fs/reiserfs/journal.c

   1 /*
   2  * Write ahead logging implementation copyright Chris Mason 2000
   3  *
   4  * The background commits make this code very interrelated, and
   5  * overly complex.  I need to rethink things a bit....The major players:
   6  *
   7  * journal_begin -- call with the number of blocks you expect to log.
   8  *                  If the current transaction is too
   9  *                  old, it will block until the current transaction is
  10  *                  finished, and then start a new one.
  11  *                  Usually, your transaction will get joined in with
  12  *                  previous ones for speed.
  13  *
  14  * journal_join  -- same as journal_begin, but won't block on the current
  15  *                  transaction regardless of age.  Don't ever call
  16  *                  this.  Ever.  There are only two places it should be
  17  *                  called from, and they are both inside this file.
  18  *
  19  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  20  *                       that might make them get sent to disk
  21  *                       and then marks them BH_JDirty.  Puts the buffer head
  22  *                       into the current transaction hash.
  23  *
  24  * journal_end -- if the current transaction is batchable, it does nothing
  25  *                   otherwise, it could do an async/synchronous commit, or
  26  *                   a full flush of all log and real blocks in the
  27  *                   transaction.
  28  *
  29  * flush_old_commits -- if the current transaction is too old, it is ended and
  30  *                      commit blocks are sent to disk.  Forces commit blocks
  31  *                      to disk for all backgrounded commits that have been
  32  *                      around too long.
  33  *                   -- Note, if you call this as an immediate flush from
  34  *                      from within kupdate, it will ignore the immediate flag
  35  */
  36
  37 #include <linux/time.h>
  38 #include <linux/semaphore.h>
  39 #include <linux/vmalloc.h>
  40 #include "reiserfs.h"
  41 #include <linux/kernel.h>
  42 #include <linux/errno.h>
  43 #include <linux/fcntl.h>
  44 #include <linux/stat.h>
  45 #include <linux/string.h>
  46 #include <linux/buffer_head.h>
  47 #include <linux/workqueue.h>
  48 #include <linux/writeback.h>
  49 #include <linux/blkdev.h>
  50 #include <linux/backing-dev.h>
  51 #include <linux/uaccess.h>
  52 #include <linux/slab.h>
  53
  54
  55 /* gets a struct reiserfs_journal_list * from a list head */
  56 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  57                                j_list))
  58 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  59                                j_working_list))
  60
  61 /* must be correct to keep the desc and commit structs at 4k */
  62 #define JOURNAL_TRANS_HALF 1018
  63 #define BUFNR 64                /*read ahead */
  64
  65 /* cnode stat bits.  Move these into reiserfs_fs.h */
  66
  67 /* this block was freed, and can't be written.  */
  68 #define BLOCK_FREED 2
  69 /* this block was freed during this transaction, and can't be written */
  70 #define BLOCK_FREED_HOLDER 3
  71
  72 /* used in flush_journal_list */
  73 #define BLOCK_NEEDS_FLUSH 4
  74 #define BLOCK_DIRTIED 5
  75
  76 /* journal list state bits */
  77 #define LIST_TOUCHED 1
  78 #define LIST_DIRTY   2
  79 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  80
  81 /* flags for do_journal_end */
  82 #define FLUSH_ALL   1           /* flush commit and real blocks */
  83 #define COMMIT_NOW  2           /* end and commit this transaction */
  84 #define WAIT        4           /* wait for the log blocks to hit the disk */
  85
  86 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  87 static int flush_journal_list(struct super_block *s,
  88                               struct reiserfs_journal_list *jl, int flushall);
  89 static int flush_commit_list(struct super_block *s,
  90                              struct reiserfs_journal_list *jl, int flushall);
  91 static int can_dirty(struct reiserfs_journal_cnode *cn);
  92 static int journal_join(struct reiserfs_transaction_handle *th,
  93                         struct super_block *sb);
  94 static void release_journal_dev(struct super_block *super,
  95                                struct reiserfs_journal *journal);
  96 static int dirty_one_transaction(struct super_block *s,
  97                                  struct reiserfs_journal_list *jl);
  98 static void flush_async_commits(struct work_struct *work);
  99 static void queue_log_writer(struct super_block *s);
 100
 101 /* values for join in do_journal_begin_r */
 102 enum {
 103         JBEGIN_REG = 0,         /* regular journal begin */
 104         /* join the running transaction if at all possible */
 105         JBEGIN_JOIN = 1,
 106         /* called from cleanup code, ignores aborted flag */
 107         JBEGIN_ABORT = 2,
 108 };
 109
 110 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 111                               struct super_block *sb,
 112                               unsigned long nblocks, int join);
 113
 114 static void init_journal_hash(struct super_block *sb)
 115 {
 116         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 117         memset(journal->j_hash_table, 0,
 118                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 119 }
 120
 121 /*
 122  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 123  * I can't allow refile_buffer to make schedule happen after I've freed a
 124  * block.  Look at remove_from_transaction and journal_mark_freed for
 125  * more details.
 126  */
 127 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 128 {
 129         if (bh) {
 130                 clear_buffer_dirty(bh);
 131                 clear_buffer_journal_test(bh);
 132         }
 133         return 0;
 134 }
 135
 136 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 137                                                          *sb)
 138 {
 139         struct reiserfs_bitmap_node *bn;
 140         static int id;
 141
 142         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 143         if (!bn) {
 144                 return NULL;
 145         }
 146         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 147         if (!bn->data) {
 148                 kfree(bn);
 149                 return NULL;
 150         }
 151         bn->id = id++;
 152         INIT_LIST_HEAD(&bn->list);
 153         return bn;
 154 }
 155
 156 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 157 {
 158         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 159         struct reiserfs_bitmap_node *bn = NULL;
 160         struct list_head *entry = journal->j_bitmap_nodes.next;
 161
 162         journal->j_used_bitmap_nodes++;
 163 repeat:
 164
 165         if (entry != &journal->j_bitmap_nodes) {
 166                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 167                 list_del(entry);
 168                 memset(bn->data, 0, sb->s_blocksize);
 169                 journal->j_free_bitmap_nodes--;
 170                 return bn;
 171         }
 172         bn = allocate_bitmap_node(sb);
 173         if (!bn) {
 174                 yield();
 175                 goto repeat;
 176         }
 177         return bn;
 178 }
 179 static inline void free_bitmap_node(struct super_block *sb,
 180                                     struct reiserfs_bitmap_node *bn)
 181 {
 182         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 183         journal->j_used_bitmap_nodes--;
 184         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 185                 kfree(bn->data);
 186                 kfree(bn);
 187         } else {
 188                 list_add(&bn->list, &journal->j_bitmap_nodes);
 189                 journal->j_free_bitmap_nodes++;
 190         }
 191 }
 192
 193 static void allocate_bitmap_nodes(struct super_block *sb)
 194 {
 195         int i;
 196         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 197         struct reiserfs_bitmap_node *bn = NULL;
 198         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 199                 bn = allocate_bitmap_node(sb);
 200                 if (bn) {
 201                         list_add(&bn->list, &journal->j_bitmap_nodes);
 202                         journal->j_free_bitmap_nodes++;
 203                 } else {
 204                         /* this is ok, we'll try again when more are needed */
 205                         break;
 206                 }
 207         }
 208 }
 209
 210 static int set_bit_in_list_bitmap(struct super_block *sb,
 211                                   b_blocknr_t block,
 212                                   struct reiserfs_list_bitmap *jb)
 213 {
 214         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 215         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 216
 217         if (!jb->bitmaps[bmap_nr]) {
 218                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 219         }
 220         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 221         return 0;
 222 }
 223
 224 static void cleanup_bitmap_list(struct super_block *sb,
 225                                 struct reiserfs_list_bitmap *jb)
 226 {
 227         int i;
 228         if (jb->bitmaps == NULL)
 229                 return;
 230
 231         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 232                 if (jb->bitmaps[i]) {
 233                         free_bitmap_node(sb, jb->bitmaps[i]);
 234                         jb->bitmaps[i] = NULL;
 235                 }
 236         }
 237 }
 238
 239 /*
 240  * only call this on FS unmount.
 241  */
 242 static int free_list_bitmaps(struct super_block *sb,
 243                              struct reiserfs_list_bitmap *jb_array)
 244 {
 245         int i;
 246         struct reiserfs_list_bitmap *jb;
 247         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 248                 jb = jb_array + i;
 249                 jb->journal_list = NULL;
 250                 cleanup_bitmap_list(sb, jb);
 251                 vfree(jb->bitmaps);
 252                 jb->bitmaps = NULL;
 253         }
 254         return 0;
 255 }
 256
 257 static int free_bitmap_nodes(struct super_block *sb)
 258 {
 259         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 260         struct list_head *next = journal->j_bitmap_nodes.next;
 261         struct reiserfs_bitmap_node *bn;
 262
 263         while (next != &journal->j_bitmap_nodes) {
 264                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 265                 list_del(next);
 266                 kfree(bn->data);
 267                 kfree(bn);
 268                 next = journal->j_bitmap_nodes.next;
 269                 journal->j_free_bitmap_nodes--;
 270         }
 271
 272         return 0;
 273 }
 274
 275 /*
 276  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 277  * jb_array is the array to be filled in.
 278  */
 279 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 280                                    struct reiserfs_list_bitmap *jb_array,
 281                                    unsigned int bmap_nr)
 282 {
 283         int i;
 284         int failed = 0;
 285         struct reiserfs_list_bitmap *jb;
 286         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 287
 288         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 289                 jb = jb_array + i;
 290                 jb->journal_list = NULL;
 291                 jb->bitmaps = vzalloc(mem);
 292                 if (!jb->bitmaps) {
 293                         reiserfs_warning(sb, "clm-2000", "unable to "
 294                                          "allocate bitmaps for journal lists");
 295                         failed = 1;
 296                         break;
 297                 }
 298         }
 299         if (failed) {
 300                 free_list_bitmaps(sb, jb_array);
 301                 return -1;
 302         }
 303         return 0;
 304 }
 305
 306 /*
 307  * find an available list bitmap.  If you can't find one, flush a commit list
 308  * and try again
 309  */
 310 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 311                                                     struct reiserfs_journal_list
 312                                                     *jl)
 313 {
 314         int i, j;
 315         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 316         struct reiserfs_list_bitmap *jb = NULL;
 317
 318         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 319                 i = journal->j_list_bitmap_index;
 320                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 321                 jb = journal->j_list_bitmap + i;
 322                 if (journal->j_list_bitmap[i].journal_list) {
 323                         flush_commit_list(sb,
 324                                           journal->j_list_bitmap[i].
 325                                           journal_list, 1);
 326                         if (!journal->j_list_bitmap[i].journal_list) {
 327                                 break;
 328                         }
 329                 } else {
 330                         break;
 331                 }
 332         }
 333         /* double check to make sure if flushed correctly */
 334         if (jb->journal_list)
 335                 return NULL;
 336         jb->journal_list = jl;
 337         return jb;
 338 }
 339
 340 /*
 341  * allocates a new chunk of X nodes, and links them all together as a list.
 342  * Uses the cnode->next and cnode->prev pointers
 343  * returns NULL on failure
 344  */
 345 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 346 {
 347         struct reiserfs_journal_cnode *head;
 348         int i;
 349         if (num_cnodes <= 0) {
 350                 return NULL;
 351         }
 352         head = vzalloc(num_cnodes * sizeof(struct reiserfs_journal_cnode));
 353         if (!head) {
 354                 return NULL;
 355         }
 356         head[0].prev = NULL;
 357         head[0].next = head + 1;
 358         for (i = 1; i < num_cnodes; i++) {
 359                 head[i].prev = head + (i - 1);
 360                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 361         }
 362         head[num_cnodes - 1].next = NULL;
 363         return head;
 364 }
 365
 366 /* pulls a cnode off the free list, or returns NULL on failure */
 367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 368 {
 369         struct reiserfs_journal_cnode *cn;
 370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 371
 372         reiserfs_check_lock_depth(sb, "get_cnode");
 373
 374         if (journal->j_cnode_free <= 0) {
 375                 return NULL;
 376         }
 377         journal->j_cnode_used++;
 378         journal->j_cnode_free--;
 379         cn = journal->j_cnode_free_list;
 380         if (!cn) {
 381                 return cn;
 382         }
 383         if (cn->next) {
 384                 cn->next->prev = NULL;
 385         }
 386         journal->j_cnode_free_list = cn->next;
 387         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 388         return cn;
 389 }
 390
 391 /*
 392  * returns a cnode to the free list
 393  */
 394 static void free_cnode(struct super_block *sb,
 395                        struct reiserfs_journal_cnode *cn)
 396 {
 397         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 398
 399         reiserfs_check_lock_depth(sb, "free_cnode");
 400
 401         journal->j_cnode_used--;
 402         journal->j_cnode_free++;
 403         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 404         cn->next = journal->j_cnode_free_list;
 405         if (journal->j_cnode_free_list) {
 406                 journal->j_cnode_free_list->prev = cn;
 407         }
 408         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 409         journal->j_cnode_free_list = cn;
 410 }
 411
 412 static void clear_prepared_bits(struct buffer_head *bh)
 413 {
 414         clear_buffer_journal_prepared(bh);
 415         clear_buffer_journal_restore_dirty(bh);
 416 }
 417
 418 /*
 419  * return a cnode with same dev, block number and size in table,
 420  * or null if not found
 421  */
 422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 423                                                                   super_block
 424                                                                   *sb,
 425                                                                   struct
 426                                                                   reiserfs_journal_cnode
 427                                                                   **table,
 428                                                                   long bl)
 429 {
 430         struct reiserfs_journal_cnode *cn;
 431         cn = journal_hash(table, sb, bl);
 432         while (cn) {
 433                 if (cn->blocknr == bl && cn->sb == sb)
 434                         return cn;
 435                 cn = cn->hnext;
 436         }
 437         return (struct reiserfs_journal_cnode *)0;
 438 }
 439
 440 /*
 441  * this actually means 'can this block be reallocated yet?'.  If you set
 442  * search_all, a block can only be allocated if it is not in the current
 443  * transaction, was not freed by the current transaction, and has no chance
 444  * of ever being overwritten by a replay after crashing.
 445  *
 446  * If you don't set search_all, a block can only be allocated if it is not
 447  * in the current transaction.  Since deleting a block removes it from the
 448  * current transaction, this case should never happen.  If you don't set
 449  * search_all, make sure you never write the block without logging it.
 450  *
 451  * next_zero_bit is a suggestion about the next block to try for find_forward.
 452  * when bl is rejected because it is set in a journal list bitmap, we search
 453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
 454  * that through next_zero_bit for find_forward to try.
 455  *
 456  * Just because we return something in next_zero_bit does not mean we won't
 457  * reject it on the next call to reiserfs_in_journal
 458  */
 459 int reiserfs_in_journal(struct super_block *sb,
 460                         unsigned int bmap_nr, int bit_nr, int search_all,
 461                         b_blocknr_t * next_zero_bit)
 462 {
 463         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 464         struct reiserfs_journal_cnode *cn;
 465         struct reiserfs_list_bitmap *jb;
 466         int i;
 467         unsigned long bl;
 468
 469         *next_zero_bit = 0;     /* always start this at zero. */
 470
 471         PROC_INFO_INC(sb, journal.in_journal);
 472         /*
 473          * If we aren't doing a search_all, this is a metablock, and it
 474          * will be logged before use.  if we crash before the transaction
 475          * that freed it commits,  this transaction won't have committed
 476          * either, and the block will never be written
 477          */
 478         if (search_all) {
 479                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 480                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 481                         jb = journal->j_list_bitmap + i;
 482                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 483                             test_bit(bit_nr,
 484                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 485                                      data)) {
 486                                 *next_zero_bit =
 487                                     find_next_zero_bit((unsigned long *)
 488                                                        (jb->bitmaps[bmap_nr]->
 489                                                         data),
 490                                                        sb->s_blocksize << 3,
 491                                                        bit_nr + 1);
 492                                 return 1;
 493                         }
 494                 }
 495         }
 496
 497         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 498         /* is it in any old transactions? */
 499         if (search_all
 500             && (cn =
 501                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 502                 return 1;
 503         }
 504
 505         /* is it in the current transaction.  This should never happen */
 506         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 507                 BUG();
 508                 return 1;
 509         }
 510
 511         PROC_INFO_INC(sb, journal.in_journal_reusable);
 512         /* safe for reuse */
 513         return 0;
 514 }
 515
 516 /* insert cn into table */
 517 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 518                                        struct reiserfs_journal_cnode *cn)
 519 {
 520         struct reiserfs_journal_cnode *cn_orig;
 521
 522         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 523         cn->hnext = cn_orig;
 524         cn->hprev = NULL;
 525         if (cn_orig) {
 526                 cn_orig->hprev = cn;
 527         }
 528         journal_hash(table, cn->sb, cn->blocknr) = cn;
 529 }
 530
 531 /* lock the current transaction */
 532 static inline void lock_journal(struct super_block *sb)
 533 {
 534         PROC_INFO_INC(sb, journal.lock_journal);
 535
 536         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 537 }
 538
 539 /* unlock the current transaction */
 540 static inline void unlock_journal(struct super_block *sb)
 541 {
 542         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 543 }
 544
 545 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 546 {
 547         jl->j_refcount++;
 548 }
 549
 550 static inline void put_journal_list(struct super_block *s,
 551                                     struct reiserfs_journal_list *jl)
 552 {
 553         if (jl->j_refcount < 1) {
 554                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 555                                jl->j_trans_id, jl->j_refcount);
 556         }
 557         if (--jl->j_refcount == 0)
 558                 kfree(jl);
 559 }
 560
 561 /*
 562  * this used to be much more involved, and I'm keeping it just in case
 563  * things get ugly again.  it gets called by flush_commit_list, and
 564  * cleans up any data stored about blocks freed during a transaction.
