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  *                      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
  60 /* must be correct to keep the desc and commit structs at 4k */
  61 #define JOURNAL_TRANS_HALF 1018
  62 #define BUFNR 64                /*read ahead */
  63
  64 /* cnode stat bits.  Move these into reiserfs_fs.h */
  65
  66 /* this block was freed, and can't be written.  */
  67 #define BLOCK_FREED 2
  68 /* this block was freed during this transaction, and can't be written */
  69 #define BLOCK_FREED_HOLDER 3
  70
  71 /* used in flush_journal_list */
  72 #define BLOCK_NEEDS_FLUSH 4
  73 #define BLOCK_DIRTIED 5
  74
  75 /* journal list state bits */
  76 #define LIST_TOUCHED 1
  77 #define LIST_DIRTY   2
  78 #define LIST_COMMIT_PENDING  4  /* someone will commit this list */
  79
  80 /* flags for do_journal_end */
  81 #define FLUSH_ALL   1           /* flush commit and real blocks */
  82 #define COMMIT_NOW  2           /* end and commit this transaction */
  83 #define WAIT        4           /* wait for the log blocks to hit the disk */
  84
  85 static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  86 static int flush_journal_list(struct super_block *s,
  87                               struct reiserfs_journal_list *jl, int flushall);
  88 static int flush_commit_list(struct super_block *s,
  89                              struct reiserfs_journal_list *jl, int flushall);
  90 static int can_dirty(struct reiserfs_journal_cnode *cn);
  91 static int journal_join(struct reiserfs_transaction_handle *th,
  92                         struct super_block *sb);
  93 static void release_journal_dev(struct super_block *super,
  94                                struct reiserfs_journal *journal);
  95 static void dirty_one_transaction(struct super_block *s,
  96                                  struct reiserfs_journal_list *jl);
  97 static void flush_async_commits(struct work_struct *work);
  98 static void queue_log_writer(struct super_block *s);
  99
 100 /* values for join in do_journal_begin_r */
 101 enum {
 102         JBEGIN_REG = 0,         /* regular journal begin */
 103         /* join the running transaction if at all possible */
 104         JBEGIN_JOIN = 1,
 105         /* called from cleanup code, ignores aborted flag */
 106         JBEGIN_ABORT = 2,
 107 };
 108
 109 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 110                               struct super_block *sb,
 111                               unsigned long nblocks, int join);
 112
 113 static void init_journal_hash(struct super_block *sb)
 114 {
 115         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 116         memset(journal->j_hash_table, 0,
 117                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 118 }
 119
 120 /*
 121  * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 122  * I can't allow refile_buffer to make schedule happen after I've freed a
 123  * block.  Look at remove_from_transaction and journal_mark_freed for
 124  * more details.
 125  */
 126 static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 127 {
 128         if (bh) {
 129                 clear_buffer_dirty(bh);
 130                 clear_buffer_journal_test(bh);
 131         }
 132         return 0;
 133 }
 134
 135 static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 136                                                          *sb)
 137 {
 138         struct reiserfs_bitmap_node *bn;
 139         static int id;
 140
 141         bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 142         if (!bn) {
 143                 return NULL;
 144         }
 145         bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 146         if (!bn->data) {
 147                 kfree(bn);
 148                 return NULL;
 149         }
 150         bn->id = id++;
 151         INIT_LIST_HEAD(&bn->list);
 152         return bn;
 153 }
 154
 155 static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 156 {
 157         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 158         struct reiserfs_bitmap_node *bn = NULL;
 159         struct list_head *entry = journal->j_bitmap_nodes.next;
 160
 161         journal->j_used_bitmap_nodes++;
 162 repeat:
 163
 164         if (entry != &journal->j_bitmap_nodes) {
 165                 bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 166                 list_del(entry);
 167                 memset(bn->data, 0, sb->s_blocksize);
 168                 journal->j_free_bitmap_nodes--;
 169                 return bn;
 170         }
 171         bn = allocate_bitmap_node(sb);
 172         if (!bn) {
 173                 yield();
 174                 goto repeat;
 175         }
 176         return bn;
 177 }
 178 static inline void free_bitmap_node(struct super_block *sb,
 179                                     struct reiserfs_bitmap_node *bn)
 180 {
 181         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 182         journal->j_used_bitmap_nodes--;
 183         if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 184                 kfree(bn->data);
 185                 kfree(bn);
 186         } else {
 187                 list_add(&bn->list, &journal->j_bitmap_nodes);
 188                 journal->j_free_bitmap_nodes++;
 189         }
 190 }
 191
 192 static void allocate_bitmap_nodes(struct super_block *sb)
 193 {
 194         int i;
 195         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 196         struct reiserfs_bitmap_node *bn = NULL;
 197         for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 198                 bn = allocate_bitmap_node(sb);
 199                 if (bn) {
 200                         list_add(&bn->list, &journal->j_bitmap_nodes);
 201                         journal->j_free_bitmap_nodes++;
 202                 } else {
 203                         /* this is ok, we'll try again when more are needed */
 204                         break;
 205                 }
 206         }
 207 }
 208
 209 static int set_bit_in_list_bitmap(struct super_block *sb,
 210                                   b_blocknr_t block,
 211                                   struct reiserfs_list_bitmap *jb)
 212 {
 213         unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 214         unsigned int bit_nr = block % (sb->s_blocksize << 3);
 215
 216         if (!jb->bitmaps[bmap_nr]) {
 217                 jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 218         }
 219         set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 220         return 0;
 221 }
 222
 223 static void cleanup_bitmap_list(struct super_block *sb,
 224                                 struct reiserfs_list_bitmap *jb)
 225 {
 226         int i;
 227         if (jb->bitmaps == NULL)
 228                 return;
 229
 230         for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 231                 if (jb->bitmaps[i]) {
 232                         free_bitmap_node(sb, jb->bitmaps[i]);
 233                         jb->bitmaps[i] = NULL;
 234                 }
 235         }
 236 }
 237
 238 /*
 239  * only call this on FS unmount.
 240  */
 241 static int free_list_bitmaps(struct super_block *sb,
 242                              struct reiserfs_list_bitmap *jb_array)
 243 {
 244         int i;
 245         struct reiserfs_list_bitmap *jb;
 246         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 247                 jb = jb_array + i;
 248                 jb->journal_list = NULL;
 249                 cleanup_bitmap_list(sb, jb);
 250                 vfree(jb->bitmaps);
 251                 jb->bitmaps = NULL;
 252         }
 253         return 0;
 254 }
 255
 256 static int free_bitmap_nodes(struct super_block *sb)
 257 {
 258         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 259         struct list_head *next = journal->j_bitmap_nodes.next;
 260         struct reiserfs_bitmap_node *bn;
 261
 262         while (next != &journal->j_bitmap_nodes) {
 263                 bn = list_entry(next, struct reiserfs_bitmap_node, list);
 264                 list_del(next);
 265                 kfree(bn->data);
 266                 kfree(bn);
 267                 next = journal->j_bitmap_nodes.next;
 268                 journal->j_free_bitmap_nodes--;
 269         }
 270
 271         return 0;
 272 }
 273
 274 /*
 275  * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 276  * jb_array is the array to be filled in.
 277  */
 278 int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 279                                    struct reiserfs_list_bitmap *jb_array,
 280                                    unsigned int bmap_nr)
 281 {
 282         int i;
 283         int failed = 0;
 284         struct reiserfs_list_bitmap *jb;
 285         int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 286
 287         for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 288                 jb = jb_array + i;
 289                 jb->journal_list = NULL;
 290                 jb->bitmaps = vzalloc(mem);
 291                 if (!jb->bitmaps) {
 292                         reiserfs_warning(sb, "clm-2000", "unable to "
 293                                          "allocate bitmaps for journal lists");
 294                         failed = 1;
 295                         break;
 296                 }
 297         }
 298         if (failed) {
 299                 free_list_bitmaps(sb, jb_array);
 300                 return -1;
 301         }
 302         return 0;
 303 }
 304
 305 /*
 306  * find an available list bitmap.  If you can't find one, flush a commit list
 307  * and try again
 308  */
 309 static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 310                                                     struct reiserfs_journal_list
 311                                                     *jl)
 312 {
 313         int i, j;
 314         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 315         struct reiserfs_list_bitmap *jb = NULL;
 316
 317         for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 318                 i = journal->j_list_bitmap_index;
 319                 journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 320                 jb = journal->j_list_bitmap + i;
 321                 if (journal->j_list_bitmap[i].journal_list) {
 322                         flush_commit_list(sb,
 323                                           journal->j_list_bitmap[i].
 324                                           journal_list, 1);
 325                         if (!journal->j_list_bitmap[i].journal_list) {
 326                                 break;
 327                         }
 328                 } else {
 329                         break;
 330                 }
 331         }
 332         /* double check to make sure if flushed correctly */
 333         if (jb->journal_list)
 334                 return NULL;
 335         jb->journal_list = jl;
 336         return jb;
 337 }
 338
 339 /*
 340  * allocates a new chunk of X nodes, and links them all together as a list.
 341  * Uses the cnode->next and cnode->prev pointers
 342  * returns NULL on failure
 343  */
 344 static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 345 {
 346         struct reiserfs_journal_cnode *head;
 347         int i;
 348         if (num_cnodes <= 0) {
 349                 return NULL;
 350         }
 351         head = vzalloc(array_size(num_cnodes,
 352                                   sizeof(struct reiserfs_journal_cnode)));
 353         if (!head) {
 354                 return NULL;
 355         }
 356         head[0].prev = NULL;
 357         head[0].next = head + 1;
 358         for (i = 1; i < num_cnodes; i++) {
 359                 head[i].prev = head + (i - 1);
 360                 head[i].next = head + (i + 1);  /* if last one, overwrite it after the if */
 361         }
 362         head[num_cnodes - 1].next = NULL;
 363         return head;
 364 }
 365
 366 /* pulls a cnode off the free list, or returns NULL on failure */
 367 static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 368 {
 369         struct reiserfs_journal_cnode *cn;
 370         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 371
 372         reiserfs_check_lock_depth(sb, "get_cnode");
 373
 374         if (journal->j_cnode_free <= 0) {
 375                 return NULL;
 376         }
 377         journal->j_cnode_used++;
 378         journal->j_cnode_free--;
 379         cn = journal->j_cnode_free_list;
 380         if (!cn) {
 381                 return cn;
 382         }
 383         if (cn->next) {
 384                 cn->next->prev = NULL;
 385         }
 386         journal->j_cnode_free_list = cn->next;
 387         memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 388         return cn;
 389 }
 390
 391 /*
 392  * returns a cnode to the free list
 393  */
 394 static void free_cnode(struct super_block *sb,
 395                        struct reiserfs_journal_cnode *cn)
 396 {
 397         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 398
 399         reiserfs_check_lock_depth(sb, "free_cnode");
 400
 401         journal->j_cnode_used--;
 402         journal->j_cnode_free++;
 403         /* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 404         cn->next = journal->j_cnode_free_list;
 405         if (journal->j_cnode_free_list) {
 406                 journal->j_cnode_free_list->prev = cn;
 407         }
 408         cn->prev = NULL;        /* not needed with the memset, but I might kill the memset, and forget to do this */
 409         journal->j_cnode_free_list = cn;
 410 }
 411
 412 static void clear_prepared_bits(struct buffer_head *bh)
 413 {
 414         clear_buffer_journal_prepared(bh);
 415         clear_buffer_journal_restore_dirty(bh);
 416 }
 417
 418 /*
 419  * return a cnode with same dev, block number and size in table,
 420  * or null if not found
 421  */
 422 static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 423                                                                   super_block
 424                                                                   *sb,
 425                                                                   struct
 426                                                                   reiserfs_journal_cnode
 427                                                                   **table,
 428                                                                   long bl)
 429 {
 430         struct reiserfs_journal_cnode *cn;
 431         cn = journal_hash(table, sb, bl);
 432         while (cn) {
 433                 if (cn->blocknr == bl && cn->sb == sb)
 434                         return cn;
 435                 cn = cn->hnext;
 436         }
 437         return (struct reiserfs_journal_cnode *)0;
 438 }
 439
 440 /*
 441  * this actually means 'can this block be reallocated yet?'.  If you set
 442  * search_all, a block can only be allocated if it is not in the current
 443  * transaction, was not freed by the current transaction, and has no chance
 444  * of ever being overwritten by a replay after crashing.
 445  *
 446  * If you don't set search_all, a block can only be allocated if it is not
 447  * in the current transaction.  Since deleting a block removes it from the
 448  * current transaction, this case should never happen.  If you don't set
 449  * search_all, make sure you never write the block without logging it.
 450  *
 451  * next_zero_bit is a suggestion about the next block to try for find_forward.
 452  * when bl is rejected because it is set in a journal list bitmap, we search
 453  * for the next zero bit in the bitmap that rejected bl.  Then, we return
 454  * that through next_zero_bit for find_forward to try.
 455  *
 456  * Just because we return something in next_zero_bit does not mean we won't
 457  * reject it on the next call to reiserfs_in_journal
 458  */
 459 int reiserfs_in_journal(struct super_block *sb,
 460                         unsigned int bmap_nr, int bit_nr, int search_all,
 461                         b_blocknr_t * next_zero_bit)
 462 {
 463         struct reiserfs_journal *journal = SB_JOURNAL(sb);
 464         struct reiserfs_list_bitmap *jb;
 465         int i;
 466         unsigned long bl;
 467
 468         *next_zero_bit = 0;     /* always start this at zero. */
 469
 470         PROC_INFO_INC(sb, journal.in_journal);
 471         /*
 472          * If we aren't doing a search_all, this is a metablock, and it
 473          * will be logged before use.  if we crash before the transaction
 474          * that freed it commits,  this transaction won't have committed
 475          * either, and the block will never be written
 476          */
 477         if (search_all) {
 478                 for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 479                         PROC_INFO_INC(sb, journal.in_journal_bitmap);
 480                         jb = journal->j_list_bitmap + i;
 481                         if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 482                             test_bit(bit_nr,
 483                                      (unsigned long *)jb->bitmaps[bmap_nr]->
 484                                      data)) {
 485                                 *next_zero_bit =
 486                                     find_next_zero_bit((unsigned long *)
 487                                                        (jb->bitmaps[bmap_nr]->
 488                                                         data),
 489                                                        sb->s_blocksize << 3,
 490                                                        bit_nr + 1);
 491                                 return 1;
 492                         }
 493                 }
 494         }
 495
 496         bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 497         /* is it in any old transactions? */
 498         if (search_all
 499             && (get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 500                 return 1;
 501         }
 502
 503         /* is it in the current transaction.  This should never happen */
 504         if ((get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 505                 BUG();
 506                 return 1;
 507         }
 508
 509         PROC_INFO_INC(sb, journal.in_journal_reusable);
 510         /* safe for reuse */
 511         return 0;
 512 }
 513
 514 /* insert cn into table */
 515 static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 516                                        struct reiserfs_journal_cnode *cn)
 517 {
 518         struct reiserfs_journal_cnode *cn_orig;
 519
 520         cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 521         cn->hnext = cn_orig;
 522         cn->hprev = NULL;
 523         if (cn_orig) {
 524                 cn_orig->hprev = cn;
 525         }
 526         journal_hash(table, cn->sb, cn->blocknr) = cn;
 527 }
 528
 529 /* lock the current transaction */
 530 static inline void lock_journal(struct super_block *sb)
 531 {
 532         PROC_INFO_INC(sb, journal.lock_journal);
 533
 534         reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 535 }
 536
 537 /* unlock the current transaction */
 538 static inline void unlock_journal(struct super_block *sb)
 539 {
 540         mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 541 }
 542
 543 static inline void get_journal_list(struct reiserfs_journal_list *jl)
 544 {
 545         jl->j_refcount++;
 546 }
 547
 548 static inline void put_journal_list(struct super_block *s,
 549                                     struct reiserfs_journal_list *jl)
 550 {
 551         if (jl->j_refcount < 1) {
 552                 reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 553                                jl->j_trans_id, jl->j_refcount);
 554         }
 555         if (--jl->j_refcount == 0)
 556                 kfree(jl);
 557 }
 558
 559 /*
 560  * this used to be much more involved, and I'm keeping it just in case
 561  * things get ugly again.  it gets called by flush_commit_list, and
 562  * cleans up any data stored about blocks freed during a transaction.
 563  */
 564 static void cleanup_freed_for_journal_list(struct super_block *sb,
 565                                            struct reiserfs_journal_list *jl)
 566 {
 567
 568         struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 569         if (jb) {
 570                 cleanup_bitmap_list(sb, jb);
 571         }
 572         jl->j_list_bitmap->journal_list = NULL;
 573         jl->j_list_bitmap = NULL;
 574 }
 575
 576 static int journal_list_still_alive(struct super_block *s,
 577                                     unsigned int trans_id)
 578 {
 579         struct reiserfs_journal *journal = SB_JOURNAL(s);
 580         struct list_head *entry = &journal->j_journal_list;
 581         struct reiserfs_journal_list *jl;
 582
 583         if (!list_empty(entry)) {
 584                 jl = JOURNAL_LIST_ENTRY(entry->next);
 585                 if (jl->j_trans_id <= trans_id) {
 586                         return 1;
 587                 }
 588         }
 589         return 0;
 590 }
 591
 592 /*
 593  * If page->mapping was null, we failed to truncate this page for
 594  * some reason.  Most likely because it was truncated after being
 595  * logged via data=journal.
 596  *
 597  * This does a check to see if the buffer belongs to one of these
 598  * lost pages before doing the final put_bh.  If page->mapping was
 599  * null, it tries to free buffers on the page, which should make the
 600  * final put_page drop the page from the lru.
