Linux Audio

Check our new training course

Loading...
v6.2
   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
  85static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  86static int flush_journal_list(struct super_block *s,
  87			      struct reiserfs_journal_list *jl, int flushall);
  88static int flush_commit_list(struct super_block *s,
  89			     struct reiserfs_journal_list *jl, int flushall);
  90static int can_dirty(struct reiserfs_journal_cnode *cn);
  91static int journal_join(struct reiserfs_transaction_handle *th,
  92			struct super_block *sb);
  93static void release_journal_dev(struct super_block *super,
  94			       struct reiserfs_journal *journal);
  95static void dirty_one_transaction(struct super_block *s,
  96				 struct reiserfs_journal_list *jl);
  97static void flush_async_commits(struct work_struct *work);
  98static void queue_log_writer(struct super_block *s);
  99
 100/* values for join in do_journal_begin_r */
 101enum {
 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
 109static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 110			      struct super_block *sb,
 111			      unsigned long nblocks, int join);
 112
 113static 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 */
 126static 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
 135static 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
 155static 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++;
 162repeat:
 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}
 178static 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
 192static 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
 209static 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
 223static 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 */
 241static 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
 256static 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 */
 278int 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 */
 309static 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 */
 344static 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 */
 367static 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 */
 394static 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
 412static 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 */
 422static 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 */
 459int 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 */
 515static 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 */
 530static 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 */
 538static inline void unlock_journal(struct super_block *sb)
 539{
 540	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 541}
 542
 543static inline void get_journal_list(struct reiserfs_journal_list *jl)
 544{
 545	jl->j_refcount++;
 546}
 547
 548static 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 */
 564static 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
 576static 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 */
 602static void release_buffer_page(struct buffer_head *bh)
 603{
 604	struct folio *folio = page_folio(bh->b_page);
 605	if (!folio->mapping && folio_trylock(folio)) {
 606		folio_get(folio);
 607		put_bh(bh);
 608		if (!folio->mapping)
 609			try_to_free_buffers(folio);
 610		folio_unlock(folio);
 611		folio_put(folio);
 612	} else {
 613		put_bh(bh);
 614	}
 615}
 616
 617static 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
 633static 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
 643static 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, bh);
 654}
 655
 656static 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, bh);
 664}
 665
 666#define CHUNK_SIZE 32
 667struct buffer_chunk {
 668	struct buffer_head *bh[CHUNK_SIZE];
 669	int nr;
 670};
 671
 672static 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
 681static 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
 690static 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
 709static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 710static 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 */
 727void 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
 744static 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 {
 758no_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
 780int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 781{
 782	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 783}
 784int 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)
 790static 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		}
 837loop_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 invalidate_folio here.
 862		 * reiserfs_invalidate_folio 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			write_dirty_buffer(bh, 0);
 872			spin_lock(lock);
 873		}
 874		put_bh(bh);
 875		cond_resched_lock(lock);
 876	}
 877	spin_unlock(lock);
 878	return ret;
 879}
 880
 881static 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
 891find_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
 946static 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		wait_var_event_timeout(&j->j_async_throttle,
 955				       atomic_read(&j->j_async_throttle) == 0,
 956				       HZ / 10);
 957		reiserfs_write_lock_nested(s, depth);
 958	}
 959
 960	return 0;
 961}
 962
 963/*
 964 * if this journal list still has commit blocks unflushed, send them to disk.
 965 *
 966 * log areas must be flushed in order (transaction 2 can't commit before
 967 * transaction 1) Before the commit block can by written, every other log
 968 * block must be safely on disk
 969 */
 970static int flush_commit_list(struct super_block *s,
 971			     struct reiserfs_journal_list *jl, int flushall)
 972{
 973	int i;
 974	b_blocknr_t bn;
 975	struct buffer_head *tbh = NULL;
 976	unsigned int trans_id = jl->j_trans_id;
 977	struct reiserfs_journal *journal = SB_JOURNAL(s);
 978	int retval = 0;
 979	int write_len;
 980	int depth;
 981
 982	reiserfs_check_lock_depth(s, "flush_commit_list");
 983
 984	if (atomic_read(&jl->j_older_commits_done)) {
 985		return 0;
 986	}
 987
 988	/*
 989	 * before we can put our commit blocks on disk, we have to make
 990	 * sure everyone older than us is on disk too
 991	 */
 992	BUG_ON(jl->j_len <= 0);
 993	BUG_ON(trans_id == journal->j_trans_id);
 994
 995	get_journal_list(jl);
 996	if (flushall) {
 997		if (flush_older_commits(s, jl) == 1) {
 998			/*
 999			 * list disappeared during flush_older_commits.
1000			 * return
1001			 */
1002			goto put_jl;
1003		}
1004	}
1005
1006	/* make sure nobody is trying to flush this one at the same time */
1007	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1008
1009	if (!journal_list_still_alive(s, trans_id)) {
1010		mutex_unlock(&jl->j_commit_mutex);
1011		goto put_jl;
1012	}
1013	BUG_ON(jl->j_trans_id == 0);
1014
1015	/* this commit is done, exit */
1016	if (atomic_read(&jl->j_commit_left) <= 0) {
1017		if (flushall) {
1018			atomic_set(&jl->j_older_commits_done, 1);
1019		}
1020		mutex_unlock(&jl->j_commit_mutex);
1021		goto put_jl;
1022	}
1023
1024	if (!list_empty(&jl->j_bh_list)) {
1025		int ret;
1026
1027		/*
1028		 * We might sleep in numerous places inside
1029		 * write_ordered_buffers. Relax the write lock.
1030		 */
1031		depth = reiserfs_write_unlock_nested(s);
1032		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1033					    journal, jl, &jl->j_bh_list);
1034		if (ret < 0 && retval == 0)
1035			retval = ret;
1036		reiserfs_write_lock_nested(s, depth);
1037	}
1038	BUG_ON(!list_empty(&jl->j_bh_list));
1039	/*
1040	 * for the description block and all the log blocks, submit any buffers
1041	 * that haven't already reached the disk.  Try to write at least 256
1042	 * log blocks. later on, we will only wait on blocks that correspond
1043	 * to this transaction, but while we're unplugging we might as well
1044	 * get a chunk of data on there.
1045	 */
1046	atomic_inc(&journal->j_async_throttle);
1047	write_len = jl->j_len + 1;
1048	if (write_len < 256)
1049		write_len = 256;
1050	for (i = 0 ; i < write_len ; i++) {
1051		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1052		    SB_ONDISK_JOURNAL_SIZE(s);
1053		tbh = journal_find_get_block(s, bn);
1054		if (tbh) {
1055			if (buffer_dirty(tbh)) {
1056		            depth = reiserfs_write_unlock_nested(s);
1057			    write_dirty_buffer(tbh, 0);
1058			    reiserfs_write_lock_nested(s, depth);
1059			}
1060			put_bh(tbh) ;
1061		}
1062	}
1063	if (atomic_dec_and_test(&journal->j_async_throttle))
1064		wake_up_var(&journal->j_async_throttle);
1065
1066	for (i = 0; i < (jl->j_len + 1); i++) {
1067		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1068		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1069		tbh = journal_find_get_block(s, bn);
1070
1071		depth = reiserfs_write_unlock_nested(s);
1072		__wait_on_buffer(tbh);
1073		reiserfs_write_lock_nested(s, depth);
1074		/*
1075		 * since we're using ll_rw_blk above, it might have skipped
1076		 * over a locked buffer.  Double check here
1077		 */
1078		/* redundant, sync_dirty_buffer() checks */
1079		if (buffer_dirty(tbh)) {
1080			depth = reiserfs_write_unlock_nested(s);
1081			sync_dirty_buffer(tbh);
1082			reiserfs_write_lock_nested(s, depth);
1083		}
1084		if (unlikely(!buffer_uptodate(tbh))) {
1085#ifdef CONFIG_REISERFS_CHECK
1086			reiserfs_warning(s, "journal-601",
1087					 "buffer write failed");
1088#endif
1089			retval = -EIO;
1090		}
1091		/* once for journal_find_get_block */
1092		put_bh(tbh);
1093		/* once due to original getblk in do_journal_end */
1094		put_bh(tbh);
1095		atomic_dec(&jl->j_commit_left);
1096	}
1097
1098	BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1099
1100	/*
1101	 * If there was a write error in the journal - we can't commit
1102	 * this transaction - it will be invalid and, if successful,
1103	 * will just end up propagating the write error out to
1104	 * the file system.
1105	 */
1106	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1107		if (buffer_dirty(jl->j_commit_bh))
1108			BUG();
1109		mark_buffer_dirty(jl->j_commit_bh) ;
1110		depth = reiserfs_write_unlock_nested(s);
1111		if (reiserfs_barrier_flush(s))
1112			__sync_dirty_buffer(jl->j_commit_bh,
1113					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1114		else
1115			sync_dirty_buffer(jl->j_commit_bh);
1116		reiserfs_write_lock_nested(s, depth);
1117	}
1118
1119	/*
1120	 * If there was a write error in the journal - we can't commit this
1121	 * transaction - it will be invalid and, if successful, will just end
1122	 * up propagating the write error out to the filesystem.
1123	 */
1124	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1125#ifdef CONFIG_REISERFS_CHECK
1126		reiserfs_warning(s, "journal-615", "buffer write failed");
1127#endif
1128		retval = -EIO;
1129	}
1130	bforget(jl->j_commit_bh);
1131	if (journal->j_last_commit_id != 0 &&
1132	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1133		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1134				 journal->j_last_commit_id, jl->j_trans_id);
1135	}
1136	journal->j_last_commit_id = jl->j_trans_id;
1137
1138	/*
1139	 * now, every commit block is on the disk.  It is safe to allow
1140	 * blocks freed during this transaction to be reallocated
1141	 */
1142	cleanup_freed_for_journal_list(s, jl);
1143
1144	retval = retval ? retval : journal->j_errno;
1145
1146	/* mark the metadata dirty */
1147	if (!retval)
1148		dirty_one_transaction(s, jl);
1149	atomic_dec(&jl->j_commit_left);
1150
1151	if (flushall) {
1152		atomic_set(&jl->j_older_commits_done, 1);
1153	}
1154	mutex_unlock(&jl->j_commit_mutex);
1155put_jl:
1156	put_journal_list(s, jl);
1157
1158	if (retval)
1159		reiserfs_abort(s, retval, "Journal write error in %s",
1160			       __func__);
1161	return retval;
1162}
1163
1164/*
1165 * flush_journal_list frequently needs to find a newer transaction for a
1166 * given block.  This does that, or returns NULL if it can't find anything
1167 */
1168static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1169							  reiserfs_journal_cnode
1170							  *cn)
1171{
1172	struct super_block *sb = cn->sb;
1173	b_blocknr_t blocknr = cn->blocknr;
1174
1175	cn = cn->hprev;
1176	while (cn) {
1177		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1178			return cn->jlist;
1179		}
1180		cn = cn->hprev;
1181	}
1182	return NULL;
1183}
1184
1185static void remove_journal_hash(struct super_block *,
1186				struct reiserfs_journal_cnode **,
1187				struct reiserfs_journal_list *, unsigned long,
1188				int);
1189
1190/*
1191 * once all the real blocks have been flushed, it is safe to remove them
1192 * from the journal list for this transaction.  Aside from freeing the
1193 * cnode, this also allows the block to be reallocated for data blocks
1194 * if it had been deleted.
1195 */
1196static void remove_all_from_journal_list(struct super_block *sb,
1197					 struct reiserfs_journal_list *jl,
1198					 int debug)
1199{
1200	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1201	struct reiserfs_journal_cnode *cn, *last;
1202	cn = jl->j_realblock;
1203
1204	/*
1205	 * which is better, to lock once around the whole loop, or
1206	 * to lock for each call to remove_journal_hash?
1207	 */
1208	while (cn) {
1209		if (cn->blocknr != 0) {
1210			if (debug) {
1211				reiserfs_warning(sb, "reiserfs-2201",
1212						 "block %u, bh is %d, state %ld",
1213						 cn->blocknr, cn->bh ? 1 : 0,
1214						 cn->state);
1215			}
1216			cn->state = 0;
1217			remove_journal_hash(sb, journal->j_list_hash_table,
1218					    jl, cn->blocknr, 1);
1219		}
1220		last = cn;
1221		cn = cn->next;
1222		free_cnode(sb, last);
1223	}
1224	jl->j_realblock = NULL;
1225}
1226
1227/*
1228 * if this timestamp is greater than the timestamp we wrote last to the
1229 * header block, write it to the header block.  once this is done, I can
1230 * safely say the log area for this transaction won't ever be replayed,
1231 * and I can start releasing blocks in this transaction for reuse as data
1232 * blocks.  called by flush_journal_list, before it calls
1233 * remove_all_from_journal_list
1234 */
1235static int _update_journal_header_block(struct super_block *sb,
1236					unsigned long offset,
1237					unsigned int trans_id)
1238{
1239	struct reiserfs_journal_header *jh;
1240	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1241	int depth;
1242
1243	if (reiserfs_is_journal_aborted(journal))
1244		return -EIO;
1245
1246	if (trans_id >= journal->j_last_flush_trans_id) {
1247		if (buffer_locked((journal->j_header_bh))) {
1248			depth = reiserfs_write_unlock_nested(sb);
1249			__wait_on_buffer(journal->j_header_bh);
1250			reiserfs_write_lock_nested(sb, depth);
1251			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1252#ifdef CONFIG_REISERFS_CHECK
1253				reiserfs_warning(sb, "journal-699",
1254						 "buffer write failed");
1255#endif
1256				return -EIO;
1257			}
1258		}
1259		journal->j_last_flush_trans_id = trans_id;
1260		journal->j_first_unflushed_offset = offset;
1261		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1262							b_data);
1263		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1264		jh->j_first_unflushed_offset = cpu_to_le32(offset);
1265		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1266
1267		set_buffer_dirty(journal->j_header_bh);
1268		depth = reiserfs_write_unlock_nested(sb);
1269
1270		if (reiserfs_barrier_flush(sb))
1271			__sync_dirty_buffer(journal->j_header_bh,
1272					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1273		else
1274			sync_dirty_buffer(journal->j_header_bh);
1275
1276		reiserfs_write_lock_nested(sb, depth);
1277		if (!buffer_uptodate(journal->j_header_bh)) {
1278			reiserfs_warning(sb, "journal-837",
1279					 "IO error during journal replay");
1280			return -EIO;
1281		}
1282	}
1283	return 0;
1284}
1285
1286static int update_journal_header_block(struct super_block *sb,
1287				       unsigned long offset,
1288				       unsigned int trans_id)
1289{
1290	return _update_journal_header_block(sb, offset, trans_id);
1291}
1292
1293/*
1294** flush any and all journal lists older than you are
1295** can only be called from flush_journal_list
1296*/
1297static int flush_older_journal_lists(struct super_block *sb,
1298				     struct reiserfs_journal_list *jl)
1299{
1300	struct list_head *entry;
1301	struct reiserfs_journal_list *other_jl;
1302	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1303	unsigned int trans_id = jl->j_trans_id;
1304
1305	/*
1306	 * we know we are the only ones flushing things, no extra race
1307	 * protection is required.
1308	 */
1309restart:
1310	entry = journal->j_journal_list.next;
1311	/* Did we wrap? */
1312	if (entry == &journal->j_journal_list)
1313		return 0;
1314	other_jl = JOURNAL_LIST_ENTRY(entry);
1315	if (other_jl->j_trans_id < trans_id) {
1316		BUG_ON(other_jl->j_refcount <= 0);
1317		/* do not flush all */
1318		flush_journal_list(sb, other_jl, 0);
1319
1320		/* other_jl is now deleted from the list */
1321		goto restart;
1322	}
1323	return 0;
1324}
1325
1326static void del_from_work_list(struct super_block *s,
1327			       struct reiserfs_journal_list *jl)
1328{
1329	struct reiserfs_journal *journal = SB_JOURNAL(s);
1330	if (!list_empty(&jl->j_working_list)) {
1331		list_del_init(&jl->j_working_list);
1332		journal->j_num_work_lists--;
1333	}
1334}
1335
1336/*
1337 * flush a journal list, both commit and real blocks
1338 *
1339 * always set flushall to 1, unless you are calling from inside
1340 * flush_journal_list
1341 *
1342 * IMPORTANT.  This can only be called while there are no journal writers,
1343 * and the journal is locked.  That means it can only be called from
1344 * do_journal_end, or by journal_release
1345 */
1346static int flush_journal_list(struct super_block *s,
1347			      struct reiserfs_journal_list *jl, int flushall)
1348{
1349	struct reiserfs_journal_list *pjl;
1350	struct reiserfs_journal_cnode *cn;
1351	int count;
1352	int was_jwait = 0;
1353	int was_dirty = 0;
1354	struct buffer_head *saved_bh;
1355	unsigned long j_len_saved = jl->j_len;
1356	struct reiserfs_journal *journal = SB_JOURNAL(s);
1357	int err = 0;
1358	int depth;
1359
1360	BUG_ON(j_len_saved <= 0);
1361
1362	if (atomic_read(&journal->j_wcount) != 0) {
1363		reiserfs_warning(s, "clm-2048", "called with wcount %d",
1364				 atomic_read(&journal->j_wcount));
1365	}
1366
1367	/* if flushall == 0, the lock is already held */
1368	if (flushall) {
1369		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1370	} else if (mutex_trylock(&journal->j_flush_mutex)) {
1371		BUG();
1372	}
1373
1374	count = 0;
1375	if (j_len_saved > journal->j_trans_max) {
1376		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1377			       j_len_saved, jl->j_trans_id);
1378		return 0;
1379	}
1380
1381	/* if all the work is already done, get out of here */
1382	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1383	    atomic_read(&jl->j_commit_left) <= 0) {
1384		goto flush_older_and_return;
1385	}
1386
1387	/*
1388	 * start by putting the commit list on disk.  This will also flush
1389	 * the commit lists of any olders transactions
1390	 */
1391	flush_commit_list(s, jl, 1);
1392
1393	if (!(jl->j_state & LIST_DIRTY)
1394	    && !reiserfs_is_journal_aborted(journal))
1395		BUG();
1396
1397	/* are we done now? */
1398	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1399	    atomic_read(&jl->j_commit_left) <= 0) {
1400		goto flush_older_and_return;
1401	}
1402
1403	/*
1404	 * loop through each cnode, see if we need to write it,
1405	 * or wait on a more recent transaction, or just ignore it
1406	 */
1407	if (atomic_read(&journal->j_wcount) != 0) {
1408		reiserfs_panic(s, "journal-844", "journal list is flushing, "
1409			       "wcount is not 0");
1410	}
1411	cn = jl->j_realblock;
1412	while (cn) {
1413		was_jwait = 0;
1414		was_dirty = 0;
1415		saved_bh = NULL;
1416		/* blocknr of 0 is no longer in the hash, ignore it */
1417		if (cn->blocknr == 0) {
1418			goto free_cnode;
1419		}
1420
1421		/*
1422		 * This transaction failed commit.
1423		 * Don't write out to the disk
1424		 */
1425		if (!(jl->j_state & LIST_DIRTY))
1426			goto free_cnode;
1427
1428		pjl = find_newer_jl_for_cn(cn);
1429		/*
1430		 * the order is important here.  We check pjl to make sure we
1431		 * don't clear BH_JDirty_wait if we aren't the one writing this
1432		 * block to disk
1433		 */
1434		if (!pjl && cn->bh) {
1435			saved_bh = cn->bh;
1436
1437			/*
1438			 * we do this to make sure nobody releases the
1439			 * buffer while we are working with it
1440			 */
1441			get_bh(saved_bh);
1442
1443			if (buffer_journal_dirty(saved_bh)) {
1444				BUG_ON(!can_dirty(cn));
1445				was_jwait = 1;
1446				was_dirty = 1;
1447			} else if (can_dirty(cn)) {
1448				/*
1449				 * everything with !pjl && jwait
1450				 * should be writable
1451				 */
1452				BUG();
1453			}
1454		}
1455
1456		/*
1457		 * if someone has this block in a newer transaction, just make
1458		 * sure they are committed, and don't try writing it to disk
1459		 */
1460		if (pjl) {
1461			if (atomic_read(&pjl->j_commit_left))
1462				flush_commit_list(s, pjl, 1);
1463			goto free_cnode;
1464		}
1465
1466		/*
1467		 * bh == NULL when the block got to disk on its own, OR,
1468		 * the block got freed in a future transaction
1469		 */
1470		if (saved_bh == NULL) {
1471			goto free_cnode;
1472		}
1473
1474		/*
1475		 * this should never happen.  kupdate_one_transaction has
1476		 * this list locked while it works, so we should never see a
1477		 * buffer here that is not marked JDirty_wait
1478		 */
1479		if ((!was_jwait) && !buffer_locked(saved_bh)) {
1480			reiserfs_warning(s, "journal-813",
1481					 "BAD! buffer %llu %cdirty %cjwait, "
1482					 "not in a newer transaction",
1483					 (unsigned long long)saved_bh->
1484					 b_blocknr, was_dirty ? ' ' : '!',
1485					 was_jwait ? ' ' : '!');
1486		}
1487		if (was_dirty) {
1488			/*
1489			 * we inc again because saved_bh gets decremented
1490			 * at free_cnode
1491			 */
1492			get_bh(saved_bh);
1493			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1494			lock_buffer(saved_bh);
1495			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1496			if (buffer_dirty(saved_bh))
1497				submit_logged_buffer(saved_bh);
1498			else
1499				unlock_buffer(saved_bh);
1500			count++;
1501		} else {
1502			reiserfs_warning(s, "clm-2082",
1503					 "Unable to flush buffer %llu in %s",
1504					 (unsigned long long)saved_bh->
1505					 b_blocknr, __func__);
1506		}
1507free_cnode:
1508		cn = cn->next;
1509		if (saved_bh) {
1510			/*
1511			 * we incremented this to keep others from
1512			 * taking the buffer head away
1513			 */
1514			put_bh(saved_bh);
1515			if (atomic_read(&saved_bh->b_count) < 0) {
1516				reiserfs_warning(s, "journal-945",
1517						 "saved_bh->b_count < 0");
1518			}
1519		}
1520	}
1521	if (count > 0) {
1522		cn = jl->j_realblock;
1523		while (cn) {
1524			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1525				if (!cn->bh) {
1526					reiserfs_panic(s, "journal-1011",
1527						       "cn->bh is NULL");
1528				}
1529
1530				depth = reiserfs_write_unlock_nested(s);
1531				__wait_on_buffer(cn->bh);
1532				reiserfs_write_lock_nested(s, depth);
1533
1534				if (!cn->bh) {
1535					reiserfs_panic(s, "journal-1012",
1536						       "cn->bh is NULL");
1537				}
1538				if (unlikely(!buffer_uptodate(cn->bh))) {
1539#ifdef CONFIG_REISERFS_CHECK
1540					reiserfs_warning(s, "journal-949",
1541							 "buffer write failed");
1542#endif
1543					err = -EIO;
1544				}
1545				/*
1546				 * note, we must clear the JDirty_wait bit
1547				 * after the up to date check, otherwise we
1548				 * race against our flushpage routine
1549				 */
1550				BUG_ON(!test_clear_buffer_journal_dirty
1551				       (cn->bh));
1552
1553				/* drop one ref for us */
1554				put_bh(cn->bh);
1555				/* drop one ref for journal_mark_dirty */
1556				release_buffer_page(cn->bh);
1557			}
1558			cn = cn->next;
1559		}
1560	}
1561
1562	if (err)
1563		reiserfs_abort(s, -EIO,
1564			       "Write error while pushing transaction to disk in %s",
1565			       __func__);
1566flush_older_and_return:
1567
1568	/*
1569	 * before we can update the journal header block, we _must_ flush all
1570	 * real blocks from all older transactions to disk.  This is because
1571	 * once the header block is updated, this transaction will not be
1572	 * replayed after a crash
1573	 */
1574	if (flushall) {
1575		flush_older_journal_lists(s, jl);
1576	}
1577
1578	err = journal->j_errno;
1579	/*
1580	 * before we can remove everything from the hash tables for this
1581	 * transaction, we must make sure it can never be replayed
1582	 *
1583	 * since we are only called from do_journal_end, we know for sure there
1584	 * are no allocations going on while we are flushing journal lists.  So,
1585	 * we only need to update the journal header block for the last list
1586	 * being flushed
1587	 */
1588	if (!err && flushall) {
1589		err =
1590		    update_journal_header_block(s,
1591						(jl->j_start + jl->j_len +
1592						 2) % SB_ONDISK_JOURNAL_SIZE(s),
1593						jl->j_trans_id);
1594		if (err)
1595			reiserfs_abort(s, -EIO,
1596				       "Write error while updating journal header in %s",
1597				       __func__);
1598	}
1599	remove_all_from_journal_list(s, jl, 0);
1600	list_del_init(&jl->j_list);
1601	journal->j_num_lists--;
1602	del_from_work_list(s, jl);
1603
1604	if (journal->j_last_flush_id != 0 &&
1605	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1606		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1607				 journal->j_last_flush_id, jl->j_trans_id);
1608	}
1609	journal->j_last_flush_id = jl->j_trans_id;
1610
1611	/*
1612	 * not strictly required since we are freeing the list, but it should
1613	 * help find code using dead lists later on
1614	 */
1615	jl->j_len = 0;
1616	atomic_set(&jl->j_nonzerolen, 0);
1617	jl->j_start = 0;
1618	jl->j_realblock = NULL;
1619	jl->j_commit_bh = NULL;
1620	jl->j_trans_id = 0;
1621	jl->j_state = 0;
1622	put_journal_list(s, jl);
1623	if (flushall)
1624		mutex_unlock(&journal->j_flush_mutex);
1625	return err;
1626}
1627
1628static int write_one_transaction(struct super_block *s,
1629				 struct reiserfs_journal_list *jl,
1630				 struct buffer_chunk *chunk)
1631{
1632	struct reiserfs_journal_cnode *cn;
1633	int ret = 0;
1634
1635	jl->j_state |= LIST_TOUCHED;
1636	del_from_work_list(s, jl);
1637	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1638		return 0;
1639	}
1640
1641	cn = jl->j_realblock;
1642	while (cn) {
1643		/*
1644		 * if the blocknr == 0, this has been cleared from the hash,
1645		 * skip it
1646		 */
1647		if (cn->blocknr == 0) {
1648			goto next;
1649		}
1650		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1651			struct buffer_head *tmp_bh;
1652			/*
1653			 * we can race against journal_mark_freed when we try
1654			 * to lock_buffer(cn->bh), so we have to inc the buffer
1655			 * count, and recheck things after locking
1656			 */
1657			tmp_bh = cn->bh;
1658			get_bh(tmp_bh);
1659			lock_buffer(tmp_bh);
1660			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1661				if (!buffer_journal_dirty(tmp_bh) ||
1662				    buffer_journal_prepared(tmp_bh))
1663					BUG();
1664				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1665				ret++;
1666			} else {
1667				/* note, cn->bh might be null now */
1668				unlock_buffer(tmp_bh);
1669			}
1670			put_bh(tmp_bh);
1671		}
1672next:
1673		cn = cn->next;
1674		cond_resched();
1675	}
1676	return ret;
1677}
1678
1679/* used by flush_commit_list */
1680static void dirty_one_transaction(struct super_block *s,
1681				 struct reiserfs_journal_list *jl)
1682{
1683	struct reiserfs_journal_cnode *cn;
1684	struct reiserfs_journal_list *pjl;
1685
1686	jl->j_state |= LIST_DIRTY;
1687	cn = jl->j_realblock;
1688	while (cn) {
1689		/*
1690		 * look for a more recent transaction that logged this
1691		 * buffer.  Only the most recent transaction with a buffer in
1692		 * it is allowed to send that buffer to disk
1693		 */
1694		pjl = find_newer_jl_for_cn(cn);
1695		if (!pjl && cn->blocknr && cn->bh
1696		    && buffer_journal_dirty(cn->bh)) {
1697			BUG_ON(!can_dirty(cn));
1698			/*
1699			 * if the buffer is prepared, it will either be logged
1700			 * or restored.  If restored, we need to make sure
1701			 * it actually gets marked dirty
1702			 */
1703			clear_buffer_journal_new(cn->bh);
1704			if (buffer_journal_prepared(cn->bh)) {
1705				set_buffer_journal_restore_dirty(cn->bh);
1706			} else {
1707				set_buffer_journal_test(cn->bh);
1708				mark_buffer_dirty(cn->bh);
1709			}
1710		}
1711		cn = cn->next;
1712	}
1713}
1714
1715static int kupdate_transactions(struct super_block *s,
1716				struct reiserfs_journal_list *jl,
1717				struct reiserfs_journal_list **next_jl,
1718				unsigned int *next_trans_id,
1719				int num_blocks, int num_trans)
1720{
1721	int ret = 0;
1722	int written = 0;
1723	int transactions_flushed = 0;
1724	unsigned int orig_trans_id = jl->j_trans_id;
1725	struct buffer_chunk chunk;
1726	struct list_head *entry;
1727	struct reiserfs_journal *journal = SB_JOURNAL(s);
1728	chunk.nr = 0;
1729
1730	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1731	if (!journal_list_still_alive(s, orig_trans_id)) {
1732		goto done;
1733	}
1734
1735	/*
1736	 * we've got j_flush_mutex held, nobody is going to delete any
1737	 * of these lists out from underneath us
1738	 */
1739	while ((num_trans && transactions_flushed < num_trans) ||
1740	       (!num_trans && written < num_blocks)) {
1741
1742		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1743		    atomic_read(&jl->j_commit_left)
1744		    || !(jl->j_state & LIST_DIRTY)) {
1745			del_from_work_list(s, jl);
1746			break;
1747		}
1748		ret = write_one_transaction(s, jl, &chunk);
1749
1750		if (ret < 0)
1751			goto done;
1752		transactions_flushed++;
1753		written += ret;
1754		entry = jl->j_list.next;
1755
1756		/* did we wrap? */
1757		if (entry == &journal->j_journal_list) {
1758			break;
1759		}
1760		jl = JOURNAL_LIST_ENTRY(entry);
1761
1762		/* don't bother with older transactions */
1763		if (jl->j_trans_id <= orig_trans_id)
1764			break;
1765	}
1766	if (chunk.nr) {
1767		write_chunk(&chunk);
1768	}
1769
1770done:
1771	mutex_unlock(&journal->j_flush_mutex);
1772	return ret;
1773}
1774
1775/*
1776 * for o_sync and fsync heavy applications, they tend to use
1777 * all the journa list slots with tiny transactions.  These
1778 * trigger lots and lots of calls to update the header block, which
1779 * adds seeks and slows things down.
1780 *
1781 * This function tries to clear out a large chunk of the journal lists
1782 * at once, which makes everything faster since only the newest journal
1783 * list updates the header block
1784 */
1785static int flush_used_journal_lists(struct super_block *s,
1786				    struct reiserfs_journal_list *jl)
1787{
1788	unsigned long len = 0;
1789	unsigned long cur_len;
1790	int i;
1791	int limit = 256;
1792	struct reiserfs_journal_list *tjl;
1793	struct reiserfs_journal_list *flush_jl;
1794	unsigned int trans_id;
1795	struct reiserfs_journal *journal = SB_JOURNAL(s);
1796
1797	flush_jl = tjl = jl;
1798
1799	/* in data logging mode, try harder to flush a lot of blocks */
1800	if (reiserfs_data_log(s))
1801		limit = 1024;
1802	/* flush for 256 transactions or limit blocks, whichever comes first */
1803	for (i = 0; i < 256 && len < limit; i++) {
1804		if (atomic_read(&tjl->j_commit_left) ||
1805		    tjl->j_trans_id < jl->j_trans_id) {
1806			break;
1807		}
1808		cur_len = atomic_read(&tjl->j_nonzerolen);
1809		if (cur_len > 0) {
1810			tjl->j_state &= ~LIST_TOUCHED;
1811		}
1812		len += cur_len;
1813		flush_jl = tjl;
1814		if (tjl->j_list.next == &journal->j_journal_list)
1815			break;
1816		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1817	}
1818	get_journal_list(jl);
1819	get_journal_list(flush_jl);
1820
1821	/*
1822	 * try to find a group of blocks we can flush across all the
1823	 * transactions, but only bother if we've actually spanned
1824	 * across multiple lists
1825	 */
1826	if (flush_jl != jl)
1827		kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1828
1829	flush_journal_list(s, flush_jl, 1);
1830	put_journal_list(s, flush_jl);
1831	put_journal_list(s, jl);
1832	return 0;
1833}
1834
1835/*
1836 * removes any nodes in table with name block and dev as bh.
1837 * only touchs the hnext and hprev pointers.
1838 */
1839static void remove_journal_hash(struct super_block *sb,
1840			 struct reiserfs_journal_cnode **table,
1841			 struct reiserfs_journal_list *jl,
1842			 unsigned long block, int remove_freed)
1843{
1844	struct reiserfs_journal_cnode *cur;
1845	struct reiserfs_journal_cnode **head;
1846
1847	head = &(journal_hash(table, sb, block));
1848	if (!head) {
1849		return;
1850	}
1851	cur = *head;
1852	while (cur) {
1853		if (cur->blocknr == block && cur->sb == sb
1854		    && (jl == NULL || jl == cur->jlist)
1855		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1856			if (cur->hnext) {
1857				cur->hnext->hprev = cur->hprev;
1858			}
1859			if (cur->hprev) {
1860				cur->hprev->hnext = cur->hnext;
1861			} else {
1862				*head = cur->hnext;
1863			}
1864			cur->blocknr = 0;
1865			cur->sb = NULL;
1866			cur->state = 0;
1867			/*
1868			 * anybody who clears the cur->bh will also
1869			 * dec the nonzerolen
1870			 */
1871			if (cur->bh && cur->jlist)
1872				atomic_dec(&cur->jlist->j_nonzerolen);
1873			cur->bh = NULL;
1874			cur->jlist = NULL;
1875		}
1876		cur = cur->hnext;
1877	}
1878}
1879
1880static void free_journal_ram(struct super_block *sb)
1881{
1882	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1883	kfree(journal->j_current_jl);
1884	journal->j_num_lists--;
1885
1886	vfree(journal->j_cnode_free_orig);
1887	free_list_bitmaps(sb, journal->j_list_bitmap);
1888	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */
1889	if (journal->j_header_bh) {
1890		brelse(journal->j_header_bh);
1891	}
1892	/*
1893	 * j_header_bh is on the journal dev, make sure
1894	 * not to release the journal dev until we brelse j_header_bh
1895	 */
1896	release_journal_dev(sb, journal);
1897	vfree(journal);
1898}
1899
1900/*
1901 * call on unmount.  Only set error to 1 if you haven't made your way out
1902 * of read_super() yet.  Any other caller must keep error at 0.