 565  */
 566 static void cleanup_freed_for_journal_list(struct super_block *sb,
 567                                            struct reiserfs_journal_list *jl)
 568 {
 569
 570         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 571         if (jb) {
 572                 cleanup_bitmap_list(sb, jb);
 573         }
 574         jl->j_list_bitmap->journal_list = NULL;
 575         jl->j_list_bitmap = NULL;
 576 }
 577
 578 static int journal_list_still_alive(struct super_block *s,
 579                                     unsigned int trans_id)
 580 {
 581         struct reiserfs_journal *journal = SB_JOURNAL(s);
 582         struct list_head *entry = &journal->j_journal_list;
 583         struct reiserfs_journal_list *jl;
 584
 585         if (!list_empty(entry)) {
 586                 jl = JOURNAL_LIST_ENTRY(entry->next);
 587                 if (jl->j_trans_id <= trans_id) {
 588                         return 1;
 589                 }
 590         }
 591         return 0;
 592 }
 593
 594 /*
 595  * If page->mapping was null, we failed to truncate this page for
 596  * some reason.  Most likely because it was truncated after being
 597  * logged via data=journal.
 598  *
 599  * This does a check to see if the buffer belongs to one of these
 600  * lost pages before doing the final put_bh.  If page->mapping was
 601  * null, it tries to free buffers on the page, which should make the
 602  * final page_cache_release drop the page from the lru.
 603  */
 604 static void release_buffer_page(struct buffer_head *bh)
 605 {
 606         struct page *page = bh->b_page;
 607         if (!page->mapping && trylock_page(page)) {
 608                 page_cache_get(page);
 609                 put_bh(bh);
 610                 if (!page->mapping)
 611                         try_to_free_buffers(page);
 612                 unlock_page(page);
 613                 page_cache_release(page);
 614         } else {
 615                 put_bh(bh);
 616         }
 617 }
 618
 619 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 620 {
 621         char b[BDEVNAME_SIZE];
 622
 623         if (buffer_journaled(bh)) {
 624                 reiserfs_warning(NULL, "clm-2084",
 625                                  "pinned buffer %lu:%s sent to disk",
 626                                  bh->b_blocknr, bdevname(bh->b_bdev, b));
 627         }
 628         if (uptodate)
 629                 set_buffer_uptodate(bh);
 630         else
 631                 clear_buffer_uptodate(bh);
 632
 633         unlock_buffer(bh);
 634         release_buffer_page(bh);
 635 }
 636
 637 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 638 {
 639         if (uptodate)
 640                 set_buffer_uptodate(bh);
 641         else
 642                 clear_buffer_uptodate(bh);
 643         unlock_buffer(bh);
 644         put_bh(bh);
 645 }
 646
 647 static void submit_logged_buffer(struct buffer_head *bh)
 648 {
 649         get_bh(bh);
 650         bh->b_end_io = reiserfs_end_buffer_io_sync;
 651         clear_buffer_journal_new(bh);
 652         clear_buffer_dirty(bh);
 653         if (!test_clear_buffer_journal_test(bh))
 654                 BUG();
 655         if (!buffer_uptodate(bh))
 656                 BUG();
 657         submit_bh(WRITE, bh);
 658 }
 659
 660 static void submit_ordered_buffer(struct buffer_head *bh)
 661 {
 662         get_bh(bh);
 663         bh->b_end_io = reiserfs_end_ordered_io;
 664         clear_buffer_dirty(bh);
 665         if (!buffer_uptodate(bh))
 666                 BUG();
 667         submit_bh(WRITE, bh);
 668 }
 669
 670 #define CHUNK_SIZE 32
 671 struct buffer_chunk {
 672         struct buffer_head *bh[CHUNK_SIZE];
 673         int nr;
 674 };
 675
 676 static void write_chunk(struct buffer_chunk *chunk)
 677 {
 678         int i;
 679         for (i = 0; i < chunk->nr; i++) {
 680                 submit_logged_buffer(chunk->bh[i]);
 681         }
 682         chunk->nr = 0;
 683 }
 684
 685 static void write_ordered_chunk(struct buffer_chunk *chunk)
 686 {
 687         int i;
 688         for (i = 0; i < chunk->nr; i++) {
 689                 submit_ordered_buffer(chunk->bh[i]);
 690         }
 691         chunk->nr = 0;
 692 }
 693
 694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 695                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 696 {
 697         int ret = 0;
 698         BUG_ON(chunk->nr >= CHUNK_SIZE);
 699         chunk->bh[chunk->nr++] = bh;
 700         if (chunk->nr >= CHUNK_SIZE) {
 701                 ret = 1;
 702                 if (lock)
 703                         spin_unlock(lock);
 704                 fn(chunk);
 705                 if (lock)
 706                         spin_lock(lock);
 707         }
 708         return ret;
 709 }
 710
 711 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 712 static struct reiserfs_jh *alloc_jh(void)
 713 {
 714         struct reiserfs_jh *jh;
 715         while (1) {
 716                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 717                 if (jh) {
 718                         atomic_inc(&nr_reiserfs_jh);
 719                         return jh;
 720                 }
 721                 yield();
 722         }
 723 }
 724
 725 /*
 726  * we want to free the jh when the buffer has been written
 727  * and waited on
 728  */
 729 void reiserfs_free_jh(struct buffer_head *bh)
 730 {
 731         struct reiserfs_jh *jh;
 732
 733         jh = bh->b_private;
 734         if (jh) {
 735                 bh->b_private = NULL;
 736                 jh->bh = NULL;
 737                 list_del_init(&jh->list);
 738                 kfree(jh);
 739                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 740                         BUG();
 741                 atomic_dec(&nr_reiserfs_jh);
 742                 put_bh(bh);
 743         }
 744 }
 745
 746 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 747                            int tail)
 748 {
 749         struct reiserfs_jh *jh;
 750
 751         if (bh->b_private) {
 752                 spin_lock(&j->j_dirty_buffers_lock);
 753                 if (!bh->b_private) {
 754                         spin_unlock(&j->j_dirty_buffers_lock);
 755                         goto no_jh;
 756                 }
 757                 jh = bh->b_private;
 758                 list_del_init(&jh->list);
 759         } else {
 760 no_jh:
 761                 get_bh(bh);
 762                 jh = alloc_jh();
 763                 spin_lock(&j->j_dirty_buffers_lock);
 764                 /*
 765                  * buffer must be locked for __add_jh, should be able to have
 766                  * two adds at the same time
 767                  */
 768                 BUG_ON(bh->b_private);
 769                 jh->bh = bh;
 770                 bh->b_private = jh;
 771         }
 772         jh->jl = j->j_current_jl;
 773         if (tail)
 774                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 775         else {
 776                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 777         }
 778         spin_unlock(&j->j_dirty_buffers_lock);
 779         return 0;
 780 }
 781
 782 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 783 {
 784         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 785 }
 786 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 787 {
 788         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 789 }
 790
 791 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 792 static int write_ordered_buffers(spinlock_t * lock,
 793                                  struct reiserfs_journal *j,
 794                                  struct reiserfs_journal_list *jl,
 795                                  struct list_head *list)
 796 {
 797         struct buffer_head *bh;
 798         struct reiserfs_jh *jh;
 799         int ret = j->j_errno;
 800         struct buffer_chunk chunk;
 801         struct list_head tmp;
 802         INIT_LIST_HEAD(&tmp);
 803
 804         chunk.nr = 0;
 805         spin_lock(lock);
 806         while (!list_empty(list)) {
 807                 jh = JH_ENTRY(list->next);
 808                 bh = jh->bh;
 809                 get_bh(bh);
 810                 if (!trylock_buffer(bh)) {
 811                         if (!buffer_dirty(bh)) {
 812                                 list_move(&jh->list, &tmp);
 813                                 goto loop_next;
 814                         }
 815                         spin_unlock(lock);
 816                         if (chunk.nr)
 817                                 write_ordered_chunk(&chunk);
 818                         wait_on_buffer(bh);
 819                         cond_resched();
 820                         spin_lock(lock);
 821                         goto loop_next;
 822                 }
 823                 /*
 824                  * in theory, dirty non-uptodate buffers should never get here,
 825                  * but the upper layer io error paths still have a few quirks.
 826                  * Handle them here as gracefully as we can
 827                  */
 828                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 829                         clear_buffer_dirty(bh);
 830                         ret = -EIO;
 831                 }
 832                 if (buffer_dirty(bh)) {
 833                         list_move(&jh->list, &tmp);
 834                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 835                 } else {
 836                         reiserfs_free_jh(bh);
 837                         unlock_buffer(bh);
 838                 }
 839 loop_next:
 840                 put_bh(bh);
 841                 cond_resched_lock(lock);
 842         }
 843         if (chunk.nr) {
 844                 spin_unlock(lock);
 845                 write_ordered_chunk(&chunk);
 846                 spin_lock(lock);
 847         }
 848         while (!list_empty(&tmp)) {
 849                 jh = JH_ENTRY(tmp.prev);
 850                 bh = jh->bh;
 851                 get_bh(bh);
 852                 reiserfs_free_jh(bh);
 853
 854                 if (buffer_locked(bh)) {
 855                         spin_unlock(lock);
 856                         wait_on_buffer(bh);
 857                         spin_lock(lock);
 858                 }
 859                 if (!buffer_uptodate(bh)) {
 860                         ret = -EIO;
 861                 }
 862                 /*
 863                  * ugly interaction with invalidatepage here.
 864                  * reiserfs_invalidate_page will pin any buffer that has a
 865                  * valid journal head from an older transaction.  If someone
 866                  * else sets our buffer dirty after we write it in the first
 867                  * loop, and then someone truncates the page away, nobody
 868                  * will ever write the buffer. We're safe if we write the
 869                  * page one last time after freeing the journal header.
 870                  */
 871                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 872                         spin_unlock(lock);
 873                         ll_rw_block(WRITE, 1, &bh);
 874                         spin_lock(lock);
 875                 }
 876                 put_bh(bh);
 877                 cond_resched_lock(lock);
 878         }
 879         spin_unlock(lock);
 880         return ret;
 881 }
 882
 883 static int flush_older_commits(struct super_block *s,
 884                                struct reiserfs_journal_list *jl)
 885 {
 886         struct reiserfs_journal *journal = SB_JOURNAL(s);
 887         struct reiserfs_journal_list *other_jl;
 888         struct reiserfs_journal_list *first_jl;
 889         struct list_head *entry;
 890         unsigned int trans_id = jl->j_trans_id;
 891         unsigned int other_trans_id;
 892         unsigned int first_trans_id;
 893
 894 find_first:
 895         /*
 896          * first we walk backwards to find the oldest uncommitted transation
 897          */
 898         first_jl = jl;
 899         entry = jl->j_list.prev;
 900         while (1) {
 901                 other_jl = JOURNAL_LIST_ENTRY(entry);
 902                 if (entry == &journal->j_journal_list ||
 903                     atomic_read(&other_jl->j_older_commits_done))
 904                         break;
 905
 906                 first_jl = other_jl;
 907                 entry = other_jl->j_list.prev;
 908         }
 909
 910         /* if we didn't find any older uncommitted transactions, return now */
 911         if (first_jl == jl) {
 912                 return 0;
 913         }
 914
 915         first_trans_id = first_jl->j_trans_id;
 916
 917         entry = &first_jl->j_list;
 918         while (1) {
 919                 other_jl = JOURNAL_LIST_ENTRY(entry);
 920                 other_trans_id = other_jl->j_trans_id;
 921
 922                 if (other_trans_id < trans_id) {
 923                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 924                                 flush_commit_list(s, other_jl, 0);
 925
 926                                 /* list we were called with is gone, return */
 927                                 if (!journal_list_still_alive(s, trans_id))
 928                                         return 1;
 929
 930                                 /*
 931                                  * the one we just flushed is gone, this means
 932                                  * all older lists are also gone, so first_jl
 933                                  * is no longer valid either.  Go back to the
 934                                  * beginning.
 935                                  */
 936                                 if (!journal_list_still_alive
 937                                     (s, other_trans_id)) {
 938                                         goto find_first;
 939                                 }
 940                         }
 941                         entry = entry->next;
 942                         if (entry == &journal->j_journal_list)
 943                                 return 0;
 944                 } else {
 945                         return 0;
 946                 }
 947         }
 948         return 0;
 949 }
 950
 951 static int reiserfs_async_progress_wait(struct super_block *s)
 952 {
 953         struct reiserfs_journal *j = SB_JOURNAL(s);
 954
 955         if (atomic_read(&j->j_async_throttle)) {
 956                 int depth;
 957
 958                 depth = reiserfs_write_unlock_nested(s);
 959                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 960                 reiserfs_write_lock_nested(s, depth);
 961         }
 962
 963         return 0;
 964 }
 965
 966 /*
 967  * if this journal list still has commit blocks unflushed, send them to disk.
 968  *
 969  * log areas must be flushed in order (transaction 2 can't commit before
 970  * transaction 1) Before the commit block can by written, every other log
 971  * block must be safely on disk
 972  */
 973 static int flush_commit_list(struct super_block *s,
 974                              struct reiserfs_journal_list *jl, int flushall)
 975 {
 976         int i;
 977         b_blocknr_t bn;
 978         struct buffer_head *tbh = NULL;
 979         unsigned int trans_id = jl->j_trans_id;
 980         struct reiserfs_journal *journal = SB_JOURNAL(s);
 981         int retval = 0;
 982         int write_len;
 983         int depth;
 984
 985         reiserfs_check_lock_depth(s, "flush_commit_list");
 986
 987         if (atomic_read(&jl->j_older_commits_done)) {
 988                 return 0;
 989         }
 990
 991         /*
 992          * before we can put our commit blocks on disk, we have to make
 993          * sure everyone older than us is on disk too
 994          */
 995         BUG_ON(jl->j_len <= 0);
 996         BUG_ON(trans_id == journal->j_trans_id);
 997
 998         get_journal_list(jl);
 999         if (flushall) {
1000                 if (flush_older_commits(s, jl) == 1) {
1001                         /*
1002                          * list disappeared during flush_older_commits.
1003                          * return
1004                          */
1005                         goto put_jl;
1006                 }
1007         }
1008
1009         /* make sure nobody is trying to flush this one at the same time */
1010         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1011
1012         if (!journal_list_still_alive(s, trans_id)) {
1013                 mutex_unlock(&jl->j_commit_mutex);
1014                 goto put_jl;
1015         }
1016         BUG_ON(jl->j_trans_id == 0);
1017
1018         /* this commit is done, exit */
1019         if (atomic_read(&jl->j_commit_left) <= 0) {
1020                 if (flushall) {
1021                         atomic_set(&jl->j_older_commits_done, 1);
1022                 }
1023                 mutex_unlock(&jl->j_commit_mutex);
1024                 goto put_jl;
1025         }
1026
1027         if (!list_empty(&jl->j_bh_list)) {
1028                 int ret;
1029
1030                 /*
1031                  * We might sleep in numerous places inside
1032                  * write_ordered_buffers. Relax the write lock.
1033                  */
1034                 depth = reiserfs_write_unlock_nested(s);
1035                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1036                                             journal, jl, &jl->j_bh_list);
1037                 if (ret < 0 && retval == 0)
1038                         retval = ret;
1039                 reiserfs_write_lock_nested(s, depth);
1040         }
1041         BUG_ON(!list_empty(&jl->j_bh_list));
1042         /*
1043          * for the description block and all the log blocks, submit any buffers
1044          * that haven't already reached the disk.  Try to write at least 256
1045          * log blocks. later on, we will only wait on blocks that correspond
1046          * to this transaction, but while we're unplugging we might as well
1047          * get a chunk of data on there.
1048          */
1049         atomic_inc(&journal->j_async_throttle);
1050         write_len = jl->j_len + 1;
1051         if (write_len < 256)
1052                 write_len = 256;
1053         for (i = 0 ; i < write_len ; i++) {
1054                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1055                     SB_ONDISK_JOURNAL_SIZE(s);
1056                 tbh = journal_find_get_block(s, bn);
1057                 if (tbh) {
1058                         if (buffer_dirty(tbh)) {
1059                             depth = reiserfs_write_unlock_nested(s);
1060                             ll_rw_block(WRITE, 1, &tbh);
1061                             reiserfs_write_lock_nested(s, depth);
1062                         }
1063                         put_bh(tbh) ;
1064                 }
1065         }
1066         atomic_dec(&journal->j_async_throttle);
1067
1068         for (i = 0; i < (jl->j_len + 1); i++) {
1069                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1070                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1071                 tbh = journal_find_get_block(s, bn);
1072
1073                 depth = reiserfs_write_unlock_nested(s);
1074                 __wait_on_buffer(tbh);
1075                 reiserfs_write_lock_nested(s, depth);
1076                 /*
1077                  * since we're using ll_rw_blk above, it might have skipped
1078                  * over a locked buffer.  Double check here
1079                  */
1080                 /* redundant, sync_dirty_buffer() checks */
1081                 if (buffer_dirty(tbh)) {
1082                         depth = reiserfs_write_unlock_nested(s);
1083                         sync_dirty_buffer(tbh);
1084                         reiserfs_write_lock_nested(s, depth);
1085                 }
1086                 if (unlikely(!buffer_uptodate(tbh))) {
1087 #ifdef CONFIG_REISERFS_CHECK
1088                         reiserfs_warning(s, "journal-601",
1089                                          "buffer write failed");
1090 #endif
1091                         retval = -EIO;
1092                 }
1093                 /* once for journal_find_get_block */
1094                 put_bh(tbh);
1095                 /* once due to original getblk in do_journal_end */
1096                 put_bh(tbh);
1097                 atomic_dec(&jl->j_commit_left);
1098         }
1099
1100         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1101
1102         /*
1103          * If there was a write error in the journal - we can't commit
1104          * this transaction - it will be invalid and, if successful,
1105          * will just end up propagating the write error out to
1106          * the file system.
1107          */
1108         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1109                 if (buffer_dirty(jl->j_commit_bh))
1110                         BUG();
1111                 mark_buffer_dirty(jl->j_commit_bh) ;
1112                 depth = reiserfs_write_unlock_nested(s);
1113                 if (reiserfs_barrier_flush(s))
1114                         __sync_dirty_buffer(jl->j_commit_bh, WRITE_FLUSH_FUA);
1115                 else
1116                         sync_dirty_buffer(jl->j_commit_bh);
1117                 reiserfs_write_lock_nested(s, depth);
1118         }
1119
1120         /*
1121          * If there was a write error in the journal - we can't commit this
1122          * transaction - it will be invalid and, if successful, will just end
1123          * up propagating the write error out to the filesystem.
