fs/reiserfs/journal.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Write ahead logging implementation copyright Chris Mason 2000
   4  *
   5  * The background commits make this code very interrelated, and
   6  * overly complex.  I need to rethink things a bit....The major players:
   7  *
   8  * journal_begin -- call with the number of blocks you expect to log.
   9  *                  If the current transaction is too
  10  *                  old, it will block until the current transaction is
  11  *                  finished, and then start a new one.
  12  *                  Usually, your transaction will get joined in with
  13  *                  previous ones for speed.
  14  *
  15  * journal_join  -- same as journal_begin, but won't block on the current
  16  *                  transaction regardless of age.  Don't ever call
  17  *                  this.  Ever.  There are only two places it should be
  18  *                  called from, and they are both inside this file.
  19  *
  20  * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  21  *                       that might make them get sent to disk
  22  *                       and then marks them BH_JDirty.  Puts the buffer head
  23  *                       into the current transaction hash.
  24  *
  25  * journal_end -- if the current transaction is batchable, it does nothing
  26  *                   otherwise, it could do an async/synchronous commit, or
  27  *                   a full flush of all log and real blocks in the
  28  *                   transaction.
  29  *
  30  * flush_old_commits -- if the current transaction is too old, it is ended and
  31  *                      commit blocks are sent to disk.  Forces commit blocks
  32  *                      to disk for all backgrounded commits that have been
  33  *                      around too long.
  34  *                   -- Note, if you call this as an immediate flush from
  35  *                      from within kupdate, it will ignore the immediate flag
  36  */
  37
  38 #include <linux/time.h>
  39 #include <linux/semaphore.h>
  40 #include <linux/vmalloc.h>
  41 #include "reiserfs.h"
  42 #include <linux/kernel.h>
  43 #include <linux/errno.h>
  44 #include <linux/fcntl.h>
  45 #include <linux/stat.h>
  46 #include <linux/string.h>
  47 #include <linux/buffer_head.h>
  48 #include <linux/workqueue.h>
  49 #include <linux/writeback.h>
  50 #include <linux/blkdev.h>
  51 #include <linux/backing-dev.h>
  52 #include <linux/uaccess.h>
  53 #include <linux/slab.h>
  54
  55
  56 /* gets a struct reiserfs_journal_list * from a list head */
  57 #define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                                j_list))
  59 #define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                                j_working_list))
  61
  62 /* must be correct to keep the desc and commit structs at 4k */
  63 #define JOURNAL_TRANS_HALF 1018
  64 #define BUFNR 64                /*read ahead */
  65
  66 /* cnode stat bits.  Move these into reiserfs_fs.h */
  67
  68 /* this block was freed, and can't be written.  */
  69 #define BLOCK_FREED 2
  70 /* this block was freed during this transaction, and can't be written */
  71 #define BLOCK_FREED_HOLDER 3
  72
  73 /* used in flush_journal_list */
  74 #define BLOCK_NEEDS_FLUSH 4
  75 #define BLOCK_DIRTIED 5
  76
  77 /* journal list state bits */
  78 #define LIST_TOUCHED 1
  79 #define LIST_DIRTY   2
  80 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  81
  82 /* flags for do_journal_end */
  83 #define FLUSH_ALL   1           /* flush commit and real blocks */
  84 #define COMMIT_NOW  2           /* end and commit this transaction */
  85 #define WAIT        4           /* wait for the log blocks to hit the disk */
  86
  87 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  88 static int flush_journal_list(struct super_block *s,
  89                               struct reiserfs_journal_list *jl, int flushall);
  90 static int flush_commit_list(struct super_block *s,
  91                              struct reiserfs_journal_list *jl, int flushall);
  92 static int can_dirty(struct reiserfs_journal_cnode *cn);
  93 static int journal_join(struct reiserfs_transaction_handle *th,
  94                         struct super_block *sb);
  95 static void release_journal_dev(struct super_block *super,
  96                                struct reiserfs_journal *journal);
  97 static void dirty_one_transaction(struct super_block *s,
  98                                  struct reiserfs_journal_list *jl);
  99 static void flush_async_commits(struct work_struct *work);
 100 static void queue_log_writer(struct super_block *s);
 101
 102 /* values for join in do_journal_begin_r */
 103 enum {
 104         JBEGIN_REG = 0,         /* regular journal begin */
 105         /* join the running transaction if at all possible */
 106         JBEGIN_JOIN = 1,
 107         /* called from cleanup code, ignores aborted flag */
 108         JBEGIN_ABORT = 2,
 109 };
 110
 111 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 112                               struct super_block *sb,
 113                               unsigned long nblocks, int join);
 114
 115 static void init_journal_hash(struct super_block *sb)
 116 {
 117         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 118         memset(journal->j_hash_table, 0,
 119                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 120 }
 121
 122 /*
 123  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 124  * I can't allow refile_buffer to make schedule happen after I've freed a
 125  * block.  Look at remove_from_transaction and journal_mark_freed for
 126  * more details.
 127  */
 128 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 129 {
 130         if (bh) {
 131                 clear_buffer_dirty(bh);
 132                 clear_buffer_journal_test(bh);
 133         }
 134         return 0;
 135 }
 136
 137 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 138                                                          *sb)
 139 {
 140         struct reiserfs_bitmap_node *bn;
 141         static int id;
 142
 143         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 144         if (!bn) {
 145                 return NULL;
 146         }
 147         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 148         if (!bn->data) {
 149                 kfree(bn);
 150                 return NULL;
 151         }
 152         bn->id = id++;
 153         INIT_LIST_HEAD(&bn->list);
 154         return bn;
 155 }
 156
 157 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 158 {
 159         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 160         struct reiserfs_bitmap_node *bn = NULL;
 161         struct list_head *entry = journal->j_bitmap_nodes.next;
 162
 163         journal->j_used_bitmap_nodes++;
 164 repeat:
 165
 166         if (entry != &journal->j_bitmap_nodes) {
 167                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 168                 list_del(entry);
 169                 memset(bn->data, 0, sb->s_blocksize);
 170                 journal->j_free_bitmap_nodes--;
 171                 return bn;
 172         }
 173         bn = allocate_bitmap_node(sb);
 174         if (!bn) {
 175                 yield();
 176                 goto repeat;
 177         }
 178         return bn;
 179 }
 180 static inline void free_bitmap_node(struct super_block *sb,
 181                                     struct reiserfs_bitmap_node *bn)
 182 {
 183         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 184         journal->j_used_bitmap_nodes--;
 185         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 186                 kfree(bn->data);
 187                 kfree(bn);
 188         } else {
 189                 list_add(&bn->list, &journal->j_bitmap_nodes);
 190                 journal->j_free_bitmap_nodes++;
 191         }
 192 }
 193
 194 static void allocate_bitmap_nodes(struct super_block *sb)
 195 {
 196         int i;
 197         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 198         struct reiserfs_bitmap_node *bn = NULL;
 199         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 200                 bn = allocate_bitmap_node(sb);
 201                 if (bn) {
 202                         list_add(&bn->list, &journal->j_bitmap_nodes);
 203                         journal->j_free_bitmap_nodes++;
 204                 } else {
 205                         /* this is ok, we'll try again when more are needed */
 206                         break;
 207                 }
 208         }
 209 }
 210
 211 static int set_bit_in_list_bitmap(struct super_block *sb,
 212                                   b_blocknr_t block,
 213                                   struct reiserfs_list_bitmap *jb)
 214 {
 215         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 216         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 217
 218         if (!jb->bitmaps[bmap_nr]) {
 219                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 220         }
 221         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 222         return 0;
 223 }
 224
 225 static void cleanup_bitmap_list(struct super_block *sb,
 226                                 struct reiserfs_list_bitmap *jb)
 227 {
 228         int i;
 229         if (jb->bitmaps == NULL)
 230                 return;
 231
 232         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 233                 if (jb->bitmaps[i]) {
 234                         free_bitmap_node(sb, jb->bitmaps[i]);
 235                         jb->bitmaps[i] = NULL;
 236                 }
 237         }
 238 }
 239
 240 /*
 241  * only call this on FS unmount.
 242  */
 243 static int free_list_bitmaps(struct super_block *sb,
 244                              struct reiserfs_list_bitmap *jb_array)
 245 {
 246         int i;
 247         struct reiserfs_list_bitmap *jb;
 248         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 249                 jb = jb_array + i;
 250                 jb->journal_list = NULL;
 251                 cleanup_bitmap_list(sb, jb);
 252                 vfree(jb->bitmaps);
 253                 jb->bitmaps = NULL;
 254         }
 255         return 0;
 256 }
 257
 258 static int free_bitmap_nodes(struct super_block *sb)
 259 {
 260         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 261         struct list_head *next = journal->j_bitmap_nodes.next;
 262         struct reiserfs_bitmap_node *bn;
 263
 264         while (next != &journal->j_bitmap_nodes) {
 265                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 266                 list_del(next);
 267                 kfree(bn->data);
 268                 kfree(bn);
 269                 next = journal->j_bitmap_nodes.next;
 270                 journal->j_free_bitmap_nodes--;
 271         }
 272
 273         return 0;
 274 }
 275
 276 /*
 277  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 278  * jb_array is the array to be filled in.
 279  */
 280 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 281                                    struct reiserfs_list_bitmap *jb_array,
 282                                    unsigned int bmap_nr)
 283 {
 284         int i;
 285         int failed = 0;
 286         struct reiserfs_list_bitmap *jb;
 287         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 288
 289         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 290                 jb = jb_array + i;
 291                 jb->journal_list = NULL;
 292                 jb->bitmaps = vzalloc(mem);
 293                 if (!jb->bitmaps) {
 294                         reiserfs_warning(sb, "clm-2000", "unable to "
 295                                          "allocate bitmaps for journal lists");
 296                         failed = 1;
 297                         break;
 298                 }
 299         }
 300         if (failed) {
 301                 free_list_bitmaps(sb, jb_array);
 302                 return -1;
 303         }
 304         return 0;
 305 }
 306
 307 /*
 308  * find an available list bitmap.  If you can't find one, flush a commit list
 309  * and try again
 310  */
 311 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 312                                                     struct reiserfs_journal_list
 313                                                     *jl)
 314 {
 315         int i, j;
 316         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 317         struct reiserfs_list_bitmap *jb = NULL;
 318
 319         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 320                 i = journal->j_list_bitmap_index;
 321                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 322                 jb = journal->j_list_bitmap + i;
 323                 if (journal->j_list_bitmap[i].journal_list) {
 324                         flush_commit_list(sb,
 325                                           journal->j_list_bitmap[i].
 326                                           journal_list, 1);
 327                         if (!journal->j_list_bitmap[i].journal_list) {
 328                                 break;
 329                         }
 330                 } else {
 331                         break;
 332                 }
 333         }
 334         /* double check to make sure if flushed correctly */
 335         if (jb->journal_list)
 336                 return NULL;
 337         jb->journal_list = jl;
 338         return jb;
 339 }
 340
 341 /*
 342  * allocates a new chunk of X nodes, and links them all together as a list.
 343  * Uses the cnode->next and cnode->prev pointers
 344  * returns NULL on failure
 345  */
 346 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 347 {
 348         struct reiserfs_journal_cnode *head;
 349         int i;
 350         if (num_cnodes <= 0) {
 351                 return NULL;
 352         }
 353         head = vzalloc(array_size(num_cnodes,
 354                                   sizeof(struct reiserfs_journal_cnode)));
 355         if (!head) {
 356                 return NULL;
 357         }
 358         head[0].prev = NULL;
 359         head[0].next = head + 1;
 360         for (i = 1; i < num_cnodes; i++) {
 361                 head[i].prev = head + (i - 1);
 362                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 363         }
 364         head[num_cnodes - 1].next = NULL;
 365         return head;
 366 }
 367
 368 /* pulls a cnode off the free list, or returns NULL on failure */
 369 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 370 {
 371         struct reiserfs_journal_cnode *cn;
 372         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 373
 374         reiserfs_check_lock_depth(sb, "get_cnode");
 375
 376         if (journal->j_cnode_free <= 0) {
 377                 return NULL;
 378         }
 379         journal->j_cnode_used++;
 380         journal->j_cnode_free--;
 381         cn = journal->j_cnode_free_list;
 382         if (!cn) {
 383                 return cn;
 384         }
 385         if (cn->next) {
 386                 cn->next->prev = NULL;
 387         }
 388         journal->j_cnode_free_list = cn->next;
 389         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 390         return cn;
 391 }
 392
 393 /*
 394  * returns a cnode to the free list
 395  */
 396 static void free_cnode(struct super_block *sb,
 397                        struct reiserfs_journal_cnode *cn)
 398 {
 399         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 400
 401         reiserfs_check_lock_depth(sb, "free_cnode");
 402
 403         journal->j_cnode_used--;
 404         journal->j_cnode_free++;
 405         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 406         cn->next = journal->j_cnode_free_list;
 407         if (journal->j_cnode_free_list) {
 408                 journal->j_cnode_free_list->prev = cn;
 409         }
 410         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 411         journal->j_cnode_free_list = cn;
 412 }
 413
 414 static void clear_prepared_bits(struct buffer_head *bh)
 415 {
 416         clear_buffer_journal_prepared(bh);
 417         clear_buffer_journal_restore_dirty(bh);
 418 }
 419
 420 /*
 421  * return a cnode with same dev, block number and size in table,
 422  * or null if not found
 423  */
 424 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 425                                                                   super_block
 426                                                                   *sb,
 427                                                                   struct
 428                                                                   reiserfs_journal_cnode
 429                                                                   **table,
 430                                                                   long bl)
 431 {
 432         struct reiserfs_journal_cnode *cn;
 433         cn = journal_hash(table, sb, bl);
 434         while (cn) {
 435                 if (cn->blocknr == bl && cn->sb == sb)
 436                         return cn;
 437                 cn = cn->hnext;
 438         }
 439         return (struct reiserfs_journal_cnode *)0;
 440 }
 441
 442 /*
 443  * this actually means 'can this block be reallocated yet?'.  If you set
 444  * search_all, a block can only be allocated if it is not in the current
 445  * transaction, was not freed by the current transaction, and has no chance
 446  * of ever being overwritten by a replay after crashing.
 447  *
 448  * If you don't set search_all, a block can only be allocated if it is not
 449  * in the current transaction.  Since deleting a block removes it from the
 450  * current transaction, this case should never happen.  If you don't set
 451  * search_all, make sure you never write the block without logging it.
 452  *
 453  * next_zero_bit is a suggestion about the next block to try for find_forward.
 454  * when bl is rejected because it is set in a journal list bitmap, we search
 455  * for the next zero bit in the bitmap that rejected bl.  Then, we return
 456  * that through next_zero_bit for find_forward to try.
 457  *
 458  * Just because we return something in next_zero_bit does not mean we won't
 459  * reject it on the next call to reiserfs_in_journal
 460  */
 461 int reiserfs_in_journal(struct super_block *sb,
 462                         unsigned int bmap_nr, int bit_nr, int search_all,
 463                         b_blocknr_t * next_zero_bit)
 464 {
 465         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 466         struct reiserfs_journal_cnode *cn;
 467         struct reiserfs_list_bitmap *jb;
 468         int i;
 469         unsigned long bl;
 470
 471         *next_zero_bit = 0;     /* always start this at zero. */
 472
 473         PROC_INFO_INC(sb, journal.in_journal);
 474         /*
 475          * If we aren't doing a search_all, this is a metablock, and it
 476          * will be logged before use.  if we crash before the transaction
 477          * that freed it commits,  this transaction won't have committed
 478          * either, and the block will never be written
 479          */
 480         if (search_all) {
 481                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 482                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 483                         jb = journal->j_list_bitmap + i;
 484                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 485                             test_bit(bit_nr,
 486                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 487                                      data)) {
 488                                 *next_zero_bit =
 489                                     find_next_zero_bit((unsigned long *)
 490                                                        (jb->bitmaps[bmap_nr]->
 491                                                         data),
 492                                                        sb->s_blocksize << 3,
 493                                                        bit_nr + 1);
 494                                 return 1;
 495                         }
 496                 }
 497         }
 498
 499         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 500         /* is it in any old transactions? */
 501         if (search_all
 502             && (cn =
 503                 get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 504                 return 1;
 505         }
 506
 507         /* is it in the current transaction.  This should never happen */
 508         if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 509                 BUG();
 510                 return 1;
 511         }
 512
 513         PROC_INFO_INC(sb, journal.in_journal_reusable);
 514         /* safe for reuse */
 515         return 0;
 516 }
 517
 518 /* insert cn into table */
 519 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 520                                        struct reiserfs_journal_cnode *cn)
 521 {
 522         struct reiserfs_journal_cnode *cn_orig;
 523
 524         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 525         cn->hnext = cn_orig;
 526         cn->hprev = NULL;
 527         if (cn_orig) {
 528                 cn_orig->hprev = cn;
 529         }
 530         journal_hash(table, cn->sb, cn->blocknr) = cn;
 531 }
 532
 533 /* lock the current transaction */
 534 static inline void lock_journal(struct super_block *sb)
 535 {
 536         PROC_INFO_INC(sb, journal.lock_journal);
 537
 538         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 539 }
 540
 541 /* unlock the current transaction */
 542 static inline void unlock_journal(struct super_block *sb)
 543 {
 544         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 545 }
 546
 547 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 548 {
 549         jl->j_refcount++;
 550 }
 551
 552 static inline void put_journal_list(struct super_block *s,
 553                                     struct reiserfs_journal_list *jl)
 554 {
 555         if (jl->j_refcount < 1) {
 556                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 557                                jl->j_trans_id, jl->j_refcount);
 558         }
 559         if (--jl->j_refcount == 0)
 560                 kfree(jl);
 561 }
 562
 563 /*
 564  * this used to be much more involved, and I'm keeping it just in case
 565  * things get ugly again.  it gets called by flush_commit_list, and
 566  * cleans up any data stored about blocks freed during a transaction.