 601  */
 602 static void release_buffer_page(struct buffer_head *bh)
 603 {
 604         struct page *page = bh->b_page;
 605         if (!page->mapping && trylock_page(page)) {
 606                 get_page(page);
 607                 put_bh(bh);
 608                 if (!page->mapping)
 609                         try_to_free_buffers(page);
 610                 unlock_page(page);
 611                 put_page(page);
 612         } else {
 613                 put_bh(bh);
 614         }
 615 }
 616
 617 static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 618 {
 619         if (buffer_journaled(bh)) {
 620                 reiserfs_warning(NULL, "clm-2084",
 621                                  "pinned buffer %lu:%pg sent to disk",
 622                                  bh->b_blocknr, bh->b_bdev);
 623         }
 624         if (uptodate)
 625                 set_buffer_uptodate(bh);
 626         else
 627                 clear_buffer_uptodate(bh);
 628
 629         unlock_buffer(bh);
 630         release_buffer_page(bh);
 631 }
 632
 633 static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 634 {
 635         if (uptodate)
 636                 set_buffer_uptodate(bh);
 637         else
 638                 clear_buffer_uptodate(bh);
 639         unlock_buffer(bh);
 640         put_bh(bh);
 641 }
 642
 643 static void submit_logged_buffer(struct buffer_head *bh)
 644 {
 645         get_bh(bh);
 646         bh->b_end_io = reiserfs_end_buffer_io_sync;
 647         clear_buffer_journal_new(bh);
 648         clear_buffer_dirty(bh);
 649         if (!test_clear_buffer_journal_test(bh))
 650                 BUG();
 651         if (!buffer_uptodate(bh))
 652                 BUG();
 653         submit_bh(REQ_OP_WRITE, 0, bh);
 654 }
 655
 656 static void submit_ordered_buffer(struct buffer_head *bh)
 657 {
 658         get_bh(bh);
 659         bh->b_end_io = reiserfs_end_ordered_io;
 660         clear_buffer_dirty(bh);
 661         if (!buffer_uptodate(bh))
 662                 BUG();
 663         submit_bh(REQ_OP_WRITE, 0, bh);
 664 }
 665
 666 #define CHUNK_SIZE 32
 667 struct buffer_chunk {
 668         struct buffer_head *bh[CHUNK_SIZE];
 669         int nr;
 670 };
 671
 672 static void write_chunk(struct buffer_chunk *chunk)
 673 {
 674         int i;
 675         for (i = 0; i < chunk->nr; i++) {
 676                 submit_logged_buffer(chunk->bh[i]);
 677         }
 678         chunk->nr = 0;
 679 }
 680
 681 static void write_ordered_chunk(struct buffer_chunk *chunk)
 682 {
 683         int i;
 684         for (i = 0; i < chunk->nr; i++) {
 685                 submit_ordered_buffer(chunk->bh[i]);
 686         }
 687         chunk->nr = 0;
 688 }
 689
 690 static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 691                         spinlock_t * lock, void (fn) (struct buffer_chunk *))
 692 {
 693         int ret = 0;
 694         BUG_ON(chunk->nr >= CHUNK_SIZE);
 695         chunk->bh[chunk->nr++] = bh;
 696         if (chunk->nr >= CHUNK_SIZE) {
 697                 ret = 1;
 698                 if (lock) {
 699                         spin_unlock(lock);
 700                         fn(chunk);
 701                         spin_lock(lock);
 702                 } else {
 703                         fn(chunk);
 704                 }
 705         }
 706         return ret;
 707 }
 708
 709 static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 710 static struct reiserfs_jh *alloc_jh(void)
 711 {
 712         struct reiserfs_jh *jh;
 713         while (1) {
 714                 jh = kmalloc(sizeof(*jh), GFP_NOFS);
 715                 if (jh) {
 716                         atomic_inc(&nr_reiserfs_jh);
 717                         return jh;
 718                 }
 719                 yield();
 720         }
 721 }
 722
 723 /*
 724  * we want to free the jh when the buffer has been written
 725  * and waited on
 726  */
 727 void reiserfs_free_jh(struct buffer_head *bh)
 728 {
 729         struct reiserfs_jh *jh;
 730
 731         jh = bh->b_private;
 732         if (jh) {
 733                 bh->b_private = NULL;
 734                 jh->bh = NULL;
 735                 list_del_init(&jh->list);
 736                 kfree(jh);
 737                 if (atomic_read(&nr_reiserfs_jh) <= 0)
 738                         BUG();
 739                 atomic_dec(&nr_reiserfs_jh);
 740                 put_bh(bh);
 741         }
 742 }
 743
 744 static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 745                            int tail)
 746 {
 747         struct reiserfs_jh *jh;
 748
 749         if (bh->b_private) {
 750                 spin_lock(&j->j_dirty_buffers_lock);
 751                 if (!bh->b_private) {
 752                         spin_unlock(&j->j_dirty_buffers_lock);
 753                         goto no_jh;
 754                 }
 755                 jh = bh->b_private;
 756                 list_del_init(&jh->list);
 757         } else {
 758 no_jh:
 759                 get_bh(bh);
 760                 jh = alloc_jh();
 761                 spin_lock(&j->j_dirty_buffers_lock);
 762                 /*
 763                  * buffer must be locked for __add_jh, should be able to have
 764                  * two adds at the same time
 765                  */
 766                 BUG_ON(bh->b_private);
 767                 jh->bh = bh;
 768                 bh->b_private = jh;
 769         }
 770         jh->jl = j->j_current_jl;
 771         if (tail)
 772                 list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 773         else {
 774                 list_add_tail(&jh->list, &jh->jl->j_bh_list);
 775         }
 776         spin_unlock(&j->j_dirty_buffers_lock);
 777         return 0;
 778 }
 779
 780 int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 781 {
 782         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 783 }
 784 int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 785 {
 786         return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 787 }
 788
 789 #define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 790 static int write_ordered_buffers(spinlock_t * lock,
 791                                  struct reiserfs_journal *j,
 792                                  struct reiserfs_journal_list *jl,
 793                                  struct list_head *list)
 794 {
 795         struct buffer_head *bh;
 796         struct reiserfs_jh *jh;
 797         int ret = j->j_errno;
 798         struct buffer_chunk chunk;
 799         struct list_head tmp;
 800         INIT_LIST_HEAD(&tmp);
 801
 802         chunk.nr = 0;
 803         spin_lock(lock);
 804         while (!list_empty(list)) {
 805                 jh = JH_ENTRY(list->next);
 806                 bh = jh->bh;
 807                 get_bh(bh);
 808                 if (!trylock_buffer(bh)) {
 809                         if (!buffer_dirty(bh)) {
 810                                 list_move(&jh->list, &tmp);
 811                                 goto loop_next;
 812                         }
 813                         spin_unlock(lock);
 814                         if (chunk.nr)
 815                                 write_ordered_chunk(&chunk);
 816                         wait_on_buffer(bh);
 817                         cond_resched();
 818                         spin_lock(lock);
 819                         goto loop_next;
 820                 }
 821                 /*
 822                  * in theory, dirty non-uptodate buffers should never get here,
 823                  * but the upper layer io error paths still have a few quirks.
 824                  * Handle them here as gracefully as we can
 825                  */
 826                 if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 827                         clear_buffer_dirty(bh);
 828                         ret = -EIO;
 829                 }
 830                 if (buffer_dirty(bh)) {
 831                         list_move(&jh->list, &tmp);
 832                         add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 833                 } else {
 834                         reiserfs_free_jh(bh);
 835                         unlock_buffer(bh);
 836                 }
 837 loop_next:
 838                 put_bh(bh);
 839                 cond_resched_lock(lock);
 840         }
 841         if (chunk.nr) {
 842                 spin_unlock(lock);
 843                 write_ordered_chunk(&chunk);
 844                 spin_lock(lock);
 845         }
 846         while (!list_empty(&tmp)) {
 847                 jh = JH_ENTRY(tmp.prev);
 848                 bh = jh->bh;
 849                 get_bh(bh);
 850                 reiserfs_free_jh(bh);
 851
 852                 if (buffer_locked(bh)) {
 853                         spin_unlock(lock);
 854                         wait_on_buffer(bh);
 855                         spin_lock(lock);
 856                 }
 857                 if (!buffer_uptodate(bh)) {
 858                         ret = -EIO;
 859                 }
 860                 /*
 861                  * ugly interaction with invalidatepage here.
 862                  * reiserfs_invalidate_page will pin any buffer that has a
 863                  * valid journal head from an older transaction.  If someone
 864                  * else sets our buffer dirty after we write it in the first
 865                  * loop, and then someone truncates the page away, nobody
 866                  * will ever write the buffer. We're safe if we write the
 867                  * page one last time after freeing the journal header.
 868                  */
 869                 if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 870                         spin_unlock(lock);
 871                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 872                         spin_lock(lock);
 873                 }
 874                 put_bh(bh);
 875                 cond_resched_lock(lock);
 876         }
 877         spin_unlock(lock);
 878         return ret;
 879 }
 880
 881 static int flush_older_commits(struct super_block *s,
 882                                struct reiserfs_journal_list *jl)
 883 {
 884         struct reiserfs_journal *journal = SB_JOURNAL(s);
 885         struct reiserfs_journal_list *other_jl;
 886         struct reiserfs_journal_list *first_jl;
 887         struct list_head *entry;
 888         unsigned int trans_id = jl->j_trans_id;
 889         unsigned int other_trans_id;
 890
 891 find_first:
 892         /*
 893          * first we walk backwards to find the oldest uncommitted transation
 894          */
 895         first_jl = jl;
 896         entry = jl->j_list.prev;
 897         while (1) {
 898                 other_jl = JOURNAL_LIST_ENTRY(entry);
 899                 if (entry == &journal->j_journal_list ||
 900                     atomic_read(&other_jl->j_older_commits_done))
 901                         break;
 902
 903                 first_jl = other_jl;
 904                 entry = other_jl->j_list.prev;
 905         }
 906
 907         /* if we didn't find any older uncommitted transactions, return now */
 908         if (first_jl == jl) {
 909                 return 0;
 910         }
 911
 912         entry = &first_jl->j_list;
 913         while (1) {
 914                 other_jl = JOURNAL_LIST_ENTRY(entry);
 915                 other_trans_id = other_jl->j_trans_id;
 916
 917                 if (other_trans_id < trans_id) {
 918                         if (atomic_read(&other_jl->j_commit_left) != 0) {
 919                                 flush_commit_list(s, other_jl, 0);
 920
 921                                 /* list we were called with is gone, return */
 922                                 if (!journal_list_still_alive(s, trans_id))
 923                                         return 1;
 924
 925                                 /*
 926                                  * the one we just flushed is gone, this means
 927                                  * all older lists are also gone, so first_jl
 928                                  * is no longer valid either.  Go back to the
 929                                  * beginning.
 930                                  */
 931                                 if (!journal_list_still_alive
 932                                     (s, other_trans_id)) {
 933                                         goto find_first;
 934                                 }
 935                         }
 936                         entry = entry->next;
 937                         if (entry == &journal->j_journal_list)
 938                                 return 0;
 939                 } else {
 940                         return 0;
 941                 }
 942         }
 943         return 0;
 944 }
 945
 946 static int reiserfs_async_progress_wait(struct super_block *s)
 947 {
 948         struct reiserfs_journal *j = SB_JOURNAL(s);
 949
 950         if (atomic_read(&j->j_async_throttle)) {
 951                 int depth;
 952
 953                 depth = reiserfs_write_unlock_nested(s);
 954                 congestion_wait(BLK_RW_ASYNC, HZ / 10);
 955                 reiserfs_write_lock_nested(s, depth);
 956         }
 957
 958         return 0;
 959 }
 960
 961 /*
 962  * if this journal list still has commit blocks unflushed, send them to disk.
 963  *
 964  * log areas must be flushed in order (transaction 2 can't commit before
 965  * transaction 1) Before the commit block can by written, every other log
 966  * block must be safely on disk
 967  */
 968 static int flush_commit_list(struct super_block *s,
 969                              struct reiserfs_journal_list *jl, int flushall)
 970 {
 971         int i;
 972         b_blocknr_t bn;
 973         struct buffer_head *tbh = NULL;
 974         unsigned int trans_id = jl->j_trans_id;
 975         struct reiserfs_journal *journal = SB_JOURNAL(s);
 976         int retval = 0;
 977         int write_len;
 978         int depth;
 979
 980         reiserfs_check_lock_depth(s, "flush_commit_list");
 981
 982         if (atomic_read(&jl->j_older_commits_done)) {
 983                 return 0;
 984         }
 985
 986         /*
 987          * before we can put our commit blocks on disk, we have to make
 988          * sure everyone older than us is on disk too
 989          */
 990         BUG_ON(jl->j_len <= 0);
 991         BUG_ON(trans_id == journal->j_trans_id);
 992
 993         get_journal_list(jl);
 994         if (flushall) {
 995                 if (flush_older_commits(s, jl) == 1) {
 996                         /*
 997                          * list disappeared during flush_older_commits.
 998                          * return
 999                          */
1000                         goto put_jl;
1001                 }
1002         }
1003
1004         /* make sure nobody is trying to flush this one at the same time */
1005         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1006
1007         if (!journal_list_still_alive(s, trans_id)) {
1008                 mutex_unlock(&jl->j_commit_mutex);
1009                 goto put_jl;
1010         }
1011         BUG_ON(jl->j_trans_id == 0);
1012
1013         /* this commit is done, exit */
1014         if (atomic_read(&jl->j_commit_left) <= 0) {
1015                 if (flushall) {
1016                         atomic_set(&jl->j_older_commits_done, 1);
1017                 }
1018                 mutex_unlock(&jl->j_commit_mutex);
1019                 goto put_jl;
1020         }
1021
1022         if (!list_empty(&jl->j_bh_list)) {
1023                 int ret;
1024
1025                 /*
1026                  * We might sleep in numerous places inside
1027                  * write_ordered_buffers. Relax the write lock.
1028                  */
1029                 depth = reiserfs_write_unlock_nested(s);
1030                 ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1031                                             journal, jl, &jl->j_bh_list);
1032                 if (ret < 0 && retval == 0)
1033                         retval = ret;
1034                 reiserfs_write_lock_nested(s, depth);
1035         }
1036         BUG_ON(!list_empty(&jl->j_bh_list));
1037         /*
1038          * for the description block and all the log blocks, submit any buffers
1039          * that haven't already reached the disk.  Try to write at least 256
1040          * log blocks. later on, we will only wait on blocks that correspond
1041          * to this transaction, but while we're unplugging we might as well
1042          * get a chunk of data on there.
1043          */
1044         atomic_inc(&journal->j_async_throttle);
1045         write_len = jl->j_len + 1;
1046         if (write_len < 256)
1047                 write_len = 256;
1048         for (i = 0 ; i < write_len ; i++) {
1049                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1050                     SB_ONDISK_JOURNAL_SIZE(s);
1051                 tbh = journal_find_get_block(s, bn);
1052                 if (tbh) {
1053                         if (buffer_dirty(tbh)) {
1054                             depth = reiserfs_write_unlock_nested(s);
1055                             ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1056                             reiserfs_write_lock_nested(s, depth);
1057                         }
1058                         put_bh(tbh) ;
1059                 }
1060         }
1061         atomic_dec(&journal->j_async_throttle);
1062
1063         for (i = 0; i < (jl->j_len + 1); i++) {
1064                 bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1065                     (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1066                 tbh = journal_find_get_block(s, bn);
1067
1068                 depth = reiserfs_write_unlock_nested(s);
1069                 __wait_on_buffer(tbh);
1070                 reiserfs_write_lock_nested(s, depth);
1071                 /*
1072                  * since we're using ll_rw_blk above, it might have skipped
1073                  * over a locked buffer.  Double check here
1074                  */
1075                 /* redundant, sync_dirty_buffer() checks */
1076                 if (buffer_dirty(tbh)) {
1077                         depth = reiserfs_write_unlock_nested(s);
1078                         sync_dirty_buffer(tbh);
1079                         reiserfs_write_lock_nested(s, depth);
1080                 }
1081                 if (unlikely(!buffer_uptodate(tbh))) {
1082 #ifdef CONFIG_REISERFS_CHECK
1083                         reiserfs_warning(s, "journal-601",
1084                                          "buffer write failed");
1085 #endif
1086                         retval = -EIO;
1087                 }
1088                 /* once for journal_find_get_block */
1089                 put_bh(tbh);
1090                 /* once due to original getblk in do_journal_end */
1091                 put_bh(tbh);
1092                 atomic_dec(&jl->j_commit_left);
1093         }
1094
1095         BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1096
1097         /*
1098          * If there was a write error in the journal - we can't commit
1099          * this transaction - it will be invalid and, if successful,
1100          * will just end up propagating the write error out to
1101          * the file system.
1102          */
1103         if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1104                 if (buffer_dirty(jl->j_commit_bh))
1105                         BUG();
1106                 mark_buffer_dirty(jl->j_commit_bh) ;
1107                 depth = reiserfs_write_unlock_nested(s);
1108                 if (reiserfs_barrier_flush(s))
1109                         __sync_dirty_buffer(jl->j_commit_bh,
1110                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1111                 else
1112                         sync_dirty_buffer(jl->j_commit_bh);
1113                 reiserfs_write_lock_nested(s, depth);
1114         }
1115
1116         /*
1117          * If there was a write error in the journal - we can't commit this
1118          * transaction - it will be invalid and, if successful, will just end
1119          * up propagating the write error out to the filesystem.
1120          */
1121         if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1122 #ifdef CONFIG_REISERFS_CHECK
1123                 reiserfs_warning(s, "journal-615", "buffer write failed");
1124 #endif
1125                 retval = -EIO;
1126         }
1127         bforget(jl->j_commit_bh);
1128         if (journal->j_last_commit_id != 0 &&
1129             (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1130                 reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1131                                  journal->j_last_commit_id, jl->j_trans_id);
1132         }
1133         journal->j_last_commit_id = jl->j_trans_id;
1134
1135         /*
1136          * now, every commit block is on the disk.  It is safe to allow
1137          * blocks freed during this transaction to be reallocated
1138          */
1139         cleanup_freed_for_journal_list(s, jl);
1140
1141         retval = retval ? retval : journal->j_errno;
1142
1143         /* mark the metadata dirty */
1144         if (!retval)
1145                 dirty_one_transaction(s, jl);
1146         atomic_dec(&jl->j_commit_left);
1147
1148         if (flushall) {
1149                 atomic_set(&jl->j_older_commits_done, 1);
1150         }
1151         mutex_unlock(&jl->j_commit_mutex);
1152 put_jl:
1153         put_journal_list(s, jl);
1154
1155         if (retval)
1156                 reiserfs_abort(s, retval, "Journal write error in %s",
1157                                __func__);
1158         return retval;
1159 }
1160
1161 /*
1162  * flush_journal_list frequently needs to find a newer transaction for a
1163  * given block.  This does that, or returns NULL if it can't find anything
1164  */
1165 static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1166                                                           reiserfs_journal_cnode
1167                                                           *cn)
1168 {
1169         struct super_block *sb = cn->sb;
1170         b_blocknr_t blocknr = cn->blocknr;
1171
1172         cn = cn->hprev;
1173         while (cn) {
1174                 if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1175                         return cn->jlist;
1176                 }
1177                 cn = cn->hprev;
1178         }
1179         return NULL;
1180 }
1181
1182 static void remove_journal_hash(struct super_block *,
1183                                 struct reiserfs_journal_cnode **,
1184                                 struct reiserfs_journal_list *, unsigned long,
1185                                 int);
1186
1187 /*
1188  * once all the real blocks have been flushed, it is safe to remove them
1189  * from the journal list for this transaction.  Aside from freeing the
1190  * cnode, this also allows the block to be reallocated for data blocks
1191  * if it had been deleted.