1124          */
1125         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1126 #ifdef CONFIG_REISERFS_CHECK
1127                 reiserfs_warning(s, "journal-615", "buffer write failed");
1128 #endif
1129                 retval = -EIO;
1130         }
1131         bforget(jl->j_commit_bh);
1132         if (journal->j_last_commit_id != 0 &&
1133             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1134                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1135                                  journal->j_last_commit_id, jl->j_trans_id);
1136         }
1137         journal->j_last_commit_id = jl->j_trans_id;
1138
1139         /*
1140          * now, every commit block is on the disk.  It is safe to allow
1141          * blocks freed during this transaction to be reallocated
1142          */
1143         cleanup_freed_for_journal_list(s, jl);
1144
1145         retval = retval ? retval : journal->j_errno;
1146
1147         /* mark the metadata dirty */
1148         if (!retval)
1149                 dirty_one_transaction(s, jl);
1150         atomic_dec(&jl->j_commit_left);
1151
1152         if (flushall) {
1153                 atomic_set(&jl->j_older_commits_done, 1);
1154         }
1155         mutex_unlock(&jl->j_commit_mutex);
1156 put_jl:
1157         put_journal_list(s, jl);
1158
1159         if (retval)
1160                 reiserfs_abort(s, retval, "Journal write error in %s",
1161                                __func__);
1162         return retval;
1163 }
1164
1165 /*
1166  * flush_journal_list frequently needs to find a newer transaction for a
1167  * given block.  This does that, or returns NULL if it can't find anything
1168  */
1169 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1170                                                           reiserfs_journal_cnode
1171                                                           *cn)
1172 {
1173         struct super_block *sb = cn->sb;
1174         b_blocknr_t blocknr = cn->blocknr;
1175
1176         cn = cn->hprev;
1177         while (cn) {
1178                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1179                         return cn->jlist;
1180                 }
1181                 cn = cn->hprev;
1182         }
1183         return NULL;
1184 }
1185
1186 static void remove_journal_hash(struct super_block *,
1187                                 struct reiserfs_journal_cnode **,
1188                                 struct reiserfs_journal_list *, unsigned long,
1189                                 int);
1190
1191 /*
1192  * once all the real blocks have been flushed, it is safe to remove them
1193  * from the journal list for this transaction.  Aside from freeing the
1194  * cnode, this also allows the block to be reallocated for data blocks
1195  * if it had been deleted.
1196  */
1197 static void remove_all_from_journal_list(struct super_block *sb,
1198                                          struct reiserfs_journal_list *jl,
1199                                          int debug)
1200 {
1201         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1202         struct reiserfs_journal_cnode *cn, *last;
1203         cn = jl->j_realblock;
1204
1205         /*
1206          * which is better, to lock once around the whole loop, or
1207          * to lock for each call to remove_journal_hash?
1208          */
1209         while (cn) {
1210                 if (cn->blocknr != 0) {
1211                         if (debug) {
1212                                 reiserfs_warning(sb, "reiserfs-2201",
1213                                                  "block %u, bh is %d, state %ld",
1214                                                  cn->blocknr, cn->bh ? 1 : 0,
1215                                                  cn->state);
1216                         }
1217                         cn->state = 0;
1218                         remove_journal_hash(sb, journal->j_list_hash_table,
1219                                             jl, cn->blocknr, 1);
1220                 }
1221                 last = cn;
1222                 cn = cn->next;
1223                 free_cnode(sb, last);
1224         }
1225         jl->j_realblock = NULL;
1226 }
1227
1228 /*
1229  * if this timestamp is greater than the timestamp we wrote last to the
1230  * header block, write it to the header block.  once this is done, I can
1231  * safely say the log area for this transaction won't ever be replayed,
1232  * and I can start releasing blocks in this transaction for reuse as data
1233  * blocks.  called by flush_journal_list, before it calls
1234  * remove_all_from_journal_list
1235  */
1236 static int _update_journal_header_block(struct super_block *sb,
1237                                         unsigned long offset,
1238                                         unsigned int trans_id)
1239 {
1240         struct reiserfs_journal_header *jh;
1241         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1242         int depth;
1243
1244         if (reiserfs_is_journal_aborted(journal))
1245                 return -EIO;
1246
1247         if (trans_id >= journal->j_last_flush_trans_id) {
1248                 if (buffer_locked((journal->j_header_bh))) {
1249                         depth = reiserfs_write_unlock_nested(sb);
1250                         __wait_on_buffer(journal->j_header_bh);
1251                         reiserfs_write_lock_nested(sb, depth);
1252                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1253 #ifdef CONFIG_REISERFS_CHECK
1254                                 reiserfs_warning(sb, "journal-699",
1255                                                  "buffer write failed");
1256 #endif
1257                                 return -EIO;
1258                         }
1259                 }
1260                 journal->j_last_flush_trans_id = trans_id;
1261                 journal->j_first_unflushed_offset = offset;
1262                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1263                                                         b_data);
1264                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1265                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1266                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1267
1268                 set_buffer_dirty(journal->j_header_bh);
1269                 depth = reiserfs_write_unlock_nested(sb);
1270
1271                 if (reiserfs_barrier_flush(sb))
1272                         __sync_dirty_buffer(journal->j_header_bh, WRITE_FLUSH_FUA);
1273                 else
1274                         sync_dirty_buffer(journal->j_header_bh);
1275
1276                 reiserfs_write_lock_nested(sb, depth);
1277                 if (!buffer_uptodate(journal->j_header_bh)) {
1278                         reiserfs_warning(sb, "journal-837",
1279                                          "IO error during journal replay");
1280                         return -EIO;
1281                 }
1282         }
1283         return 0;
1284 }
1285
1286 static int update_journal_header_block(struct super_block *sb,
1287                                        unsigned long offset,
1288                                        unsigned int trans_id)
1289 {
1290         return _update_journal_header_block(sb, offset, trans_id);
1291 }
1292
1293 /*
1294 ** flush any and all journal lists older than you are
1295 ** can only be called from flush_journal_list
1296 */
1297 static int flush_older_journal_lists(struct super_block *sb,
1298                                      struct reiserfs_journal_list *jl)
1299 {
1300         struct list_head *entry;
1301         struct reiserfs_journal_list *other_jl;
1302         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1303         unsigned int trans_id = jl->j_trans_id;
1304
1305         /*
1306          * we know we are the only ones flushing things, no extra race
1307          * protection is required.
1308          */
1309 restart:
1310         entry = journal->j_journal_list.next;
1311         /* Did we wrap? */
1312         if (entry == &journal->j_journal_list)
1313                 return 0;
1314         other_jl = JOURNAL_LIST_ENTRY(entry);
1315         if (other_jl->j_trans_id < trans_id) {
1316                 BUG_ON(other_jl->j_refcount <= 0);
1317                 /* do not flush all */
1318                 flush_journal_list(sb, other_jl, 0);
1319
1320                 /* other_jl is now deleted from the list */
1321                 goto restart;
1322         }
1323         return 0;
1324 }
1325
1326 static void del_from_work_list(struct super_block *s,
1327                                struct reiserfs_journal_list *jl)
1328 {
1329         struct reiserfs_journal *journal = SB_JOURNAL(s);
1330         if (!list_empty(&jl->j_working_list)) {
1331                 list_del_init(&jl->j_working_list);
1332                 journal->j_num_work_lists--;
1333         }
1334 }
1335
1336 /*
1337  * flush a journal list, both commit and real blocks
1338  *
1339  * always set flushall to 1, unless you are calling from inside
1340  * flush_journal_list
1341  *
1342  * IMPORTANT.  This can only be called while there are no journal writers,
1343  * and the journal is locked.  That means it can only be called from
1344  * do_journal_end, or by journal_release
1345  */
1346 static int flush_journal_list(struct super_block *s,
1347                               struct reiserfs_journal_list *jl, int flushall)
1348 {
1349         struct reiserfs_journal_list *pjl;
1350         struct reiserfs_journal_cnode *cn, *last;
1351         int count;
1352         int was_jwait = 0;
1353         int was_dirty = 0;
1354         struct buffer_head *saved_bh;
1355         unsigned long j_len_saved = jl->j_len;
1356         struct reiserfs_journal *journal = SB_JOURNAL(s);
1357         int err = 0;
1358         int depth;
1359
1360         BUG_ON(j_len_saved <= 0);
1361
1362         if (atomic_read(&journal->j_wcount) != 0) {
1363                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1364                                  atomic_read(&journal->j_wcount));
1365         }
1366
1367         /* if flushall == 0, the lock is already held */
1368         if (flushall) {
1369                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1370         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1371                 BUG();
1372         }
1373
1374         count = 0;
1375         if (j_len_saved > journal->j_trans_max) {
1376                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1377                                j_len_saved, jl->j_trans_id);
1378                 return 0;
1379         }
1380
1381         /* if all the work is already done, get out of here */
1382         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1383             atomic_read(&jl->j_commit_left) <= 0) {
1384                 goto flush_older_and_return;
1385         }
1386
1387         /*
1388          * start by putting the commit list on disk.  This will also flush
1389          * the commit lists of any olders transactions
1390          */
1391         flush_commit_list(s, jl, 1);
1392
1393         if (!(jl->j_state & LIST_DIRTY)
1394             && !reiserfs_is_journal_aborted(journal))
1395                 BUG();
1396
1397         /* are we done now? */
1398         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1399             atomic_read(&jl->j_commit_left) <= 0) {
1400                 goto flush_older_and_return;
1401         }
1402
1403         /*
1404          * loop through each cnode, see if we need to write it,
1405          * or wait on a more recent transaction, or just ignore it
1406          */
1407         if (atomic_read(&journal->j_wcount) != 0) {
1408                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1409                                "wcount is not 0");
1410         }
1411         cn = jl->j_realblock;
1412         while (cn) {
1413                 was_jwait = 0;
1414                 was_dirty = 0;
1415                 saved_bh = NULL;
1416                 /* blocknr of 0 is no longer in the hash, ignore it */
1417                 if (cn->blocknr == 0) {
1418                         goto free_cnode;
1419                 }
1420
1421                 /*
1422                  * This transaction failed commit.
1423                  * Don't write out to the disk
1424                  */
1425                 if (!(jl->j_state & LIST_DIRTY))
1426                         goto free_cnode;
1427
1428                 pjl = find_newer_jl_for_cn(cn);
1429                 /*
1430                  * the order is important here.  We check pjl to make sure we
1431                  * don't clear BH_JDirty_wait if we aren't the one writing this
1432                  * block to disk
1433                  */
1434                 if (!pjl && cn->bh) {
1435                         saved_bh = cn->bh;
1436
1437                         /*
1438                          * we do this to make sure nobody releases the
1439                          * buffer while we are working with it
1440                          */
1441                         get_bh(saved_bh);
1442
1443                         if (buffer_journal_dirty(saved_bh)) {
1444                                 BUG_ON(!can_dirty(cn));
1445                                 was_jwait = 1;
1446                                 was_dirty = 1;
1447                         } else if (can_dirty(cn)) {
1448                                 /*
1449                                  * everything with !pjl && jwait
1450                                  * should be writable
1451                                  */
1452                                 BUG();
1453                         }
1454                 }
1455
1456                 /*
1457                  * if someone has this block in a newer transaction, just make
1458                  * sure they are committed, and don't try writing it to disk
1459                  */
1460                 if (pjl) {
1461                         if (atomic_read(&pjl->j_commit_left))
1462                                 flush_commit_list(s, pjl, 1);
1463                         goto free_cnode;
1464                 }
1465
1466                 /*
1467                  * bh == NULL when the block got to disk on its own, OR,
1468                  * the block got freed in a future transaction
1469                  */
1470                 if (saved_bh == NULL) {
1471                         goto free_cnode;
1472                 }
1473
1474                 /*
1475                  * this should never happen.  kupdate_one_transaction has
1476                  * this list locked while it works, so we should never see a
1477                  * buffer here that is not marked JDirty_wait
1478                  */
1479                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1480                         reiserfs_warning(s, "journal-813",
1481                                          "BAD! buffer %llu %cdirty %cjwait, "
1482                                          "not in a newer tranasction",
1483                                          (unsigned long long)saved_bh->
1484                                          b_blocknr, was_dirty ? ' ' : '!',
1485                                          was_jwait ? ' ' : '!');
1486                 }
1487                 if (was_dirty) {
1488                         /*
1489                          * we inc again because saved_bh gets decremented
1490                          * at free_cnode
1491                          */
1492                         get_bh(saved_bh);
1493                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1494                         lock_buffer(saved_bh);
1495                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1496                         if (buffer_dirty(saved_bh))
1497                                 submit_logged_buffer(saved_bh);
1498                         else
1499                                 unlock_buffer(saved_bh);
1500                         count++;
1501                 } else {
1502                         reiserfs_warning(s, "clm-2082",
1503                                          "Unable to flush buffer %llu in %s",
1504                                          (unsigned long long)saved_bh->
1505                                          b_blocknr, __func__);
1506                 }
1507 free_cnode:
1508                 last = cn;
1509                 cn = cn->next;
1510                 if (saved_bh) {
1511                         /*
1512                          * we incremented this to keep others from
1513                          * taking the buffer head away
1514                          */
1515                         put_bh(saved_bh);
1516                         if (atomic_read(&saved_bh->b_count) < 0) {
1517                                 reiserfs_warning(s, "journal-945",
1518                                                  "saved_bh->b_count < 0");
1519                         }
1520                 }
1521         }
1522         if (count > 0) {
1523                 cn = jl->j_realblock;
1524                 while (cn) {
1525                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1526                                 if (!cn->bh) {
1527                                         reiserfs_panic(s, "journal-1011",
1528                                                        "cn->bh is NULL");
1529                                 }
1530
1531                                 depth = reiserfs_write_unlock_nested(s);
1532                                 __wait_on_buffer(cn->bh);
1533                                 reiserfs_write_lock_nested(s, depth);
1534
1535                                 if (!cn->bh) {
1536                                         reiserfs_panic(s, "journal-1012",
1537                                                        "cn->bh is NULL");
1538                                 }
1539                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1540 #ifdef CONFIG_REISERFS_CHECK
1541                                         reiserfs_warning(s, "journal-949",
1542                                                          "buffer write failed");
1543 #endif
1544                                         err = -EIO;
1545                                 }
1546                                 /*
1547                                  * note, we must clear the JDirty_wait bit
1548                                  * after the up to date check, otherwise we
1549                                  * race against our flushpage routine
1550                                  */
1551                                 BUG_ON(!test_clear_buffer_journal_dirty
1552                                        (cn->bh));
1553
1554                                 /* drop one ref for us */
1555                                 put_bh(cn->bh);
1556                                 /* drop one ref for journal_mark_dirty */
1557                                 release_buffer_page(cn->bh);
1558                         }
1559                         cn = cn->next;
1560                 }
1561         }
1562
1563         if (err)
1564                 reiserfs_abort(s, -EIO,
1565                                "Write error while pushing transaction to disk in %s",
1566                                __func__);
1567 flush_older_and_return:
1568
1569         /*
1570          * before we can update the journal header block, we _must_ flush all
1571          * real blocks from all older transactions to disk.  This is because
1572          * once the header block is updated, this transaction will not be
1573          * replayed after a crash
1574          */
1575         if (flushall) {
1576                 flush_older_journal_lists(s, jl);
1577         }
1578
1579         err = journal->j_errno;
1580         /*
1581          * before we can remove everything from the hash tables for this
1582          * transaction, we must make sure it can never be replayed
1583          *
1584          * since we are only called from do_journal_end, we know for sure there
1585          * are no allocations going on while we are flushing journal lists.  So,
1586          * we only need to update the journal header block for the last list
1587          * being flushed
1588          */
1589         if (!err && flushall) {
1590                 err =
1591                     update_journal_header_block(s,
1592                                                 (jl->j_start + jl->j_len +
1593                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1594                                                 jl->j_trans_id);
1595                 if (err)
1596                         reiserfs_abort(s, -EIO,
1597                                        "Write error while updating journal header in %s",
1598                                        __func__);
1599         }
1600         remove_all_from_journal_list(s, jl, 0);
1601         list_del_init(&jl->j_list);
1602         journal->j_num_lists--;
1603         del_from_work_list(s, jl);
1604
1605         if (journal->j_last_flush_id != 0 &&
1606             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1607                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1608                                  journal->j_last_flush_id, jl->j_trans_id);
1609         }
1610         journal->j_last_flush_id = jl->j_trans_id;
1611
1612         /*
1613          * not strictly required since we are freeing the list, but it should
1614          * help find code using dead lists later on
1615          */
1616         jl->j_len = 0;
1617         atomic_set(&jl->j_nonzerolen, 0);
1618         jl->j_start = 0;
1619         jl->j_realblock = NULL;
1620         jl->j_commit_bh = NULL;
1621         jl->j_trans_id = 0;
1622         jl->j_state = 0;
1623         put_journal_list(s, jl);
1624         if (flushall)
1625                 mutex_unlock(&journal->j_flush_mutex);
1626         return err;
1627 }
1628
1629 static int write_one_transaction(struct super_block *s,
1630                                  struct reiserfs_journal_list *jl,
1631                                  struct buffer_chunk *chunk)
1632 {
1633         struct reiserfs_journal_cnode *cn;
1634         int ret = 0;
1635
1636         jl->j_state |= LIST_TOUCHED;
1637         del_from_work_list(s, jl);
1638         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1639                 return 0;
1640         }
1641
1642         cn = jl->j_realblock;
1643         while (cn) {
1644                 /*
1645                  * if the blocknr == 0, this has been cleared from the hash,
1646                  * skip it
1647                  */
1648                 if (cn->blocknr == 0) {
1649                         goto next;
1650                 }
1651                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1652                         struct buffer_head *tmp_bh;
1653                         /*
1654                          * we can race against journal_mark_freed when we try
1655                          * to lock_buffer(cn->bh), so we have to inc the buffer
1656                          * count, and recheck things after locking
1657                          */
1658                         tmp_bh = cn->bh;
1659                         get_bh(tmp_bh);
1660                         lock_buffer(tmp_bh);
1661                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1662                                 if (!buffer_journal_dirty(tmp_bh) ||
1663                                     buffer_journal_prepared(tmp_bh))
1664                                         BUG();
1665                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1666                                 ret++;
1667                         } else {
1668                                 /* note, cn->bh might be null now */
1669                                 unlock_buffer(tmp_bh);
1670                         }
1671                         put_bh(tmp_bh);
1672                 }
1673 next:
1674                 cn = cn->next;
1675                 cond_resched();
1676         }
1677         return ret;
1678 }
1679
1680 /* used by flush_commit_list */
1681 static int dirty_one_transaction(struct super_block *s,
1682                                  struct reiserfs_journal_list *jl)
1683 {
1684         struct reiserfs_journal_cnode *cn;
1685         struct reiserfs_journal_list *pjl;
1686         int ret = 0;
1687
1688         jl->j_state |= LIST_DIRTY;
1689         cn = jl->j_realblock;
1690         while (cn) {
1691                 /*
1692                  * look for a more recent transaction that logged this
1693                  * buffer.  Only the most recent transaction with a buffer in
1694                  * it is allowed to send that buffer to disk
1695                  */
1696                 pjl = find_newer_jl_for_cn(cn);
1697                 if (!pjl && cn->blocknr && cn->bh
1698                     && buffer_journal_dirty(cn->bh)) {
1699                         BUG_ON(!can_dirty(cn));
1700                         /*
1701                          * if the buffer is prepared, it will either be logged
1702                          * or restored.  If restored, we need to make sure
1703                          * it actually gets marked dirty
1704                          */
1705                         clear_buffer_journal_new(cn->bh);
1706                         if (buffer_journal_prepared(cn->bh)) {
1707                                 set_buffer_journal_restore_dirty(cn->bh);
1708                         } else {
1709                                 set_buffer_journal_test(cn->bh);
1710                                 mark_buffer_dirty(cn->bh);
1711                         }
1712                 }
1713                 cn = cn->next;
1714         }
1715         return ret;
1716 }
1717
1718 static int kupdate_transactions(struct super_block *s,
1719                                 struct reiserfs_journal_list *jl,
1720                                 struct reiserfs_journal_list **next_jl,
1721                                 unsigned int *next_trans_id,
1722                                 int num_blocks, int num_trans)
1723 {
1724         int ret = 0;
1725         int written = 0;
1726         int transactions_flushed = 0;
1727         unsigned int orig_trans_id = jl->j_trans_id;
1728         struct buffer_chunk chunk;
1729         struct list_head *entry;
1730         struct reiserfs_journal *journal = SB_JOURNAL(s);
1731         chunk.nr = 0;
1732
1733         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1734         if (!journal_list_still_alive(s, orig_trans_id)) {
1735                 goto done;
1736         }
1737
1738         /*
1739          * we've got j_flush_mutex held, nobody is going to delete any
1740          * of these lists out from underneath us
1741          */
1742         while ((num_trans && transactions_flushed < num_trans) ||
1743                (!num_trans && written < num_blocks)) {
1744
1745                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1746                     atomic_read(&jl->j_commit_left)
1747                     || !(jl->j_state & LIST_DIRTY)) {
1748                         del_from_work_list(s, jl);
1749                         break;
1750                 }
1751                 ret = write_one_transaction(s, jl, &chunk);
1752
1753                 if (ret < 0)
1754                         goto done;
1755                 transactions_flushed++;
1756                 written += ret;
1757                 entry = jl->j_list.next;
1758
1759                 /* did we wrap? */
1760                 if (entry == &journal->j_journal_list) {
1761                         break;
1762                 }
1763                 jl = JOURNAL_LIST_ENTRY(entry);
1764
1765                 /* don't bother with older transactions */
1766                 if (jl->j_trans_id <= orig_trans_id)
1767                         break;
1768         }
1769         if (chunk.nr) {
1770                 write_chunk(&chunk);
1771         }
1772
1773 done:
1774         mutex_unlock(&journal->j_flush_mutex);
1775         return ret;
1776 }
1777
1778 /*
1779  * for o_sync and fsync heavy applications, they tend to use
1780  * all the journa list slots with tiny transactions.  These
1781  * trigger lots and lots of calls to update the header block, which
1782  * adds seeks and slows things down.