 567  */
 568 static void cleanup_freed_for_journal_list(struct super_block *sb,
 569                                            struct reiserfs_journal_list *jl)
 570 {
 571
 572         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 573         if (jb) {
 574                 cleanup_bitmap_list(sb, jb);
 575         }
 576         jl->j_list_bitmap->journal_list = NULL;
 577         jl->j_list_bitmap = NULL;
 578 }
 579
 580 static int journal_list_still_alive(struct super_block *s,
 581                                     unsigned int trans_id)
 582 {
 583         struct reiserfs_journal *journal = SB_JOURNAL(s);
 584         struct list_head *entry = &journal->j_journal_list;
 585         struct reiserfs_journal_list *jl;
 586
 587         if (!list_empty(entry)) {
 588                 jl = JOURNAL_LIST_ENTRY(entry->next);
 589                 if (jl->j_trans_id <= trans_id) {
 590                         return 1;
 591                 }
 592         }
 593         return 0;
 594 }
 595
 596 /*
 597  * If page->mapping was null, we failed to truncate this page for
 598  * some reason.  Most likely because it was truncated after being
 599  * logged via data=journal.
 600  *
 601  * This does a check to see if the buffer belongs to one of these
 602  * lost pages before doing the final put_bh.  If page->mapping was
 603  * null, it tries to free buffers on the page, which should make the
 604  * final put_page drop the page from the lru.
 605  */
 606 static void release_buffer_page(struct buffer_head *bh)
 607 {
 608         struct page *page = bh->b_page;
 609         if (!page->mapping && trylock_page(page)) {
 610                 get_page(page);
 611                 put_bh(bh);
 612                 if (!page->mapping)
 613                         try_to_free_buffers(page);
 614                 unlock_page(page);
 615                 put_page(page);
 616         } else {
 617                 put_bh(bh);
 618         }
 619 }
 620
 621 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 622 {
 623         if (buffer_journaled(bh)) {
 624                 reiserfs_warning(NULL, "clm-2084",
 625                                  "pinned buffer %lu:%pg sent to disk",
 626                                  bh->b_blocknr, bh->b_bdev);
 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(REQ_OP_WRITE, 0, 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(REQ_OP_WRITE, 0, bh);
 668 }
 669
 670 #define CHUNK_SIZE 32
 671 struct buffer_chunk {
 672         struct buffer_head *bh[CHUNK_SIZE];
 673         int nr;
 674 };
 675
 676 static void write_chunk(struct buffer_chunk *chunk)
 677 {
 678         int i;
 679         for (i = 0; i < chunk->nr; i++) {
 680                 submit_logged_buffer(chunk->bh[i]);
 681         }
 682         chunk->nr = 0;
 683 }
 684
 685 static void write_ordered_chunk(struct buffer_chunk *chunk)
 686 {
 687         int i;
 688         for (i = 0; i < chunk->nr; i++) {
 689                 submit_ordered_buffer(chunk->bh[i]);
 690         }
 691         chunk->nr = 0;
 692 }
 693
 694 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 695                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 696 {
 697         int ret = 0;
 698         BUG_ON(chunk->nr >= CHUNK_SIZE);
 699         chunk->bh[chunk->nr++] = bh;
 700         if (chunk->nr >= CHUNK_SIZE) {
 701                 ret = 1;
 702                 if (lock) {
 703                         spin_unlock(lock);
 704                         fn(chunk);
 705                         spin_lock(lock);
 706                 } else {
 707                         fn(chunk);
 708                 }
 709         }
 710         return ret;
 711 }
 712
 713 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 714 static struct reiserfs_jh *alloc_jh(void)
 715 {
 716         struct reiserfs_jh *jh;
 717         while (1) {
 718                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 719                 if (jh) {
 720                         atomic_inc(&nr_reiserfs_jh);
 721                         return jh;
 722                 }
 723                 yield();
 724         }
 725 }
 726
 727 /*
 728  * we want to free the jh when the buffer has been written
 729  * and waited on
 730  */
 731 void reiserfs_free_jh(struct buffer_head *bh)
 732 {
 733         struct reiserfs_jh *jh;
 734
 735         jh = bh->b_private;
 736         if (jh) {
 737                 bh->b_private = NULL;
 738                 jh->bh = NULL;
 739                 list_del_init(&jh->list);
 740                 kfree(jh);
 741                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 742                         BUG();
 743                 atomic_dec(&nr_reiserfs_jh);
 744                 put_bh(bh);
 745         }
 746 }
 747
 748 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 749                            int tail)
 750 {
 751         struct reiserfs_jh *jh;
 752
 753         if (bh->b_private) {
 754                 spin_lock(&j->j_dirty_buffers_lock);
 755                 if (!bh->b_private) {
 756                         spin_unlock(&j->j_dirty_buffers_lock);
 757                         goto no_jh;
 758                 }
 759                 jh = bh->b_private;
 760                 list_del_init(&jh->list);
 761         } else {
 762 no_jh:
 763                 get_bh(bh);
 764                 jh = alloc_jh();
 765                 spin_lock(&j->j_dirty_buffers_lock);
 766                 /*
 767                  * buffer must be locked for __add_jh, should be able to have
 768                  * two adds at the same time
 769                  */
 770                 BUG_ON(bh->b_private);
 771                 jh->bh = bh;
 772                 bh->b_private = jh;
 773         }
 774         jh->jl = j->j_current_jl;
 775         if (tail)
 776                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 777         else {
 778                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 779         }
 780         spin_unlock(&j->j_dirty_buffers_lock);
 781         return 0;
 782 }
 783
 784 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 785 {
 786         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 787 }
 788 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 789 {
 790         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 791 }
 792
 793 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 794 static int write_ordered_buffers(spinlock_t * lock,
 795                                  struct reiserfs_journal *j,
 796                                  struct reiserfs_journal_list *jl,
 797                                  struct list_head *list)
 798 {
 799         struct buffer_head *bh;
 800         struct reiserfs_jh *jh;
 801         int ret = j->j_errno;
 802         struct buffer_chunk chunk;
 803         struct list_head tmp;
 804         INIT_LIST_HEAD(&tmp);
 805
 806         chunk.nr = 0;
 807         spin_lock(lock);
 808         while (!list_empty(list)) {
 809                 jh = JH_ENTRY(list->next);
 810                 bh = jh->bh;
 811                 get_bh(bh);
 812                 if (!trylock_buffer(bh)) {
 813                         if (!buffer_dirty(bh)) {
 814                                 list_move(&jh->list, &tmp);
 815                                 goto loop_next;
 816                         }
 817                         spin_unlock(lock);
 818                         if (chunk.nr)
 819                                 write_ordered_chunk(&chunk);
 820                         wait_on_buffer(bh);
 821                         cond_resched();
 822                         spin_lock(lock);
 823                         goto loop_next;
 824                 }
 825                 /*
 826                  * in theory, dirty non-uptodate buffers should never get here,
 827                  * but the upper layer io error paths still have a few quirks.
 828                  * Handle them here as gracefully as we can
 829                  */
 830                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 831                         clear_buffer_dirty(bh);
 832                         ret = -EIO;
 833                 }
 834                 if (buffer_dirty(bh)) {
 835                         list_move(&jh->list, &tmp);
 836                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 837                 } else {
 838                         reiserfs_free_jh(bh);
 839                         unlock_buffer(bh);
 840                 }
 841 loop_next:
 842                 put_bh(bh);
 843                 cond_resched_lock(lock);
 844         }
 845         if (chunk.nr) {
 846                 spin_unlock(lock);
 847                 write_ordered_chunk(&chunk);
 848                 spin_lock(lock);
 849         }
 850         while (!list_empty(&tmp)) {
 851                 jh = JH_ENTRY(tmp.prev);
 852                 bh = jh->bh;
 853                 get_bh(bh);
 854                 reiserfs_free_jh(bh);
 855
 856                 if (buffer_locked(bh)) {
 857                         spin_unlock(lock);
 858                         wait_on_buffer(bh);
 859                         spin_lock(lock);
 860                 }
 861                 if (!buffer_uptodate(bh)) {
 862                         ret = -EIO;
 863                 }
 864                 /*
 865                  * ugly interaction with invalidatepage here.
 866                  * reiserfs_invalidate_page will pin any buffer that has a
 867                  * valid journal head from an older transaction.  If someone
 868                  * else sets our buffer dirty after we write it in the first
 869                  * loop, and then someone truncates the page away, nobody
 870                  * will ever write the buffer. We're safe if we write the
 871                  * page one last time after freeing the journal header.
 872                  */
 873                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 874                         spin_unlock(lock);
 875                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 876                         spin_lock(lock);
 877                 }
 878                 put_bh(bh);
 879                 cond_resched_lock(lock);
 880         }
 881         spin_unlock(lock);
 882         return ret;
 883 }
 884
 885 static int flush_older_commits(struct super_block *s,
 886                                struct reiserfs_journal_list *jl)
 887 {
 888         struct reiserfs_journal *journal = SB_JOURNAL(s);
 889         struct reiserfs_journal_list *other_jl;
 890         struct reiserfs_journal_list *first_jl;
 891         struct list_head *entry;
 892         unsigned int trans_id = jl->j_trans_id;
 893         unsigned int other_trans_id;
 894         unsigned int first_trans_id;
 895
 896 find_first:
 897         /*
 898          * first we walk backwards to find the oldest uncommitted transation
 899          */
 900         first_jl = jl;
 901         entry = jl->j_list.prev;
 902         while (1) {
 903                 other_jl = JOURNAL_LIST_ENTRY(entry);
 904                 if (entry == &journal->j_journal_list ||
 905                     atomic_read(&other_jl->j_older_commits_done))
 906                         break;
 907
 908                 first_jl = other_jl;
 909                 entry = other_jl->j_list.prev;
 910         }
 911
 912         /* if we didn't find any older uncommitted transactions, return now */
 913         if (first_jl == jl) {
 914                 return 0;
 915         }
 916
 917         first_trans_id = first_jl->j_trans_id;
 918
 919         entry = &first_jl->j_list;
 920         while (1) {
 921                 other_jl = JOURNAL_LIST_ENTRY(entry);
 922                 other_trans_id = other_jl->j_trans_id;
 923
 924                 if (other_trans_id < trans_id) {
 925                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 926                                 flush_commit_list(s, other_jl, 0);
 927
 928                                 /* list we were called with is gone, return */
 929                                 if (!journal_list_still_alive(s, trans_id))
 930                                         return 1;
 931
 932                                 /*
 933                                  * the one we just flushed is gone, this means
 934                                  * all older lists are also gone, so first_jl
 935                                  * is no longer valid either.  Go back to the
 936                                  * beginning.
 937                                  */
 938                                 if (!journal_list_still_alive
 939                                     (s, other_trans_id)) {
 940                                         goto find_first;
 941                                 }
 942                         }
 943                         entry = entry->next;
 944                         if (entry == &journal->j_journal_list)
 945                                 return 0;
 946                 } else {
 947                         return 0;
 948                 }
 949         }
 950         return 0;
 951 }
 952
 953 static int reiserfs_async_progress_wait(struct super_block *s)
 954 {
 955         struct reiserfs_journal *j = SB_JOURNAL(s);
 956
 957         if (atomic_read(&j->j_async_throttle)) {
 958                 int depth;
 959
 960                 depth = reiserfs_write_unlock_nested(s);
 961                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 962                 reiserfs_write_lock_nested(s, depth);
 963         }
 964
 965         return 0;
 966 }
 967
 968 /*
 969  * if this journal list still has commit blocks unflushed, send them to disk.
 970  *
 971  * log areas must be flushed in order (transaction 2 can't commit before
 972  * transaction 1) Before the commit block can by written, every other log
 973  * block must be safely on disk
 974  */
 975 static int flush_commit_list(struct super_block *s,
 976                              struct reiserfs_journal_list *jl, int flushall)
 977 {
 978         int i;
 979         b_blocknr_t bn;
 980         struct buffer_head *tbh = NULL;
 981         unsigned int trans_id = jl->j_trans_id;
 982         struct reiserfs_journal *journal = SB_JOURNAL(s);
 983         int retval = 0;
 984         int write_len;
 985         int depth;
 986
 987         reiserfs_check_lock_depth(s, "flush_commit_list");
 988
 989         if (atomic_read(&jl->j_older_commits_done)) {
 990                 return 0;
 991         }
 992
 993         /*
 994          * before we can put our commit blocks on disk, we have to make
 995          * sure everyone older than us is on disk too
 996          */
 997         BUG_ON(jl->j_len <= 0);
 998         BUG_ON(trans_id == journal->j_trans_id);
 999
1000         get_journal_list(jl);
1001         if (flushall) {
1002                 if (flush_older_commits(s, jl) == 1) {
1003                         /*
1004                          * list disappeared during flush_older_commits.
1005                          * return
1006                          */
1007                         goto put_jl;
1008                 }
1009         }
1010
1011         /* make sure nobody is trying to flush this one at the same time */
1012         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1013
1014         if (!journal_list_still_alive(s, trans_id)) {
1015                 mutex_unlock(&jl->j_commit_mutex);
1016                 goto put_jl;
1017         }
1018         BUG_ON(jl->j_trans_id == 0);
1019
1020         /* this commit is done, exit */
1021         if (atomic_read(&jl->j_commit_left) <= 0) {
1022                 if (flushall) {
1023                         atomic_set(&jl->j_older_commits_done, 1);
1024                 }
1025                 mutex_unlock(&jl->j_commit_mutex);
1026                 goto put_jl;
1027         }
1028
1029         if (!list_empty(&jl->j_bh_list)) {
1030                 int ret;
1031
1032                 /*
1033                  * We might sleep in numerous places inside
1034                  * write_ordered_buffers. Relax the write lock.
1035                  */
1036                 depth = reiserfs_write_unlock_nested(s);
1037                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1038                                             journal, jl, &jl->j_bh_list);
1039                 if (ret < 0 && retval == 0)
1040                         retval = ret;
1041                 reiserfs_write_lock_nested(s, depth);
1042         }
1043         BUG_ON(!list_empty(&jl->j_bh_list));
1044         /*
1045          * for the description block and all the log blocks, submit any buffers
1046          * that haven't already reached the disk.  Try to write at least 256
1047          * log blocks. later on, we will only wait on blocks that correspond
1048          * to this transaction, but while we're unplugging we might as well
1049          * get a chunk of data on there.
1050          */
1051         atomic_inc(&journal->j_async_throttle);
1052         write_len = jl->j_len + 1;
1053         if (write_len < 256)
1054                 write_len = 256;
1055         for (i = 0 ; i < write_len ; i++) {
1056                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1057                     SB_ONDISK_JOURNAL_SIZE(s);
1058                 tbh = journal_find_get_block(s, bn);
1059                 if (tbh) {
1060                         if (buffer_dirty(tbh)) {
1061                             depth = reiserfs_write_unlock_nested(s);
1062                             ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1063                             reiserfs_write_lock_nested(s, depth);
1064                         }
1065                         put_bh(tbh) ;
1066                 }
1067         }
1068         atomic_dec(&journal->j_async_throttle);
1069
1070         for (i = 0; i < (jl->j_len + 1); i++) {
1071                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1072                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1073                 tbh = journal_find_get_block(s, bn);
1074
1075                 depth = reiserfs_write_unlock_nested(s);
1076                 __wait_on_buffer(tbh);
1077                 reiserfs_write_lock_nested(s, depth);
1078                 /*
1079                  * since we're using ll_rw_blk above, it might have skipped
1080                  * over a locked buffer.  Double check here
1081                  */
1082                 /* redundant, sync_dirty_buffer() checks */
1083                 if (buffer_dirty(tbh)) {
1084                         depth = reiserfs_write_unlock_nested(s);
1085                         sync_dirty_buffer(tbh);
1086                         reiserfs_write_lock_nested(s, depth);
1087                 }
1088                 if (unlikely(!buffer_uptodate(tbh))) {
1089 #ifdef CONFIG_REISERFS_CHECK
1090                         reiserfs_warning(s, "journal-601",
1091                                          "buffer write failed");
1092 #endif
1093                         retval = -EIO;
1094                 }
1095                 /* once for journal_find_get_block */
1096                 put_bh(tbh);
1097                 /* once due to original getblk in do_journal_end */
1098                 put_bh(tbh);
1099                 atomic_dec(&jl->j_commit_left);
1100         }
1101
1102         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1103
1104         /*
1105          * If there was a write error in the journal - we can't commit
1106          * this transaction - it will be invalid and, if successful,
1107          * will just end up propagating the write error out to
1108          * the file system.
1109          */
1110         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1111                 if (buffer_dirty(jl->j_commit_bh))
1112                         BUG();
1113                 mark_buffer_dirty(jl->j_commit_bh) ;
1114                 depth = reiserfs_write_unlock_nested(s);
1115                 if (reiserfs_barrier_flush(s))
1116                         __sync_dirty_buffer(jl->j_commit_bh,
1117                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1118                 else
1119                         sync_dirty_buffer(jl->j_commit_bh);
1120                 reiserfs_write_lock_nested(s, depth);
1121         }
1122
1123         /*
1124          * If there was a write error in the journal - we can't commit this
1125          * transaction - it will be invalid and, if successful, will just end
1126          * up propagating the write error out to the filesystem.