1903 */
1904static int do_journal_release(struct reiserfs_transaction_handle *th,
1905			      struct super_block *sb, int error)
1906{
1907	struct reiserfs_transaction_handle myth;
1908	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1909
1910	/*
1911	 * we only want to flush out transactions if we were
1912	 * called with error == 0
1913	 */
1914	if (!error && !sb_rdonly(sb)) {
1915		/* end the current trans */
1916		BUG_ON(!th->t_trans_id);
1917		do_journal_end(th, FLUSH_ALL);
1918
1919		/*
1920		 * make sure something gets logged to force
1921		 * our way into the flush code
1922		 */
1923		if (!journal_join(&myth, sb)) {
1924			reiserfs_prepare_for_journal(sb,
1925						     SB_BUFFER_WITH_SB(sb),
1926						     1);
1927			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1928			do_journal_end(&myth, FLUSH_ALL);
1929		}
1930	}
1931
1932	/* this also catches errors during the do_journal_end above */
1933	if (!error && reiserfs_is_journal_aborted(journal)) {
1934		memset(&myth, 0, sizeof(myth));
1935		if (!journal_join_abort(&myth, sb)) {
1936			reiserfs_prepare_for_journal(sb,
1937						     SB_BUFFER_WITH_SB(sb),
1938						     1);
1939			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1940			do_journal_end(&myth, FLUSH_ALL);
1941		}
1942	}
1943
1944
1945	/*
1946	 * We must release the write lock here because
1947	 * the workqueue job (flush_async_commit) needs this lock
1948	 */
1949	reiserfs_write_unlock(sb);
1950
1951	/*
1952	 * Cancel flushing of old commits. Note that neither of these works
1953	 * will be requeued because superblock is being shutdown and doesn't
1954	 * have SB_ACTIVE set.
1955	 */
1956	reiserfs_cancel_old_flush(sb);
1957	/* wait for all commits to finish */
1958	cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1959
1960	free_journal_ram(sb);
1961
1962	reiserfs_write_lock(sb);
1963
1964	return 0;
1965}
1966
1967/* * call on unmount.  flush all journal trans, release all alloc'd ram */
1968int journal_release(struct reiserfs_transaction_handle *th,
1969		    struct super_block *sb)
1970{
1971	return do_journal_release(th, sb, 0);
1972}
1973
1974/* only call from an error condition inside reiserfs_read_super!  */
1975int journal_release_error(struct reiserfs_transaction_handle *th,
1976			  struct super_block *sb)
1977{
1978	return do_journal_release(th, sb, 1);
1979}
1980
1981/*
1982 * compares description block with commit block.
1983 * returns 1 if they differ, 0 if they are the same
1984 */
1985static int journal_compare_desc_commit(struct super_block *sb,
1986				       struct reiserfs_journal_desc *desc,
1987				       struct reiserfs_journal_commit *commit)
1988{
1989	if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1990	    get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1991	    get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1992	    get_commit_trans_len(commit) <= 0) {
1993		return 1;
1994	}
1995	return 0;
1996}
1997
1998/*
1999 * returns 0 if it did not find a description block
2000 * returns -1 if it found a corrupt commit block
2001 * returns 1 if both desc and commit were valid
2002 * NOTE: only called during fs mount
2003 */
2004static int journal_transaction_is_valid(struct super_block *sb,
2005					struct buffer_head *d_bh,
2006					unsigned int *oldest_invalid_trans_id,
2007					unsigned long *newest_mount_id)
2008{
2009	struct reiserfs_journal_desc *desc;
2010	struct reiserfs_journal_commit *commit;
2011	struct buffer_head *c_bh;
2012	unsigned long offset;
2013
2014	if (!d_bh)
2015		return 0;
2016
2017	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2018	if (get_desc_trans_len(desc) > 0
2019	    && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2020		if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2021		    && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2022			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2023				       "journal-986: transaction "
2024				       "is valid returning because trans_id %d is greater than "
2025				       "oldest_invalid %lu",
2026				       get_desc_trans_id(desc),
2027				       *oldest_invalid_trans_id);
2028			return 0;
2029		}
2030		if (newest_mount_id
2031		    && *newest_mount_id > get_desc_mount_id(desc)) {
2032			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2033				       "journal-1087: transaction "
2034				       "is valid returning because mount_id %d is less than "
2035				       "newest_mount_id %lu",
2036				       get_desc_mount_id(desc),
2037				       *newest_mount_id);
2038			return -1;
2039		}
2040		if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2041			reiserfs_warning(sb, "journal-2018",
2042					 "Bad transaction length %d "
2043					 "encountered, ignoring transaction",
2044					 get_desc_trans_len(desc));
2045			return -1;
2046		}
2047		offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2048
2049		/*
2050		 * ok, we have a journal description block,
2051		 * let's see if the transaction was valid
2052		 */
2053		c_bh =
2054		    journal_bread(sb,
2055				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2056				  ((offset + get_desc_trans_len(desc) +
2057				    1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2058		if (!c_bh)
2059			return 0;
2060		commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2061		if (journal_compare_desc_commit(sb, desc, commit)) {
2062			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2063				       "journal_transaction_is_valid, commit offset %ld had bad "
2064				       "time %d or length %d",
2065				       c_bh->b_blocknr -
2066				       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2067				       get_commit_trans_id(commit),
2068				       get_commit_trans_len(commit));
2069			brelse(c_bh);
2070			if (oldest_invalid_trans_id) {
2071				*oldest_invalid_trans_id =
2072				    get_desc_trans_id(desc);
2073				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2074					       "journal-1004: "
2075					       "transaction_is_valid setting oldest invalid trans_id "
2076					       "to %d",
2077					       get_desc_trans_id(desc));
2078			}
2079			return -1;
2080		}
2081		brelse(c_bh);
2082		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2083			       "journal-1006: found valid "
2084			       "transaction start offset %llu, len %d id %d",
2085			       d_bh->b_blocknr -
2086			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2087			       get_desc_trans_len(desc),
2088			       get_desc_trans_id(desc));
2089		return 1;
2090	} else {
2091		return 0;
2092	}
2093}
2094
2095static void brelse_array(struct buffer_head **heads, int num)
2096{
2097	int i;
2098	for (i = 0; i < num; i++) {
2099		brelse(heads[i]);
2100	}
2101}
2102
2103/*
2104 * given the start, and values for the oldest acceptable transactions,
2105 * this either reads in a replays a transaction, or returns because the
2106 * transaction is invalid, or too old.
2107 * NOTE: only called during fs mount
2108 */
2109static int journal_read_transaction(struct super_block *sb,
2110				    unsigned long cur_dblock,
2111				    unsigned long oldest_start,
2112				    unsigned int oldest_trans_id,
2113				    unsigned long newest_mount_id)
2114{
2115	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2116	struct reiserfs_journal_desc *desc;
2117	struct reiserfs_journal_commit *commit;
2118	unsigned int trans_id = 0;
2119	struct buffer_head *c_bh;
2120	struct buffer_head *d_bh;
2121	struct buffer_head **log_blocks = NULL;
2122	struct buffer_head **real_blocks = NULL;
2123	unsigned int trans_offset;
2124	int i;
2125	int trans_half;
2126
2127	d_bh = journal_bread(sb, cur_dblock);
2128	if (!d_bh)
2129		return 1;
2130	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2131	trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2132	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2133		       "journal_read_transaction, offset %llu, len %d mount_id %d",
2134		       d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2135		       get_desc_trans_len(desc), get_desc_mount_id(desc));
2136	if (get_desc_trans_id(desc) < oldest_trans_id) {
2137		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2138			       "journal_read_trans skipping because %lu is too old",
2139			       cur_dblock -
2140			       SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2141		brelse(d_bh);
2142		return 1;
2143	}
2144	if (get_desc_mount_id(desc) != newest_mount_id) {
2145		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2146			       "journal_read_trans skipping because %d is != "
2147			       "newest_mount_id %lu", get_desc_mount_id(desc),
2148			       newest_mount_id);
2149		brelse(d_bh);
2150		return 1;
2151	}
2152	c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2153			     ((trans_offset + get_desc_trans_len(desc) + 1) %
2154			      SB_ONDISK_JOURNAL_SIZE(sb)));
2155	if (!c_bh) {
2156		brelse(d_bh);
2157		return 1;
2158	}
2159	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2160	if (journal_compare_desc_commit(sb, desc, commit)) {
2161		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2162			       "journal_read_transaction, "
2163			       "commit offset %llu had bad time %d or length %d",
2164			       c_bh->b_blocknr -
2165			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2166			       get_commit_trans_id(commit),
2167			       get_commit_trans_len(commit));
2168		brelse(c_bh);
2169		brelse(d_bh);
2170		return 1;
2171	}
2172
2173	if (bdev_read_only(sb->s_bdev)) {
2174		reiserfs_warning(sb, "clm-2076",
2175				 "device is readonly, unable to replay log");
2176		brelse(c_bh);
2177		brelse(d_bh);
2178		return -EROFS;
2179	}
2180
2181	trans_id = get_desc_trans_id(desc);
2182	/*
2183	 * now we know we've got a good transaction, and it was
2184	 * inside the valid time ranges
2185	 */
2186	log_blocks = kmalloc_array(get_desc_trans_len(desc),
2187				   sizeof(struct buffer_head *),
2188				   GFP_NOFS);
2189	real_blocks = kmalloc_array(get_desc_trans_len(desc),
2190				    sizeof(struct buffer_head *),
2191				    GFP_NOFS);
2192	if (!log_blocks || !real_blocks) {
2193		brelse(c_bh);
2194		brelse(d_bh);
2195		kfree(log_blocks);
2196		kfree(real_blocks);
2197		reiserfs_warning(sb, "journal-1169",
2198				 "kmalloc failed, unable to mount FS");
2199		return -1;
2200	}
2201	/* get all the buffer heads */
2202	trans_half = journal_trans_half(sb->s_blocksize);
2203	for (i = 0; i < get_desc_trans_len(desc); i++) {
2204		log_blocks[i] =
2205		    journal_getblk(sb,
2206				   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2207				   (trans_offset + 1 +
2208				    i) % SB_ONDISK_JOURNAL_SIZE(sb));
2209		if (i < trans_half) {
2210			real_blocks[i] =
2211			    sb_getblk(sb,
2212				      le32_to_cpu(desc->j_realblock[i]));
2213		} else {
2214			real_blocks[i] =
2215			    sb_getblk(sb,
2216				      le32_to_cpu(commit->
2217						  j_realblock[i - trans_half]));
2218		}
2219		if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2220			reiserfs_warning(sb, "journal-1207",
2221					 "REPLAY FAILURE fsck required! "
2222					 "Block to replay is outside of "
2223					 "filesystem");
2224			goto abort_replay;
2225		}
2226		/* make sure we don't try to replay onto log or reserved area */
2227		if (is_block_in_log_or_reserved_area
2228		    (sb, real_blocks[i]->b_blocknr)) {
2229			reiserfs_warning(sb, "journal-1204",
2230					 "REPLAY FAILURE fsck required! "
2231					 "Trying to replay onto a log block");
2232abort_replay:
2233			brelse_array(log_blocks, i);
2234			brelse_array(real_blocks, i);
2235			brelse(c_bh);
2236			brelse(d_bh);
2237			kfree(log_blocks);
2238			kfree(real_blocks);
2239			return -1;
2240		}
2241	}
2242	/* read in the log blocks, memcpy to the corresponding real block */
2243	bh_read_batch(get_desc_trans_len(desc), log_blocks);
2244	for (i = 0; i < get_desc_trans_len(desc); i++) {
2245
2246		wait_on_buffer(log_blocks[i]);
2247		if (!buffer_uptodate(log_blocks[i])) {
2248			reiserfs_warning(sb, "journal-1212",
2249					 "REPLAY FAILURE fsck required! "
2250					 "buffer write failed");
2251			brelse_array(log_blocks + i,
2252				     get_desc_trans_len(desc) - i);
2253			brelse_array(real_blocks, get_desc_trans_len(desc));
2254			brelse(c_bh);
2255			brelse(d_bh);
2256			kfree(log_blocks);
2257			kfree(real_blocks);
2258			return -1;
2259		}
2260		memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2261		       real_blocks[i]->b_size);
2262		set_buffer_uptodate(real_blocks[i]);
2263		brelse(log_blocks[i]);
2264	}
2265	/* flush out the real blocks */
2266	for (i = 0; i < get_desc_trans_len(desc); i++) {
2267		set_buffer_dirty(real_blocks[i]);
2268		write_dirty_buffer(real_blocks[i], 0);
2269	}
2270	for (i = 0; i < get_desc_trans_len(desc); i++) {
2271		wait_on_buffer(real_blocks[i]);
2272		if (!buffer_uptodate(real_blocks[i])) {
2273			reiserfs_warning(sb, "journal-1226",
2274					 "REPLAY FAILURE, fsck required! "
2275					 "buffer write failed");
2276			brelse_array(real_blocks + i,
2277				     get_desc_trans_len(desc) - i);
2278			brelse(c_bh);
2279			brelse(d_bh);
2280			kfree(log_blocks);
2281			kfree(real_blocks);
2282			return -1;
2283		}
2284		brelse(real_blocks[i]);
2285	}
2286	cur_dblock =
2287	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2288	    ((trans_offset + get_desc_trans_len(desc) +
2289	      2) % SB_ONDISK_JOURNAL_SIZE(sb));
2290	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2291		       "journal-1095: setting journal " "start to offset %ld",
2292		       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2293
2294	/*
2295	 * init starting values for the first transaction, in case
2296	 * this is the last transaction to be replayed.
2297	 */
2298	journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2299	journal->j_last_flush_trans_id = trans_id;
2300	journal->j_trans_id = trans_id + 1;
2301	/* check for trans_id overflow */
2302	if (journal->j_trans_id == 0)
2303		journal->j_trans_id = 10;
2304	brelse(c_bh);
2305	brelse(d_bh);
2306	kfree(log_blocks);
2307	kfree(real_blocks);
2308	return 0;
2309}
2310
2311/*
2312 * This function reads blocks starting from block and to max_block of bufsize
2313 * size (but no more than BUFNR blocks at a time). This proved to improve
2314 * mounting speed on self-rebuilding raid5 arrays at least.
2315 * Right now it is only used from journal code. But later we might use it
2316 * from other places.
2317 * Note: Do not use journal_getblk/sb_getblk functions here!
2318 */
2319static struct buffer_head *reiserfs_breada(struct block_device *dev,
2320					   b_blocknr_t block, int bufsize,
2321					   b_blocknr_t max_block)
2322{
2323	struct buffer_head *bhlist[BUFNR];
2324	unsigned int blocks = BUFNR;
2325	struct buffer_head *bh;
2326	int i, j;
2327
2328	bh = __getblk(dev, block, bufsize);
2329	if (buffer_uptodate(bh))
2330		return (bh);
2331
2332	if (block + BUFNR > max_block) {
2333		blocks = max_block - block;
2334	}
2335	bhlist[0] = bh;
2336	j = 1;
2337	for (i = 1; i < blocks; i++) {
2338		bh = __getblk(dev, block + i, bufsize);
2339		if (buffer_uptodate(bh)) {
2340			brelse(bh);
2341			break;
2342		} else
2343			bhlist[j++] = bh;
2344	}
2345	bh = bhlist[0];
2346	bh_read_nowait(bh, 0);
2347	bh_readahead_batch(j - 1, &bhlist[1], 0);
2348	for (i = 1; i < j; i++)
2349		brelse(bhlist[i]);
 
2350	wait_on_buffer(bh);
2351	if (buffer_uptodate(bh))
2352		return bh;
2353	brelse(bh);
2354	return NULL;
2355}
2356
2357/*
2358 * read and replay the log
2359 * on a clean unmount, the journal header's next unflushed pointer will be
2360 * to an invalid transaction.  This tests that before finding all the
2361 * transactions in the log, which makes normal mount times fast.
2362 *
2363 * After a crash, this starts with the next unflushed transaction, and
2364 * replays until it finds one too old, or invalid.
2365 *
2366 * On exit, it sets things up so the first transaction will work correctly.
2367 * NOTE: only called during fs mount
2368 */
2369static int journal_read(struct super_block *sb)
2370{
2371	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2372	struct reiserfs_journal_desc *desc;
2373	unsigned int oldest_trans_id = 0;
2374	unsigned int oldest_invalid_trans_id = 0;
2375	time64_t start;
2376	unsigned long oldest_start = 0;
2377	unsigned long cur_dblock = 0;
2378	unsigned long newest_mount_id = 9;
2379	struct buffer_head *d_bh;
2380	struct reiserfs_journal_header *jh;
2381	int valid_journal_header = 0;
2382	int replay_count = 0;
2383	int continue_replay = 1;
2384	int ret;
2385
2386	cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2387	reiserfs_info(sb, "checking transaction log (%pg)\n",
2388		      journal->j_dev_bd);
2389	start = ktime_get_seconds();
2390
2391	/*
2392	 * step 1, read in the journal header block.  Check the transaction
2393	 * it says is the first unflushed, and if that transaction is not
2394	 * valid, replay is done
2395	 */
2396	journal->j_header_bh = journal_bread(sb,
2397					     SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2398					     + SB_ONDISK_JOURNAL_SIZE(sb));
2399	if (!journal->j_header_bh) {
2400		return 1;
2401	}
2402	jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2403	if (le32_to_cpu(jh->j_first_unflushed_offset) <
2404	    SB_ONDISK_JOURNAL_SIZE(sb)
2405	    && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2406		oldest_start =
2407		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2408		    le32_to_cpu(jh->j_first_unflushed_offset);
2409		oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2410		newest_mount_id = le32_to_cpu(jh->j_mount_id);
2411		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2412			       "journal-1153: found in "
2413			       "header: first_unflushed_offset %d, last_flushed_trans_id "
2414			       "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2415			       le32_to_cpu(jh->j_last_flush_trans_id));
2416		valid_journal_header = 1;
2417
2418		/*
2419		 * now, we try to read the first unflushed offset.  If it
2420		 * is not valid, there is nothing more we can do, and it
2421		 * makes no sense to read through the whole log.
2422		 */
2423		d_bh =
2424		    journal_bread(sb,
2425				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2426				  le32_to_cpu(jh->j_first_unflushed_offset));
2427		ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2428		if (!ret) {
2429			continue_replay = 0;
2430		}
2431		brelse(d_bh);
2432		goto start_log_replay;
2433	}
2434
2435	/*
2436	 * ok, there are transactions that need to be replayed.  start
2437	 * with the first log block, find all the valid transactions, and
2438	 * pick out the oldest.
2439	 */
2440	while (continue_replay
2441	       && cur_dblock <
2442	       (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2443		SB_ONDISK_JOURNAL_SIZE(sb))) {
2444		/*
2445		 * Note that it is required for blocksize of primary fs
2446		 * device and journal device to be the same
2447		 */
2448		d_bh =
2449		    reiserfs_breada(journal->j_dev_bd, cur_dblock,
2450				    sb->s_blocksize,
2451				    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2452				    SB_ONDISK_JOURNAL_SIZE(sb));
2453		ret =
2454		    journal_transaction_is_valid(sb, d_bh,
2455						 &oldest_invalid_trans_id,
2456						 &newest_mount_id);
2457		if (ret == 1) {
2458			desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2459			if (oldest_start == 0) {	/* init all oldest_ values */
2460				oldest_trans_id = get_desc_trans_id(desc);
2461				oldest_start = d_bh->b_blocknr;
2462				newest_mount_id = get_desc_mount_id(desc);
2463				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2464					       "journal-1179: Setting "
2465					       "oldest_start to offset %llu, trans_id %lu",
2466					       oldest_start -
2467					       SB_ONDISK_JOURNAL_1st_BLOCK
2468					       (sb), oldest_trans_id);
2469			} else if (oldest_trans_id > get_desc_trans_id(desc)) {
2470				/* one we just read was older */
2471				oldest_trans_id = get_desc_trans_id(desc);
2472				oldest_start = d_bh->b_blocknr;
2473				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2474					       "journal-1180: Resetting "
2475					       "oldest_start to offset %lu, trans_id %lu",
2476					       oldest_start -
2477					       SB_ONDISK_JOURNAL_1st_BLOCK
2478					       (sb), oldest_trans_id);
2479			}
2480			if (newest_mount_id < get_desc_mount_id(desc)) {
2481				newest_mount_id = get_desc_mount_id(desc);
2482				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2483					       "journal-1299: Setting "
2484					       "newest_mount_id to %d",
2485					       get_desc_mount_id(desc));
2486			}
2487			cur_dblock += get_desc_trans_len(desc) + 2;
2488		} else {
2489			cur_dblock++;
2490		}
2491		brelse(d_bh);
2492	}
2493
2494start_log_replay:
2495	cur_dblock = oldest_start;
2496	if (oldest_trans_id) {
2497		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2498			       "journal-1206: Starting replay "
2499			       "from offset %llu, trans_id %lu",
2500			       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2501			       oldest_trans_id);
2502
2503	}
2504	replay_count = 0;
2505	while (continue_replay && oldest_trans_id > 0) {
2506		ret =
2507		    journal_read_transaction(sb, cur_dblock, oldest_start,
2508					     oldest_trans_id, newest_mount_id);
2509		if (ret < 0) {
2510			return ret;
2511		} else if (ret != 0) {
2512			break;
2513		}
2514		cur_dblock =
2515		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2516		replay_count++;
2517		if (cur_dblock == oldest_start)
2518			break;
2519	}
2520
2521	if (oldest_trans_id == 0) {
2522		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2523			       "journal-1225: No valid " "transactions found");
2524	}
2525	/*
2526	 * j_start does not get set correctly if we don't replay any
2527	 * transactions.  if we had a valid journal_header, set j_start
2528	 * to the first unflushed transaction value, copy the trans_id
2529	 * from the header
2530	 */
2531	if (valid_journal_header && replay_count == 0) {
2532		journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2533		journal->j_trans_id =
2534		    le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2535		/* check for trans_id overflow */
2536		if (journal->j_trans_id == 0)
2537			journal->j_trans_id = 10;
2538		journal->j_last_flush_trans_id =
2539		    le32_to_cpu(jh->j_last_flush_trans_id);
2540		journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2541	} else {
2542		journal->j_mount_id = newest_mount_id + 1;
2543	}
2544	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2545		       "newest_mount_id to %lu", journal->j_mount_id);
2546	journal->j_first_unflushed_offset = journal->j_start;
2547	if (replay_count > 0) {
2548		reiserfs_info(sb,
2549			      "replayed %d transactions in %lu seconds\n",
2550			      replay_count, ktime_get_seconds() - start);
2551	}
2552	/* needed to satisfy the locking in _update_journal_header_block */
2553	reiserfs_write_lock(sb);
2554	if (!bdev_read_only(sb->s_bdev) &&
2555	    _update_journal_header_block(sb, journal->j_start,
2556					 journal->j_last_flush_trans_id)) {
2557		reiserfs_write_unlock(sb);
2558		/*
2559		 * replay failed, caller must call free_journal_ram and abort
2560		 * the mount
2561		 */
2562		return -1;
2563	}
2564	reiserfs_write_unlock(sb);
2565	return 0;
2566}
2567
2568static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2569{
2570	struct reiserfs_journal_list *jl;
2571	jl = kzalloc(sizeof(struct reiserfs_journal_list),
2572		     GFP_NOFS | __GFP_NOFAIL);
2573	INIT_LIST_HEAD(&jl->j_list);
2574	INIT_LIST_HEAD(&jl->j_working_list);
2575	INIT_LIST_HEAD(&jl->j_tail_bh_list);
2576	INIT_LIST_HEAD(&jl->j_bh_list);
2577	mutex_init(&jl->j_commit_mutex);
2578	SB_JOURNAL(s)->j_num_lists++;
2579	get_journal_list(jl);
2580	return jl;
2581}
2582
2583static void journal_list_init(struct super_block *sb)
2584{
2585	SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2586}
2587
2588static void release_journal_dev(struct super_block *super,
2589			       struct reiserfs_journal *journal)
2590{
2591	if (journal->j_dev_bd != NULL) {
2592		blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2593		journal->j_dev_bd = NULL;
2594	}
2595}
2596
2597static int journal_init_dev(struct super_block *super,
2598			    struct reiserfs_journal *journal,
2599			    const char *jdev_name)
2600{
2601	int result;
2602	dev_t jdev;
2603	fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
 
2604
2605	result = 0;
2606
2607	journal->j_dev_bd = NULL;
2608	jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2609	    new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2610
2611	if (bdev_read_only(super->s_bdev))
2612		blkdev_mode = FMODE_READ;
2613
2614	/* there is no "jdev" option and journal is on separate device */
2615	if ((!jdev_name || !jdev_name[0])) {
2616		if (jdev == super->s_dev)
2617			blkdev_mode &= ~FMODE_EXCL;
2618		journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2619						      journal);
2620		journal->j_dev_mode = blkdev_mode;
2621		if (IS_ERR(journal->j_dev_bd)) {
2622			result = PTR_ERR(journal->j_dev_bd);
2623			journal->j_dev_bd = NULL;
2624			reiserfs_warning(super, "sh-458",
2625					 "cannot init journal device unknown-block(%u,%u): %i",
2626					 MAJOR(jdev), MINOR(jdev), result);
2627			return result;
2628		} else if (jdev != super->s_dev)
2629			set_blocksize(journal->j_dev_bd, super->s_blocksize);
2630
2631		return 0;
2632	}
2633
2634	journal->j_dev_mode = blkdev_mode;
2635	journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2636	if (IS_ERR(journal->j_dev_bd)) {
2637		result = PTR_ERR(journal->j_dev_bd);
2638		journal->j_dev_bd = NULL;
2639		reiserfs_warning(super, "sh-457",
2640				 "journal_init_dev: Cannot open '%s': %i",
2641				 jdev_name, result);
2642		return result;
2643	}
2644
2645	set_blocksize(journal->j_dev_bd, super->s_blocksize);
2646	reiserfs_info(super,
2647		      "journal_init_dev: journal device: %pg\n",
2648		      journal->j_dev_bd);
2649	return 0;
2650}
2651
2652/*
2653 * When creating/tuning a file system user can assign some
2654 * journal params within boundaries which depend on the ratio
2655 * blocksize/standard_blocksize.
2656 *
2657 * For blocks >= standard_blocksize transaction size should
2658 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2659 * then JOURNAL_TRANS_MAX_DEFAULT.
2660 *
2661 * For blocks < standard_blocksize these boundaries should be
2662 * decreased proportionally.
2663 */
2664#define REISERFS_STANDARD_BLKSIZE (4096)
2665
2666static int check_advise_trans_params(struct super_block *sb,
2667				     struct reiserfs_journal *journal)
2668{
2669        if (journal->j_trans_max) {
2670		/* Non-default journal params.  Do sanity check for them. */
2671	        int ratio = 1;
2672		if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2673		        ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2674
2675		if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2676		    journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2677		    SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2678		    JOURNAL_MIN_RATIO) {
2679			reiserfs_warning(sb, "sh-462",
2680					 "bad transaction max size (%u). "
2681					 "FSCK?", journal->j_trans_max);
2682			return 1;
2683		}
2684		if (journal->j_max_batch != (journal->j_trans_max) *
2685		        JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2686			reiserfs_warning(sb, "sh-463",
2687					 "bad transaction max batch (%u). "
2688					 "FSCK?", journal->j_max_batch);
2689			return 1;
2690		}
2691	} else {
2692		/*
2693		 * Default journal params.
2694		 * The file system was created by old version
2695		 * of mkreiserfs, so some fields contain zeros,
2696		 * and we need to advise proper values for them
2697		 */
2698		if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2699			reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2700					 sb->s_blocksize);
2701			return 1;
2702		}
2703		journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2704		journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2705		journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2706	}
2707	return 0;
2708}
2709
2710/* must be called once on fs mount.  calls journal_read for you */
2711int journal_init(struct super_block *sb, const char *j_dev_name,
2712		 int old_format, unsigned int commit_max_age)
2713{
2714	int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2715	struct buffer_head *bhjh;
2716	struct reiserfs_super_block *rs;
2717	struct reiserfs_journal_header *jh;
2718	struct reiserfs_journal *journal;
2719	struct reiserfs_journal_list *jl;
2720	int ret;
2721
2722	journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2723	if (!journal) {
2724		reiserfs_warning(sb, "journal-1256",
2725				 "unable to get memory for journal structure");
2726		return 1;
2727	}
2728	INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2729	INIT_LIST_HEAD(&journal->j_prealloc_list);
2730	INIT_LIST_HEAD(&journal->j_working_list);
2731	INIT_LIST_HEAD(&journal->j_journal_list);
2732	journal->j_persistent_trans = 0;
2733	if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2734					   reiserfs_bmap_count(sb)))
2735		goto free_and_return;
2736
2737	allocate_bitmap_nodes(sb);
2738
2739	/* reserved for journal area support */
2740	SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2741						 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2742						 / sb->s_blocksize +
2743						 reiserfs_bmap_count(sb) +
2744						 1 :
2745						 REISERFS_DISK_OFFSET_IN_BYTES /
2746						 sb->s_blocksize + 2);
2747
2748	/*
2749	 * Sanity check to see is the standard journal fitting
2750	 * within first bitmap (actual for small blocksizes)
2751	 */
2752	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2753	    (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2754	     SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2755		reiserfs_warning(sb, "journal-1393",
2756				 "journal does not fit for area addressed "
2757				 "by first of bitmap blocks. It starts at "
2758				 "%u and its size is %u. Block size %ld",
2759				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2760				 SB_ONDISK_JOURNAL_SIZE(sb),
2761				 sb->s_blocksize);
2762		goto free_and_return;
2763	}
2764
2765	/*
2766	 * Sanity check to see if journal first block is correct.
2767	 * If journal first block is invalid it can cause
2768	 * zeroing important superblock members.
2769	 */
2770	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2771	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) < SB_JOURNAL_1st_RESERVED_BLOCK(sb)) {
2772		reiserfs_warning(sb, "journal-1393",
2773				 "journal 1st super block is invalid: 1st reserved block %d, but actual 1st block is %d",
2774				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2775				 SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2776		goto free_and_return;
2777	}
2778
2779	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2780		reiserfs_warning(sb, "sh-462",
2781				 "unable to initialize journal device");
2782		goto free_and_return;
2783	}
2784
2785	rs = SB_DISK_SUPER_BLOCK(sb);
2786
2787	/* read journal header */
2788	bhjh = journal_bread(sb,
2789			     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2790			     SB_ONDISK_JOURNAL_SIZE(sb));
2791	if (!bhjh) {
2792		reiserfs_warning(sb, "sh-459",
2793				 "unable to read journal header");
2794		goto free_and_return;
2795	}
2796	jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2797
2798	/* make sure that journal matches to the super block */
2799	if (is_reiserfs_jr(rs)
2800	    && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2801		sb_jp_journal_magic(rs))) {
2802		reiserfs_warning(sb, "sh-460",
2803				 "journal header magic %x (device %pg) does "
2804				 "not match to magic found in super block %x",
2805				 jh->jh_journal.jp_journal_magic,
2806				 journal->j_dev_bd,
2807				 sb_jp_journal_magic(rs));
2808		brelse(bhjh);
2809		goto free_and_return;
2810	}
2811
2812	journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2813	journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2814	journal->j_max_commit_age =
2815	    le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2816	journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2817
2818	if (check_advise_trans_params(sb, journal) != 0)
2819	        goto free_and_return;
2820	journal->j_default_max_commit_age = journal->j_max_commit_age;
2821
2822	if (commit_max_age != 0) {
2823		journal->j_max_commit_age = commit_max_age;
2824		journal->j_max_trans_age = commit_max_age;
2825	}
2826
2827	reiserfs_info(sb, "journal params: device %pg, size %u, "
2828		      "journal first block %u, max trans len %u, max batch %u, "
2829		      "max commit age %u, max trans age %u\n",
2830		      journal->j_dev_bd,
2831		      SB_ONDISK_JOURNAL_SIZE(sb),
2832		      SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2833		      journal->j_trans_max,
2834		      journal->j_max_batch,
2835		      journal->j_max_commit_age, journal->j_max_trans_age);
2836
2837	brelse(bhjh);
2838
2839	journal->j_list_bitmap_index = 0;
2840	journal_list_init(sb);
2841
2842	memset(journal->j_list_hash_table, 0,
2843	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2844
2845	INIT_LIST_HEAD(&journal->j_dirty_buffers);
2846	spin_lock_init(&journal->j_dirty_buffers_lock);
2847
2848	journal->j_start = 0;
2849	journal->j_len = 0;
2850	journal->j_len_alloc = 0;
2851	atomic_set(&journal->j_wcount, 0);
2852	atomic_set(&journal->j_async_throttle, 0);
2853	journal->j_bcount = 0;
2854	journal->j_trans_start_time = 0;
2855	journal->j_last = NULL;
2856	journal->j_first = NULL;
2857	init_waitqueue_head(&journal->j_join_wait);
2858	mutex_init(&journal->j_mutex);
2859	mutex_init(&journal->j_flush_mutex);
2860
2861	journal->j_trans_id = 10;
2862	journal->j_mount_id = 10;
2863	journal->j_state = 0;
2864	atomic_set(&journal->j_jlock, 0);
2865	journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2866	journal->j_cnode_free_orig = journal->j_cnode_free_list;
2867	journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2868	journal->j_cnode_used = 0;
2869	journal->j_must_wait = 0;
2870
2871	if (journal->j_cnode_free == 0) {
2872		reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2873		                 "allocation failed (%ld bytes). Journal is "
2874		                 "too large for available memory. Usually "
2875		                 "this is due to a journal that is too large.",
2876		                 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2877        	goto free_and_return;
2878	}
2879
2880	init_journal_hash(sb);
2881	jl = journal->j_current_jl;
2882
2883	/*
2884	 * get_list_bitmap() may call flush_commit_list() which
2885	 * requires the lock. Calling flush_commit_list() shouldn't happen
2886	 * this early but I like to be paranoid.
2887	 */
2888	reiserfs_write_lock(sb);
2889	jl->j_list_bitmap = get_list_bitmap(sb, jl);
2890	reiserfs_write_unlock(sb);
2891	if (!jl->j_list_bitmap) {
2892		reiserfs_warning(sb, "journal-2005",
2893				 "get_list_bitmap failed for journal list 0");
2894		goto free_and_return;
2895	}
2896
2897	ret = journal_read(sb);
2898	if (ret < 0) {
2899		reiserfs_warning(sb, "reiserfs-2006",
2900				 "Replay Failure, unable to mount");
2901		goto free_and_return;
2902	}
2903
2904	INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2905	journal->j_work_sb = sb;
2906	return 0;
2907free_and_return:
2908	free_journal_ram(sb);
2909	return 1;
2910}
2911
2912/*
2913 * test for a polite end of the current transaction.  Used by file_write,
2914 * and should be used by delete to make sure they don't write more than
2915 * can fit inside a single transaction
2916 */
2917int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2918				   int new_alloc)
2919{
2920	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2921	time64_t now = ktime_get_seconds();
2922	/* cannot restart while nested */
2923	BUG_ON(!th->t_trans_id);
2924	if (th->t_refcount > 1)
2925		return 0;
2926	if (journal->j_must_wait > 0 ||
2927	    (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2928	    atomic_read(&journal->j_jlock) ||
2929	    (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2930	    journal->j_cnode_free < (journal->j_trans_max * 3)) {
2931		return 1;
2932	}
2933
2934	journal->j_len_alloc += new_alloc;
2935	th->t_blocks_allocated += new_alloc ;
2936	return 0;
2937}
2938
2939/* this must be called inside a transaction */
2940void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2941{
2942	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2943	BUG_ON(!th->t_trans_id);
2944	journal->j_must_wait = 1;
2945	set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2946	return;
2947}
2948
2949/* this must be called without a transaction started */
2950void reiserfs_allow_writes(struct super_block *s)
2951{
2952	struct reiserfs_journal *journal = SB_JOURNAL(s);
2953	clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2954	wake_up(&journal->j_join_wait);
2955}
2956
2957/* this must be called without a transaction started */
2958void reiserfs_wait_on_write_block(struct super_block *s)
2959{
2960	struct reiserfs_journal *journal = SB_JOURNAL(s);
2961	wait_event(journal->j_join_wait,
2962		   !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2963}
2964
2965static void queue_log_writer(struct super_block *s)
2966{
2967	wait_queue_entry_t wait;
2968	struct reiserfs_journal *journal = SB_JOURNAL(s);
2969	set_bit(J_WRITERS_QUEUED, &journal->j_state);
2970
2971	/*
2972	 * we don't want to use wait_event here because
2973	 * we only want to wait once.