1783  *
1784  * This function tries to clear out a large chunk of the journal lists
1785  * at once, which makes everything faster since only the newest journal
1786  * list updates the header block
1787  */
1788 static int flush_used_journal_lists(struct super_block *s,
1789                                     struct reiserfs_journal_list *jl)
1790 {
1791         unsigned long len = 0;
1792         unsigned long cur_len;
1793         int ret;
1794         int i;
1795         int limit = 256;
1796         struct reiserfs_journal_list *tjl;
1797         struct reiserfs_journal_list *flush_jl;
1798         unsigned int trans_id;
1799         struct reiserfs_journal *journal = SB_JOURNAL(s);
1800
1801         flush_jl = tjl = jl;
1802
1803         /* in data logging mode, try harder to flush a lot of blocks */
1804         if (reiserfs_data_log(s))
1805                 limit = 1024;
1806         /* flush for 256 transactions or limit blocks, whichever comes first */
1807         for (i = 0; i < 256 && len < limit; i++) {
1808                 if (atomic_read(&tjl->j_commit_left) ||
1809                     tjl->j_trans_id < jl->j_trans_id) {
1810                         break;
1811                 }
1812                 cur_len = atomic_read(&tjl->j_nonzerolen);
1813                 if (cur_len > 0) {
1814                         tjl->j_state &= ~LIST_TOUCHED;
1815                 }
1816                 len += cur_len;
1817                 flush_jl = tjl;
1818                 if (tjl->j_list.next == &journal->j_journal_list)
1819                         break;
1820                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1821         }
1822         get_journal_list(jl);
1823         get_journal_list(flush_jl);
1824
1825         /*
1826          * try to find a group of blocks we can flush across all the
1827          * transactions, but only bother if we've actually spanned
1828          * across multiple lists
1829          */
1830         if (flush_jl != jl) {
1831                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1832         }
1833         flush_journal_list(s, flush_jl, 1);
1834         put_journal_list(s, flush_jl);
1835         put_journal_list(s, jl);
1836         return 0;
1837 }
1838
1839 /*
1840  * removes any nodes in table with name block and dev as bh.
1841  * only touchs the hnext and hprev pointers.
1842  */
1843 void remove_journal_hash(struct super_block *sb,
1844                          struct reiserfs_journal_cnode **table,
1845                          struct reiserfs_journal_list *jl,
1846                          unsigned long block, int remove_freed)
1847 {
1848         struct reiserfs_journal_cnode *cur;
1849         struct reiserfs_journal_cnode **head;
1850
1851         head = &(journal_hash(table, sb, block));
1852         if (!head) {
1853                 return;
1854         }
1855         cur = *head;
1856         while (cur) {
1857                 if (cur->blocknr == block && cur->sb == sb
1858                     && (jl == NULL || jl == cur->jlist)
1859                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1860                         if (cur->hnext) {
1861                                 cur->hnext->hprev = cur->hprev;
1862                         }
1863                         if (cur->hprev) {
1864                                 cur->hprev->hnext = cur->hnext;
1865                         } else {
1866                                 *head = cur->hnext;
1867                         }
1868                         cur->blocknr = 0;
1869                         cur->sb = NULL;
1870                         cur->state = 0;
1871                         /*
1872                          * anybody who clears the cur->bh will also
1873                          * dec the nonzerolen
1874                          */
1875                         if (cur->bh && cur->jlist)
1876                                 atomic_dec(&cur->jlist->j_nonzerolen);
1877                         cur->bh = NULL;
1878                         cur->jlist = NULL;
1879                 }
1880                 cur = cur->hnext;
1881         }
1882 }
1883
1884 static void free_journal_ram(struct super_block *sb)
1885 {
1886         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1887         kfree(journal->j_current_jl);
1888         journal->j_num_lists--;
1889
1890         vfree(journal->j_cnode_free_orig);
1891         free_list_bitmaps(sb, journal->j_list_bitmap);
1892         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1893         if (journal->j_header_bh) {
1894                 brelse(journal->j_header_bh);
1895         }
1896         /*
1897          * j_header_bh is on the journal dev, make sure
1898          * not to release the journal dev until we brelse j_header_bh
1899          */
1900         release_journal_dev(sb, journal);
1901         vfree(journal);
1902 }
1903
1904 /*
1905  * call on unmount.  Only set error to 1 if you haven't made your way out
1906  * of read_super() yet.  Any other caller must keep error at 0.
1907  */
1908 static int do_journal_release(struct reiserfs_transaction_handle *th,
1909                               struct super_block *sb, int error)
1910 {
1911         struct reiserfs_transaction_handle myth;
1912         int flushed = 0;
1913         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1914
1915         /*
1916          * we only want to flush out transactions if we were
1917          * called with error == 0
1918          */
1919         if (!error && !(sb->s_flags & MS_RDONLY)) {
1920                 /* end the current trans */
1921                 BUG_ON(!th->t_trans_id);
1922                 do_journal_end(th, FLUSH_ALL);
1923
1924                 /*
1925                  * make sure something gets logged to force
1926                  * our way into the flush code
1927                  */
1928                 if (!journal_join(&myth, sb)) {
1929                         reiserfs_prepare_for_journal(sb,
1930                                                      SB_BUFFER_WITH_SB(sb),
1931                                                      1);
1932                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1933                         do_journal_end(&myth, FLUSH_ALL);
1934                         flushed = 1;
1935                 }
1936         }
1937
1938         /* this also catches errors during the do_journal_end above */
1939         if (!error && reiserfs_is_journal_aborted(journal)) {
1940                 memset(&myth, 0, sizeof(myth));
1941                 if (!journal_join_abort(&myth, sb)) {
1942                         reiserfs_prepare_for_journal(sb,
1943                                                      SB_BUFFER_WITH_SB(sb),
1944                                                      1);
1945                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1946                         do_journal_end(&myth, FLUSH_ALL);
1947                 }
1948         }
1949
1950         /* wait for all commits to finish */
1951         cancel_delayed_work(&SB_JOURNAL(sb)->j_work);
1952
1953         /*
1954          * We must release the write lock here because
1955          * the workqueue job (flush_async_commit) needs this lock
1956          */
1957         reiserfs_write_unlock(sb);
1958
1959         cancel_delayed_work_sync(&REISERFS_SB(sb)->old_work);
1960         flush_workqueue(REISERFS_SB(sb)->commit_wq);
1961
1962         free_journal_ram(sb);
1963
1964         reiserfs_write_lock(sb);
1965
1966         return 0;
1967 }
1968
1969 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1970 int journal_release(struct reiserfs_transaction_handle *th,
1971                     struct super_block *sb)
1972 {
1973         return do_journal_release(th, sb, 0);
1974 }
1975
1976 /* only call from an error condition inside reiserfs_read_super!  */
1977 int journal_release_error(struct reiserfs_transaction_handle *th,
1978                           struct super_block *sb)
1979 {
1980         return do_journal_release(th, sb, 1);
1981 }
1982
1983 /*
1984  * compares description block with commit block.
1985  * returns 1 if they differ, 0 if they are the same
1986  */
1987 static int journal_compare_desc_commit(struct super_block *sb,
1988                                        struct reiserfs_journal_desc *desc,
1989                                        struct reiserfs_journal_commit *commit)
1990 {
1991         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1992             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1993             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1994             get_commit_trans_len(commit) <= 0) {
1995                 return 1;
1996         }
1997         return 0;
1998 }
1999
2000 /*
2001  * returns 0 if it did not find a description block
2002  * returns -1 if it found a corrupt commit block
2003  * returns 1 if both desc and commit were valid
2004  * NOTE: only called during fs mount
2005  */
2006 static int journal_transaction_is_valid(struct super_block *sb,
2007                                         struct buffer_head *d_bh,
2008                                         unsigned int *oldest_invalid_trans_id,
2009                                         unsigned long *newest_mount_id)
2010 {
2011         struct reiserfs_journal_desc *desc;
2012         struct reiserfs_journal_commit *commit;
2013         struct buffer_head *c_bh;
2014         unsigned long offset;
2015
2016         if (!d_bh)
2017                 return 0;
2018
2019         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2020         if (get_desc_trans_len(desc) > 0
2021             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2022                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2023                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2024                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2025                                        "journal-986: transaction "
2026                                        "is valid returning because trans_id %d is greater than "
2027                                        "oldest_invalid %lu",
2028                                        get_desc_trans_id(desc),
2029                                        *oldest_invalid_trans_id);
2030                         return 0;
2031                 }
2032                 if (newest_mount_id
2033                     && *newest_mount_id > get_desc_mount_id(desc)) {
2034                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2035                                        "journal-1087: transaction "
2036                                        "is valid returning because mount_id %d is less than "
2037                                        "newest_mount_id %lu",
2038                                        get_desc_mount_id(desc),
2039                                        *newest_mount_id);
2040                         return -1;
2041                 }
2042                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2043                         reiserfs_warning(sb, "journal-2018",
2044                                          "Bad transaction length %d "
2045                                          "encountered, ignoring transaction",
2046                                          get_desc_trans_len(desc));
2047                         return -1;
2048                 }
2049                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2050
2051                 /*
2052                  * ok, we have a journal description block,
2053                  * let's see if the transaction was valid
2054                  */
2055                 c_bh =
2056                     journal_bread(sb,
2057                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2058                                   ((offset + get_desc_trans_len(desc) +
2059                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2060                 if (!c_bh)
2061                         return 0;
2062                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2063                 if (journal_compare_desc_commit(sb, desc, commit)) {
2064                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2065                                        "journal_transaction_is_valid, commit offset %ld had bad "
2066                                        "time %d or length %d",
2067                                        c_bh->b_blocknr -
2068                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2069                                        get_commit_trans_id(commit),
2070                                        get_commit_trans_len(commit));
2071                         brelse(c_bh);
2072                         if (oldest_invalid_trans_id) {
2073                                 *oldest_invalid_trans_id =
2074                                     get_desc_trans_id(desc);
2075                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2076                                                "journal-1004: "
2077                                                "transaction_is_valid setting oldest invalid trans_id "
2078                                                "to %d",
2079                                                get_desc_trans_id(desc));
2080                         }
2081                         return -1;
2082                 }
2083                 brelse(c_bh);
2084                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2085                                "journal-1006: found valid "
2086                                "transaction start offset %llu, len %d id %d",
2087                                d_bh->b_blocknr -
2088                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2089                                get_desc_trans_len(desc),
2090                                get_desc_trans_id(desc));
2091                 return 1;
2092         } else {
2093                 return 0;
2094         }
2095 }
2096
2097 static void brelse_array(struct buffer_head **heads, int num)
2098 {
2099         int i;
2100         for (i = 0; i < num; i++) {
2101                 brelse(heads[i]);
2102         }
2103 }
2104
2105 /*
2106  * given the start, and values for the oldest acceptable transactions,
2107  * this either reads in a replays a transaction, or returns because the
2108  * transaction is invalid, or too old.
2109  * NOTE: only called during fs mount
2110  */
2111 static int journal_read_transaction(struct super_block *sb,
2112                                     unsigned long cur_dblock,
2113                                     unsigned long oldest_start,
2114                                     unsigned int oldest_trans_id,
2115                                     unsigned long newest_mount_id)
2116 {
2117         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2118         struct reiserfs_journal_desc *desc;
2119         struct reiserfs_journal_commit *commit;
2120         unsigned int trans_id = 0;
2121         struct buffer_head *c_bh;
2122         struct buffer_head *d_bh;
2123         struct buffer_head **log_blocks = NULL;
2124         struct buffer_head **real_blocks = NULL;
2125         unsigned int trans_offset;
2126         int i;
2127         int trans_half;
2128
2129         d_bh = journal_bread(sb, cur_dblock);
2130         if (!d_bh)
2131                 return 1;
2132         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2133         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2134         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2135                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2136                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2137                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2138         if (get_desc_trans_id(desc) < oldest_trans_id) {
2139                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2140                                "journal_read_trans skipping because %lu is too old",
2141                                cur_dblock -
2142                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2143                 brelse(d_bh);
2144                 return 1;
2145         }
2146         if (get_desc_mount_id(desc) != newest_mount_id) {
2147                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2148                                "journal_read_trans skipping because %d is != "
2149                                "newest_mount_id %lu", get_desc_mount_id(desc),
2150                                newest_mount_id);
2151                 brelse(d_bh);
2152                 return 1;
2153         }
2154         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2155                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2156                               SB_ONDISK_JOURNAL_SIZE(sb)));
2157         if (!c_bh) {
2158                 brelse(d_bh);
2159                 return 1;
2160         }
2161         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2162         if (journal_compare_desc_commit(sb, desc, commit)) {
2163                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2164                                "journal_read_transaction, "
2165                                "commit offset %llu had bad time %d or length %d",
2166                                c_bh->b_blocknr -
2167                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2168                                get_commit_trans_id(commit),
2169                                get_commit_trans_len(commit));
2170                 brelse(c_bh);
2171                 brelse(d_bh);
2172                 return 1;
2173         }
2174
2175         if (bdev_read_only(sb->s_bdev)) {
2176                 reiserfs_warning(sb, "clm-2076",
2177                                  "device is readonly, unable to replay log");
2178                 brelse(c_bh);
2179                 brelse(d_bh);
2180                 return -EROFS;
2181         }
2182
2183         trans_id = get_desc_trans_id(desc);
2184         /*
2185          * now we know we've got a good transaction, and it was
2186          * inside the valid time ranges
2187          */
2188         log_blocks = kmalloc(get_desc_trans_len(desc) *
2189                              sizeof(struct buffer_head *), GFP_NOFS);
2190         real_blocks = kmalloc(get_desc_trans_len(desc) *
2191                               sizeof(struct buffer_head *), GFP_NOFS);
2192         if (!log_blocks || !real_blocks) {
2193                 brelse(c_bh);
2194                 brelse(d_bh);
2195                 kfree(log_blocks);
2196                 kfree(real_blocks);
2197                 reiserfs_warning(sb, "journal-1169",
2198                                  "kmalloc failed, unable to mount FS");
2199                 return -1;
2200         }
2201         /* get all the buffer heads */
2202         trans_half = journal_trans_half(sb->s_blocksize);
2203         for (i = 0; i < get_desc_trans_len(desc); i++) {
2204                 log_blocks[i] =
2205                     journal_getblk(sb,
2206                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2207                                    (trans_offset + 1 +
2208                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2209                 if (i < trans_half) {
2210                         real_blocks[i] =
2211                             sb_getblk(sb,
2212                                       le32_to_cpu(desc->j_realblock[i]));
2213                 } else {
2214                         real_blocks[i] =
2215                             sb_getblk(sb,
2216                                       le32_to_cpu(commit->
2217                                                   j_realblock[i - trans_half]));
2218                 }
2219                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2220                         reiserfs_warning(sb, "journal-1207",
2221                                          "REPLAY FAILURE fsck required! "
2222                                          "Block to replay is outside of "
2223                                          "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                     (sb, real_blocks[i]->b_blocknr)) {
2229                         reiserfs_warning(sb, "journal-1204",
2230                                          "REPLAY FAILURE fsck required! "
2231                                          "Trying to replay onto a log block");
2232 abort_replay:
2233                         brelse_array(log_blocks, i);
2234                         brelse_array(real_blocks, i);
2235                         brelse(c_bh);
2236                         brelse(d_bh);
2237                         kfree(log_blocks);
2238                         kfree(real_blocks);
2239                         return -1;
2240                 }
2241         }
2242         /* read in the log blocks, memcpy to the corresponding real block */
2243         ll_rw_block(READ, get_desc_trans_len(desc), log_blocks);
2244         for (i = 0; i < get_desc_trans_len(desc); i++) {
2245
2246                 wait_on_buffer(log_blocks[i]);
2247                 if (!buffer_uptodate(log_blocks[i])) {
2248                         reiserfs_warning(sb, "journal-1212",
2249                                          "REPLAY FAILURE fsck required! "
2250                                          "buffer write failed");
2251                         brelse_array(log_blocks + i,
2252                                      get_desc_trans_len(desc) - i);
2253                         brelse_array(real_blocks, get_desc_trans_len(desc));
2254                         brelse(c_bh);
2255                         brelse(d_bh);
2256                         kfree(log_blocks);
2257                         kfree(real_blocks);
2258                         return -1;
2259                 }
2260                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2261                        real_blocks[i]->b_size);
2262                 set_buffer_uptodate(real_blocks[i]);
2263                 brelse(log_blocks[i]);
2264         }
2265         /* flush out the real blocks */
2266         for (i = 0; i < get_desc_trans_len(desc); i++) {
2267                 set_buffer_dirty(real_blocks[i]);
2268                 write_dirty_buffer(real_blocks[i], WRITE);
2269         }
2270         for (i = 0; i < get_desc_trans_len(desc); i++) {
2271                 wait_on_buffer(real_blocks[i]);
2272                 if (!buffer_uptodate(real_blocks[i])) {
2273                         reiserfs_warning(sb, "journal-1226",
2274                                          "REPLAY FAILURE, fsck required! "
2275                                          "buffer write failed");
2276                         brelse_array(real_blocks + i,
2277                                      get_desc_trans_len(desc) - i);
2278                         brelse(c_bh);
2279                         brelse(d_bh);
2280                         kfree(log_blocks);
2281                         kfree(real_blocks);
2282                         return -1;
2283                 }
2284                 brelse(real_blocks[i]);
2285         }
2286         cur_dblock =
2287             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2288             ((trans_offset + get_desc_trans_len(desc) +
2289               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2290         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2291                        "journal-1095: setting journal " "start to offset %ld",
2292                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2293
2294         /*
2295          * init starting values for the first transaction, in case
2296          * this is the last transaction to be replayed.