1127          */
1128         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1129 #ifdef CONFIG_REISERFS_CHECK
1130                 reiserfs_warning(s, "journal-615", "buffer write failed");
1131 #endif
1132                 retval = -EIO;
1133         }
1134         bforget(jl->j_commit_bh);
1135         if (journal->j_last_commit_id != 0 &&
1136             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1137                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1138                                  journal->j_last_commit_id, jl->j_trans_id);
1139         }
1140         journal->j_last_commit_id = jl->j_trans_id;
1141
1142         /*
1143          * now, every commit block is on the disk.  It is safe to allow
1144          * blocks freed during this transaction to be reallocated
1145          */
1146         cleanup_freed_for_journal_list(s, jl);
1147
1148         retval = retval ? retval : journal->j_errno;
1149
1150         /* mark the metadata dirty */
1151         if (!retval)
1152                 dirty_one_transaction(s, jl);
1153         atomic_dec(&jl->j_commit_left);
1154
1155         if (flushall) {
1156                 atomic_set(&jl->j_older_commits_done, 1);
1157         }
1158         mutex_unlock(&jl->j_commit_mutex);
1159 put_jl:
1160         put_journal_list(s, jl);
1161
1162         if (retval)
1163                 reiserfs_abort(s, retval, "Journal write error in %s",
1164                                __func__);
1165         return retval;
1166 }
1167
1168 /*
1169  * flush_journal_list frequently needs to find a newer transaction for a
1170  * given block.  This does that, or returns NULL if it can't find anything
1171  */
1172 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1173                                                           reiserfs_journal_cnode
1174                                                           *cn)
1175 {
1176         struct super_block *sb = cn->sb;
1177         b_blocknr_t blocknr = cn->blocknr;
1178
1179         cn = cn->hprev;
1180         while (cn) {
1181                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1182                         return cn->jlist;
1183                 }
1184                 cn = cn->hprev;
1185         }
1186         return NULL;
1187 }
1188
1189 static void remove_journal_hash(struct super_block *,
1190                                 struct reiserfs_journal_cnode **,
1191                                 struct reiserfs_journal_list *, unsigned long,
1192                                 int);
1193
1194 /*
1195  * once all the real blocks have been flushed, it is safe to remove them
1196  * from the journal list for this transaction.  Aside from freeing the
1197  * cnode, this also allows the block to be reallocated for data blocks
1198  * if it had been deleted.
1199  */
1200 static void remove_all_from_journal_list(struct super_block *sb,
1201                                          struct reiserfs_journal_list *jl,
1202                                          int debug)
1203 {
1204         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1205         struct reiserfs_journal_cnode *cn, *last;
1206         cn = jl->j_realblock;
1207
1208         /*
1209          * which is better, to lock once around the whole loop, or
1210          * to lock for each call to remove_journal_hash?
1211          */
1212         while (cn) {
1213                 if (cn->blocknr != 0) {
1214                         if (debug) {
1215                                 reiserfs_warning(sb, "reiserfs-2201",
1216                                                  "block %u, bh is %d, state %ld",
1217                                                  cn->blocknr, cn->bh ? 1 : 0,
1218                                                  cn->state);
1219                         }
1220                         cn->state = 0;
1221                         remove_journal_hash(sb, journal->j_list_hash_table,
1222                                             jl, cn->blocknr, 1);
1223                 }
1224                 last = cn;
1225                 cn = cn->next;
1226                 free_cnode(sb, last);
1227         }
1228         jl->j_realblock = NULL;
1229 }
1230
1231 /*
1232  * if this timestamp is greater than the timestamp we wrote last to the
1233  * header block, write it to the header block.  once this is done, I can
1234  * safely say the log area for this transaction won't ever be replayed,
1235  * and I can start releasing blocks in this transaction for reuse as data
1236  * blocks.  called by flush_journal_list, before it calls
1237  * remove_all_from_journal_list
1238  */
1239 static int _update_journal_header_block(struct super_block *sb,
1240                                         unsigned long offset,
1241                                         unsigned int trans_id)
1242 {
1243         struct reiserfs_journal_header *jh;
1244         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1245         int depth;
1246
1247         if (reiserfs_is_journal_aborted(journal))
1248                 return -EIO;
1249
1250         if (trans_id >= journal->j_last_flush_trans_id) {
1251                 if (buffer_locked((journal->j_header_bh))) {
1252                         depth = reiserfs_write_unlock_nested(sb);
1253                         __wait_on_buffer(journal->j_header_bh);
1254                         reiserfs_write_lock_nested(sb, depth);
1255                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1256 #ifdef CONFIG_REISERFS_CHECK
1257                                 reiserfs_warning(sb, "journal-699",
1258                                                  "buffer write failed");
1259 #endif
1260                                 return -EIO;
1261                         }
1262                 }
1263                 journal->j_last_flush_trans_id = trans_id;
1264                 journal->j_first_unflushed_offset = offset;
1265                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1266                                                         b_data);
1267                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1268                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1269                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1270
1271                 set_buffer_dirty(journal->j_header_bh);
1272                 depth = reiserfs_write_unlock_nested(sb);
1273
1274                 if (reiserfs_barrier_flush(sb))
1275                         __sync_dirty_buffer(journal->j_header_bh,
1276                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1277                 else
1278                         sync_dirty_buffer(journal->j_header_bh);
1279
1280                 reiserfs_write_lock_nested(sb, depth);
1281                 if (!buffer_uptodate(journal->j_header_bh)) {
1282                         reiserfs_warning(sb, "journal-837",
1283                                          "IO error during journal replay");
1284                         return -EIO;
1285                 }
1286         }
1287         return 0;
1288 }
1289
1290 static int update_journal_header_block(struct super_block *sb,
1291                                        unsigned long offset,
1292                                        unsigned int trans_id)
1293 {
1294         return _update_journal_header_block(sb, offset, trans_id);
1295 }
1296
1297 /*
1298 ** flush any and all journal lists older than you are
1299 ** can only be called from flush_journal_list
1300 */
1301 static int flush_older_journal_lists(struct super_block *sb,
1302                                      struct reiserfs_journal_list *jl)
1303 {
1304         struct list_head *entry;
1305         struct reiserfs_journal_list *other_jl;
1306         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1307         unsigned int trans_id = jl->j_trans_id;
1308
1309         /*
1310          * we know we are the only ones flushing things, no extra race
1311          * protection is required.
1312          */
1313 restart:
1314         entry = journal->j_journal_list.next;
1315         /* Did we wrap? */
1316         if (entry == &journal->j_journal_list)
1317                 return 0;
1318         other_jl = JOURNAL_LIST_ENTRY(entry);
1319         if (other_jl->j_trans_id < trans_id) {
1320                 BUG_ON(other_jl->j_refcount <= 0);
1321                 /* do not flush all */
1322                 flush_journal_list(sb, other_jl, 0);
1323
1324                 /* other_jl is now deleted from the list */
1325                 goto restart;
1326         }
1327         return 0;
1328 }
1329
1330 static void del_from_work_list(struct super_block *s,
1331                                struct reiserfs_journal_list *jl)
1332 {
1333         struct reiserfs_journal *journal = SB_JOURNAL(s);
1334         if (!list_empty(&jl->j_working_list)) {
1335                 list_del_init(&jl->j_working_list);
1336                 journal->j_num_work_lists--;
1337         }
1338 }
1339
1340 /*
1341  * flush a journal list, both commit and real blocks
1342  *
1343  * always set flushall to 1, unless you are calling from inside
1344  * flush_journal_list
1345  *
1346  * IMPORTANT.  This can only be called while there are no journal writers,
1347  * and the journal is locked.  That means it can only be called from
1348  * do_journal_end, or by journal_release
1349  */
1350 static int flush_journal_list(struct super_block *s,
1351                               struct reiserfs_journal_list *jl, int flushall)
1352 {
1353         struct reiserfs_journal_list *pjl;
1354         struct reiserfs_journal_cnode *cn, *last;
1355         int count;
1356         int was_jwait = 0;
1357         int was_dirty = 0;
1358         struct buffer_head *saved_bh;
1359         unsigned long j_len_saved = jl->j_len;
1360         struct reiserfs_journal *journal = SB_JOURNAL(s);
1361         int err = 0;
1362         int depth;
1363
1364         BUG_ON(j_len_saved <= 0);
1365
1366         if (atomic_read(&journal->j_wcount) != 0) {
1367                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1368                                  atomic_read(&journal->j_wcount));
1369         }
1370
1371         /* if flushall == 0, the lock is already held */
1372         if (flushall) {
1373                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1374         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1375                 BUG();
1376         }
1377
1378         count = 0;
1379         if (j_len_saved > journal->j_trans_max) {
1380                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1381                                j_len_saved, jl->j_trans_id);
1382                 return 0;
1383         }
1384
1385         /* if all the work is already done, get out of here */
1386         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1387             atomic_read(&jl->j_commit_left) <= 0) {
1388                 goto flush_older_and_return;
1389         }
1390
1391         /*
1392          * start by putting the commit list on disk.  This will also flush
1393          * the commit lists of any olders transactions
1394          */
1395         flush_commit_list(s, jl, 1);
1396
1397         if (!(jl->j_state & LIST_DIRTY)
1398             && !reiserfs_is_journal_aborted(journal))
1399                 BUG();
1400
1401         /* are we done now? */
1402         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1403             atomic_read(&jl->j_commit_left) <= 0) {
1404                 goto flush_older_and_return;
1405         }
1406
1407         /*
1408          * loop through each cnode, see if we need to write it,
1409          * or wait on a more recent transaction, or just ignore it
1410          */
1411         if (atomic_read(&journal->j_wcount) != 0) {
1412                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1413                                "wcount is not 0");
1414         }
1415         cn = jl->j_realblock;
1416         while (cn) {
1417                 was_jwait = 0;
1418                 was_dirty = 0;
1419                 saved_bh = NULL;
1420                 /* blocknr of 0 is no longer in the hash, ignore it */
1421                 if (cn->blocknr == 0) {
1422                         goto free_cnode;
1423                 }
1424
1425                 /*
1426                  * This transaction failed commit.
1427                  * Don't write out to the disk
1428                  */
1429                 if (!(jl->j_state & LIST_DIRTY))
1430                         goto free_cnode;
1431
1432                 pjl = find_newer_jl_for_cn(cn);
1433                 /*
1434                  * the order is important here.  We check pjl to make sure we
1435                  * don't clear BH_JDirty_wait if we aren't the one writing this
1436                  * block to disk
1437                  */
1438                 if (!pjl && cn->bh) {
1439                         saved_bh = cn->bh;
1440
1441                         /*
1442                          * we do this to make sure nobody releases the
1443                          * buffer while we are working with it
1444                          */
1445                         get_bh(saved_bh);
1446
1447                         if (buffer_journal_dirty(saved_bh)) {
1448                                 BUG_ON(!can_dirty(cn));
1449                                 was_jwait = 1;
1450                                 was_dirty = 1;
1451                         } else if (can_dirty(cn)) {
1452                                 /*
1453                                  * everything with !pjl && jwait
1454                                  * should be writable
1455                                  */
1456                                 BUG();
1457                         }
1458                 }
1459
1460                 /*
1461                  * if someone has this block in a newer transaction, just make
1462                  * sure they are committed, and don't try writing it to disk
1463                  */
1464                 if (pjl) {
1465                         if (atomic_read(&pjl->j_commit_left))
1466                                 flush_commit_list(s, pjl, 1);
1467                         goto free_cnode;
1468                 }
1469
1470                 /*
1471                  * bh == NULL when the block got to disk on its own, OR,
1472                  * the block got freed in a future transaction
1473                  */
1474                 if (saved_bh == NULL) {
1475                         goto free_cnode;
1476                 }
1477
1478                 /*
1479                  * this should never happen.  kupdate_one_transaction has
1480                  * this list locked while it works, so we should never see a
1481                  * buffer here that is not marked JDirty_wait
1482                  */
1483                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1484                         reiserfs_warning(s, "journal-813",
1485                                          "BAD! buffer %llu %cdirty %cjwait, "
1486                                          "not in a newer transaction",
1487                                          (unsigned long long)saved_bh->
1488                                          b_blocknr, was_dirty ? ' ' : '!',
1489                                          was_jwait ? ' ' : '!');
1490                 }
1491                 if (was_dirty) {
1492                         /*
1493                          * we inc again because saved_bh gets decremented
1494                          * at free_cnode
1495                          */
1496                         get_bh(saved_bh);
1497                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1498                         lock_buffer(saved_bh);
1499                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1500                         if (buffer_dirty(saved_bh))
1501                                 submit_logged_buffer(saved_bh);
1502                         else
1503                                 unlock_buffer(saved_bh);
1504                         count++;
1505                 } else {
1506                         reiserfs_warning(s, "clm-2082",
1507                                          "Unable to flush buffer %llu in %s",
1508                                          (unsigned long long)saved_bh->
1509                                          b_blocknr, __func__);
1510                 }
1511 free_cnode:
1512                 last = cn;
1513                 cn = cn->next;
1514                 if (saved_bh) {
1515                         /*
1516                          * we incremented this to keep others from
1517                          * taking the buffer head away
1518                          */
1519                         put_bh(saved_bh);
1520                         if (atomic_read(&saved_bh->b_count) < 0) {
1521                                 reiserfs_warning(s, "journal-945",
1522                                                  "saved_bh->b_count < 0");
1523                         }
1524                 }
1525         }
1526         if (count > 0) {
1527                 cn = jl->j_realblock;
1528                 while (cn) {
1529                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1530                                 if (!cn->bh) {
1531                                         reiserfs_panic(s, "journal-1011",
1532                                                        "cn->bh is NULL");
1533                                 }
1534
1535                                 depth = reiserfs_write_unlock_nested(s);
1536                                 __wait_on_buffer(cn->bh);
1537                                 reiserfs_write_lock_nested(s, depth);
1538
1539                                 if (!cn->bh) {
1540                                         reiserfs_panic(s, "journal-1012",
1541                                                        "cn->bh is NULL");
1542                                 }
1543                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1544 #ifdef CONFIG_REISERFS_CHECK
1545                                         reiserfs_warning(s, "journal-949",
1546                                                          "buffer write failed");
1547 #endif
1548                                         err = -EIO;
1549                                 }
1550                                 /*
1551                                  * note, we must clear the JDirty_wait bit
1552                                  * after the up to date check, otherwise we
1553                                  * race against our flushpage routine
1554                                  */
1555                                 BUG_ON(!test_clear_buffer_journal_dirty
1556                                        (cn->bh));
1557
1558                                 /* drop one ref for us */
1559                                 put_bh(cn->bh);
1560                                 /* drop one ref for journal_mark_dirty */
1561                                 release_buffer_page(cn->bh);
1562                         }
1563                         cn = cn->next;
1564                 }
1565         }
1566
1567         if (err)
1568                 reiserfs_abort(s, -EIO,
1569                                "Write error while pushing transaction to disk in %s",
1570                                __func__);
1571 flush_older_and_return:
1572
1573         /*
1574          * before we can update the journal header block, we _must_ flush all
1575          * real blocks from all older transactions to disk.  This is because
1576          * once the header block is updated, this transaction will not be
1577          * replayed after a crash
1578          */
1579         if (flushall) {
1580                 flush_older_journal_lists(s, jl);
1581         }
1582
1583         err = journal->j_errno;
1584         /*
1585          * before we can remove everything from the hash tables for this
1586          * transaction, we must make sure it can never be replayed
1587          *
1588          * since we are only called from do_journal_end, we know for sure there
1589          * are no allocations going on while we are flushing journal lists.  So,
1590          * we only need to update the journal header block for the last list
1591          * being flushed
1592          */
1593         if (!err && flushall) {
1594                 err =
1595                     update_journal_header_block(s,
1596                                                 (jl->j_start + jl->j_len +
1597                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1598                                                 jl->j_trans_id);
1599                 if (err)
1600                         reiserfs_abort(s, -EIO,
1601                                        "Write error while updating journal header in %s",
1602                                        __func__);
1603         }
1604         remove_all_from_journal_list(s, jl, 0);
1605         list_del_init(&jl->j_list);
1606         journal->j_num_lists--;
1607         del_from_work_list(s, jl);
1608
1609         if (journal->j_last_flush_id != 0 &&
1610             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1611                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1612                                  journal->j_last_flush_id, jl->j_trans_id);
1613         }
1614         journal->j_last_flush_id = jl->j_trans_id;
1615
1616         /*
1617          * not strictly required since we are freeing the list, but it should
1618          * help find code using dead lists later on
1619          */
1620         jl->j_len = 0;
1621         atomic_set(&jl->j_nonzerolen, 0);
1622         jl->j_start = 0;
1623         jl->j_realblock = NULL;
1624         jl->j_commit_bh = NULL;
1625         jl->j_trans_id = 0;
1626         jl->j_state = 0;
1627         put_journal_list(s, jl);
1628         if (flushall)
1629                 mutex_unlock(&journal->j_flush_mutex);
1630         return err;
1631 }
1632
1633 static int write_one_transaction(struct super_block *s,
1634                                  struct reiserfs_journal_list *jl,
1635                                  struct buffer_chunk *chunk)
1636 {
1637         struct reiserfs_journal_cnode *cn;
1638         int ret = 0;
1639
1640         jl->j_state |= LIST_TOUCHED;
1641         del_from_work_list(s, jl);
1642         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1643                 return 0;
1644         }
1645
1646         cn = jl->j_realblock;
1647         while (cn) {
1648                 /*
1649                  * if the blocknr == 0, this has been cleared from the hash,
1650                  * skip it
1651                  */
1652                 if (cn->blocknr == 0) {
1653                         goto next;
1654                 }
1655                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1656                         struct buffer_head *tmp_bh;
1657                         /*
1658                          * we can race against journal_mark_freed when we try
1659                          * to lock_buffer(cn->bh), so we have to inc the buffer
1660                          * count, and recheck things after locking
1661                          */
1662                         tmp_bh = cn->bh;
1663                         get_bh(tmp_bh);
1664                         lock_buffer(tmp_bh);
1665                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1666                                 if (!buffer_journal_dirty(tmp_bh) ||
1667                                     buffer_journal_prepared(tmp_bh))
1668                                         BUG();
1669                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1670                                 ret++;
1671                         } else {
1672                                 /* note, cn->bh might be null now */
1673                                 unlock_buffer(tmp_bh);
1674                         }
1675                         put_bh(tmp_bh);
1676                 }
1677 next:
1678                 cn = cn->next;
1679                 cond_resched();
1680         }
1681         return ret;
1682 }
1683
1684 /* used by flush_commit_list */
1685 static void dirty_one_transaction(struct super_block *s,
1686                                  struct reiserfs_journal_list *jl)
1687 {
1688         struct reiserfs_journal_cnode *cn;
1689         struct reiserfs_journal_list *pjl;
1690
1691         jl->j_state |= LIST_DIRTY;
1692         cn = jl->j_realblock;
1693         while (cn) {
1694                 /*
1695                  * look for a more recent transaction that logged this
1696                  * buffer.  Only the most recent transaction with a buffer in
1697                  * it is allowed to send that buffer to disk
1698                  */
1699                 pjl = find_newer_jl_for_cn(cn);
1700                 if (!pjl && cn->blocknr && cn->bh
1701                     && buffer_journal_dirty(cn->bh)) {
1702                         BUG_ON(!can_dirty(cn));
1703                         /*
1704                          * if the buffer is prepared, it will either be logged
1705                          * or restored.  If restored, we need to make sure
1706                          * it actually gets marked dirty
1707                          */
1708                         clear_buffer_journal_new(cn->bh);
1709                         if (buffer_journal_prepared(cn->bh)) {
1710                                 set_buffer_journal_restore_dirty(cn->bh);
1711                         } else {
1712                                 set_buffer_journal_test(cn->bh);
1713                                 mark_buffer_dirty(cn->bh);
1714                         }
1715                 }
1716                 cn = cn->next;
1717         }
1718 }
1719
1720 static int kupdate_transactions(struct super_block *s,
1721                                 struct reiserfs_journal_list *jl,
1722                                 struct reiserfs_journal_list **next_jl,
1723                                 unsigned int *next_trans_id,
1724                                 int num_blocks, int num_trans)
1725 {
1726         int ret = 0;
1727         int written = 0;
1728         int transactions_flushed = 0;
1729         unsigned int orig_trans_id = jl->j_trans_id;
1730         struct buffer_chunk chunk;
1731         struct list_head *entry;
1732         struct reiserfs_journal *journal = SB_JOURNAL(s);
1733         chunk.nr = 0;
1734
1735         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1736         if (!journal_list_still_alive(s, orig_trans_id)) {
1737                 goto done;
1738         }
1739
1740         /*
1741          * we've got j_flush_mutex held, nobody is going to delete any
1742          * of these lists out from underneath us
1743          */
1744         while ((num_trans && transactions_flushed < num_trans) ||
1745                (!num_trans && written < num_blocks)) {
1746
1747                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1748                     atomic_read(&jl->j_commit_left)
1749                     || !(jl->j_state & LIST_DIRTY)) {
1750                         del_from_work_list(s, jl);
1751                         break;
1752                 }
1753                 ret = write_one_transaction(s, jl, &chunk);
1754
1755                 if (ret < 0)
1756                         goto done;
1757                 transactions_flushed++;
1758                 written += ret;
1759                 entry = jl->j_list.next;
1760
1761                 /* did we wrap? */
1762                 if (entry == &journal->j_journal_list) {
1763                         break;
1764                 }
1765                 jl = JOURNAL_LIST_ENTRY(entry);
1766
1767                 /* don't bother with older transactions */
1768                 if (jl->j_trans_id <= orig_trans_id)
1769                         break;
1770         }
1771         if (chunk.nr) {
1772                 write_chunk(&chunk);
1773         }
1774
1775 done:
1776         mutex_unlock(&journal->j_flush_mutex);
1777         return ret;
1778 }
1779
1780 /*
1781  * for o_sync and fsync heavy applications, they tend to use
1782  * all the journa list slots with tiny transactions.  These
1783  * trigger lots and lots of calls to update the header block, which
1784  * adds seeks and slows things down.