1192  */
1193 static void remove_all_from_journal_list(struct super_block *sb,
1194                                          struct reiserfs_journal_list *jl,
1195                                          int debug)
1196 {
1197         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1198         struct reiserfs_journal_cnode *cn, *last;
1199         cn = jl->j_realblock;
1200
1201         /*
1202          * which is better, to lock once around the whole loop, or
1203          * to lock for each call to remove_journal_hash?
1204          */
1205         while (cn) {
1206                 if (cn->blocknr != 0) {
1207                         if (debug) {
1208                                 reiserfs_warning(sb, "reiserfs-2201",
1209                                                  "block %u, bh is %d, state %ld",
1210                                                  cn->blocknr, cn->bh ? 1 : 0,
1211                                                  cn->state);
1212                         }
1213                         cn->state = 0;
1214                         remove_journal_hash(sb, journal->j_list_hash_table,
1215                                             jl, cn->blocknr, 1);
1216                 }
1217                 last = cn;
1218                 cn = cn->next;
1219                 free_cnode(sb, last);
1220         }
1221         jl->j_realblock = NULL;
1222 }
1223
1224 /*
1225  * if this timestamp is greater than the timestamp we wrote last to the
1226  * header block, write it to the header block.  once this is done, I can
1227  * safely say the log area for this transaction won't ever be replayed,
1228  * and I can start releasing blocks in this transaction for reuse as data
1229  * blocks.  called by flush_journal_list, before it calls
1230  * remove_all_from_journal_list
1231  */
1232 static int _update_journal_header_block(struct super_block *sb,
1233                                         unsigned long offset,
1234                                         unsigned int trans_id)
1235 {
1236         struct reiserfs_journal_header *jh;
1237         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1238         int depth;
1239
1240         if (reiserfs_is_journal_aborted(journal))
1241                 return -EIO;
1242
1243         if (trans_id >= journal->j_last_flush_trans_id) {
1244                 if (buffer_locked((journal->j_header_bh))) {
1245                         depth = reiserfs_write_unlock_nested(sb);
1246                         __wait_on_buffer(journal->j_header_bh);
1247                         reiserfs_write_lock_nested(sb, depth);
1248                         if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1249 #ifdef CONFIG_REISERFS_CHECK
1250                                 reiserfs_warning(sb, "journal-699",
1251                                                  "buffer write failed");
1252 #endif
1253                                 return -EIO;
1254                         }
1255                 }
1256                 journal->j_last_flush_trans_id = trans_id;
1257                 journal->j_first_unflushed_offset = offset;
1258                 jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1259                                                         b_data);
1260                 jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1261                 jh->j_first_unflushed_offset = cpu_to_le32(offset);
1262                 jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1263
1264                 set_buffer_dirty(journal->j_header_bh);
1265                 depth = reiserfs_write_unlock_nested(sb);
1266
1267                 if (reiserfs_barrier_flush(sb))
1268                         __sync_dirty_buffer(journal->j_header_bh,
1269                                         REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1270                 else
1271                         sync_dirty_buffer(journal->j_header_bh);
1272
1273                 reiserfs_write_lock_nested(sb, depth);
1274                 if (!buffer_uptodate(journal->j_header_bh)) {
1275                         reiserfs_warning(sb, "journal-837",
1276                                          "IO error during journal replay");
1277                         return -EIO;
1278                 }
1279         }
1280         return 0;
1281 }
1282
1283 static int update_journal_header_block(struct super_block *sb,
1284                                        unsigned long offset,
1285                                        unsigned int trans_id)
1286 {
1287         return _update_journal_header_block(sb, offset, trans_id);
1288 }
1289
1290 /*
1291 ** flush any and all journal lists older than you are
1292 ** can only be called from flush_journal_list
1293 */
1294 static int flush_older_journal_lists(struct super_block *sb,
1295                                      struct reiserfs_journal_list *jl)
1296 {
1297         struct list_head *entry;
1298         struct reiserfs_journal_list *other_jl;
1299         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1300         unsigned int trans_id = jl->j_trans_id;
1301
1302         /*
1303          * we know we are the only ones flushing things, no extra race
1304          * protection is required.
1305          */
1306 restart:
1307         entry = journal->j_journal_list.next;
1308         /* Did we wrap? */
1309         if (entry == &journal->j_journal_list)
1310                 return 0;
1311         other_jl = JOURNAL_LIST_ENTRY(entry);
1312         if (other_jl->j_trans_id < trans_id) {
1313                 BUG_ON(other_jl->j_refcount <= 0);
1314                 /* do not flush all */
1315                 flush_journal_list(sb, other_jl, 0);
1316
1317                 /* other_jl is now deleted from the list */
1318                 goto restart;
1319         }
1320         return 0;
1321 }
1322
1323 static void del_from_work_list(struct super_block *s,
1324                                struct reiserfs_journal_list *jl)
1325 {
1326         struct reiserfs_journal *journal = SB_JOURNAL(s);
1327         if (!list_empty(&jl->j_working_list)) {
1328                 list_del_init(&jl->j_working_list);
1329                 journal->j_num_work_lists--;
1330         }
1331 }
1332
1333 /*
1334  * flush a journal list, both commit and real blocks
1335  *
1336  * always set flushall to 1, unless you are calling from inside
1337  * flush_journal_list
1338  *
1339  * IMPORTANT.  This can only be called while there are no journal writers,
1340  * and the journal is locked.  That means it can only be called from
1341  * do_journal_end, or by journal_release
1342  */
1343 static int flush_journal_list(struct super_block *s,
1344                               struct reiserfs_journal_list *jl, int flushall)
1345 {
1346         struct reiserfs_journal_list *pjl;
1347         struct reiserfs_journal_cnode *cn;
1348         int count;
1349         int was_jwait = 0;
1350         int was_dirty = 0;
1351         struct buffer_head *saved_bh;
1352         unsigned long j_len_saved = jl->j_len;
1353         struct reiserfs_journal *journal = SB_JOURNAL(s);
1354         int err = 0;
1355         int depth;
1356
1357         BUG_ON(j_len_saved <= 0);
1358
1359         if (atomic_read(&journal->j_wcount) != 0) {
1360                 reiserfs_warning(s, "clm-2048", "called with wcount %d",
1361                                  atomic_read(&journal->j_wcount));
1362         }
1363
1364         /* if flushall == 0, the lock is already held */
1365         if (flushall) {
1366                 reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1367         } else if (mutex_trylock(&journal->j_flush_mutex)) {
1368                 BUG();
1369         }
1370
1371         count = 0;
1372         if (j_len_saved > journal->j_trans_max) {
1373                 reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1374                                j_len_saved, jl->j_trans_id);
1375                 return 0;
1376         }
1377
1378         /* if all the work is already done, get out of here */
1379         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1380             atomic_read(&jl->j_commit_left) <= 0) {
1381                 goto flush_older_and_return;
1382         }
1383
1384         /*
1385          * start by putting the commit list on disk.  This will also flush
1386          * the commit lists of any olders transactions
1387          */
1388         flush_commit_list(s, jl, 1);
1389
1390         if (!(jl->j_state & LIST_DIRTY)
1391             && !reiserfs_is_journal_aborted(journal))
1392                 BUG();
1393
1394         /* are we done now? */
1395         if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1396             atomic_read(&jl->j_commit_left) <= 0) {
1397                 goto flush_older_and_return;
1398         }
1399
1400         /*
1401          * loop through each cnode, see if we need to write it,
1402          * or wait on a more recent transaction, or just ignore it
1403          */
1404         if (atomic_read(&journal->j_wcount) != 0) {
1405                 reiserfs_panic(s, "journal-844", "journal list is flushing, "
1406                                "wcount is not 0");
1407         }
1408         cn = jl->j_realblock;
1409         while (cn) {
1410                 was_jwait = 0;
1411                 was_dirty = 0;
1412                 saved_bh = NULL;
1413                 /* blocknr of 0 is no longer in the hash, ignore it */
1414                 if (cn->blocknr == 0) {
1415                         goto free_cnode;
1416                 }
1417
1418                 /*
1419                  * This transaction failed commit.
1420                  * Don't write out to the disk
1421                  */
1422                 if (!(jl->j_state & LIST_DIRTY))
1423                         goto free_cnode;
1424
1425                 pjl = find_newer_jl_for_cn(cn);
1426                 /*
1427                  * the order is important here.  We check pjl to make sure we
1428                  * don't clear BH_JDirty_wait if we aren't the one writing this
1429                  * block to disk
1430                  */
1431                 if (!pjl && cn->bh) {
1432                         saved_bh = cn->bh;
1433
1434                         /*
1435                          * we do this to make sure nobody releases the
1436                          * buffer while we are working with it
1437                          */
1438                         get_bh(saved_bh);
1439
1440                         if (buffer_journal_dirty(saved_bh)) {
1441                                 BUG_ON(!can_dirty(cn));
1442                                 was_jwait = 1;
1443                                 was_dirty = 1;
1444                         } else if (can_dirty(cn)) {
1445                                 /*
1446                                  * everything with !pjl && jwait
1447                                  * should be writable
1448                                  */
1449                                 BUG();
1450                         }
1451                 }
1452
1453                 /*
1454                  * if someone has this block in a newer transaction, just make
1455                  * sure they are committed, and don't try writing it to disk
1456                  */
1457                 if (pjl) {
1458                         if (atomic_read(&pjl->j_commit_left))
1459                                 flush_commit_list(s, pjl, 1);
1460                         goto free_cnode;
1461                 }
1462
1463                 /*
1464                  * bh == NULL when the block got to disk on its own, OR,
1465                  * the block got freed in a future transaction
1466                  */
1467                 if (saved_bh == NULL) {
1468                         goto free_cnode;
1469                 }
1470
1471                 /*
1472                  * this should never happen.  kupdate_one_transaction has
1473                  * this list locked while it works, so we should never see a
1474                  * buffer here that is not marked JDirty_wait
1475                  */
1476                 if ((!was_jwait) && !buffer_locked(saved_bh)) {
1477                         reiserfs_warning(s, "journal-813",
1478                                          "BAD! buffer %llu %cdirty %cjwait, "
1479                                          "not in a newer transaction",
1480                                          (unsigned long long)saved_bh->
1481                                          b_blocknr, was_dirty ? ' ' : '!',
1482                                          was_jwait ? ' ' : '!');
1483                 }
1484                 if (was_dirty) {
1485                         /*
1486                          * we inc again because saved_bh gets decremented
1487                          * at free_cnode
1488                          */
1489                         get_bh(saved_bh);
1490                         set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1491                         lock_buffer(saved_bh);
1492                         BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1493                         if (buffer_dirty(saved_bh))
1494                                 submit_logged_buffer(saved_bh);
1495                         else
1496                                 unlock_buffer(saved_bh);
1497                         count++;
1498                 } else {
1499                         reiserfs_warning(s, "clm-2082",
1500                                          "Unable to flush buffer %llu in %s",
1501                                          (unsigned long long)saved_bh->
1502                                          b_blocknr, __func__);
1503                 }
1504 free_cnode:
1505                 cn = cn->next;
1506                 if (saved_bh) {
1507                         /*
1508                          * we incremented this to keep others from
1509                          * taking the buffer head away
1510                          */
1511                         put_bh(saved_bh);
1512                         if (atomic_read(&saved_bh->b_count) < 0) {
1513                                 reiserfs_warning(s, "journal-945",
1514                                                  "saved_bh->b_count < 0");
1515                         }
1516                 }
1517         }
1518         if (count > 0) {
1519                 cn = jl->j_realblock;
1520                 while (cn) {
1521                         if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1522                                 if (!cn->bh) {
1523                                         reiserfs_panic(s, "journal-1011",
1524                                                        "cn->bh is NULL");
1525                                 }
1526
1527                                 depth = reiserfs_write_unlock_nested(s);
1528                                 __wait_on_buffer(cn->bh);
1529                                 reiserfs_write_lock_nested(s, depth);
1530
1531                                 if (!cn->bh) {
1532                                         reiserfs_panic(s, "journal-1012",
1533                                                        "cn->bh is NULL");
1534                                 }
1535                                 if (unlikely(!buffer_uptodate(cn->bh))) {
1536 #ifdef CONFIG_REISERFS_CHECK
1537                                         reiserfs_warning(s, "journal-949",
1538                                                          "buffer write failed");
1539 #endif
1540                                         err = -EIO;
1541                                 }
1542                                 /*
1543                                  * note, we must clear the JDirty_wait bit
1544                                  * after the up to date check, otherwise we
1545                                  * race against our flushpage routine
1546                                  */
1547                                 BUG_ON(!test_clear_buffer_journal_dirty
1548                                        (cn->bh));
1549
1550                                 /* drop one ref for us */
1551                                 put_bh(cn->bh);
1552                                 /* drop one ref for journal_mark_dirty */
1553                                 release_buffer_page(cn->bh);
1554                         }
1555                         cn = cn->next;
1556                 }
1557         }
1558
1559         if (err)
1560                 reiserfs_abort(s, -EIO,
1561                                "Write error while pushing transaction to disk in %s",
1562                                __func__);
1563 flush_older_and_return:
1564
1565         /*
1566          * before we can update the journal header block, we _must_ flush all
1567          * real blocks from all older transactions to disk.  This is because
1568          * once the header block is updated, this transaction will not be
1569          * replayed after a crash
1570          */
1571         if (flushall) {
1572                 flush_older_journal_lists(s, jl);
1573         }
1574
1575         err = journal->j_errno;
1576         /*
1577          * before we can remove everything from the hash tables for this
1578          * transaction, we must make sure it can never be replayed
1579          *
1580          * since we are only called from do_journal_end, we know for sure there
1581          * are no allocations going on while we are flushing journal lists.  So,
1582          * we only need to update the journal header block for the last list
1583          * being flushed
1584          */
1585         if (!err && flushall) {
1586                 err =
1587                     update_journal_header_block(s,
1588                                                 (jl->j_start + jl->j_len +
1589                                                  2) % SB_ONDISK_JOURNAL_SIZE(s),
1590                                                 jl->j_trans_id);
1591                 if (err)
1592                         reiserfs_abort(s, -EIO,
1593                                        "Write error while updating journal header in %s",
1594                                        __func__);
1595         }
1596         remove_all_from_journal_list(s, jl, 0);
1597         list_del_init(&jl->j_list);
1598         journal->j_num_lists--;
1599         del_from_work_list(s, jl);
1600
1601         if (journal->j_last_flush_id != 0 &&
1602             (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1603                 reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1604                                  journal->j_last_flush_id, jl->j_trans_id);
1605         }
1606         journal->j_last_flush_id = jl->j_trans_id;
1607
1608         /*
1609          * not strictly required since we are freeing the list, but it should
1610          * help find code using dead lists later on
1611          */
1612         jl->j_len = 0;
1613         atomic_set(&jl->j_nonzerolen, 0);
1614         jl->j_start = 0;
1615         jl->j_realblock = NULL;
1616         jl->j_commit_bh = NULL;
1617         jl->j_trans_id = 0;
1618         jl->j_state = 0;
1619         put_journal_list(s, jl);
1620         if (flushall)
1621                 mutex_unlock(&journal->j_flush_mutex);
1622         return err;
1623 }
1624
1625 static int write_one_transaction(struct super_block *s,
1626                                  struct reiserfs_journal_list *jl,
1627                                  struct buffer_chunk *chunk)
1628 {
1629         struct reiserfs_journal_cnode *cn;
1630         int ret = 0;
1631
1632         jl->j_state |= LIST_TOUCHED;
1633         del_from_work_list(s, jl);
1634         if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1635                 return 0;
1636         }
1637
1638         cn = jl->j_realblock;
1639         while (cn) {
1640                 /*
1641                  * if the blocknr == 0, this has been cleared from the hash,
1642                  * skip it
1643                  */
1644                 if (cn->blocknr == 0) {
1645                         goto next;
1646                 }
1647                 if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1648                         struct buffer_head *tmp_bh;
1649                         /*
1650                          * we can race against journal_mark_freed when we try
1651                          * to lock_buffer(cn->bh), so we have to inc the buffer
1652                          * count, and recheck things after locking
1653                          */
1654                         tmp_bh = cn->bh;
1655                         get_bh(tmp_bh);
1656                         lock_buffer(tmp_bh);
1657                         if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1658                                 if (!buffer_journal_dirty(tmp_bh) ||
1659                                     buffer_journal_prepared(tmp_bh))
1660                                         BUG();
1661                                 add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1662                                 ret++;
1663                         } else {
1664                                 /* note, cn->bh might be null now */
1665                                 unlock_buffer(tmp_bh);
1666                         }
1667                         put_bh(tmp_bh);
1668                 }
1669 next:
1670                 cn = cn->next;
1671                 cond_resched();
1672         }
1673         return ret;
1674 }
1675
1676 /* used by flush_commit_list */
1677 static void dirty_one_transaction(struct super_block *s,
1678                                  struct reiserfs_journal_list *jl)
1679 {
1680         struct reiserfs_journal_cnode *cn;
1681         struct reiserfs_journal_list *pjl;
1682
1683         jl->j_state |= LIST_DIRTY;
1684         cn = jl->j_realblock;
1685         while (cn) {
1686                 /*
1687                  * look for a more recent transaction that logged this
1688                  * buffer.  Only the most recent transaction with a buffer in
1689                  * it is allowed to send that buffer to disk
1690                  */
1691                 pjl = find_newer_jl_for_cn(cn);
1692                 if (!pjl && cn->blocknr && cn->bh
1693                     && buffer_journal_dirty(cn->bh)) {
1694                         BUG_ON(!can_dirty(cn));
1695                         /*
1696                          * if the buffer is prepared, it will either be logged
1697                          * or restored.  If restored, we need to make sure
1698                          * it actually gets marked dirty
1699                          */
1700                         clear_buffer_journal_new(cn->bh);
1701                         if (buffer_journal_prepared(cn->bh)) {
1702                                 set_buffer_journal_restore_dirty(cn->bh);
1703                         } else {
1704                                 set_buffer_journal_test(cn->bh);
1705                                 mark_buffer_dirty(cn->bh);
1706                         }
1707                 }
1708                 cn = cn->next;
1709         }
1710 }
1711
1712 static int kupdate_transactions(struct super_block *s,
1713                                 struct reiserfs_journal_list *jl,
1714                                 struct reiserfs_journal_list **next_jl,
1715                                 unsigned int *next_trans_id,
1716                                 int num_blocks, int num_trans)
1717 {
1718         int ret = 0;
1719         int written = 0;
1720         int transactions_flushed = 0;
1721         unsigned int orig_trans_id = jl->j_trans_id;
1722         struct buffer_chunk chunk;
1723         struct list_head *entry;
1724         struct reiserfs_journal *journal = SB_JOURNAL(s);
1725         chunk.nr = 0;
1726
1727         reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1728         if (!journal_list_still_alive(s, orig_trans_id)) {
1729                 goto done;
1730         }
1731
1732         /*
1733          * we've got j_flush_mutex held, nobody is going to delete any
1734          * of these lists out from underneath us
1735          */
1736         while ((num_trans && transactions_flushed < num_trans) ||
1737                (!num_trans && written < num_blocks)) {
1738
1739                 if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1740                     atomic_read(&jl->j_commit_left)
1741                     || !(jl->j_state & LIST_DIRTY)) {
1742                         del_from_work_list(s, jl);
1743                         break;
1744                 }
1745                 ret = write_one_transaction(s, jl, &chunk);
1746
1747                 if (ret < 0)
1748                         goto done;
1749                 transactions_flushed++;
1750                 written += ret;
1751                 entry = jl->j_list.next;
1752
1753                 /* did we wrap? */
1754                 if (entry == &journal->j_journal_list) {
1755                         break;
1756                 }
1757                 jl = JOURNAL_LIST_ENTRY(entry);
1758
1759                 /* don't bother with older transactions */
1760                 if (jl->j_trans_id <= orig_trans_id)
1761                         break;
1762         }
1763         if (chunk.nr) {
1764                 write_chunk(&chunk);
1765         }
1766
1767 done:
1768         mutex_unlock(&journal->j_flush_mutex);
1769         return ret;
1770 }
1771
1772 /*
1773  * for o_sync and fsync heavy applications, they tend to use
1774  * all the journa list slots with tiny transactions.  These
1775  * trigger lots and lots of calls to update the header block, which
1776  * adds seeks and slows things down.