2297          */
2298         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2299         journal->j_last_flush_trans_id = trans_id;
2300         journal->j_trans_id = trans_id + 1;
2301         /* check for trans_id overflow */
2302         if (journal->j_trans_id == 0)
2303                 journal->j_trans_id = 10;
2304         brelse(c_bh);
2305         brelse(d_bh);
2306         kfree(log_blocks);
2307         kfree(real_blocks);
2308         return 0;
2309 }
2310
2311 /*
2312  * This function reads blocks starting from block and to max_block of bufsize
2313  * size (but no more than BUFNR blocks at a time). This proved to improve
2314  * mounting speed on self-rebuilding raid5 arrays at least.
2315  * Right now it is only used from journal code. But later we might use it
2316  * from other places.
2317  * Note: Do not use journal_getblk/sb_getblk functions here!
2318  */
2319 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2320                                            b_blocknr_t block, int bufsize,
2321                                            b_blocknr_t max_block)
2322 {
2323         struct buffer_head *bhlist[BUFNR];
2324         unsigned int blocks = BUFNR;
2325         struct buffer_head *bh;
2326         int i, j;
2327
2328         bh = __getblk(dev, block, bufsize);
2329         if (buffer_uptodate(bh))
2330                 return (bh);
2331
2332         if (block + BUFNR > max_block) {
2333                 blocks = max_block - block;
2334         }
2335         bhlist[0] = bh;
2336         j = 1;
2337         for (i = 1; i < blocks; i++) {
2338                 bh = __getblk(dev, block + i, bufsize);
2339                 if (buffer_uptodate(bh)) {
2340                         brelse(bh);
2341                         break;
2342                 } else
2343                         bhlist[j++] = bh;
2344         }
2345         ll_rw_block(READ, j, bhlist);
2346         for (i = 1; i < j; i++)
2347                 brelse(bhlist[i]);
2348         bh = bhlist[0];
2349         wait_on_buffer(bh);
2350         if (buffer_uptodate(bh))
2351                 return bh;
2352         brelse(bh);
2353         return NULL;
2354 }
2355
2356 /*
2357  * read and replay the log
2358  * on a clean unmount, the journal header's next unflushed pointer will be
2359  * to an invalid transaction.  This tests that before finding all the
2360  * transactions in the log, which makes normal mount times fast.
2361  *
2362  * After a crash, this starts with the next unflushed transaction, and
2363  * replays until it finds one too old, or invalid.
2364  *
2365  * On exit, it sets things up so the first transaction will work correctly.
2366  * NOTE: only called during fs mount
2367  */
2368 static int journal_read(struct super_block *sb)
2369 {
2370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2371         struct reiserfs_journal_desc *desc;
2372         unsigned int oldest_trans_id = 0;
2373         unsigned int oldest_invalid_trans_id = 0;
2374         time_t start;
2375         unsigned long oldest_start = 0;
2376         unsigned long cur_dblock = 0;
2377         unsigned long newest_mount_id = 9;
2378         struct buffer_head *d_bh;
2379         struct reiserfs_journal_header *jh;
2380         int valid_journal_header = 0;
2381         int replay_count = 0;
2382         int continue_replay = 1;
2383         int ret;
2384         char b[BDEVNAME_SIZE];
2385
2386         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2387         reiserfs_info(sb, "checking transaction log (%s)\n",
2388                       bdevname(journal->j_dev_bd, b));
2389         start = get_seconds();
2390
2391         /*
2392          * step 1, read in the journal header block.  Check the transaction
2393          * it says is the first unflushed, and if that transaction is not
2394          * valid, replay is done
2395          */
2396         journal->j_header_bh = journal_bread(sb,
2397                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2398                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2399         if (!journal->j_header_bh) {
2400                 return 1;
2401         }
2402         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2403         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2404             SB_ONDISK_JOURNAL_SIZE(sb)
2405             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2406                 oldest_start =
2407                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2408                     le32_to_cpu(jh->j_first_unflushed_offset);
2409                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2410                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2411                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2412                                "journal-1153: found in "
2413                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2414                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2415                                le32_to_cpu(jh->j_last_flush_trans_id));
2416                 valid_journal_header = 1;
2417
2418                 /*
2419                  * now, we try to read the first unflushed offset.  If it
2420                  * is not valid, there is nothing more we can do, and it
2421                  * makes no sense to read through the whole log.
2422                  */
2423                 d_bh =
2424                     journal_bread(sb,
2425                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2426                                   le32_to_cpu(jh->j_first_unflushed_offset));
2427                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2428                 if (!ret) {
2429                         continue_replay = 0;
2430                 }
2431                 brelse(d_bh);
2432                 goto start_log_replay;
2433         }
2434
2435         /*
2436          * ok, there are transactions that need to be replayed.  start
2437          * with the first log block, find all the valid transactions, and
2438          * pick out the oldest.
2439          */
2440         while (continue_replay
2441                && cur_dblock <
2442                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2443                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2444                 /*
2445                  * Note that it is required for blocksize of primary fs
2446                  * device and journal device to be the same
2447                  */
2448                 d_bh =
2449                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2450                                     sb->s_blocksize,
2451                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2452                                     SB_ONDISK_JOURNAL_SIZE(sb));
2453                 ret =
2454                     journal_transaction_is_valid(sb, d_bh,
2455                                                  &oldest_invalid_trans_id,
2456                                                  &newest_mount_id);
2457                 if (ret == 1) {
2458                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2459                         if (oldest_start == 0) {        /* init all oldest_ values */
2460                                 oldest_trans_id = get_desc_trans_id(desc);
2461                                 oldest_start = d_bh->b_blocknr;
2462                                 newest_mount_id = get_desc_mount_id(desc);
2463                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2464                                                "journal-1179: Setting "
2465                                                "oldest_start to offset %llu, trans_id %lu",
2466                                                oldest_start -
2467                                                SB_ONDISK_JOURNAL_1st_BLOCK
2468                                                (sb), oldest_trans_id);
2469                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2470                                 /* one we just read was older */
2471                                 oldest_trans_id = get_desc_trans_id(desc);
2472                                 oldest_start = d_bh->b_blocknr;
2473                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2474                                                "journal-1180: Resetting "
2475                                                "oldest_start to offset %lu, trans_id %lu",
2476                                                oldest_start -
2477                                                SB_ONDISK_JOURNAL_1st_BLOCK
2478                                                (sb), oldest_trans_id);
2479                         }
2480                         if (newest_mount_id < get_desc_mount_id(desc)) {
2481                                 newest_mount_id = get_desc_mount_id(desc);
2482                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2483                                                "journal-1299: Setting "
2484                                                "newest_mount_id to %d",
2485                                                get_desc_mount_id(desc));
2486                         }
2487                         cur_dblock += get_desc_trans_len(desc) + 2;
2488                 } else {
2489                         cur_dblock++;
2490                 }
2491                 brelse(d_bh);
2492         }
2493
2494 start_log_replay:
2495         cur_dblock = oldest_start;
2496         if (oldest_trans_id) {
2497                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2498                                "journal-1206: Starting replay "
2499                                "from offset %llu, trans_id %lu",
2500                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2501                                oldest_trans_id);
2502
2503         }
2504         replay_count = 0;
2505         while (continue_replay && oldest_trans_id > 0) {
2506                 ret =
2507                     journal_read_transaction(sb, cur_dblock, oldest_start,
2508                                              oldest_trans_id, newest_mount_id);
2509                 if (ret < 0) {
2510                         return ret;
2511                 } else if (ret != 0) {
2512                         break;
2513                 }
2514                 cur_dblock =
2515                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2516                 replay_count++;
2517                 if (cur_dblock == oldest_start)
2518                         break;
2519         }
2520
2521         if (oldest_trans_id == 0) {
2522                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2523                                "journal-1225: No valid " "transactions found");
2524         }
2525         /*
2526          * j_start does not get set correctly if we don't replay any
2527          * transactions.  if we had a valid journal_header, set j_start
2528          * to the first unflushed transaction value, copy the trans_id
2529          * from the header
2530          */
2531         if (valid_journal_header && replay_count == 0) {
2532                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2533                 journal->j_trans_id =
2534                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2535                 /* check for trans_id overflow */
2536                 if (journal->j_trans_id == 0)
2537                         journal->j_trans_id = 10;
2538                 journal->j_last_flush_trans_id =
2539                     le32_to_cpu(jh->j_last_flush_trans_id);
2540                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2541         } else {
2542                 journal->j_mount_id = newest_mount_id + 1;
2543         }
2544         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2545                        "newest_mount_id to %lu", journal->j_mount_id);
2546         journal->j_first_unflushed_offset = journal->j_start;
2547         if (replay_count > 0) {
2548                 reiserfs_info(sb,
2549                               "replayed %d transactions in %lu seconds\n",
2550                               replay_count, get_seconds() - start);
2551         }
2552         /* needed to satisfy the locking in _update_journal_header_block */
2553         reiserfs_write_lock(sb);
2554         if (!bdev_read_only(sb->s_bdev) &&
2555             _update_journal_header_block(sb, journal->j_start,
2556                                          journal->j_last_flush_trans_id)) {
2557                 reiserfs_write_unlock(sb);
2558                 /*
2559                  * replay failed, caller must call free_journal_ram and abort
2560                  * the mount
2561                  */
2562                 return -1;
2563         }
2564         reiserfs_write_unlock(sb);
2565         return 0;
2566 }
2567
2568 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2569 {
2570         struct reiserfs_journal_list *jl;
2571         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2572                      GFP_NOFS | __GFP_NOFAIL);
2573         INIT_LIST_HEAD(&jl->j_list);
2574         INIT_LIST_HEAD(&jl->j_working_list);
2575         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2576         INIT_LIST_HEAD(&jl->j_bh_list);
2577         mutex_init(&jl->j_commit_mutex);
2578         SB_JOURNAL(s)->j_num_lists++;
2579         get_journal_list(jl);
2580         return jl;
2581 }
2582
2583 static void journal_list_init(struct super_block *sb)
2584 {
2585         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2586 }
2587
2588 static void release_journal_dev(struct super_block *super,
2589                                struct reiserfs_journal *journal)
2590 {
2591         if (journal->j_dev_bd != NULL) {
2592                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2593                 journal->j_dev_bd = NULL;
2594         }
2595 }
2596
2597 static int journal_init_dev(struct super_block *super,
2598                             struct reiserfs_journal *journal,
2599                             const char *jdev_name)
2600 {
2601         int result;
2602         dev_t jdev;
2603         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2604         char b[BDEVNAME_SIZE];
2605
2606         result = 0;
2607
2608         journal->j_dev_bd = NULL;
2609         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2610             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2611
2612         if (bdev_read_only(super->s_bdev))
2613                 blkdev_mode = FMODE_READ;
2614
2615         /* there is no "jdev" option and journal is on separate device */
2616         if ((!jdev_name || !jdev_name[0])) {
2617                 if (jdev == super->s_dev)
2618                         blkdev_mode &= ~FMODE_EXCL;
2619                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2620                                                       journal);
2621                 journal->j_dev_mode = blkdev_mode;
2622                 if (IS_ERR(journal->j_dev_bd)) {
2623                         result = PTR_ERR(journal->j_dev_bd);
2624                         journal->j_dev_bd = NULL;
2625                         reiserfs_warning(super, "sh-458",
2626                                          "cannot init journal device '%s': %i",
2627                                          __bdevname(jdev, b), result);
2628                         return result;
2629                 } else if (jdev != super->s_dev)
2630                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2631
2632                 return 0;
2633         }
2634
2635         journal->j_dev_mode = blkdev_mode;
2636         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2637         if (IS_ERR(journal->j_dev_bd)) {
2638                 result = PTR_ERR(journal->j_dev_bd);
2639                 journal->j_dev_bd = NULL;
2640                 reiserfs_warning(super,
2641                                  "journal_init_dev: Cannot open '%s': %i",
2642                                  jdev_name, result);
2643                 return result;
2644         }
2645
2646         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2647         reiserfs_info(super,
2648                       "journal_init_dev: journal device: %s\n",
2649                       bdevname(journal->j_dev_bd, b));
2650         return 0;
2651 }
2652
2653 /*
2654  * When creating/tuning a file system user can assign some
2655  * journal params within boundaries which depend on the ratio
2656  * blocksize/standard_blocksize.
2657  *
2658  * For blocks >= standard_blocksize transaction size should
2659  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2660  * then JOURNAL_TRANS_MAX_DEFAULT.
2661  *
2662  * For blocks < standard_blocksize these boundaries should be
2663  * decreased proportionally.
2664  */
2665 #define REISERFS_STANDARD_BLKSIZE (4096)
2666
2667 static int check_advise_trans_params(struct super_block *sb,
2668                                      struct reiserfs_journal *journal)
2669 {
2670         if (journal->j_trans_max) {
2671                 /* Non-default journal params.  Do sanity check for them. */
2672                 int ratio = 1;
2673                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2674                         ratio = REISERFS_STANDARD_BLKSIZE / 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(sb) / journal->j_trans_max <
2679                     JOURNAL_MIN_RATIO) {
2680                         reiserfs_warning(sb, "sh-462",
2681                                          "bad transaction max size (%u). "
2682                                          "FSCK?", 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(sb, "sh-463",
2688                                          "bad transaction max batch (%u). "
2689                                          "FSCK?", journal->j_max_batch);
2690                         return 1;
2691                 }
2692         } else {
2693                 /*
2694                  * Default journal params.