1785  *
1786  * This function tries to clear out a large chunk of the journal lists
1787  * at once, which makes everything faster since only the newest journal
1788  * list updates the header block
1789  */
1790 static int flush_used_journal_lists(struct super_block *s,
1791                                     struct reiserfs_journal_list *jl)
1792 {
1793         unsigned long len = 0;
1794         unsigned long cur_len;
1795         int ret;
1796         int i;
1797         int limit = 256;
1798         struct reiserfs_journal_list *tjl;
1799         struct reiserfs_journal_list *flush_jl;
1800         unsigned int trans_id;
1801         struct reiserfs_journal *journal = SB_JOURNAL(s);
1802
1803         flush_jl = tjl = jl;
1804
1805         /* in data logging mode, try harder to flush a lot of blocks */
1806         if (reiserfs_data_log(s))
1807                 limit = 1024;
1808         /* flush for 256 transactions or limit blocks, whichever comes first */
1809         for (i = 0; i < 256 && len < limit; i++) {
1810                 if (atomic_read(&tjl->j_commit_left) ||
1811                     tjl->j_trans_id < jl->j_trans_id) {
1812                         break;
1813                 }
1814                 cur_len = atomic_read(&tjl->j_nonzerolen);
1815                 if (cur_len > 0) {
1816                         tjl->j_state &= ~LIST_TOUCHED;
1817                 }
1818                 len += cur_len;
1819                 flush_jl = tjl;
1820                 if (tjl->j_list.next == &journal->j_journal_list)
1821                         break;
1822                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1823         }
1824         get_journal_list(jl);
1825         get_journal_list(flush_jl);
1826
1827         /*
1828          * try to find a group of blocks we can flush across all the
1829          * transactions, but only bother if we've actually spanned
1830          * across multiple lists
1831          */
1832         if (flush_jl != jl) {
1833                 ret = kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1834         }
1835         flush_journal_list(s, flush_jl, 1);
1836         put_journal_list(s, flush_jl);
1837         put_journal_list(s, jl);
1838         return 0;
1839 }
1840
1841 /*
1842  * removes any nodes in table with name block and dev as bh.
1843  * only touchs the hnext and hprev pointers.
1844  */
1845 static void remove_journal_hash(struct super_block *sb,
1846                          struct reiserfs_journal_cnode **table,
1847                          struct reiserfs_journal_list *jl,
1848                          unsigned long block, int remove_freed)
1849 {
1850         struct reiserfs_journal_cnode *cur;
1851         struct reiserfs_journal_cnode **head;
1852
1853         head = &(journal_hash(table, sb, block));
1854         if (!head) {
1855                 return;
1856         }
1857         cur = *head;
1858         while (cur) {
1859                 if (cur->blocknr == block && cur->sb == sb
1860                     && (jl == NULL || jl == cur->jlist)
1861                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1862                         if (cur->hnext) {
1863                                 cur->hnext->hprev = cur->hprev;
1864                         }
1865                         if (cur->hprev) {
1866                                 cur->hprev->hnext = cur->hnext;
1867                         } else {
1868                                 *head = cur->hnext;
1869                         }
1870                         cur->blocknr = 0;
1871                         cur->sb = NULL;
1872                         cur->state = 0;
1873                         /*
1874                          * anybody who clears the cur->bh will also
1875                          * dec the nonzerolen
1876                          */
1877                         if (cur->bh && cur->jlist)
1878                                 atomic_dec(&cur->jlist->j_nonzerolen);
1879                         cur->bh = NULL;
1880                         cur->jlist = NULL;
1881                 }
1882                 cur = cur->hnext;
1883         }
1884 }
1885
1886 static void free_journal_ram(struct super_block *sb)
1887 {
1888         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1889         kfree(journal->j_current_jl);
1890         journal->j_num_lists--;
1891
1892         vfree(journal->j_cnode_free_orig);
1893         free_list_bitmaps(sb, journal->j_list_bitmap);
1894         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1895         if (journal->j_header_bh) {
1896                 brelse(journal->j_header_bh);
1897         }
1898         /*
1899          * j_header_bh is on the journal dev, make sure
1900          * not to release the journal dev until we brelse j_header_bh
1901          */
1902         release_journal_dev(sb, journal);
1903         vfree(journal);
1904 }
1905
1906 /*
1907  * call on unmount.  Only set error to 1 if you haven't made your way out
1908  * of read_super() yet.  Any other caller must keep error at 0.
1909  */
1910 static int do_journal_release(struct reiserfs_transaction_handle *th,
1911                               struct super_block *sb, int error)
1912 {
1913         struct reiserfs_transaction_handle myth;
1914         int flushed = 0;
1915         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1916
1917         /*
1918          * we only want to flush out transactions if we were
1919          * called with error == 0
1920          */
1921         if (!error && !sb_rdonly(sb)) {
1922                 /* end the current trans */
1923                 BUG_ON(!th->t_trans_id);
1924                 do_journal_end(th, FLUSH_ALL);
1925
1926                 /*
1927                  * make sure something gets logged to force
1928                  * our way into the flush code
1929                  */
1930                 if (!journal_join(&myth, sb)) {
1931                         reiserfs_prepare_for_journal(sb,
1932                                                      SB_BUFFER_WITH_SB(sb),
1933                                                      1);
1934                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1935                         do_journal_end(&myth, FLUSH_ALL);
1936                         flushed = 1;
1937                 }
1938         }
1939
1940         /* this also catches errors during the do_journal_end above */
1941         if (!error && reiserfs_is_journal_aborted(journal)) {
1942                 memset(&myth, 0, sizeof(myth));
1943                 if (!journal_join_abort(&myth, sb)) {
1944                         reiserfs_prepare_for_journal(sb,
1945                                                      SB_BUFFER_WITH_SB(sb),
1946                                                      1);
1947                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1948                         do_journal_end(&myth, FLUSH_ALL);
1949                 }
1950         }
1951
1952
1953         /*
1954          * We must release the write lock here because
1955          * the workqueue job (flush_async_commit) needs this lock
1956          */
1957         reiserfs_write_unlock(sb);
1958
1959         /*
1960          * Cancel flushing of old commits. Note that neither of these works
1961          * will be requeued because superblock is being shutdown and doesn't
1962          * have SB_ACTIVE set.
1963          */
1964         reiserfs_cancel_old_flush(sb);
1965         /* wait for all commits to finish */
1966         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1967
1968         free_journal_ram(sb);
1969
1970         reiserfs_write_lock(sb);
1971
1972         return 0;
1973 }
1974
1975 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1976 int journal_release(struct reiserfs_transaction_handle *th,
1977                     struct super_block *sb)
1978 {
1979         return do_journal_release(th, sb, 0);
1980 }
1981
1982 /* only call from an error condition inside reiserfs_read_super!  */
1983 int journal_release_error(struct reiserfs_transaction_handle *th,
1984                           struct super_block *sb)
1985 {
1986         return do_journal_release(th, sb, 1);
1987 }
1988
1989 /*
1990  * compares description block with commit block.
1991  * returns 1 if they differ, 0 if they are the same
1992  */
1993 static int journal_compare_desc_commit(struct super_block *sb,
1994                                        struct reiserfs_journal_desc *desc,
1995                                        struct reiserfs_journal_commit *commit)
1996 {
1997         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1998             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1999             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
2000             get_commit_trans_len(commit) <= 0) {
2001                 return 1;
2002         }
2003         return 0;
2004 }
2005
2006 /*
2007  * returns 0 if it did not find a description block
2008  * returns -1 if it found a corrupt commit block
2009  * returns 1 if both desc and commit were valid
2010  * NOTE: only called during fs mount
2011  */
2012 static int journal_transaction_is_valid(struct super_block *sb,
2013                                         struct buffer_head *d_bh,
2014                                         unsigned int *oldest_invalid_trans_id,
2015                                         unsigned long *newest_mount_id)
2016 {
2017         struct reiserfs_journal_desc *desc;
2018         struct reiserfs_journal_commit *commit;
2019         struct buffer_head *c_bh;
2020         unsigned long offset;
2021
2022         if (!d_bh)
2023                 return 0;
2024
2025         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2026         if (get_desc_trans_len(desc) > 0
2027             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2028                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2029                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2030                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2031                                        "journal-986: transaction "
2032                                        "is valid returning because trans_id %d is greater than "
2033                                        "oldest_invalid %lu",
2034                                        get_desc_trans_id(desc),
2035                                        *oldest_invalid_trans_id);
2036                         return 0;
2037                 }
2038                 if (newest_mount_id
2039                     && *newest_mount_id > get_desc_mount_id(desc)) {
2040                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2041                                        "journal-1087: transaction "
2042                                        "is valid returning because mount_id %d is less than "
2043                                        "newest_mount_id %lu",
2044                                        get_desc_mount_id(desc),
2045                                        *newest_mount_id);
2046                         return -1;
2047                 }
2048                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2049                         reiserfs_warning(sb, "journal-2018",
2050                                          "Bad transaction length %d "
2051                                          "encountered, ignoring transaction",
2052                                          get_desc_trans_len(desc));
2053                         return -1;
2054                 }
2055                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2056
2057                 /*
2058                  * ok, we have a journal description block,
2059                  * let's see if the transaction was valid
2060                  */
2061                 c_bh =
2062                     journal_bread(sb,
2063                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2064                                   ((offset + get_desc_trans_len(desc) +
2065                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2066                 if (!c_bh)
2067                         return 0;
2068                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2069                 if (journal_compare_desc_commit(sb, desc, commit)) {
2070                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2071                                        "journal_transaction_is_valid, commit offset %ld had bad "
2072                                        "time %d or length %d",
2073                                        c_bh->b_blocknr -
2074                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2075                                        get_commit_trans_id(commit),
2076                                        get_commit_trans_len(commit));
2077                         brelse(c_bh);
2078                         if (oldest_invalid_trans_id) {
2079                                 *oldest_invalid_trans_id =
2080                                     get_desc_trans_id(desc);
2081                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2082                                                "journal-1004: "
2083                                                "transaction_is_valid setting oldest invalid trans_id "
2084                                                "to %d",
2085                                                get_desc_trans_id(desc));
2086                         }
2087                         return -1;
2088                 }
2089                 brelse(c_bh);
2090                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2091                                "journal-1006: found valid "
2092                                "transaction start offset %llu, len %d id %d",
2093                                d_bh->b_blocknr -
2094                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2095                                get_desc_trans_len(desc),
2096                                get_desc_trans_id(desc));
2097                 return 1;
2098         } else {
2099                 return 0;
2100         }
2101 }
2102
2103 static void brelse_array(struct buffer_head **heads, int num)
2104 {
2105         int i;
2106         for (i = 0; i < num; i++) {
2107                 brelse(heads[i]);
2108         }
2109 }
2110
2111 /*
2112  * given the start, and values for the oldest acceptable transactions,
2113  * this either reads in a replays a transaction, or returns because the
2114  * transaction is invalid, or too old.
2115  * NOTE: only called during fs mount
2116  */
2117 static int journal_read_transaction(struct super_block *sb,
2118                                     unsigned long cur_dblock,
2119                                     unsigned long oldest_start,
2120                                     unsigned int oldest_trans_id,
2121                                     unsigned long newest_mount_id)
2122 {
2123         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2124         struct reiserfs_journal_desc *desc;
2125         struct reiserfs_journal_commit *commit;
2126         unsigned int trans_id = 0;
2127         struct buffer_head *c_bh;
2128         struct buffer_head *d_bh;
2129         struct buffer_head **log_blocks = NULL;
2130         struct buffer_head **real_blocks = NULL;
2131         unsigned int trans_offset;
2132         int i;
2133         int trans_half;
2134
2135         d_bh = journal_bread(sb, cur_dblock);
2136         if (!d_bh)
2137                 return 1;
2138         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2139         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2140         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2141                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2142                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2143                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2144         if (get_desc_trans_id(desc) < oldest_trans_id) {
2145                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2146                                "journal_read_trans skipping because %lu is too old",
2147                                cur_dblock -
2148                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2149                 brelse(d_bh);
2150                 return 1;
2151         }
2152         if (get_desc_mount_id(desc) != newest_mount_id) {
2153                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2154                                "journal_read_trans skipping because %d is != "
2155                                "newest_mount_id %lu", get_desc_mount_id(desc),
2156                                newest_mount_id);
2157                 brelse(d_bh);
2158                 return 1;
2159         }
2160         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2161                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2162                               SB_ONDISK_JOURNAL_SIZE(sb)));
2163         if (!c_bh) {
2164                 brelse(d_bh);
2165                 return 1;
2166         }
2167         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2168         if (journal_compare_desc_commit(sb, desc, commit)) {
2169                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2170                                "journal_read_transaction, "
2171                                "commit offset %llu had bad time %d or length %d",
2172                                c_bh->b_blocknr -
2173                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2174                                get_commit_trans_id(commit),
2175                                get_commit_trans_len(commit));
2176                 brelse(c_bh);
2177                 brelse(d_bh);
2178                 return 1;
2179         }
2180
2181         if (bdev_read_only(sb->s_bdev)) {
2182                 reiserfs_warning(sb, "clm-2076",
2183                                  "device is readonly, unable to replay log");
2184                 brelse(c_bh);
2185                 brelse(d_bh);
2186                 return -EROFS;
2187         }
2188
2189         trans_id = get_desc_trans_id(desc);
2190         /*
2191          * now we know we've got a good transaction, and it was
2192          * inside the valid time ranges
2193          */
2194         log_blocks = kmalloc_array(get_desc_trans_len(desc),
2195                                    sizeof(struct buffer_head *),
2196                                    GFP_NOFS);
2197         real_blocks = kmalloc_array(get_desc_trans_len(desc),
2198                                     sizeof(struct buffer_head *),
2199                                     GFP_NOFS);
2200         if (!log_blocks || !real_blocks) {
2201                 brelse(c_bh);
2202                 brelse(d_bh);
2203                 kfree(log_blocks);
2204                 kfree(real_blocks);
2205                 reiserfs_warning(sb, "journal-1169",
2206                                  "kmalloc failed, unable to mount FS");
2207                 return -1;
2208         }
2209         /* get all the buffer heads */
2210         trans_half = journal_trans_half(sb->s_blocksize);
2211         for (i = 0; i < get_desc_trans_len(desc); i++) {
2212                 log_blocks[i] =
2213                     journal_getblk(sb,
2214                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2215                                    (trans_offset + 1 +
2216                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2217                 if (i < trans_half) {
2218                         real_blocks[i] =
2219                             sb_getblk(sb,
2220                                       le32_to_cpu(desc->j_realblock[i]));
2221                 } else {
2222                         real_blocks[i] =
2223                             sb_getblk(sb,
2224                                       le32_to_cpu(commit->
2225                                                   j_realblock[i - trans_half]));
2226                 }
2227                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2228                         reiserfs_warning(sb, "journal-1207",
2229                                          "REPLAY FAILURE fsck required! "
2230                                          "Block to replay is outside of "
2231                                          "filesystem");
2232                         goto abort_replay;
2233                 }
2234                 /* make sure we don't try to replay onto log or reserved area */
2235                 if (is_block_in_log_or_reserved_area
2236                     (sb, real_blocks[i]->b_blocknr)) {
2237                         reiserfs_warning(sb, "journal-1204",
2238                                          "REPLAY FAILURE fsck required! "
2239                                          "Trying to replay onto a log block");
2240 abort_replay:
2241                         brelse_array(log_blocks, i);
2242                         brelse_array(real_blocks, i);
2243                         brelse(c_bh);
2244                         brelse(d_bh);
2245                         kfree(log_blocks);
2246                         kfree(real_blocks);
2247                         return -1;
2248                 }
2249         }
2250         /* read in the log blocks, memcpy to the corresponding real block */
2251         ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2252         for (i = 0; i < get_desc_trans_len(desc); i++) {
2253
2254                 wait_on_buffer(log_blocks[i]);
2255                 if (!buffer_uptodate(log_blocks[i])) {
2256                         reiserfs_warning(sb, "journal-1212",
2257                                          "REPLAY FAILURE fsck required! "
2258                                          "buffer write failed");
2259                         brelse_array(log_blocks + i,
2260                                      get_desc_trans_len(desc) - i);
2261                         brelse_array(real_blocks, get_desc_trans_len(desc));
2262                         brelse(c_bh);
2263                         brelse(d_bh);
2264                         kfree(log_blocks);
2265                         kfree(real_blocks);
2266                         return -1;
2267                 }
2268                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2269                        real_blocks[i]->b_size);
2270                 set_buffer_uptodate(real_blocks[i]);
2271                 brelse(log_blocks[i]);
2272         }
2273         /* flush out the real blocks */
2274         for (i = 0; i < get_desc_trans_len(desc); i++) {
2275                 set_buffer_dirty(real_blocks[i]);
2276                 write_dirty_buffer(real_blocks[i], 0);
2277         }
2278         for (i = 0; i < get_desc_trans_len(desc); i++) {
2279                 wait_on_buffer(real_blocks[i]);
2280                 if (!buffer_uptodate(real_blocks[i])) {
2281                         reiserfs_warning(sb, "journal-1226",
2282                                          "REPLAY FAILURE, fsck required! "
2283                                          "buffer write failed");
2284                         brelse_array(real_blocks + i,
2285                                      get_desc_trans_len(desc) - i);
2286                         brelse(c_bh);
2287                         brelse(d_bh);
2288                         kfree(log_blocks);
2289                         kfree(real_blocks);
2290                         return -1;
2291                 }
2292                 brelse(real_blocks[i]);
2293         }
2294         cur_dblock =
2295             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2296             ((trans_offset + get_desc_trans_len(desc) +
2297               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2298         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2299                        "journal-1095: setting journal " "start to offset %ld",
2300                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2301
2302         /*
2303          * init starting values for the first transaction, in case
2304          * this is the last transaction to be replayed.