1777  *
1778  * This function tries to clear out a large chunk of the journal lists
1779  * at once, which makes everything faster since only the newest journal
1780  * list updates the header block
1781  */
1782 static int flush_used_journal_lists(struct super_block *s,
1783                                     struct reiserfs_journal_list *jl)
1784 {
1785         unsigned long len = 0;
1786         unsigned long cur_len;
1787         int i;
1788         int limit = 256;
1789         struct reiserfs_journal_list *tjl;
1790         struct reiserfs_journal_list *flush_jl;
1791         unsigned int trans_id;
1792         struct reiserfs_journal *journal = SB_JOURNAL(s);
1793
1794         flush_jl = tjl = jl;
1795
1796         /* in data logging mode, try harder to flush a lot of blocks */
1797         if (reiserfs_data_log(s))
1798                 limit = 1024;
1799         /* flush for 256 transactions or limit blocks, whichever comes first */
1800         for (i = 0; i < 256 && len < limit; i++) {
1801                 if (atomic_read(&tjl->j_commit_left) ||
1802                     tjl->j_trans_id < jl->j_trans_id) {
1803                         break;
1804                 }
1805                 cur_len = atomic_read(&tjl->j_nonzerolen);
1806                 if (cur_len > 0) {
1807                         tjl->j_state &= ~LIST_TOUCHED;
1808                 }
1809                 len += cur_len;
1810                 flush_jl = tjl;
1811                 if (tjl->j_list.next == &journal->j_journal_list)
1812                         break;
1813                 tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1814         }
1815         get_journal_list(jl);
1816         get_journal_list(flush_jl);
1817
1818         /*
1819          * try to find a group of blocks we can flush across all the
1820          * transactions, but only bother if we've actually spanned
1821          * across multiple lists
1822          */
1823         if (flush_jl != jl)
1824                 kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1825
1826         flush_journal_list(s, flush_jl, 1);
1827         put_journal_list(s, flush_jl);
1828         put_journal_list(s, jl);
1829         return 0;
1830 }
1831
1832 /*
1833  * removes any nodes in table with name block and dev as bh.
1834  * only touchs the hnext and hprev pointers.
1835  */
1836 static void remove_journal_hash(struct super_block *sb,
1837                          struct reiserfs_journal_cnode **table,
1838                          struct reiserfs_journal_list *jl,
1839                          unsigned long block, int remove_freed)
1840 {
1841         struct reiserfs_journal_cnode *cur;
1842         struct reiserfs_journal_cnode **head;
1843
1844         head = &(journal_hash(table, sb, block));
1845         if (!head) {
1846                 return;
1847         }
1848         cur = *head;
1849         while (cur) {
1850                 if (cur->blocknr == block && cur->sb == sb
1851                     && (jl == NULL || jl == cur->jlist)
1852                     && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1853                         if (cur->hnext) {
1854                                 cur->hnext->hprev = cur->hprev;
1855                         }
1856                         if (cur->hprev) {
1857                                 cur->hprev->hnext = cur->hnext;
1858                         } else {
1859                                 *head = cur->hnext;
1860                         }
1861                         cur->blocknr = 0;
1862                         cur->sb = NULL;
1863                         cur->state = 0;
1864                         /*
1865                          * anybody who clears the cur->bh will also
1866                          * dec the nonzerolen
1867                          */
1868                         if (cur->bh && cur->jlist)
1869                                 atomic_dec(&cur->jlist->j_nonzerolen);
1870                         cur->bh = NULL;
1871                         cur->jlist = NULL;
1872                 }
1873                 cur = cur->hnext;
1874         }
1875 }
1876
1877 static void free_journal_ram(struct super_block *sb)
1878 {
1879         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1880         kfree(journal->j_current_jl);
1881         journal->j_num_lists--;
1882
1883         vfree(journal->j_cnode_free_orig);
1884         free_list_bitmaps(sb, journal->j_list_bitmap);
1885         free_bitmap_nodes(sb);  /* must be after free_list_bitmaps */
1886         if (journal->j_header_bh) {
1887                 brelse(journal->j_header_bh);
1888         }
1889         /*
1890          * j_header_bh is on the journal dev, make sure
1891          * not to release the journal dev until we brelse j_header_bh
1892          */
1893         release_journal_dev(sb, journal);
1894         vfree(journal);
1895 }
1896
1897 /*
1898  * call on unmount.  Only set error to 1 if you haven't made your way out
1899  * of read_super() yet.  Any other caller must keep error at 0.
1900  */
1901 static int do_journal_release(struct reiserfs_transaction_handle *th,
1902                               struct super_block *sb, int error)
1903 {
1904         struct reiserfs_transaction_handle myth;
1905         struct reiserfs_journal *journal = SB_JOURNAL(sb);
1906
1907         /*
1908          * we only want to flush out transactions if we were
1909          * called with error == 0
1910          */
1911         if (!error && !sb_rdonly(sb)) {
1912                 /* end the current trans */
1913                 BUG_ON(!th->t_trans_id);
1914                 do_journal_end(th, FLUSH_ALL);
1915
1916                 /*
1917                  * make sure something gets logged to force
1918                  * our way into the flush code
1919                  */
1920                 if (!journal_join(&myth, sb)) {
1921                         reiserfs_prepare_for_journal(sb,
1922                                                      SB_BUFFER_WITH_SB(sb),
1923                                                      1);
1924                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1925                         do_journal_end(&myth, FLUSH_ALL);
1926                 }
1927         }
1928
1929         /* this also catches errors during the do_journal_end above */
1930         if (!error && reiserfs_is_journal_aborted(journal)) {
1931                 memset(&myth, 0, sizeof(myth));
1932                 if (!journal_join_abort(&myth, sb)) {
1933                         reiserfs_prepare_for_journal(sb,
1934                                                      SB_BUFFER_WITH_SB(sb),
1935                                                      1);
1936                         journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1937                         do_journal_end(&myth, FLUSH_ALL);
1938                 }
1939         }
1940
1941
1942         /*
1943          * We must release the write lock here because
1944          * the workqueue job (flush_async_commit) needs this lock
1945          */
1946         reiserfs_write_unlock(sb);
1947
1948         /*
1949          * Cancel flushing of old commits. Note that neither of these works
1950          * will be requeued because superblock is being shutdown and doesn't
1951          * have SB_ACTIVE set.
1952          */
1953         reiserfs_cancel_old_flush(sb);
1954         /* wait for all commits to finish */
1955         cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1956
1957         free_journal_ram(sb);
1958
1959         reiserfs_write_lock(sb);
1960
1961         return 0;
1962 }
1963
1964 /* * call on unmount.  flush all journal trans, release all alloc'd ram */
1965 int journal_release(struct reiserfs_transaction_handle *th,
1966                     struct super_block *sb)
1967 {
1968         return do_journal_release(th, sb, 0);
1969 }
1970
1971 /* only call from an error condition inside reiserfs_read_super!  */
1972 int journal_release_error(struct reiserfs_transaction_handle *th,
1973                           struct super_block *sb)
1974 {
1975         return do_journal_release(th, sb, 1);
1976 }
1977
1978 /*
1979  * compares description block with commit block.
1980  * returns 1 if they differ, 0 if they are the same
1981  */
1982 static int journal_compare_desc_commit(struct super_block *sb,
1983                                        struct reiserfs_journal_desc *desc,
1984                                        struct reiserfs_journal_commit *commit)
1985 {
1986         if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1987             get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1988             get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1989             get_commit_trans_len(commit) <= 0) {
1990                 return 1;
1991         }
1992         return 0;
1993 }
1994
1995 /*
1996  * returns 0 if it did not find a description block
1997  * returns -1 if it found a corrupt commit block
1998  * returns 1 if both desc and commit were valid
1999  * NOTE: only called during fs mount
2000  */
2001 static int journal_transaction_is_valid(struct super_block *sb,
2002                                         struct buffer_head *d_bh,
2003                                         unsigned int *oldest_invalid_trans_id,
2004                                         unsigned long *newest_mount_id)
2005 {
2006         struct reiserfs_journal_desc *desc;
2007         struct reiserfs_journal_commit *commit;
2008         struct buffer_head *c_bh;
2009         unsigned long offset;
2010
2011         if (!d_bh)
2012                 return 0;
2013
2014         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2015         if (get_desc_trans_len(desc) > 0
2016             && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2017                 if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2018                     && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2019                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2020                                        "journal-986: transaction "
2021                                        "is valid returning because trans_id %d is greater than "
2022                                        "oldest_invalid %lu",
2023                                        get_desc_trans_id(desc),
2024                                        *oldest_invalid_trans_id);
2025                         return 0;
2026                 }
2027                 if (newest_mount_id
2028                     && *newest_mount_id > get_desc_mount_id(desc)) {
2029                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2030                                        "journal-1087: transaction "
2031                                        "is valid returning because mount_id %d is less than "
2032                                        "newest_mount_id %lu",
2033                                        get_desc_mount_id(desc),
2034                                        *newest_mount_id);
2035                         return -1;
2036                 }
2037                 if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2038                         reiserfs_warning(sb, "journal-2018",
2039                                          "Bad transaction length %d "
2040                                          "encountered, ignoring transaction",
2041                                          get_desc_trans_len(desc));
2042                         return -1;
2043                 }
2044                 offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2045
2046                 /*
2047                  * ok, we have a journal description block,
2048                  * let's see if the transaction was valid
2049                  */
2050                 c_bh =
2051                     journal_bread(sb,
2052                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2053                                   ((offset + get_desc_trans_len(desc) +
2054                                     1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2055                 if (!c_bh)
2056                         return 0;
2057                 commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2058                 if (journal_compare_desc_commit(sb, desc, commit)) {
2059                         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2060                                        "journal_transaction_is_valid, commit offset %ld had bad "
2061                                        "time %d or length %d",
2062                                        c_bh->b_blocknr -
2063                                        SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2064                                        get_commit_trans_id(commit),
2065                                        get_commit_trans_len(commit));
2066                         brelse(c_bh);
2067                         if (oldest_invalid_trans_id) {
2068                                 *oldest_invalid_trans_id =
2069                                     get_desc_trans_id(desc);
2070                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2071                                                "journal-1004: "
2072                                                "transaction_is_valid setting oldest invalid trans_id "
2073                                                "to %d",
2074                                                get_desc_trans_id(desc));
2075                         }
2076                         return -1;
2077                 }
2078                 brelse(c_bh);
2079                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2080                                "journal-1006: found valid "
2081                                "transaction start offset %llu, len %d id %d",
2082                                d_bh->b_blocknr -
2083                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2084                                get_desc_trans_len(desc),
2085                                get_desc_trans_id(desc));
2086                 return 1;
2087         } else {
2088                 return 0;
2089         }
2090 }
2091
2092 static void brelse_array(struct buffer_head **heads, int num)
2093 {
2094         int i;
2095         for (i = 0; i < num; i++) {
2096                 brelse(heads[i]);
2097         }
2098 }
2099
2100 /*
2101  * given the start, and values for the oldest acceptable transactions,
2102  * this either reads in a replays a transaction, or returns because the
2103  * transaction is invalid, or too old.
2104  * NOTE: only called during fs mount
2105  */
2106 static int journal_read_transaction(struct super_block *sb,
2107                                     unsigned long cur_dblock,
2108                                     unsigned long oldest_start,
2109                                     unsigned int oldest_trans_id,
2110                                     unsigned long newest_mount_id)
2111 {
2112         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2113         struct reiserfs_journal_desc *desc;
2114         struct reiserfs_journal_commit *commit;
2115         unsigned int trans_id = 0;
2116         struct buffer_head *c_bh;
2117         struct buffer_head *d_bh;
2118         struct buffer_head **log_blocks = NULL;
2119         struct buffer_head **real_blocks = NULL;
2120         unsigned int trans_offset;
2121         int i;
2122         int trans_half;
2123
2124         d_bh = journal_bread(sb, cur_dblock);
2125         if (!d_bh)
2126                 return 1;
2127         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2128         trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2129         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2130                        "journal_read_transaction, offset %llu, len %d mount_id %d",
2131                        d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2132                        get_desc_trans_len(desc), get_desc_mount_id(desc));
2133         if (get_desc_trans_id(desc) < oldest_trans_id) {
2134                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2135                                "journal_read_trans skipping because %lu is too old",
2136                                cur_dblock -
2137                                SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2138                 brelse(d_bh);
2139                 return 1;
2140         }
2141         if (get_desc_mount_id(desc) != newest_mount_id) {
2142                 reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2143                                "journal_read_trans skipping because %d is != "
2144                                "newest_mount_id %lu", get_desc_mount_id(desc),
2145                                newest_mount_id);
2146                 brelse(d_bh);
2147                 return 1;
2148         }
2149         c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2150                              ((trans_offset + get_desc_trans_len(desc) + 1) %
2151                               SB_ONDISK_JOURNAL_SIZE(sb)));
2152         if (!c_bh) {
2153                 brelse(d_bh);
2154                 return 1;
2155         }
2156         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2157         if (journal_compare_desc_commit(sb, desc, commit)) {
2158                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2159                                "journal_read_transaction, "
2160                                "commit offset %llu had bad time %d or length %d",
2161                                c_bh->b_blocknr -
2162                                SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2163                                get_commit_trans_id(commit),
2164                                get_commit_trans_len(commit));
2165                 brelse(c_bh);
2166                 brelse(d_bh);
2167                 return 1;
2168         }
2169
2170         if (bdev_read_only(sb->s_bdev)) {
2171                 reiserfs_warning(sb, "clm-2076",
2172                                  "device is readonly, unable to replay log");
2173                 brelse(c_bh);
2174                 brelse(d_bh);
2175                 return -EROFS;
2176         }
2177
2178         trans_id = get_desc_trans_id(desc);
2179         /*
2180          * now we know we've got a good transaction, and it was
2181          * inside the valid time ranges
2182          */
2183         log_blocks = kmalloc_array(get_desc_trans_len(desc),
2184                                    sizeof(struct buffer_head *),
2185                                    GFP_NOFS);
2186         real_blocks = kmalloc_array(get_desc_trans_len(desc),
2187                                     sizeof(struct buffer_head *),
2188                                     GFP_NOFS);
2189         if (!log_blocks || !real_blocks) {
2190                 brelse(c_bh);
2191                 brelse(d_bh);
2192                 kfree(log_blocks);
2193                 kfree(real_blocks);
2194                 reiserfs_warning(sb, "journal-1169",
2195                                  "kmalloc failed, unable to mount FS");
2196                 return -1;
2197         }
2198         /* get all the buffer heads */
2199         trans_half = journal_trans_half(sb->s_blocksize);
2200         for (i = 0; i < get_desc_trans_len(desc); i++) {
2201                 log_blocks[i] =
2202                     journal_getblk(sb,
2203                                    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2204                                    (trans_offset + 1 +
2205                                     i) % SB_ONDISK_JOURNAL_SIZE(sb));
2206                 if (i < trans_half) {
2207                         real_blocks[i] =
2208                             sb_getblk(sb,
2209                                       le32_to_cpu(desc->j_realblock[i]));
2210                 } else {
2211                         real_blocks[i] =
2212                             sb_getblk(sb,
2213                                       le32_to_cpu(commit->
2214                                                   j_realblock[i - trans_half]));
2215                 }
2216                 if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2217                         reiserfs_warning(sb, "journal-1207",
2218                                          "REPLAY FAILURE fsck required! "
2219                                          "Block to replay is outside of "
2220                                          "filesystem");
2221                         goto abort_replay;
2222                 }
2223                 /* make sure we don't try to replay onto log or reserved area */
2224                 if (is_block_in_log_or_reserved_area
2225                     (sb, real_blocks[i]->b_blocknr)) {
2226                         reiserfs_warning(sb, "journal-1204",
2227                                          "REPLAY FAILURE fsck required! "
2228                                          "Trying to replay onto a log block");
2229 abort_replay:
2230                         brelse_array(log_blocks, i);
2231                         brelse_array(real_blocks, i);
2232                         brelse(c_bh);
2233                         brelse(d_bh);
2234                         kfree(log_blocks);
2235                         kfree(real_blocks);
2236                         return -1;
2237                 }
2238         }
2239         /* read in the log blocks, memcpy to the corresponding real block */
2240         ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2241         for (i = 0; i < get_desc_trans_len(desc); i++) {
2242
2243                 wait_on_buffer(log_blocks[i]);
2244                 if (!buffer_uptodate(log_blocks[i])) {
2245                         reiserfs_warning(sb, "journal-1212",
2246                                          "REPLAY FAILURE fsck required! "
2247                                          "buffer write failed");
2248                         brelse_array(log_blocks + i,
2249                                      get_desc_trans_len(desc) - i);
2250                         brelse_array(real_blocks, get_desc_trans_len(desc));
2251                         brelse(c_bh);
2252                         brelse(d_bh);
2253                         kfree(log_blocks);
2254                         kfree(real_blocks);
2255                         return -1;
2256                 }
2257                 memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2258                        real_blocks[i]->b_size);
2259                 set_buffer_uptodate(real_blocks[i]);
2260                 brelse(log_blocks[i]);
2261         }
2262         /* flush out the real blocks */
2263         for (i = 0; i < get_desc_trans_len(desc); i++) {
2264                 set_buffer_dirty(real_blocks[i]);
2265                 write_dirty_buffer(real_blocks[i], 0);
2266         }
2267         for (i = 0; i < get_desc_trans_len(desc); i++) {
2268                 wait_on_buffer(real_blocks[i]);
2269                 if (!buffer_uptodate(real_blocks[i])) {
2270                         reiserfs_warning(sb, "journal-1226",
2271                                          "REPLAY FAILURE, fsck required! "
2272                                          "buffer write failed");
2273                         brelse_array(real_blocks + i,
2274                                      get_desc_trans_len(desc) - i);
2275                         brelse(c_bh);
2276                         brelse(d_bh);
2277                         kfree(log_blocks);
2278                         kfree(real_blocks);
2279                         return -1;
2280                 }
2281                 brelse(real_blocks[i]);
2282         }
2283         cur_dblock =
2284             SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2285             ((trans_offset + get_desc_trans_len(desc) +
2286               2) % SB_ONDISK_JOURNAL_SIZE(sb));
2287         reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2288                        "journal-1095: setting journal " "start to offset %ld",
2289                        cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2290
2291         /*
2292          * init starting values for the first transaction, in case
2293          * this is the last transaction to be replayed.
