involve new blocks into
4002                                                  * the transaction */
4003         th->t_refcount--;
4004         current->journal_info = th->t_handle_save;
4005 #endif
4006
4007         /* setup description block */
4008         d_bh =
4009             journal_getblk(p_s_sb,
4010                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4011                            journal->j_start);
4012         set_buffer_uptodate(d_bh);
4013         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4014         memset(d_bh->b_data, 0, d_bh->b_size);
4015         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4016         set_desc_trans_id(desc, journal->j_trans_id);
4017
4018         /* setup commit block.  Don't write (keep it clean too) this one until after everyone else is written */
4019         c_bh = journal_getblk(p_s_sb, SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4020                               ((journal->j_start + journal->j_len +
4021                                 1) % SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4022         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4023         memset(c_bh->b_data, 0, c_bh->b_size);
4024         set_commit_trans_id(commit, journal->j_trans_id);
4025         set_buffer_uptodate(c_bh);
4026
4027         /* init this journal list */
4028         jl = journal->j_current_jl;
4029
4030         /* we lock the commit before doing anything because
4031          * we want to make sure nobody tries to run flush_commit_list until
4032          * the new transaction is fully setup, and we've already flushed the
4033          * ordered bh list
4034          */
4035         down(&jl->j_commit_lock);
4036
4037         /* save the transaction id in case we need to commit it later */
4038         commit_trans_id = jl->j_trans_id;
4039
4040         atomic_set(&jl->j_older_commits_done, 0);
4041         jl->j_trans_id = journal->j_trans_id;
4042         jl->j_timestamp = journal->j_trans_start_time;
4043         jl->j_commit_bh = c_bh;
4044         jl->j_start = journal->j_start;
4045         jl->j_len = journal->j_len;
4046         atomic_set(&jl->j_nonzerolen, journal->j_len);
4047         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4048         jl->j_realblock = NULL;
4049
4050         /* The ENTIRE FOR LOOP MUST not cause schedule to occur.
4051          **  for each real block, add it to the journal list hash,
4052          ** copy into real block index array in the commit or desc block
4053          */
4054         trans_half = journal_trans_half(p_s_sb->s_blocksize);
4055         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4056                 if (buffer_journaled(cn->bh)) {
4057                         jl_cn = get_cnode(p_s_sb);
4058                         if (!jl_cn) {
4059                                 reiserfs_panic(p_s_sb,
4060                                                "journal-1676, get_cnode returned NULL\n");
4061                         }
4062                         if (i == 0) {
4063                                 jl->j_realblock = jl_cn;
4064                         }
4065                         jl_cn->prev = last_cn;
4066                         jl_cn->next = NULL;
4067                         if (last_cn) {
4068                                 last_cn->next = jl_cn;
4069                         }
4070                         last_cn = jl_cn;
4071                         /* make sure the block we are trying to log is not a block
4072                            of journal or reserved area */
4073
4074                         if (is_block_in_log_or_reserved_area
4075                             (p_s_sb, cn->bh->b_blocknr)) {
4076                                 reiserfs_panic(p_s_sb,
4077                                                "journal-2332: Trying to log block %lu, which is a log block\n",
4078                                                cn->bh->b_blocknr);
4079                         }
4080                         jl_cn->blocknr = cn->bh->b_blocknr;
4081                         jl_cn->state = 0;
4082                         jl_cn->sb = p_s_sb;
4083                         jl_cn->bh = cn->bh;
4084                         jl_cn->jlist = jl;
4085                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4086                         if (i < trans_half) {
4087                                 desc->j_realblock[i] =
4088                                     cpu_to_le32(cn->bh->b_blocknr);
4089                         } else {
4090                                 commit->j_realblock[i - trans_half] =
4091                                     cpu_to_le32(cn->bh->b_blocknr);
4092                         }
4093                 } else {
4094                         i--;
4095                 }
4096         }
4097         set_desc_trans_len(desc, journal->j_len);
4098         set_desc_mount_id(desc, journal->j_mount_id);
4099         set_desc_trans_id(desc, journal->j_trans_id);
4100         set_commit_trans_len(commit, journal->j_len);
4101
4102         /* special check in case all buffers in the journal were marked for not logging */
4103         BUG_ON(journal->j_len == 0);