2305          */
2306         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2307         journal->j_last_flush_trans_id = trans_id;
2308         journal->j_trans_id = trans_id + 1;
2309         /* check for trans_id overflow */
2310         if (journal->j_trans_id == 0)
2311                 journal->j_trans_id = 10;
2312         brelse(c_bh);
2313         brelse(d_bh);
2314         kfree(log_blocks);
2315         kfree(real_blocks);
2316         return 0;
2317 }
2318
2319 /*
2320  * This function reads blocks starting from block and to max_block of bufsize
2321  * size (but no more than BUFNR blocks at a time). This proved to improve
2322  * mounting speed on self-rebuilding raid5 arrays at least.
2323  * Right now it is only used from journal code. But later we might use it
2324  * from other places.
2325  * Note: Do not use journal_getblk/sb_getblk functions here!
2326  */
2327 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2328                                            b_blocknr_t block, int bufsize,
2329                                            b_blocknr_t max_block)
2330 {
2331         struct buffer_head *bhlist[BUFNR];
2332         unsigned int blocks = BUFNR;
2333         struct buffer_head *bh;
2334         int i, j;
2335
2336         bh = __getblk(dev, block, bufsize);
2337         if (buffer_uptodate(bh))
2338                 return (bh);
2339
2340         if (block + BUFNR > max_block) {
2341                 blocks = max_block - block;
2342         }
2343         bhlist[0] = bh;
2344         j = 1;
2345         for (i = 1; i < blocks; i++) {
2346                 bh = __getblk(dev, block + i, bufsize);
2347                 if (buffer_uptodate(bh)) {
2348                         brelse(bh);
2349                         break;
2350                 } else
2351                         bhlist[j++] = bh;
2352         }
2353         ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2354         for (i = 1; i < j; i++)
2355                 brelse(bhlist[i]);
2356         bh = bhlist[0];
2357         wait_on_buffer(bh);
2358         if (buffer_uptodate(bh))
2359                 return bh;
2360         brelse(bh);
2361         return NULL;
2362 }
2363
2364 /*
2365  * read and replay the log
2366  * on a clean unmount, the journal header's next unflushed pointer will be
2367  * to an invalid transaction.  This tests that before finding all the
2368  * transactions in the log, which makes normal mount times fast.
2369  *
2370  * After a crash, this starts with the next unflushed transaction, and
2371  * replays until it finds one too old, or invalid.
2372  *
2373  * On exit, it sets things up so the first transaction will work correctly.
2374  * NOTE: only called during fs mount
2375  */
2376 static int journal_read(struct super_block *sb)
2377 {
2378         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2379         struct reiserfs_journal_desc *desc;
2380         unsigned int oldest_trans_id = 0;
2381         unsigned int oldest_invalid_trans_id = 0;
2382         time64_t start;
2383         unsigned long oldest_start = 0;
2384         unsigned long cur_dblock = 0;
2385         unsigned long newest_mount_id = 9;
2386         struct buffer_head *d_bh;
2387         struct reiserfs_journal_header *jh;
2388         int valid_journal_header = 0;
2389         int replay_count = 0;
2390         int continue_replay = 1;
2391         int ret;
2392
2393         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2394         reiserfs_info(sb, "checking transaction log (%pg)\n",
2395                       journal->j_dev_bd);
2396         start = ktime_get_seconds();
2397
2398         /*
2399          * step 1, read in the journal header block.  Check the transaction
2400          * it says is the first unflushed, and if that transaction is not
2401          * valid, replay is done
2402          */
2403         journal->j_header_bh = journal_bread(sb,
2404                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2405                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2406         if (!journal->j_header_bh) {
2407                 return 1;
2408         }
2409         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2410         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2411             SB_ONDISK_JOURNAL_SIZE(sb)
2412             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2413                 oldest_start =
2414                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2415                     le32_to_cpu(jh->j_first_unflushed_offset);
2416                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2417                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2418                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2419                                "journal-1153: found in "
2420                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2421                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2422                                le32_to_cpu(jh->j_last_flush_trans_id));
2423                 valid_journal_header = 1;
2424
2425                 /*
2426                  * now, we try to read the first unflushed offset.  If it
2427                  * is not valid, there is nothing more we can do, and it
2428                  * makes no sense to read through the whole log.
2429                  */
2430                 d_bh =
2431                     journal_bread(sb,
2432                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2433                                   le32_to_cpu(jh->j_first_unflushed_offset));
2434                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2435                 if (!ret) {
2436                         continue_replay = 0;
2437                 }
2438                 brelse(d_bh);
2439                 goto start_log_replay;
2440         }
2441
2442         /*
2443          * ok, there are transactions that need to be replayed.  start
2444          * with the first log block, find all the valid transactions, and
2445          * pick out the oldest.
2446          */
2447         while (continue_replay
2448                && cur_dblock <
2449                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2450                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2451                 /*
2452                  * Note that it is required for blocksize of primary fs
2453                  * device and journal device to be the same
2454                  */
2455                 d_bh =
2456                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2457                                     sb->s_blocksize,
2458                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2459                                     SB_ONDISK_JOURNAL_SIZE(sb));
2460                 ret =
2461                     journal_transaction_is_valid(sb, d_bh,
2462                                                  &oldest_invalid_trans_id,
2463                                                  &newest_mount_id);
2464                 if (ret == 1) {
2465                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2466                         if (oldest_start == 0) {        /* init all oldest_ values */
2467                                 oldest_trans_id = get_desc_trans_id(desc);
2468                                 oldest_start = d_bh->b_blocknr;
2469                                 newest_mount_id = get_desc_mount_id(desc);
2470                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2471                                                "journal-1179: Setting "
2472                                                "oldest_start to offset %llu, trans_id %lu",
2473                                                oldest_start -
2474                                                SB_ONDISK_JOURNAL_1st_BLOCK
2475                                                (sb), oldest_trans_id);
2476                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2477                                 /* one we just read was older */
2478                                 oldest_trans_id = get_desc_trans_id(desc);
2479                                 oldest_start = d_bh->b_blocknr;
2480                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2481                                                "journal-1180: Resetting "
2482                                                "oldest_start to offset %lu, trans_id %lu",
2483                                                oldest_start -
2484                                                SB_ONDISK_JOURNAL_1st_BLOCK
2485                                                (sb), oldest_trans_id);
2486                         }
2487                         if (newest_mount_id < get_desc_mount_id(desc)) {
2488                                 newest_mount_id = get_desc_mount_id(desc);
2489                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2490                                                "journal-1299: Setting "
2491                                                "newest_mount_id to %d",
2492                                                get_desc_mount_id(desc));
2493                         }
2494                         cur_dblock += get_desc_trans_len(desc) + 2;
2495                 } else {
2496                         cur_dblock++;
2497                 }
2498                 brelse(d_bh);
2499         }
2500
2501 start_log_replay:
2502         cur_dblock = oldest_start;
2503         if (oldest_trans_id) {
2504                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2505                                "journal-1206: Starting replay "
2506                                "from offset %llu, trans_id %lu",
2507                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2508                                oldest_trans_id);
2509
2510         }
2511         replay_count = 0;
2512         while (continue_replay && oldest_trans_id > 0) {
2513                 ret =
2514                     journal_read_transaction(sb, cur_dblock, oldest_start,
2515                                              oldest_trans_id, newest_mount_id);
2516                 if (ret < 0) {
2517                         return ret;
2518                 } else if (ret != 0) {
2519                         break;
2520                 }
2521                 cur_dblock =
2522                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2523                 replay_count++;
2524                 if (cur_dblock == oldest_start)
2525                         break;
2526         }
2527
2528         if (oldest_trans_id == 0) {
2529                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2530                                "journal-1225: No valid " "transactions found");
2531         }
2532         /*
2533          * j_start does not get set correctly if we don't replay any
2534          * transactions.  if we had a valid journal_header, set j_start
2535          * to the first unflushed transaction value, copy the trans_id
2536          * from the header
2537          */
2538         if (valid_journal_header && replay_count == 0) {
2539                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2540                 journal->j_trans_id =
2541                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2542                 /* check for trans_id overflow */
2543                 if (journal->j_trans_id == 0)
2544                         journal->j_trans_id = 10;
2545                 journal->j_last_flush_trans_id =
2546                     le32_to_cpu(jh->j_last_flush_trans_id);
2547                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2548         } else {
2549                 journal->j_mount_id = newest_mount_id + 1;
2550         }
2551         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2552                        "newest_mount_id to %lu", journal->j_mount_id);
2553         journal->j_first_unflushed_offset = journal->j_start;
2554         if (replay_count > 0) {
2555                 reiserfs_info(sb,
2556                               "replayed %d transactions in %lu seconds\n",
2557                               replay_count, ktime_get_seconds() - start);
2558         }
2559         /* needed to satisfy the locking in _update_journal_header_block */
2560         reiserfs_write_lock(sb);
2561         if (!bdev_read_only(sb->s_bdev) &&
2562             _update_journal_header_block(sb, journal->j_start,
2563                                          journal->j_last_flush_trans_id)) {
2564                 reiserfs_write_unlock(sb);
2565                 /*
2566                  * replay failed, caller must call free_journal_ram and abort
2567                  * the mount
2568                  */
2569                 return -1;
2570         }
2571         reiserfs_write_unlock(sb);
2572         return 0;
2573 }
2574
2575 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2576 {
2577         struct reiserfs_journal_list *jl;
2578         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2579                      GFP_NOFS | __GFP_NOFAIL);
2580         INIT_LIST_HEAD(&jl->j_list);
2581         INIT_LIST_HEAD(&jl->j_working_list);
2582         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2583         INIT_LIST_HEAD(&jl->j_bh_list);
2584         mutex_init(&jl->j_commit_mutex);
2585         SB_JOURNAL(s)->j_num_lists++;
2586         get_journal_list(jl);
2587         return jl;
2588 }
2589
2590 static void journal_list_init(struct super_block *sb)
2591 {
2592         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2593 }
2594
2595 static void release_journal_dev(struct super_block *super,
2596                                struct reiserfs_journal *journal)
2597 {
2598         if (journal->j_dev_bd != NULL) {
2599                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2600                 journal->j_dev_bd = NULL;
2601         }
2602 }
2603
2604 static int journal_init_dev(struct super_block *super,
2605                             struct reiserfs_journal *journal,
2606                             const char *jdev_name)
2607 {
2608         int result;
2609         dev_t jdev;
2610         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2611         char b[BDEVNAME_SIZE];
2612
2613         result = 0;
2614
2615         journal->j_dev_bd = NULL;
2616         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2617             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2618
2619         if (bdev_read_only(super->s_bdev))
2620                 blkdev_mode = FMODE_READ;
2621
2622         /* there is no "jdev" option and journal is on separate device */
2623         if ((!jdev_name || !jdev_name[0])) {
2624                 if (jdev == super->s_dev)
2625                         blkdev_mode &= ~FMODE_EXCL;
2626                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2627                                                       journal);
2628                 journal->j_dev_mode = blkdev_mode;
2629                 if (IS_ERR(journal->j_dev_bd)) {
2630                         result = PTR_ERR(journal->j_dev_bd);
2631                         journal->j_dev_bd = NULL;
2632                         reiserfs_warning(super, "sh-458",
2633                                          "cannot init journal device '%s': %i",
2634                                          __bdevname(jdev, b), result);
2635                         return result;
2636                 } else if (jdev != super->s_dev)
2637                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2638
2639                 return 0;
2640         }
2641
2642         journal->j_dev_mode = blkdev_mode;
2643         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2644         if (IS_ERR(journal->j_dev_bd)) {
2645                 result = PTR_ERR(journal->j_dev_bd);
2646                 journal->j_dev_bd = NULL;
2647                 reiserfs_warning(super, "sh-457",
2648                                  "journal_init_dev: Cannot open '%s': %i",
2649                                  jdev_name, result);
2650                 return result;
2651         }
2652
2653         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2654         reiserfs_info(super,
2655                       "journal_init_dev: journal device: %pg\n",
2656                       journal->j_dev_bd);
2657         return 0;
2658 }
2659
2660 /*
2661  * When creating/tuning a file system user can assign some
2662  * journal params within boundaries which depend on the ratio
2663  * blocksize/standard_blocksize.
2664  *
2665  * For blocks >= standard_blocksize transaction size should
2666  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2667  * then JOURNAL_TRANS_MAX_DEFAULT.
2668  *
2669  * For blocks < standard_blocksize these boundaries should be
2670  * decreased proportionally.
2671  */
2672 #define REISERFS_STANDARD_BLKSIZE (4096)
2673
2674 static int check_advise_trans_params(struct super_block *sb,
2675                                      struct reiserfs_journal *journal)
2676 {
2677         if (journal->j_trans_max) {
2678                 /* Non-default journal params.  Do sanity check for them. */
2679                 int ratio = 1;
2680                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2681                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2682
2683                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2684                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2685                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2686                     JOURNAL_MIN_RATIO) {
2687                         reiserfs_warning(sb, "sh-462",
2688                                          "bad transaction max size (%u). "
2689                                          "FSCK?", journal->j_trans_max);
2690                         return 1;
2691                 }
2692                 if (journal->j_max_batch != (journal->j_trans_max) *
2693                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2694                         reiserfs_warning(sb, "sh-463",
2695                                          "bad transaction max batch (%u). "
2696                                          "FSCK?", journal->j_max_batch);
2697                         return 1;
2698                 }
2699         } else {
2700                 /*
2701                  * Default journal params.