2695                  * The file system was created by old version
2696                  * of mkreiserfs, so some fields contain zeros,
2697                  * and we need to advise proper values for them
2698                  */
2699                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2700                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2701                                          sb->s_blocksize);
2702                         return 1;
2703                 }
2704                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2705                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2706                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2707         }
2708         return 0;
2709 }
2710
2711 /* must be called once on fs mount.  calls journal_read for you */
2712 int journal_init(struct super_block *sb, const char *j_dev_name,
2713                  int old_format, unsigned int commit_max_age)
2714 {
2715         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2716         struct buffer_head *bhjh;
2717         struct reiserfs_super_block *rs;
2718         struct reiserfs_journal_header *jh;
2719         struct reiserfs_journal *journal;
2720         struct reiserfs_journal_list *jl;
2721         char b[BDEVNAME_SIZE];
2722         int ret;
2723
2724         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2725         if (!journal) {
2726                 reiserfs_warning(sb, "journal-1256",
2727                                  "unable to get memory for journal structure");
2728                 return 1;
2729         }
2730         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2731         INIT_LIST_HEAD(&journal->j_prealloc_list);
2732         INIT_LIST_HEAD(&journal->j_working_list);
2733         INIT_LIST_HEAD(&journal->j_journal_list);
2734         journal->j_persistent_trans = 0;
2735         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2736                                            reiserfs_bmap_count(sb)))
2737                 goto free_and_return;
2738
2739         allocate_bitmap_nodes(sb);
2740
2741         /* reserved for journal area support */
2742         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2743                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2744                                                  / sb->s_blocksize +
2745                                                  reiserfs_bmap_count(sb) +
2746                                                  1 :
2747                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2748                                                  sb->s_blocksize + 2);
2749
2750         /*
2751          * Sanity check to see is the standard journal fitting
2752          * within first bitmap (actual for small blocksizes)
2753          */
2754         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2755             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2756              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2757                 reiserfs_warning(sb, "journal-1393",
2758                                  "journal does not fit for area addressed "
2759                                  "by first of bitmap blocks. It starts at "
2760                                  "%u and its size is %u. Block size %ld",
2761                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2762                                  SB_ONDISK_JOURNAL_SIZE(sb),
2763                                  sb->s_blocksize);
2764                 goto free_and_return;
2765         }
2766
2767         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2768                 reiserfs_warning(sb, "sh-462",
2769                                  "unable to initialize jornal device");
2770                 goto free_and_return;
2771         }
2772
2773         rs = SB_DISK_SUPER_BLOCK(sb);
2774
2775         /* read journal header */
2776         bhjh = journal_bread(sb,
2777                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2778                              SB_ONDISK_JOURNAL_SIZE(sb));
2779         if (!bhjh) {
2780                 reiserfs_warning(sb, "sh-459",
2781                                  "unable to read journal header");
2782                 goto free_and_return;
2783         }
2784         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2785
2786         /* make sure that journal matches to the super block */
2787         if (is_reiserfs_jr(rs)
2788             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2789                 sb_jp_journal_magic(rs))) {
2790                 reiserfs_warning(sb, "sh-460",
2791                                  "journal header magic %x (device %s) does "
2792                                  "not match to magic found in super block %x",
2793                                  jh->jh_journal.jp_journal_magic,
2794                                  bdevname(journal->j_dev_bd, b),
2795                                  sb_jp_journal_magic(rs));
2796                 brelse(bhjh);
2797                 goto free_and_return;
2798         }
2799
2800         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2801         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2802         journal->j_max_commit_age =
2803             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2804         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2805
2806         if (check_advise_trans_params(sb, journal) != 0)
2807                 goto free_and_return;
2808         journal->j_default_max_commit_age = journal->j_max_commit_age;
2809
2810         if (commit_max_age != 0) {
2811                 journal->j_max_commit_age = commit_max_age;
2812                 journal->j_max_trans_age = commit_max_age;
2813         }
2814
2815         reiserfs_info(sb, "journal params: device %s, size %u, "
2816                       "journal first block %u, max trans len %u, max batch %u, "
2817                       "max commit age %u, max trans age %u\n",
2818                       bdevname(journal->j_dev_bd, b),
2819                       SB_ONDISK_JOURNAL_SIZE(sb),
2820                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2821                       journal->j_trans_max,
2822                       journal->j_max_batch,
2823                       journal->j_max_commit_age, journal->j_max_trans_age);
2824
2825         brelse(bhjh);
2826
2827         journal->j_list_bitmap_index = 0;
2828         journal_list_init(sb);
2829
2830         memset(journal->j_list_hash_table, 0,
2831                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2832
2833         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2834         spin_lock_init(&journal->j_dirty_buffers_lock);
2835
2836         journal->j_start = 0;
2837         journal->j_len = 0;
2838         journal->j_len_alloc = 0;
2839         atomic_set(&journal->j_wcount, 0);
2840         atomic_set(&journal->j_async_throttle, 0);
2841         journal->j_bcount = 0;
2842         journal->j_trans_start_time = 0;
2843         journal->j_last = NULL;
2844         journal->j_first = NULL;
2845         init_waitqueue_head(&journal->j_join_wait);
2846         mutex_init(&journal->j_mutex);
2847         mutex_init(&journal->j_flush_mutex);
2848
2849         journal->j_trans_id = 10;
2850         journal->j_mount_id = 10;
2851         journal->j_state = 0;
2852         atomic_set(&journal->j_jlock, 0);
2853         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2854         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2855         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2856         journal->j_cnode_used = 0;
2857         journal->j_must_wait = 0;
2858
2859         if (journal->j_cnode_free == 0) {
2860                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2861                                  "allocation failed (%ld bytes). Journal is "
2862                                  "too large for available memory. Usually "
2863                                  "this is due to a journal that is too large.",
2864                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2865                 goto free_and_return;
2866         }
2867
2868         init_journal_hash(sb);
2869         jl = journal->j_current_jl;
2870
2871         /*
2872          * get_list_bitmap() may call flush_commit_list() which
2873          * requires the lock. Calling flush_commit_list() shouldn't happen
2874          * this early but I like to be paranoid.
2875          */
2876         reiserfs_write_lock(sb);
2877         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2878         reiserfs_write_unlock(sb);
2879         if (!jl->j_list_bitmap) {
2880                 reiserfs_warning(sb, "journal-2005",
2881                                  "get_list_bitmap failed for journal list 0");
2882                 goto free_and_return;
2883         }
2884
2885         ret = journal_read(sb);
2886         if (ret < 0) {
2887                 reiserfs_warning(sb, "reiserfs-2006",
2888                                  "Replay Failure, unable to mount");
2889                 goto free_and_return;
2890         }
2891
2892         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2893         journal->j_work_sb = sb;
2894         return 0;
2895 free_and_return:
2896         free_journal_ram(sb);
2897         return 1;
2898 }
2899
2900 /*
2901  * test for a polite end of the current transaction.  Used by file_write,
2902  * and should be used by delete to make sure they don't write more than
2903  * can fit inside a single transaction
2904  */
2905 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2906                                    int new_alloc)
2907 {
2908         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2909         time_t now = get_seconds();
2910         /* cannot restart while nested */
2911         BUG_ON(!th->t_trans_id);
2912         if (th->t_refcount > 1)
2913                 return 0;
2914         if (journal->j_must_wait > 0 ||
2915             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2916             atomic_read(&journal->j_jlock) ||
2917             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2918             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2919                 return 1;
2920         }
2921
2922         journal->j_len_alloc += new_alloc;
2923         th->t_blocks_allocated += new_alloc ;
2924         return 0;
2925 }
2926
2927 /* this must be called inside a transaction */
2928 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2929 {
2930         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2931         BUG_ON(!th->t_trans_id);
2932         journal->j_must_wait = 1;
2933         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2934         return;
2935 }
2936
2937 /* this must be called without a transaction started */
2938 void reiserfs_allow_writes(struct super_block *s)
2939 {
2940         struct reiserfs_journal *journal = SB_JOURNAL(s);
2941         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2942         wake_up(&journal->j_join_wait);
2943 }
2944
2945 /* this must be called without a transaction started */
2946 void reiserfs_wait_on_write_block(struct super_block *s)
2947 {
2948         struct reiserfs_journal *journal = SB_JOURNAL(s);
2949         wait_event(journal->j_join_wait,
2950                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2951 }
2952
2953 static void queue_log_writer(struct super_block *s)
2954 {
2955         wait_queue_t wait;
2956         struct reiserfs_journal *journal = SB_JOURNAL(s);
2957         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2958
2959         /*
2960          * we don't want to use wait_event here because
2961          * we only want to wait once.
2962          */
2963         init_waitqueue_entry(&wait, current);
2964         add_wait_queue(&journal->j_join_wait, &wait);
2965         set_current_state(TASK_UNINTERRUPTIBLE);
2966         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2967                 int depth = reiserfs_write_unlock_nested(s);
2968                 schedule();
2969                 reiserfs_write_lock_nested(s, depth);
2970         }
2971         __set_current_state(TASK_RUNNING);
2972         remove_wait_queue(&journal->j_join_wait, &wait);
2973 }
2974
2975 static void wake_queued_writers(struct super_block *s)
2976 {
2977         struct reiserfs_journal *journal = SB_JOURNAL(s);
2978         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2979                 wake_up(&journal->j_join_wait);
2980 }
2981
2982 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2983 {
2984         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2985         unsigned long bcount = journal->j_bcount;
2986         while (1) {
2987                 int depth;
2988
2989                 depth = reiserfs_write_unlock_nested(sb);
2990                 schedule_timeout_uninterruptible(1);
2991                 reiserfs_write_lock_nested(sb, depth);
2992
2993                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2994                 while ((atomic_read(&journal->j_wcount) > 0 ||
2995                         atomic_read(&journal->j_jlock)) &&
2996                        journal->j_trans_id == trans_id) {
2997                         queue_log_writer(sb);
2998                 }
2999                 if (journal->j_trans_id != trans_id)
3000                         break;
3001                 if (bcount == journal->j_bcount)
3002                         break;
3003                 bcount = journal->j_bcount;
3004         }
3005 }
3006
3007 /*
3008  * join == true if you must join an existing transaction.
3009  * join == false if you can deal with waiting for others to finish
3010  *
3011  * this will block until the transaction is joinable.  send the number of
3012  * blocks you expect to use in nblocks.
3013 */
3014 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3015                               struct super_block *sb, unsigned long nblocks,
3016                               int join)
3017 {
3018         time_t now = get_seconds();
3019         unsigned int old_trans_id;
3020         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3021         struct reiserfs_transaction_handle myth;
3022         int sched_count = 0;
3023         int retval;
3024         int depth;
3025
3026         reiserfs_check_lock_depth(sb, "journal_begin");
3027         BUG_ON(nblocks > journal->j_trans_max);
3028
3029         PROC_INFO_INC(sb, journal.journal_being);
3030         /* set here for journal_join */
3031         th->t_refcount = 1;
3032         th->t_super = sb;
3033
3034 relock:
3035         lock_journal(sb);
3036         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3037                 unlock_journal(sb);
3038                 retval = journal->j_errno;
3039                 goto out_fail;
3040         }
3041         journal->j_bcount++;
3042
3043         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3044                 unlock_journal(sb);
3045                 depth = reiserfs_write_unlock_nested(sb);
3046                 reiserfs_wait_on_write_block(sb);
3047                 reiserfs_write_lock_nested(sb, depth);
3048                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3049                 goto relock;
3050         }
3051         now = get_seconds();
3052
3053         /*
3054          * if there is no room in the journal OR
3055          * if this transaction is too old, and we weren't called joinable,
3056          * wait for it to finish before beginning we don't sleep if there
3057          * aren't other writers
3058          */
3059
3060         if ((!join && journal->j_must_wait > 0) ||
3061             (!join
3062              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3063             || (!join && atomic_read(&journal->j_wcount) > 0
3064                 && journal->j_trans_start_time > 0
3065                 && (now - journal->j_trans_start_time) >
3066                 journal->j_max_trans_age) || (!join
3067                                               && atomic_read(&journal->j_jlock))
3068             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3069
3070                 old_trans_id = journal->j_trans_id;
3071                 /* allow others to finish this transaction */
3072                 unlock_journal(sb);
3073
3074                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3075                     journal->j_max_batch &&
3076                     ((journal->j_len + nblocks + 2) * 100) <
3077                     (journal->j_len_alloc * 75)) {
3078                         if (atomic_read(&journal->j_wcount) > 10) {
3079                                 sched_count++;
3080                                 queue_log_writer(sb);
3081                                 goto relock;
3082                         }
3083                 }
3084                 /*
3085                  * don't mess with joining the transaction if all we
3086                  * have to do is wait for someone else to do a commit
3087                  */
3088                 if (atomic_read(&journal->j_jlock)) {
3089                         while (journal->j_trans_id == old_trans_id &&
3090                                atomic_read(&journal->j_jlock)) {
3091                                 queue_log_writer(sb);
3092                         }
3093                         goto relock;
3094                 }
3095                 retval = journal_join(&myth, sb);
3096                 if (retval)
3097                         goto out_fail;
3098
3099                 /* someone might have ended the transaction while we joined */
3100                 if (old_trans_id != journal->j_trans_id) {
3101                         retval = do_journal_end(&myth, 0);
3102                 } else {
3103                         retval = do_journal_end(&myth, COMMIT_NOW);
3104                 }
3105
3106                 if (retval)
3107                         goto out_fail;
3108
3109                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3110                 goto relock;
3111         }
3112         /* we are the first writer, set trans_id */
3113         if (journal->j_trans_start_time == 0) {
3114                 journal->j_trans_start_time = get_seconds();
3115         }
3116         atomic_inc(&journal->j_wcount);
3117         journal->j_len_alloc += nblocks;
3118         th->t_blocks_logged = 0;
3119         th->t_blocks_allocated = nblocks;
3120         th->t_trans_id = journal->j_trans_id;
3121         unlock_journal(sb);
3122         INIT_LIST_HEAD(&th->t_list);
3123         return 0;
3124
3125 out_fail:
3126         memset(th, 0, sizeof(*th));
3127         /*
3128          * Re-set th->t_super, so we can properly keep track of how many
3129          * persistent transactions there are. We need to do this so if this
3130          * call is part of a failed restart_transaction, we can free it later
3131          */
3132         th->t_super = sb;
3133         return retval;
3134 }
3135
3136 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3137                                                                     super_block
3138                                                                     *s,
3139                                                                     int nblocks)
3140 {
3141         int ret;
3142         struct reiserfs_transaction_handle *th;
3143
3144         /*
3145          * if we're nesting into an existing transaction.  It will be
3146          * persistent on its own
3147          */
3148         if (reiserfs_transaction_running(s)) {
3149                 th = current->journal_info;
3150                 th->t_refcount++;
3151                 BUG_ON(th->t_refcount < 2);
3152
3153                 return th;
3154         }
3155         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3156         if (!th)
3157                 return NULL;
3158         ret = journal_begin(th, s, nblocks);
3159         if (ret) {
3160                 kfree(th);
3161                 return NULL;
3162         }
3163
3164         SB_JOURNAL(s)->j_persistent_trans++;
3165         return th;
3166 }
3167
3168 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3169 {
3170         struct super_block *s = th->t_super;
3171         int ret = 0;
3172         if (th->t_trans_id)
3173                 ret = journal_end(th);
3174         else
3175                 ret = -EIO;
3176         if (th->t_refcount == 0) {
3177                 SB_JOURNAL(s)->j_persistent_trans--;
3178                 kfree(th);
3179         }
3180         return ret;
3181 }
3182
3183 static int journal_join(struct reiserfs_transaction_handle *th,
3184                         struct super_block *sb)
3185 {
3186         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3187
3188         /*
3189          * this keeps do_journal_end from NULLing out the
3190          * current->journal_info pointer
3191          */
3192         th->t_handle_save = cur_th;
3193         BUG_ON(cur_th && cur_th->t_refcount > 1);
3194         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3195 }
3196
3197 int journal_join_abort(struct reiserfs_transaction_handle *th,
3198                        struct super_block *sb)
3199 {
3200         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3201
3202         /*
3203          * this keeps do_journal_end from NULLing out the
3204          * current->journal_info pointer
3205          */
3206         th->t_handle_save = cur_th;
3207         BUG_ON(cur_th && cur_th->t_refcount > 1);
3208         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3209 }
3210
3211 int journal_begin(struct reiserfs_transaction_handle *th,
3212                   struct super_block *sb, unsigned long nblocks)
3213 {
3214         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3215         int ret;
3216
3217         th->t_handle_save = NULL;
3218         if (cur_th) {
3219                 /* we are nesting into the current transaction */
3220                 if (cur_th->t_super == sb) {
3221                         BUG_ON(!cur_th->t_refcount);
3222                         cur_th->t_refcount++;
3223                         memcpy(th, cur_th, sizeof(*th));
3224                         if (th->t_refcount <= 1)
3225                                 reiserfs_warning(sb, "reiserfs-2005",
3226                                                  "BAD: refcount <= 1, but "
3227                                                  "journal_info != 0");
3228                         return 0;
3229                 } else {
3230                         /*
3231                          * we've ended up with a handle from a different
3232                          * filesystem.  save it and restore on journal_end.
3233                          * This should never really happen...
3234                          */
3235                         reiserfs_warning(sb, "clm-2100",
3236                                          "nesting info a different FS");
3237                         th->t_handle_save = current->journal_info;
3238                         current->journal_info = th;
3239                 }
3240         } else {
3241                 current->journal_info = th;
3242         }
3243         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3244         BUG_ON(current->journal_info != th);
3245
3246         /*
3247          * I guess this boils down to being the reciprocal of clm-2100 above.
3248          * If do_journal_begin_r fails, we need to put it back, since
3249          * journal_end won't be called to do it. */
3250         if (ret)
3251                 current->journal_info = th->t_handle_save;
3252         else
3253                 BUG_ON(!th->t_refcount);
3254
3255         return ret;
3256 }
3257
3258 /*
3259  * puts bh into the current transaction.  If it was already there, reorders
3260  * removes the old pointers from the hash, and puts new ones in (to make
3261  * sure replay happen in the right order).
3262  *
3263  * if it was dirty, cleans and files onto the clean list.  I can't let it
3264  * be dirty again until the transaction is committed.
3265  *
3266  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3267  */
3268 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3269                        struct buffer_head *bh)
3270 {
3271         struct super_block *sb = th->t_super;
3272         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3273         struct reiserfs_journal_cnode *cn = NULL;
3274         int count_already_incd = 0;
3275         int prepared = 0;
3276         BUG_ON(!th->t_trans_id);
3277
3278         PROC_INFO_INC(sb, journal.mark_dirty);
3279         if (th->t_trans_id != journal->j_trans_id) {
3280                 reiserfs_panic(th->t_super, "journal-1577",
3281                                "handle trans id %ld != current trans id %ld",
3282                                th->t_trans_id, journal->j_trans_id);
3283         }
3284
3285         prepared = test_clear_buffer_journal_prepared(bh);
3286         clear_buffer_journal_restore_dirty(bh);
3287         /* already in this transaction, we are done */
3288         if (buffer_journaled(bh)) {
3289                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3290                 return 0;
3291         }
3292
3293         /*
3294          * this must be turned into a panic instead of a warning.  We can't
3295          * allow a dirty or journal_dirty or locked buffer to be logged, as
3296          * some changes could get to disk too early.  NOT GOOD.
3297          */
3298         if (!prepared || buffer_dirty(bh)) {
3299                 reiserfs_warning(sb, "journal-1777",
3300                                  "buffer %llu bad state "
3301                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3302                                  (unsigned long long)bh->b_blocknr,
3303                                  prepared ? ' ' : '!',
3304                                  buffer_locked(bh) ? ' ' : '!',
3305                                  buffer_dirty(bh) ? ' ' : '!',
3306                                  buffer_journal_dirty(bh) ? ' ' : '!');
3307         }
3308
3309         if (atomic_read(&journal->j_wcount) <= 0) {
3310                 reiserfs_warning(sb, "journal-1409",
3311                                  "returning because j_wcount was %d",
3312                                  atomic_read(&journal->j_wcount));
3313                 return 1;
3314         }
3315         /*
3316          * this error means I've screwed up, and we've overflowed
3317          * the transaction.  Nothing can be done here, except make the
3318          * FS readonly or panic.