2294          */
2295         journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2296         journal->j_last_flush_trans_id = trans_id;
2297         journal->j_trans_id = trans_id + 1;
2298         /* check for trans_id overflow */
2299         if (journal->j_trans_id == 0)
2300                 journal->j_trans_id = 10;
2301         brelse(c_bh);
2302         brelse(d_bh);
2303         kfree(log_blocks);
2304         kfree(real_blocks);
2305         return 0;
2306 }
2307
2308 /*
2309  * This function reads blocks starting from block and to max_block of bufsize
2310  * size (but no more than BUFNR blocks at a time). This proved to improve
2311  * mounting speed on self-rebuilding raid5 arrays at least.
2312  * Right now it is only used from journal code. But later we might use it
2313  * from other places.
2314  * Note: Do not use journal_getblk/sb_getblk functions here!
2315  */
2316 static struct buffer_head *reiserfs_breada(struct block_device *dev,
2317                                            b_blocknr_t block, int bufsize,
2318                                            b_blocknr_t max_block)
2319 {
2320         struct buffer_head *bhlist[BUFNR];
2321         unsigned int blocks = BUFNR;
2322         struct buffer_head *bh;
2323         int i, j;
2324
2325         bh = __getblk(dev, block, bufsize);
2326         if (buffer_uptodate(bh))
2327                 return (bh);
2328
2329         if (block + BUFNR > max_block) {
2330                 blocks = max_block - block;
2331         }
2332         bhlist[0] = bh;
2333         j = 1;
2334         for (i = 1; i < blocks; i++) {
2335                 bh = __getblk(dev, block + i, bufsize);
2336                 if (buffer_uptodate(bh)) {
2337                         brelse(bh);
2338                         break;
2339                 } else
2340                         bhlist[j++] = bh;
2341         }
2342         ll_rw_block(REQ_OP_READ, 0, j, bhlist);
2343         for (i = 1; i < j; i++)
2344                 brelse(bhlist[i]);
2345         bh = bhlist[0];
2346         wait_on_buffer(bh);
2347         if (buffer_uptodate(bh))
2348                 return bh;
2349         brelse(bh);
2350         return NULL;
2351 }
2352
2353 /*
2354  * read and replay the log
2355  * on a clean unmount, the journal header's next unflushed pointer will be
2356  * to an invalid transaction.  This tests that before finding all the
2357  * transactions in the log, which makes normal mount times fast.
2358  *
2359  * After a crash, this starts with the next unflushed transaction, and
2360  * replays until it finds one too old, or invalid.
2361  *
2362  * On exit, it sets things up so the first transaction will work correctly.
2363  * NOTE: only called during fs mount
2364  */
2365 static int journal_read(struct super_block *sb)
2366 {
2367         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2368         struct reiserfs_journal_desc *desc;
2369         unsigned int oldest_trans_id = 0;
2370         unsigned int oldest_invalid_trans_id = 0;
2371         time64_t start;
2372         unsigned long oldest_start = 0;
2373         unsigned long cur_dblock = 0;
2374         unsigned long newest_mount_id = 9;
2375         struct buffer_head *d_bh;
2376         struct reiserfs_journal_header *jh;
2377         int valid_journal_header = 0;
2378         int replay_count = 0;
2379         int continue_replay = 1;
2380         int ret;
2381
2382         cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2383         reiserfs_info(sb, "checking transaction log (%pg)\n",
2384                       journal->j_dev_bd);
2385         start = ktime_get_seconds();
2386
2387         /*
2388          * step 1, read in the journal header block.  Check the transaction
2389          * it says is the first unflushed, and if that transaction is not
2390          * valid, replay is done
2391          */
2392         journal->j_header_bh = journal_bread(sb,
2393                                              SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2394                                              + SB_ONDISK_JOURNAL_SIZE(sb));
2395         if (!journal->j_header_bh) {
2396                 return 1;
2397         }
2398         jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2399         if (le32_to_cpu(jh->j_first_unflushed_offset) <
2400             SB_ONDISK_JOURNAL_SIZE(sb)
2401             && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2402                 oldest_start =
2403                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2404                     le32_to_cpu(jh->j_first_unflushed_offset);
2405                 oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2406                 newest_mount_id = le32_to_cpu(jh->j_mount_id);
2407                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2408                                "journal-1153: found in "
2409                                "header: first_unflushed_offset %d, last_flushed_trans_id "
2410                                "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2411                                le32_to_cpu(jh->j_last_flush_trans_id));
2412                 valid_journal_header = 1;
2413
2414                 /*
2415                  * now, we try to read the first unflushed offset.  If it
2416                  * is not valid, there is nothing more we can do, and it
2417                  * makes no sense to read through the whole log.
2418                  */
2419                 d_bh =
2420                     journal_bread(sb,
2421                                   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2422                                   le32_to_cpu(jh->j_first_unflushed_offset));
2423                 ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2424                 if (!ret) {
2425                         continue_replay = 0;
2426                 }
2427                 brelse(d_bh);
2428                 goto start_log_replay;
2429         }
2430
2431         /*
2432          * ok, there are transactions that need to be replayed.  start
2433          * with the first log block, find all the valid transactions, and
2434          * pick out the oldest.
2435          */
2436         while (continue_replay
2437                && cur_dblock <
2438                (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2439                 SB_ONDISK_JOURNAL_SIZE(sb))) {
2440                 /*
2441                  * Note that it is required for blocksize of primary fs
2442                  * device and journal device to be the same
2443                  */
2444                 d_bh =
2445                     reiserfs_breada(journal->j_dev_bd, cur_dblock,
2446                                     sb->s_blocksize,
2447                                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2448                                     SB_ONDISK_JOURNAL_SIZE(sb));
2449                 ret =
2450                     journal_transaction_is_valid(sb, d_bh,
2451                                                  &oldest_invalid_trans_id,
2452                                                  &newest_mount_id);
2453                 if (ret == 1) {
2454                         desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2455                         if (oldest_start == 0) {        /* init all oldest_ values */
2456                                 oldest_trans_id = get_desc_trans_id(desc);
2457                                 oldest_start = d_bh->b_blocknr;
2458                                 newest_mount_id = get_desc_mount_id(desc);
2459                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2460                                                "journal-1179: Setting "
2461                                                "oldest_start to offset %llu, trans_id %lu",
2462                                                oldest_start -
2463                                                SB_ONDISK_JOURNAL_1st_BLOCK
2464                                                (sb), oldest_trans_id);
2465                         } else if (oldest_trans_id > get_desc_trans_id(desc)) {
2466                                 /* one we just read was older */
2467                                 oldest_trans_id = get_desc_trans_id(desc);
2468                                 oldest_start = d_bh->b_blocknr;
2469                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2470                                                "journal-1180: Resetting "
2471                                                "oldest_start to offset %lu, trans_id %lu",
2472                                                oldest_start -
2473                                                SB_ONDISK_JOURNAL_1st_BLOCK
2474                                                (sb), oldest_trans_id);
2475                         }
2476                         if (newest_mount_id < get_desc_mount_id(desc)) {
2477                                 newest_mount_id = get_desc_mount_id(desc);
2478                                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2479                                                "journal-1299: Setting "
2480                                                "newest_mount_id to %d",
2481                                                get_desc_mount_id(desc));
2482                         }
2483                         cur_dblock += get_desc_trans_len(desc) + 2;
2484                 } else {
2485                         cur_dblock++;
2486                 }
2487                 brelse(d_bh);
2488         }
2489
2490 start_log_replay:
2491         cur_dblock = oldest_start;
2492         if (oldest_trans_id) {
2493                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2494                                "journal-1206: Starting replay "
2495                                "from offset %llu, trans_id %lu",
2496                                cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2497                                oldest_trans_id);
2498
2499         }
2500         replay_count = 0;
2501         while (continue_replay && oldest_trans_id > 0) {
2502                 ret =
2503                     journal_read_transaction(sb, cur_dblock, oldest_start,
2504                                              oldest_trans_id, newest_mount_id);
2505                 if (ret < 0) {
2506                         return ret;
2507                 } else if (ret != 0) {
2508                         break;
2509                 }
2510                 cur_dblock =
2511                     SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2512                 replay_count++;
2513                 if (cur_dblock == oldest_start)
2514                         break;
2515         }
2516
2517         if (oldest_trans_id == 0) {
2518                 reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2519                                "journal-1225: No valid " "transactions found");
2520         }
2521         /*
2522          * j_start does not get set correctly if we don't replay any
2523          * transactions.  if we had a valid journal_header, set j_start
2524          * to the first unflushed transaction value, copy the trans_id
2525          * from the header
2526          */
2527         if (valid_journal_header && replay_count == 0) {
2528                 journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2529                 journal->j_trans_id =
2530                     le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2531                 /* check for trans_id overflow */
2532                 if (journal->j_trans_id == 0)
2533                         journal->j_trans_id = 10;
2534                 journal->j_last_flush_trans_id =
2535                     le32_to_cpu(jh->j_last_flush_trans_id);
2536                 journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2537         } else {
2538                 journal->j_mount_id = newest_mount_id + 1;
2539         }
2540         reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2541                        "newest_mount_id to %lu", journal->j_mount_id);
2542         journal->j_first_unflushed_offset = journal->j_start;
2543         if (replay_count > 0) {
2544                 reiserfs_info(sb,
2545                               "replayed %d transactions in %lu seconds\n",
2546                               replay_count, ktime_get_seconds() - start);
2547         }
2548         /* needed to satisfy the locking in _update_journal_header_block */
2549         reiserfs_write_lock(sb);
2550         if (!bdev_read_only(sb->s_bdev) &&
2551             _update_journal_header_block(sb, journal->j_start,
2552                                          journal->j_last_flush_trans_id)) {
2553                 reiserfs_write_unlock(sb);
2554                 /*
2555                  * replay failed, caller must call free_journal_ram and abort
2556                  * the mount
2557                  */
2558                 return -1;
2559         }
2560         reiserfs_write_unlock(sb);
2561         return 0;
2562 }
2563
2564 static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2565 {
2566         struct reiserfs_journal_list *jl;
2567         jl = kzalloc(sizeof(struct reiserfs_journal_list),
2568                      GFP_NOFS | __GFP_NOFAIL);
2569         INIT_LIST_HEAD(&jl->j_list);
2570         INIT_LIST_HEAD(&jl->j_working_list);
2571         INIT_LIST_HEAD(&jl->j_tail_bh_list);
2572         INIT_LIST_HEAD(&jl->j_bh_list);
2573         mutex_init(&jl->j_commit_mutex);
2574         SB_JOURNAL(s)->j_num_lists++;
2575         get_journal_list(jl);
2576         return jl;
2577 }
2578
2579 static void journal_list_init(struct super_block *sb)
2580 {
2581         SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2582 }
2583
2584 static void release_journal_dev(struct super_block *super,
2585                                struct reiserfs_journal *journal)
2586 {
2587         if (journal->j_dev_bd != NULL) {
2588                 blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2589                 journal->j_dev_bd = NULL;
2590         }
2591 }
2592
2593 static int journal_init_dev(struct super_block *super,
2594                             struct reiserfs_journal *journal,
2595                             const char *jdev_name)
2596 {
2597         int result;
2598         dev_t jdev;
2599         fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2600
2601         result = 0;
2602
2603         journal->j_dev_bd = NULL;
2604         jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2605             new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2606
2607         if (bdev_read_only(super->s_bdev))
2608                 blkdev_mode = FMODE_READ;
2609
2610         /* there is no "jdev" option and journal is on separate device */
2611         if ((!jdev_name || !jdev_name[0])) {
2612                 if (jdev == super->s_dev)
2613                         blkdev_mode &= ~FMODE_EXCL;
2614                 journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2615                                                       journal);
2616                 journal->j_dev_mode = blkdev_mode;
2617                 if (IS_ERR(journal->j_dev_bd)) {
2618                         result = PTR_ERR(journal->j_dev_bd);
2619                         journal->j_dev_bd = NULL;
2620                         reiserfs_warning(super, "sh-458",
2621                                          "cannot init journal device unknown-block(%u,%u): %i",
2622                                          MAJOR(jdev), MINOR(jdev), result);
2623                         return result;
2624                 } else if (jdev != super->s_dev)
2625                         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2626
2627                 return 0;
2628         }
2629
2630         journal->j_dev_mode = blkdev_mode;
2631         journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2632         if (IS_ERR(journal->j_dev_bd)) {
2633                 result = PTR_ERR(journal->j_dev_bd);
2634                 journal->j_dev_bd = NULL;
2635                 reiserfs_warning(super, "sh-457",
2636                                  "journal_init_dev: Cannot open '%s': %i",
2637                                  jdev_name, result);
2638                 return result;
2639         }
2640
2641         set_blocksize(journal->j_dev_bd, super->s_blocksize);
2642         reiserfs_info(super,
2643                       "journal_init_dev: journal device: %pg\n",
2644                       journal->j_dev_bd);
2645         return 0;
2646 }
2647
2648 /*
2649  * When creating/tuning a file system user can assign some
2650  * journal params within boundaries which depend on the ratio
2651  * blocksize/standard_blocksize.
2652  *
2653  * For blocks >= standard_blocksize transaction size should
2654  * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2655  * then JOURNAL_TRANS_MAX_DEFAULT.
2656  *
2657  * For blocks < standard_blocksize these boundaries should be
2658  * decreased proportionally.
2659  */
2660 #define REISERFS_STANDARD_BLKSIZE (4096)
2661
2662 static int check_advise_trans_params(struct super_block *sb,
2663                                      struct reiserfs_journal *journal)
2664 {
2665         if (journal->j_trans_max) {
2666                 /* Non-default journal params.  Do sanity check for them. */
2667                 int ratio = 1;
2668                 if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2669                         ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2670
2671                 if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2672                     journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2673                     SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2674                     JOURNAL_MIN_RATIO) {
2675                         reiserfs_warning(sb, "sh-462",
2676                                          "bad transaction max size (%u). "
2677                                          "FSCK?", journal->j_trans_max);
2678                         return 1;
2679                 }
2680                 if (journal->j_max_batch != (journal->j_trans_max) *
2681                         JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2682                         reiserfs_warning(sb, "sh-463",
2683                                          "bad transaction max batch (%u). "
2684                                          "FSCK?", journal->j_max_batch);
2685                         return 1;
2686                 }
2687         } else {
2688                 /*
2689                  * Default journal params.