2974	 */
2975	init_waitqueue_entry(&wait, current);
2976	add_wait_queue(&journal->j_join_wait, &wait);
2977	set_current_state(TASK_UNINTERRUPTIBLE);
2978	if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2979		int depth = reiserfs_write_unlock_nested(s);
2980		schedule();
2981		reiserfs_write_lock_nested(s, depth);
2982	}
2983	__set_current_state(TASK_RUNNING);
2984	remove_wait_queue(&journal->j_join_wait, &wait);
2985}
2986
2987static void wake_queued_writers(struct super_block *s)
2988{
2989	struct reiserfs_journal *journal = SB_JOURNAL(s);
2990	if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2991		wake_up(&journal->j_join_wait);
2992}
2993
2994static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2995{
2996	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2997	unsigned long bcount = journal->j_bcount;
2998	while (1) {
2999		int depth;
3000
3001		depth = reiserfs_write_unlock_nested(sb);
3002		schedule_timeout_uninterruptible(1);
3003		reiserfs_write_lock_nested(sb, depth);
3004
3005		journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
3006		while ((atomic_read(&journal->j_wcount) > 0 ||
3007			atomic_read(&journal->j_jlock)) &&
3008		       journal->j_trans_id == trans_id) {
3009			queue_log_writer(sb);
3010		}
3011		if (journal->j_trans_id != trans_id)
3012			break;
3013		if (bcount == journal->j_bcount)
3014			break;
3015		bcount = journal->j_bcount;
3016	}
3017}
3018
3019/*
3020 * join == true if you must join an existing transaction.
3021 * join == false if you can deal with waiting for others to finish
3022 *
3023 * this will block until the transaction is joinable.  send the number of
3024 * blocks you expect to use in nblocks.
3025*/
3026static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3027			      struct super_block *sb, unsigned long nblocks,
3028			      int join)
3029{
3030	time64_t now = ktime_get_seconds();
3031	unsigned int old_trans_id;
3032	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3033	struct reiserfs_transaction_handle myth;
3034	int sched_count = 0;
3035	int retval;
3036	int depth;
3037
3038	reiserfs_check_lock_depth(sb, "journal_begin");
3039	BUG_ON(nblocks > journal->j_trans_max);
3040
3041	PROC_INFO_INC(sb, journal.journal_being);
3042	/* set here for journal_join */
3043	th->t_refcount = 1;
3044	th->t_super = sb;
3045
3046relock:
3047	lock_journal(sb);
3048	if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3049		unlock_journal(sb);
3050		retval = journal->j_errno;
3051		goto out_fail;
3052	}
3053	journal->j_bcount++;
3054
3055	if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3056		unlock_journal(sb);
3057		depth = reiserfs_write_unlock_nested(sb);
3058		reiserfs_wait_on_write_block(sb);
3059		reiserfs_write_lock_nested(sb, depth);
3060		PROC_INFO_INC(sb, journal.journal_relock_writers);
3061		goto relock;
3062	}
3063	now = ktime_get_seconds();
3064
3065	/*
3066	 * if there is no room in the journal OR
3067	 * if this transaction is too old, and we weren't called joinable,
3068	 * wait for it to finish before beginning we don't sleep if there
3069	 * aren't other writers
3070	 */
3071
3072	if ((!join && journal->j_must_wait > 0) ||
3073	    (!join
3074	     && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3075	    || (!join && atomic_read(&journal->j_wcount) > 0
3076		&& journal->j_trans_start_time > 0
3077		&& (now - journal->j_trans_start_time) >
3078		journal->j_max_trans_age) || (!join
3079					      && atomic_read(&journal->j_jlock))
3080	    || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3081
3082		old_trans_id = journal->j_trans_id;
3083		/* allow others to finish this transaction */
3084		unlock_journal(sb);
3085
3086		if (!join && (journal->j_len_alloc + nblocks + 2) >=
3087		    journal->j_max_batch &&
3088		    ((journal->j_len + nblocks + 2) * 100) <
3089		    (journal->j_len_alloc * 75)) {
3090			if (atomic_read(&journal->j_wcount) > 10) {
3091				sched_count++;
3092				queue_log_writer(sb);
3093				goto relock;
3094			}
3095		}
3096		/*
3097		 * don't mess with joining the transaction if all we
3098		 * have to do is wait for someone else to do a commit
3099		 */
3100		if (atomic_read(&journal->j_jlock)) {
3101			while (journal->j_trans_id == old_trans_id &&
3102			       atomic_read(&journal->j_jlock)) {
3103				queue_log_writer(sb);
3104			}
3105			goto relock;
3106		}
3107		retval = journal_join(&myth, sb);
3108		if (retval)
3109			goto out_fail;
3110
3111		/* someone might have ended the transaction while we joined */
3112		if (old_trans_id != journal->j_trans_id) {
3113			retval = do_journal_end(&myth, 0);
3114		} else {
3115			retval = do_journal_end(&myth, COMMIT_NOW);
3116		}
3117
3118		if (retval)
3119			goto out_fail;
3120
3121		PROC_INFO_INC(sb, journal.journal_relock_wcount);
3122		goto relock;
3123	}
3124	/* we are the first writer, set trans_id */
3125	if (journal->j_trans_start_time == 0) {
3126		journal->j_trans_start_time = ktime_get_seconds();
3127	}
3128	atomic_inc(&journal->j_wcount);
3129	journal->j_len_alloc += nblocks;
3130	th->t_blocks_logged = 0;
3131	th->t_blocks_allocated = nblocks;
3132	th->t_trans_id = journal->j_trans_id;
3133	unlock_journal(sb);
3134	INIT_LIST_HEAD(&th->t_list);
3135	return 0;
3136
3137out_fail:
3138	memset(th, 0, sizeof(*th));
3139	/*
3140	 * Re-set th->t_super, so we can properly keep track of how many
3141	 * persistent transactions there are. We need to do this so if this
3142	 * call is part of a failed restart_transaction, we can free it later
3143	 */
3144	th->t_super = sb;
3145	return retval;
3146}
3147
3148struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3149								    super_block
3150								    *s,
3151								    int nblocks)
3152{
3153	int ret;
3154	struct reiserfs_transaction_handle *th;
3155
3156	/*
3157	 * if we're nesting into an existing transaction.  It will be
3158	 * persistent on its own
3159	 */
3160	if (reiserfs_transaction_running(s)) {
3161		th = current->journal_info;
3162		th->t_refcount++;
3163		BUG_ON(th->t_refcount < 2);
3164
3165		return th;
3166	}
3167	th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3168	if (!th)
3169		return NULL;
3170	ret = journal_begin(th, s, nblocks);
3171	if (ret) {
3172		kfree(th);
3173		return NULL;
3174	}
3175
3176	SB_JOURNAL(s)->j_persistent_trans++;
3177	return th;
3178}
3179
3180int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3181{
3182	struct super_block *s = th->t_super;
3183	int ret = 0;
3184	if (th->t_trans_id)
3185		ret = journal_end(th);
3186	else
3187		ret = -EIO;
3188	if (th->t_refcount == 0) {
3189		SB_JOURNAL(s)->j_persistent_trans--;
3190		kfree(th);
3191	}
3192	return ret;
3193}
3194
3195static int journal_join(struct reiserfs_transaction_handle *th,
3196			struct super_block *sb)
3197{
3198	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3199
3200	/*
3201	 * this keeps do_journal_end from NULLing out the
3202	 * current->journal_info pointer
3203	 */
3204	th->t_handle_save = cur_th;
3205	BUG_ON(cur_th && cur_th->t_refcount > 1);
3206	return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3207}
3208
3209int journal_join_abort(struct reiserfs_transaction_handle *th,
3210		       struct super_block *sb)
3211{
3212	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3213
3214	/*
3215	 * this keeps do_journal_end from NULLing out the
3216	 * current->journal_info pointer
3217	 */
3218	th->t_handle_save = cur_th;
3219	BUG_ON(cur_th && cur_th->t_refcount > 1);
3220	return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3221}
3222
3223int journal_begin(struct reiserfs_transaction_handle *th,
3224		  struct super_block *sb, unsigned long nblocks)
3225{
3226	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3227	int ret;
3228
3229	th->t_handle_save = NULL;
3230	if (cur_th) {
3231		/* we are nesting into the current transaction */
3232		if (cur_th->t_super == sb) {
3233			BUG_ON(!cur_th->t_refcount);
3234			cur_th->t_refcount++;
3235			memcpy(th, cur_th, sizeof(*th));
3236			if (th->t_refcount <= 1)
3237				reiserfs_warning(sb, "reiserfs-2005",
3238						 "BAD: refcount <= 1, but "
3239						 "journal_info != 0");
3240			return 0;
3241		} else {
3242			/*
3243			 * we've ended up with a handle from a different
3244			 * filesystem.  save it and restore on journal_end.
3245			 * This should never really happen...
3246			 */
3247			reiserfs_warning(sb, "clm-2100",
3248					 "nesting info a different FS");
3249			th->t_handle_save = current->journal_info;
3250			current->journal_info = th;
3251		}
3252	} else {
3253		current->journal_info = th;
3254	}
3255	ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3256	BUG_ON(current->journal_info != th);
3257
3258	/*
3259	 * I guess this boils down to being the reciprocal of clm-2100 above.
3260	 * If do_journal_begin_r fails, we need to put it back, since
3261	 * journal_end won't be called to do it. */
3262	if (ret)
3263		current->journal_info = th->t_handle_save;
3264	else
3265		BUG_ON(!th->t_refcount);
3266
3267	return ret;
3268}
3269
3270/*
3271 * puts bh into the current transaction.  If it was already there, reorders
3272 * removes the old pointers from the hash, and puts new ones in (to make
3273 * sure replay happen in the right order).
3274 *
3275 * if it was dirty, cleans and files onto the clean list.  I can't let it
3276 * be dirty again until the transaction is committed.
3277 *
3278 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3279 */
3280int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3281		       struct buffer_head *bh)
3282{
3283	struct super_block *sb = th->t_super;
3284	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3285	struct reiserfs_journal_cnode *cn = NULL;
3286	int count_already_incd = 0;
3287	int prepared = 0;
3288	BUG_ON(!th->t_trans_id);
3289
3290	PROC_INFO_INC(sb, journal.mark_dirty);
3291	if (th->t_trans_id != journal->j_trans_id) {
3292		reiserfs_panic(th->t_super, "journal-1577",
3293			       "handle trans id %ld != current trans id %ld",
3294			       th->t_trans_id, journal->j_trans_id);
3295	}
3296
3297	prepared = test_clear_buffer_journal_prepared(bh);
3298	clear_buffer_journal_restore_dirty(bh);
3299	/* already in this transaction, we are done */
3300	if (buffer_journaled(bh)) {
3301		PROC_INFO_INC(sb, journal.mark_dirty_already);
3302		return 0;
3303	}
3304
3305	/*
3306	 * this must be turned into a panic instead of a warning.  We can't
3307	 * allow a dirty or journal_dirty or locked buffer to be logged, as
3308	 * some changes could get to disk too early.  NOT GOOD.
3309	 */
3310	if (!prepared || buffer_dirty(bh)) {
3311		reiserfs_warning(sb, "journal-1777",
3312				 "buffer %llu bad state "
3313				 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3314				 (unsigned long long)bh->b_blocknr,
3315				 prepared ? ' ' : '!',
3316				 buffer_locked(bh) ? ' ' : '!',
3317				 buffer_dirty(bh) ? ' ' : '!',
3318				 buffer_journal_dirty(bh) ? ' ' : '!');
3319	}
3320
3321	if (atomic_read(&journal->j_wcount) <= 0) {
3322		reiserfs_warning(sb, "journal-1409",
3323				 "returning because j_wcount was %d",
3324				 atomic_read(&journal->j_wcount));
3325		return 1;
3326	}
3327	/*
3328	 * this error means I've screwed up, and we've overflowed
3329	 * the transaction.  Nothing can be done here, except make the
3330	 * FS readonly or panic.
3331	 */
3332	if (journal->j_len >= journal->j_trans_max) {
3333		reiserfs_panic(th->t_super, "journal-1413",
3334			       "j_len (%lu) is too big",
3335			       journal->j_len);
3336	}
3337
3338	if (buffer_journal_dirty(bh)) {
3339		count_already_incd = 1;
3340		PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3341		clear_buffer_journal_dirty(bh);
3342	}
3343
3344	if (journal->j_len > journal->j_len_alloc) {
3345		journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3346	}
3347
3348	set_buffer_journaled(bh);
3349
3350	/* now put this guy on the end */
3351	if (!cn) {
3352		cn = get_cnode(sb);
3353		if (!cn) {
3354			reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3355		}
3356
3357		if (th->t_blocks_logged == th->t_blocks_allocated) {
3358			th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3359			journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3360		}
3361		th->t_blocks_logged++;
3362		journal->j_len++;
3363
3364		cn->bh = bh;
3365		cn->blocknr = bh->b_blocknr;
3366		cn->sb = sb;
3367		cn->jlist = NULL;
3368		insert_journal_hash(journal->j_hash_table, cn);
3369		if (!count_already_incd) {
3370			get_bh(bh);
3371		}
3372	}
3373	cn->next = NULL;
3374	cn->prev = journal->j_last;
3375	cn->bh = bh;
3376	if (journal->j_last) {
3377		journal->j_last->next = cn;
3378		journal->j_last = cn;
3379	} else {
3380		journal->j_first = cn;
3381		journal->j_last = cn;
3382	}
3383	reiserfs_schedule_old_flush(sb);
3384	return 0;
3385}
3386
3387int journal_end(struct reiserfs_transaction_handle *th)
3388{
3389	struct super_block *sb = th->t_super;
3390	if (!current->journal_info && th->t_refcount > 1)
3391		reiserfs_warning(sb, "REISER-NESTING",
3392				 "th NULL, refcount %d", th->t_refcount);
3393
3394	if (!th->t_trans_id) {
3395		WARN_ON(1);
3396		return -EIO;
3397	}
3398
3399	th->t_refcount--;
3400	if (th->t_refcount > 0) {
3401		struct reiserfs_transaction_handle *cur_th =
3402		    current->journal_info;
3403
3404		/*
3405		 * we aren't allowed to close a nested transaction on a
3406		 * different filesystem from the one in the task struct
3407		 */
3408		BUG_ON(cur_th->t_super != th->t_super);
3409
3410		if (th != cur_th) {
3411			memcpy(current->journal_info, th, sizeof(*th));
3412			th->t_trans_id = 0;
3413		}
3414		return 0;
3415	} else {
3416		return do_journal_end(th, 0);
3417	}
3418}
3419
3420/*
3421 * removes from the current transaction, relsing and descrementing any counters.
3422 * also files the removed buffer directly onto the clean list
3423 *
3424 * called by journal_mark_freed when a block has been deleted
3425 *
3426 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3427 */
3428static int remove_from_transaction(struct super_block *sb,
3429				   b_blocknr_t blocknr, int already_cleaned)
3430{
3431	struct buffer_head *bh;
3432	struct reiserfs_journal_cnode *cn;
3433	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3434	int ret = 0;
3435
3436	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3437	if (!cn || !cn->bh) {
3438		return ret;
3439	}
3440	bh = cn->bh;
3441	if (cn->prev) {
3442		cn->prev->next = cn->next;
3443	}
3444	if (cn->next) {
3445		cn->next->prev = cn->prev;
3446	}
3447	if (cn == journal->j_first) {
3448		journal->j_first = cn->next;
3449	}
3450	if (cn == journal->j_last) {
3451		journal->j_last = cn->prev;
3452	}
3453	remove_journal_hash(sb, journal->j_hash_table, NULL,
3454			    bh->b_blocknr, 0);
3455	clear_buffer_journaled(bh);	/* don't log this one */
3456
3457	if (!already_cleaned) {
3458		clear_buffer_journal_dirty(bh);
3459		clear_buffer_dirty(bh);
3460		clear_buffer_journal_test(bh);
3461		put_bh(bh);
3462		if (atomic_read(&bh->b_count) < 0) {
3463			reiserfs_warning(sb, "journal-1752",
3464					 "b_count < 0");
3465		}
3466		ret = 1;
3467	}
3468	journal->j_len--;
3469	journal->j_len_alloc--;
3470	free_cnode(sb, cn);
3471	return ret;
3472}
3473
3474/*
3475 * for any cnode in a journal list, it can only be dirtied of all the
3476 * transactions that include it are committed to disk.
3477 * this checks through each transaction, and returns 1 if you are allowed
3478 * to dirty, and 0 if you aren't
3479 *
3480 * it is called by dirty_journal_list, which is called after
3481 * flush_commit_list has gotten all the log blocks for a given
3482 * transaction on disk
3483 *
3484 */
3485static int can_dirty(struct reiserfs_journal_cnode *cn)
3486{
3487	struct super_block *sb = cn->sb;
3488	b_blocknr_t blocknr = cn->blocknr;
3489	struct reiserfs_journal_cnode *cur = cn->hprev;
3490	int can_dirty = 1;
3491
3492	/*
3493	 * first test hprev.  These are all newer than cn, so any node here
3494	 * with the same block number and dev means this node can't be sent
3495	 * to disk right now.
3496	 */
3497	while (cur && can_dirty) {
3498		if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3499		    cur->blocknr == blocknr) {
3500			can_dirty = 0;
3501		}
3502		cur = cur->hprev;
3503	}
3504	/*
3505	 * then test hnext.  These are all older than cn.  As long as they
3506	 * are committed to the log, it is safe to write cn to disk
3507	 */
3508	cur = cn->hnext;
3509	while (cur && can_dirty) {
3510		if (cur->jlist && cur->jlist->j_len > 0 &&
3511		    atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3512		    cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3513			can_dirty = 0;
3514		}
3515		cur = cur->hnext;
3516	}
3517	return can_dirty;
3518}
3519
3520/*
3521 * syncs the commit blocks, but does not force the real buffers to disk
3522 * will wait until the current transaction is done/committed before returning
3523 */
3524int journal_end_sync(struct reiserfs_transaction_handle *th)
3525{
3526	struct super_block *sb = th->t_super;
3527	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3528
3529	BUG_ON(!th->t_trans_id);
3530	/* you can sync while nested, very, very bad */
3531	BUG_ON(th->t_refcount > 1);
3532	if (journal->j_len == 0) {
3533		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3534					     1);
3535		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3536	}
3537	return do_journal_end(th, COMMIT_NOW | WAIT);
3538}
3539
3540/* writeback the pending async commits to disk */
3541static void flush_async_commits(struct work_struct *work)
3542{
3543	struct reiserfs_journal *journal =
3544		container_of(work, struct reiserfs_journal, j_work.work);
3545	struct super_block *sb = journal->j_work_sb;
3546	struct reiserfs_journal_list *jl;
3547	struct list_head *entry;
3548
3549	reiserfs_write_lock(sb);
3550	if (!list_empty(&journal->j_journal_list)) {
3551		/* last entry is the youngest, commit it and you get everything */
3552		entry = journal->j_journal_list.prev;
3553		jl = JOURNAL_LIST_ENTRY(entry);
3554		flush_commit_list(sb, jl, 1);
3555	}
3556	reiserfs_write_unlock(sb);
3557}
3558
3559/*
3560 * flushes any old transactions to disk
3561 * ends the current transaction if it is too old
3562 */
3563void reiserfs_flush_old_commits(struct super_block *sb)
3564{
3565	time64_t now;
3566	struct reiserfs_transaction_handle th;
3567	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3568
3569	now = ktime_get_seconds();
3570	/*
3571	 * safety check so we don't flush while we are replaying the log during
3572	 * mount
3573	 */
3574	if (list_empty(&journal->j_journal_list))
3575		return;
3576
3577	/*
3578	 * check the current transaction.  If there are no writers, and it is
3579	 * too old, finish it, and force the commit blocks to disk
3580	 */
3581	if (atomic_read(&journal->j_wcount) <= 0 &&
3582	    journal->j_trans_start_time > 0 &&
3583	    journal->j_len > 0 &&
3584	    (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3585		if (!journal_join(&th, sb)) {
3586			reiserfs_prepare_for_journal(sb,
3587						     SB_BUFFER_WITH_SB(sb),
3588						     1);
3589			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3590
3591			/*
3592			 * we're only being called from kreiserfsd, it makes
3593			 * no sense to do an async commit so that kreiserfsd
3594			 * can do it later
3595			 */
3596			do_journal_end(&th, COMMIT_NOW | WAIT);
3597		}
3598	}
3599}
3600
3601/*
3602 * returns 0 if do_journal_end should return right away, returns 1 if
3603 * do_journal_end should finish the commit
3604 *
3605 * if the current transaction is too old, but still has writers, this will
3606 * wait on j_join_wait until all the writers are done.  By the time it
3607 * wakes up, the transaction it was called has already ended, so it just
3608 * flushes the commit list and returns 0.
3609 *
3610 * Won't batch when flush or commit_now is set.  Also won't batch when
3611 * others are waiting on j_join_wait.
3612 *
3613 * Note, we can't allow the journal_end to proceed while there are still
3614 * writers in the log.
3615 */
3616static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3617{
3618
3619	time64_t now;
3620	int flush = flags & FLUSH_ALL;
3621	int commit_now = flags & COMMIT_NOW;
3622	int wait_on_commit = flags & WAIT;
3623	struct reiserfs_journal_list *jl;
3624	struct super_block *sb = th->t_super;
3625	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3626
3627	BUG_ON(!th->t_trans_id);
3628
3629	if (th->t_trans_id != journal->j_trans_id) {
3630		reiserfs_panic(th->t_super, "journal-1577",
3631			       "handle trans id %ld != current trans id %ld",
3632			       th->t_trans_id, journal->j_trans_id);
3633	}
3634
3635	journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3636	/* <= 0 is allowed.  unmounting might not call begin */
3637	if (atomic_read(&journal->j_wcount) > 0)
3638		atomic_dec(&journal->j_wcount);
3639
3640	/*
3641	 * BUG, deal with case where j_len is 0, but people previously
3642	 * freed blocks need to be released will be dealt with by next
3643	 * transaction that actually writes something, but should be taken
3644	 * care of in this trans
3645	 */
3646	BUG_ON(journal->j_len == 0);
3647
3648	/*
3649	 * if wcount > 0, and we are called to with flush or commit_now,
3650	 * we wait on j_join_wait.  We will wake up when the last writer has
3651	 * finished the transaction, and started it on its way to the disk.
3652	 * Then, we flush the commit or journal list, and just return 0
3653	 * because the rest of journal end was already done for this
3654	 * transaction.
3655	 */
3656	if (atomic_read(&journal->j_wcount) > 0) {
3657		if (flush || commit_now) {
3658			unsigned trans_id;
3659
3660			jl = journal->j_current_jl;
3661			trans_id = jl->j_trans_id;
3662			if (wait_on_commit)
3663				jl->j_state |= LIST_COMMIT_PENDING;
3664			atomic_set(&journal->j_jlock, 1);
3665			if (flush) {
3666				journal->j_next_full_flush = 1;
3667			}
3668			unlock_journal(sb);
3669
3670			/*
3671			 * sleep while the current transaction is
3672			 * still j_jlocked
3673			 */
3674			while (journal->j_trans_id == trans_id) {
3675				if (atomic_read(&journal->j_jlock)) {
3676					queue_log_writer(sb);
3677				} else {
3678					lock_journal(sb);
3679					if (journal->j_trans_id == trans_id) {
3680						atomic_set(&journal->j_jlock,
3681							   1);
3682					}
3683					unlock_journal(sb);
3684				}
3685			}
3686			BUG_ON(journal->j_trans_id == trans_id);
3687
3688			if (commit_now
3689			    && journal_list_still_alive(sb, trans_id)
3690			    && wait_on_commit) {
3691				flush_commit_list(sb, jl, 1);
3692			}
3693			return 0;
3694		}
3695		unlock_journal(sb);
3696		return 0;
3697	}
3698
3699	/* deal with old transactions where we are the last writers */
3700	now = ktime_get_seconds();
3701	if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3702		commit_now = 1;
3703		journal->j_next_async_flush = 1;
3704	}
3705	/* don't batch when someone is waiting on j_join_wait */
3706	/* don't batch when syncing the commit or flushing the whole trans */
3707	if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3708	    && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3709	    && journal->j_len_alloc < journal->j_max_batch
3710	    && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3711		journal->j_bcount++;
3712		unlock_journal(sb);
3713		return 0;
3714	}
3715
3716	if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3717		reiserfs_panic(sb, "journal-003",
3718			       "j_start (%ld) is too high",
3719			       journal->j_start);
3720	}
3721	return 1;
3722}
3723
3724/*
3725 * Does all the work that makes deleting blocks safe.
3726 * when deleting a block mark BH_JNew, just remove it from the current
3727 * transaction, clean it's buffer_head and move on.
3728 *
3729 * otherwise:
3730 * set a bit for the block in the journal bitmap.  That will prevent it from
3731 * being allocated for unformatted nodes before this transaction has finished.
3732 *
3733 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3734 * That will prevent any old transactions with this block from trying to flush
3735 * to the real location.  Since we aren't removing the cnode from the
3736 * journal_list_hash, *the block can't be reallocated yet.
3737 *
3738 * Then remove it from the current transaction, decrementing any counters and
3739 * filing it on the clean list.
3740 */
3741int journal_mark_freed(struct reiserfs_transaction_handle *th,
3742		       struct super_block *sb, b_blocknr_t blocknr)
3743{
3744	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3745	struct reiserfs_journal_cnode *cn = NULL;
3746	struct buffer_head *bh = NULL;
3747	struct reiserfs_list_bitmap *jb = NULL;
3748	int cleaned = 0;
3749	BUG_ON(!th->t_trans_id);
3750
3751	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3752	if (cn && cn->bh) {
3753		bh = cn->bh;
3754		get_bh(bh);
3755	}
3756	/* if it is journal new, we just remove it from this transaction */
3757	if (bh && buffer_journal_new(bh)) {
3758		clear_buffer_journal_new(bh);
3759		clear_prepared_bits(bh);
3760		reiserfs_clean_and_file_buffer(bh);
3761		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3762	} else {
3763		/*
3764		 * set the bit for this block in the journal bitmap
3765		 * for this transaction
3766		 */
3767		jb = journal->j_current_jl->j_list_bitmap;
3768		if (!jb) {
3769			reiserfs_panic(sb, "journal-1702",
3770				       "journal_list_bitmap is NULL");
3771		}
3772		set_bit_in_list_bitmap(sb, blocknr, jb);
3773
3774		/* Note, the entire while loop is not allowed to schedule.  */
3775
3776		if (bh) {
3777			clear_prepared_bits(bh);
3778			reiserfs_clean_and_file_buffer(bh);
3779		}
3780		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3781
3782		/*
3783		 * find all older transactions with this block,
3784		 * make sure they don't try to write it out
3785		 */
3786		cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3787					  blocknr);
3788		while (cn) {
3789			if (sb == cn->sb && blocknr == cn->blocknr) {
3790				set_bit(BLOCK_FREED, &cn->state);
3791				if (cn->bh) {
3792					/*
3793					 * remove_from_transaction will brelse
3794					 * the buffer if it was in the current
3795					 * trans
3796					 */
3797					if (!cleaned) {
3798						clear_buffer_journal_dirty(cn->
3799									   bh);
3800						clear_buffer_dirty(cn->bh);
3801						clear_buffer_journal_test(cn->
3802									  bh);
3803						cleaned = 1;
3804						put_bh(cn->bh);
3805						if (atomic_read
3806						    (&cn->bh->b_count) < 0) {
3807							reiserfs_warning(sb,
3808								 "journal-2138",
3809								 "cn->bh->b_count < 0");
3810						}
3811					}
3812					/*
3813					 * since we are clearing the bh,
3814					 * we MUST dec nonzerolen
3815					 */
3816					if (cn->jlist) {
3817						atomic_dec(&cn->jlist->
3818							   j_nonzerolen);
3819					}
3820					cn->bh = NULL;
3821				}
3822			}
3823			cn = cn->hnext;
3824		}
3825	}
3826
3827	if (bh)
3828		release_buffer_page(bh); /* get_hash grabs the buffer */
3829	return 0;
3830}
3831
3832void reiserfs_update_inode_transaction(struct inode *inode)
3833{
3834	struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3835	REISERFS_I(inode)->i_jl = journal->j_current_jl;
3836	REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3837}
3838
3839/*
3840 * returns -1 on error, 0 if no commits/barriers were done and 1
3841 * if a transaction was actually committed and the barrier was done
3842 */
3843static int __commit_trans_jl(struct inode *inode, unsigned long id,
3844			     struct reiserfs_journal_list *jl)
3845{
3846	struct reiserfs_transaction_handle th;
3847	struct super_block *sb = inode->i_sb;
3848	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3849	int ret = 0;
3850
3851	/*
3852	 * is it from the current transaction,
3853	 * or from an unknown transaction?
3854	 */
3855	if (id == journal->j_trans_id) {
3856		jl = journal->j_current_jl;
3857		/*
3858		 * try to let other writers come in and
3859		 * grow this transaction
3860		 */
3861		let_transaction_grow(sb, id);
3862		if (journal->j_trans_id != id) {
3863			goto flush_commit_only;
3864		}
3865
3866		ret = journal_begin(&th, sb, 1);
3867		if (ret)
3868			return ret;
3869
3870		/* someone might have ended this transaction while we joined */
3871		if (journal->j_trans_id != id) {
3872			reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3873						     1);
3874			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3875			ret = journal_end(&th);
3876			goto flush_commit_only;
3877		}
3878
3879		ret = journal_end_sync(&th);
3880		if (!ret)
3881			ret = 1;
3882
3883	} else {
3884		/*
3885		 * this gets tricky, we have to make sure the journal list in
3886		 * the inode still exists.  We know the list is still around
3887		 * if we've got a larger transaction id than the oldest list
3888		 */
3889flush_commit_only:
3890		if (journal_list_still_alive(inode->i_sb, id)) {
3891			/*
3892			 * we only set ret to 1 when we know for sure
3893			 * the barrier hasn't been started yet on the commit
3894			 * block.
3895			 */
3896			if (atomic_read(&jl->j_commit_left) > 1)
3897				ret = 1;
3898			flush_commit_list(sb, jl, 1);
3899			if (journal->j_errno)
3900				ret = journal->j_errno;
3901		}
3902	}
3903	/* otherwise the list is gone, and long since committed */
3904	return ret;
3905}
3906
3907int reiserfs_commit_for_inode(struct inode *inode)
3908{
3909	unsigned int id = REISERFS_I(inode)->i_trans_id;
3910	struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3911
3912	/*
3913	 * for the whole inode, assume unset id means it was
3914	 * changed in the current transaction.  More conservative
3915	 */
3916	if (!id || !jl) {
3917		reiserfs_update_inode_transaction(inode);
3918		id = REISERFS_I(inode)->i_trans_id;
3919		/* jl will be updated in __commit_trans_jl */
3920	}
3921
3922	return __commit_trans_jl(inode, id, jl);
3923}
3924
3925void reiserfs_restore_prepared_buffer(struct super_block *sb,
3926				      struct buffer_head *bh)
3927{
3928	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3929	PROC_INFO_INC(sb, journal.restore_prepared);
3930	if (!bh) {
3931		return;
3932	}
3933	if (test_clear_buffer_journal_restore_dirty(bh) &&
3934	    buffer_journal_dirty(bh)) {
3935		struct reiserfs_journal_cnode *cn;
3936		reiserfs_write_lock(sb);
3937		cn = get_journal_hash_dev(sb,
3938					  journal->j_list_hash_table,
3939					  bh->b_blocknr);
3940		if (cn && can_dirty(cn)) {
3941			set_buffer_journal_test(bh);
3942			mark_buffer_dirty(bh);
3943		}
3944		reiserfs_write_unlock(sb);
3945	}
3946	clear_buffer_journal_prepared(bh);
3947}
3948
3949extern struct tree_balance *cur_tb;
3950/*
3951 * before we can change a metadata block, we have to make sure it won't
3952 * be written to disk while we are altering it.  So, we must:
3953 * clean it
3954 * wait on it.
3955 */
3956int reiserfs_prepare_for_journal(struct super_block *sb,
3957				 struct buffer_head *bh, int wait)
3958{
3959	PROC_INFO_INC(sb, journal.prepare);
3960
3961	if (!trylock_buffer(bh)) {
3962		if (!wait)
3963			return 0;
3964		lock_buffer(bh);
3965	}
3966	set_buffer_journal_prepared(bh);
3967	if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3968		clear_buffer_journal_test(bh);
3969		set_buffer_journal_restore_dirty(bh);
3970	}
3971	unlock_buffer(bh);
3972	return 1;
3973}
3974
3975/*
3976 * long and ugly.  If flush, will not return until all commit
3977 * blocks and all real buffers in the trans are on disk.
3978 * If no_async, won't return until all commit blocks are on disk.
3979 *
3980 * keep reading, there are comments as you go along
3981 *
3982 * If the journal is aborted, we just clean up. Things like flushing
3983 * journal lists, etc just won't happen.
3984 */
3985static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3986{
3987	struct super_block *sb = th->t_super;
3988	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3989	struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3990	struct reiserfs_journal_cnode *last_cn = NULL;
3991	struct reiserfs_journal_desc *desc;
3992	struct reiserfs_journal_commit *commit;
3993	struct buffer_head *c_bh;	/* commit bh */
3994	struct buffer_head *d_bh;	/* desc bh */
3995	int cur_write_start = 0;	/* start index of current log write */
3996	int i;
3997	int flush;
3998	int wait_on_commit;
3999	struct reiserfs_journal_list *jl, *temp_jl;
4000	struct list_head *entry, *safe;
4001	unsigned long jindex;
4002	unsigned int commit_trans_id;
4003	int trans_half;
4004	int depth;
4005
4006	BUG_ON(th->t_refcount > 1);
4007	BUG_ON(!th->t_trans_id);
4008	BUG_ON(!th->t_super);
4009
4010	/*
4011	 * protect flush_older_commits from doing mistakes if the
4012	 * transaction ID counter gets overflowed.