4104
4105         /* we're about to dirty all the log blocks, mark the description block
4106          * dirty now too.  Don't mark the commit block dirty until all the
4107          * others are on disk
4108          */
4109         mark_buffer_dirty(d_bh);
4110
4111         /* first data block is j_start + 1, so add one to cur_write_start wherever you use it */
4112         cur_write_start = journal->j_start;
4113         cn = journal->j_first;
4114         jindex = 1;             /* start at one so we don't get the desc again */
4115         while (cn) {
4116                 clear_buffer_journal_new(cn->bh);
4117                 /* copy all the real blocks into log area.  dirty log blocks */
4118                 if (buffer_journaled(cn->bh)) {
4119                         struct buffer_head *tmp_bh;
4120                         char *addr;
4121                         struct page *page;
4122                         tmp_bh =
4123                             journal_getblk(p_s_sb,
4124                                            SB_ONDISK_JOURNAL_1st_BLOCK(p_s_sb) +
4125                                            ((cur_write_start +
4126                                              jindex) %
4127                                             SB_ONDISK_JOURNAL_SIZE(p_s_sb)));
4128                         set_buffer_uptodate(tmp_bh);
4129                         page = cn->bh->b_page;
4130                         addr = kmap(page);
4131                         memcpy(tmp_bh->b_data,
4132                                addr + offset_in_page(cn->bh->b_data),
4133                                cn->bh->b_size);
4134                         kunmap(page);
4135                         mark_buffer_dirty(tmp_bh);
4136                         jindex++;
4137                         set_buffer_journal_dirty(cn->bh);
4138                         clear_buffer_journaled(cn->bh);
4139                 } else {
4140                         /* JDirty cleared sometime during transaction.  don't log this one */
4141                         reiserfs_warning(p_s_sb,
4142                                          "journal-2048: do_journal_end: BAD, buffer in journal hash, but not JDirty!");
4143                         brelse(cn->bh);
4144                 }
4145                 next = cn->next;
4146                 free_cnode(p_s_sb, cn);
4147                 cn = next;
4148                 cond_resched();
4149         }
4150
4151         /* we are done  with both the c_bh and d_bh, but
4152          ** c_bh must be written after all other commit blocks,
4153          ** so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4154          */
4155
4156         journal->j_current_jl = alloc_journal_list(p_s_sb);
4157
4158         /* now it is safe to insert this transaction on the main list */
4159         list_add_tail(&jl->j_list, &journal->j_journal_list);
4160         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4161         journal->j_num_work_lists++;
4162
4163         /* reset journal values for the next transaction */
4164         old_start = journal->j_start;
4165         journal->j_start =
4166             (journal->j_start + journal->j_len +
4167              2) % SB_ONDISK_JOURNAL_SIZE(p_s_sb);
4168         atomic_set(&(journal->j_wcount), 0);
4169         journal->j_bcount = 0;
4170         journal->j_last = NULL;
4171         journal->j_first = NULL;
4172         journal->j_len = 0;
4173         journal->j_trans_start_time = 0;
4174         /* check for trans_id overflow */
4175         if (++journal->j_trans_id == 0)
4176                 journal->j_trans_id = 10;
4177         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4178         journal->j_must_wait = 0;
4179         journal->j_len_alloc = 0;
4180         journal->j_next_full_flush = 0;
4181         journal->j_next_async_flush = 0;
4182         init_journal_hash(p_s_sb);
4183
4184         // make sure reiserfs_add_jh sees the new current_jl before we
4185         // write out the tails
4186         smp_mb();
4187
4188         /* tail conversion targets have to hit the disk before we end the
4189          * transaction.  Otherwise a later transaction might repack the tail
4190          * before this transaction commits, leaving the data block unflushed and
4191          * clean, if we crash before the later transaction commits, the data block
4192          * is lost.
4193          */
4194         if (!list_empty(&jl->j_tail_bh_list)) {
4195                 unlock_kernel();
4196                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4197                                       journal, jl, &jl->j_tail_bh_list);
4198                 lock_kernel();
4199         }
4200         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4201         up(&jl->j_commit_lock);
4202
4203         /* honor the flush wishes from the caller, simple commits can
4204          ** be done outside the journal lock, they are done below
4205          **
4206          ** if we don't flush the commit list right now, we put it into
4207          ** the work queue so the people waiting on the async progress work
4208          ** queue don't wait for this proc to flush journal lists and such.