2690                  * The file system was created by old version
2691                  * of mkreiserfs, so some fields contain zeros,
2692                  * and we need to advise proper values for them
2693                  */
2694                 if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2695                         reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2696                                          sb->s_blocksize);
2697                         return 1;
2698                 }
2699                 journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2700                 journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2701                 journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2702         }
2703         return 0;
2704 }
2705
2706 /* must be called once on fs mount.  calls journal_read for you */
2707 int journal_init(struct super_block *sb, const char *j_dev_name,
2708                  int old_format, unsigned int commit_max_age)
2709 {
2710         int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2711         struct buffer_head *bhjh;
2712         struct reiserfs_super_block *rs;
2713         struct reiserfs_journal_header *jh;
2714         struct reiserfs_journal *journal;
2715         struct reiserfs_journal_list *jl;
2716         int ret;
2717
2718         journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2719         if (!journal) {
2720                 reiserfs_warning(sb, "journal-1256",
2721                                  "unable to get memory for journal structure");
2722                 return 1;
2723         }
2724         INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2725         INIT_LIST_HEAD(&journal->j_prealloc_list);
2726         INIT_LIST_HEAD(&journal->j_working_list);
2727         INIT_LIST_HEAD(&journal->j_journal_list);
2728         journal->j_persistent_trans = 0;
2729         if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2730                                            reiserfs_bmap_count(sb)))
2731                 goto free_and_return;
2732
2733         allocate_bitmap_nodes(sb);
2734
2735         /* reserved for journal area support */
2736         SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2737                                                  REISERFS_OLD_DISK_OFFSET_IN_BYTES
2738                                                  / sb->s_blocksize +
2739                                                  reiserfs_bmap_count(sb) +
2740                                                  1 :
2741                                                  REISERFS_DISK_OFFSET_IN_BYTES /
2742                                                  sb->s_blocksize + 2);
2743
2744         /*
2745          * Sanity check to see is the standard journal fitting
2746          * within first bitmap (actual for small blocksizes)
2747          */
2748         if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2749             (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2750              SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2751                 reiserfs_warning(sb, "journal-1393",
2752                                  "journal does not fit for area addressed "
2753                                  "by first of bitmap blocks. It starts at "
2754                                  "%u and its size is %u. Block size %ld",
2755                                  SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2756                                  SB_ONDISK_JOURNAL_SIZE(sb),
2757                                  sb->s_blocksize);
2758                 goto free_and_return;
2759         }
2760
2761         if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2762                 reiserfs_warning(sb, "sh-462",
2763                                  "unable to initialize journal device");
2764                 goto free_and_return;
2765         }
2766
2767         rs = SB_DISK_SUPER_BLOCK(sb);
2768
2769         /* read journal header */
2770         bhjh = journal_bread(sb,
2771                              SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2772                              SB_ONDISK_JOURNAL_SIZE(sb));
2773         if (!bhjh) {
2774                 reiserfs_warning(sb, "sh-459",
2775                                  "unable to read journal header");
2776                 goto free_and_return;
2777         }
2778         jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2779
2780         /* make sure that journal matches to the super block */
2781         if (is_reiserfs_jr(rs)
2782             && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2783                 sb_jp_journal_magic(rs))) {
2784                 reiserfs_warning(sb, "sh-460",
2785                                  "journal header magic %x (device %pg) does "
2786                                  "not match to magic found in super block %x",
2787                                  jh->jh_journal.jp_journal_magic,
2788                                  journal->j_dev_bd,
2789                                  sb_jp_journal_magic(rs));
2790                 brelse(bhjh);
2791                 goto free_and_return;
2792         }
2793
2794         journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2795         journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2796         journal->j_max_commit_age =
2797             le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2798         journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2799
2800         if (check_advise_trans_params(sb, journal) != 0)
2801                 goto free_and_return;
2802         journal->j_default_max_commit_age = journal->j_max_commit_age;
2803
2804         if (commit_max_age != 0) {
2805                 journal->j_max_commit_age = commit_max_age;
2806                 journal->j_max_trans_age = commit_max_age;
2807         }
2808
2809         reiserfs_info(sb, "journal params: device %pg, size %u, "
2810                       "journal first block %u, max trans len %u, max batch %u, "
2811                       "max commit age %u, max trans age %u\n",
2812                       journal->j_dev_bd,
2813                       SB_ONDISK_JOURNAL_SIZE(sb),
2814                       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2815                       journal->j_trans_max,
2816                       journal->j_max_batch,
2817                       journal->j_max_commit_age, journal->j_max_trans_age);
2818
2819         brelse(bhjh);
2820
2821         journal->j_list_bitmap_index = 0;
2822         journal_list_init(sb);
2823
2824         memset(journal->j_list_hash_table, 0,
2825                JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2826
2827         INIT_LIST_HEAD(&journal->j_dirty_buffers);
2828         spin_lock_init(&journal->j_dirty_buffers_lock);
2829
2830         journal->j_start = 0;
2831         journal->j_len = 0;
2832         journal->j_len_alloc = 0;
2833         atomic_set(&journal->j_wcount, 0);
2834         atomic_set(&journal->j_async_throttle, 0);
2835         journal->j_bcount = 0;
2836         journal->j_trans_start_time = 0;
2837         journal->j_last = NULL;
2838         journal->j_first = NULL;
2839         init_waitqueue_head(&journal->j_join_wait);
2840         mutex_init(&journal->j_mutex);
2841         mutex_init(&journal->j_flush_mutex);
2842
2843         journal->j_trans_id = 10;
2844         journal->j_mount_id = 10;
2845         journal->j_state = 0;
2846         atomic_set(&journal->j_jlock, 0);
2847         journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2848         journal->j_cnode_free_orig = journal->j_cnode_free_list;
2849         journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2850         journal->j_cnode_used = 0;
2851         journal->j_must_wait = 0;
2852
2853         if (journal->j_cnode_free == 0) {
2854                 reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2855                                  "allocation failed (%ld bytes). Journal is "
2856                                  "too large for available memory. Usually "
2857                                  "this is due to a journal that is too large.",
2858                                  sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2859                 goto free_and_return;
2860         }
2861
2862         init_journal_hash(sb);
2863         jl = journal->j_current_jl;
2864
2865         /*
2866          * get_list_bitmap() may call flush_commit_list() which
2867          * requires the lock. Calling flush_commit_list() shouldn't happen
2868          * this early but I like to be paranoid.
2869          */
2870         reiserfs_write_lock(sb);
2871         jl->j_list_bitmap = get_list_bitmap(sb, jl);
2872         reiserfs_write_unlock(sb);
2873         if (!jl->j_list_bitmap) {
2874                 reiserfs_warning(sb, "journal-2005",
2875                                  "get_list_bitmap failed for journal list 0");
2876                 goto free_and_return;
2877         }
2878
2879         ret = journal_read(sb);
2880         if (ret < 0) {
2881                 reiserfs_warning(sb, "reiserfs-2006",
2882                                  "Replay Failure, unable to mount");
2883                 goto free_and_return;
2884         }
2885
2886         INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2887         journal->j_work_sb = sb;
2888         return 0;
2889 free_and_return:
2890         free_journal_ram(sb);
2891         return 1;
2892 }
2893
2894 /*
2895  * test for a polite end of the current transaction.  Used by file_write,
2896  * and should be used by delete to make sure they don't write more than
2897  * can fit inside a single transaction
2898  */
2899 int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2900                                    int new_alloc)
2901 {
2902         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2903         time64_t now = ktime_get_seconds();
2904         /* cannot restart while nested */
2905         BUG_ON(!th->t_trans_id);
2906         if (th->t_refcount > 1)
2907                 return 0;
2908         if (journal->j_must_wait > 0 ||
2909             (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2910             atomic_read(&journal->j_jlock) ||
2911             (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2912             journal->j_cnode_free < (journal->j_trans_max * 3)) {
2913                 return 1;
2914         }
2915
2916         journal->j_len_alloc += new_alloc;
2917         th->t_blocks_allocated += new_alloc ;
2918         return 0;
2919 }
2920
2921 /* this must be called inside a transaction */
2922 void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2923 {
2924         struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2925         BUG_ON(!th->t_trans_id);
2926         journal->j_must_wait = 1;
2927         set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2928         return;
2929 }
2930
2931 /* this must be called without a transaction started */
2932 void reiserfs_allow_writes(struct super_block *s)
2933 {
2934         struct reiserfs_journal *journal = SB_JOURNAL(s);
2935         clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2936         wake_up(&journal->j_join_wait);
2937 }
2938
2939 /* this must be called without a transaction started */
2940 void reiserfs_wait_on_write_block(struct super_block *s)
2941 {
2942         struct reiserfs_journal *journal = SB_JOURNAL(s);
2943         wait_event(journal->j_join_wait,
2944                    !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2945 }
2946
2947 static void queue_log_writer(struct super_block *s)
2948 {
2949         wait_queue_entry_t wait;
2950         struct reiserfs_journal *journal = SB_JOURNAL(s);
2951         set_bit(J_WRITERS_QUEUED, &journal->j_state);
2952
2953         /*
2954          * we don't want to use wait_event here because
2955          * we only want to wait once.
2956          */
2957         init_waitqueue_entry(&wait, current);
2958         add_wait_queue(&journal->j_join_wait, &wait);
2959         set_current_state(TASK_UNINTERRUPTIBLE);
2960         if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2961                 int depth = reiserfs_write_unlock_nested(s);
2962                 schedule();
2963                 reiserfs_write_lock_nested(s, depth);
2964         }
2965         __set_current_state(TASK_RUNNING);
2966         remove_wait_queue(&journal->j_join_wait, &wait);
2967 }
2968
2969 static void wake_queued_writers(struct super_block *s)
2970 {
2971         struct reiserfs_journal *journal = SB_JOURNAL(s);
2972         if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2973                 wake_up(&journal->j_join_wait);
2974 }
2975
2976 static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2977 {
2978         struct reiserfs_journal *journal = SB_JOURNAL(sb);
2979         unsigned long bcount = journal->j_bcount;
2980         while (1) {
2981                 int depth;
2982
2983                 depth = reiserfs_write_unlock_nested(sb);
2984                 schedule_timeout_uninterruptible(1);
2985                 reiserfs_write_lock_nested(sb, depth);
2986
2987                 journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2988                 while ((atomic_read(&journal->j_wcount) > 0 ||
2989                         atomic_read(&journal->j_jlock)) &&
2990                        journal->j_trans_id == trans_id) {
2991                         queue_log_writer(sb);
2992                 }
2993                 if (journal->j_trans_id != trans_id)
2994                         break;
2995                 if (bcount == journal->j_bcount)
2996                         break;
2997                 bcount = journal->j_bcount;
2998         }
2999 }
3000
3001 /*
3002  * join == true if you must join an existing transaction.
3003  * join == false if you can deal with waiting for others to finish
3004  *
3005  * this will block until the transaction is joinable.  send the number of
3006  * blocks you expect to use in nblocks.
3007 */
3008 static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3009                               struct super_block *sb, unsigned long nblocks,
3010                               int join)
3011 {
3012         time64_t now = ktime_get_seconds();
3013         unsigned int old_trans_id;
3014         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3015         struct reiserfs_transaction_handle myth;
3016         int sched_count = 0;
3017         int retval;
3018         int depth;
3019
3020         reiserfs_check_lock_depth(sb, "journal_begin");
3021         BUG_ON(nblocks > journal->j_trans_max);
3022
3023         PROC_INFO_INC(sb, journal.journal_being);
3024         /* set here for journal_join */
3025         th->t_refcount = 1;
3026         th->t_super = sb;
3027
3028 relock:
3029         lock_journal(sb);
3030         if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3031                 unlock_journal(sb);
3032                 retval = journal->j_errno;
3033                 goto out_fail;
3034         }
3035         journal->j_bcount++;
3036
3037         if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3038                 unlock_journal(sb);
3039                 depth = reiserfs_write_unlock_nested(sb);
3040                 reiserfs_wait_on_write_block(sb);
3041                 reiserfs_write_lock_nested(sb, depth);
3042                 PROC_INFO_INC(sb, journal.journal_relock_writers);
3043                 goto relock;
3044         }
3045         now = ktime_get_seconds();
3046
3047         /*
3048          * if there is no room in the journal OR
3049          * if this transaction is too old, and we weren't called joinable,
3050          * wait for it to finish before beginning we don't sleep if there
3051          * aren't other writers
3052          */
3053
3054         if ((!join && journal->j_must_wait > 0) ||
3055             (!join
3056              && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3057             || (!join && atomic_read(&journal->j_wcount) > 0
3058                 && journal->j_trans_start_time > 0
3059                 && (now - journal->j_trans_start_time) >
3060                 journal->j_max_trans_age) || (!join
3061                                               && atomic_read(&journal->j_jlock))
3062             || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3063
3064                 old_trans_id = journal->j_trans_id;
3065                 /* allow others to finish this transaction */
3066                 unlock_journal(sb);
3067
3068                 if (!join && (journal->j_len_alloc + nblocks + 2) >=
3069                     journal->j_max_batch &&
3070                     ((journal->j_len + nblocks + 2) * 100) <
3071                     (journal->j_len_alloc * 75)) {
3072                         if (atomic_read(&journal->j_wcount) > 10) {
3073                                 sched_count++;
3074                                 queue_log_writer(sb);
3075                                 goto relock;
3076                         }
3077                 }
3078                 /*
3079                  * don't mess with joining the transaction if all we
3080                  * have to do is wait for someone else to do a commit
3081                  */
3082                 if (atomic_read(&journal->j_jlock)) {
3083                         while (journal->j_trans_id == old_trans_id &&
3084                                atomic_read(&journal->j_jlock)) {
3085                                 queue_log_writer(sb);
3086                         }
3087                         goto relock;
3088                 }
3089                 retval = journal_join(&myth, sb);
3090                 if (retval)
3091                         goto out_fail;
3092
3093                 /* someone might have ended the transaction while we joined */
3094                 if (old_trans_id != journal->j_trans_id) {
3095                         retval = do_journal_end(&myth, 0);
3096                 } else {
3097                         retval = do_journal_end(&myth, COMMIT_NOW);
3098                 }
3099
3100                 if (retval)
3101                         goto out_fail;
3102
3103                 PROC_INFO_INC(sb, journal.journal_relock_wcount);
3104                 goto relock;
3105         }
3106         /* we are the first writer, set trans_id */
3107         if (journal->j_trans_start_time == 0) {
3108                 journal->j_trans_start_time = ktime_get_seconds();
3109         }
3110         atomic_inc(&journal->j_wcount);
3111         journal->j_len_alloc += nblocks;
3112         th->t_blocks_logged = 0;
3113         th->t_blocks_allocated = nblocks;
3114         th->t_trans_id = journal->j_trans_id;
3115         unlock_journal(sb);
3116         INIT_LIST_HEAD(&th->t_list);
3117         return 0;
3118
3119 out_fail:
3120         memset(th, 0, sizeof(*th));
3121         /*
3122          * Re-set th->t_super, so we can properly keep track of how many
3123          * persistent transactions there are. We need to do this so if this
3124          * call is part of a failed restart_transaction, we can free it later
3125          */
3126         th->t_super = sb;
3127         return retval;
3128 }
3129
3130 struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3131                                                                     super_block
3132                                                                     *s,
3133                                                                     int nblocks)
3134 {
3135         int ret;
3136         struct reiserfs_transaction_handle *th;
3137
3138         /*
3139          * if we're nesting into an existing transaction.  It will be
3140          * persistent on its own
3141          */
3142         if (reiserfs_transaction_running(s)) {
3143                 th = current->journal_info;
3144                 th->t_refcount++;
3145                 BUG_ON(th->t_refcount < 2);
3146
3147                 return th;
3148         }
3149         th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3150         if (!th)
3151                 return NULL;
3152         ret = journal_begin(th, s, nblocks);
3153         if (ret) {
3154                 kfree(th);
3155                 return NULL;
3156         }
3157
3158         SB_JOURNAL(s)->j_persistent_trans++;
3159         return th;
3160 }
3161
3162 int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3163 {
3164         struct super_block *s = th->t_super;
3165         int ret = 0;
3166         if (th->t_trans_id)
3167                 ret = journal_end(th);
3168         else
3169                 ret = -EIO;
3170         if (th->t_refcount == 0) {
3171                 SB_JOURNAL(s)->j_persistent_trans--;
3172                 kfree(th);
3173         }
3174         return ret;
3175 }
3176
3177 static int journal_join(struct reiserfs_transaction_handle *th,
3178                         struct super_block *sb)
3179 {
3180         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3181
3182         /*
3183          * this keeps do_journal_end from NULLing out the
3184          * current->journal_info pointer
3185          */
3186         th->t_handle_save = cur_th;
3187         BUG_ON(cur_th && cur_th->t_refcount > 1);
3188         return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3189 }
3190
3191 int journal_join_abort(struct reiserfs_transaction_handle *th,
3192                        struct super_block *sb)
3193 {
3194         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3195
3196         /*
3197          * this keeps do_journal_end from NULLing out the
3198          * current->journal_info pointer
3199          */
3200         th->t_handle_save = cur_th;
3201         BUG_ON(cur_th && cur_th->t_refcount > 1);
3202         return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3203 }
3204
3205 int journal_begin(struct reiserfs_transaction_handle *th,
3206                   struct super_block *sb, unsigned long nblocks)
3207 {
3208         struct reiserfs_transaction_handle *cur_th = current->journal_info;
3209         int ret;
3210
3211         th->t_handle_save = NULL;
3212         if (cur_th) {
3213                 /* we are nesting into the current transaction */
3214                 if (cur_th->t_super == sb) {
3215                         BUG_ON(!cur_th->t_refcount);
3216                         cur_th->t_refcount++;
3217                         memcpy(th, cur_th, sizeof(*th));
3218                         if (th->t_refcount <= 1)
3219                                 reiserfs_warning(sb, "reiserfs-2005",
3220                                                  "BAD: refcount <= 1, but "
3221                                                  "journal_info != 0");
3222                         return 0;
3223                 } else {
3224                         /*
3225                          * we've ended up with a handle from a different
3226                          * filesystem.  save it and restore on journal_end.
3227                          * This should never really happen...
3228                          */
3229                         reiserfs_warning(sb, "clm-2100",
3230                                          "nesting info a different FS");
3231                         th->t_handle_save = current->journal_info;
3232                         current->journal_info = th;
3233                 }
3234         } else {
3235                 current->journal_info = th;
3236         }
3237         ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3238         BUG_ON(current->journal_info != th);
3239
3240         /*
3241          * I guess this boils down to being the reciprocal of clm-2100 above.
3242          * If do_journal_begin_r fails, we need to put it back, since
3243          * journal_end won't be called to do it. */
3244         if (ret)
3245                 current->journal_info = th->t_handle_save;
3246         else
3247                 BUG_ON(!th->t_refcount);
3248
3249         return ret;
3250 }
3251
3252 /*
3253  * puts bh into the current transaction.  If it was already there, reorders
3254  * removes the old pointers from the hash, and puts new ones in (to make
3255  * sure replay happen in the right order).
3256  *
3257  * if it was dirty, cleans and files onto the clean list.  I can't let it
3258  * be dirty again until the transaction is committed.
3259  *
3260  * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3261  */
3262 int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3263                        struct buffer_head *bh)
3264 {
3265         struct super_block *sb = th->t_super;
3266         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3267         struct reiserfs_journal_cnode *cn = NULL;
3268         int count_already_incd = 0;
3269         int prepared = 0;
3270         BUG_ON(!th->t_trans_id);
3271
3272         PROC_INFO_INC(sb, journal.mark_dirty);
3273         if (th->t_trans_id != journal->j_trans_id) {
3274                 reiserfs_panic(th->t_super, "journal-1577",
3275                                "handle trans id %ld != current trans id %ld",
3276                                th->t_trans_id, journal->j_trans_id);
3277         }
3278
3279         prepared = test_clear_buffer_journal_prepared(bh);
3280         clear_buffer_journal_restore_dirty(bh);
3281         /* already in this transaction, we are done */
3282         if (buffer_journaled(bh)) {
3283                 PROC_INFO_INC(sb, journal.mark_dirty_already);
3284                 return 0;
3285         }
3286
3287         /*
3288          * this must be turned into a panic instead of a warning.  We can't
3289          * allow a dirty or journal_dirty or locked buffer to be logged, as
3290          * some changes could get to disk too early.  NOT GOOD.
3291          */
3292         if (!prepared || buffer_dirty(bh)) {
3293                 reiserfs_warning(sb, "journal-1777",
3294                                  "buffer %llu bad state "
3295                                  "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3296                                  (unsigned long long)bh->b_blocknr,
3297                                  prepared ? ' ' : '!',
3298                                  buffer_locked(bh) ? ' ' : '!',
3299                                  buffer_dirty(bh) ? ' ' : '!',
3300                                  buffer_journal_dirty(bh) ? ' ' : '!');
3301         }
3302
3303         if (atomic_read(&journal->j_wcount) <= 0) {
3304                 reiserfs_warning(sb, "journal-1409",
3305                                  "returning because j_wcount was %d",
3306                                  atomic_read(&journal->j_wcount));
3307                 return 1;
3308         }
3309         /*
3310          * this error means I've screwed up, and we've overflowed
3311          * the transaction.  Nothing can be done here, except make the
3312          * FS readonly or panic.