4013	 */
4014	if (th->t_trans_id == ~0U)
4015		flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4016	flush = flags & FLUSH_ALL;
4017	wait_on_commit = flags & WAIT;
4018
4019	current->journal_info = th->t_handle_save;
4020	reiserfs_check_lock_depth(sb, "journal end");
4021	if (journal->j_len == 0) {
4022		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4023					     1);
4024		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4025	}
4026
4027	lock_journal(sb);
4028	if (journal->j_next_full_flush) {
4029		flags |= FLUSH_ALL;
4030		flush = 1;
4031	}
4032	if (journal->j_next_async_flush) {
4033		flags |= COMMIT_NOW | WAIT;
4034		wait_on_commit = 1;
4035	}
4036
4037	/*
4038	 * check_journal_end locks the journal, and unlocks if it does
4039	 * not return 1 it tells us if we should continue with the
4040	 * journal_end, or just return
4041	 */
4042	if (!check_journal_end(th, flags)) {
4043		reiserfs_schedule_old_flush(sb);
4044		wake_queued_writers(sb);
4045		reiserfs_async_progress_wait(sb);
4046		goto out;
4047	}
4048
4049	/* check_journal_end might set these, check again */
4050	if (journal->j_next_full_flush) {
4051		flush = 1;
4052	}
4053
4054	/*
4055	 * j must wait means we have to flush the log blocks, and the
4056	 * real blocks for this transaction
4057	 */
4058	if (journal->j_must_wait > 0) {
4059		flush = 1;
4060	}
4061#ifdef REISERFS_PREALLOCATE
4062	/*
4063	 * quota ops might need to nest, setup the journal_info pointer
4064	 * for them and raise the refcount so that it is > 0.
4065	 */
4066	current->journal_info = th;
4067	th->t_refcount++;
4068
4069	/* it should not involve new blocks into the transaction */
4070	reiserfs_discard_all_prealloc(th);
4071
4072	th->t_refcount--;
4073	current->journal_info = th->t_handle_save;
4074#endif
4075
4076	/* setup description block */
4077	d_bh =
4078	    journal_getblk(sb,
4079			   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4080			   journal->j_start);
4081	set_buffer_uptodate(d_bh);
4082	desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4083	memset(d_bh->b_data, 0, d_bh->b_size);
4084	memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4085	set_desc_trans_id(desc, journal->j_trans_id);
4086
4087	/*
4088	 * setup commit block.  Don't write (keep it clean too) this one
4089	 * until after everyone else is written
4090	 */
4091	c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4092			      ((journal->j_start + journal->j_len +
4093				1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4094	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4095	memset(c_bh->b_data, 0, c_bh->b_size);
4096	set_commit_trans_id(commit, journal->j_trans_id);
4097	set_buffer_uptodate(c_bh);
4098
4099	/* init this journal list */
4100	jl = journal->j_current_jl;
4101
4102	/*
4103	 * we lock the commit before doing anything because
4104	 * we want to make sure nobody tries to run flush_commit_list until
4105	 * the new transaction is fully setup, and we've already flushed the
4106	 * ordered bh list
4107	 */
4108	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4109
4110	/* save the transaction id in case we need to commit it later */
4111	commit_trans_id = jl->j_trans_id;
4112
4113	atomic_set(&jl->j_older_commits_done, 0);
4114	jl->j_trans_id = journal->j_trans_id;
4115	jl->j_timestamp = journal->j_trans_start_time;
4116	jl->j_commit_bh = c_bh;
4117	jl->j_start = journal->j_start;
4118	jl->j_len = journal->j_len;
4119	atomic_set(&jl->j_nonzerolen, journal->j_len);
4120	atomic_set(&jl->j_commit_left, journal->j_len + 2);
4121	jl->j_realblock = NULL;
4122
4123	/*
4124	 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4125	 * for each real block, add it to the journal list hash,
4126	 * copy into real block index array in the commit or desc block
4127	 */
4128	trans_half = journal_trans_half(sb->s_blocksize);
4129	for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4130		if (buffer_journaled(cn->bh)) {
4131			jl_cn = get_cnode(sb);
4132			if (!jl_cn) {
4133				reiserfs_panic(sb, "journal-1676",
4134					       "get_cnode returned NULL");
4135			}
4136			if (i == 0) {
4137				jl->j_realblock = jl_cn;
4138			}
4139			jl_cn->prev = last_cn;
4140			jl_cn->next = NULL;
4141			if (last_cn) {
4142				last_cn->next = jl_cn;
4143			}
4144			last_cn = jl_cn;
4145			/*
4146			 * make sure the block we are trying to log
4147			 * is not a block of journal or reserved area
4148			 */
4149			if (is_block_in_log_or_reserved_area
4150			    (sb, cn->bh->b_blocknr)) {
4151				reiserfs_panic(sb, "journal-2332",
4152					       "Trying to log block %lu, "
4153					       "which is a log block",
4154					       cn->bh->b_blocknr);
4155			}
4156			jl_cn->blocknr = cn->bh->b_blocknr;
4157			jl_cn->state = 0;
4158			jl_cn->sb = sb;
4159			jl_cn->bh = cn->bh;
4160			jl_cn->jlist = jl;
4161			insert_journal_hash(journal->j_list_hash_table, jl_cn);
4162			if (i < trans_half) {
4163				desc->j_realblock[i] =
4164				    cpu_to_le32(cn->bh->b_blocknr);
4165			} else {
4166				commit->j_realblock[i - trans_half] =
4167				    cpu_to_le32(cn->bh->b_blocknr);
4168			}
4169		} else {
4170			i--;
4171		}
4172	}
4173	set_desc_trans_len(desc, journal->j_len);
4174	set_desc_mount_id(desc, journal->j_mount_id);
4175	set_desc_trans_id(desc, journal->j_trans_id);
4176	set_commit_trans_len(commit, journal->j_len);
4177
4178	/*
4179	 * special check in case all buffers in the journal
4180	 * were marked for not logging
4181	 */
4182	BUG_ON(journal->j_len == 0);
4183
4184	/*
4185	 * we're about to dirty all the log blocks, mark the description block
4186	 * dirty now too.  Don't mark the commit block dirty until all the
4187	 * others are on disk
4188	 */
4189	mark_buffer_dirty(d_bh);
4190
4191	/*
4192	 * first data block is j_start + 1, so add one to
4193	 * cur_write_start wherever you use it
4194	 */
4195	cur_write_start = journal->j_start;
4196	cn = journal->j_first;
4197	jindex = 1;	/* start at one so we don't get the desc again */
4198	while (cn) {
4199		clear_buffer_journal_new(cn->bh);
4200		/* copy all the real blocks into log area.  dirty log blocks */
4201		if (buffer_journaled(cn->bh)) {
4202			struct buffer_head *tmp_bh;
4203			char *addr;
4204			struct page *page;
4205			tmp_bh =
4206			    journal_getblk(sb,
4207					   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4208					   ((cur_write_start +
4209					     jindex) %
4210					    SB_ONDISK_JOURNAL_SIZE(sb)));
4211			set_buffer_uptodate(tmp_bh);
4212			page = cn->bh->b_page;
4213			addr = kmap(page);
4214			memcpy(tmp_bh->b_data,
4215			       addr + offset_in_page(cn->bh->b_data),
4216			       cn->bh->b_size);
4217			kunmap(page);
4218			mark_buffer_dirty(tmp_bh);
4219			jindex++;
4220			set_buffer_journal_dirty(cn->bh);
4221			clear_buffer_journaled(cn->bh);
4222		} else {
4223			/*
4224			 * JDirty cleared sometime during transaction.
4225			 * don't log this one
4226			 */
4227			reiserfs_warning(sb, "journal-2048",
4228					 "BAD, buffer in journal hash, "
4229					 "but not JDirty!");
4230			brelse(cn->bh);
4231		}
4232		next = cn->next;
4233		free_cnode(sb, cn);
4234		cn = next;
4235		reiserfs_cond_resched(sb);
4236	}
4237
4238	/*
4239	 * we are done with both the c_bh and d_bh, but
4240	 * c_bh must be written after all other commit blocks,
4241	 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4242	 */
4243
4244	journal->j_current_jl = alloc_journal_list(sb);
4245
4246	/* now it is safe to insert this transaction on the main list */
4247	list_add_tail(&jl->j_list, &journal->j_journal_list);
4248	list_add_tail(&jl->j_working_list, &journal->j_working_list);
4249	journal->j_num_work_lists++;
4250
4251	/* reset journal values for the next transaction */
4252	journal->j_start =
4253	    (journal->j_start + journal->j_len +
4254	     2) % SB_ONDISK_JOURNAL_SIZE(sb);
4255	atomic_set(&journal->j_wcount, 0);
4256	journal->j_bcount = 0;
4257	journal->j_last = NULL;
4258	journal->j_first = NULL;
4259	journal->j_len = 0;
4260	journal->j_trans_start_time = 0;
4261	/* check for trans_id overflow */
4262	if (++journal->j_trans_id == 0)
4263		journal->j_trans_id = 10;
4264	journal->j_current_jl->j_trans_id = journal->j_trans_id;
4265	journal->j_must_wait = 0;
4266	journal->j_len_alloc = 0;
4267	journal->j_next_full_flush = 0;
4268	journal->j_next_async_flush = 0;
4269	init_journal_hash(sb);
4270
4271	/*
4272	 * make sure reiserfs_add_jh sees the new current_jl before we
4273	 * write out the tails
4274	 */
4275	smp_mb();
4276
4277	/*
4278	 * tail conversion targets have to hit the disk before we end the
4279	 * transaction.  Otherwise a later transaction might repack the tail
4280	 * before this transaction commits, leaving the data block unflushed
4281	 * and clean, if we crash before the later transaction commits, the
4282	 * data block is lost.
4283	 */
4284	if (!list_empty(&jl->j_tail_bh_list)) {
4285		depth = reiserfs_write_unlock_nested(sb);
4286		write_ordered_buffers(&journal->j_dirty_buffers_lock,
4287				      journal, jl, &jl->j_tail_bh_list);
4288		reiserfs_write_lock_nested(sb, depth);
4289	}
4290	BUG_ON(!list_empty(&jl->j_tail_bh_list));
4291	mutex_unlock(&jl->j_commit_mutex);
4292
4293	/*
4294	 * honor the flush wishes from the caller, simple commits can
4295	 * be done outside the journal lock, they are done below
4296	 *
4297	 * if we don't flush the commit list right now, we put it into
4298	 * the work queue so the people waiting on the async progress work
4299	 * queue don't wait for this proc to flush journal lists and such.
4300	 */
4301	if (flush) {
4302		flush_commit_list(sb, jl, 1);
4303		flush_journal_list(sb, jl, 1);
4304	} else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4305		/*
4306		 * Avoid queueing work when sb is being shut down. Transaction
4307		 * will be flushed on journal shutdown.
4308		 */
4309		if (sb->s_flags & SB_ACTIVE)
4310			queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4311					   &journal->j_work, HZ / 10);
4312	}
4313
4314	/*
4315	 * if the next transaction has any chance of wrapping, flush
4316	 * transactions that might get overwritten.  If any journal lists
4317	 * are very old flush them as well.
4318	 */
4319first_jl:
4320	list_for_each_safe(entry, safe, &journal->j_journal_list) {
4321		temp_jl = JOURNAL_LIST_ENTRY(entry);
4322		if (journal->j_start <= temp_jl->j_start) {
4323			if ((journal->j_start + journal->j_trans_max + 1) >=
4324			    temp_jl->j_start) {
4325				flush_used_journal_lists(sb, temp_jl);
4326				goto first_jl;
4327			} else if ((journal->j_start +
4328				    journal->j_trans_max + 1) <
4329				   SB_ONDISK_JOURNAL_SIZE(sb)) {
4330				/*
4331				 * if we don't cross into the next
4332				 * transaction and we don't wrap, there is
4333				 * no way we can overlap any later transactions
4334				 * break now
4335				 */
4336				break;
4337			}
4338		} else if ((journal->j_start +
4339			    journal->j_trans_max + 1) >
4340			   SB_ONDISK_JOURNAL_SIZE(sb)) {
4341			if (((journal->j_start + journal->j_trans_max + 1) %
4342			     SB_ONDISK_JOURNAL_SIZE(sb)) >=
4343			    temp_jl->j_start) {
4344				flush_used_journal_lists(sb, temp_jl);
4345				goto first_jl;
4346			} else {
4347				/*
4348				* we don't overlap anything from out start
4349				* to the end of the log, and our wrapped
4350				* portion doesn't overlap anything at
4351				* the start of the log.  We can break
4352				*/
4353				break;
4354			}
4355		}
4356	}
4357
4358	journal->j_current_jl->j_list_bitmap =
4359	    get_list_bitmap(sb, journal->j_current_jl);
4360
4361	if (!(journal->j_current_jl->j_list_bitmap)) {
4362		reiserfs_panic(sb, "journal-1996",
4363			       "could not get a list bitmap");
4364	}
4365
4366	atomic_set(&journal->j_jlock, 0);
4367	unlock_journal(sb);
4368	/* wake up any body waiting to join. */
4369	clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4370	wake_up(&journal->j_join_wait);
4371
4372	if (!flush && wait_on_commit &&
4373	    journal_list_still_alive(sb, commit_trans_id)) {
4374		flush_commit_list(sb, jl, 1);
4375	}
4376out:
4377	reiserfs_check_lock_depth(sb, "journal end2");
4378
4379	memset(th, 0, sizeof(*th));
4380	/*
4381	 * Re-set th->t_super, so we can properly keep track of how many
4382	 * persistent transactions there are. We need to do this so if this
4383	 * call is part of a failed restart_transaction, we can free it later
4384	 */
4385	th->t_super = sb;
4386
4387	return journal->j_errno;
4388}
4389
4390/* Send the file system read only and refuse new transactions */
4391void reiserfs_abort_journal(struct super_block *sb, int errno)
4392{
4393	struct reiserfs_journal *journal = SB_JOURNAL(sb);
4394	if (test_bit(J_ABORTED, &journal->j_state))
4395		return;
4396
4397	if (!journal->j_errno)
4398		journal->j_errno = errno;
4399
4400	sb->s_flags |= SB_RDONLY;
4401	set_bit(J_ABORTED, &journal->j_state);
4402
4403#ifdef CONFIG_REISERFS_CHECK
4404	dump_stack();
4405#endif
4406}
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Write ahead logging implementation copyright Chris Mason 2000
   4 *
   5 * The background commits make this code very interrelated, and
   6 * overly complex.  I need to rethink things a bit....The major players:
   7 *
   8 * journal_begin -- call with the number of blocks you expect to log.
   9 *                  If the current transaction is too
  10 *		    old, it will block until the current transaction is
  11 *		    finished, and then start a new one.
  12 *		    Usually, your transaction will get joined in with
  13 *                  previous ones for speed.
  14 *
  15 * journal_join  -- same as journal_begin, but won't block on the current
  16 *                  transaction regardless of age.  Don't ever call
  17 *                  this.  Ever.  There are only two places it should be
  18 *                  called from, and they are both inside this file.
  19 *
  20 * journal_mark_dirty -- adds blocks into this transaction.  clears any flags
  21 *                       that might make them get sent to disk
  22 *                       and then marks them BH_JDirty.  Puts the buffer head
  23 *                       into the current transaction hash.
  24 *
  25 * journal_end -- if the current transaction is batchable, it does nothing
  26 *                   otherwise, it could do an async/synchronous commit, or
  27 *                   a full flush of all log and real blocks in the
  28 *                   transaction.
  29 *
  30 * flush_old_commits -- if the current transaction is too old, it is ended and
  31 *                      commit blocks are sent to disk.  Forces commit blocks
  32 *                      to disk for all backgrounded commits that have been
  33 *                      around too long.
  34 *		     -- Note, if you call this as an immediate flush from
  35 *		        from within kupdate, it will ignore the immediate flag
  36 */
  37
  38#include <linux/time.h>
  39#include <linux/semaphore.h>
  40#include <linux/vmalloc.h>
  41#include "reiserfs.h"
  42#include <linux/kernel.h>
  43#include <linux/errno.h>
  44#include <linux/fcntl.h>
  45#include <linux/stat.h>
  46#include <linux/string.h>
  47#include <linux/buffer_head.h>
  48#include <linux/workqueue.h>
  49#include <linux/writeback.h>
  50#include <linux/blkdev.h>
  51#include <linux/backing-dev.h>
  52#include <linux/uaccess.h>
  53#include <linux/slab.h>
  54
  55
  56/* gets a struct reiserfs_journal_list * from a list head */
  57#define JOURNAL_LIST_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  58                               j_list))
  59#define JOURNAL_WORK_ENTRY(h) (list_entry((h), struct reiserfs_journal_list, \
  60                               j_working_list))
  61
  62/* must be correct to keep the desc and commit structs at 4k */
  63#define JOURNAL_TRANS_HALF 1018
  64#define BUFNR 64		/*read ahead */
  65
  66/* cnode stat bits.  Move these into reiserfs_fs.h */
  67
  68/* this block was freed, and can't be written.  */
  69#define BLOCK_FREED 2
  70/* this block was freed during this transaction, and can't be written */
  71#define BLOCK_FREED_HOLDER 3
  72
  73/* used in flush_journal_list */
  74#define BLOCK_NEEDS_FLUSH 4
  75#define BLOCK_DIRTIED 5
  76
  77/* journal list state bits */
  78#define LIST_TOUCHED 1
  79#define LIST_DIRTY   2
  80#define LIST_COMMIT_PENDING  4	/* someone will commit this list */
  81
  82/* flags for do_journal_end */
  83#define FLUSH_ALL   1		/* flush commit and real blocks */
  84#define COMMIT_NOW  2		/* end and commit this transaction */
  85#define WAIT        4		/* wait for the log blocks to hit the disk */
  86
  87static int do_journal_end(struct reiserfs_transaction_handle *, int flags);
  88static int flush_journal_list(struct super_block *s,
  89			      struct reiserfs_journal_list *jl, int flushall);
  90static int flush_commit_list(struct super_block *s,
  91			     struct reiserfs_journal_list *jl, int flushall);
  92static int can_dirty(struct reiserfs_journal_cnode *cn);
  93static int journal_join(struct reiserfs_transaction_handle *th,
  94			struct super_block *sb);
  95static void release_journal_dev(struct super_block *super,
  96			       struct reiserfs_journal *journal);
  97static void dirty_one_transaction(struct super_block *s,
  98				 struct reiserfs_journal_list *jl);
  99static void flush_async_commits(struct work_struct *work);
 100static void queue_log_writer(struct super_block *s);
 101
 102/* values for join in do_journal_begin_r */
 103enum {
 104	JBEGIN_REG = 0,		/* regular journal begin */
 105	/* join the running transaction if at all possible */
 106	JBEGIN_JOIN = 1,
 107	/* called from cleanup code, ignores aborted flag */
 108	JBEGIN_ABORT = 2,
 109};
 110
 111static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
 112			      struct super_block *sb,
 113			      unsigned long nblocks, int join);
 114
 115static void init_journal_hash(struct super_block *sb)
 116{
 117	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 118	memset(journal->j_hash_table, 0,
 119	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
 120}
 121
 122/*
 123 * clears BH_Dirty and sticks the buffer on the clean list.  Called because
 124 * I can't allow refile_buffer to make schedule happen after I've freed a
 125 * block.  Look at remove_from_transaction and journal_mark_freed for
 126 * more details.
 127 */
 128static int reiserfs_clean_and_file_buffer(struct buffer_head *bh)
 129{
 130	if (bh) {
 131		clear_buffer_dirty(bh);
 132		clear_buffer_journal_test(bh);
 133	}
 134	return 0;
 135}
 136
 137static struct reiserfs_bitmap_node *allocate_bitmap_node(struct super_block
 138							 *sb)
 139{
 140	struct reiserfs_bitmap_node *bn;
 141	static int id;
 142
 143	bn = kmalloc(sizeof(struct reiserfs_bitmap_node), GFP_NOFS);
 144	if (!bn) {
 145		return NULL;
 146	}
 147	bn->data = kzalloc(sb->s_blocksize, GFP_NOFS);
 148	if (!bn->data) {
 149		kfree(bn);
 150		return NULL;
 151	}
 152	bn->id = id++;
 153	INIT_LIST_HEAD(&bn->list);
 154	return bn;
 155}
 156
 157static struct reiserfs_bitmap_node *get_bitmap_node(struct super_block *sb)
 158{
 159	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 160	struct reiserfs_bitmap_node *bn = NULL;
 161	struct list_head *entry = journal->j_bitmap_nodes.next;
 162
 163	journal->j_used_bitmap_nodes++;
 164repeat:
 165
 166	if (entry != &journal->j_bitmap_nodes) {
 167		bn = list_entry(entry, struct reiserfs_bitmap_node, list);
 168		list_del(entry);
 169		memset(bn->data, 0, sb->s_blocksize);
 170		journal->j_free_bitmap_nodes--;
 171		return bn;
 172	}
 173	bn = allocate_bitmap_node(sb);
 174	if (!bn) {
 175		yield();
 176		goto repeat;
 177	}
 178	return bn;
 179}
 180static inline void free_bitmap_node(struct super_block *sb,
 181				    struct reiserfs_bitmap_node *bn)
 182{
 183	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 184	journal->j_used_bitmap_nodes--;
 185	if (journal->j_free_bitmap_nodes > REISERFS_MAX_BITMAP_NODES) {
 186		kfree(bn->data);
 187		kfree(bn);
 188	} else {
 189		list_add(&bn->list, &journal->j_bitmap_nodes);
 190		journal->j_free_bitmap_nodes++;
 191	}
 192}
 193
 194static void allocate_bitmap_nodes(struct super_block *sb)
 195{
 196	int i;
 197	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 198	struct reiserfs_bitmap_node *bn = NULL;
 199	for (i = 0; i < REISERFS_MIN_BITMAP_NODES; i++) {
 200		bn = allocate_bitmap_node(sb);
 201		if (bn) {
 202			list_add(&bn->list, &journal->j_bitmap_nodes);
 203			journal->j_free_bitmap_nodes++;
 204		} else {
 205			/* this is ok, we'll try again when more are needed */
 206			break;
 207		}
 208	}
 209}
 210
 211static int set_bit_in_list_bitmap(struct super_block *sb,
 212				  b_blocknr_t block,
 213				  struct reiserfs_list_bitmap *jb)
 214{
 215	unsigned int bmap_nr = block / (sb->s_blocksize << 3);
 216	unsigned int bit_nr = block % (sb->s_blocksize << 3);
 217
 218	if (!jb->bitmaps[bmap_nr]) {
 219		jb->bitmaps[bmap_nr] = get_bitmap_node(sb);
 220	}
 221	set_bit(bit_nr, (unsigned long *)jb->bitmaps[bmap_nr]->data);
 222	return 0;
 223}
 224
 225static void cleanup_bitmap_list(struct super_block *sb,
 226				struct reiserfs_list_bitmap *jb)
 227{
 228	int i;
 229	if (jb->bitmaps == NULL)
 230		return;
 231
 232	for (i = 0; i < reiserfs_bmap_count(sb); i++) {
 233		if (jb->bitmaps[i]) {
 234			free_bitmap_node(sb, jb->bitmaps[i]);
 235			jb->bitmaps[i] = NULL;
 236		}
 237	}
 238}
 239
 240/*
 241 * only call this on FS unmount.
 242 */
 243static int free_list_bitmaps(struct super_block *sb,
 244			     struct reiserfs_list_bitmap *jb_array)
 245{
 246	int i;
 247	struct reiserfs_list_bitmap *jb;
 248	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 249		jb = jb_array + i;
 250		jb->journal_list = NULL;
 251		cleanup_bitmap_list(sb, jb);
 252		vfree(jb->bitmaps);
 253		jb->bitmaps = NULL;
 254	}
 255	return 0;
 256}
 257
 258static int free_bitmap_nodes(struct super_block *sb)
 259{
 260	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 261	struct list_head *next = journal->j_bitmap_nodes.next;
 262	struct reiserfs_bitmap_node *bn;
 263
 264	while (next != &journal->j_bitmap_nodes) {
 265		bn = list_entry(next, struct reiserfs_bitmap_node, list);
 266		list_del(next);
 267		kfree(bn->data);
 268		kfree(bn);
 269		next = journal->j_bitmap_nodes.next;
 270		journal->j_free_bitmap_nodes--;
 271	}
 272
 273	return 0;
 274}
 275
 276/*
 277 * get memory for JOURNAL_NUM_BITMAPS worth of bitmaps.
 278 * jb_array is the array to be filled in.
 279 */
 280int reiserfs_allocate_list_bitmaps(struct super_block *sb,
 281				   struct reiserfs_list_bitmap *jb_array,
 282				   unsigned int bmap_nr)
 283{
 284	int i;
 285	int failed = 0;
 286	struct reiserfs_list_bitmap *jb;
 287	int mem = bmap_nr * sizeof(struct reiserfs_bitmap_node *);
 288
 289	for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 290		jb = jb_array + i;
 291		jb->journal_list = NULL;
 292		jb->bitmaps = vzalloc(mem);
 293		if (!jb->bitmaps) {
 294			reiserfs_warning(sb, "clm-2000", "unable to "
 295					 "allocate bitmaps for journal lists");
 296			failed = 1;
 297			break;
 298		}
 299	}
 300	if (failed) {
 301		free_list_bitmaps(sb, jb_array);
 302		return -1;
 303	}
 304	return 0;
 305}
 306
 307/*
 308 * find an available list bitmap.  If you can't find one, flush a commit list
 309 * and try again
 310 */
 311static struct reiserfs_list_bitmap *get_list_bitmap(struct super_block *sb,
 312						    struct reiserfs_journal_list
 313						    *jl)
 314{
 315	int i, j;
 316	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 317	struct reiserfs_list_bitmap *jb = NULL;
 318
 319	for (j = 0; j < (JOURNAL_NUM_BITMAPS * 3); j++) {
 320		i = journal->j_list_bitmap_index;
 321		journal->j_list_bitmap_index = (i + 1) % JOURNAL_NUM_BITMAPS;
 322		jb = journal->j_list_bitmap + i;
 323		if (journal->j_list_bitmap[i].journal_list) {
 324			flush_commit_list(sb,
 325					  journal->j_list_bitmap[i].
 326					  journal_list, 1);
 327			if (!journal->j_list_bitmap[i].journal_list) {
 328				break;
 329			}
 330		} else {
 331			break;
 332		}
 333	}
 334	/* double check to make sure if flushed correctly */
 335	if (jb->journal_list)
 336		return NULL;
 337	jb->journal_list = jl;
 338	return jb;
 339}
 340
 341/*
 342 * allocates a new chunk of X nodes, and links them all together as a list.
 343 * Uses the cnode->next and cnode->prev pointers
 344 * returns NULL on failure
 345 */
 346static struct reiserfs_journal_cnode *allocate_cnodes(int num_cnodes)
 347{
 348	struct reiserfs_journal_cnode *head;
 349	int i;
 350	if (num_cnodes <= 0) {
 351		return NULL;
 352	}
 353	head = vzalloc(array_size(num_cnodes,
 354				  sizeof(struct reiserfs_journal_cnode)));
 355	if (!head) {
 356		return NULL;
 357	}
 358	head[0].prev = NULL;
 359	head[0].next = head + 1;
 360	for (i = 1; i < num_cnodes; i++) {
 361		head[i].prev = head + (i - 1);
 362		head[i].next = head + (i + 1);	/* if last one, overwrite it after the if */
 363	}
 364	head[num_cnodes - 1].next = NULL;
 365	return head;
 366}
 367
 368/* pulls a cnode off the free list, or returns NULL on failure */
 369static struct reiserfs_journal_cnode *get_cnode(struct super_block *sb)
 370{
 371	struct reiserfs_journal_cnode *cn;
 372	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 373
 374	reiserfs_check_lock_depth(sb, "get_cnode");
 375
 376	if (journal->j_cnode_free <= 0) {
 377		return NULL;
 378	}
 379	journal->j_cnode_used++;
 380	journal->j_cnode_free--;
 381	cn = journal->j_cnode_free_list;
 382	if (!cn) {
 383		return cn;
 384	}
 385	if (cn->next) {
 386		cn->next->prev = NULL;
 387	}
 388	journal->j_cnode_free_list = cn->next;
 389	memset(cn, 0, sizeof(struct reiserfs_journal_cnode));
 390	return cn;
 391}
 392
 393/*
 394 * returns a cnode to the free list
 395 */
 396static void free_cnode(struct super_block *sb,
 397		       struct reiserfs_journal_cnode *cn)
 398{
 399	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 400
 401	reiserfs_check_lock_depth(sb, "free_cnode");
 402
 403	journal->j_cnode_used--;
 404	journal->j_cnode_free++;
 405	/* memset(cn, 0, sizeof(struct reiserfs_journal_cnode)) ; */
 406	cn->next = journal->j_cnode_free_list;
 407	if (journal->j_cnode_free_list) {
 408		journal->j_cnode_free_list->prev = cn;
 409	}
 410	cn->prev = NULL;	/* not needed with the memset, but I might kill the memset, and forget to do this */
 411	journal->j_cnode_free_list = cn;
 412}
 413
 414static void clear_prepared_bits(struct buffer_head *bh)
 415{
 416	clear_buffer_journal_prepared(bh);
 417	clear_buffer_journal_restore_dirty(bh);
 418}
 419
 420/*
 421 * return a cnode with same dev, block number and size in table,
 422 * or null if not found
 423 */
 424static inline struct reiserfs_journal_cnode *get_journal_hash_dev(struct
 425								  super_block
 426								  *sb,
 427								  struct
 428								  reiserfs_journal_cnode
 429								  **table,
 430								  long bl)
 431{
 432	struct reiserfs_journal_cnode *cn;
 433	cn = journal_hash(table, sb, bl);
 434	while (cn) {
 435		if (cn->blocknr == bl && cn->sb == sb)
 436			return cn;
 437		cn = cn->hnext;
 438	}
 439	return (struct reiserfs_journal_cnode *)0;
 440}
 441
 442/*
 443 * this actually means 'can this block be reallocated yet?'.  If you set
 444 * search_all, a block can only be allocated if it is not in the current
 445 * transaction, was not freed by the current transaction, and has no chance
 446 * of ever being overwritten by a replay after crashing.
 447 *
 448 * If you don't set search_all, a block can only be allocated if it is not
 449 * in the current transaction.  Since deleting a block removes it from the
 450 * current transaction, this case should never happen.  If you don't set
 451 * search_all, make sure you never write the block without logging it.
 452 *
 453 * next_zero_bit is a suggestion about the next block to try for find_forward.
 454 * when bl is rejected because it is set in a journal list bitmap, we search
 455 * for the next zero bit in the bitmap that rejected bl.  Then, we return
 456 * that through next_zero_bit for find_forward to try.
 457 *
 458 * Just because we return something in next_zero_bit does not mean we won't
 459 * reject it on the next call to reiserfs_in_journal
 460 */
 461int reiserfs_in_journal(struct super_block *sb,
 462			unsigned int bmap_nr, int bit_nr, int search_all,
 463			b_blocknr_t * next_zero_bit)
 464{
 465	struct reiserfs_journal *journal = SB_JOURNAL(sb);
 466	struct reiserfs_journal_cnode *cn;
 467	struct reiserfs_list_bitmap *jb;
 468	int i;
 469	unsigned long bl;
 470
 471	*next_zero_bit = 0;	/* always start this at zero. */
 472
 473	PROC_INFO_INC(sb, journal.in_journal);
 474	/*
 475	 * If we aren't doing a search_all, this is a metablock, and it
 476	 * will be logged before use.  if we crash before the transaction
 477	 * that freed it commits,  this transaction won't have committed
 478	 * either, and the block will never be written
 479	 */
 480	if (search_all) {
 481		for (i = 0; i < JOURNAL_NUM_BITMAPS; i++) {
 482			PROC_INFO_INC(sb, journal.in_journal_bitmap);
 483			jb = journal->j_list_bitmap + i;
 484			if (jb->journal_list && jb->bitmaps[bmap_nr] &&
 485			    test_bit(bit_nr,
 486				     (unsigned long *)jb->bitmaps[bmap_nr]->
 487				     data)) {
 488				*next_zero_bit =
 489				    find_next_zero_bit((unsigned long *)
 490						       (jb->bitmaps[bmap_nr]->
 491							data),
 492						       sb->s_blocksize << 3,
 493						       bit_nr + 1);
 494				return 1;
 495			}
 496		}
 497	}
 498
 499	bl = bmap_nr * (sb->s_blocksize << 3) + bit_nr;
 500	/* is it in any old transactions? */
 501	if (search_all
 502	    && (cn =
 503		get_journal_hash_dev(sb, journal->j_list_hash_table, bl))) {
 504		return 1;
 505	}
 506
 507	/* is it in the current transaction.  This should never happen */
 508	if ((cn = get_journal_hash_dev(sb, journal->j_hash_table, bl))) {
 509		BUG();
 510		return 1;
 511	}
 512
 513	PROC_INFO_INC(sb, journal.in_journal_reusable);
 514	/* safe for reuse */
 515	return 0;
 516}
 517
 518/* insert cn into table */
 519static inline void insert_journal_hash(struct reiserfs_journal_cnode **table,
 520				       struct reiserfs_journal_cnode *cn)
 521{
 522	struct reiserfs_journal_cnode *cn_orig;
 523
 524	cn_orig = journal_hash(table, cn->sb, cn->blocknr);
 525	cn->hnext = cn_orig;
 526	cn->hprev = NULL;
 527	if (cn_orig) {
 528		cn_orig->hprev = cn;
 529	}
 530	journal_hash(table, cn->sb, cn->blocknr) = cn;
 531}
 532
 533/* lock the current transaction */
 534static inline void lock_journal(struct super_block *sb)
 535{
 536	PROC_INFO_INC(sb, journal.lock_journal);
 537
 538	reiserfs_mutex_lock_safe(&SB_JOURNAL(sb)->j_mutex, sb);
 539}
 540
 541/* unlock the current transaction */
 542static inline void unlock_journal(struct super_block *sb)
 543{
 544	mutex_unlock(&SB_JOURNAL(sb)->j_mutex);
 545}
 546
 547static inline void get_journal_list(struct reiserfs_journal_list *jl)
 548{
 549	jl->j_refcount++;
 550}
 551
 552static inline void put_journal_list(struct super_block *s,
 553				    struct reiserfs_journal_list *jl)
 554{
 555	if (jl->j_refcount < 1) {
 556		reiserfs_panic(s, "journal-2", "trans id %u, refcount at %d",
 557			       jl->j_trans_id, jl->j_refcount);
 558	}
 559	if (--jl->j_refcount == 0)
 560		kfree(jl);
 561}
 562
 563/*
 564 * this used to be much more involved, and I'm keeping it just in case
 565 * things get ugly again.  it gets called by flush_commit_list, and
 566 * cleans up any data stored about blocks freed during a transaction.
 567 */
 568static void cleanup_freed_for_journal_list(struct super_block *sb,
 569					   struct reiserfs_journal_list *jl)
 570{
 571
 572	struct reiserfs_list_bitmap *jb = jl->j_list_bitmap;
 573	if (jb) {
 574		cleanup_bitmap_list(sb, jb);
 575	}
 576	jl->j_list_bitmap->journal_list = NULL;
 577	jl->j_list_bitmap = NULL;
 578}
 579
 580static int journal_list_still_alive(struct super_block *s,
 581				    unsigned int trans_id)
 582{
 583	struct reiserfs_journal *journal = SB_JOURNAL(s);
 584	struct list_head *entry = &journal->j_journal_list;
 585	struct reiserfs_journal_list *jl;
 586
 587	if (!list_empty(entry)) {
 588		jl = JOURNAL_LIST_ENTRY(entry->next);
 589		if (jl->j_trans_id <= trans_id) {
 590			return 1;
 591		}
 592	}
 593	return 0;
 594}
 595
 596/*
 597 * If page->mapping was null, we failed to truncate this page for
 598 * some reason.  Most likely because it was truncated after being
 599 * logged via data=journal.