2702                  * The file system was created by old version
2703                  * of mkreiserfs, so some fields contain zeros,
2704                  * and we need to advise proper values for them
2705                  */
2706                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2707                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2708                                          sb->s_blocksize);
2709                         return 1;
2710                 }
2711                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2712                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2713                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2714         }
2715         return 0;
2716 }
2717
2718 /* must be called once on fs mount.  calls journal_read for you */
2719 int journal_init(struct super_block *sb, const char *j_dev_name,
2720                  int old_format, unsigned int commit_max_age)
2721 {
2722         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2723         struct buffer_head *bhjh;
2724         struct reiserfs_super_block *rs;
2725         struct reiserfs_journal_header *jh;
2726         struct reiserfs_journal *journal;
2727         struct reiserfs_journal_list *jl;
2728         int ret;
2729
2730         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2731         if (!journal) {
2732                 reiserfs_warning(sb, "journal-1256",
2733                                  "unable to get memory for journal structure");
2734                 return 1;
2735         }
2736         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2737         INIT_LIST_HEAD(&journal->j_prealloc_list);
2738         INIT_LIST_HEAD(&journal->j_working_list);
2739         INIT_LIST_HEAD(&journal->j_journal_list);
2740         journal->j_persistent_trans = 0;
2741         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2742                                            reiserfs_bmap_count(sb)))
2743                 goto free_and_return;
2744
2745         allocate_bitmap_nodes(sb);
2746
2747         /* reserved for journal area support */
2748         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2749                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2750                                                  / sb->s_blocksize +
2751                                                  reiserfs_bmap_count(sb) +
2752                                                  1 :
2753                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2754                                                  sb->s_blocksize + 2);
2755
2756         /*
2757          * Sanity check to see is the standard journal fitting
2758          * within first bitmap (actual for small blocksizes)
2759          */
2760         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2761             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2762              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2763                 reiserfs_warning(sb, "journal-1393",
2764                                  "journal does not fit for area addressed "
2765                                  "by first of bitmap blocks. It starts at "
2766                                  "%u and its size is %u. Block size %ld",
2767                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2768                                  SB_ONDISK_JOURNAL_SIZE(sb),
2769                                  sb->s_blocksize);
2770                 goto free_and_return;
2771         }
2772
2773         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2774                 reiserfs_warning(sb, "sh-462",
2775                                  "unable to initialize journal device");
2776                 goto free_and_return;
2777         }
2778
2779         rs = SB_DISK_SUPER_BLOCK(sb);
2780
2781         /* read journal header */
2782         bhjh = journal_bread(sb,
2783                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2784                              SB_ONDISK_JOURNAL_SIZE(sb));
2785         if (!bhjh) {
2786                 reiserfs_warning(sb, "sh-459",
2787                                  "unable to read journal header");
2788                 goto free_and_return;
2789         }
2790         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2791
2792         /* make sure that journal matches to the super block */
2793         if (is_reiserfs_jr(rs)
2794             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2795                 sb_jp_journal_magic(rs))) {
2796                 reiserfs_warning(sb, "sh-460",
2797                                  "journal header magic %x (device %pg) does "
2798                                  "not match to magic found in super block %x",
2799                                  jh->jh_journal.jp_journal_magic,
2800                                  journal->j_dev_bd,
2801                                  sb_jp_journal_magic(rs));
2802                 brelse(bhjh);
2803                 goto free_and_return;
2804         }
2805
2806         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2807         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2808         journal->j_max_commit_age =
2809             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2810         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2811
2812         if (check_advise_trans_params(sb, journal) != 0)
2813                 goto free_and_return;
2814         journal->j_default_max_commit_age = journal->j_max_commit_age;
2815
2816         if (commit_max_age != 0) {
2817                 journal->j_max_commit_age = commit_max_age;
2818                 journal->j_max_trans_age = commit_max_age;
2819         }
2820
2821         reiserfs_info(sb, "journal params: device %pg, size %u, "
2822                       "journal first block %u, max trans len %u, max batch %u, "
2823                       "max commit age %u, max trans age %u\n",
2824                       journal->j_dev_bd,
2825                       SB_ONDISK_JOURNAL_SIZE(sb),
2826                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2827                       journal->j_trans_max,
2828                       journal->j_max_batch,
2829                       journal->j_max_commit_age, journal->j_max_trans_age);
2830
2831         brelse(bhjh);
2832
2833         journal->j_list_bitmap_index = 0;
2834         journal_list_init(sb);
2835
2836         memset(journal->j_list_hash_table, 0,
2837                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2838
2839         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2840         spin_lock_init(&journal->j_dirty_buffers_lock);
2841
2842         journal->j_start = 0;
2843         journal->j_len = 0;
2844         journal->j_len_alloc = 0;
2845         atomic_set(&journal->j_wcount, 0);
2846         atomic_set(&journal->j_async_throttle, 0);
2847         journal->j_bcount = 0;
2848         journal->j_trans_start_time = 0;
2849         journal->j_last = NULL;
2850         journal->j_first = NULL;
2851         init_waitqueue_head(&journal->j_join_wait);
2852         mutex_init(&journal->j_mutex);
2853         mutex_init(&journal->j_flush_mutex);
2854
2855         journal->j_trans_id = 10;
2856         journal->j_mount_id = 10;
2857         journal->j_state = 0;
2858         atomic_set(&journal->j_jlock, 0);
2859         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2860         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2861         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2862         journal->j_cnode_used = 0;
2863         journal->j_must_wait = 0;
2864
2865         if (journal->j_cnode_free == 0) {
2866                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2867                                  "allocation failed (%ld bytes). Journal is "
2868                                  "too large for available memory. Usually "
2869                                  "this is due to a journal that is too large.",
2870                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2871                 goto free_and_return;
2872         }
2873
2874         init_journal_hash(sb);
2875         jl = journal->j_current_jl;
2876
2877         /*
2878          * get_list_bitmap() may call flush_commit_list() which
2879          * requires the lock. Calling flush_commit_list() shouldn't happen
2880          * this early but I like to be paranoid.
2881          */
2882         reiserfs_write_lock(sb);
2883         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2884         reiserfs_write_unlock(sb);
2885         if (!jl->j_list_bitmap) {
2886                 reiserfs_warning(sb, "journal-2005",
2887                                  "get_list_bitmap failed for journal list 0");
2888                 goto free_and_return;
2889         }
2890
2891         ret = journal_read(sb);
2892         if (ret < 0) {
2893                 reiserfs_warning(sb, "reiserfs-2006",
2894                                  "Replay Failure, unable to mount");
2895                 goto free_and_return;
2896         }
2897
2898         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2899         journal->j_work_sb = sb;
2900         return 0;
2901 free_and_return:
2902         free_journal_ram(sb);
2903         return 1;
2904 }
2905
2906 /*
2907  * test for a polite end of the current transaction.  Used by file_write,
2908  * and should be used by delete to make sure they don't write more than
2909  * can fit inside a single transaction
2910  */
2911 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2912                                    int new_alloc)
2913 {
2914         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2915         time64_t now = ktime_get_seconds();
2916         /* cannot restart while nested */
2917         BUG_ON(!th->t_trans_id);
2918         if (th->t_refcount > 1)
2919                 return 0;
2920         if (journal->j_must_wait > 0 ||
2921             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2922             atomic_read(&journal->j_jlock) ||
2923             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2924             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2925                 return 1;
2926         }
2927
2928         journal->j_len_alloc += new_alloc;
2929         th->t_blocks_allocated += new_alloc ;
2930         return 0;
2931 }
2932
2933 /* this must be called inside a transaction */
2934 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2935 {
2936         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2937         BUG_ON(!th->t_trans_id);
2938         journal->j_must_wait = 1;
2939         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2940         return;
2941 }
2942
2943 /* this must be called without a transaction started */
2944 void reiserfs_allow_writes(struct super_block *s)
2945 {
2946         struct reiserfs_journal *journal = SB_JOURNAL(s);
2947         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2948         wake_up(&journal->j_join_wait);
2949 }
2950
2951 /* this must be called without a transaction started */
2952 void reiserfs_wait_on_write_block(struct super_block *s)
2953 {
2954         struct reiserfs_journal *journal = SB_JOURNAL(s);
2955         wait_event(journal->j_join_wait,
2956                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2957 }
2958
2959 static void queue_log_writer(struct super_block *s)
2960 {
2961         wait_queue_entry_t wait;
2962         struct reiserfs_journal *journal = SB_JOURNAL(s);
2963         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2964
2965         /*
2966          * we don't want to use wait_event here because
2967          * we only want to wait once.
2968          */
2969         init_waitqueue_entry(&wait, current);
2970         add_wait_queue(&journal->j_join_wait, &wait);
2971         set_current_state(TASK_UNINTERRUPTIBLE);
2972         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2973                 int depth = reiserfs_write_unlock_nested(s);
2974                 schedule();
2975                 reiserfs_write_lock_nested(s, depth);
2976         }
2977         __set_current_state(TASK_RUNNING);
2978         remove_wait_queue(&journal->j_join_wait, &wait);
2979 }
2980
2981 static void wake_queued_writers(struct super_block *s)
2982 {
2983         struct reiserfs_journal *journal = SB_JOURNAL(s);
2984         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2985                 wake_up(&journal->j_join_wait);
2986 }
2987
2988 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2989 {
2990         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2991         unsigned long bcount = journal->j_bcount;
2992         while (1) {
2993                 int depth;
2994
2995                 depth = reiserfs_write_unlock_nested(sb);
2996                 schedule_timeout_uninterruptible(1);
2997                 reiserfs_write_lock_nested(sb, depth);
2998
2999                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3000                 while ((atomic_read(&journal->j_wcount) > 0 ||
3001                         atomic_read(&journal->j_jlock)) &&
3002                        journal->j_trans_id == trans_id) {
3003                         queue_log_writer(sb);
3004                 }
3005                 if (journal->j_trans_id != trans_id)
3006                         break;
3007                 if (bcount == journal->j_bcount)
3008                         break;
3009                 bcount = journal->j_bcount;
3010         }
3011 }
3012
3013 /*
3014  * join == true if you must join an existing transaction.
3015  * join == false if you can deal with waiting for others to finish
3016  *
3017  * this will block until the transaction is joinable.  send the number of
3018  * blocks you expect to use in nblocks.
3019 */
3020 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3021                               struct super_block *sb, unsigned long nblocks,
3022                               int join)
3023 {
3024         time64_t now = ktime_get_seconds();
3025         unsigned int old_trans_id;
3026         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3027         struct reiserfs_transaction_handle myth;
3028         int sched_count = 0;
3029         int retval;
3030         int depth;
3031
3032         reiserfs_check_lock_depth(sb, "journal_begin");
3033         BUG_ON(nblocks > journal->j_trans_max);
3034
3035         PROC_INFO_INC(sb, journal.journal_being);
3036         /* set here for journal_join */
3037         th->t_refcount = 1;
3038         th->t_super = sb;
3039
3040 relock:
3041         lock_journal(sb);
3042         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3043                 unlock_journal(sb);
3044                 retval = journal->j_errno;
3045                 goto out_fail;
3046         }
3047         journal->j_bcount++;
3048
3049         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3050                 unlock_journal(sb);
3051                 depth = reiserfs_write_unlock_nested(sb);
3052                 reiserfs_wait_on_write_block(sb);
3053                 reiserfs_write_lock_nested(sb, depth);
3054                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3055                 goto relock;
3056         }
3057         now = ktime_get_seconds();
3058
3059         /*
3060          * if there is no room in the journal OR
3061          * if this transaction is too old, and we weren't called joinable,
3062          * wait for it to finish before beginning we don't sleep if there
3063          * aren't other writers
3064          */
3065
3066         if ((!join && journal->j_must_wait > 0) ||
3067             (!join
3068              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3069             || (!join && atomic_read(&journal->j_wcount) > 0
3070                 && journal->j_trans_start_time > 0
3071                 && (now - journal->j_trans_start_time) >
3072                 journal->j_max_trans_age) || (!join
3073                                               && atomic_read(&journal->j_jlock))
3074             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3075
3076                 old_trans_id = journal->j_trans_id;
3077                 /* allow others to finish this transaction */
3078                 unlock_journal(sb);
3079
3080                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3081                     journal->j_max_batch &&
3082                     ((journal->j_len + nblocks + 2) * 100) <
3083                     (journal->j_len_alloc * 75)) {
3084                         if (atomic_read(&journal->j_wcount) > 10) {
3085                                 sched_count++;
3086                                 queue_log_writer(sb);
3087                                 goto relock;
3088                         }
3089                 }
3090                 /*
3091                  * don't mess with joining the transaction if all we
3092                  * have to do is wait for someone else to do a commit
3093                  */
3094                 if (atomic_read(&journal->j_jlock)) {
3095                         while (journal->j_trans_id == old_trans_id &&
3096                                atomic_read(&journal->j_jlock)) {
3097                                 queue_log_writer(sb);
3098                         }
3099                         goto relock;
3100                 }
3101                 retval = journal_join(&myth, sb);
3102                 if (retval)
3103                         goto out_fail;
3104
3105                 /* someone might have ended the transaction while we joined */
3106                 if (old_trans_id != journal->j_trans_id) {
3107                         retval = do_journal_end(&myth, 0);
3108                 } else {
3109                         retval = do_journal_end(&myth, COMMIT_NOW);
3110                 }
3111
3112                 if (retval)
3113                         goto out_fail;
3114
3115                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3116                 goto relock;
3117         }
3118         /* we are the first writer, set trans_id */
3119         if (journal->j_trans_start_time == 0) {
3120                 journal->j_trans_start_time = ktime_get_seconds();
3121         }
3122         atomic_inc(&journal->j_wcount);
3123         journal->j_len_alloc += nblocks;
3124         th->t_blocks_logged = 0;
3125         th->t_blocks_allocated = nblocks;
3126         th->t_trans_id = journal->j_trans_id;
3127         unlock_journal(sb);
3128         INIT_LIST_HEAD(&th->t_list);
3129         return 0;
3130
3131 out_fail:
3132         memset(th, 0, sizeof(*th));
3133         /*
3134          * Re-set th->t_super, so we can properly keep track of how many
3135          * persistent transactions there are. We need to do this so if this
3136          * call is part of a failed restart_transaction, we can free it later
3137          */
3138         th->t_super = sb;
3139         return retval;
3140 }
3141
3142 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3143                                                                     super_block
3144                                                                     *s,
3145                                                                     int nblocks)
3146 {
3147         int ret;
3148         struct reiserfs_transaction_handle *th;
3149
3150         /*
3151          * if we're nesting into an existing transaction.  It will be
3152          * persistent on its own
3153          */
3154         if (reiserfs_transaction_running(s)) {
3155                 th = current->journal_info;
3156                 th->t_refcount++;
3157                 BUG_ON(th->t_refcount < 2);
3158
3159                 return th;
3160         }
3161         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3162         if (!th)
3163                 return NULL;
3164         ret = journal_begin(th, s, nblocks);
3165         if (ret) {
3166                 kfree(th);
3167                 return NULL;
3168         }
3169
3170         SB_JOURNAL(s)->j_persistent_trans++;
3171         return th;
3172 }
3173
3174 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3175 {
3176         struct super_block *s = th->t_super;
3177         int ret = 0;
3178         if (th->t_trans_id)
3179                 ret = journal_end(th);
3180         else
3181                 ret = -EIO;
3182         if (th->t_refcount == 0) {
3183                 SB_JOURNAL(s)->j_persistent_trans--;
3184                 kfree(th);
3185         }
3186         return ret;
3187 }
3188
3189 static int journal_join(struct reiserfs_transaction_handle *th,
3190                         struct super_block *sb)
3191 {
3192         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3193
3194         /*
3195          * this keeps do_journal_end from NULLing out the
3196          * current->journal_info pointer
3197          */
3198         th->t_handle_save = cur_th;
3199         BUG_ON(cur_th && cur_th->t_refcount > 1);
3200         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3201 }
3202
3203 int journal_join_abort(struct reiserfs_transaction_handle *th,
3204                        struct super_block *sb)
3205 {
3206         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3207
3208         /*
3209          * this keeps do_journal_end from NULLing out the
3210          * current->journal_info pointer
3211          */
3212         th->t_handle_save = cur_th;
3213         BUG_ON(cur_th && cur_th->t_refcount > 1);
3214         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3215 }
3216
3217 int journal_begin(struct reiserfs_transaction_handle *th,
3218                   struct super_block *sb, unsigned long nblocks)
3219 {
3220         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3221         int ret;
3222
3223         th->t_handle_save = NULL;
3224         if (cur_th) {
3225                 /* we are nesting into the current transaction */
3226                 if (cur_th->t_super == sb) {
3227                         BUG_ON(!cur_th->t_refcount);
3228                         cur_th->t_refcount++;
3229                         memcpy(th, cur_th, sizeof(*th));
3230                         if (th->t_refcount <= 1)
3231                                 reiserfs_warning(sb, "reiserfs-2005",
3232                                                  "BAD: refcount <= 1, but "
3233                                                  "journal_info != 0");
3234                         return 0;
3235                 } else {
3236                         /*
3237                          * we've ended up with a handle from a different
3238                          * filesystem.  save it and restore on journal_end.
3239                          * This should never really happen...
3240                          */
3241                         reiserfs_warning(sb, "clm-2100",
3242                                          "nesting info a different FS");
3243                         th->t_handle_save = current->journal_info;
3244                         current->journal_info = th;
3245                 }
3246         } else {
3247                 current->journal_info = th;
3248         }
3249         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3250         BUG_ON(current->journal_info != th);
3251
3252         /*
3253          * I guess this boils down to being the reciprocal of clm-2100 above.
3254          * If do_journal_begin_r fails, we need to put it back, since
3255          * journal_end won't be called to do it. */
3256         if (ret)
3257                 current->journal_info = th->t_handle_save;
3258         else
3259                 BUG_ON(!th->t_refcount);
3260
3261         return ret;
3262 }
3263
3264 /*
3265  * puts bh into the current transaction.  If it was already there, reorders
3266  * removes the old pointers from the hash, and puts new ones in (to make
3267  * sure replay happen in the right order).
3268  *
3269  * if it was dirty, cleans and files onto the clean list.  I can't let it
3270  * be dirty again until the transaction is committed.
3271  *
3272  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3273  */
3274 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3275                        struct buffer_head *bh)
3276 {
3277         struct super_block *sb = th->t_super;
3278         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3279         struct reiserfs_journal_cnode *cn = NULL;
3280         int count_already_incd = 0;
3281         int prepared = 0;
3282         BUG_ON(!th->t_trans_id);
3283
3284         PROC_INFO_INC(sb, journal.mark_dirty);
3285         if (th->t_trans_id != journal->j_trans_id) {
3286                 reiserfs_panic(th->t_super, "journal-1577",
3287                                "handle trans id %ld != current trans id %ld",
3288                                th->t_trans_id, journal->j_trans_id);
3289         }
3290
3291         prepared = test_clear_buffer_journal_prepared(bh);
3292         clear_buffer_journal_restore_dirty(bh);
3293         /* already in this transaction, we are done */
3294         if (buffer_journaled(bh)) {
3295                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3296                 return 0;
3297         }
3298
3299         /*
3300          * this must be turned into a panic instead of a warning.  We can't
3301          * allow a dirty or journal_dirty or locked buffer to be logged, as
3302          * some changes could get to disk too early.  NOT GOOD.
3303          */
3304         if (!prepared || buffer_dirty(bh)) {
3305                 reiserfs_warning(sb, "journal-1777",
3306                                  "buffer %llu bad state "
3307                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3308                                  (unsigned long long)bh->b_blocknr,
3309                                  prepared ? ' ' : '!',
3310                                  buffer_locked(bh) ? ' ' : '!',
3311                                  buffer_dirty(bh) ? ' ' : '!',
3312                                  buffer_journal_dirty(bh) ? ' ' : '!');
3313         }
3314
3315         if (atomic_read(&journal->j_wcount) <= 0) {
3316                 reiserfs_warning(sb, "journal-1409",
3317                                  "returning because j_wcount was %d",
3318                                  atomic_read(&journal->j_wcount));
3319                 return 1;
3320         }
3321         /*
3322          * this error means I've screwed up, and we've overflowed
3323          * the transaction.  Nothing can be done here, except make the
3324          * FS readonly or panic.