3313          */
3314         if (journal->j_len >= journal->j_trans_max) {
3315                 reiserfs_panic(th->t_super, "journal-1413",
3316                                "j_len (%lu) is too big",
3317                                journal->j_len);
3318         }
3319
3320         if (buffer_journal_dirty(bh)) {
3321                 count_already_incd = 1;
3322                 PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3323                 clear_buffer_journal_dirty(bh);
3324         }
3325
3326         if (journal->j_len > journal->j_len_alloc) {
3327                 journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3328         }
3329
3330         set_buffer_journaled(bh);
3331
3332         /* now put this guy on the end */
3333         if (!cn) {
3334                 cn = get_cnode(sb);
3335                 if (!cn) {
3336                         reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3337                 }
3338
3339                 if (th->t_blocks_logged == th->t_blocks_allocated) {
3340                         th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3341                         journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3342                 }
3343                 th->t_blocks_logged++;
3344                 journal->j_len++;
3345
3346                 cn->bh = bh;
3347                 cn->blocknr = bh->b_blocknr;
3348                 cn->sb = sb;
3349                 cn->jlist = NULL;
3350                 insert_journal_hash(journal->j_hash_table, cn);
3351                 if (!count_already_incd) {
3352                         get_bh(bh);
3353                 }
3354         }
3355         cn->next = NULL;
3356         cn->prev = journal->j_last;
3357         cn->bh = bh;
3358         if (journal->j_last) {
3359                 journal->j_last->next = cn;
3360                 journal->j_last = cn;
3361         } else {
3362                 journal->j_first = cn;
3363                 journal->j_last = cn;
3364         }
3365         reiserfs_schedule_old_flush(sb);
3366         return 0;
3367 }
3368
3369 int journal_end(struct reiserfs_transaction_handle *th)
3370 {
3371         struct super_block *sb = th->t_super;
3372         if (!current->journal_info && th->t_refcount > 1)
3373                 reiserfs_warning(sb, "REISER-NESTING",
3374                                  "th NULL, refcount %d", th->t_refcount);
3375
3376         if (!th->t_trans_id) {
3377                 WARN_ON(1);
3378                 return -EIO;
3379         }
3380
3381         th->t_refcount--;
3382         if (th->t_refcount > 0) {
3383                 struct reiserfs_transaction_handle *cur_th =
3384                     current->journal_info;
3385
3386                 /*
3387                  * we aren't allowed to close a nested transaction on a
3388                  * different filesystem from the one in the task struct
3389                  */
3390                 BUG_ON(cur_th->t_super != th->t_super);
3391
3392                 if (th != cur_th) {
3393                         memcpy(current->journal_info, th, sizeof(*th));
3394                         th->t_trans_id = 0;
3395                 }
3396                 return 0;
3397         } else {
3398                 return do_journal_end(th, 0);
3399         }
3400 }
3401
3402 /*
3403  * removes from the current transaction, relsing and descrementing any counters.
3404  * also files the removed buffer directly onto the clean list
3405  *
3406  * called by journal_mark_freed when a block has been deleted
3407  *
3408  * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3409  */
3410 static int remove_from_transaction(struct super_block *sb,
3411                                    b_blocknr_t blocknr, int already_cleaned)
3412 {
3413         struct buffer_head *bh;
3414         struct reiserfs_journal_cnode *cn;
3415         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3416         int ret = 0;
3417
3418         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3419         if (!cn || !cn->bh) {
3420                 return ret;
3421         }
3422         bh = cn->bh;
3423         if (cn->prev) {
3424                 cn->prev->next = cn->next;
3425         }
3426         if (cn->next) {
3427                 cn->next->prev = cn->prev;
3428         }
3429         if (cn == journal->j_first) {
3430                 journal->j_first = cn->next;
3431         }
3432         if (cn == journal->j_last) {
3433                 journal->j_last = cn->prev;
3434         }
3435         remove_journal_hash(sb, journal->j_hash_table, NULL,
3436                             bh->b_blocknr, 0);
3437         clear_buffer_journaled(bh);     /* don't log this one */
3438
3439         if (!already_cleaned) {
3440                 clear_buffer_journal_dirty(bh);
3441                 clear_buffer_dirty(bh);
3442                 clear_buffer_journal_test(bh);
3443                 put_bh(bh);
3444                 if (atomic_read(&bh->b_count) < 0) {
3445                         reiserfs_warning(sb, "journal-1752",
3446                                          "b_count < 0");
3447                 }
3448                 ret = 1;
3449         }
3450         journal->j_len--;
3451         journal->j_len_alloc--;
3452         free_cnode(sb, cn);
3453         return ret;
3454 }
3455
3456 /*
3457  * for any cnode in a journal list, it can only be dirtied of all the
3458  * transactions that include it are committed to disk.
3459  * this checks through each transaction, and returns 1 if you are allowed
3460  * to dirty, and 0 if you aren't
3461  *
3462  * it is called by dirty_journal_list, which is called after
3463  * flush_commit_list has gotten all the log blocks for a given
3464  * transaction on disk
3465  *
3466  */
3467 static int can_dirty(struct reiserfs_journal_cnode *cn)
3468 {
3469         struct super_block *sb = cn->sb;
3470         b_blocknr_t blocknr = cn->blocknr;
3471         struct reiserfs_journal_cnode *cur = cn->hprev;
3472         int can_dirty = 1;
3473
3474         /*
3475          * first test hprev.  These are all newer than cn, so any node here
3476          * with the same block number and dev means this node can't be sent
3477          * to disk right now.
3478          */
3479         while (cur && can_dirty) {
3480                 if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3481                     cur->blocknr == blocknr) {
3482                         can_dirty = 0;
3483                 }
3484                 cur = cur->hprev;
3485         }
3486         /*
3487          * then test hnext.  These are all older than cn.  As long as they
3488          * are committed to the log, it is safe to write cn to disk
3489          */
3490         cur = cn->hnext;
3491         while (cur && can_dirty) {
3492                 if (cur->jlist && cur->jlist->j_len > 0 &&
3493                     atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3494                     cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3495                         can_dirty = 0;
3496                 }
3497                 cur = cur->hnext;
3498         }
3499         return can_dirty;
3500 }
3501
3502 /*
3503  * syncs the commit blocks, but does not force the real buffers to disk
3504  * will wait until the current transaction is done/committed before returning
3505  */
3506 int journal_end_sync(struct reiserfs_transaction_handle *th)
3507 {
3508         struct super_block *sb = th->t_super;
3509         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3510
3511         BUG_ON(!th->t_trans_id);
3512         /* you can sync while nested, very, very bad */
3513         BUG_ON(th->t_refcount > 1);
3514         if (journal->j_len == 0) {
3515                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3516                                              1);
3517                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3518         }
3519         return do_journal_end(th, COMMIT_NOW | WAIT);
3520 }
3521
3522 /* writeback the pending async commits to disk */
3523 static void flush_async_commits(struct work_struct *work)
3524 {
3525         struct reiserfs_journal *journal =
3526                 container_of(work, struct reiserfs_journal, j_work.work);
3527         struct super_block *sb = journal->j_work_sb;
3528         struct reiserfs_journal_list *jl;
3529         struct list_head *entry;
3530
3531         reiserfs_write_lock(sb);
3532         if (!list_empty(&journal->j_journal_list)) {
3533                 /* last entry is the youngest, commit it and you get everything */
3534                 entry = journal->j_journal_list.prev;
3535                 jl = JOURNAL_LIST_ENTRY(entry);
3536                 flush_commit_list(sb, jl, 1);
3537         }
3538         reiserfs_write_unlock(sb);
3539 }
3540
3541 /*
3542  * flushes any old transactions to disk
3543  * ends the current transaction if it is too old
3544  */
3545 void reiserfs_flush_old_commits(struct super_block *sb)
3546 {
3547         time64_t now;
3548         struct reiserfs_transaction_handle th;
3549         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3550
3551         now = ktime_get_seconds();
3552         /*
3553          * safety check so we don't flush while we are replaying the log during
3554          * mount
3555          */
3556         if (list_empty(&journal->j_journal_list))
3557                 return;
3558
3559         /*
3560          * check the current transaction.  If there are no writers, and it is
3561          * too old, finish it, and force the commit blocks to disk
3562          */
3563         if (atomic_read(&journal->j_wcount) <= 0 &&
3564             journal->j_trans_start_time > 0 &&
3565             journal->j_len > 0 &&
3566             (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3567                 if (!journal_join(&th, sb)) {
3568                         reiserfs_prepare_for_journal(sb,
3569                                                      SB_BUFFER_WITH_SB(sb),
3570                                                      1);
3571                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3572
3573                         /*
3574                          * we're only being called from kreiserfsd, it makes
3575                          * no sense to do an async commit so that kreiserfsd
3576                          * can do it later
3577                          */
3578                         do_journal_end(&th, COMMIT_NOW | WAIT);
3579                 }
3580         }
3581 }
3582
3583 /*
3584  * returns 0 if do_journal_end should return right away, returns 1 if
3585  * do_journal_end should finish the commit
3586  *
3587  * if the current transaction is too old, but still has writers, this will
3588  * wait on j_join_wait until all the writers are done.  By the time it
3589  * wakes up, the transaction it was called has already ended, so it just
3590  * flushes the commit list and returns 0.
3591  *
3592  * Won't batch when flush or commit_now is set.  Also won't batch when
3593  * others are waiting on j_join_wait.
3594  *
3595  * Note, we can't allow the journal_end to proceed while there are still
3596  * writers in the log.
3597  */
3598 static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3599 {
3600
3601         time64_t now;
3602         int flush = flags & FLUSH_ALL;
3603         int commit_now = flags & COMMIT_NOW;
3604         int wait_on_commit = flags & WAIT;
3605         struct reiserfs_journal_list *jl;
3606         struct super_block *sb = th->t_super;
3607         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3608
3609         BUG_ON(!th->t_trans_id);
3610
3611         if (th->t_trans_id != journal->j_trans_id) {
3612                 reiserfs_panic(th->t_super, "journal-1577",
3613                                "handle trans id %ld != current trans id %ld",
3614                                th->t_trans_id, journal->j_trans_id);
3615         }
3616
3617         journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3618         /* <= 0 is allowed.  unmounting might not call begin */
3619         if (atomic_read(&journal->j_wcount) > 0)
3620                 atomic_dec(&journal->j_wcount);
3621
3622         /*
3623          * BUG, deal with case where j_len is 0, but people previously
3624          * freed blocks need to be released will be dealt with by next
3625          * transaction that actually writes something, but should be taken
3626          * care of in this trans
3627          */
3628         BUG_ON(journal->j_len == 0);
3629
3630         /*
3631          * if wcount > 0, and we are called to with flush or commit_now,
3632          * we wait on j_join_wait.  We will wake up when the last writer has
3633          * finished the transaction, and started it on its way to the disk.
3634          * Then, we flush the commit or journal list, and just return 0
3635          * because the rest of journal end was already done for this
3636          * transaction.
3637          */
3638         if (atomic_read(&journal->j_wcount) > 0) {
3639                 if (flush || commit_now) {
3640                         unsigned trans_id;
3641
3642                         jl = journal->j_current_jl;
3643                         trans_id = jl->j_trans_id;
3644                         if (wait_on_commit)
3645                                 jl->j_state |= LIST_COMMIT_PENDING;
3646                         atomic_set(&journal->j_jlock, 1);
3647                         if (flush) {
3648                                 journal->j_next_full_flush = 1;
3649                         }
3650                         unlock_journal(sb);
3651
3652                         /*
3653                          * sleep while the current transaction is
3654                          * still j_jlocked
3655                          */
3656                         while (journal->j_trans_id == trans_id) {
3657                                 if (atomic_read(&journal->j_jlock)) {
3658                                         queue_log_writer(sb);
3659                                 } else {
3660                                         lock_journal(sb);
3661                                         if (journal->j_trans_id == trans_id) {
3662                                                 atomic_set(&journal->j_jlock,
3663                                                            1);
3664                                         }
3665                                         unlock_journal(sb);
3666                                 }
3667                         }
3668                         BUG_ON(journal->j_trans_id == trans_id);
3669
3670                         if (commit_now
3671                             && journal_list_still_alive(sb, trans_id)
3672                             && wait_on_commit) {
3673                                 flush_commit_list(sb, jl, 1);
3674                         }
3675                         return 0;
3676                 }
3677                 unlock_journal(sb);
3678                 return 0;
3679         }
3680
3681         /* deal with old transactions where we are the last writers */
3682         now = ktime_get_seconds();
3683         if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3684                 commit_now = 1;
3685                 journal->j_next_async_flush = 1;
3686         }
3687         /* don't batch when someone is waiting on j_join_wait */
3688         /* don't batch when syncing the commit or flushing the whole trans */
3689         if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3690             && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3691             && journal->j_len_alloc < journal->j_max_batch
3692             && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3693                 journal->j_bcount++;
3694                 unlock_journal(sb);
3695                 return 0;
3696         }
3697
3698         if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3699                 reiserfs_panic(sb, "journal-003",
3700                                "j_start (%ld) is too high",
3701                                journal->j_start);
3702         }
3703         return 1;
3704 }
3705
3706 /*
3707  * Does all the work that makes deleting blocks safe.
3708  * when deleting a block mark BH_JNew, just remove it from the current
3709  * transaction, clean it's buffer_head and move on.
3710  *
3711  * otherwise:
3712  * set a bit for the block in the journal bitmap.  That will prevent it from
3713  * being allocated for unformatted nodes before this transaction has finished.
3714  *
3715  * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3716  * That will prevent any old transactions with this block from trying to flush
3717  * to the real location.  Since we aren't removing the cnode from the
3718  * journal_list_hash, *the block can't be reallocated yet.
3719  *
3720  * Then remove it from the current transaction, decrementing any counters and
3721  * filing it on the clean list.
3722  */
3723 int journal_mark_freed(struct reiserfs_transaction_handle *th,
3724                        struct super_block *sb, b_blocknr_t blocknr)
3725 {
3726         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3727         struct reiserfs_journal_cnode *cn = NULL;
3728         struct buffer_head *bh = NULL;
3729         struct reiserfs_list_bitmap *jb = NULL;
3730         int cleaned = 0;
3731         BUG_ON(!th->t_trans_id);
3732
3733         cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3734         if (cn && cn->bh) {
3735                 bh = cn->bh;
3736                 get_bh(bh);
3737         }
3738         /* if it is journal new, we just remove it from this transaction */
3739         if (bh && buffer_journal_new(bh)) {
3740                 clear_buffer_journal_new(bh);
3741                 clear_prepared_bits(bh);
3742                 reiserfs_clean_and_file_buffer(bh);
3743                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3744         } else {
3745                 /*
3746                  * set the bit for this block in the journal bitmap
3747                  * for this transaction
3748                  */
3749                 jb = journal->j_current_jl->j_list_bitmap;
3750                 if (!jb) {
3751                         reiserfs_panic(sb, "journal-1702",
3752                                        "journal_list_bitmap is NULL");
3753                 }
3754                 set_bit_in_list_bitmap(sb, blocknr, jb);
3755
3756                 /* Note, the entire while loop is not allowed to schedule.  */
3757
3758                 if (bh) {
3759                         clear_prepared_bits(bh);
3760                         reiserfs_clean_and_file_buffer(bh);
3761                 }
3762                 cleaned = remove_from_transaction(sb, blocknr, cleaned);
3763
3764                 /*
3765                  * find all older transactions with this block,
3766                  * make sure they don't try to write it out
3767                  */
3768                 cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3769                                           blocknr);
3770                 while (cn) {
3771                         if (sb == cn->sb && blocknr == cn->blocknr) {
3772                                 set_bit(BLOCK_FREED, &cn->state);
3773                                 if (cn->bh) {
3774                                         /*
3775                                          * remove_from_transaction will brelse
3776                                          * the buffer if it was in the current
3777                                          * trans
3778                                          */
3779                                         if (!cleaned) {
3780                                                 clear_buffer_journal_dirty(cn->
3781                                                                            bh);
3782                                                 clear_buffer_dirty(cn->bh);
3783                                                 clear_buffer_journal_test(cn->
3784                                                                           bh);
3785                                                 cleaned = 1;
3786                                                 put_bh(cn->bh);
3787                                                 if (atomic_read
3788                                                     (&cn->bh->b_count) < 0) {
3789                                                         reiserfs_warning(sb,
3790                                                                  "journal-2138",
3791                                                                  "cn->bh->b_count < 0");
3792                                                 }
3793                                         }
3794                                         /*
3795                                          * since we are clearing the bh,
3796                                          * we MUST dec nonzerolen
3797                                          */
3798                                         if (cn->jlist) {
3799                                                 atomic_dec(&cn->jlist->
3800                                                            j_nonzerolen);
3801                                         }
3802                                         cn->bh = NULL;
3803                                 }
3804                         }
3805                         cn = cn->hnext;
3806                 }
3807         }
3808
3809         if (bh)
3810                 release_buffer_page(bh); /* get_hash grabs the buffer */
3811         return 0;
3812 }
3813
3814 void reiserfs_update_inode_transaction(struct inode *inode)
3815 {
3816         struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3817         REISERFS_I(inode)->i_jl = journal->j_current_jl;
3818         REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3819 }
3820
3821 /*
3822  * returns -1 on error, 0 if no commits/barriers were done and 1
3823  * if a transaction was actually committed and the barrier was done
3824  */
3825 static int __commit_trans_jl(struct inode *inode, unsigned long id,
3826                              struct reiserfs_journal_list *jl)
3827 {
3828         struct reiserfs_transaction_handle th;
3829         struct super_block *sb = inode->i_sb;
3830         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3831         int ret = 0;
3832
3833         /*
3834          * is it from the current transaction,
3835          * or from an unknown transaction?
3836          */
3837         if (id == journal->j_trans_id) {
3838                 jl = journal->j_current_jl;
3839                 /*
3840                  * try to let other writers come in and
3841                  * grow this transaction
3842                  */
3843                 let_transaction_grow(sb, id);
3844                 if (journal->j_trans_id != id) {
3845                         goto flush_commit_only;
3846                 }
3847
3848                 ret = journal_begin(&th, sb, 1);
3849                 if (ret)
3850                         return ret;
3851
3852                 /* someone might have ended this transaction while we joined */
3853                 if (journal->j_trans_id != id) {
3854                         reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3855                                                      1);
3856                         journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3857                         ret = journal_end(&th);
3858                         goto flush_commit_only;
3859                 }
3860
3861                 ret = journal_end_sync(&th);
3862                 if (!ret)
3863                         ret = 1;
3864
3865         } else {
3866                 /*
3867                  * this gets tricky, we have to make sure the journal list in
3868                  * the inode still exists.  We know the list is still around
3869                  * if we've got a larger transaction id than the oldest list
3870                  */
3871 flush_commit_only:
3872                 if (journal_list_still_alive(inode->i_sb, id)) {
3873                         /*
3874                          * we only set ret to 1 when we know for sure
3875                          * the barrier hasn't been started yet on the commit
3876                          * block.