 600 *
 601 * This does a check to see if the buffer belongs to one of these
 602 * lost pages before doing the final put_bh.  If page->mapping was
 603 * null, it tries to free buffers on the page, which should make the
 604 * final put_page drop the page from the lru.
 605 */
 606static void release_buffer_page(struct buffer_head *bh)
 607{
 608	struct page *page = bh->b_page;
 609	if (!page->mapping && trylock_page(page)) {
 610		get_page(page);
 611		put_bh(bh);
 612		if (!page->mapping)
 613			try_to_free_buffers(page);
 614		unlock_page(page);
 615		put_page(page);
 616	} else {
 617		put_bh(bh);
 618	}
 619}
 620
 621static void reiserfs_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 622{
 623	if (buffer_journaled(bh)) {
 624		reiserfs_warning(NULL, "clm-2084",
 625				 "pinned buffer %lu:%pg sent to disk",
 626				 bh->b_blocknr, bh->b_bdev);
 627	}
 628	if (uptodate)
 629		set_buffer_uptodate(bh);
 630	else
 631		clear_buffer_uptodate(bh);
 632
 633	unlock_buffer(bh);
 634	release_buffer_page(bh);
 635}
 636
 637static void reiserfs_end_ordered_io(struct buffer_head *bh, int uptodate)
 638{
 639	if (uptodate)
 640		set_buffer_uptodate(bh);
 641	else
 642		clear_buffer_uptodate(bh);
 643	unlock_buffer(bh);
 644	put_bh(bh);
 645}
 646
 647static void submit_logged_buffer(struct buffer_head *bh)
 648{
 649	get_bh(bh);
 650	bh->b_end_io = reiserfs_end_buffer_io_sync;
 651	clear_buffer_journal_new(bh);
 652	clear_buffer_dirty(bh);
 653	if (!test_clear_buffer_journal_test(bh))
 654		BUG();
 655	if (!buffer_uptodate(bh))
 656		BUG();
 657	submit_bh(REQ_OP_WRITE, 0, bh);
 658}
 659
 660static void submit_ordered_buffer(struct buffer_head *bh)
 661{
 662	get_bh(bh);
 663	bh->b_end_io = reiserfs_end_ordered_io;
 664	clear_buffer_dirty(bh);
 665	if (!buffer_uptodate(bh))
 666		BUG();
 667	submit_bh(REQ_OP_WRITE, 0, bh);
 668}
 669
 670#define CHUNK_SIZE 32
 671struct buffer_chunk {
 672	struct buffer_head *bh[CHUNK_SIZE];
 673	int nr;
 674};
 675
 676static void write_chunk(struct buffer_chunk *chunk)
 677{
 678	int i;
 679	for (i = 0; i < chunk->nr; i++) {
 680		submit_logged_buffer(chunk->bh[i]);
 681	}
 682	chunk->nr = 0;
 683}
 684
 685static void write_ordered_chunk(struct buffer_chunk *chunk)
 686{
 687	int i;
 688	for (i = 0; i < chunk->nr; i++) {
 689		submit_ordered_buffer(chunk->bh[i]);
 690	}
 691	chunk->nr = 0;
 692}
 693
 694static int add_to_chunk(struct buffer_chunk *chunk, struct buffer_head *bh,
 695			spinlock_t * lock, void (fn) (struct buffer_chunk *))
 696{
 697	int ret = 0;
 698	BUG_ON(chunk->nr >= CHUNK_SIZE);
 699	chunk->bh[chunk->nr++] = bh;
 700	if (chunk->nr >= CHUNK_SIZE) {
 701		ret = 1;
 702		if (lock) {
 703			spin_unlock(lock);
 704			fn(chunk);
 705			spin_lock(lock);
 706		} else {
 707			fn(chunk);
 708		}
 709	}
 710	return ret;
 711}
 712
 713static atomic_t nr_reiserfs_jh = ATOMIC_INIT(0);
 714static struct reiserfs_jh *alloc_jh(void)
 715{
 716	struct reiserfs_jh *jh;
 717	while (1) {
 718		jh = kmalloc(sizeof(*jh), GFP_NOFS);
 719		if (jh) {
 720			atomic_inc(&nr_reiserfs_jh);
 721			return jh;
 722		}
 723		yield();
 724	}
 725}
 726
 727/*
 728 * we want to free the jh when the buffer has been written
 729 * and waited on
 730 */
 731void reiserfs_free_jh(struct buffer_head *bh)
 732{
 733	struct reiserfs_jh *jh;
 734
 735	jh = bh->b_private;
 736	if (jh) {
 737		bh->b_private = NULL;
 738		jh->bh = NULL;
 739		list_del_init(&jh->list);
 740		kfree(jh);
 741		if (atomic_read(&nr_reiserfs_jh) <= 0)
 742			BUG();
 743		atomic_dec(&nr_reiserfs_jh);
 744		put_bh(bh);
 745	}
 746}
 747
 748static inline int __add_jh(struct reiserfs_journal *j, struct buffer_head *bh,
 749			   int tail)
 750{
 751	struct reiserfs_jh *jh;
 752
 753	if (bh->b_private) {
 754		spin_lock(&j->j_dirty_buffers_lock);
 755		if (!bh->b_private) {
 756			spin_unlock(&j->j_dirty_buffers_lock);
 757			goto no_jh;
 758		}
 759		jh = bh->b_private;
 760		list_del_init(&jh->list);
 761	} else {
 762no_jh:
 763		get_bh(bh);
 764		jh = alloc_jh();
 765		spin_lock(&j->j_dirty_buffers_lock);
 766		/*
 767		 * buffer must be locked for __add_jh, should be able to have
 768		 * two adds at the same time
 769		 */
 770		BUG_ON(bh->b_private);
 771		jh->bh = bh;
 772		bh->b_private = jh;
 773	}
 774	jh->jl = j->j_current_jl;
 775	if (tail)
 776		list_add_tail(&jh->list, &jh->jl->j_tail_bh_list);
 777	else {
 778		list_add_tail(&jh->list, &jh->jl->j_bh_list);
 779	}
 780	spin_unlock(&j->j_dirty_buffers_lock);
 781	return 0;
 782}
 783
 784int reiserfs_add_tail_list(struct inode *inode, struct buffer_head *bh)
 785{
 786	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 1);
 787}
 788int reiserfs_add_ordered_list(struct inode *inode, struct buffer_head *bh)
 789{
 790	return __add_jh(SB_JOURNAL(inode->i_sb), bh, 0);
 791}
 792
 793#define JH_ENTRY(l) list_entry((l), struct reiserfs_jh, list)
 794static int write_ordered_buffers(spinlock_t * lock,
 795				 struct reiserfs_journal *j,
 796				 struct reiserfs_journal_list *jl,
 797				 struct list_head *list)
 798{
 799	struct buffer_head *bh;
 800	struct reiserfs_jh *jh;
 801	int ret = j->j_errno;
 802	struct buffer_chunk chunk;
 803	struct list_head tmp;
 804	INIT_LIST_HEAD(&tmp);
 805
 806	chunk.nr = 0;
 807	spin_lock(lock);
 808	while (!list_empty(list)) {
 809		jh = JH_ENTRY(list->next);
 810		bh = jh->bh;
 811		get_bh(bh);
 812		if (!trylock_buffer(bh)) {
 813			if (!buffer_dirty(bh)) {
 814				list_move(&jh->list, &tmp);
 815				goto loop_next;
 816			}
 817			spin_unlock(lock);
 818			if (chunk.nr)
 819				write_ordered_chunk(&chunk);
 820			wait_on_buffer(bh);
 821			cond_resched();
 822			spin_lock(lock);
 823			goto loop_next;
 824		}
 825		/*
 826		 * in theory, dirty non-uptodate buffers should never get here,
 827		 * but the upper layer io error paths still have a few quirks.
 828		 * Handle them here as gracefully as we can
 829		 */
 830		if (!buffer_uptodate(bh) && buffer_dirty(bh)) {
 831			clear_buffer_dirty(bh);
 832			ret = -EIO;
 833		}
 834		if (buffer_dirty(bh)) {
 835			list_move(&jh->list, &tmp);
 836			add_to_chunk(&chunk, bh, lock, write_ordered_chunk);
 837		} else {
 838			reiserfs_free_jh(bh);
 839			unlock_buffer(bh);
 840		}
 841loop_next:
 842		put_bh(bh);
 843		cond_resched_lock(lock);
 844	}
 845	if (chunk.nr) {
 846		spin_unlock(lock);
 847		write_ordered_chunk(&chunk);
 848		spin_lock(lock);
 849	}
 850	while (!list_empty(&tmp)) {
 851		jh = JH_ENTRY(tmp.prev);
 852		bh = jh->bh;
 853		get_bh(bh);
 854		reiserfs_free_jh(bh);
 855
 856		if (buffer_locked(bh)) {
 857			spin_unlock(lock);
 858			wait_on_buffer(bh);
 859			spin_lock(lock);
 860		}
 861		if (!buffer_uptodate(bh)) {
 862			ret = -EIO;
 863		}
 864		/*
 865		 * ugly interaction with invalidatepage here.
 866		 * reiserfs_invalidate_page will pin any buffer that has a
 867		 * valid journal head from an older transaction.  If someone
 868		 * else sets our buffer dirty after we write it in the first
 869		 * loop, and then someone truncates the page away, nobody
 870		 * will ever write the buffer. We're safe if we write the
 871		 * page one last time after freeing the journal header.
 872		 */
 873		if (buffer_dirty(bh) && unlikely(bh->b_page->mapping == NULL)) {
 874			spin_unlock(lock);
 875			ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
 876			spin_lock(lock);
 877		}
 878		put_bh(bh);
 879		cond_resched_lock(lock);
 880	}
 881	spin_unlock(lock);
 882	return ret;
 883}
 884
 885static int flush_older_commits(struct super_block *s,
 886			       struct reiserfs_journal_list *jl)
 887{
 888	struct reiserfs_journal *journal = SB_JOURNAL(s);
 889	struct reiserfs_journal_list *other_jl;
 890	struct reiserfs_journal_list *first_jl;
 891	struct list_head *entry;
 892	unsigned int trans_id = jl->j_trans_id;
 893	unsigned int other_trans_id;
 894
 895find_first:
 896	/*
 897	 * first we walk backwards to find the oldest uncommitted transation
 898	 */
 899	first_jl = jl;
 900	entry = jl->j_list.prev;
 901	while (1) {
 902		other_jl = JOURNAL_LIST_ENTRY(entry);
 903		if (entry == &journal->j_journal_list ||
 904		    atomic_read(&other_jl->j_older_commits_done))
 905			break;
 906
 907		first_jl = other_jl;
 908		entry = other_jl->j_list.prev;
 909	}
 910
 911	/* if we didn't find any older uncommitted transactions, return now */
 912	if (first_jl == jl) {
 913		return 0;
 914	}
 915
 916	entry = &first_jl->j_list;
 917	while (1) {
 918		other_jl = JOURNAL_LIST_ENTRY(entry);
 919		other_trans_id = other_jl->j_trans_id;
 920
 921		if (other_trans_id < trans_id) {
 922			if (atomic_read(&other_jl->j_commit_left) != 0) {
 923				flush_commit_list(s, other_jl, 0);
 924
 925				/* list we were called with is gone, return */
 926				if (!journal_list_still_alive(s, trans_id))
 927					return 1;
 928
 929				/*
 930				 * the one we just flushed is gone, this means
 931				 * all older lists are also gone, so first_jl
 932				 * is no longer valid either.  Go back to the
 933				 * beginning.
 934				 */
 935				if (!journal_list_still_alive
 936				    (s, other_trans_id)) {
 937					goto find_first;
 938				}
 939			}
 940			entry = entry->next;
 941			if (entry == &journal->j_journal_list)
 942				return 0;
 943		} else {
 944			return 0;
 945		}
 946	}
 947	return 0;
 948}
 949
 950static int reiserfs_async_progress_wait(struct super_block *s)
 951{
 952	struct reiserfs_journal *j = SB_JOURNAL(s);
 953
 954	if (atomic_read(&j->j_async_throttle)) {
 955		int depth;
 956
 957		depth = reiserfs_write_unlock_nested(s);
 958		congestion_wait(BLK_RW_ASYNC, HZ / 10);
 
 
 959		reiserfs_write_lock_nested(s, depth);
 960	}
 961
 962	return 0;
 963}
 964
 965/*
 966 * if this journal list still has commit blocks unflushed, send them to disk.
 967 *
 968 * log areas must be flushed in order (transaction 2 can't commit before
 969 * transaction 1) Before the commit block can by written, every other log
 970 * block must be safely on disk
 971 */
 972static int flush_commit_list(struct super_block *s,
 973			     struct reiserfs_journal_list *jl, int flushall)
 974{
 975	int i;
 976	b_blocknr_t bn;
 977	struct buffer_head *tbh = NULL;
 978	unsigned int trans_id = jl->j_trans_id;
 979	struct reiserfs_journal *journal = SB_JOURNAL(s);
 980	int retval = 0;
 981	int write_len;
 982	int depth;
 983
 984	reiserfs_check_lock_depth(s, "flush_commit_list");
 985
 986	if (atomic_read(&jl->j_older_commits_done)) {
 987		return 0;
 988	}
 989
 990	/*
 991	 * before we can put our commit blocks on disk, we have to make
 992	 * sure everyone older than us is on disk too
 993	 */
 994	BUG_ON(jl->j_len <= 0);
 995	BUG_ON(trans_id == journal->j_trans_id);
 996
 997	get_journal_list(jl);
 998	if (flushall) {
 999		if (flush_older_commits(s, jl) == 1) {
1000			/*
1001			 * list disappeared during flush_older_commits.
1002			 * return
1003			 */
1004			goto put_jl;
1005		}
1006	}
1007
1008	/* make sure nobody is trying to flush this one at the same time */
1009	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, s);
1010
1011	if (!journal_list_still_alive(s, trans_id)) {
1012		mutex_unlock(&jl->j_commit_mutex);
1013		goto put_jl;
1014	}
1015	BUG_ON(jl->j_trans_id == 0);
1016
1017	/* this commit is done, exit */
1018	if (atomic_read(&jl->j_commit_left) <= 0) {
1019		if (flushall) {
1020			atomic_set(&jl->j_older_commits_done, 1);
1021		}
1022		mutex_unlock(&jl->j_commit_mutex);
1023		goto put_jl;
1024	}
1025
1026	if (!list_empty(&jl->j_bh_list)) {
1027		int ret;
1028
1029		/*
1030		 * We might sleep in numerous places inside
1031		 * write_ordered_buffers. Relax the write lock.
1032		 */
1033		depth = reiserfs_write_unlock_nested(s);
1034		ret = write_ordered_buffers(&journal->j_dirty_buffers_lock,
1035					    journal, jl, &jl->j_bh_list);
1036		if (ret < 0 && retval == 0)
1037			retval = ret;
1038		reiserfs_write_lock_nested(s, depth);
1039	}
1040	BUG_ON(!list_empty(&jl->j_bh_list));
1041	/*
1042	 * for the description block and all the log blocks, submit any buffers
1043	 * that haven't already reached the disk.  Try to write at least 256
1044	 * log blocks. later on, we will only wait on blocks that correspond
1045	 * to this transaction, but while we're unplugging we might as well
1046	 * get a chunk of data on there.
1047	 */
1048	atomic_inc(&journal->j_async_throttle);
1049	write_len = jl->j_len + 1;
1050	if (write_len < 256)
1051		write_len = 256;
1052	for (i = 0 ; i < write_len ; i++) {
1053		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) + (jl->j_start + i) %
1054		    SB_ONDISK_JOURNAL_SIZE(s);
1055		tbh = journal_find_get_block(s, bn);
1056		if (tbh) {
1057			if (buffer_dirty(tbh)) {
1058		            depth = reiserfs_write_unlock_nested(s);
1059			    ll_rw_block(REQ_OP_WRITE, 0, 1, &tbh);
1060			    reiserfs_write_lock_nested(s, depth);
1061			}
1062			put_bh(tbh) ;
1063		}
1064	}
1065	atomic_dec(&journal->j_async_throttle);
 
1066
1067	for (i = 0; i < (jl->j_len + 1); i++) {
1068		bn = SB_ONDISK_JOURNAL_1st_BLOCK(s) +
1069		    (jl->j_start + i) % SB_ONDISK_JOURNAL_SIZE(s);
1070		tbh = journal_find_get_block(s, bn);
1071
1072		depth = reiserfs_write_unlock_nested(s);
1073		__wait_on_buffer(tbh);
1074		reiserfs_write_lock_nested(s, depth);
1075		/*
1076		 * since we're using ll_rw_blk above, it might have skipped
1077		 * over a locked buffer.  Double check here
1078		 */
1079		/* redundant, sync_dirty_buffer() checks */
1080		if (buffer_dirty(tbh)) {
1081			depth = reiserfs_write_unlock_nested(s);
1082			sync_dirty_buffer(tbh);
1083			reiserfs_write_lock_nested(s, depth);
1084		}
1085		if (unlikely(!buffer_uptodate(tbh))) {
1086#ifdef CONFIG_REISERFS_CHECK
1087			reiserfs_warning(s, "journal-601",
1088					 "buffer write failed");
1089#endif
1090			retval = -EIO;
1091		}
1092		/* once for journal_find_get_block */
1093		put_bh(tbh);
1094		/* once due to original getblk in do_journal_end */
1095		put_bh(tbh);
1096		atomic_dec(&jl->j_commit_left);
1097	}
1098
1099	BUG_ON(atomic_read(&jl->j_commit_left) != 1);
1100
1101	/*
1102	 * If there was a write error in the journal - we can't commit
1103	 * this transaction - it will be invalid and, if successful,
1104	 * will just end up propagating the write error out to
1105	 * the file system.
1106	 */
1107	if (likely(!retval && !reiserfs_is_journal_aborted (journal))) {
1108		if (buffer_dirty(jl->j_commit_bh))
1109			BUG();
1110		mark_buffer_dirty(jl->j_commit_bh) ;
1111		depth = reiserfs_write_unlock_nested(s);
1112		if (reiserfs_barrier_flush(s))
1113			__sync_dirty_buffer(jl->j_commit_bh,
1114					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1115		else
1116			sync_dirty_buffer(jl->j_commit_bh);
1117		reiserfs_write_lock_nested(s, depth);
1118	}
1119
1120	/*
1121	 * If there was a write error in the journal - we can't commit this
1122	 * transaction - it will be invalid and, if successful, will just end
1123	 * up propagating the write error out to the filesystem.
1124	 */
1125	if (unlikely(!buffer_uptodate(jl->j_commit_bh))) {
1126#ifdef CONFIG_REISERFS_CHECK
1127		reiserfs_warning(s, "journal-615", "buffer write failed");
1128#endif
1129		retval = -EIO;
1130	}
1131	bforget(jl->j_commit_bh);
1132	if (journal->j_last_commit_id != 0 &&
1133	    (jl->j_trans_id - journal->j_last_commit_id) != 1) {
1134		reiserfs_warning(s, "clm-2200", "last commit %lu, current %lu",
1135				 journal->j_last_commit_id, jl->j_trans_id);
1136	}
1137	journal->j_last_commit_id = jl->j_trans_id;
1138
1139	/*
1140	 * now, every commit block is on the disk.  It is safe to allow
1141	 * blocks freed during this transaction to be reallocated
1142	 */
1143	cleanup_freed_for_journal_list(s, jl);
1144
1145	retval = retval ? retval : journal->j_errno;
1146
1147	/* mark the metadata dirty */
1148	if (!retval)
1149		dirty_one_transaction(s, jl);
1150	atomic_dec(&jl->j_commit_left);
1151
1152	if (flushall) {
1153		atomic_set(&jl->j_older_commits_done, 1);
1154	}
1155	mutex_unlock(&jl->j_commit_mutex);
1156put_jl:
1157	put_journal_list(s, jl);
1158
1159	if (retval)
1160		reiserfs_abort(s, retval, "Journal write error in %s",
1161			       __func__);
1162	return retval;
1163}
1164
1165/*
1166 * flush_journal_list frequently needs to find a newer transaction for a
1167 * given block.  This does that, or returns NULL if it can't find anything
1168 */
1169static struct reiserfs_journal_list *find_newer_jl_for_cn(struct
1170							  reiserfs_journal_cnode
1171							  *cn)
1172{
1173	struct super_block *sb = cn->sb;
1174	b_blocknr_t blocknr = cn->blocknr;
1175
1176	cn = cn->hprev;
1177	while (cn) {
1178		if (cn->sb == sb && cn->blocknr == blocknr && cn->jlist) {
1179			return cn->jlist;
1180		}
1181		cn = cn->hprev;
1182	}
1183	return NULL;
1184}
1185
1186static void remove_journal_hash(struct super_block *,
1187				struct reiserfs_journal_cnode **,
1188				struct reiserfs_journal_list *, unsigned long,
1189				int);
1190
1191/*
1192 * once all the real blocks have been flushed, it is safe to remove them
1193 * from the journal list for this transaction.  Aside from freeing the
1194 * cnode, this also allows the block to be reallocated for data blocks
1195 * if it had been deleted.
1196 */
1197static void remove_all_from_journal_list(struct super_block *sb,
1198					 struct reiserfs_journal_list *jl,
1199					 int debug)
1200{
1201	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1202	struct reiserfs_journal_cnode *cn, *last;
1203	cn = jl->j_realblock;
1204
1205	/*
1206	 * which is better, to lock once around the whole loop, or
1207	 * to lock for each call to remove_journal_hash?
1208	 */
1209	while (cn) {
1210		if (cn->blocknr != 0) {
1211			if (debug) {
1212				reiserfs_warning(sb, "reiserfs-2201",
1213						 "block %u, bh is %d, state %ld",
1214						 cn->blocknr, cn->bh ? 1 : 0,
1215						 cn->state);
1216			}
1217			cn->state = 0;
1218			remove_journal_hash(sb, journal->j_list_hash_table,
1219					    jl, cn->blocknr, 1);
1220		}
1221		last = cn;
1222		cn = cn->next;
1223		free_cnode(sb, last);
1224	}
1225	jl->j_realblock = NULL;
1226}
1227
1228/*
1229 * if this timestamp is greater than the timestamp we wrote last to the
1230 * header block, write it to the header block.  once this is done, I can
1231 * safely say the log area for this transaction won't ever be replayed,
1232 * and I can start releasing blocks in this transaction for reuse as data
1233 * blocks.  called by flush_journal_list, before it calls
1234 * remove_all_from_journal_list
1235 */
1236static int _update_journal_header_block(struct super_block *sb,
1237					unsigned long offset,
1238					unsigned int trans_id)
1239{
1240	struct reiserfs_journal_header *jh;
1241	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1242	int depth;
1243
1244	if (reiserfs_is_journal_aborted(journal))
1245		return -EIO;
1246
1247	if (trans_id >= journal->j_last_flush_trans_id) {
1248		if (buffer_locked((journal->j_header_bh))) {
1249			depth = reiserfs_write_unlock_nested(sb);
1250			__wait_on_buffer(journal->j_header_bh);
1251			reiserfs_write_lock_nested(sb, depth);
1252			if (unlikely(!buffer_uptodate(journal->j_header_bh))) {
1253#ifdef CONFIG_REISERFS_CHECK
1254				reiserfs_warning(sb, "journal-699",
1255						 "buffer write failed");
1256#endif
1257				return -EIO;
1258			}
1259		}
1260		journal->j_last_flush_trans_id = trans_id;
1261		journal->j_first_unflushed_offset = offset;
1262		jh = (struct reiserfs_journal_header *)(journal->j_header_bh->
1263							b_data);
1264		jh->j_last_flush_trans_id = cpu_to_le32(trans_id);
1265		jh->j_first_unflushed_offset = cpu_to_le32(offset);
1266		jh->j_mount_id = cpu_to_le32(journal->j_mount_id);
1267
1268		set_buffer_dirty(journal->j_header_bh);
1269		depth = reiserfs_write_unlock_nested(sb);
1270
1271		if (reiserfs_barrier_flush(sb))
1272			__sync_dirty_buffer(journal->j_header_bh,
1273					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1274		else
1275			sync_dirty_buffer(journal->j_header_bh);
1276
1277		reiserfs_write_lock_nested(sb, depth);
1278		if (!buffer_uptodate(journal->j_header_bh)) {
1279			reiserfs_warning(sb, "journal-837",
1280					 "IO error during journal replay");
1281			return -EIO;
1282		}
1283	}
1284	return 0;
1285}
1286
1287static int update_journal_header_block(struct super_block *sb,
1288				       unsigned long offset,
1289				       unsigned int trans_id)
1290{
1291	return _update_journal_header_block(sb, offset, trans_id);
1292}
1293
1294/*
1295** flush any and all journal lists older than you are
1296** can only be called from flush_journal_list
1297*/
1298static int flush_older_journal_lists(struct super_block *sb,
1299				     struct reiserfs_journal_list *jl)
1300{
1301	struct list_head *entry;
1302	struct reiserfs_journal_list *other_jl;
1303	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1304	unsigned int trans_id = jl->j_trans_id;
1305
1306	/*
1307	 * we know we are the only ones flushing things, no extra race
1308	 * protection is required.
1309	 */
1310restart:
1311	entry = journal->j_journal_list.next;
1312	/* Did we wrap? */
1313	if (entry == &journal->j_journal_list)
1314		return 0;
1315	other_jl = JOURNAL_LIST_ENTRY(entry);
1316	if (other_jl->j_trans_id < trans_id) {
1317		BUG_ON(other_jl->j_refcount <= 0);
1318		/* do not flush all */
1319		flush_journal_list(sb, other_jl, 0);
1320
1321		/* other_jl is now deleted from the list */
1322		goto restart;
1323	}
1324	return 0;
1325}
1326
1327static void del_from_work_list(struct super_block *s,
1328			       struct reiserfs_journal_list *jl)
1329{
1330	struct reiserfs_journal *journal = SB_JOURNAL(s);
1331	if (!list_empty(&jl->j_working_list)) {
1332		list_del_init(&jl->j_working_list);
1333		journal->j_num_work_lists--;
1334	}
1335}
1336
1337/*
1338 * flush a journal list, both commit and real blocks
1339 *
1340 * always set flushall to 1, unless you are calling from inside
1341 * flush_journal_list
1342 *
1343 * IMPORTANT.  This can only be called while there are no journal writers,
1344 * and the journal is locked.  That means it can only be called from
1345 * do_journal_end, or by journal_release
1346 */
1347static int flush_journal_list(struct super_block *s,
1348			      struct reiserfs_journal_list *jl, int flushall)
1349{
1350	struct reiserfs_journal_list *pjl;
1351	struct reiserfs_journal_cnode *cn;
1352	int count;
1353	int was_jwait = 0;
1354	int was_dirty = 0;
1355	struct buffer_head *saved_bh;
1356	unsigned long j_len_saved = jl->j_len;
1357	struct reiserfs_journal *journal = SB_JOURNAL(s);
1358	int err = 0;
1359	int depth;
1360
1361	BUG_ON(j_len_saved <= 0);
1362
1363	if (atomic_read(&journal->j_wcount) != 0) {
1364		reiserfs_warning(s, "clm-2048", "called with wcount %d",
1365				 atomic_read(&journal->j_wcount));
1366	}
1367
1368	/* if flushall == 0, the lock is already held */
1369	if (flushall) {
1370		reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1371	} else if (mutex_trylock(&journal->j_flush_mutex)) {
1372		BUG();
1373	}
1374
1375	count = 0;
1376	if (j_len_saved > journal->j_trans_max) {
1377		reiserfs_panic(s, "journal-715", "length is %lu, trans id %lu",
1378			       j_len_saved, jl->j_trans_id);
1379		return 0;
1380	}
1381
1382	/* if all the work is already done, get out of here */
1383	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1384	    atomic_read(&jl->j_commit_left) <= 0) {
1385		goto flush_older_and_return;
1386	}
1387
1388	/*
1389	 * start by putting the commit list on disk.  This will also flush
1390	 * the commit lists of any olders transactions
1391	 */
1392	flush_commit_list(s, jl, 1);
1393
1394	if (!(jl->j_state & LIST_DIRTY)
1395	    && !reiserfs_is_journal_aborted(journal))
1396		BUG();
1397
1398	/* are we done now? */
1399	if (atomic_read(&jl->j_nonzerolen) <= 0 &&
1400	    atomic_read(&jl->j_commit_left) <= 0) {
1401		goto flush_older_and_return;
1402	}
1403
1404	/*
1405	 * loop through each cnode, see if we need to write it,
1406	 * or wait on a more recent transaction, or just ignore it
1407	 */
1408	if (atomic_read(&journal->j_wcount) != 0) {
1409		reiserfs_panic(s, "journal-844", "journal list is flushing, "
1410			       "wcount is not 0");
1411	}
1412	cn = jl->j_realblock;
1413	while (cn) {
1414		was_jwait = 0;
1415		was_dirty = 0;
1416		saved_bh = NULL;
1417		/* blocknr of 0 is no longer in the hash, ignore it */
1418		if (cn->blocknr == 0) {
1419			goto free_cnode;
1420		}
1421
1422		/*
1423		 * This transaction failed commit.