3325          */
3326         if (journal->j_len >= journal->j_trans_max) {
3327                 reiserfs_panic(th->t_super, "journal-1413",
3328                                "j_len (%lu) is too big",
3329                                journal->j_len);
3330         }
3331
3332         if (buffer_journal_dirty(bh)) {
3333                 count_already_incd = 1;
3334                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3335                 clear_buffer_journal_dirty(bh);
3336         }
3337
3338         if (journal->j_len > journal->j_len_alloc) {
3339                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3340         }
3341
3342         set_buffer_journaled(bh);
3343
3344         /* now put this guy on the end */
3345         if (!cn) {
3346                 cn = get_cnode(sb);
3347                 if (!cn) {
3348                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3349                 }
3350
3351                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3352                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3353                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3354                 }
3355                 th->t_blocks_logged++;
3356                 journal->j_len++;
3357
3358                 cn->bh = bh;
3359                 cn->blocknr = bh->b_blocknr;
3360                 cn->sb = sb;
3361                 cn->jlist = NULL;
3362                 insert_journal_hash(journal->j_hash_table, cn);
3363                 if (!count_already_incd) {
3364                         get_bh(bh);
3365                 }
3366         }
3367         cn->next = NULL;
3368         cn->prev = journal->j_last;
3369         cn->bh = bh;
3370         if (journal->j_last) {
3371                 journal->j_last->next = cn;
3372                 journal->j_last = cn;
3373         } else {
3374                 journal->j_first = cn;
3375                 journal->j_last = cn;
3376         }
3377         reiserfs_schedule_old_flush(sb);
3378         return 0;
3379 }
3380
3381 int journal_end(struct reiserfs_transaction_handle *th)
3382 {
3383         struct super_block *sb = th->t_super;
3384         if (!current->journal_info && th->t_refcount > 1)
3385                 reiserfs_warning(sb, "REISER-NESTING",
3386                                  "th NULL, refcount %d", th->t_refcount);
3387
3388         if (!th->t_trans_id) {
3389                 WARN_ON(1);
3390                 return -EIO;
3391         }
3392
3393         th->t_refcount--;
3394         if (th->t_refcount > 0) {
3395                 struct reiserfs_transaction_handle *cur_th =
3396                     current->journal_info;
3397
3398                 /*
3399                  * we aren't allowed to close a nested transaction on a
3400                  * different filesystem from the one in the task struct
3401                  */
3402                 BUG_ON(cur_th->t_super != th->t_super);
3403
3404                 if (th != cur_th) {
3405                         memcpy(current->journal_info, th, sizeof(*th));
3406                         th->t_trans_id = 0;
3407                 }
3408                 return 0;
3409         } else {
3410                 return do_journal_end(th, 0);
3411         }
3412 }
3413
3414 /*
3415  * removes from the current transaction, relsing and descrementing any counters.
3416  * also files the removed buffer directly onto the clean list
3417  *
3418  * called by journal_mark_freed when a block has been deleted
3419  *
3420  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3421  */
3422 static int remove_from_transaction(struct super_block *sb,
3423                                    b_blocknr_t blocknr, int already_cleaned)
3424 {
3425         struct buffer_head *bh;
3426         struct reiserfs_journal_cnode *cn;
3427         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3428         int ret = 0;
3429
3430         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3431         if (!cn || !cn->bh) {
3432                 return ret;
3433         }
3434         bh = cn->bh;
3435         if (cn->prev) {
3436                 cn->prev->next = cn->next;
3437         }
3438         if (cn->next) {
3439                 cn->next->prev = cn->prev;
3440         }
3441         if (cn == journal->j_first) {
3442                 journal->j_first = cn->next;
3443         }
3444         if (cn == journal->j_last) {
3445                 journal->j_last = cn->prev;
3446         }
3447         remove_journal_hash(sb, journal->j_hash_table, NULL,
3448                             bh->b_blocknr, 0);
3449         clear_buffer_journaled(bh);     /* don't log this one */
3450
3451         if (!already_cleaned) {
3452                 clear_buffer_journal_dirty(bh);
3453                 clear_buffer_dirty(bh);
3454                 clear_buffer_journal_test(bh);
3455                 put_bh(bh);
3456                 if (atomic_read(&bh->b_count) < 0) {
3457                         reiserfs_warning(sb, "journal-1752",
3458                                          "b_count < 0");
3459                 }
3460                 ret = 1;
3461         }
3462         journal->j_len--;
3463         journal->j_len_alloc--;
3464         free_cnode(sb, cn);
3465         return ret;
3466 }
3467
3468 /*
3469  * for any cnode in a journal list, it can only be dirtied of all the
3470  * transactions that include it are committed to disk.
3471  * this checks through each transaction, and returns 1 if you are allowed
3472  * to dirty, and 0 if you aren't
3473  *
3474  * it is called by dirty_journal_list, which is called after
3475  * flush_commit_list has gotten all the log blocks for a given
3476  * transaction on disk
3477  *
3478  */
3479 static int can_dirty(struct reiserfs_journal_cnode *cn)
3480 {
3481         struct super_block *sb = cn->sb;
3482         b_blocknr_t blocknr = cn->blocknr;
3483         struct reiserfs_journal_cnode *cur = cn->hprev;
3484         int can_dirty = 1;
3485
3486         /*
3487          * first test hprev.  These are all newer than cn, so any node here
3488          * with the same block number and dev means this node can't be sent
3489          * to disk right now.
3490          */
3491         while (cur && can_dirty) {
3492                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3493                     cur->blocknr == blocknr) {
3494                         can_dirty = 0;
3495                 }
3496                 cur = cur->hprev;
3497         }
3498         /*
3499          * then test hnext.  These are all older than cn.  As long as they
3500          * are committed to the log, it is safe to write cn to disk
3501          */
3502         cur = cn->hnext;
3503         while (cur && can_dirty) {
3504                 if (cur->jlist && cur->jlist->j_len > 0 &&
3505                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3506                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3507                         can_dirty = 0;
3508                 }
3509                 cur = cur->hnext;
3510         }
3511         return can_dirty;
3512 }
3513
3514 /*
3515  * syncs the commit blocks, but does not force the real buffers to disk
3516  * will wait until the current transaction is done/committed before returning
3517  */
3518 int journal_end_sync(struct reiserfs_transaction_handle *th)
3519 {
3520         struct super_block *sb = th->t_super;
3521         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3522
3523         BUG_ON(!th->t_trans_id);
3524         /* you can sync while nested, very, very bad */
3525         BUG_ON(th->t_refcount > 1);
3526         if (journal->j_len == 0) {
3527                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3528                                              1);
3529                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3530         }
3531         return do_journal_end(th, COMMIT_NOW | WAIT);
3532 }
3533
3534 /* writeback the pending async commits to disk */
3535 static void flush_async_commits(struct work_struct *work)
3536 {
3537         struct reiserfs_journal *journal =
3538                 container_of(work, struct reiserfs_journal, j_work.work);
3539         struct super_block *sb = journal->j_work_sb;
3540         struct reiserfs_journal_list *jl;
3541         struct list_head *entry;
3542
3543         reiserfs_write_lock(sb);
3544         if (!list_empty(&journal->j_journal_list)) {
3545                 /* last entry is the youngest, commit it and you get everything */
3546                 entry = journal->j_journal_list.prev;
3547                 jl = JOURNAL_LIST_ENTRY(entry);
3548                 flush_commit_list(sb, jl, 1);
3549         }
3550         reiserfs_write_unlock(sb);
3551 }
3552
3553 /*
3554  * flushes any old transactions to disk
3555  * ends the current transaction if it is too old
3556  */
3557 void reiserfs_flush_old_commits(struct super_block *sb)
3558 {
3559         time64_t now;
3560         struct reiserfs_transaction_handle th;
3561         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3562
3563         now = ktime_get_seconds();
3564         /*
3565          * safety check so we don't flush while we are replaying the log during
3566          * mount
3567          */
3568         if (list_empty(&journal->j_journal_list))
3569                 return;
3570
3571         /*
3572          * check the current transaction.  If there are no writers, and it is
3573          * too old, finish it, and force the commit blocks to disk
3574          */
3575         if (atomic_read(&journal->j_wcount) <= 0 &&
3576             journal->j_trans_start_time > 0 &&
3577             journal->j_len > 0 &&
3578             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3579                 if (!journal_join(&th, sb)) {
3580                         reiserfs_prepare_for_journal(sb,
3581                                                      SB_BUFFER_WITH_SB(sb),
3582                                                      1);
3583                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3584
3585                         /*
3586                          * we're only being called from kreiserfsd, it makes
3587                          * no sense to do an async commit so that kreiserfsd
3588                          * can do it later
3589                          */
3590                         do_journal_end(&th, COMMIT_NOW | WAIT);
3591                 }
3592         }
3593 }
3594
3595 /*
3596  * returns 0 if do_journal_end should return right away, returns 1 if
3597  * do_journal_end should finish the commit
3598  *
3599  * if the current transaction is too old, but still has writers, this will
3600  * wait on j_join_wait until all the writers are done.  By the time it
3601  * wakes up, the transaction it was called has already ended, so it just
3602  * flushes the commit list and returns 0.
3603  *
3604  * Won't batch when flush or commit_now is set.  Also won't batch when
3605  * others are waiting on j_join_wait.
3606  *
3607  * Note, we can't allow the journal_end to proceed while there are still
3608  * writers in the log.
3609  */
3610 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3611 {
3612
3613         time64_t now;
3614         int flush = flags & FLUSH_ALL;
3615         int commit_now = flags & COMMIT_NOW;
3616         int wait_on_commit = flags & WAIT;
3617         struct reiserfs_journal_list *jl;
3618         struct super_block *sb = th->t_super;
3619         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3620
3621         BUG_ON(!th->t_trans_id);
3622
3623         if (th->t_trans_id != journal->j_trans_id) {
3624                 reiserfs_panic(th->t_super, "journal-1577",
3625                                "handle trans id %ld != current trans id %ld",
3626                                th->t_trans_id, journal->j_trans_id);
3627         }
3628
3629         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3630         /* <= 0 is allowed.  unmounting might not call begin */
3631         if (atomic_read(&journal->j_wcount) > 0)
3632                 atomic_dec(&journal->j_wcount);
3633
3634         /*
3635          * BUG, deal with case where j_len is 0, but people previously
3636          * freed blocks need to be released will be dealt with by next
3637          * transaction that actually writes something, but should be taken
3638          * care of in this trans
3639          */
3640         BUG_ON(journal->j_len == 0);
3641
3642         /*
3643          * if wcount > 0, and we are called to with flush or commit_now,
3644          * we wait on j_join_wait.  We will wake up when the last writer has
3645          * finished the transaction, and started it on its way to the disk.
3646          * Then, we flush the commit or journal list, and just return 0
3647          * because the rest of journal end was already done for this
3648          * transaction.
3649          */
3650         if (atomic_read(&journal->j_wcount) > 0) {
3651                 if (flush || commit_now) {
3652                         unsigned trans_id;
3653
3654                         jl = journal->j_current_jl;
3655                         trans_id = jl->j_trans_id;
3656                         if (wait_on_commit)
3657                                 jl->j_state |= LIST_COMMIT_PENDING;
3658                         atomic_set(&journal->j_jlock, 1);
3659                         if (flush) {
3660                                 journal->j_next_full_flush = 1;
3661                         }
3662                         unlock_journal(sb);
3663
3664                         /*
3665                          * sleep while the current transaction is
3666                          * still j_jlocked
3667                          */
3668                         while (journal->j_trans_id == trans_id) {
3669                                 if (atomic_read(&journal->j_jlock)) {
3670                                         queue_log_writer(sb);
3671                                 } else {
3672                                         lock_journal(sb);
3673                                         if (journal->j_trans_id == trans_id) {
3674                                                 atomic_set(&journal->j_jlock,
3675                                                            1);
3676                                         }
3677                                         unlock_journal(sb);
3678                                 }
3679                         }
3680                         BUG_ON(journal->j_trans_id == trans_id);
3681
3682                         if (commit_now
3683                             && journal_list_still_alive(sb, trans_id)
3684                             && wait_on_commit) {
3685                                 flush_commit_list(sb, jl, 1);
3686                         }
3687                         return 0;
3688                 }
3689                 unlock_journal(sb);
3690                 return 0;
3691         }
3692
3693         /* deal with old transactions where we are the last writers */
3694         now = ktime_get_seconds();
3695         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3696                 commit_now = 1;
3697                 journal->j_next_async_flush = 1;
3698         }
3699         /* don't batch when someone is waiting on j_join_wait */
3700         /* don't batch when syncing the commit or flushing the whole trans */
3701         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3702             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3703             && journal->j_len_alloc < journal->j_max_batch
3704             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3705                 journal->j_bcount++;
3706                 unlock_journal(sb);
3707                 return 0;
3708         }
3709
3710         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3711                 reiserfs_panic(sb, "journal-003",
3712                                "j_start (%ld) is too high",
3713                                journal->j_start);
3714         }
3715         return 1;
3716 }
3717
3718 /*
3719  * Does all the work that makes deleting blocks safe.
3720  * when deleting a block mark BH_JNew, just remove it from the current
3721  * transaction, clean it's buffer_head and move on.
3722  *
3723  * otherwise:
3724  * set a bit for the block in the journal bitmap.  That will prevent it from
3725  * being allocated for unformatted nodes before this transaction has finished.
3726  *
3727  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3728  * That will prevent any old transactions with this block from trying to flush
3729  * to the real location.  Since we aren't removing the cnode from the
3730  * journal_list_hash, *the block can't be reallocated yet.
3731  *
3732  * Then remove it from the current transaction, decrementing any counters and
3733  * filing it on the clean list.
3734  */
3735 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3736                        struct super_block *sb, b_blocknr_t blocknr)
3737 {
3738         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3739         struct reiserfs_journal_cnode *cn = NULL;
3740         struct buffer_head *bh = NULL;
3741         struct reiserfs_list_bitmap *jb = NULL;
3742         int cleaned = 0;
3743         BUG_ON(!th->t_trans_id);
3744
3745         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3746         if (cn && cn->bh) {
3747                 bh = cn->bh;
3748                 get_bh(bh);
3749         }
3750         /* if it is journal new, we just remove it from this transaction */
3751         if (bh && buffer_journal_new(bh)) {
3752                 clear_buffer_journal_new(bh);
3753                 clear_prepared_bits(bh);
3754                 reiserfs_clean_and_file_buffer(bh);
3755                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3756         } else {
3757                 /*
3758                  * set the bit for this block in the journal bitmap
3759                  * for this transaction
3760                  */
3761                 jb = journal->j_current_jl->j_list_bitmap;
3762                 if (!jb) {
3763                         reiserfs_panic(sb, "journal-1702",
3764                                        "journal_list_bitmap is NULL");
3765                 }
3766                 set_bit_in_list_bitmap(sb, blocknr, jb);
3767
3768                 /* Note, the entire while loop is not allowed to schedule.  */
3769
3770                 if (bh) {
3771                         clear_prepared_bits(bh);
3772                         reiserfs_clean_and_file_buffer(bh);
3773                 }
3774                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3775
3776                 /*
3777                  * find all older transactions with this block,
3778                  * make sure they don't try to write it out
3779                  */
3780                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3781                                           blocknr);
3782                 while (cn) {
3783                         if (sb == cn->sb && blocknr == cn->blocknr) {
3784                                 set_bit(BLOCK_FREED, &cn->state);
3785                                 if (cn->bh) {
3786                                         /*
3787                                          * remove_from_transaction will brelse
3788                                          * the buffer if it was in the current
3789                                          * trans
3790                                          */
3791                                         if (!cleaned) {
3792                                                 clear_buffer_journal_dirty(cn->
3793                                                                            bh);
3794                                                 clear_buffer_dirty(cn->bh);
3795                                                 clear_buffer_journal_test(cn->
3796                                                                           bh);
3797                                                 cleaned = 1;
3798                                                 put_bh(cn->bh);
3799                                                 if (atomic_read
3800                                                     (&cn->bh->b_count) < 0) {
3801                                                         reiserfs_warning(sb,
3802                                                                  "journal-2138",
3803                                                                  "cn->bh->b_count < 0");
3804                                                 }
3805                                         }
3806                                         /*
3807                                          * since we are clearing the bh,
3808                                          * we MUST dec nonzerolen
3809                                          */
3810                                         if (cn->jlist) {
3811                                                 atomic_dec(&cn->jlist->
3812                                                            j_nonzerolen);
3813                                         }
3814                                         cn->bh = NULL;
3815                                 }
3816                         }
3817                         cn = cn->hnext;
3818                 }
3819         }
3820
3821         if (bh)
3822                 release_buffer_page(bh); /* get_hash grabs the buffer */
3823         return 0;
3824 }
3825
3826 void reiserfs_update_inode_transaction(struct inode *inode)
3827 {
3828         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3829         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3830         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3831 }
3832
3833 /*
3834  * returns -1 on error, 0 if no commits/barriers were done and 1
3835  * if a transaction was actually committed and the barrier was done
3836  */
3837 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3838                              struct reiserfs_journal_list *jl)
3839 {
3840         struct reiserfs_transaction_handle th;
3841         struct super_block *sb = inode->i_sb;
3842         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3843         int ret = 0;
3844
3845         /*
3846          * is it from the current transaction,
3847          * or from an unknown transaction?
3848          */
3849         if (id == journal->j_trans_id) {
3850                 jl = journal->j_current_jl;
3851                 /*
3852                  * try to let other writers come in and
3853                  * grow this transaction
3854                  */
3855                 let_transaction_grow(sb, id);
3856                 if (journal->j_trans_id != id) {
3857                         goto flush_commit_only;
3858                 }
3859
3860                 ret = journal_begin(&th, sb, 1);
3861                 if (ret)
3862                         return ret;
3863
3864                 /* someone might have ended this transaction while we joined */
3865                 if (journal->j_trans_id != id) {
3866                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3867                                                      1);
3868                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3869                         ret = journal_end(&th);
3870                         goto flush_commit_only;
3871                 }
3872
3873                 ret = journal_end_sync(&th);
3874                 if (!ret)
3875                         ret = 1;
3876
3877         } else {
3878                 /*
3879                  * this gets tricky, we have to make sure the journal list in
3880                  * the inode still exists.  We know the list is still around
3881                  * if we've got a larger transaction id than the oldest list
3882                  */
3883 flush_commit_only:
3884                 if (journal_list_still_alive(inode->i_sb, id)) {
3885                         /*
3886                          * we only set ret to 1 when we know for sure
3887                          * the barrier hasn't been started yet on the commit
3888                          * block.