3319          */
3320         if (journal->j_len >= journal->j_trans_max) {
3321                 reiserfs_panic(th->t_super, "journal-1413",
3322                                "j_len (%lu) is too big",
3323                                journal->j_len);
3324         }
3325
3326         if (buffer_journal_dirty(bh)) {
3327                 count_already_incd = 1;
3328                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3329                 clear_buffer_journal_dirty(bh);
3330         }
3331
3332         if (journal->j_len > journal->j_len_alloc) {
3333                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3334         }
3335
3336         set_buffer_journaled(bh);
3337
3338         /* now put this guy on the end */
3339         if (!cn) {
3340                 cn = get_cnode(sb);
3341                 if (!cn) {
3342                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3343                 }
3344
3345                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3346                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3347                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3348                 }
3349                 th->t_blocks_logged++;
3350                 journal->j_len++;
3351
3352                 cn->bh = bh;
3353                 cn->blocknr = bh->b_blocknr;
3354                 cn->sb = sb;
3355                 cn->jlist = NULL;
3356                 insert_journal_hash(journal->j_hash_table, cn);
3357                 if (!count_already_incd) {
3358                         get_bh(bh);
3359                 }
3360         }
3361         cn->next = NULL;
3362         cn->prev = journal->j_last;
3363         cn->bh = bh;
3364         if (journal->j_last) {
3365                 journal->j_last->next = cn;
3366                 journal->j_last = cn;
3367         } else {
3368                 journal->j_first = cn;
3369                 journal->j_last = cn;
3370         }
3371         reiserfs_schedule_old_flush(sb);
3372         return 0;
3373 }
3374
3375 int journal_end(struct reiserfs_transaction_handle *th)
3376 {
3377         struct super_block *sb = th->t_super;
3378         if (!current->journal_info && th->t_refcount > 1)
3379                 reiserfs_warning(sb, "REISER-NESTING",
3380                                  "th NULL, refcount %d", th->t_refcount);
3381
3382         if (!th->t_trans_id) {
3383                 WARN_ON(1);
3384                 return -EIO;
3385         }
3386
3387         th->t_refcount--;
3388         if (th->t_refcount > 0) {
3389                 struct reiserfs_transaction_handle *cur_th =
3390                     current->journal_info;
3391
3392                 /*
3393                  * we aren't allowed to close a nested transaction on a
3394                  * different filesystem from the one in the task struct
3395                  */
3396                 BUG_ON(cur_th->t_super != th->t_super);
3397
3398                 if (th != cur_th) {
3399                         memcpy(current->journal_info, th, sizeof(*th));
3400                         th->t_trans_id = 0;
3401                 }
3402                 return 0;
3403         } else {
3404                 return do_journal_end(th, 0);
3405         }
3406 }
3407
3408 /*
3409  * removes from the current transaction, relsing and descrementing any counters.
3410  * also files the removed buffer directly onto the clean list
3411  *
3412  * called by journal_mark_freed when a block has been deleted
3413  *
3414  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3415  */
3416 static int remove_from_transaction(struct super_block *sb,
3417                                    b_blocknr_t blocknr, int already_cleaned)
3418 {
3419         struct buffer_head *bh;
3420         struct reiserfs_journal_cnode *cn;
3421         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3422         int ret = 0;
3423
3424         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3425         if (!cn || !cn->bh) {
3426                 return ret;
3427         }
3428         bh = cn->bh;
3429         if (cn->prev) {
3430                 cn->prev->next = cn->next;
3431         }
3432         if (cn->next) {
3433                 cn->next->prev = cn->prev;
3434         }
3435         if (cn == journal->j_first) {
3436                 journal->j_first = cn->next;
3437         }
3438         if (cn == journal->j_last) {
3439                 journal->j_last = cn->prev;
3440         }
3441         if (bh)
3442                 remove_journal_hash(sb, journal->j_hash_table, NULL,
3443                                     bh->b_blocknr, 0);
3444         clear_buffer_journaled(bh);     /* don't log this one */
3445
3446         if (!already_cleaned) {
3447                 clear_buffer_journal_dirty(bh);
3448                 clear_buffer_dirty(bh);
3449                 clear_buffer_journal_test(bh);
3450                 put_bh(bh);
3451                 if (atomic_read(&bh->b_count) < 0) {
3452                         reiserfs_warning(sb, "journal-1752",
3453                                          "b_count < 0");
3454                 }
3455                 ret = 1;
3456         }
3457         journal->j_len--;
3458         journal->j_len_alloc--;
3459         free_cnode(sb, cn);
3460         return ret;
3461 }
3462
3463 /*
3464  * for any cnode in a journal list, it can only be dirtied of all the
3465  * transactions that include it are committed to disk.
3466  * this checks through each transaction, and returns 1 if you are allowed
3467  * to dirty, and 0 if you aren't
3468  *
3469  * it is called by dirty_journal_list, which is called after
3470  * flush_commit_list has gotten all the log blocks for a given
3471  * transaction on disk
3472  *
3473  */
3474 static int can_dirty(struct reiserfs_journal_cnode *cn)
3475 {
3476         struct super_block *sb = cn->sb;
3477         b_blocknr_t blocknr = cn->blocknr;
3478         struct reiserfs_journal_cnode *cur = cn->hprev;
3479         int can_dirty = 1;
3480
3481         /*
3482          * first test hprev.  These are all newer than cn, so any node here
3483          * with the same block number and dev means this node can't be sent
3484          * to disk right now.
3485          */
3486         while (cur && can_dirty) {
3487                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3488                     cur->blocknr == blocknr) {
3489                         can_dirty = 0;
3490                 }
3491                 cur = cur->hprev;
3492         }
3493         /*
3494          * then test hnext.  These are all older than cn.  As long as they
3495          * are committed to the log, it is safe to write cn to disk
3496          */
3497         cur = cn->hnext;
3498         while (cur && can_dirty) {
3499                 if (cur->jlist && cur->jlist->j_len > 0 &&
3500                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3501                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3502                         can_dirty = 0;
3503                 }
3504                 cur = cur->hnext;
3505         }
3506         return can_dirty;
3507 }
3508
3509 /*
3510  * syncs the commit blocks, but does not force the real buffers to disk
3511  * will wait until the current transaction is done/committed before returning
3512  */
3513 int journal_end_sync(struct reiserfs_transaction_handle *th)
3514 {
3515         struct super_block *sb = th->t_super;
3516         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3517
3518         BUG_ON(!th->t_trans_id);
3519         /* you can sync while nested, very, very bad */
3520         BUG_ON(th->t_refcount > 1);
3521         if (journal->j_len == 0) {
3522                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3523                                              1);
3524                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3525         }
3526         return do_journal_end(th, COMMIT_NOW | WAIT);
3527 }
3528
3529 /* writeback the pending async commits to disk */
3530 static void flush_async_commits(struct work_struct *work)
3531 {
3532         struct reiserfs_journal *journal =
3533                 container_of(work, struct reiserfs_journal, j_work.work);
3534         struct super_block *sb = journal->j_work_sb;
3535         struct reiserfs_journal_list *jl;
3536         struct list_head *entry;
3537
3538         reiserfs_write_lock(sb);
3539         if (!list_empty(&journal->j_journal_list)) {
3540                 /* last entry is the youngest, commit it and you get everything */
3541                 entry = journal->j_journal_list.prev;
3542                 jl = JOURNAL_LIST_ENTRY(entry);
3543                 flush_commit_list(sb, jl, 1);
3544         }
3545         reiserfs_write_unlock(sb);
3546 }
3547
3548 /*
3549  * flushes any old transactions to disk
3550  * ends the current transaction if it is too old
3551  */
3552 void reiserfs_flush_old_commits(struct super_block *sb)
3553 {
3554         time_t now;
3555         struct reiserfs_transaction_handle th;
3556         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3557
3558         now = get_seconds();
3559         /*
3560          * safety check so we don't flush while we are replaying the log during
3561          * mount
3562          */
3563         if (list_empty(&journal->j_journal_list))
3564                 return;
3565
3566         /*
3567          * check the current transaction.  If there are no writers, and it is
3568          * too old, finish it, and force the commit blocks to disk
3569          */
3570         if (atomic_read(&journal->j_wcount) <= 0 &&
3571             journal->j_trans_start_time > 0 &&
3572             journal->j_len > 0 &&
3573             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3574                 if (!journal_join(&th, sb)) {
3575                         reiserfs_prepare_for_journal(sb,
3576                                                      SB_BUFFER_WITH_SB(sb),
3577                                                      1);
3578                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3579
3580                         /*
3581                          * we're only being called from kreiserfsd, it makes
3582                          * no sense to do an async commit so that kreiserfsd
3583                          * can do it later
3584                          */
3585                         do_journal_end(&th, COMMIT_NOW | WAIT);
3586                 }
3587         }
3588 }
3589
3590 /*
3591  * returns 0 if do_journal_end should return right away, returns 1 if
3592  * do_journal_end should finish the commit
3593  *
3594  * if the current transaction is too old, but still has writers, this will
3595  * wait on j_join_wait until all the writers are done.  By the time it
3596  * wakes up, the transaction it was called has already ended, so it just
3597  * flushes the commit list and returns 0.
3598  *
3599  * Won't batch when flush or commit_now is set.  Also won't batch when
3600  * others are waiting on j_join_wait.
3601  *
3602  * Note, we can't allow the journal_end to proceed while there are still
3603  * writers in the log.
3604  */
3605 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3606 {
3607
3608         time_t now;
3609         int flush = flags & FLUSH_ALL;
3610         int commit_now = flags & COMMIT_NOW;
3611         int wait_on_commit = flags & WAIT;
3612         struct reiserfs_journal_list *jl;
3613         struct super_block *sb = th->t_super;
3614         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3615
3616         BUG_ON(!th->t_trans_id);
3617
3618         if (th->t_trans_id != journal->j_trans_id) {
3619                 reiserfs_panic(th->t_super, "journal-1577",
3620                                "handle trans id %ld != current trans id %ld",
3621                                th->t_trans_id, journal->j_trans_id);
3622         }
3623
3624         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3625         /* <= 0 is allowed.  unmounting might not call begin */
3626         if (atomic_read(&journal->j_wcount) > 0)
3627                 atomic_dec(&journal->j_wcount);
3628
3629         /*
3630          * BUG, deal with case where j_len is 0, but people previously
3631          * freed blocks need to be released will be dealt with by next
3632          * transaction that actually writes something, but should be taken
3633          * care of in this trans
3634          */
3635         BUG_ON(journal->j_len == 0);
3636
3637         /*
3638          * if wcount > 0, and we are called to with flush or commit_now,
3639          * we wait on j_join_wait.  We will wake up when the last writer has
3640          * finished the transaction, and started it on its way to the disk.
3641          * Then, we flush the commit or journal list, and just return 0
3642          * because the rest of journal end was already done for this
3643          * transaction.
3644          */
3645         if (atomic_read(&journal->j_wcount) > 0) {
3646                 if (flush || commit_now) {
3647                         unsigned trans_id;
3648
3649                         jl = journal->j_current_jl;
3650                         trans_id = jl->j_trans_id;
3651                         if (wait_on_commit)
3652                                 jl->j_state |= LIST_COMMIT_PENDING;
3653                         atomic_set(&journal->j_jlock, 1);
3654                         if (flush) {
3655                                 journal->j_next_full_flush = 1;
3656                         }
3657                         unlock_journal(sb);
3658
3659                         /*
3660                          * sleep while the current transaction is
3661                          * still j_jlocked
3662                          */
3663                         while (journal->j_trans_id == trans_id) {
3664                                 if (atomic_read(&journal->j_jlock)) {
3665                                         queue_log_writer(sb);
3666                                 } else {
3667                                         lock_journal(sb);
3668                                         if (journal->j_trans_id == trans_id) {
3669                                                 atomic_set(&journal->j_jlock,
3670                                                            1);
3671                                         }
3672                                         unlock_journal(sb);
3673                                 }
3674                         }
3675                         BUG_ON(journal->j_trans_id == trans_id);
3676
3677                         if (commit_now
3678                             && journal_list_still_alive(sb, trans_id)
3679                             && wait_on_commit) {
3680                                 flush_commit_list(sb, jl, 1);
3681                         }
3682                         return 0;
3683                 }
3684                 unlock_journal(sb);
3685                 return 0;
3686         }
3687
3688         /* deal with old transactions where we are the last writers */
3689         now = get_seconds();
3690         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3691                 commit_now = 1;
3692                 journal->j_next_async_flush = 1;
3693         }
3694         /* don't batch when someone is waiting on j_join_wait */
3695         /* don't batch when syncing the commit or flushing the whole trans */
3696         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3697             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3698             && journal->j_len_alloc < journal->j_max_batch
3699             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3700                 journal->j_bcount++;
3701                 unlock_journal(sb);
3702                 return 0;
3703         }
3704
3705         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3706                 reiserfs_panic(sb, "journal-003",
3707                                "j_start (%ld) is too high",
3708                                journal->j_start);
3709         }
3710         return 1;
3711 }
3712
3713 /*
3714  * Does all the work that makes deleting blocks safe.
3715  * when deleting a block mark BH_JNew, just remove it from the current
3716  * transaction, clean it's buffer_head and move on.
3717  *
3718  * otherwise:
3719  * set a bit for the block in the journal bitmap.  That will prevent it from
3720  * being allocated for unformatted nodes before this transaction has finished.
3721  *
3722  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3723  * That will prevent any old transactions with this block from trying to flush
3724  * to the real location.  Since we aren't removing the cnode from the
3725  * journal_list_hash, *the block can't be reallocated yet.
3726  *
3727  * Then remove it from the current transaction, decrementing any counters and
3728  * filing it on the clean list.
3729  */
3730 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3731                        struct super_block *sb, b_blocknr_t blocknr)
3732 {
3733         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3734         struct reiserfs_journal_cnode *cn = NULL;
3735         struct buffer_head *bh = NULL;
3736         struct reiserfs_list_bitmap *jb = NULL;
3737         int cleaned = 0;
3738         BUG_ON(!th->t_trans_id);
3739
3740         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3741         if (cn && cn->bh) {
3742                 bh = cn->bh;
3743                 get_bh(bh);
3744         }
3745         /* if it is journal new, we just remove it from this transaction */
3746         if (bh && buffer_journal_new(bh)) {
3747                 clear_buffer_journal_new(bh);
3748                 clear_prepared_bits(bh);
3749                 reiserfs_clean_and_file_buffer(bh);
3750                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3751         } else {
3752                 /*
3753                  * set the bit for this block in the journal bitmap
3754                  * for this transaction
3755                  */
3756                 jb = journal->j_current_jl->j_list_bitmap;
3757                 if (!jb) {
3758                         reiserfs_panic(sb, "journal-1702",
3759                                        "journal_list_bitmap is NULL");
3760                 }
3761                 set_bit_in_list_bitmap(sb, blocknr, jb);
3762
3763                 /* Note, the entire while loop is not allowed to schedule.  */
3764
3765                 if (bh) {
3766                         clear_prepared_bits(bh);
3767                         reiserfs_clean_and_file_buffer(bh);
3768                 }
3769                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3770
3771                 /*
3772                  * find all older transactions with this block,
3773                  * make sure they don't try to write it out
3774                  */
3775                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3776                                           blocknr);
3777                 while (cn) {
3778                         if (sb == cn->sb && blocknr == cn->blocknr) {
3779                                 set_bit(BLOCK_FREED, &cn->state);
3780                                 if (cn->bh) {
3781                                         /*
3782                                          * remove_from_transaction will brelse
3783                                          * the buffer if it was in the current
3784                                          * trans
3785                                          */
3786                                         if (!cleaned) {
3787                                                 clear_buffer_journal_dirty(cn->
3788                                                                            bh);
3789                                                 clear_buffer_dirty(cn->bh);
3790                                                 clear_buffer_journal_test(cn->
3791                                                                           bh);
3792                                                 cleaned = 1;
3793                                                 put_bh(cn->bh);
3794                                                 if (atomic_read
3795                                                     (&cn->bh->b_count) < 0) {
3796                                                         reiserfs_warning(sb,
3797                                                                  "journal-2138",
3798                                                                  "cn->bh->b_count < 0");
3799                                                 }
3800                                         }
3801                                         /*
3802                                          * since we are clearing the bh,
3803                                          * we MUST dec nonzerolen
3804                                          */
3805                                         if (cn->jlist) {
3806                                                 atomic_dec(&cn->jlist->
3807                                                            j_nonzerolen);
3808                                         }
3809                                         cn->bh = NULL;
3810                                 }
3811                         }
3812                         cn = cn->hnext;
3813                 }
3814         }
3815
3816         if (bh)
3817                 release_buffer_page(bh); /* get_hash grabs the buffer */
3818         return 0;
3819 }
3820
3821 void reiserfs_update_inode_transaction(struct inode *inode)
3822 {
3823         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3824         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3825         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3826 }
3827
3828 /*
3829  * returns -1 on error, 0 if no commits/barriers were done and 1
3830  * if a transaction was actually committed and the barrier was done
3831  */
3832 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3833                              struct reiserfs_journal_list *jl)
3834 {
3835         struct reiserfs_transaction_handle th;
3836         struct super_block *sb = inode->i_sb;
3837         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3838         int ret = 0;
3839
3840         /*
3841          * is it from the current transaction,
3842          * or from an unknown transaction?