1424		 * Don't write out to the disk
1425		 */
1426		if (!(jl->j_state & LIST_DIRTY))
1427			goto free_cnode;
1428
1429		pjl = find_newer_jl_for_cn(cn);
1430		/*
1431		 * the order is important here.  We check pjl to make sure we
1432		 * don't clear BH_JDirty_wait if we aren't the one writing this
1433		 * block to disk
1434		 */
1435		if (!pjl && cn->bh) {
1436			saved_bh = cn->bh;
1437
1438			/*
1439			 * we do this to make sure nobody releases the
1440			 * buffer while we are working with it
1441			 */
1442			get_bh(saved_bh);
1443
1444			if (buffer_journal_dirty(saved_bh)) {
1445				BUG_ON(!can_dirty(cn));
1446				was_jwait = 1;
1447				was_dirty = 1;
1448			} else if (can_dirty(cn)) {
1449				/*
1450				 * everything with !pjl && jwait
1451				 * should be writable
1452				 */
1453				BUG();
1454			}
1455		}
1456
1457		/*
1458		 * if someone has this block in a newer transaction, just make
1459		 * sure they are committed, and don't try writing it to disk
1460		 */
1461		if (pjl) {
1462			if (atomic_read(&pjl->j_commit_left))
1463				flush_commit_list(s, pjl, 1);
1464			goto free_cnode;
1465		}
1466
1467		/*
1468		 * bh == NULL when the block got to disk on its own, OR,
1469		 * the block got freed in a future transaction
1470		 */
1471		if (saved_bh == NULL) {
1472			goto free_cnode;
1473		}
1474
1475		/*
1476		 * this should never happen.  kupdate_one_transaction has
1477		 * this list locked while it works, so we should never see a
1478		 * buffer here that is not marked JDirty_wait
1479		 */
1480		if ((!was_jwait) && !buffer_locked(saved_bh)) {
1481			reiserfs_warning(s, "journal-813",
1482					 "BAD! buffer %llu %cdirty %cjwait, "
1483					 "not in a newer transaction",
1484					 (unsigned long long)saved_bh->
1485					 b_blocknr, was_dirty ? ' ' : '!',
1486					 was_jwait ? ' ' : '!');
1487		}
1488		if (was_dirty) {
1489			/*
1490			 * we inc again because saved_bh gets decremented
1491			 * at free_cnode
1492			 */
1493			get_bh(saved_bh);
1494			set_bit(BLOCK_NEEDS_FLUSH, &cn->state);
1495			lock_buffer(saved_bh);
1496			BUG_ON(cn->blocknr != saved_bh->b_blocknr);
1497			if (buffer_dirty(saved_bh))
1498				submit_logged_buffer(saved_bh);
1499			else
1500				unlock_buffer(saved_bh);
1501			count++;
1502		} else {
1503			reiserfs_warning(s, "clm-2082",
1504					 "Unable to flush buffer %llu in %s",
1505					 (unsigned long long)saved_bh->
1506					 b_blocknr, __func__);
1507		}
1508free_cnode:
1509		cn = cn->next;
1510		if (saved_bh) {
1511			/*
1512			 * we incremented this to keep others from
1513			 * taking the buffer head away
1514			 */
1515			put_bh(saved_bh);
1516			if (atomic_read(&saved_bh->b_count) < 0) {
1517				reiserfs_warning(s, "journal-945",
1518						 "saved_bh->b_count < 0");
1519			}
1520		}
1521	}
1522	if (count > 0) {
1523		cn = jl->j_realblock;
1524		while (cn) {
1525			if (test_bit(BLOCK_NEEDS_FLUSH, &cn->state)) {
1526				if (!cn->bh) {
1527					reiserfs_panic(s, "journal-1011",
1528						       "cn->bh is NULL");
1529				}
1530
1531				depth = reiserfs_write_unlock_nested(s);
1532				__wait_on_buffer(cn->bh);
1533				reiserfs_write_lock_nested(s, depth);
1534
1535				if (!cn->bh) {
1536					reiserfs_panic(s, "journal-1012",
1537						       "cn->bh is NULL");
1538				}
1539				if (unlikely(!buffer_uptodate(cn->bh))) {
1540#ifdef CONFIG_REISERFS_CHECK
1541					reiserfs_warning(s, "journal-949",
1542							 "buffer write failed");
1543#endif
1544					err = -EIO;
1545				}
1546				/*
1547				 * note, we must clear the JDirty_wait bit
1548				 * after the up to date check, otherwise we
1549				 * race against our flushpage routine
1550				 */
1551				BUG_ON(!test_clear_buffer_journal_dirty
1552				       (cn->bh));
1553
1554				/* drop one ref for us */
1555				put_bh(cn->bh);
1556				/* drop one ref for journal_mark_dirty */
1557				release_buffer_page(cn->bh);
1558			}
1559			cn = cn->next;
1560		}
1561	}
1562
1563	if (err)
1564		reiserfs_abort(s, -EIO,
1565			       "Write error while pushing transaction to disk in %s",
1566			       __func__);
1567flush_older_and_return:
1568
1569	/*
1570	 * before we can update the journal header block, we _must_ flush all
1571	 * real blocks from all older transactions to disk.  This is because
1572	 * once the header block is updated, this transaction will not be
1573	 * replayed after a crash
1574	 */
1575	if (flushall) {
1576		flush_older_journal_lists(s, jl);
1577	}
1578
1579	err = journal->j_errno;
1580	/*
1581	 * before we can remove everything from the hash tables for this
1582	 * transaction, we must make sure it can never be replayed
1583	 *
1584	 * since we are only called from do_journal_end, we know for sure there
1585	 * are no allocations going on while we are flushing journal lists.  So,
1586	 * we only need to update the journal header block for the last list
1587	 * being flushed
1588	 */
1589	if (!err && flushall) {
1590		err =
1591		    update_journal_header_block(s,
1592						(jl->j_start + jl->j_len +
1593						 2) % SB_ONDISK_JOURNAL_SIZE(s),
1594						jl->j_trans_id);
1595		if (err)
1596			reiserfs_abort(s, -EIO,
1597				       "Write error while updating journal header in %s",
1598				       __func__);
1599	}
1600	remove_all_from_journal_list(s, jl, 0);
1601	list_del_init(&jl->j_list);
1602	journal->j_num_lists--;
1603	del_from_work_list(s, jl);
1604
1605	if (journal->j_last_flush_id != 0 &&
1606	    (jl->j_trans_id - journal->j_last_flush_id) != 1) {
1607		reiserfs_warning(s, "clm-2201", "last flush %lu, current %lu",
1608				 journal->j_last_flush_id, jl->j_trans_id);
1609	}
1610	journal->j_last_flush_id = jl->j_trans_id;
1611
1612	/*
1613	 * not strictly required since we are freeing the list, but it should
1614	 * help find code using dead lists later on
1615	 */
1616	jl->j_len = 0;
1617	atomic_set(&jl->j_nonzerolen, 0);
1618	jl->j_start = 0;
1619	jl->j_realblock = NULL;
1620	jl->j_commit_bh = NULL;
1621	jl->j_trans_id = 0;
1622	jl->j_state = 0;
1623	put_journal_list(s, jl);
1624	if (flushall)
1625		mutex_unlock(&journal->j_flush_mutex);
1626	return err;
1627}
1628
1629static int write_one_transaction(struct super_block *s,
1630				 struct reiserfs_journal_list *jl,
1631				 struct buffer_chunk *chunk)
1632{
1633	struct reiserfs_journal_cnode *cn;
1634	int ret = 0;
1635
1636	jl->j_state |= LIST_TOUCHED;
1637	del_from_work_list(s, jl);
1638	if (jl->j_len == 0 || atomic_read(&jl->j_nonzerolen) == 0) {
1639		return 0;
1640	}
1641
1642	cn = jl->j_realblock;
1643	while (cn) {
1644		/*
1645		 * if the blocknr == 0, this has been cleared from the hash,
1646		 * skip it
1647		 */
1648		if (cn->blocknr == 0) {
1649			goto next;
1650		}
1651		if (cn->bh && can_dirty(cn) && buffer_dirty(cn->bh)) {
1652			struct buffer_head *tmp_bh;
1653			/*
1654			 * we can race against journal_mark_freed when we try
1655			 * to lock_buffer(cn->bh), so we have to inc the buffer
1656			 * count, and recheck things after locking
1657			 */
1658			tmp_bh = cn->bh;
1659			get_bh(tmp_bh);
1660			lock_buffer(tmp_bh);
1661			if (cn->bh && can_dirty(cn) && buffer_dirty(tmp_bh)) {
1662				if (!buffer_journal_dirty(tmp_bh) ||
1663				    buffer_journal_prepared(tmp_bh))
1664					BUG();
1665				add_to_chunk(chunk, tmp_bh, NULL, write_chunk);
1666				ret++;
1667			} else {
1668				/* note, cn->bh might be null now */
1669				unlock_buffer(tmp_bh);
1670			}
1671			put_bh(tmp_bh);
1672		}
1673next:
1674		cn = cn->next;
1675		cond_resched();
1676	}
1677	return ret;
1678}
1679
1680/* used by flush_commit_list */
1681static void dirty_one_transaction(struct super_block *s,
1682				 struct reiserfs_journal_list *jl)
1683{
1684	struct reiserfs_journal_cnode *cn;
1685	struct reiserfs_journal_list *pjl;
1686
1687	jl->j_state |= LIST_DIRTY;
1688	cn = jl->j_realblock;
1689	while (cn) {
1690		/*
1691		 * look for a more recent transaction that logged this
1692		 * buffer.  Only the most recent transaction with a buffer in
1693		 * it is allowed to send that buffer to disk
1694		 */
1695		pjl = find_newer_jl_for_cn(cn);
1696		if (!pjl && cn->blocknr && cn->bh
1697		    && buffer_journal_dirty(cn->bh)) {
1698			BUG_ON(!can_dirty(cn));
1699			/*
1700			 * if the buffer is prepared, it will either be logged
1701			 * or restored.  If restored, we need to make sure
1702			 * it actually gets marked dirty
1703			 */
1704			clear_buffer_journal_new(cn->bh);
1705			if (buffer_journal_prepared(cn->bh)) {
1706				set_buffer_journal_restore_dirty(cn->bh);
1707			} else {
1708				set_buffer_journal_test(cn->bh);
1709				mark_buffer_dirty(cn->bh);
1710			}
1711		}
1712		cn = cn->next;
1713	}
1714}
1715
1716static int kupdate_transactions(struct super_block *s,
1717				struct reiserfs_journal_list *jl,
1718				struct reiserfs_journal_list **next_jl,
1719				unsigned int *next_trans_id,
1720				int num_blocks, int num_trans)
1721{
1722	int ret = 0;
1723	int written = 0;
1724	int transactions_flushed = 0;
1725	unsigned int orig_trans_id = jl->j_trans_id;
1726	struct buffer_chunk chunk;
1727	struct list_head *entry;
1728	struct reiserfs_journal *journal = SB_JOURNAL(s);
1729	chunk.nr = 0;
1730
1731	reiserfs_mutex_lock_safe(&journal->j_flush_mutex, s);
1732	if (!journal_list_still_alive(s, orig_trans_id)) {
1733		goto done;
1734	}
1735
1736	/*
1737	 * we've got j_flush_mutex held, nobody is going to delete any
1738	 * of these lists out from underneath us
1739	 */
1740	while ((num_trans && transactions_flushed < num_trans) ||
1741	       (!num_trans && written < num_blocks)) {
1742
1743		if (jl->j_len == 0 || (jl->j_state & LIST_TOUCHED) ||
1744		    atomic_read(&jl->j_commit_left)
1745		    || !(jl->j_state & LIST_DIRTY)) {
1746			del_from_work_list(s, jl);
1747			break;
1748		}
1749		ret = write_one_transaction(s, jl, &chunk);
1750
1751		if (ret < 0)
1752			goto done;
1753		transactions_flushed++;
1754		written += ret;
1755		entry = jl->j_list.next;
1756
1757		/* did we wrap? */
1758		if (entry == &journal->j_journal_list) {
1759			break;
1760		}
1761		jl = JOURNAL_LIST_ENTRY(entry);
1762
1763		/* don't bother with older transactions */
1764		if (jl->j_trans_id <= orig_trans_id)
1765			break;
1766	}
1767	if (chunk.nr) {
1768		write_chunk(&chunk);
1769	}
1770
1771done:
1772	mutex_unlock(&journal->j_flush_mutex);
1773	return ret;
1774}
1775
1776/*
1777 * for o_sync and fsync heavy applications, they tend to use
1778 * all the journa list slots with tiny transactions.  These
1779 * trigger lots and lots of calls to update the header block, which
1780 * adds seeks and slows things down.
1781 *
1782 * This function tries to clear out a large chunk of the journal lists
1783 * at once, which makes everything faster since only the newest journal
1784 * list updates the header block
1785 */
1786static int flush_used_journal_lists(struct super_block *s,
1787				    struct reiserfs_journal_list *jl)
1788{
1789	unsigned long len = 0;
1790	unsigned long cur_len;
1791	int i;
1792	int limit = 256;
1793	struct reiserfs_journal_list *tjl;
1794	struct reiserfs_journal_list *flush_jl;
1795	unsigned int trans_id;
1796	struct reiserfs_journal *journal = SB_JOURNAL(s);
1797
1798	flush_jl = tjl = jl;
1799
1800	/* in data logging mode, try harder to flush a lot of blocks */
1801	if (reiserfs_data_log(s))
1802		limit = 1024;
1803	/* flush for 256 transactions or limit blocks, whichever comes first */
1804	for (i = 0; i < 256 && len < limit; i++) {
1805		if (atomic_read(&tjl->j_commit_left) ||
1806		    tjl->j_trans_id < jl->j_trans_id) {
1807			break;
1808		}
1809		cur_len = atomic_read(&tjl->j_nonzerolen);
1810		if (cur_len > 0) {
1811			tjl->j_state &= ~LIST_TOUCHED;
1812		}
1813		len += cur_len;
1814		flush_jl = tjl;
1815		if (tjl->j_list.next == &journal->j_journal_list)
1816			break;
1817		tjl = JOURNAL_LIST_ENTRY(tjl->j_list.next);
1818	}
1819	get_journal_list(jl);
1820	get_journal_list(flush_jl);
1821
1822	/*
1823	 * try to find a group of blocks we can flush across all the
1824	 * transactions, but only bother if we've actually spanned
1825	 * across multiple lists
1826	 */
1827	if (flush_jl != jl)
1828		kupdate_transactions(s, jl, &tjl, &trans_id, len, i);
1829
1830	flush_journal_list(s, flush_jl, 1);
1831	put_journal_list(s, flush_jl);
1832	put_journal_list(s, jl);
1833	return 0;
1834}
1835
1836/*
1837 * removes any nodes in table with name block and dev as bh.
1838 * only touchs the hnext and hprev pointers.
1839 */
1840static void remove_journal_hash(struct super_block *sb,
1841			 struct reiserfs_journal_cnode **table,
1842			 struct reiserfs_journal_list *jl,
1843			 unsigned long block, int remove_freed)
1844{
1845	struct reiserfs_journal_cnode *cur;
1846	struct reiserfs_journal_cnode **head;
1847
1848	head = &(journal_hash(table, sb, block));
1849	if (!head) {
1850		return;
1851	}
1852	cur = *head;
1853	while (cur) {
1854		if (cur->blocknr == block && cur->sb == sb
1855		    && (jl == NULL || jl == cur->jlist)
1856		    && (!test_bit(BLOCK_FREED, &cur->state) || remove_freed)) {
1857			if (cur->hnext) {
1858				cur->hnext->hprev = cur->hprev;
1859			}
1860			if (cur->hprev) {
1861				cur->hprev->hnext = cur->hnext;
1862			} else {
1863				*head = cur->hnext;
1864			}
1865			cur->blocknr = 0;
1866			cur->sb = NULL;
1867			cur->state = 0;
1868			/*
1869			 * anybody who clears the cur->bh will also
1870			 * dec the nonzerolen
1871			 */
1872			if (cur->bh && cur->jlist)
1873				atomic_dec(&cur->jlist->j_nonzerolen);
1874			cur->bh = NULL;
1875			cur->jlist = NULL;
1876		}
1877		cur = cur->hnext;
1878	}
1879}
1880
1881static void free_journal_ram(struct super_block *sb)
1882{
1883	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1884	kfree(journal->j_current_jl);
1885	journal->j_num_lists--;
1886
1887	vfree(journal->j_cnode_free_orig);
1888	free_list_bitmaps(sb, journal->j_list_bitmap);
1889	free_bitmap_nodes(sb);	/* must be after free_list_bitmaps */
1890	if (journal->j_header_bh) {
1891		brelse(journal->j_header_bh);
1892	}
1893	/*
1894	 * j_header_bh is on the journal dev, make sure
1895	 * not to release the journal dev until we brelse j_header_bh
1896	 */
1897	release_journal_dev(sb, journal);
1898	vfree(journal);
1899}
1900
1901/*
1902 * call on unmount.  Only set error to 1 if you haven't made your way out
1903 * of read_super() yet.  Any other caller must keep error at 0.
1904 */
1905static int do_journal_release(struct reiserfs_transaction_handle *th,
1906			      struct super_block *sb, int error)
1907{
1908	struct reiserfs_transaction_handle myth;
1909	struct reiserfs_journal *journal = SB_JOURNAL(sb);
1910
1911	/*
1912	 * we only want to flush out transactions if we were
1913	 * called with error == 0
1914	 */
1915	if (!error && !sb_rdonly(sb)) {
1916		/* end the current trans */
1917		BUG_ON(!th->t_trans_id);
1918		do_journal_end(th, FLUSH_ALL);
1919
1920		/*
1921		 * make sure something gets logged to force
1922		 * our way into the flush code
1923		 */
1924		if (!journal_join(&myth, sb)) {
1925			reiserfs_prepare_for_journal(sb,
1926						     SB_BUFFER_WITH_SB(sb),
1927						     1);
1928			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1929			do_journal_end(&myth, FLUSH_ALL);
1930		}
1931	}
1932
1933	/* this also catches errors during the do_journal_end above */
1934	if (!error && reiserfs_is_journal_aborted(journal)) {
1935		memset(&myth, 0, sizeof(myth));
1936		if (!journal_join_abort(&myth, sb)) {
1937			reiserfs_prepare_for_journal(sb,
1938						     SB_BUFFER_WITH_SB(sb),
1939						     1);
1940			journal_mark_dirty(&myth, SB_BUFFER_WITH_SB(sb));
1941			do_journal_end(&myth, FLUSH_ALL);
1942		}
1943	}
1944
1945
1946	/*
1947	 * We must release the write lock here because
1948	 * the workqueue job (flush_async_commit) needs this lock
1949	 */
1950	reiserfs_write_unlock(sb);
1951
1952	/*
1953	 * Cancel flushing of old commits. Note that neither of these works
1954	 * will be requeued because superblock is being shutdown and doesn't
1955	 * have SB_ACTIVE set.
1956	 */
1957	reiserfs_cancel_old_flush(sb);
1958	/* wait for all commits to finish */
1959	cancel_delayed_work_sync(&SB_JOURNAL(sb)->j_work);
1960
1961	free_journal_ram(sb);
1962
1963	reiserfs_write_lock(sb);
1964
1965	return 0;
1966}
1967
1968/* * call on unmount.  flush all journal trans, release all alloc'd ram */
1969int journal_release(struct reiserfs_transaction_handle *th,
1970		    struct super_block *sb)
1971{
1972	return do_journal_release(th, sb, 0);
1973}
1974
1975/* only call from an error condition inside reiserfs_read_super!  */
1976int journal_release_error(struct reiserfs_transaction_handle *th,
1977			  struct super_block *sb)
1978{
1979	return do_journal_release(th, sb, 1);
1980}
1981
1982/*
1983 * compares description block with commit block.
1984 * returns 1 if they differ, 0 if they are the same
1985 */
1986static int journal_compare_desc_commit(struct super_block *sb,
1987				       struct reiserfs_journal_desc *desc,
1988				       struct reiserfs_journal_commit *commit)
1989{
1990	if (get_commit_trans_id(commit) != get_desc_trans_id(desc) ||
1991	    get_commit_trans_len(commit) != get_desc_trans_len(desc) ||
1992	    get_commit_trans_len(commit) > SB_JOURNAL(sb)->j_trans_max ||
1993	    get_commit_trans_len(commit) <= 0) {
1994		return 1;
1995	}
1996	return 0;
1997}
1998
1999/*
2000 * returns 0 if it did not find a description block
2001 * returns -1 if it found a corrupt commit block
2002 * returns 1 if both desc and commit were valid
2003 * NOTE: only called during fs mount
2004 */
2005static int journal_transaction_is_valid(struct super_block *sb,
2006					struct buffer_head *d_bh,
2007					unsigned int *oldest_invalid_trans_id,
2008					unsigned long *newest_mount_id)
2009{
2010	struct reiserfs_journal_desc *desc;
2011	struct reiserfs_journal_commit *commit;
2012	struct buffer_head *c_bh;
2013	unsigned long offset;
2014
2015	if (!d_bh)
2016		return 0;
2017
2018	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2019	if (get_desc_trans_len(desc) > 0
2020	    && !memcmp(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8)) {
2021		if (oldest_invalid_trans_id && *oldest_invalid_trans_id
2022		    && get_desc_trans_id(desc) > *oldest_invalid_trans_id) {
2023			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2024				       "journal-986: transaction "
2025				       "is valid returning because trans_id %d is greater than "
2026				       "oldest_invalid %lu",
2027				       get_desc_trans_id(desc),
2028				       *oldest_invalid_trans_id);
2029			return 0;
2030		}
2031		if (newest_mount_id
2032		    && *newest_mount_id > get_desc_mount_id(desc)) {
2033			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2034				       "journal-1087: transaction "
2035				       "is valid returning because mount_id %d is less than "
2036				       "newest_mount_id %lu",
2037				       get_desc_mount_id(desc),
2038				       *newest_mount_id);
2039			return -1;
2040		}
2041		if (get_desc_trans_len(desc) > SB_JOURNAL(sb)->j_trans_max) {
2042			reiserfs_warning(sb, "journal-2018",
2043					 "Bad transaction length %d "
2044					 "encountered, ignoring transaction",
2045					 get_desc_trans_len(desc));
2046			return -1;
2047		}
2048		offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2049
2050		/*
2051		 * ok, we have a journal description block,
2052		 * let's see if the transaction was valid
2053		 */
2054		c_bh =
2055		    journal_bread(sb,
2056				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2057				  ((offset + get_desc_trans_len(desc) +
2058				    1) % SB_ONDISK_JOURNAL_SIZE(sb)));
2059		if (!c_bh)
2060			return 0;
2061		commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2062		if (journal_compare_desc_commit(sb, desc, commit)) {
2063			reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2064				       "journal_transaction_is_valid, commit offset %ld had bad "
2065				       "time %d or length %d",
2066				       c_bh->b_blocknr -
2067				       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2068				       get_commit_trans_id(commit),
2069				       get_commit_trans_len(commit));
2070			brelse(c_bh);
2071			if (oldest_invalid_trans_id) {
2072				*oldest_invalid_trans_id =
2073				    get_desc_trans_id(desc);
2074				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2075					       "journal-1004: "
2076					       "transaction_is_valid setting oldest invalid trans_id "
2077					       "to %d",
2078					       get_desc_trans_id(desc));
2079			}
2080			return -1;
2081		}
2082		brelse(c_bh);
2083		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2084			       "journal-1006: found valid "
2085			       "transaction start offset %llu, len %d id %d",
2086			       d_bh->b_blocknr -
2087			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2088			       get_desc_trans_len(desc),
2089			       get_desc_trans_id(desc));
2090		return 1;
2091	} else {
2092		return 0;
2093	}
2094}
2095
2096static void brelse_array(struct buffer_head **heads, int num)
2097{
2098	int i;
2099	for (i = 0; i < num; i++) {
2100		brelse(heads[i]);
2101	}
2102}
2103
2104/*
2105 * given the start, and values for the oldest acceptable transactions,
2106 * this either reads in a replays a transaction, or returns because the
2107 * transaction is invalid, or too old.
2108 * NOTE: only called during fs mount
2109 */
2110static int journal_read_transaction(struct super_block *sb,
2111				    unsigned long cur_dblock,
2112				    unsigned long oldest_start,
2113				    unsigned int oldest_trans_id,
2114				    unsigned long newest_mount_id)
2115{
2116	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2117	struct reiserfs_journal_desc *desc;
2118	struct reiserfs_journal_commit *commit;
2119	unsigned int trans_id = 0;
2120	struct buffer_head *c_bh;
2121	struct buffer_head *d_bh;
2122	struct buffer_head **log_blocks = NULL;
2123	struct buffer_head **real_blocks = NULL;
2124	unsigned int trans_offset;
2125	int i;
2126	int trans_half;
2127
2128	d_bh = journal_bread(sb, cur_dblock);
2129	if (!d_bh)
2130		return 1;
2131	desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2132	trans_offset = d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2133	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1037: "
2134		       "journal_read_transaction, offset %llu, len %d mount_id %d",
2135		       d_bh->b_blocknr - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2136		       get_desc_trans_len(desc), get_desc_mount_id(desc));
2137	if (get_desc_trans_id(desc) < oldest_trans_id) {
2138		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1039: "
2139			       "journal_read_trans skipping because %lu is too old",
2140			       cur_dblock -
2141			       SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2142		brelse(d_bh);
2143		return 1;
2144	}
2145	if (get_desc_mount_id(desc) != newest_mount_id) {
2146		reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1146: "
2147			       "journal_read_trans skipping because %d is != "
2148			       "newest_mount_id %lu", get_desc_mount_id(desc),
2149			       newest_mount_id);
2150		brelse(d_bh);
2151		return 1;
2152	}
2153	c_bh = journal_bread(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2154			     ((trans_offset + get_desc_trans_len(desc) + 1) %
2155			      SB_ONDISK_JOURNAL_SIZE(sb)));
2156	if (!c_bh) {
2157		brelse(d_bh);
2158		return 1;
2159	}
2160	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
2161	if (journal_compare_desc_commit(sb, desc, commit)) {
2162		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2163			       "journal_read_transaction, "
2164			       "commit offset %llu had bad time %d or length %d",
2165			       c_bh->b_blocknr -
2166			       SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2167			       get_commit_trans_id(commit),
2168			       get_commit_trans_len(commit));
2169		brelse(c_bh);
2170		brelse(d_bh);
2171		return 1;
2172	}
2173
2174	if (bdev_read_only(sb->s_bdev)) {
2175		reiserfs_warning(sb, "clm-2076",
2176				 "device is readonly, unable to replay log");
2177		brelse(c_bh);
2178		brelse(d_bh);
2179		return -EROFS;
2180	}
2181
2182	trans_id = get_desc_trans_id(desc);
2183	/*
2184	 * now we know we've got a good transaction, and it was
2185	 * inside the valid time ranges
2186	 */
2187	log_blocks = kmalloc_array(get_desc_trans_len(desc),
2188				   sizeof(struct buffer_head *),
2189				   GFP_NOFS);
2190	real_blocks = kmalloc_array(get_desc_trans_len(desc),
2191				    sizeof(struct buffer_head *),
2192				    GFP_NOFS);
2193	if (!log_blocks || !real_blocks) {
2194		brelse(c_bh);
2195		brelse(d_bh);
2196		kfree(log_blocks);
2197		kfree(real_blocks);
2198		reiserfs_warning(sb, "journal-1169",
2199				 "kmalloc failed, unable to mount FS");
2200		return -1;
2201	}
2202	/* get all the buffer heads */
2203	trans_half = journal_trans_half(sb->s_blocksize);
2204	for (i = 0; i < get_desc_trans_len(desc); i++) {
2205		log_blocks[i] =
2206		    journal_getblk(sb,
2207				   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2208				   (trans_offset + 1 +
2209				    i) % SB_ONDISK_JOURNAL_SIZE(sb));
2210		if (i < trans_half) {
2211			real_blocks[i] =
2212			    sb_getblk(sb,
2213				      le32_to_cpu(desc->j_realblock[i]));
2214		} else {
2215			real_blocks[i] =
2216			    sb_getblk(sb,
2217				      le32_to_cpu(commit->
2218						  j_realblock[i - trans_half]));
2219		}
2220		if (real_blocks[i]->b_blocknr > SB_BLOCK_COUNT(sb)) {
2221			reiserfs_warning(sb, "journal-1207",
2222					 "REPLAY FAILURE fsck required! "
2223					 "Block to replay is outside of "
2224					 "filesystem");
2225			goto abort_replay;
2226		}
2227		/* make sure we don't try to replay onto log or reserved area */
2228		if (is_block_in_log_or_reserved_area
2229		    (sb, real_blocks[i]->b_blocknr)) {
2230			reiserfs_warning(sb, "journal-1204",
2231					 "REPLAY FAILURE fsck required! "
2232					 "Trying to replay onto a log block");
2233abort_replay:
2234			brelse_array(log_blocks, i);
2235			brelse_array(real_blocks, i);
2236			brelse(c_bh);
2237			brelse(d_bh);
2238			kfree(log_blocks);
2239			kfree(real_blocks);
2240			return -1;
2241		}
2242	}
2243	/* read in the log blocks, memcpy to the corresponding real block */
2244	ll_rw_block(REQ_OP_READ, 0, get_desc_trans_len(desc), log_blocks);
2245	for (i = 0; i < get_desc_trans_len(desc); i++) {
2246
2247		wait_on_buffer(log_blocks[i]);
2248		if (!buffer_uptodate(log_blocks[i])) {
2249			reiserfs_warning(sb, "journal-1212",
2250					 "REPLAY FAILURE fsck required! "
2251					 "buffer write failed");
2252			brelse_array(log_blocks + i,
2253				     get_desc_trans_len(desc) - i);
2254			brelse_array(real_blocks, get_desc_trans_len(desc));
2255			brelse(c_bh);
2256			brelse(d_bh);
2257			kfree(log_blocks);
2258			kfree(real_blocks);
2259			return -1;
2260		}
2261		memcpy(real_blocks[i]->b_data, log_blocks[i]->b_data,
2262		       real_blocks[i]->b_size);
2263		set_buffer_uptodate(real_blocks[i]);
2264		brelse(log_blocks[i]);
2265	}
2266	/* flush out the real blocks */
2267	for (i = 0; i < get_desc_trans_len(desc); i++) {
2268		set_buffer_dirty(real_blocks[i]);
2269		write_dirty_buffer(real_blocks[i], 0);
2270	}
2271	for (i = 0; i < get_desc_trans_len(desc); i++) {
2272		wait_on_buffer(real_blocks[i]);
2273		if (!buffer_uptodate(real_blocks[i])) {
2274			reiserfs_warning(sb, "journal-1226",
2275					 "REPLAY FAILURE, fsck required! "
2276					 "buffer write failed");
2277			brelse_array(real_blocks + i,
2278				     get_desc_trans_len(desc) - i);
2279			brelse(c_bh);
2280			brelse(d_bh);
2281			kfree(log_blocks);
2282			kfree(real_blocks);
2283			return -1;
2284		}
2285		brelse(real_blocks[i]);
2286	}
2287	cur_dblock =
2288	    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2289	    ((trans_offset + get_desc_trans_len(desc) +
2290	      2) % SB_ONDISK_JOURNAL_SIZE(sb));
2291	reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2292		       "journal-1095: setting journal " "start to offset %ld",
2293		       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb));
2294
2295	/*
2296	 * init starting values for the first transaction, in case
2297	 * this is the last transaction to be replayed.
2298	 */
2299	journal->j_start = cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2300	journal->j_last_flush_trans_id = trans_id;
2301	journal->j_trans_id = trans_id + 1;
2302	/* check for trans_id overflow */
2303	if (journal->j_trans_id == 0)
2304		journal->j_trans_id = 10;
2305	brelse(c_bh);
2306	brelse(d_bh);
2307	kfree(log_blocks);
2308	kfree(real_blocks);
2309	return 0;
2310}
2311
2312/*
2313 * This function reads blocks starting from block and to max_block of bufsize
2314 * size (but no more than BUFNR blocks at a time). This proved to improve
2315 * mounting speed on self-rebuilding raid5 arrays at least.
2316 * Right now it is only used from journal code. But later we might use it
2317 * from other places.
2318 * Note: Do not use journal_getblk/sb_getblk functions here!
2319 */
2320static struct buffer_head *reiserfs_breada(struct block_device *dev,
2321					   b_blocknr_t block, int bufsize,
2322					   b_blocknr_t max_block)
2323{
2324	struct buffer_head *bhlist[BUFNR];
2325	unsigned int blocks = BUFNR;
2326	struct buffer_head *bh;
2327	int i, j;
2328
2329	bh = __getblk(dev, block, bufsize);
2330	if (buffer_uptodate(bh))
2331		return (bh);
2332
2333	if (block + BUFNR > max_block) {
2334		blocks = max_block - block;
2335	}
2336	bhlist[0] = bh;
2337	j = 1;
2338	for (i = 1; i < blocks; i++) {
2339		bh = __getblk(dev, block + i, bufsize);
2340		if (buffer_uptodate(bh)) {
2341			brelse(bh);
2342			break;
2343		} else
2344			bhlist[j++] = bh;
2345	}
2346	ll_rw_block(REQ_OP_READ, 0, j, bhlist);
 
 
2347	for (i = 1; i < j; i++)
2348		brelse(bhlist[i]);
2349	bh = bhlist[0];
2350	wait_on_buffer(bh);
2351	if (buffer_uptodate(bh))
2352		return bh;
2353	brelse(bh);
2354	return NULL;
2355}
2356
2357/*
2358 * read and replay the log
2359 * on a clean unmount, the journal header's next unflushed pointer will be
2360 * to an invalid transaction.  This tests that before finding all the
2361 * transactions in the log, which makes normal mount times fast.
2362 *
2363 * After a crash, this starts with the next unflushed transaction, and
2364 * replays until it finds one too old, or invalid.
2365 *
2366 * On exit, it sets things up so the first transaction will work correctly.
2367 * NOTE: only called during fs mount
2368 */
2369static int journal_read(struct super_block *sb)
2370{
2371	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2372	struct reiserfs_journal_desc *desc;
2373	unsigned int oldest_trans_id = 0;
2374	unsigned int oldest_invalid_trans_id = 0;
2375	time64_t start;
2376	unsigned long oldest_start = 0;
2377	unsigned long cur_dblock = 0;
2378	unsigned long newest_mount_id = 9;
2379	struct buffer_head *d_bh;
2380	struct reiserfs_journal_header *jh;
2381	int valid_journal_header = 0;
2382	int replay_count = 0;
2383	int continue_replay = 1;
2384	int ret;
2385
2386	cur_dblock = SB_ONDISK_JOURNAL_1st_BLOCK(sb);
2387	reiserfs_info(sb, "checking transaction log (%pg)\n",
2388		      journal->j_dev_bd);
2389	start = ktime_get_seconds();
2390
2391	/*
2392	 * step 1, read in the journal header block.  Check the transaction
2393	 * it says is the first unflushed, and if that transaction is not
2394	 * valid, replay is done
2395	 */
2396	journal->j_header_bh = journal_bread(sb,
2397					     SB_ONDISK_JOURNAL_1st_BLOCK(sb)
2398					     + SB_ONDISK_JOURNAL_SIZE(sb));
2399	if (!journal->j_header_bh) {
2400		return 1;
2401	}
2402	jh = (struct reiserfs_journal_header *)(journal->j_header_bh->b_data);
2403	if (le32_to_cpu(jh->j_first_unflushed_offset) <
2404	    SB_ONDISK_JOURNAL_SIZE(sb)
2405	    && le32_to_cpu(jh->j_last_flush_trans_id) > 0) {
2406		oldest_start =
2407		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2408		    le32_to_cpu(jh->j_first_unflushed_offset);
2409		oldest_trans_id = le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2410		newest_mount_id = le32_to_cpu(jh->j_mount_id);
2411		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2412			       "journal-1153: found in "
2413			       "header: first_unflushed_offset %d, last_flushed_trans_id "
2414			       "%lu", le32_to_cpu(jh->j_first_unflushed_offset),
2415			       le32_to_cpu(jh->j_last_flush_trans_id));
2416		valid_journal_header = 1;
2417
2418		/*
2419		 * now, we try to read the first unflushed offset.  If it
2420		 * is not valid, there is nothing more we can do, and it
2421		 * makes no sense to read through the whole log.
2422		 */
2423		d_bh =
2424		    journal_bread(sb,
2425				  SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2426				  le32_to_cpu(jh->j_first_unflushed_offset));
2427		ret = journal_transaction_is_valid(sb, d_bh, NULL, NULL);
2428		if (!ret) {
2429			continue_replay = 0;
2430		}
2431		brelse(d_bh);
2432		goto start_log_replay;
2433	}
2434
2435	/*
2436	 * ok, there are transactions that need to be replayed.  start
2437	 * with the first log block, find all the valid transactions, and
2438	 * pick out the oldest.
2439	 */
2440	while (continue_replay
2441	       && cur_dblock <
2442	       (SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2443		SB_ONDISK_JOURNAL_SIZE(sb))) {
2444		/*
2445		 * Note that it is required for blocksize of primary fs
2446		 * device and journal device to be the same
2447		 */
2448		d_bh =
2449		    reiserfs_breada(journal->j_dev_bd, cur_dblock,
2450				    sb->s_blocksize,
2451				    SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2452				    SB_ONDISK_JOURNAL_SIZE(sb));
2453		ret =
2454		    journal_transaction_is_valid(sb, d_bh,
2455						 &oldest_invalid_trans_id,
2456						 &newest_mount_id);
2457		if (ret == 1) {
2458			desc = (struct reiserfs_journal_desc *)d_bh->b_data;
2459			if (oldest_start == 0) {	/* init all oldest_ values */
2460				oldest_trans_id = get_desc_trans_id(desc);
2461				oldest_start = d_bh->b_blocknr;
2462				newest_mount_id = get_desc_mount_id(desc);
2463				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2464					       "journal-1179: Setting "
2465					       "oldest_start to offset %llu, trans_id %lu",
2466					       oldest_start -
2467					       SB_ONDISK_JOURNAL_1st_BLOCK
2468					       (sb), oldest_trans_id);
2469			} else if (oldest_trans_id > get_desc_trans_id(desc)) {
2470				/* one we just read was older */
2471				oldest_trans_id = get_desc_trans_id(desc);
2472				oldest_start = d_bh->b_blocknr;
2473				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2474					       "journal-1180: Resetting "
2475					       "oldest_start to offset %lu, trans_id %lu",
2476					       oldest_start -
2477					       SB_ONDISK_JOURNAL_1st_BLOCK
2478					       (sb), oldest_trans_id);
2479			}
2480			if (newest_mount_id < get_desc_mount_id(desc)) {
2481				newest_mount_id = get_desc_mount_id(desc);
2482				reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2483					       "journal-1299: Setting "
2484					       "newest_mount_id to %d",
2485					       get_desc_mount_id(desc));
2486			}
2487			cur_dblock += get_desc_trans_len(desc) + 2;
2488		} else {
2489			cur_dblock++;
2490		}
2491		brelse(d_bh);
2492	}
2493
2494start_log_replay:
2495	cur_dblock = oldest_start;
2496	if (oldest_trans_id) {
2497		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2498			       "journal-1206: Starting replay "
2499			       "from offset %llu, trans_id %lu",
2500			       cur_dblock - SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2501			       oldest_trans_id);
2502
2503	}
2504	replay_count = 0;
2505	while (continue_replay && oldest_trans_id > 0) {
2506		ret =
2507		    journal_read_transaction(sb, cur_dblock, oldest_start,
2508					     oldest_trans_id, newest_mount_id);
2509		if (ret < 0) {
2510			return ret;
2511		} else if (ret != 0) {
2512			break;
2513		}
2514		cur_dblock =
2515		    SB_ONDISK_JOURNAL_1st_BLOCK(sb) + journal->j_start;
2516		replay_count++;
2517		if (cur_dblock == oldest_start)
2518			break;
2519	}
2520
2521	if (oldest_trans_id == 0) {
2522		reiserfs_debug(sb, REISERFS_DEBUG_CODE,
2523			       "journal-1225: No valid " "transactions found");
2524	}
2525	/*
2526	 * j_start does not get set correctly if we don't replay any
2527	 * transactions.  if we had a valid journal_header, set j_start
2528	 * to the first unflushed transaction value, copy the trans_id
2529	 * from the header
2530	 */
2531	if (valid_journal_header && replay_count == 0) {
2532		journal->j_start = le32_to_cpu(jh->j_first_unflushed_offset);
2533		journal->j_trans_id =
2534		    le32_to_cpu(jh->j_last_flush_trans_id) + 1;
2535		/* check for trans_id overflow */
2536		if (journal->j_trans_id == 0)
2537			journal->j_trans_id = 10;
2538		journal->j_last_flush_trans_id =
2539		    le32_to_cpu(jh->j_last_flush_trans_id);
2540		journal->j_mount_id = le32_to_cpu(jh->j_mount_id) + 1;
2541	} else {
2542		journal->j_mount_id = newest_mount_id + 1;
2543	}
2544	reiserfs_debug(sb, REISERFS_DEBUG_CODE, "journal-1299: Setting "
2545		       "newest_mount_id to %lu", journal->j_mount_id);
2546	journal->j_first_unflushed_offset = journal->j_start;
2547	if (replay_count > 0) {
2548		reiserfs_info(sb,
2549			      "replayed %d transactions in %lu seconds\n",
2550			      replay_count, ktime_get_seconds() - start);
2551	}
2552	/* needed to satisfy the locking in _update_journal_header_block */
2553	reiserfs_write_lock(sb);
2554	if (!bdev_read_only(sb->s_bdev) &&
2555	    _update_journal_header_block(sb, journal->j_start,
2556					 journal->j_last_flush_trans_id)) {
2557		reiserfs_write_unlock(sb);
2558		/*
2559		 * replay failed, caller must call free_journal_ram and abort
2560		 * the mount
2561		 */
2562		return -1;
2563	}
2564	reiserfs_write_unlock(sb);
2565	return 0;
2566}
2567
2568static struct reiserfs_journal_list *alloc_journal_list(struct super_block *s)
2569{
2570	struct reiserfs_journal_list *jl;
2571	jl = kzalloc(sizeof(struct reiserfs_journal_list),
2572		     GFP_NOFS | __GFP_NOFAIL);
2573	INIT_LIST_HEAD(&jl->j_list);
2574	INIT_LIST_HEAD(&jl->j_working_list);
2575	INIT_LIST_HEAD(&jl->j_tail_bh_list);
2576	INIT_LIST_HEAD(&jl->j_bh_list);
2577	mutex_init(&jl->j_commit_mutex);
2578	SB_JOURNAL(s)->j_num_lists++;
2579	get_journal_list(jl);
2580	return jl;
2581}
2582
2583static void journal_list_init(struct super_block *sb)
2584{
2585	SB_JOURNAL(sb)->j_current_jl = alloc_journal_list(sb);
2586}
2587
2588static void release_journal_dev(struct super_block *super,
2589			       struct reiserfs_journal *journal)
2590{
2591	if (journal->j_dev_bd != NULL) {
2592		blkdev_put(journal->j_dev_bd, journal->j_dev_mode);
2593		journal->j_dev_bd = NULL;
2594	}
2595}
2596
2597static int journal_init_dev(struct super_block *super,
2598			    struct reiserfs_journal *journal,
2599			    const char *jdev_name)
2600{
2601	int result;
2602	dev_t jdev;
2603	fmode_t blkdev_mode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
2604	char b[BDEVNAME_SIZE];
2605
2606	result = 0;
2607
2608	journal->j_dev_bd = NULL;
2609	jdev = SB_ONDISK_JOURNAL_DEVICE(super) ?