3889                          */
3890                         if (atomic_read(&jl->j_commit_left) > 1)
3891                                 ret = 1;
3892                         flush_commit_list(sb, jl, 1);
3893                         if (journal->j_errno)
3894                                 ret = journal->j_errno;
3895                 }
3896         }
3897         /* otherwise the list is gone, and long since committed */
3898         return ret;
3899 }
3900
3901 int reiserfs_commit_for_inode(struct inode *inode)
3902 {
3903         unsigned int id = REISERFS_I(inode)->i_trans_id;
3904         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3905
3906         /*
3907          * for the whole inode, assume unset id means it was
3908          * changed in the current transaction.  More conservative
3909          */
3910         if (!id || !jl) {
3911                 reiserfs_update_inode_transaction(inode);
3912                 id = REISERFS_I(inode)->i_trans_id;
3913                 /* jl will be updated in __commit_trans_jl */
3914         }
3915
3916         return __commit_trans_jl(inode, id, jl);
3917 }
3918
3919 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3920                                       struct buffer_head *bh)
3921 {
3922         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3923         PROC_INFO_INC(sb, journal.restore_prepared);
3924         if (!bh) {
3925                 return;
3926         }
3927         if (test_clear_buffer_journal_restore_dirty(bh) &&
3928             buffer_journal_dirty(bh)) {
3929                 struct reiserfs_journal_cnode *cn;
3930                 reiserfs_write_lock(sb);
3931                 cn = get_journal_hash_dev(sb,
3932                                           journal->j_list_hash_table,
3933                                           bh->b_blocknr);
3934                 if (cn && can_dirty(cn)) {
3935                         set_buffer_journal_test(bh);
3936                         mark_buffer_dirty(bh);
3937                 }
3938                 reiserfs_write_unlock(sb);
3939         }
3940         clear_buffer_journal_prepared(bh);
3941 }
3942
3943 extern struct tree_balance *cur_tb;
3944 /*
3945  * before we can change a metadata block, we have to make sure it won't
3946  * be written to disk while we are altering it.  So, we must:
3947  * clean it
3948  * wait on it.
3949  */
3950 int reiserfs_prepare_for_journal(struct super_block *sb,
3951                                  struct buffer_head *bh, int wait)
3952 {
3953         PROC_INFO_INC(sb, journal.prepare);
3954
3955         if (!trylock_buffer(bh)) {
3956                 if (!wait)
3957                         return 0;
3958                 lock_buffer(bh);
3959         }
3960         set_buffer_journal_prepared(bh);
3961         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3962                 clear_buffer_journal_test(bh);
3963                 set_buffer_journal_restore_dirty(bh);
3964         }
3965         unlock_buffer(bh);
3966         return 1;
3967 }
3968
3969 /*
3970  * long and ugly.  If flush, will not return until all commit
3971  * blocks and all real buffers in the trans are on disk.
3972  * If no_async, won't return until all commit blocks are on disk.
3973  *
3974  * keep reading, there are comments as you go along
3975  *
3976  * If the journal is aborted, we just clean up. Things like flushing
3977  * journal lists, etc just won't happen.
3978  */
3979 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3980 {
3981         struct super_block *sb = th->t_super;
3982         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3983         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3984         struct reiserfs_journal_cnode *last_cn = NULL;
3985         struct reiserfs_journal_desc *desc;
3986         struct reiserfs_journal_commit *commit;
3987         struct buffer_head *c_bh;       /* commit bh */
3988         struct buffer_head *d_bh;       /* desc bh */
3989         int cur_write_start = 0;        /* start index of current log write */
3990         int old_start;
3991         int i;
3992         int flush;
3993         int wait_on_commit;
3994         struct reiserfs_journal_list *jl, *temp_jl;
3995         struct list_head *entry, *safe;
3996         unsigned long jindex;
3997         unsigned int commit_trans_id;
3998         int trans_half;
3999         int depth;
4000
4001         BUG_ON(th->t_refcount > 1);
4002         BUG_ON(!th->t_trans_id);
4003         BUG_ON(!th->t_super);
4004
4005         /*
4006          * protect flush_older_commits from doing mistakes if the
4007          * transaction ID counter gets overflowed.
4008          */
4009         if (th->t_trans_id == ~0U)
4010                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4011         flush = flags & FLUSH_ALL;
4012         wait_on_commit = flags & WAIT;
4013
4014         current->journal_info = th->t_handle_save;
4015         reiserfs_check_lock_depth(sb, "journal end");
4016         if (journal->j_len == 0) {
4017                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4018                                              1);
4019                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4020         }
4021
4022         lock_journal(sb);
4023         if (journal->j_next_full_flush) {
4024                 flags |= FLUSH_ALL;
4025                 flush = 1;
4026         }
4027         if (journal->j_next_async_flush) {
4028                 flags |= COMMIT_NOW | WAIT;
4029                 wait_on_commit = 1;
4030         }
4031
4032         /*
4033          * check_journal_end locks the journal, and unlocks if it does
4034          * not return 1 it tells us if we should continue with the
4035          * journal_end, or just return
4036          */
4037         if (!check_journal_end(th, flags)) {
4038                 reiserfs_schedule_old_flush(sb);
4039                 wake_queued_writers(sb);
4040                 reiserfs_async_progress_wait(sb);
4041                 goto out;
4042         }
4043
4044         /* check_journal_end might set these, check again */
4045         if (journal->j_next_full_flush) {
4046                 flush = 1;
4047         }
4048
4049         /*
4050          * j must wait means we have to flush the log blocks, and the
4051          * real blocks for this transaction
4052          */
4053         if (journal->j_must_wait > 0) {
4054                 flush = 1;
4055         }
4056 #ifdef REISERFS_PREALLOCATE
4057         /*
4058          * quota ops might need to nest, setup the journal_info pointer
4059          * for them and raise the refcount so that it is > 0.
4060          */
4061         current->journal_info = th;
4062         th->t_refcount++;
4063
4064         /* it should not involve new blocks into the transaction */
4065         reiserfs_discard_all_prealloc(th);
4066
4067         th->t_refcount--;
4068         current->journal_info = th->t_handle_save;
4069 #endif
4070
4071         /* setup description block */
4072         d_bh =
4073             journal_getblk(sb,
4074                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4075                            journal->j_start);
4076         set_buffer_uptodate(d_bh);
4077         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4078         memset(d_bh->b_data, 0, d_bh->b_size);
4079         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4080         set_desc_trans_id(desc, journal->j_trans_id);
4081
4082         /*
4083          * setup commit block.  Don't write (keep it clean too) this one
4084          * until after everyone else is written
4085          */
4086         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4087                               ((journal->j_start + journal->j_len +
4088                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4089         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4090         memset(c_bh->b_data, 0, c_bh->b_size);
4091         set_commit_trans_id(commit, journal->j_trans_id);
4092         set_buffer_uptodate(c_bh);
4093
4094         /* init this journal list */
4095         jl = journal->j_current_jl;
4096
4097         /*
4098          * we lock the commit before doing anything because
4099          * we want to make sure nobody tries to run flush_commit_list until
4100          * the new transaction is fully setup, and we've already flushed the
4101          * ordered bh list
4102          */
4103         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4104
4105         /* save the transaction id in case we need to commit it later */
4106         commit_trans_id = jl->j_trans_id;
4107
4108         atomic_set(&jl->j_older_commits_done, 0);
4109         jl->j_trans_id = journal->j_trans_id;
4110         jl->j_timestamp = journal->j_trans_start_time;
4111         jl->j_commit_bh = c_bh;
4112         jl->j_start = journal->j_start;
4113         jl->j_len = journal->j_len;
4114         atomic_set(&jl->j_nonzerolen, journal->j_len);
4115         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4116         jl->j_realblock = NULL;
4117
4118         /*
4119          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4120          * for each real block, add it to the journal list hash,
4121          * copy into real block index array in the commit or desc block
4122          */
4123         trans_half = journal_trans_half(sb->s_blocksize);
4124         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4125                 if (buffer_journaled(cn->bh)) {
4126                         jl_cn = get_cnode(sb);
4127                         if (!jl_cn) {
4128                                 reiserfs_panic(sb, "journal-1676",
4129                                                "get_cnode returned NULL");
4130                         }
4131                         if (i == 0) {
4132                                 jl->j_realblock = jl_cn;
4133                         }
4134                         jl_cn->prev = last_cn;
4135                         jl_cn->next = NULL;
4136                         if (last_cn) {
4137                                 last_cn->next = jl_cn;
4138                         }
4139                         last_cn = jl_cn;
4140                         /*
4141                          * make sure the block we are trying to log
4142                          * is not a block of journal or reserved area
4143                          */
4144                         if (is_block_in_log_or_reserved_area
4145                             (sb, cn->bh->b_blocknr)) {
4146                                 reiserfs_panic(sb, "journal-2332",
4147                                                "Trying to log block %lu, "
4148                                                "which is a log block",
4149                                                cn->bh->b_blocknr);
4150                         }
4151                         jl_cn->blocknr = cn->bh->b_blocknr;
4152                         jl_cn->state = 0;
4153                         jl_cn->sb = sb;
4154                         jl_cn->bh = cn->bh;
4155                         jl_cn->jlist = jl;
4156                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4157                         if (i < trans_half) {
4158                                 desc->j_realblock[i] =
4159                                     cpu_to_le32(cn->bh->b_blocknr);
4160                         } else {
4161                                 commit->j_realblock[i - trans_half] =
4162                                     cpu_to_le32(cn->bh->b_blocknr);
4163                         }
4164                 } else {
4165                         i--;
4166                 }
4167         }
4168         set_desc_trans_len(desc, journal->j_len);
4169         set_desc_mount_id(desc, journal->j_mount_id);
4170         set_desc_trans_id(desc, journal->j_trans_id);
4171         set_commit_trans_len(commit, journal->j_len);
4172
4173         /*
4174          * special check in case all buffers in the journal
4175          * were marked for not logging
4176          */
4177         BUG_ON(journal->j_len == 0);
4178
4179         /*
4180          * we're about to dirty all the log blocks, mark the description block
4181          * dirty now too.  Don't mark the commit block dirty until all the
4182          * others are on disk
4183          */
4184         mark_buffer_dirty(d_bh);
4185
4186         /*
4187          * first data block is j_start + 1, so add one to
4188          * cur_write_start wherever you use it
4189          */
4190         cur_write_start = journal->j_start;
4191         cn = journal->j_first;
4192         jindex = 1;     /* start at one so we don't get the desc again */
4193         while (cn) {
4194                 clear_buffer_journal_new(cn->bh);
4195                 /* copy all the real blocks into log area.  dirty log blocks */
4196                 if (buffer_journaled(cn->bh)) {
4197                         struct buffer_head *tmp_bh;
4198                         char *addr;
4199                         struct page *page;
4200                         tmp_bh =
4201                             journal_getblk(sb,
4202                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4203                                            ((cur_write_start +
4204                                              jindex) %
4205                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4206                         set_buffer_uptodate(tmp_bh);
4207                         page = cn->bh->b_page;
4208                         addr = kmap(page);
4209                         memcpy(tmp_bh->b_data,
4210                                addr + offset_in_page(cn->bh->b_data),
4211                                cn->bh->b_size);
4212                         kunmap(page);
4213                         mark_buffer_dirty(tmp_bh);
4214                         jindex++;
4215                         set_buffer_journal_dirty(cn->bh);
4216                         clear_buffer_journaled(cn->bh);
4217                 } else {
4218                         /*
4219                          * JDirty cleared sometime during transaction.
4220                          * don't log this one
4221                          */
4222                         reiserfs_warning(sb, "journal-2048",
4223                                          "BAD, buffer in journal hash, "
4224                                          "but not JDirty!");
4225                         brelse(cn->bh);
4226                 }
4227                 next = cn->next;
4228                 free_cnode(sb, cn);
4229                 cn = next;
4230                 reiserfs_cond_resched(sb);
4231         }
4232
4233         /*
4234          * we are done with both the c_bh and d_bh, but
4235          * c_bh must be written after all other commit blocks,
4236          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4237          */
4238
4239         journal->j_current_jl = alloc_journal_list(sb);
4240
4241         /* now it is safe to insert this transaction on the main list */
4242         list_add_tail(&jl->j_list, &journal->j_journal_list);
4243         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4244         journal->j_num_work_lists++;
4245
4246         /* reset journal values for the next transaction */
4247         old_start = journal->j_start;
4248         journal->j_start =
4249             (journal->j_start + journal->j_len +
4250              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4251         atomic_set(&journal->j_wcount, 0);
4252         journal->j_bcount = 0;
4253         journal->j_last = NULL;
4254         journal->j_first = NULL;
4255         journal->j_len = 0;
4256         journal->j_trans_start_time = 0;
4257         /* check for trans_id overflow */
4258         if (++journal->j_trans_id == 0)
4259                 journal->j_trans_id = 10;
4260         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4261         journal->j_must_wait = 0;
4262         journal->j_len_alloc = 0;
4263         journal->j_next_full_flush = 0;
4264         journal->j_next_async_flush = 0;
4265         init_journal_hash(sb);
4266
4267         /*
4268          * make sure reiserfs_add_jh sees the new current_jl before we
4269          * write out the tails
4270          */
4271         smp_mb();
4272
4273         /*
4274          * tail conversion targets have to hit the disk before we end the
4275          * transaction.  Otherwise a later transaction might repack the tail
4276          * before this transaction commits, leaving the data block unflushed
4277          * and clean, if we crash before the later transaction commits, the
4278          * data block is lost.
4279          */
4280         if (!list_empty(&jl->j_tail_bh_list)) {
4281                 depth = reiserfs_write_unlock_nested(sb);
4282                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4283                                       journal, jl, &jl->j_tail_bh_list);
4284                 reiserfs_write_lock_nested(sb, depth);
4285         }
4286         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4287         mutex_unlock(&jl->j_commit_mutex);
4288
4289         /*
4290          * honor the flush wishes from the caller, simple commits can
4291          * be done outside the journal lock, they are done below
4292          *
4293          * if we don't flush the commit list right now, we put it into
4294          * the work queue so the people waiting on the async progress work
4295          * queue don't wait for this proc to flush journal lists and such.
4296          */
4297         if (flush) {
4298                 flush_commit_list(sb, jl, 1);
4299                 flush_journal_list(sb, jl, 1);
4300         } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4301                 /*
4302                  * Avoid queueing work when sb is being shut down. Transaction
4303                  * will be flushed on journal shutdown.
4304                  */
4305                 if (sb->s_flags & SB_ACTIVE)
4306                         queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4307                                            &journal->j_work, HZ / 10);
4308         }
4309
4310         /*
4311          * if the next transaction has any chance of wrapping, flush
4312          * transactions that might get overwritten.  If any journal lists
4313          * are very old flush them as well.
4314          */
4315 first_jl:
4316         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4317                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4318                 if (journal->j_start <= temp_jl->j_start) {
4319                         if ((journal->j_start + journal->j_trans_max + 1) >=
4320                             temp_jl->j_start) {
4321                                 flush_used_journal_lists(sb, temp_jl);
4322                                 goto first_jl;
4323                         } else if ((journal->j_start +
4324                                     journal->j_trans_max + 1) <
4325                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4326                                 /*
4327                                  * if we don't cross into the next
4328                                  * transaction and we don't wrap, there is
4329                                  * no way we can overlap any later transactions
4330                                  * break now
4331                                  */
4332                                 break;
4333                         }
4334                 } else if ((journal->j_start +
4335                             journal->j_trans_max + 1) >
4336                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4337                         if (((journal->j_start + journal->j_trans_max + 1) %
4338                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4339                             temp_jl->j_start) {
4340                                 flush_used_journal_lists(sb, temp_jl);
4341                                 goto first_jl;
4342                         } else {
4343                                 /*
4344                                 * we don't overlap anything from out start
4345                                 * to the end of the log, and our wrapped
4346                                 * portion doesn't overlap anything at
4347                                 * the start of the log.  We can break
4348                                 */
4349                                 break;
4350                         }
4351                 }
4352         }
4353
4354         journal->j_current_jl->j_list_bitmap =
4355             get_list_bitmap(sb, journal->j_current_jl);
4356
4357         if (!(journal->j_current_jl->j_list_bitmap)) {
4358                 reiserfs_panic(sb, "journal-1996",
4359                                "could not get a list bitmap");
4360         }
4361
4362         atomic_set(&journal->j_jlock, 0);
4363         unlock_journal(sb);
4364         /* wake up any body waiting to join. */
4365         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4366         wake_up(&journal->j_join_wait);
4367
4368         if (!flush && wait_on_commit &&
4369             journal_list_still_alive(sb, commit_trans_id)) {
4370                 flush_commit_list(sb, jl, 1);
4371         }
4372 out:
4373         reiserfs_check_lock_depth(sb, "journal end2");
4374
4375         memset(th, 0, sizeof(*th));
4376         /*
4377          * Re-set th->t_super, so we can properly keep track of how many
4378          * persistent transactions there are. We need to do this so if this
4379          * call is part of a failed restart_transaction, we can free it later
4380          */
4381         th->t_super = sb;
4382
4383         return journal->j_errno;
4384 }
4385
4386 /* Send the file system read only and refuse new transactions */
4387 void reiserfs_abort_journal(struct super_block *sb, int errno)
4388 {
4389         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4390         if (test_bit(J_ABORTED, &journal->j_state))
4391                 return;
4392
4393         if (!journal->j_errno)
4394                 journal->j_errno = errno;
4395
4396         sb->s_flags |= SB_RDONLY;
4397         set_bit(J_ABORTED, &journal->j_state);
4398
4399 #ifdef CONFIG_REISERFS_CHECK
4400         dump_stack();
4401 #endif
4402 }