3877                          */
3878                         if (atomic_read(&jl->j_commit_left) > 1)
3879                                 ret = 1;
3880                         flush_commit_list(sb, jl, 1);
3881                         if (journal->j_errno)
3882                                 ret = journal->j_errno;
3883                 }
3884         }
3885         /* otherwise the list is gone, and long since committed */
3886         return ret;
3887 }
3888
3889 int reiserfs_commit_for_inode(struct inode *inode)
3890 {
3891         unsigned int id = REISERFS_I(inode)->i_trans_id;
3892         struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3893
3894         /*
3895          * for the whole inode, assume unset id means it was
3896          * changed in the current transaction.  More conservative
3897          */
3898         if (!id || !jl) {
3899                 reiserfs_update_inode_transaction(inode);
3900                 id = REISERFS_I(inode)->i_trans_id;
3901                 /* jl will be updated in __commit_trans_jl */
3902         }
3903
3904         return __commit_trans_jl(inode, id, jl);
3905 }
3906
3907 void reiserfs_restore_prepared_buffer(struct super_block *sb,
3908                                       struct buffer_head *bh)
3909 {
3910         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3911         PROC_INFO_INC(sb, journal.restore_prepared);
3912         if (!bh) {
3913                 return;
3914         }
3915         if (test_clear_buffer_journal_restore_dirty(bh) &&
3916             buffer_journal_dirty(bh)) {
3917                 struct reiserfs_journal_cnode *cn;
3918                 reiserfs_write_lock(sb);
3919                 cn = get_journal_hash_dev(sb,
3920                                           journal->j_list_hash_table,
3921                                           bh->b_blocknr);
3922                 if (cn && can_dirty(cn)) {
3923                         set_buffer_journal_test(bh);
3924                         mark_buffer_dirty(bh);
3925                 }
3926                 reiserfs_write_unlock(sb);
3927         }
3928         clear_buffer_journal_prepared(bh);
3929 }
3930
3931 extern struct tree_balance *cur_tb;
3932 /*
3933  * before we can change a metadata block, we have to make sure it won't
3934  * be written to disk while we are altering it.  So, we must:
3935  * clean it
3936  * wait on it.
3937  */
3938 int reiserfs_prepare_for_journal(struct super_block *sb,
3939                                  struct buffer_head *bh, int wait)
3940 {
3941         PROC_INFO_INC(sb, journal.prepare);
3942
3943         if (!trylock_buffer(bh)) {
3944                 if (!wait)
3945                         return 0;
3946                 lock_buffer(bh);
3947         }
3948         set_buffer_journal_prepared(bh);
3949         if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3950                 clear_buffer_journal_test(bh);
3951                 set_buffer_journal_restore_dirty(bh);
3952         }
3953         unlock_buffer(bh);
3954         return 1;
3955 }
3956
3957 /*
3958  * long and ugly.  If flush, will not return until all commit
3959  * blocks and all real buffers in the trans are on disk.
3960  * If no_async, won't return until all commit blocks are on disk.
3961  *
3962  * keep reading, there are comments as you go along
3963  *
3964  * If the journal is aborted, we just clean up. Things like flushing
3965  * journal lists, etc just won't happen.
3966  */
3967 static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3968 {
3969         struct super_block *sb = th->t_super;
3970         struct reiserfs_journal *journal = SB_JOURNAL(sb);
3971         struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3972         struct reiserfs_journal_cnode *last_cn = NULL;
3973         struct reiserfs_journal_desc *desc;
3974         struct reiserfs_journal_commit *commit;
3975         struct buffer_head *c_bh;       /* commit bh */
3976         struct buffer_head *d_bh;       /* desc bh */
3977         int cur_write_start = 0;        /* start index of current log write */
3978         int i;
3979         int flush;
3980         int wait_on_commit;
3981         struct reiserfs_journal_list *jl, *temp_jl;
3982         struct list_head *entry, *safe;
3983         unsigned long jindex;
3984         unsigned int commit_trans_id;
3985         int trans_half;
3986         int depth;
3987
3988         BUG_ON(th->t_refcount > 1);
3989         BUG_ON(!th->t_trans_id);
3990         BUG_ON(!th->t_super);
3991
3992         /*
3993          * protect flush_older_commits from doing mistakes if the
3994          * transaction ID counter gets overflowed.
3995          */
3996         if (th->t_trans_id == ~0U)
3997                 flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
3998         flush = flags & FLUSH_ALL;
3999         wait_on_commit = flags & WAIT;
4000
4001         current->journal_info = th->t_handle_save;
4002         reiserfs_check_lock_depth(sb, "journal end");
4003         if (journal->j_len == 0) {
4004                 reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4005                                              1);
4006                 journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4007         }
4008
4009         lock_journal(sb);
4010         if (journal->j_next_full_flush) {
4011                 flags |= FLUSH_ALL;
4012                 flush = 1;
4013         }
4014         if (journal->j_next_async_flush) {
4015                 flags |= COMMIT_NOW | WAIT;
4016                 wait_on_commit = 1;
4017         }
4018
4019         /*
4020          * check_journal_end locks the journal, and unlocks if it does
4021          * not return 1 it tells us if we should continue with the
4022          * journal_end, or just return
4023          */
4024         if (!check_journal_end(th, flags)) {
4025                 reiserfs_schedule_old_flush(sb);
4026                 wake_queued_writers(sb);
4027                 reiserfs_async_progress_wait(sb);
4028                 goto out;
4029         }
4030
4031         /* check_journal_end might set these, check again */
4032         if (journal->j_next_full_flush) {
4033                 flush = 1;
4034         }
4035
4036         /*
4037          * j must wait means we have to flush the log blocks, and the
4038          * real blocks for this transaction
4039          */
4040         if (journal->j_must_wait > 0) {
4041                 flush = 1;
4042         }
4043 #ifdef REISERFS_PREALLOCATE
4044         /*
4045          * quota ops might need to nest, setup the journal_info pointer
4046          * for them and raise the refcount so that it is > 0.
4047          */
4048         current->journal_info = th;
4049         th->t_refcount++;
4050
4051         /* it should not involve new blocks into the transaction */
4052         reiserfs_discard_all_prealloc(th);
4053
4054         th->t_refcount--;
4055         current->journal_info = th->t_handle_save;
4056 #endif
4057
4058         /* setup description block */
4059         d_bh =
4060             journal_getblk(sb,
4061                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4062                            journal->j_start);
4063         set_buffer_uptodate(d_bh);
4064         desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4065         memset(d_bh->b_data, 0, d_bh->b_size);
4066         memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4067         set_desc_trans_id(desc, journal->j_trans_id);
4068
4069         /*
4070          * setup commit block.  Don't write (keep it clean too) this one
4071          * until after everyone else is written
4072          */
4073         c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4074                               ((journal->j_start + journal->j_len +
4075                                 1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4076         commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4077         memset(c_bh->b_data, 0, c_bh->b_size);
4078         set_commit_trans_id(commit, journal->j_trans_id);
4079         set_buffer_uptodate(c_bh);
4080
4081         /* init this journal list */
4082         jl = journal->j_current_jl;
4083
4084         /*
4085          * we lock the commit before doing anything because
4086          * we want to make sure nobody tries to run flush_commit_list until
4087          * the new transaction is fully setup, and we've already flushed the
4088          * ordered bh list
4089          */
4090         reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4091
4092         /* save the transaction id in case we need to commit it later */
4093         commit_trans_id = jl->j_trans_id;
4094
4095         atomic_set(&jl->j_older_commits_done, 0);
4096         jl->j_trans_id = journal->j_trans_id;
4097         jl->j_timestamp = journal->j_trans_start_time;
4098         jl->j_commit_bh = c_bh;
4099         jl->j_start = journal->j_start;
4100         jl->j_len = journal->j_len;
4101         atomic_set(&jl->j_nonzerolen, journal->j_len);
4102         atomic_set(&jl->j_commit_left, journal->j_len + 2);
4103         jl->j_realblock = NULL;
4104
4105         /*
4106          * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4107          * for each real block, add it to the journal list hash,
4108          * copy into real block index array in the commit or desc block
4109          */
4110         trans_half = journal_trans_half(sb->s_blocksize);
4111         for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4112                 if (buffer_journaled(cn->bh)) {
4113                         jl_cn = get_cnode(sb);
4114                         if (!jl_cn) {
4115                                 reiserfs_panic(sb, "journal-1676",
4116                                                "get_cnode returned NULL");
4117                         }
4118                         if (i == 0) {
4119                                 jl->j_realblock = jl_cn;
4120                         }
4121                         jl_cn->prev = last_cn;
4122                         jl_cn->next = NULL;
4123                         if (last_cn) {
4124                                 last_cn->next = jl_cn;
4125                         }
4126                         last_cn = jl_cn;
4127                         /*
4128                          * make sure the block we are trying to log
4129                          * is not a block of journal or reserved area
4130                          */
4131                         if (is_block_in_log_or_reserved_area
4132                             (sb, cn->bh->b_blocknr)) {
4133                                 reiserfs_panic(sb, "journal-2332",
4134                                                "Trying to log block %lu, "
4135                                                "which is a log block",
4136                                                cn->bh->b_blocknr);
4137                         }
4138                         jl_cn->blocknr = cn->bh->b_blocknr;
4139                         jl_cn->state = 0;
4140                         jl_cn->sb = sb;
4141                         jl_cn->bh = cn->bh;
4142                         jl_cn->jlist = jl;
4143                         insert_journal_hash(journal->j_list_hash_table, jl_cn);
4144                         if (i < trans_half) {
4145                                 desc->j_realblock[i] =
4146                                     cpu_to_le32(cn->bh->b_blocknr);
4147                         } else {
4148                                 commit->j_realblock[i - trans_half] =
4149                                     cpu_to_le32(cn->bh->b_blocknr);
4150                         }
4151                 } else {
4152                         i--;
4153                 }
4154         }
4155         set_desc_trans_len(desc, journal->j_len);
4156         set_desc_mount_id(desc, journal->j_mount_id);
4157         set_desc_trans_id(desc, journal->j_trans_id);
4158         set_commit_trans_len(commit, journal->j_len);
4159
4160         /*
4161          * special check in case all buffers in the journal
4162          * were marked for not logging
4163          */
4164         BUG_ON(journal->j_len == 0);
4165
4166         /*
4167          * we're about to dirty all the log blocks, mark the description block
4168          * dirty now too.  Don't mark the commit block dirty until all the
4169          * others are on disk
4170          */
4171         mark_buffer_dirty(d_bh);
4172
4173         /*
4174          * first data block is j_start + 1, so add one to
4175          * cur_write_start wherever you use it
4176          */
4177         cur_write_start = journal->j_start;
4178         cn = journal->j_first;
4179         jindex = 1;     /* start at one so we don't get the desc again */
4180         while (cn) {
4181                 clear_buffer_journal_new(cn->bh);
4182                 /* copy all the real blocks into log area.  dirty log blocks */
4183                 if (buffer_journaled(cn->bh)) {
4184                         struct buffer_head *tmp_bh;
4185                         char *addr;
4186                         struct page *page;
4187                         tmp_bh =
4188                             journal_getblk(sb,
4189                                            SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4190                                            ((cur_write_start +
4191                                              jindex) %
4192                                             SB_ONDISK_JOURNAL_SIZE(sb)));
4193                         set_buffer_uptodate(tmp_bh);
4194                         page = cn->bh->b_page;
4195                         addr = kmap(page);
4196                         memcpy(tmp_bh->b_data,
4197                                addr + offset_in_page(cn->bh->b_data),
4198                                cn->bh->b_size);
4199                         kunmap(page);
4200                         mark_buffer_dirty(tmp_bh);
4201                         jindex++;
4202                         set_buffer_journal_dirty(cn->bh);
4203                         clear_buffer_journaled(cn->bh);
4204                 } else {
4205                         /*
4206                          * JDirty cleared sometime during transaction.
4207                          * don't log this one
4208                          */
4209                         reiserfs_warning(sb, "journal-2048",
4210                                          "BAD, buffer in journal hash, "
4211                                          "but not JDirty!");
4212                         brelse(cn->bh);
4213                 }
4214                 next = cn->next;
4215                 free_cnode(sb, cn);
4216                 cn = next;
4217                 reiserfs_cond_resched(sb);
4218         }
4219
4220         /*
4221          * we are done with both the c_bh and d_bh, but
4222          * c_bh must be written after all other commit blocks,
4223          * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4224          */
4225
4226         journal->j_current_jl = alloc_journal_list(sb);
4227
4228         /* now it is safe to insert this transaction on the main list */
4229         list_add_tail(&jl->j_list, &journal->j_journal_list);
4230         list_add_tail(&jl->j_working_list, &journal->j_working_list);
4231         journal->j_num_work_lists++;
4232
4233         /* reset journal values for the next transaction */
4234         journal->j_start =
4235             (journal->j_start + journal->j_len +
4236              2) % SB_ONDISK_JOURNAL_SIZE(sb);
4237         atomic_set(&journal->j_wcount, 0);
4238         journal->j_bcount = 0;
4239         journal->j_last = NULL;
4240         journal->j_first = NULL;
4241         journal->j_len = 0;
4242         journal->j_trans_start_time = 0;
4243         /* check for trans_id overflow */
4244         if (++journal->j_trans_id == 0)
4245                 journal->j_trans_id = 10;
4246         journal->j_current_jl->j_trans_id = journal->j_trans_id;
4247         journal->j_must_wait = 0;
4248         journal->j_len_alloc = 0;
4249         journal->j_next_full_flush = 0;
4250         journal->j_next_async_flush = 0;
4251         init_journal_hash(sb);
4252
4253         /*
4254          * make sure reiserfs_add_jh sees the new current_jl before we
4255          * write out the tails
4256          */
4257         smp_mb();
4258
4259         /*
4260          * tail conversion targets have to hit the disk before we end the
4261          * transaction.  Otherwise a later transaction might repack the tail
4262          * before this transaction commits, leaving the data block unflushed
4263          * and clean, if we crash before the later transaction commits, the
4264          * data block is lost.
4265          */
4266         if (!list_empty(&jl->j_tail_bh_list)) {
4267                 depth = reiserfs_write_unlock_nested(sb);
4268                 write_ordered_buffers(&journal->j_dirty_buffers_lock,
4269                                       journal, jl, &jl->j_tail_bh_list);
4270                 reiserfs_write_lock_nested(sb, depth);
4271         }
4272         BUG_ON(!list_empty(&jl->j_tail_bh_list));
4273         mutex_unlock(&jl->j_commit_mutex);
4274
4275         /*
4276          * honor the flush wishes from the caller, simple commits can
4277          * be done outside the journal lock, they are done below
4278          *
4279          * if we don't flush the commit list right now, we put it into
4280          * the work queue so the people waiting on the async progress work
4281          * queue don't wait for this proc to flush journal lists and such.
4282          */
4283         if (flush) {
4284                 flush_commit_list(sb, jl, 1);
4285                 flush_journal_list(sb, jl, 1);
4286         } else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4287                 /*
4288                  * Avoid queueing work when sb is being shut down. Transaction
4289                  * will be flushed on journal shutdown.
4290                  */
4291                 if (sb->s_flags & SB_ACTIVE)
4292                         queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4293                                            &journal->j_work, HZ / 10);
4294         }
4295
4296         /*
4297          * if the next transaction has any chance of wrapping, flush
4298          * transactions that might get overwritten.  If any journal lists
4299          * are very old flush them as well.
4300          */
4301 first_jl:
4302         list_for_each_safe(entry, safe, &journal->j_journal_list) {
4303                 temp_jl = JOURNAL_LIST_ENTRY(entry);
4304                 if (journal->j_start <= temp_jl->j_start) {
4305                         if ((journal->j_start + journal->j_trans_max + 1) >=
4306                             temp_jl->j_start) {
4307                                 flush_used_journal_lists(sb, temp_jl);
4308                                 goto first_jl;
4309                         } else if ((journal->j_start +
4310                                     journal->j_trans_max + 1) <
4311                                    SB_ONDISK_JOURNAL_SIZE(sb)) {
4312                                 /*
4313                                  * if we don't cross into the next
4314                                  * transaction and we don't wrap, there is
4315                                  * no way we can overlap any later transactions
4316                                  * break now
4317                                  */
4318                                 break;
4319                         }
4320                 } else if ((journal->j_start +
4321                             journal->j_trans_max + 1) >
4322                            SB_ONDISK_JOURNAL_SIZE(sb)) {
4323                         if (((journal->j_start + journal->j_trans_max + 1) %
4324                              SB_ONDISK_JOURNAL_SIZE(sb)) >=
4325                             temp_jl->j_start) {
4326                                 flush_used_journal_lists(sb, temp_jl);
4327                                 goto first_jl;
4328                         } else {
4329                                 /*
4330                                 * we don't overlap anything from out start
4331                                 * to the end of the log, and our wrapped
4332                                 * portion doesn't overlap anything at
4333                                 * the start of the log.  We can break
4334                                 */
4335                                 break;
4336                         }
4337                 }
4338         }
4339
4340         journal->j_current_jl->j_list_bitmap =
4341             get_list_bitmap(sb, journal->j_current_jl);
4342
4343         if (!(journal->j_current_jl->j_list_bitmap)) {
4344                 reiserfs_panic(sb, "journal-1996",
4345                                "could not get a list bitmap");
4346         }
4347
4348         atomic_set(&journal->j_jlock, 0);
4349         unlock_journal(sb);
4350         /* wake up any body waiting to join. */
4351         clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4352         wake_up(&journal->j_join_wait);
4353
4354         if (!flush && wait_on_commit &&
4355             journal_list_still_alive(sb, commit_trans_id)) {
4356                 flush_commit_list(sb, jl, 1);
4357         }
4358 out:
4359         reiserfs_check_lock_depth(sb, "journal end2");
4360
4361         memset(th, 0, sizeof(*th));
4362         /*
4363          * Re-set th->t_super, so we can properly keep track of how many
4364          * persistent transactions there are. We need to do this so if this
4365          * call is part of a failed restart_transaction, we can free it later
4366          */
4367         th->t_super = sb;
4368
4369         return journal->j_errno;
4370 }
4371
4372 /* Send the file system read only and refuse new transactions */
4373 void reiserfs_abort_journal(struct super_block *sb, int errno)
4374 {
4375         struct reiserfs_journal *journal = SB_JOURNAL(sb);
4376         if (test_bit(J_ABORTED, &journal->j_state))
4377                 return;
4378
4379         if (!journal->j_errno)
4380                 journal->j_errno = errno;
4381
4382         sb->s_flags |= SB_RDONLY;
4383         set_bit(J_ABORTED, &journal->j_state);
4384
4385 #ifdef CONFIG_REISERFS_CHECK
4386         dump_stack();
4387 #endif
4388 }