2610	    new_decode_dev(SB_ONDISK_JOURNAL_DEVICE(super)) : super->s_dev;
2611
2612	if (bdev_read_only(super->s_bdev))
2613		blkdev_mode = FMODE_READ;
2614
2615	/* there is no "jdev" option and journal is on separate device */
2616	if ((!jdev_name || !jdev_name[0])) {
2617		if (jdev == super->s_dev)
2618			blkdev_mode &= ~FMODE_EXCL;
2619		journal->j_dev_bd = blkdev_get_by_dev(jdev, blkdev_mode,
2620						      journal);
2621		journal->j_dev_mode = blkdev_mode;
2622		if (IS_ERR(journal->j_dev_bd)) {
2623			result = PTR_ERR(journal->j_dev_bd);
2624			journal->j_dev_bd = NULL;
2625			reiserfs_warning(super, "sh-458",
2626					 "cannot init journal device '%s': %i",
2627					 __bdevname(jdev, b), result);
2628			return result;
2629		} else if (jdev != super->s_dev)
2630			set_blocksize(journal->j_dev_bd, super->s_blocksize);
2631
2632		return 0;
2633	}
2634
2635	journal->j_dev_mode = blkdev_mode;
2636	journal->j_dev_bd = blkdev_get_by_path(jdev_name, blkdev_mode, journal);
2637	if (IS_ERR(journal->j_dev_bd)) {
2638		result = PTR_ERR(journal->j_dev_bd);
2639		journal->j_dev_bd = NULL;
2640		reiserfs_warning(super, "sh-457",
2641				 "journal_init_dev: Cannot open '%s': %i",
2642				 jdev_name, result);
2643		return result;
2644	}
2645
2646	set_blocksize(journal->j_dev_bd, super->s_blocksize);
2647	reiserfs_info(super,
2648		      "journal_init_dev: journal device: %pg\n",
2649		      journal->j_dev_bd);
2650	return 0;
2651}
2652
2653/*
2654 * When creating/tuning a file system user can assign some
2655 * journal params within boundaries which depend on the ratio
2656 * blocksize/standard_blocksize.
2657 *
2658 * For blocks >= standard_blocksize transaction size should
2659 * be not less then JOURNAL_TRANS_MIN_DEFAULT, and not more
2660 * then JOURNAL_TRANS_MAX_DEFAULT.
2661 *
2662 * For blocks < standard_blocksize these boundaries should be
2663 * decreased proportionally.
2664 */
2665#define REISERFS_STANDARD_BLKSIZE (4096)
2666
2667static int check_advise_trans_params(struct super_block *sb,
2668				     struct reiserfs_journal *journal)
2669{
2670        if (journal->j_trans_max) {
2671		/* Non-default journal params.  Do sanity check for them. */
2672	        int ratio = 1;
2673		if (sb->s_blocksize < REISERFS_STANDARD_BLKSIZE)
2674		        ratio = REISERFS_STANDARD_BLKSIZE / sb->s_blocksize;
2675
2676		if (journal->j_trans_max > JOURNAL_TRANS_MAX_DEFAULT / ratio ||
2677		    journal->j_trans_max < JOURNAL_TRANS_MIN_DEFAULT / ratio ||
2678		    SB_ONDISK_JOURNAL_SIZE(sb) / journal->j_trans_max <
2679		    JOURNAL_MIN_RATIO) {
2680			reiserfs_warning(sb, "sh-462",
2681					 "bad transaction max size (%u). "
2682					 "FSCK?", journal->j_trans_max);
2683			return 1;
2684		}
2685		if (journal->j_max_batch != (journal->j_trans_max) *
2686		        JOURNAL_MAX_BATCH_DEFAULT/JOURNAL_TRANS_MAX_DEFAULT) {
2687			reiserfs_warning(sb, "sh-463",
2688					 "bad transaction max batch (%u). "
2689					 "FSCK?", journal->j_max_batch);
2690			return 1;
2691		}
2692	} else {
2693		/*
2694		 * Default journal params.
2695		 * The file system was created by old version
2696		 * of mkreiserfs, so some fields contain zeros,
2697		 * and we need to advise proper values for them
2698		 */
2699		if (sb->s_blocksize != REISERFS_STANDARD_BLKSIZE) {
2700			reiserfs_warning(sb, "sh-464", "bad blocksize (%u)",
2701					 sb->s_blocksize);
2702			return 1;
2703		}
2704		journal->j_trans_max = JOURNAL_TRANS_MAX_DEFAULT;
2705		journal->j_max_batch = JOURNAL_MAX_BATCH_DEFAULT;
2706		journal->j_max_commit_age = JOURNAL_MAX_COMMIT_AGE;
2707	}
2708	return 0;
2709}
2710
2711/* must be called once on fs mount.  calls journal_read for you */
2712int journal_init(struct super_block *sb, const char *j_dev_name,
2713		 int old_format, unsigned int commit_max_age)
2714{
2715	int num_cnodes = SB_ONDISK_JOURNAL_SIZE(sb) * 2;
2716	struct buffer_head *bhjh;
2717	struct reiserfs_super_block *rs;
2718	struct reiserfs_journal_header *jh;
2719	struct reiserfs_journal *journal;
2720	struct reiserfs_journal_list *jl;
2721	int ret;
2722
2723	journal = SB_JOURNAL(sb) = vzalloc(sizeof(struct reiserfs_journal));
2724	if (!journal) {
2725		reiserfs_warning(sb, "journal-1256",
2726				 "unable to get memory for journal structure");
2727		return 1;
2728	}
2729	INIT_LIST_HEAD(&journal->j_bitmap_nodes);
2730	INIT_LIST_HEAD(&journal->j_prealloc_list);
2731	INIT_LIST_HEAD(&journal->j_working_list);
2732	INIT_LIST_HEAD(&journal->j_journal_list);
2733	journal->j_persistent_trans = 0;
2734	if (reiserfs_allocate_list_bitmaps(sb, journal->j_list_bitmap,
2735					   reiserfs_bmap_count(sb)))
2736		goto free_and_return;
2737
2738	allocate_bitmap_nodes(sb);
2739
2740	/* reserved for journal area support */
2741	SB_JOURNAL_1st_RESERVED_BLOCK(sb) = (old_format ?
2742						 REISERFS_OLD_DISK_OFFSET_IN_BYTES
2743						 / sb->s_blocksize +
2744						 reiserfs_bmap_count(sb) +
2745						 1 :
2746						 REISERFS_DISK_OFFSET_IN_BYTES /
2747						 sb->s_blocksize + 2);
2748
2749	/*
2750	 * Sanity check to see is the standard journal fitting
2751	 * within first bitmap (actual for small blocksizes)
2752	 */
2753	if (!SB_ONDISK_JOURNAL_DEVICE(sb) &&
2754	    (SB_JOURNAL_1st_RESERVED_BLOCK(sb) +
2755	     SB_ONDISK_JOURNAL_SIZE(sb) > sb->s_blocksize * 8)) {
2756		reiserfs_warning(sb, "journal-1393",
2757				 "journal does not fit for area addressed "
2758				 "by first of bitmap blocks. It starts at "
2759				 "%u and its size is %u. Block size %ld",
2760				 SB_JOURNAL_1st_RESERVED_BLOCK(sb),
2761				 SB_ONDISK_JOURNAL_SIZE(sb),
2762				 sb->s_blocksize);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2763		goto free_and_return;
2764	}
2765
2766	if (journal_init_dev(sb, journal, j_dev_name) != 0) {
2767		reiserfs_warning(sb, "sh-462",
2768				 "unable to initialize journal device");
2769		goto free_and_return;
2770	}
2771
2772	rs = SB_DISK_SUPER_BLOCK(sb);
2773
2774	/* read journal header */
2775	bhjh = journal_bread(sb,
2776			     SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
2777			     SB_ONDISK_JOURNAL_SIZE(sb));
2778	if (!bhjh) {
2779		reiserfs_warning(sb, "sh-459",
2780				 "unable to read journal header");
2781		goto free_and_return;
2782	}
2783	jh = (struct reiserfs_journal_header *)(bhjh->b_data);
2784
2785	/* make sure that journal matches to the super block */
2786	if (is_reiserfs_jr(rs)
2787	    && (le32_to_cpu(jh->jh_journal.jp_journal_magic) !=
2788		sb_jp_journal_magic(rs))) {
2789		reiserfs_warning(sb, "sh-460",
2790				 "journal header magic %x (device %pg) does "
2791				 "not match to magic found in super block %x",
2792				 jh->jh_journal.jp_journal_magic,
2793				 journal->j_dev_bd,
2794				 sb_jp_journal_magic(rs));
2795		brelse(bhjh);
2796		goto free_and_return;
2797	}
2798
2799	journal->j_trans_max = le32_to_cpu(jh->jh_journal.jp_journal_trans_max);
2800	journal->j_max_batch = le32_to_cpu(jh->jh_journal.jp_journal_max_batch);
2801	journal->j_max_commit_age =
2802	    le32_to_cpu(jh->jh_journal.jp_journal_max_commit_age);
2803	journal->j_max_trans_age = JOURNAL_MAX_TRANS_AGE;
2804
2805	if (check_advise_trans_params(sb, journal) != 0)
2806	        goto free_and_return;
2807	journal->j_default_max_commit_age = journal->j_max_commit_age;
2808
2809	if (commit_max_age != 0) {
2810		journal->j_max_commit_age = commit_max_age;
2811		journal->j_max_trans_age = commit_max_age;
2812	}
2813
2814	reiserfs_info(sb, "journal params: device %pg, size %u, "
2815		      "journal first block %u, max trans len %u, max batch %u, "
2816		      "max commit age %u, max trans age %u\n",
2817		      journal->j_dev_bd,
2818		      SB_ONDISK_JOURNAL_SIZE(sb),
2819		      SB_ONDISK_JOURNAL_1st_BLOCK(sb),
2820		      journal->j_trans_max,
2821		      journal->j_max_batch,
2822		      journal->j_max_commit_age, journal->j_max_trans_age);
2823
2824	brelse(bhjh);
2825
2826	journal->j_list_bitmap_index = 0;
2827	journal_list_init(sb);
2828
2829	memset(journal->j_list_hash_table, 0,
2830	       JOURNAL_HASH_SIZE * sizeof(struct reiserfs_journal_cnode *));
2831
2832	INIT_LIST_HEAD(&journal->j_dirty_buffers);
2833	spin_lock_init(&journal->j_dirty_buffers_lock);
2834
2835	journal->j_start = 0;
2836	journal->j_len = 0;
2837	journal->j_len_alloc = 0;
2838	atomic_set(&journal->j_wcount, 0);
2839	atomic_set(&journal->j_async_throttle, 0);
2840	journal->j_bcount = 0;
2841	journal->j_trans_start_time = 0;
2842	journal->j_last = NULL;
2843	journal->j_first = NULL;
2844	init_waitqueue_head(&journal->j_join_wait);
2845	mutex_init(&journal->j_mutex);
2846	mutex_init(&journal->j_flush_mutex);
2847
2848	journal->j_trans_id = 10;
2849	journal->j_mount_id = 10;
2850	journal->j_state = 0;
2851	atomic_set(&journal->j_jlock, 0);
2852	journal->j_cnode_free_list = allocate_cnodes(num_cnodes);
2853	journal->j_cnode_free_orig = journal->j_cnode_free_list;
2854	journal->j_cnode_free = journal->j_cnode_free_list ? num_cnodes : 0;
2855	journal->j_cnode_used = 0;
2856	journal->j_must_wait = 0;
2857
2858	if (journal->j_cnode_free == 0) {
2859		reiserfs_warning(sb, "journal-2004", "Journal cnode memory "
2860		                 "allocation failed (%ld bytes). Journal is "
2861		                 "too large for available memory. Usually "
2862		                 "this is due to a journal that is too large.",
2863		                 sizeof (struct reiserfs_journal_cnode) * num_cnodes);
2864        	goto free_and_return;
2865	}
2866
2867	init_journal_hash(sb);
2868	jl = journal->j_current_jl;
2869
2870	/*
2871	 * get_list_bitmap() may call flush_commit_list() which
2872	 * requires the lock. Calling flush_commit_list() shouldn't happen
2873	 * this early but I like to be paranoid.
2874	 */
2875	reiserfs_write_lock(sb);
2876	jl->j_list_bitmap = get_list_bitmap(sb, jl);
2877	reiserfs_write_unlock(sb);
2878	if (!jl->j_list_bitmap) {
2879		reiserfs_warning(sb, "journal-2005",
2880				 "get_list_bitmap failed for journal list 0");
2881		goto free_and_return;
2882	}
2883
2884	ret = journal_read(sb);
2885	if (ret < 0) {
2886		reiserfs_warning(sb, "reiserfs-2006",
2887				 "Replay Failure, unable to mount");
2888		goto free_and_return;
2889	}
2890
2891	INIT_DELAYED_WORK(&journal->j_work, flush_async_commits);
2892	journal->j_work_sb = sb;
2893	return 0;
2894free_and_return:
2895	free_journal_ram(sb);
2896	return 1;
2897}
2898
2899/*
2900 * test for a polite end of the current transaction.  Used by file_write,
2901 * and should be used by delete to make sure they don't write more than
2902 * can fit inside a single transaction
2903 */
2904int journal_transaction_should_end(struct reiserfs_transaction_handle *th,
2905				   int new_alloc)
2906{
2907	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2908	time64_t now = ktime_get_seconds();
2909	/* cannot restart while nested */
2910	BUG_ON(!th->t_trans_id);
2911	if (th->t_refcount > 1)
2912		return 0;
2913	if (journal->j_must_wait > 0 ||
2914	    (journal->j_len_alloc + new_alloc) >= journal->j_max_batch ||
2915	    atomic_read(&journal->j_jlock) ||
2916	    (now - journal->j_trans_start_time) > journal->j_max_trans_age ||
2917	    journal->j_cnode_free < (journal->j_trans_max * 3)) {
2918		return 1;
2919	}
2920
2921	journal->j_len_alloc += new_alloc;
2922	th->t_blocks_allocated += new_alloc ;
2923	return 0;
2924}
2925
2926/* this must be called inside a transaction */
2927void reiserfs_block_writes(struct reiserfs_transaction_handle *th)
2928{
2929	struct reiserfs_journal *journal = SB_JOURNAL(th->t_super);
2930	BUG_ON(!th->t_trans_id);
2931	journal->j_must_wait = 1;
2932	set_bit(J_WRITERS_BLOCKED, &journal->j_state);
2933	return;
2934}
2935
2936/* this must be called without a transaction started */
2937void reiserfs_allow_writes(struct super_block *s)
2938{
2939	struct reiserfs_journal *journal = SB_JOURNAL(s);
2940	clear_bit(J_WRITERS_BLOCKED, &journal->j_state);
2941	wake_up(&journal->j_join_wait);
2942}
2943
2944/* this must be called without a transaction started */
2945void reiserfs_wait_on_write_block(struct super_block *s)
2946{
2947	struct reiserfs_journal *journal = SB_JOURNAL(s);
2948	wait_event(journal->j_join_wait,
2949		   !test_bit(J_WRITERS_BLOCKED, &journal->j_state));
2950}
2951
2952static void queue_log_writer(struct super_block *s)
2953{
2954	wait_queue_entry_t wait;
2955	struct reiserfs_journal *journal = SB_JOURNAL(s);
2956	set_bit(J_WRITERS_QUEUED, &journal->j_state);
2957
2958	/*
2959	 * we don't want to use wait_event here because
2960	 * we only want to wait once.
2961	 */
2962	init_waitqueue_entry(&wait, current);
2963	add_wait_queue(&journal->j_join_wait, &wait);
2964	set_current_state(TASK_UNINTERRUPTIBLE);
2965	if (test_bit(J_WRITERS_QUEUED, &journal->j_state)) {
2966		int depth = reiserfs_write_unlock_nested(s);
2967		schedule();
2968		reiserfs_write_lock_nested(s, depth);
2969	}
2970	__set_current_state(TASK_RUNNING);
2971	remove_wait_queue(&journal->j_join_wait, &wait);
2972}
2973
2974static void wake_queued_writers(struct super_block *s)
2975{
2976	struct reiserfs_journal *journal = SB_JOURNAL(s);
2977	if (test_and_clear_bit(J_WRITERS_QUEUED, &journal->j_state))
2978		wake_up(&journal->j_join_wait);
2979}
2980
2981static void let_transaction_grow(struct super_block *sb, unsigned int trans_id)
2982{
2983	struct reiserfs_journal *journal = SB_JOURNAL(sb);
2984	unsigned long bcount = journal->j_bcount;
2985	while (1) {
2986		int depth;
2987
2988		depth = reiserfs_write_unlock_nested(sb);
2989		schedule_timeout_uninterruptible(1);
2990		reiserfs_write_lock_nested(sb, depth);
2991
2992		journal->j_current_jl->j_state |= LIST_COMMIT_PENDING;
2993		while ((atomic_read(&journal->j_wcount) > 0 ||
2994			atomic_read(&journal->j_jlock)) &&
2995		       journal->j_trans_id == trans_id) {
2996			queue_log_writer(sb);
2997		}
2998		if (journal->j_trans_id != trans_id)
2999			break;
3000		if (bcount == journal->j_bcount)
3001			break;
3002		bcount = journal->j_bcount;
3003	}
3004}
3005
3006/*
3007 * join == true if you must join an existing transaction.
3008 * join == false if you can deal with waiting for others to finish
3009 *
3010 * this will block until the transaction is joinable.  send the number of
3011 * blocks you expect to use in nblocks.
3012*/
3013static int do_journal_begin_r(struct reiserfs_transaction_handle *th,
3014			      struct super_block *sb, unsigned long nblocks,
3015			      int join)
3016{
3017	time64_t now = ktime_get_seconds();
3018	unsigned int old_trans_id;
3019	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3020	struct reiserfs_transaction_handle myth;
3021	int sched_count = 0;
3022	int retval;
3023	int depth;
3024
3025	reiserfs_check_lock_depth(sb, "journal_begin");
3026	BUG_ON(nblocks > journal->j_trans_max);
3027
3028	PROC_INFO_INC(sb, journal.journal_being);
3029	/* set here for journal_join */
3030	th->t_refcount = 1;
3031	th->t_super = sb;
3032
3033relock:
3034	lock_journal(sb);
3035	if (join != JBEGIN_ABORT && reiserfs_is_journal_aborted(journal)) {
3036		unlock_journal(sb);
3037		retval = journal->j_errno;
3038		goto out_fail;
3039	}
3040	journal->j_bcount++;
3041
3042	if (test_bit(J_WRITERS_BLOCKED, &journal->j_state)) {
3043		unlock_journal(sb);
3044		depth = reiserfs_write_unlock_nested(sb);
3045		reiserfs_wait_on_write_block(sb);
3046		reiserfs_write_lock_nested(sb, depth);
3047		PROC_INFO_INC(sb, journal.journal_relock_writers);
3048		goto relock;
3049	}
3050	now = ktime_get_seconds();
3051
3052	/*
3053	 * if there is no room in the journal OR
3054	 * if this transaction is too old, and we weren't called joinable,
3055	 * wait for it to finish before beginning we don't sleep if there
3056	 * aren't other writers
3057	 */
3058
3059	if ((!join && journal->j_must_wait > 0) ||
3060	    (!join
3061	     && (journal->j_len_alloc + nblocks + 2) >= journal->j_max_batch)
3062	    || (!join && atomic_read(&journal->j_wcount) > 0
3063		&& journal->j_trans_start_time > 0
3064		&& (now - journal->j_trans_start_time) >
3065		journal->j_max_trans_age) || (!join
3066					      && atomic_read(&journal->j_jlock))
3067	    || (!join && journal->j_cnode_free < (journal->j_trans_max * 3))) {
3068
3069		old_trans_id = journal->j_trans_id;
3070		/* allow others to finish this transaction */
3071		unlock_journal(sb);
3072
3073		if (!join && (journal->j_len_alloc + nblocks + 2) >=
3074		    journal->j_max_batch &&
3075		    ((journal->j_len + nblocks + 2) * 100) <
3076		    (journal->j_len_alloc * 75)) {
3077			if (atomic_read(&journal->j_wcount) > 10) {
3078				sched_count++;
3079				queue_log_writer(sb);
3080				goto relock;
3081			}
3082		}
3083		/*
3084		 * don't mess with joining the transaction if all we
3085		 * have to do is wait for someone else to do a commit
3086		 */
3087		if (atomic_read(&journal->j_jlock)) {
3088			while (journal->j_trans_id == old_trans_id &&
3089			       atomic_read(&journal->j_jlock)) {
3090				queue_log_writer(sb);
3091			}
3092			goto relock;
3093		}
3094		retval = journal_join(&myth, sb);
3095		if (retval)
3096			goto out_fail;
3097
3098		/* someone might have ended the transaction while we joined */
3099		if (old_trans_id != journal->j_trans_id) {
3100			retval = do_journal_end(&myth, 0);
3101		} else {
3102			retval = do_journal_end(&myth, COMMIT_NOW);
3103		}
3104
3105		if (retval)
3106			goto out_fail;
3107
3108		PROC_INFO_INC(sb, journal.journal_relock_wcount);
3109		goto relock;
3110	}
3111	/* we are the first writer, set trans_id */
3112	if (journal->j_trans_start_time == 0) {
3113		journal->j_trans_start_time = ktime_get_seconds();
3114	}
3115	atomic_inc(&journal->j_wcount);
3116	journal->j_len_alloc += nblocks;
3117	th->t_blocks_logged = 0;
3118	th->t_blocks_allocated = nblocks;
3119	th->t_trans_id = journal->j_trans_id;
3120	unlock_journal(sb);
3121	INIT_LIST_HEAD(&th->t_list);
3122	return 0;
3123
3124out_fail:
3125	memset(th, 0, sizeof(*th));
3126	/*
3127	 * Re-set th->t_super, so we can properly keep track of how many
3128	 * persistent transactions there are. We need to do this so if this
3129	 * call is part of a failed restart_transaction, we can free it later
3130	 */
3131	th->t_super = sb;
3132	return retval;
3133}
3134
3135struct reiserfs_transaction_handle *reiserfs_persistent_transaction(struct
3136								    super_block
3137								    *s,
3138								    int nblocks)
3139{
3140	int ret;
3141	struct reiserfs_transaction_handle *th;
3142
3143	/*
3144	 * if we're nesting into an existing transaction.  It will be
3145	 * persistent on its own
3146	 */
3147	if (reiserfs_transaction_running(s)) {
3148		th = current->journal_info;
3149		th->t_refcount++;
3150		BUG_ON(th->t_refcount < 2);
3151
3152		return th;
3153	}
3154	th = kmalloc(sizeof(struct reiserfs_transaction_handle), GFP_NOFS);
3155	if (!th)
3156		return NULL;
3157	ret = journal_begin(th, s, nblocks);
3158	if (ret) {
3159		kfree(th);
3160		return NULL;
3161	}
3162
3163	SB_JOURNAL(s)->j_persistent_trans++;
3164	return th;
3165}
3166
3167int reiserfs_end_persistent_transaction(struct reiserfs_transaction_handle *th)
3168{
3169	struct super_block *s = th->t_super;
3170	int ret = 0;
3171	if (th->t_trans_id)
3172		ret = journal_end(th);
3173	else
3174		ret = -EIO;
3175	if (th->t_refcount == 0) {
3176		SB_JOURNAL(s)->j_persistent_trans--;
3177		kfree(th);
3178	}
3179	return ret;
3180}
3181
3182static int journal_join(struct reiserfs_transaction_handle *th,
3183			struct super_block *sb)
3184{
3185	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3186
3187	/*
3188	 * this keeps do_journal_end from NULLing out the
3189	 * current->journal_info pointer
3190	 */
3191	th->t_handle_save = cur_th;
3192	BUG_ON(cur_th && cur_th->t_refcount > 1);
3193	return do_journal_begin_r(th, sb, 1, JBEGIN_JOIN);
3194}
3195
3196int journal_join_abort(struct reiserfs_transaction_handle *th,
3197		       struct super_block *sb)
3198{
3199	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3200
3201	/*
3202	 * this keeps do_journal_end from NULLing out the
3203	 * current->journal_info pointer
3204	 */
3205	th->t_handle_save = cur_th;
3206	BUG_ON(cur_th && cur_th->t_refcount > 1);
3207	return do_journal_begin_r(th, sb, 1, JBEGIN_ABORT);
3208}
3209
3210int journal_begin(struct reiserfs_transaction_handle *th,
3211		  struct super_block *sb, unsigned long nblocks)
3212{
3213	struct reiserfs_transaction_handle *cur_th = current->journal_info;
3214	int ret;
3215
3216	th->t_handle_save = NULL;
3217	if (cur_th) {
3218		/* we are nesting into the current transaction */
3219		if (cur_th->t_super == sb) {
3220			BUG_ON(!cur_th->t_refcount);
3221			cur_th->t_refcount++;
3222			memcpy(th, cur_th, sizeof(*th));
3223			if (th->t_refcount <= 1)
3224				reiserfs_warning(sb, "reiserfs-2005",
3225						 "BAD: refcount <= 1, but "
3226						 "journal_info != 0");
3227			return 0;
3228		} else {
3229			/*
3230			 * we've ended up with a handle from a different
3231			 * filesystem.  save it and restore on journal_end.
3232			 * This should never really happen...
3233			 */
3234			reiserfs_warning(sb, "clm-2100",
3235					 "nesting info a different FS");
3236			th->t_handle_save = current->journal_info;
3237			current->journal_info = th;
3238		}
3239	} else {
3240		current->journal_info = th;
3241	}
3242	ret = do_journal_begin_r(th, sb, nblocks, JBEGIN_REG);
3243	BUG_ON(current->journal_info != th);
3244
3245	/*
3246	 * I guess this boils down to being the reciprocal of clm-2100 above.
3247	 * If do_journal_begin_r fails, we need to put it back, since
3248	 * journal_end won't be called to do it. */
3249	if (ret)
3250		current->journal_info = th->t_handle_save;
3251	else
3252		BUG_ON(!th->t_refcount);
3253
3254	return ret;
3255}
3256
3257/*
3258 * puts bh into the current transaction.  If it was already there, reorders
3259 * removes the old pointers from the hash, and puts new ones in (to make
3260 * sure replay happen in the right order).
3261 *
3262 * if it was dirty, cleans and files onto the clean list.  I can't let it
3263 * be dirty again until the transaction is committed.
3264 *
3265 * if j_len, is bigger than j_len_alloc, it pushes j_len_alloc to 10 + j_len.
3266 */
3267int journal_mark_dirty(struct reiserfs_transaction_handle *th,
3268		       struct buffer_head *bh)
3269{
3270	struct super_block *sb = th->t_super;
3271	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3272	struct reiserfs_journal_cnode *cn = NULL;
3273	int count_already_incd = 0;
3274	int prepared = 0;
3275	BUG_ON(!th->t_trans_id);
3276
3277	PROC_INFO_INC(sb, journal.mark_dirty);
3278	if (th->t_trans_id != journal->j_trans_id) {
3279		reiserfs_panic(th->t_super, "journal-1577",
3280			       "handle trans id %ld != current trans id %ld",
3281			       th->t_trans_id, journal->j_trans_id);
3282	}
3283
3284	prepared = test_clear_buffer_journal_prepared(bh);
3285	clear_buffer_journal_restore_dirty(bh);
3286	/* already in this transaction, we are done */
3287	if (buffer_journaled(bh)) {
3288		PROC_INFO_INC(sb, journal.mark_dirty_already);
3289		return 0;
3290	}
3291
3292	/*
3293	 * this must be turned into a panic instead of a warning.  We can't
3294	 * allow a dirty or journal_dirty or locked buffer to be logged, as
3295	 * some changes could get to disk too early.  NOT GOOD.
3296	 */
3297	if (!prepared || buffer_dirty(bh)) {
3298		reiserfs_warning(sb, "journal-1777",
3299				 "buffer %llu bad state "
3300				 "%cPREPARED %cLOCKED %cDIRTY %cJDIRTY_WAIT",
3301				 (unsigned long long)bh->b_blocknr,
3302				 prepared ? ' ' : '!',
3303				 buffer_locked(bh) ? ' ' : '!',
3304				 buffer_dirty(bh) ? ' ' : '!',
3305				 buffer_journal_dirty(bh) ? ' ' : '!');
3306	}
3307
3308	if (atomic_read(&journal->j_wcount) <= 0) {
3309		reiserfs_warning(sb, "journal-1409",
3310				 "returning because j_wcount was %d",
3311				 atomic_read(&journal->j_wcount));
3312		return 1;
3313	}
3314	/*
3315	 * this error means I've screwed up, and we've overflowed
3316	 * the transaction.  Nothing can be done here, except make the
3317	 * FS readonly or panic.
3318	 */
3319	if (journal->j_len >= journal->j_trans_max) {
3320		reiserfs_panic(th->t_super, "journal-1413",
3321			       "j_len (%lu) is too big",
3322			       journal->j_len);
3323	}
3324
3325	if (buffer_journal_dirty(bh)) {
3326		count_already_incd = 1;
3327		PROC_INFO_INC(sb, journal.mark_dirty_notjournal);
3328		clear_buffer_journal_dirty(bh);
3329	}
3330
3331	if (journal->j_len > journal->j_len_alloc) {
3332		journal->j_len_alloc = journal->j_len + JOURNAL_PER_BALANCE_CNT;
3333	}
3334
3335	set_buffer_journaled(bh);
3336
3337	/* now put this guy on the end */
3338	if (!cn) {
3339		cn = get_cnode(sb);
3340		if (!cn) {
3341			reiserfs_panic(sb, "journal-4", "get_cnode failed!");
3342		}
3343
3344		if (th->t_blocks_logged == th->t_blocks_allocated) {
3345			th->t_blocks_allocated += JOURNAL_PER_BALANCE_CNT;
3346			journal->j_len_alloc += JOURNAL_PER_BALANCE_CNT;
3347		}
3348		th->t_blocks_logged++;
3349		journal->j_len++;
3350
3351		cn->bh = bh;
3352		cn->blocknr = bh->b_blocknr;
3353		cn->sb = sb;
3354		cn->jlist = NULL;
3355		insert_journal_hash(journal->j_hash_table, cn);
3356		if (!count_already_incd) {
3357			get_bh(bh);
3358		}
3359	}
3360	cn->next = NULL;
3361	cn->prev = journal->j_last;
3362	cn->bh = bh;
3363	if (journal->j_last) {
3364		journal->j_last->next = cn;
3365		journal->j_last = cn;
3366	} else {
3367		journal->j_first = cn;
3368		journal->j_last = cn;
3369	}
3370	reiserfs_schedule_old_flush(sb);
3371	return 0;
3372}
3373
3374int journal_end(struct reiserfs_transaction_handle *th)
3375{
3376	struct super_block *sb = th->t_super;
3377	if (!current->journal_info && th->t_refcount > 1)
3378		reiserfs_warning(sb, "REISER-NESTING",
3379				 "th NULL, refcount %d", th->t_refcount);
3380
3381	if (!th->t_trans_id) {
3382		WARN_ON(1);
3383		return -EIO;
3384	}
3385
3386	th->t_refcount--;
3387	if (th->t_refcount > 0) {
3388		struct reiserfs_transaction_handle *cur_th =
3389		    current->journal_info;
3390
3391		/*
3392		 * we aren't allowed to close a nested transaction on a
3393		 * different filesystem from the one in the task struct
3394		 */
3395		BUG_ON(cur_th->t_super != th->t_super);
3396
3397		if (th != cur_th) {
3398			memcpy(current->journal_info, th, sizeof(*th));
3399			th->t_trans_id = 0;
3400		}
3401		return 0;
3402	} else {
3403		return do_journal_end(th, 0);
3404	}
3405}
3406
3407/*
3408 * removes from the current transaction, relsing and descrementing any counters.
3409 * also files the removed buffer directly onto the clean list
3410 *
3411 * called by journal_mark_freed when a block has been deleted
3412 *
3413 * returns 1 if it cleaned and relsed the buffer. 0 otherwise
3414 */
3415static int remove_from_transaction(struct super_block *sb,
3416				   b_blocknr_t blocknr, int already_cleaned)
3417{
3418	struct buffer_head *bh;
3419	struct reiserfs_journal_cnode *cn;
3420	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3421	int ret = 0;
3422
3423	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3424	if (!cn || !cn->bh) {
3425		return ret;
3426	}
3427	bh = cn->bh;
3428	if (cn->prev) {
3429		cn->prev->next = cn->next;
3430	}
3431	if (cn->next) {
3432		cn->next->prev = cn->prev;
3433	}
3434	if (cn == journal->j_first) {
3435		journal->j_first = cn->next;
3436	}
3437	if (cn == journal->j_last) {
3438		journal->j_last = cn->prev;
3439	}
3440	remove_journal_hash(sb, journal->j_hash_table, NULL,
3441			    bh->b_blocknr, 0);
3442	clear_buffer_journaled(bh);	/* don't log this one */
3443
3444	if (!already_cleaned) {
3445		clear_buffer_journal_dirty(bh);
3446		clear_buffer_dirty(bh);
3447		clear_buffer_journal_test(bh);
3448		put_bh(bh);
3449		if (atomic_read(&bh->b_count) < 0) {
3450			reiserfs_warning(sb, "journal-1752",
3451					 "b_count < 0");
3452		}
3453		ret = 1;
3454	}
3455	journal->j_len--;
3456	journal->j_len_alloc--;
3457	free_cnode(sb, cn);
3458	return ret;
3459}
3460
3461/*
3462 * for any cnode in a journal list, it can only be dirtied of all the
3463 * transactions that include it are committed to disk.