4209          */
4210         if (flush) {
4211                 flush_commit_list(p_s_sb, jl, 1);
4212                 flush_journal_list(p_s_sb, jl, 1);
4213         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4214                 queue_delayed_work(commit_wq, &journal->j_work, HZ / 10);
4215
4216         /* if the next transaction has any chance of wrapping, flush
4217          ** transactions that might get overwritten.  If any journal lists are very
4218          ** old flush them as well.
4219          */
4220       first_jl:
4221         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4222                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4223                 if (journal->j_start <= temp_jl->j_start) {
4224                         if ((journal->j_start + journal->j_trans_max + 1) >=
4225                             temp_jl->j_start) {
4226                                 flush_used_journal_lists(p_s_sb, temp_jl);
4227                                 goto first_jl;
4228                         } else if ((journal->j_start +
4229                                     journal->j_trans_max + 1) <
4230                                    SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4231                                 /* if we don't cross into the next transaction and we don't
4232                                  * wrap, there is no way we can overlap any later transactions
4233                                  * break now
4234                                  */
4235                                 break;
4236                         }
4237                 } else if ((journal->j_start +
4238                             journal->j_trans_max + 1) >
4239                            SB_ONDISK_JOURNAL_SIZE(p_s_sb)) {
4240                         if (((journal->j_start + journal->j_trans_max + 1) %
4241                              SB_ONDISK_JOURNAL_SIZE(p_s_sb)) >=
4242                             temp_jl->j_start) {
4243                                 flush_used_journal_lists(p_s_sb, temp_jl);
4244                                 goto first_jl;
4245                         } else {
4246                                 /* we don't overlap anything from out start to the end of the
4247                                  * log, and our wrapped portion doesn't overlap anything at
4248                                  * the start of the log.  We can break
4249                                  */
4250                                 break;
4251                         }
4252                 }
4253         }
4254         flush_old_journal_lists(p_s_sb);
4255
4256         journal->j_current_jl->j_list_bitmap =
4257             get_list_bitmap(p_s_sb, journal->j_current_jl);
4258
4259         if (!(journal->j_current_jl->j_list_bitmap)) {
4260                 reiserfs_panic(p_s_sb,
4261                                "journal-1996: do_journal_end, could not get a list bitmap\n");
4262         }
4263
4264         atomic_set(&(journal->j_jlock), 0);
4265         unlock_journal(p_s_sb);
4266         /* wake up any body waiting to join. */
4267         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4268         wake_up(&(journal->j_join_wait));
4269
4270         if (!flush && wait_on_commit &&
4271             journal_list_still_alive(p_s_sb, commit_trans_id)) {
4272                 flush_commit_list(p_s_sb, jl, 1);
4273         }
4274       out:
4275         reiserfs_check_lock_depth(p_s_sb, "journal end2");
4276
4277         memset(th, 0, sizeof(*th));
4278         /* Re-set th->t_super, so we can properly keep track of how many
4279          * persistent transactions there are. We need to do this so if this
4280          * call is part of a failed restart_transaction, we can free it later */
4281         th->t_super = p_s_sb;
4282
4283         return journal->j_errno;
4284 }
4285
4286 static void __reiserfs_journal_abort_hard(struct super_block *sb)
4287 {
4288         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4289         if (test_bit(J_ABORTED, &journal->j_state))
4290                 return;
4291
4292         printk(KERN_CRIT "REISERFS: Aborting journal for filesystem on %s\n",
4293                reiserfs_bdevname(sb));
4294
4295         sb->s_flags |= MS_RDONLY;
4296         set_bit(J_ABORTED, &journal->j_state);
4297
4298 #ifdef CONFIG_REISERFS_CHECK
4299         dump_stack();
4300 #endif
4301 }
4302
4303 static void __reiserfs_journal_abort_soft(struct super_block *sb, int errno)
4304 {
4305         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4306         if (test_bit(J_ABORTED, &journal->j_state))
4307                 return;
4308
4309         if (!journal->j_errno)
4310                 journal->j_errno = errno;
4311
4312         __reiserfs_journal_abort_hard(sb);
4313 }
4314
4315 void reiserfs_journal_abort(struct super_block *sb, int errno)
4316 {
4317         return __reiserfs_journal_abort_soft(sb, errno);
4318 }