3843          */
3844         if (id == journal->j_trans_id) {
3845                 jl = journal->j_current_jl;
3846                 /*
3847                  * try to let other writers come in and
3848                  * grow this transaction
3849                  */
3850                 let_transaction_grow(sb, id);
3851                 if (journal->j_trans_id != id) {
3852                         goto flush_commit_only;
3853                 }
3854
3855                 ret = journal_begin(&th, sb, 1);
3856                 if (ret)
3857                         return ret;
3858
3859                 /* someone might have ended this transaction while we joined */
3860                 if (journal->j_trans_id != id) {
3861                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3862                                                      1);
3863                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3864                         ret = journal_end(&th);
3865                         goto flush_commit_only;
3866                 }
3867
3868                 ret = journal_end_sync(&th);
3869                 if (!ret)
3870                         ret = 1;
3871
3872         } else {
3873                 /*
3874                  * this gets tricky, we have to make sure the journal list in
3875                  * the inode still exists.  We know the list is still around
3876                  * if we've got a larger transaction id than the oldest list
3877                  */
3878 flush_commit_only:
3879                 if (journal_list_still_alive(inode->i_sb, id)) {
3880                         /*
3881                          * we only set ret to 1 when we know for sure
3882                          * the barrier hasn't been started yet on the commit
3883                          * block.
3884                          */
3885                         if (atomic_read(&jl->j_commit_left) > 1)
3886                                 ret = 1;
3887                         flush_commit_list(sb, jl, 1);
3888                         if (journal->j_errno)
3889                                 ret = journal->j_errno;
3890                 }
3891         }
3892         /* otherwise the list is gone, and long since committed */
3893         return ret;
3894 }
3895
3896 int reiserfs_commit_for_inode(struct inode *inode)
3897 {
3898         unsigned int id = REISERFS_I(inode)->i_trans_id;
3899         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3900
3901         /*
3902          * for the whole inode, assume unset id means it was
3903          * changed in the current transaction.  More conservative
3904          */
3905         if (!id || !jl) {
3906                 reiserfs_update_inode_transaction(inode);
3907                 id = REISERFS_I(inode)->i_trans_id;
3908                 /* jl will be updated in __commit_trans_jl */
3909         }
3910
3911         return __commit_trans_jl(inode, id, jl);
3912 }
3913
3914 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3915                                       struct buffer_head *bh)
3916 {
3917         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3918         PROC_INFO_INC(sb, journal.restore_prepared);
3919         if (!bh) {
3920                 return;
3921         }
3922         if (test_clear_buffer_journal_restore_dirty(bh) &&
3923             buffer_journal_dirty(bh)) {
3924                 struct reiserfs_journal_cnode *cn;
3925                 reiserfs_write_lock(sb);
3926                 cn = get_journal_hash_dev(sb,
3927                                           journal->j_list_hash_table,
3928                                           bh->b_blocknr);
3929                 if (cn && can_dirty(cn)) {
3930                         set_buffer_journal_test(bh);
3931                         mark_buffer_dirty(bh);
3932                 }
3933                 reiserfs_write_unlock(sb);
3934         }
3935         clear_buffer_journal_prepared(bh);
3936 }
3937
3938 extern struct tree_balance *cur_tb;
3939 /*
3940  * before we can change a metadata block, we have to make sure it won't
3941  * be written to disk while we are altering it.  So, we must:
3942  * clean it
3943  * wait on it.
3944  */
3945 int reiserfs_prepare_for_journal(struct super_block *sb,
3946                                  struct buffer_head *bh, int wait)
3947 {
3948         PROC_INFO_INC(sb, journal.prepare);
3949
3950         if (!trylock_buffer(bh)) {
3951                 if (!wait)
3952                         return 0;
3953                 lock_buffer(bh);
3954         }
3955         set_buffer_journal_prepared(bh);
3956         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3957                 clear_buffer_journal_test(bh);
3958                 set_buffer_journal_restore_dirty(bh);
3959         }
3960         unlock_buffer(bh);
3961         return 1;
3962 }
3963
3964 /*
3965  * long and ugly.  If flush, will not return until all commit
3966  * blocks and all real buffers in the trans are on disk.
3967  * If no_async, won't return until all commit blocks are on disk.
3968  *
3969  * keep reading, there are comments as you go along
3970  *
3971  * If the journal is aborted, we just clean up. Things like flushing
3972  * journal lists, etc just won't happen.
3973  */
3974 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3975 {
3976         struct super_block *sb = th->t_super;
3977         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3978         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3979         struct reiserfs_journal_cnode *last_cn = NULL;
3980         struct reiserfs_journal_desc *desc;
3981         struct reiserfs_journal_commit *commit;
3982         struct buffer_head *c_bh;       /* commit bh */
3983         struct buffer_head *d_bh;       /* desc bh */
3984         int cur_write_start = 0;        /* start index of current log write */
3985         int old_start;
3986         int i;
3987         int flush;
3988         int wait_on_commit;
3989         struct reiserfs_journal_list *jl, *temp_jl;
3990         struct list_head *entry, *safe;
3991         unsigned long jindex;
3992         unsigned int commit_trans_id;
3993         int trans_half;
3994         int depth;
3995
3996         BUG_ON(th->t_refcount > 1);
3997         BUG_ON(!th->t_trans_id);
3998         BUG_ON(!th->t_super);
3999
4000         /*
4001          * protect flush_older_commits from doing mistakes if the
4002          * transaction ID counter gets overflowed.
4003          */
4004         if (th->t_trans_id == ~0U)
4005                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4006         flush = flags & FLUSH_ALL;
4007         wait_on_commit = flags & WAIT;
4008
4009         current->journal_info = th->t_handle_save;
4010         reiserfs_check_lock_depth(sb, "journal end");
4011         if (journal->j_len == 0) {
4012                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4013                                              1);
4014                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4015         }
4016
4017         lock_journal(sb);
4018         if (journal->j_next_full_flush) {
4019                 flags |= FLUSH_ALL;
4020                 flush = 1;
4021         }
4022         if (journal->j_next_async_flush) {
4023                 flags |= COMMIT_NOW | WAIT;
4024                 wait_on_commit = 1;
4025         }
4026
4027         /*
4028          * check_journal_end locks the journal, and unlocks if it does
4029          * not return 1 it tells us if we should continue with the
4030          * journal_end, or just return
4031          */
4032         if (!check_journal_end(th, flags)) {
4033                 reiserfs_schedule_old_flush(sb);
4034                 wake_queued_writers(sb);
4035                 reiserfs_async_progress_wait(sb);
4036                 goto out;
4037         }
4038
4039         /* check_journal_end might set these, check again */
4040         if (journal->j_next_full_flush) {
4041                 flush = 1;
4042         }
4043
4044         /*
4045          * j must wait means we have to flush the log blocks, and the
4046          * real blocks for this transaction
4047          */
4048         if (journal->j_must_wait > 0) {
4049                 flush = 1;
4050         }
4051 #ifdef REISERFS_PREALLOCATE
4052         /*
4053          * quota ops might need to nest, setup the journal_info pointer
4054          * for them and raise the refcount so that it is > 0.
4055          */
4056         current->journal_info = th;
4057         th->t_refcount++;
4058
4059         /* it should not involve new blocks into the transaction */
4060         reiserfs_discard_all_prealloc(th);
4061
4062         th->t_refcount--;
4063         current->journal_info = th->t_handle_save;
4064 #endif
4065
4066         /* setup description block */
4067         d_bh =
4068             journal_getblk(sb,
4069                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4070                            journal->j_start);
4071         set_buffer_uptodate(d_bh);
4072         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4073         memset(d_bh->b_data, 0, d_bh->b_size);
4074         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4075         set_desc_trans_id(desc, journal->j_trans_id);
4076
4077         /*
4078          * setup commit block.  Don't write (keep it clean too) this one
4079          * until after everyone else is written
4080          */
4081         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4082                               ((journal->j_start + journal->j_len +
4083                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4084         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4085         memset(c_bh->b_data, 0, c_bh->b_size);
4086         set_commit_trans_id(commit, journal->j_trans_id);
4087         set_buffer_uptodate(c_bh);
4088
4089         /* init this journal list */
4090         jl = journal->j_current_jl;
4091
4092         /*
4093          * we lock the commit before doing anything because
4094          * we want to make sure nobody tries to run flush_commit_list until
4095          * the new transaction is fully setup, and we've already flushed the
4096          * ordered bh list
4097          */
4098         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4099
4100         /* save the transaction id in case we need to commit it later */
4101         commit_trans_id = jl->j_trans_id;
4102
4103         atomic_set(&jl->j_older_commits_done, 0);
4104         jl->j_trans_id = journal->j_trans_id;
4105         jl->j_timestamp = journal->j_trans_start_time;
4106         jl->j_commit_bh = c_bh;
4107         jl->j_start = journal->j_start;
4108         jl->j_len = journal->j_len;
4109         atomic_set(&jl->j_nonzerolen, journal->j_len);
4110         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4111         jl->j_realblock = NULL;
4112
4113         /*
4114          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4115          * for each real block, add it to the journal list hash,
4116          * copy into real block index array in the commit or desc block
4117          */
4118         trans_half = journal_trans_half(sb->s_blocksize);
4119         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4120                 if (buffer_journaled(cn->bh)) {
4121                         jl_cn = get_cnode(sb);
4122                         if (!jl_cn) {
4123                                 reiserfs_panic(sb, "journal-1676",
4124                                                "get_cnode returned NULL");
4125                         }
4126                         if (i == 0) {
4127                                 jl->j_realblock = jl_cn;
4128                         }
4129                         jl_cn->prev = last_cn;
4130                         jl_cn->next = NULL;
4131                         if (last_cn) {
4132                                 last_cn->next = jl_cn;
4133                         }
4134                         last_cn = jl_cn;
4135                         /*
4136                          * make sure the block we are trying to log
4137                          * is not a block of journal or reserved area
4138                          */
4139                         if (is_block_in_log_or_reserved_area
4140                             (sb, cn->bh->b_blocknr)) {
4141                                 reiserfs_panic(sb, "journal-2332",
4142                                                "Trying to log block %lu, "
4143                                                "which is a log block",
4144                                                cn->bh->b_blocknr);
4145                         }
4146                         jl_cn->blocknr = cn->bh->b_blocknr;
4147                         jl_cn->state = 0;
4148                         jl_cn->sb = sb;
4149                         jl_cn->bh = cn->bh;
4150                         jl_cn->jlist = jl;
4151                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4152                         if (i < trans_half) {
4153                                 desc->j_realblock[i] =
4154                                     cpu_to_le32(cn->bh->b_blocknr);
4155                         } else {
4156                                 commit->j_realblock[i - trans_half] =
4157                                     cpu_to_le32(cn->bh->b_blocknr);
4158                         }
4159                 } else {
4160                         i--;
4161                 }
4162         }
4163         set_desc_trans_len(desc, journal->j_len);
4164         set_desc_mount_id(desc, journal->j_mount_id);
4165         set_desc_trans_id(desc, journal->j_trans_id);
4166         set_commit_trans_len(commit, journal->j_len);
4167
4168         /*
4169          * special check in case all buffers in the journal
4170          * were marked for not logging
4171          */
4172         BUG_ON(journal->j_len == 0);
4173
4174         /*
4175          * we're about to dirty all the log blocks, mark the description block
4176          * dirty now too.  Don't mark the commit block dirty until all the
4177          * others are on disk
4178          */
4179         mark_buffer_dirty(d_bh);
4180
4181         /*
4182          * first data block is j_start + 1, so add one to
4183          * cur_write_start wherever you use it
4184          */
4185         cur_write_start = journal->j_start;
4186         cn = journal->j_first;
4187         jindex = 1;     /* start at one so we don't get the desc again */
4188         while (cn) {
4189                 clear_buffer_journal_new(cn->bh);
4190                 /* copy all the real blocks into log area.  dirty log blocks */
4191                 if (buffer_journaled(cn->bh)) {
4192                         struct buffer_head *tmp_bh;
4193                         char *addr;
4194                         struct page *page;
4195                         tmp_bh =
4196                             journal_getblk(sb,
4197                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4198                                            ((cur_write_start +
4199                                              jindex) %
4200                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4201                         set_buffer_uptodate(tmp_bh);
4202                         page = cn->bh->b_page;
4203                         addr = kmap(page);
4204                         memcpy(tmp_bh->b_data,
4205                                addr + offset_in_page(cn->bh->b_data),
4206                                cn->bh->b_size);
4207                         kunmap(page);
4208                         mark_buffer_dirty(tmp_bh);
4209                         jindex++;
4210                         set_buffer_journal_dirty(cn->bh);
4211                         clear_buffer_journaled(cn->bh);
4212                 } else {
4213                         /*
4214                          * JDirty cleared sometime during transaction.
4215                          * don't log this one
4216                          */
4217                         reiserfs_warning(sb, "journal-2048",
4218                                          "BAD, buffer in journal hash, "
4219                                          "but not JDirty!");
4220                         brelse(cn->bh);
4221                 }
4222                 next = cn->next;
4223                 free_cnode(sb, cn);
4224                 cn = next;
4225                 reiserfs_cond_resched(sb);
4226         }
4227
4228         /*
4229          * we are done with both the c_bh and d_bh, but
4230          * c_bh must be written after all other commit blocks,
4231          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4232          */
4233
4234         journal->j_current_jl = alloc_journal_list(sb);
4235
4236         /* now it is safe to insert this transaction on the main list */
4237         list_add_tail(&jl->j_list, &journal->j_journal_list);
4238         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4239         journal->j_num_work_lists++;
4240
4241         /* reset journal values for the next transaction */
4242         old_start = journal->j_start;
4243         journal->j_start =
4244             (journal->j_start + journal->j_len +
4245              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4246         atomic_set(&journal->j_wcount, 0);
4247         journal->j_bcount = 0;
4248         journal->j_last = NULL;
4249         journal->j_first = NULL;
4250         journal->j_len = 0;
4251         journal->j_trans_start_time = 0;
4252         /* check for trans_id overflow */
4253         if (++journal->j_trans_id == 0)
4254                 journal->j_trans_id = 10;
4255         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4256         journal->j_must_wait = 0;
4257         journal->j_len_alloc = 0;
4258         journal->j_next_full_flush = 0;
4259         journal->j_next_async_flush = 0;
4260         init_journal_hash(sb);
4261
4262         /*
4263          * make sure reiserfs_add_jh sees the new current_jl before we
4264          * write out the tails
4265          */
4266         smp_mb();
4267
4268         /*
4269          * tail conversion targets have to hit the disk before we end the
4270          * transaction.  Otherwise a later transaction might repack the tail
4271          * before this transaction commits, leaving the data block unflushed
4272          * and clean, if we crash before the later transaction commits, the
4273          * data block is lost.
4274          */
4275         if (!list_empty(&jl->j_tail_bh_list)) {
4276                 depth = reiserfs_write_unlock_nested(sb);
4277                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4278                                       journal, jl, &jl->j_tail_bh_list);
4279                 reiserfs_write_lock_nested(sb, depth);
4280         }
4281         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4282         mutex_unlock(&jl->j_commit_mutex);
4283
4284         /*
4285          * honor the flush wishes from the caller, simple commits can
4286          * be done outside the journal lock, they are done below
4287          *
4288          * if we don't flush the commit list right now, we put it into
4289          * the work queue so the people waiting on the async progress work
4290          * queue don't wait for this proc to flush journal lists and such.
4291          */
4292         if (flush) {
4293                 flush_commit_list(sb, jl, 1);
4294                 flush_journal_list(sb, jl, 1);
4295         } else if (!(jl->j_state & LIST_COMMIT_PENDING))
4296                 queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4297                                    &journal->j_work, HZ / 10);
4298
4299         /*
4300          * if the next transaction has any chance of wrapping, flush
4301          * transactions that might get overwritten.  If any journal lists
4302          * are very old flush them as well.
4303          */
4304 first_jl:
4305         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4306                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4307                 if (journal->j_start <= temp_jl->j_start) {
4308                         if ((journal->j_start + journal->j_trans_max + 1) >=
4309                             temp_jl->j_start) {
4310                                 flush_used_journal_lists(sb, temp_jl);
4311                                 goto first_jl;
4312                         } else if ((journal->j_start +
4313                                     journal->j_trans_max + 1) <
4314                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4315                                 /*
4316                                  * if we don't cross into the next
4317                                  * transaction and we don't wrap, there is
4318                                  * no way we can overlap any later transactions
4319                                  * break now
4320                                  */
4321                                 break;
4322                         }
4323                 } else if ((journal->j_start +
4324                             journal->j_trans_max + 1) >
4325                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4326                         if (((journal->j_start + journal->j_trans_max + 1) %
4327                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4328                             temp_jl->j_start) {
4329                                 flush_used_journal_lists(sb, temp_jl);
4330                                 goto first_jl;
4331                         } else {
4332                                 /*
4333                                 * we don't overlap anything from out start
4334                                 * to the end of the log, and our wrapped
4335                                 * portion doesn't overlap anything at
4336                                 * the start of the log.  We can break
4337                                 */
4338                                 break;
4339                         }
4340                 }
4341         }
4342
4343         journal->j_current_jl->j_list_bitmap =
4344             get_list_bitmap(sb, journal->j_current_jl);
4345
4346         if (!(journal->j_current_jl->j_list_bitmap)) {
4347                 reiserfs_panic(sb, "journal-1996",
4348                                "could not get a list bitmap");
4349         }
4350
4351         atomic_set(&journal->j_jlock, 0);
4352         unlock_journal(sb);
4353         /* wake up any body waiting to join. */
4354         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4355         wake_up(&journal->j_join_wait);
4356
4357         if (!flush && wait_on_commit &&
4358             journal_list_still_alive(sb, commit_trans_id)) {
4359                 flush_commit_list(sb, jl, 1);
4360         }
4361 out:
4362         reiserfs_check_lock_depth(sb, "journal end2");
4363
4364         memset(th, 0, sizeof(*th));
4365         /*
4366          * Re-set th->t_super, so we can properly keep track of how many
4367          * persistent transactions there are. We need to do this so if this
4368          * call is part of a failed restart_transaction, we can free it later
4369          */
4370         th->t_super = sb;
4371
4372         return journal->j_errno;
4373 }
4374
4375 /* Send the file system read only and refuse new transactions */
4376 void reiserfs_abort_journal(struct super_block *sb, int errno)
4377 {
4378         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4379         if (test_bit(J_ABORTED, &journal->j_state))
4380                 return;
4381
4382         if (!journal->j_errno)
4383                 journal->j_errno = errno;
4384
4385         sb->s_flags |= MS_RDONLY;
4386         set_bit(J_ABORTED, &journal->j_state);
4387
4388 #ifdef CONFIG_REISERFS_CHECK
4389         dump_stack();
4390 #endif
4391 }