3464 * this checks through each transaction, and returns 1 if you are allowed
3465 * to dirty, and 0 if you aren't
3466 *
3467 * it is called by dirty_journal_list, which is called after
3468 * flush_commit_list has gotten all the log blocks for a given
3469 * transaction on disk
3470 *
3471 */
3472static int can_dirty(struct reiserfs_journal_cnode *cn)
3473{
3474	struct super_block *sb = cn->sb;
3475	b_blocknr_t blocknr = cn->blocknr;
3476	struct reiserfs_journal_cnode *cur = cn->hprev;
3477	int can_dirty = 1;
3478
3479	/*
3480	 * first test hprev.  These are all newer than cn, so any node here
3481	 * with the same block number and dev means this node can't be sent
3482	 * to disk right now.
3483	 */
3484	while (cur && can_dirty) {
3485		if (cur->jlist && cur->bh && cur->blocknr && cur->sb == sb &&
3486		    cur->blocknr == blocknr) {
3487			can_dirty = 0;
3488		}
3489		cur = cur->hprev;
3490	}
3491	/*
3492	 * then test hnext.  These are all older than cn.  As long as they
3493	 * are committed to the log, it is safe to write cn to disk
3494	 */
3495	cur = cn->hnext;
3496	while (cur && can_dirty) {
3497		if (cur->jlist && cur->jlist->j_len > 0 &&
3498		    atomic_read(&cur->jlist->j_commit_left) > 0 && cur->bh &&
3499		    cur->blocknr && cur->sb == sb && cur->blocknr == blocknr) {
3500			can_dirty = 0;
3501		}
3502		cur = cur->hnext;
3503	}
3504	return can_dirty;
3505}
3506
3507/*
3508 * syncs the commit blocks, but does not force the real buffers to disk
3509 * will wait until the current transaction is done/committed before returning
3510 */
3511int journal_end_sync(struct reiserfs_transaction_handle *th)
3512{
3513	struct super_block *sb = th->t_super;
3514	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3515
3516	BUG_ON(!th->t_trans_id);
3517	/* you can sync while nested, very, very bad */
3518	BUG_ON(th->t_refcount > 1);
3519	if (journal->j_len == 0) {
3520		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3521					     1);
3522		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
3523	}
3524	return do_journal_end(th, COMMIT_NOW | WAIT);
3525}
3526
3527/* writeback the pending async commits to disk */
3528static void flush_async_commits(struct work_struct *work)
3529{
3530	struct reiserfs_journal *journal =
3531		container_of(work, struct reiserfs_journal, j_work.work);
3532	struct super_block *sb = journal->j_work_sb;
3533	struct reiserfs_journal_list *jl;
3534	struct list_head *entry;
3535
3536	reiserfs_write_lock(sb);
3537	if (!list_empty(&journal->j_journal_list)) {
3538		/* last entry is the youngest, commit it and you get everything */
3539		entry = journal->j_journal_list.prev;
3540		jl = JOURNAL_LIST_ENTRY(entry);
3541		flush_commit_list(sb, jl, 1);
3542	}
3543	reiserfs_write_unlock(sb);
3544}
3545
3546/*
3547 * flushes any old transactions to disk
3548 * ends the current transaction if it is too old
3549 */
3550void reiserfs_flush_old_commits(struct super_block *sb)
3551{
3552	time64_t now;
3553	struct reiserfs_transaction_handle th;
3554	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3555
3556	now = ktime_get_seconds();
3557	/*
3558	 * safety check so we don't flush while we are replaying the log during
3559	 * mount
3560	 */
3561	if (list_empty(&journal->j_journal_list))
3562		return;
3563
3564	/*
3565	 * check the current transaction.  If there are no writers, and it is
3566	 * too old, finish it, and force the commit blocks to disk
3567	 */
3568	if (atomic_read(&journal->j_wcount) <= 0 &&
3569	    journal->j_trans_start_time > 0 &&
3570	    journal->j_len > 0 &&
3571	    (now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3572		if (!journal_join(&th, sb)) {
3573			reiserfs_prepare_for_journal(sb,
3574						     SB_BUFFER_WITH_SB(sb),
3575						     1);
3576			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3577
3578			/*
3579			 * we're only being called from kreiserfsd, it makes
3580			 * no sense to do an async commit so that kreiserfsd
3581			 * can do it later
3582			 */
3583			do_journal_end(&th, COMMIT_NOW | WAIT);
3584		}
3585	}
3586}
3587
3588/*
3589 * returns 0 if do_journal_end should return right away, returns 1 if
3590 * do_journal_end should finish the commit
3591 *
3592 * if the current transaction is too old, but still has writers, this will
3593 * wait on j_join_wait until all the writers are done.  By the time it
3594 * wakes up, the transaction it was called has already ended, so it just
3595 * flushes the commit list and returns 0.
3596 *
3597 * Won't batch when flush or commit_now is set.  Also won't batch when
3598 * others are waiting on j_join_wait.
3599 *
3600 * Note, we can't allow the journal_end to proceed while there are still
3601 * writers in the log.
3602 */
3603static int check_journal_end(struct reiserfs_transaction_handle *th, int flags)
3604{
3605
3606	time64_t now;
3607	int flush = flags & FLUSH_ALL;
3608	int commit_now = flags & COMMIT_NOW;
3609	int wait_on_commit = flags & WAIT;
3610	struct reiserfs_journal_list *jl;
3611	struct super_block *sb = th->t_super;
3612	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3613
3614	BUG_ON(!th->t_trans_id);
3615
3616	if (th->t_trans_id != journal->j_trans_id) {
3617		reiserfs_panic(th->t_super, "journal-1577",
3618			       "handle trans id %ld != current trans id %ld",
3619			       th->t_trans_id, journal->j_trans_id);
3620	}
3621
3622	journal->j_len_alloc -= (th->t_blocks_allocated - th->t_blocks_logged);
3623	/* <= 0 is allowed.  unmounting might not call begin */
3624	if (atomic_read(&journal->j_wcount) > 0)
3625		atomic_dec(&journal->j_wcount);
3626
3627	/*
3628	 * BUG, deal with case where j_len is 0, but people previously
3629	 * freed blocks need to be released will be dealt with by next
3630	 * transaction that actually writes something, but should be taken
3631	 * care of in this trans
3632	 */
3633	BUG_ON(journal->j_len == 0);
3634
3635	/*
3636	 * if wcount > 0, and we are called to with flush or commit_now,
3637	 * we wait on j_join_wait.  We will wake up when the last writer has
3638	 * finished the transaction, and started it on its way to the disk.
3639	 * Then, we flush the commit or journal list, and just return 0
3640	 * because the rest of journal end was already done for this
3641	 * transaction.
3642	 */
3643	if (atomic_read(&journal->j_wcount) > 0) {
3644		if (flush || commit_now) {
3645			unsigned trans_id;
3646
3647			jl = journal->j_current_jl;
3648			trans_id = jl->j_trans_id;
3649			if (wait_on_commit)
3650				jl->j_state |= LIST_COMMIT_PENDING;
3651			atomic_set(&journal->j_jlock, 1);
3652			if (flush) {
3653				journal->j_next_full_flush = 1;
3654			}
3655			unlock_journal(sb);
3656
3657			/*
3658			 * sleep while the current transaction is
3659			 * still j_jlocked
3660			 */
3661			while (journal->j_trans_id == trans_id) {
3662				if (atomic_read(&journal->j_jlock)) {
3663					queue_log_writer(sb);
3664				} else {
3665					lock_journal(sb);
3666					if (journal->j_trans_id == trans_id) {
3667						atomic_set(&journal->j_jlock,
3668							   1);
3669					}
3670					unlock_journal(sb);
3671				}
3672			}
3673			BUG_ON(journal->j_trans_id == trans_id);
3674
3675			if (commit_now
3676			    && journal_list_still_alive(sb, trans_id)
3677			    && wait_on_commit) {
3678				flush_commit_list(sb, jl, 1);
3679			}
3680			return 0;
3681		}
3682		unlock_journal(sb);
3683		return 0;
3684	}
3685
3686	/* deal with old transactions where we are the last writers */
3687	now = ktime_get_seconds();
3688	if ((now - journal->j_trans_start_time) > journal->j_max_trans_age) {
3689		commit_now = 1;
3690		journal->j_next_async_flush = 1;
3691	}
3692	/* don't batch when someone is waiting on j_join_wait */
3693	/* don't batch when syncing the commit or flushing the whole trans */
3694	if (!(journal->j_must_wait > 0) && !(atomic_read(&journal->j_jlock))
3695	    && !flush && !commit_now && (journal->j_len < journal->j_max_batch)
3696	    && journal->j_len_alloc < journal->j_max_batch
3697	    && journal->j_cnode_free > (journal->j_trans_max * 3)) {
3698		journal->j_bcount++;
3699		unlock_journal(sb);
3700		return 0;
3701	}
3702
3703	if (journal->j_start > SB_ONDISK_JOURNAL_SIZE(sb)) {
3704		reiserfs_panic(sb, "journal-003",
3705			       "j_start (%ld) is too high",
3706			       journal->j_start);
3707	}
3708	return 1;
3709}
3710
3711/*
3712 * Does all the work that makes deleting blocks safe.
3713 * when deleting a block mark BH_JNew, just remove it from the current
3714 * transaction, clean it's buffer_head and move on.
3715 *
3716 * otherwise:
3717 * set a bit for the block in the journal bitmap.  That will prevent it from
3718 * being allocated for unformatted nodes before this transaction has finished.
3719 *
3720 * mark any cnodes for this block as BLOCK_FREED, and clear their bh pointers.
3721 * That will prevent any old transactions with this block from trying to flush
3722 * to the real location.  Since we aren't removing the cnode from the
3723 * journal_list_hash, *the block can't be reallocated yet.
3724 *
3725 * Then remove it from the current transaction, decrementing any counters and
3726 * filing it on the clean list.
3727 */
3728int journal_mark_freed(struct reiserfs_transaction_handle *th,
3729		       struct super_block *sb, b_blocknr_t blocknr)
3730{
3731	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3732	struct reiserfs_journal_cnode *cn = NULL;
3733	struct buffer_head *bh = NULL;
3734	struct reiserfs_list_bitmap *jb = NULL;
3735	int cleaned = 0;
3736	BUG_ON(!th->t_trans_id);
3737
3738	cn = get_journal_hash_dev(sb, journal->j_hash_table, blocknr);
3739	if (cn && cn->bh) {
3740		bh = cn->bh;
3741		get_bh(bh);
3742	}
3743	/* if it is journal new, we just remove it from this transaction */
3744	if (bh && buffer_journal_new(bh)) {
3745		clear_buffer_journal_new(bh);
3746		clear_prepared_bits(bh);
3747		reiserfs_clean_and_file_buffer(bh);
3748		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3749	} else {
3750		/*
3751		 * set the bit for this block in the journal bitmap
3752		 * for this transaction
3753		 */
3754		jb = journal->j_current_jl->j_list_bitmap;
3755		if (!jb) {
3756			reiserfs_panic(sb, "journal-1702",
3757				       "journal_list_bitmap is NULL");
3758		}
3759		set_bit_in_list_bitmap(sb, blocknr, jb);
3760
3761		/* Note, the entire while loop is not allowed to schedule.  */
3762
3763		if (bh) {
3764			clear_prepared_bits(bh);
3765			reiserfs_clean_and_file_buffer(bh);
3766		}
3767		cleaned = remove_from_transaction(sb, blocknr, cleaned);
3768
3769		/*
3770		 * find all older transactions with this block,
3771		 * make sure they don't try to write it out
3772		 */
3773		cn = get_journal_hash_dev(sb, journal->j_list_hash_table,
3774					  blocknr);
3775		while (cn) {
3776			if (sb == cn->sb && blocknr == cn->blocknr) {
3777				set_bit(BLOCK_FREED, &cn->state);
3778				if (cn->bh) {
3779					/*
3780					 * remove_from_transaction will brelse
3781					 * the buffer if it was in the current
3782					 * trans
3783					 */
3784					if (!cleaned) {
3785						clear_buffer_journal_dirty(cn->
3786									   bh);
3787						clear_buffer_dirty(cn->bh);
3788						clear_buffer_journal_test(cn->
3789									  bh);
3790						cleaned = 1;
3791						put_bh(cn->bh);
3792						if (atomic_read
3793						    (&cn->bh->b_count) < 0) {
3794							reiserfs_warning(sb,
3795								 "journal-2138",
3796								 "cn->bh->b_count < 0");
3797						}
3798					}
3799					/*
3800					 * since we are clearing the bh,
3801					 * we MUST dec nonzerolen
3802					 */
3803					if (cn->jlist) {
3804						atomic_dec(&cn->jlist->
3805							   j_nonzerolen);
3806					}
3807					cn->bh = NULL;
3808				}
3809			}
3810			cn = cn->hnext;
3811		}
3812	}
3813
3814	if (bh)
3815		release_buffer_page(bh); /* get_hash grabs the buffer */
3816	return 0;
3817}
3818
3819void reiserfs_update_inode_transaction(struct inode *inode)
3820{
3821	struct reiserfs_journal *journal = SB_JOURNAL(inode->i_sb);
3822	REISERFS_I(inode)->i_jl = journal->j_current_jl;
3823	REISERFS_I(inode)->i_trans_id = journal->j_trans_id;
3824}
3825
3826/*
3827 * returns -1 on error, 0 if no commits/barriers were done and 1
3828 * if a transaction was actually committed and the barrier was done
3829 */
3830static int __commit_trans_jl(struct inode *inode, unsigned long id,
3831			     struct reiserfs_journal_list *jl)
3832{
3833	struct reiserfs_transaction_handle th;
3834	struct super_block *sb = inode->i_sb;
3835	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3836	int ret = 0;
3837
3838	/*
3839	 * is it from the current transaction,
3840	 * or from an unknown transaction?
3841	 */
3842	if (id == journal->j_trans_id) {
3843		jl = journal->j_current_jl;
3844		/*
3845		 * try to let other writers come in and
3846		 * grow this transaction
3847		 */
3848		let_transaction_grow(sb, id);
3849		if (journal->j_trans_id != id) {
3850			goto flush_commit_only;
3851		}
3852
3853		ret = journal_begin(&th, sb, 1);
3854		if (ret)
3855			return ret;
3856
3857		/* someone might have ended this transaction while we joined */
3858		if (journal->j_trans_id != id) {
3859			reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
3860						     1);
3861			journal_mark_dirty(&th, SB_BUFFER_WITH_SB(sb));
3862			ret = journal_end(&th);
3863			goto flush_commit_only;
3864		}
3865
3866		ret = journal_end_sync(&th);
3867		if (!ret)
3868			ret = 1;
3869
3870	} else {
3871		/*
3872		 * this gets tricky, we have to make sure the journal list in
3873		 * the inode still exists.  We know the list is still around
3874		 * if we've got a larger transaction id than the oldest list
3875		 */
3876flush_commit_only:
3877		if (journal_list_still_alive(inode->i_sb, id)) {
3878			/*
3879			 * we only set ret to 1 when we know for sure
3880			 * the barrier hasn't been started yet on the commit
3881			 * block.
3882			 */
3883			if (atomic_read(&jl->j_commit_left) > 1)
3884				ret = 1;
3885			flush_commit_list(sb, jl, 1);
3886			if (journal->j_errno)
3887				ret = journal->j_errno;
3888		}
3889	}
3890	/* otherwise the list is gone, and long since committed */
3891	return ret;
3892}
3893
3894int reiserfs_commit_for_inode(struct inode *inode)
3895{
3896	unsigned int id = REISERFS_I(inode)->i_trans_id;
3897	struct reiserfs_journal_list *jl = REISERFS_I(inode)->i_jl;
3898
3899	/*
3900	 * for the whole inode, assume unset id means it was
3901	 * changed in the current transaction.  More conservative
3902	 */
3903	if (!id || !jl) {
3904		reiserfs_update_inode_transaction(inode);
3905		id = REISERFS_I(inode)->i_trans_id;
3906		/* jl will be updated in __commit_trans_jl */
3907	}
3908
3909	return __commit_trans_jl(inode, id, jl);
3910}
3911
3912void reiserfs_restore_prepared_buffer(struct super_block *sb,
3913				      struct buffer_head *bh)
3914{
3915	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3916	PROC_INFO_INC(sb, journal.restore_prepared);
3917	if (!bh) {
3918		return;
3919	}
3920	if (test_clear_buffer_journal_restore_dirty(bh) &&
3921	    buffer_journal_dirty(bh)) {
3922		struct reiserfs_journal_cnode *cn;
3923		reiserfs_write_lock(sb);
3924		cn = get_journal_hash_dev(sb,
3925					  journal->j_list_hash_table,
3926					  bh->b_blocknr);
3927		if (cn && can_dirty(cn)) {
3928			set_buffer_journal_test(bh);
3929			mark_buffer_dirty(bh);
3930		}
3931		reiserfs_write_unlock(sb);
3932	}
3933	clear_buffer_journal_prepared(bh);
3934}
3935
3936extern struct tree_balance *cur_tb;
3937/*
3938 * before we can change a metadata block, we have to make sure it won't
3939 * be written to disk while we are altering it.  So, we must:
3940 * clean it
3941 * wait on it.
3942 */
3943int reiserfs_prepare_for_journal(struct super_block *sb,
3944				 struct buffer_head *bh, int wait)
3945{
3946	PROC_INFO_INC(sb, journal.prepare);
3947
3948	if (!trylock_buffer(bh)) {
3949		if (!wait)
3950			return 0;
3951		lock_buffer(bh);
3952	}
3953	set_buffer_journal_prepared(bh);
3954	if (test_clear_buffer_dirty(bh) && buffer_journal_dirty(bh)) {
3955		clear_buffer_journal_test(bh);
3956		set_buffer_journal_restore_dirty(bh);
3957	}
3958	unlock_buffer(bh);
3959	return 1;
3960}
3961
3962/*
3963 * long and ugly.  If flush, will not return until all commit
3964 * blocks and all real buffers in the trans are on disk.
3965 * If no_async, won't return until all commit blocks are on disk.
3966 *
3967 * keep reading, there are comments as you go along
3968 *
3969 * If the journal is aborted, we just clean up. Things like flushing
3970 * journal lists, etc just won't happen.
3971 */
3972static int do_journal_end(struct reiserfs_transaction_handle *th, int flags)
3973{
3974	struct super_block *sb = th->t_super;
3975	struct reiserfs_journal *journal = SB_JOURNAL(sb);
3976	struct reiserfs_journal_cnode *cn, *next, *jl_cn;
3977	struct reiserfs_journal_cnode *last_cn = NULL;
3978	struct reiserfs_journal_desc *desc;
3979	struct reiserfs_journal_commit *commit;
3980	struct buffer_head *c_bh;	/* commit bh */
3981	struct buffer_head *d_bh;	/* desc bh */
3982	int cur_write_start = 0;	/* start index of current log write */
3983	int i;
3984	int flush;
3985	int wait_on_commit;
3986	struct reiserfs_journal_list *jl, *temp_jl;
3987	struct list_head *entry, *safe;
3988	unsigned long jindex;
3989	unsigned int commit_trans_id;
3990	int trans_half;
3991	int depth;
3992
3993	BUG_ON(th->t_refcount > 1);
3994	BUG_ON(!th->t_trans_id);
3995	BUG_ON(!th->t_super);
3996
3997	/*
3998	 * protect flush_older_commits from doing mistakes if the
3999	 * transaction ID counter gets overflowed.
4000	 */
4001	if (th->t_trans_id == ~0U)
4002		flags |= FLUSH_ALL | COMMIT_NOW | WAIT;
4003	flush = flags & FLUSH_ALL;
4004	wait_on_commit = flags & WAIT;
4005
4006	current->journal_info = th->t_handle_save;
4007	reiserfs_check_lock_depth(sb, "journal end");
4008	if (journal->j_len == 0) {
4009		reiserfs_prepare_for_journal(sb, SB_BUFFER_WITH_SB(sb),
4010					     1);
4011		journal_mark_dirty(th, SB_BUFFER_WITH_SB(sb));
4012	}
4013
4014	lock_journal(sb);
4015	if (journal->j_next_full_flush) {
4016		flags |= FLUSH_ALL;
4017		flush = 1;
4018	}
4019	if (journal->j_next_async_flush) {
4020		flags |= COMMIT_NOW | WAIT;
4021		wait_on_commit = 1;
4022	}
4023
4024	/*
4025	 * check_journal_end locks the journal, and unlocks if it does
4026	 * not return 1 it tells us if we should continue with the
4027	 * journal_end, or just return
4028	 */
4029	if (!check_journal_end(th, flags)) {
4030		reiserfs_schedule_old_flush(sb);
4031		wake_queued_writers(sb);
4032		reiserfs_async_progress_wait(sb);
4033		goto out;
4034	}
4035
4036	/* check_journal_end might set these, check again */
4037	if (journal->j_next_full_flush) {
4038		flush = 1;
4039	}
4040
4041	/*
4042	 * j must wait means we have to flush the log blocks, and the
4043	 * real blocks for this transaction
4044	 */
4045	if (journal->j_must_wait > 0) {
4046		flush = 1;
4047	}
4048#ifdef REISERFS_PREALLOCATE
4049	/*
4050	 * quota ops might need to nest, setup the journal_info pointer
4051	 * for them and raise the refcount so that it is > 0.
4052	 */
4053	current->journal_info = th;
4054	th->t_refcount++;
4055
4056	/* it should not involve new blocks into the transaction */
4057	reiserfs_discard_all_prealloc(th);
4058
4059	th->t_refcount--;
4060	current->journal_info = th->t_handle_save;
4061#endif
4062
4063	/* setup description block */
4064	d_bh =
4065	    journal_getblk(sb,
4066			   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4067			   journal->j_start);
4068	set_buffer_uptodate(d_bh);
4069	desc = (struct reiserfs_journal_desc *)(d_bh)->b_data;
4070	memset(d_bh->b_data, 0, d_bh->b_size);
4071	memcpy(get_journal_desc_magic(d_bh), JOURNAL_DESC_MAGIC, 8);
4072	set_desc_trans_id(desc, journal->j_trans_id);
4073
4074	/*
4075	 * setup commit block.  Don't write (keep it clean too) this one
4076	 * until after everyone else is written
4077	 */
4078	c_bh = journal_getblk(sb, SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4079			      ((journal->j_start + journal->j_len +
4080				1) % SB_ONDISK_JOURNAL_SIZE(sb)));
4081	commit = (struct reiserfs_journal_commit *)c_bh->b_data;
4082	memset(c_bh->b_data, 0, c_bh->b_size);
4083	set_commit_trans_id(commit, journal->j_trans_id);
4084	set_buffer_uptodate(c_bh);
4085
4086	/* init this journal list */
4087	jl = journal->j_current_jl;
4088
4089	/*
4090	 * we lock the commit before doing anything because
4091	 * we want to make sure nobody tries to run flush_commit_list until
4092	 * the new transaction is fully setup, and we've already flushed the
4093	 * ordered bh list
4094	 */
4095	reiserfs_mutex_lock_safe(&jl->j_commit_mutex, sb);
4096
4097	/* save the transaction id in case we need to commit it later */
4098	commit_trans_id = jl->j_trans_id;
4099
4100	atomic_set(&jl->j_older_commits_done, 0);
4101	jl->j_trans_id = journal->j_trans_id;
4102	jl->j_timestamp = journal->j_trans_start_time;
4103	jl->j_commit_bh = c_bh;
4104	jl->j_start = journal->j_start;
4105	jl->j_len = journal->j_len;
4106	atomic_set(&jl->j_nonzerolen, journal->j_len);
4107	atomic_set(&jl->j_commit_left, journal->j_len + 2);
4108	jl->j_realblock = NULL;
4109
4110	/*
4111	 * The ENTIRE FOR LOOP MUST not cause schedule to occur.
4112	 * for each real block, add it to the journal list hash,
4113	 * copy into real block index array in the commit or desc block
4114	 */
4115	trans_half = journal_trans_half(sb->s_blocksize);
4116	for (i = 0, cn = journal->j_first; cn; cn = cn->next, i++) {
4117		if (buffer_journaled(cn->bh)) {
4118			jl_cn = get_cnode(sb);
4119			if (!jl_cn) {
4120				reiserfs_panic(sb, "journal-1676",
4121					       "get_cnode returned NULL");
4122			}
4123			if (i == 0) {
4124				jl->j_realblock = jl_cn;
4125			}
4126			jl_cn->prev = last_cn;
4127			jl_cn->next = NULL;
4128			if (last_cn) {
4129				last_cn->next = jl_cn;
4130			}
4131			last_cn = jl_cn;
4132			/*
4133			 * make sure the block we are trying to log
4134			 * is not a block of journal or reserved area
4135			 */
4136			if (is_block_in_log_or_reserved_area
4137			    (sb, cn->bh->b_blocknr)) {
4138				reiserfs_panic(sb, "journal-2332",
4139					       "Trying to log block %lu, "
4140					       "which is a log block",
4141					       cn->bh->b_blocknr);
4142			}
4143			jl_cn->blocknr = cn->bh->b_blocknr;
4144			jl_cn->state = 0;
4145			jl_cn->sb = sb;
4146			jl_cn->bh = cn->bh;
4147			jl_cn->jlist = jl;
4148			insert_journal_hash(journal->j_list_hash_table, jl_cn);
4149			if (i < trans_half) {
4150				desc->j_realblock[i] =
4151				    cpu_to_le32(cn->bh->b_blocknr);
4152			} else {
4153				commit->j_realblock[i - trans_half] =
4154				    cpu_to_le32(cn->bh->b_blocknr);
4155			}
4156		} else {
4157			i--;
4158		}
4159	}
4160	set_desc_trans_len(desc, journal->j_len);
4161	set_desc_mount_id(desc, journal->j_mount_id);
4162	set_desc_trans_id(desc, journal->j_trans_id);
4163	set_commit_trans_len(commit, journal->j_len);
4164
4165	/*
4166	 * special check in case all buffers in the journal
4167	 * were marked for not logging
4168	 */
4169	BUG_ON(journal->j_len == 0);
4170
4171	/*
4172	 * we're about to dirty all the log blocks, mark the description block
4173	 * dirty now too.  Don't mark the commit block dirty until all the
4174	 * others are on disk
4175	 */
4176	mark_buffer_dirty(d_bh);
4177
4178	/*
4179	 * first data block is j_start + 1, so add one to
4180	 * cur_write_start wherever you use it
4181	 */
4182	cur_write_start = journal->j_start;
4183	cn = journal->j_first;
4184	jindex = 1;	/* start at one so we don't get the desc again */
4185	while (cn) {
4186		clear_buffer_journal_new(cn->bh);
4187		/* copy all the real blocks into log area.  dirty log blocks */
4188		if (buffer_journaled(cn->bh)) {
4189			struct buffer_head *tmp_bh;
4190			char *addr;
4191			struct page *page;
4192			tmp_bh =
4193			    journal_getblk(sb,
4194					   SB_ONDISK_JOURNAL_1st_BLOCK(sb) +
4195					   ((cur_write_start +
4196					     jindex) %
4197					    SB_ONDISK_JOURNAL_SIZE(sb)));
4198			set_buffer_uptodate(tmp_bh);
4199			page = cn->bh->b_page;
4200			addr = kmap(page);
4201			memcpy(tmp_bh->b_data,
4202			       addr + offset_in_page(cn->bh->b_data),
4203			       cn->bh->b_size);
4204			kunmap(page);
4205			mark_buffer_dirty(tmp_bh);
4206			jindex++;
4207			set_buffer_journal_dirty(cn->bh);
4208			clear_buffer_journaled(cn->bh);
4209		} else {
4210			/*
4211			 * JDirty cleared sometime during transaction.
4212			 * don't log this one
4213			 */
4214			reiserfs_warning(sb, "journal-2048",
4215					 "BAD, buffer in journal hash, "
4216					 "but not JDirty!");
4217			brelse(cn->bh);
4218		}
4219		next = cn->next;
4220		free_cnode(sb, cn);
4221		cn = next;
4222		reiserfs_cond_resched(sb);
4223	}
4224
4225	/*
4226	 * we are done with both the c_bh and d_bh, but
4227	 * c_bh must be written after all other commit blocks,
4228	 * so we dirty/relse c_bh in flush_commit_list, with commit_left <= 1.
4229	 */
4230
4231	journal->j_current_jl = alloc_journal_list(sb);
4232
4233	/* now it is safe to insert this transaction on the main list */
4234	list_add_tail(&jl->j_list, &journal->j_journal_list);
4235	list_add_tail(&jl->j_working_list, &journal->j_working_list);
4236	journal->j_num_work_lists++;
4237
4238	/* reset journal values for the next transaction */
4239	journal->j_start =
4240	    (journal->j_start + journal->j_len +
4241	     2) % SB_ONDISK_JOURNAL_SIZE(sb);
4242	atomic_set(&journal->j_wcount, 0);
4243	journal->j_bcount = 0;
4244	journal->j_last = NULL;
4245	journal->j_first = NULL;
4246	journal->j_len = 0;
4247	journal->j_trans_start_time = 0;
4248	/* check for trans_id overflow */
4249	if (++journal->j_trans_id == 0)
4250		journal->j_trans_id = 10;
4251	journal->j_current_jl->j_trans_id = journal->j_trans_id;
4252	journal->j_must_wait = 0;
4253	journal->j_len_alloc = 0;
4254	journal->j_next_full_flush = 0;
4255	journal->j_next_async_flush = 0;
4256	init_journal_hash(sb);
4257
4258	/*
4259	 * make sure reiserfs_add_jh sees the new current_jl before we
4260	 * write out the tails
4261	 */
4262	smp_mb();
4263
4264	/*
4265	 * tail conversion targets have to hit the disk before we end the
4266	 * transaction.  Otherwise a later transaction might repack the tail
4267	 * before this transaction commits, leaving the data block unflushed
4268	 * and clean, if we crash before the later transaction commits, the
4269	 * data block is lost.
4270	 */
4271	if (!list_empty(&jl->j_tail_bh_list)) {
4272		depth = reiserfs_write_unlock_nested(sb);
4273		write_ordered_buffers(&journal->j_dirty_buffers_lock,
4274				      journal, jl, &jl->j_tail_bh_list);
4275		reiserfs_write_lock_nested(sb, depth);
4276	}
4277	BUG_ON(!list_empty(&jl->j_tail_bh_list));
4278	mutex_unlock(&jl->j_commit_mutex);
4279
4280	/*
4281	 * honor the flush wishes from the caller, simple commits can
4282	 * be done outside the journal lock, they are done below
4283	 *
4284	 * if we don't flush the commit list right now, we put it into
4285	 * the work queue so the people waiting on the async progress work
4286	 * queue don't wait for this proc to flush journal lists and such.
4287	 */
4288	if (flush) {
4289		flush_commit_list(sb, jl, 1);
4290		flush_journal_list(sb, jl, 1);
4291	} else if (!(jl->j_state & LIST_COMMIT_PENDING)) {
4292		/*
4293		 * Avoid queueing work when sb is being shut down. Transaction
4294		 * will be flushed on journal shutdown.
4295		 */
4296		if (sb->s_flags & SB_ACTIVE)
4297			queue_delayed_work(REISERFS_SB(sb)->commit_wq,
4298					   &journal->j_work, HZ / 10);
4299	}
4300
4301	/*
4302	 * if the next transaction has any chance of wrapping, flush
4303	 * transactions that might get overwritten.  If any journal lists
4304	 * are very old flush them as well.
4305	 */
4306first_jl:
4307	list_for_each_safe(entry, safe, &journal->j_journal_list) {
4308		temp_jl = JOURNAL_LIST_ENTRY(entry);
4309		if (journal->j_start <= temp_jl->j_start) {
4310			if ((journal->j_start + journal->j_trans_max + 1) >=
4311			    temp_jl->j_start) {
4312				flush_used_journal_lists(sb, temp_jl);
4313				goto first_jl;
4314			} else if ((journal->j_start +
4315				    journal->j_trans_max + 1) <
4316				   SB_ONDISK_JOURNAL_SIZE(sb)) {
4317				/*
4318				 * if we don't cross into the next
4319				 * transaction and we don't wrap, there is
4320				 * no way we can overlap any later transactions
4321				 * break now
4322				 */
4323				break;
4324			}
4325		} else if ((journal->j_start +
4326			    journal->j_trans_max + 1) >
4327			   SB_ONDISK_JOURNAL_SIZE(sb)) {
4328			if (((journal->j_start + journal->j_trans_max + 1) %
4329			     SB_ONDISK_JOURNAL_SIZE(sb)) >=
4330			    temp_jl->j_start) {
4331				flush_used_journal_lists(sb, temp_jl);
4332				goto first_jl;
4333			} else {
4334				/*
4335				* we don't overlap anything from out start
4336				* to the end of the log, and our wrapped
4337				* portion doesn't overlap anything at
4338				* the start of the log.  We can break
4339				*/
4340				break;
4341			}
4342		}
4343	}
4344
4345	journal->j_current_jl->j_list_bitmap =
4346	    get_list_bitmap(sb, journal->j_current_jl);
4347
4348	if (!(journal->j_current_jl->j_list_bitmap)) {
4349		reiserfs_panic(sb, "journal-1996",
4350			       "could not get a list bitmap");
4351	}
4352
4353	atomic_set(&journal->j_jlock, 0);
4354	unlock_journal(sb);
4355	/* wake up any body waiting to join. */
4356	clear_bit(J_WRITERS_QUEUED, &journal->j_state);
4357	wake_up(&journal->j_join_wait);
4358
4359	if (!flush && wait_on_commit &&
4360	    journal_list_still_alive(sb, commit_trans_id)) {
4361		flush_commit_list(sb, jl, 1);
4362	}
4363out:
4364	reiserfs_check_lock_depth(sb, "journal end2");
4365
4366	memset(th, 0, sizeof(*th));
4367	/*
4368	 * Re-set th->t_super, so we can properly keep track of how many
4369	 * persistent transactions there are. We need to do this so if this
4370	 * call is part of a failed restart_transaction, we can free it later
4371	 */
4372	th->t_super = sb;
4373
4374	return journal->j_errno;
4375}
4376
4377/* Send the file system read only and refuse new transactions */
4378void reiserfs_abort_journal(struct super_block *sb, int errno)
4379{
4380	struct reiserfs_journal *journal = SB_JOURNAL(sb);
4381	if (test_bit(J_ABORTED, &journal->j_state))
4382		return;
4383
4384	if (!journal->j_errno)
4385		journal->j_errno = errno;
4386
4387	sb->s_flags |= SB_RDONLY;
4388	set_bit(J_ABORTED, &journal->j_state);
4389
4390#ifdef CONFIG_REISERFS_CHECK
4391	dump_stack();
4392#endif
4393}