Linux Audio

Check our new training course

Loading...
v4.6
 
   1/*
   2 * linux/fs/jbd2/journal.c
   3 *
   4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
   5 *
   6 * Copyright 1998 Red Hat corp --- All Rights Reserved
   7 *
   8 * This file is part of the Linux kernel and is made available under
   9 * the terms of the GNU General Public License, version 2, or at your
  10 * option, any later version, incorporated herein by reference.
  11 *
  12 * Generic filesystem journal-writing code; part of the ext2fs
  13 * journaling system.
  14 *
  15 * This file manages journals: areas of disk reserved for logging
  16 * transactional updates.  This includes the kernel journaling thread
  17 * which is responsible for scheduling updates to the log.
  18 *
  19 * We do not actually manage the physical storage of the journal in this
  20 * file: that is left to a per-journal policy function, which allows us
  21 * to store the journal within a filesystem-specified area for ext2
  22 * journaling (ext2 can use a reserved inode for storing the log).
  23 */
  24
  25#include <linux/module.h>
  26#include <linux/time.h>
  27#include <linux/fs.h>
  28#include <linux/jbd2.h>
  29#include <linux/errno.h>
  30#include <linux/slab.h>
  31#include <linux/init.h>
  32#include <linux/mm.h>
  33#include <linux/freezer.h>
  34#include <linux/pagemap.h>
  35#include <linux/kthread.h>
  36#include <linux/poison.h>
  37#include <linux/proc_fs.h>
  38#include <linux/seq_file.h>
  39#include <linux/math64.h>
  40#include <linux/hash.h>
  41#include <linux/log2.h>
  42#include <linux/vmalloc.h>
  43#include <linux/backing-dev.h>
  44#include <linux/bitops.h>
  45#include <linux/ratelimit.h>
 
  46
  47#define CREATE_TRACE_POINTS
  48#include <trace/events/jbd2.h>
  49
  50#include <asm/uaccess.h>
  51#include <asm/page.h>
  52
  53#ifdef CONFIG_JBD2_DEBUG
  54ushort jbd2_journal_enable_debug __read_mostly;
  55EXPORT_SYMBOL(jbd2_journal_enable_debug);
  56
  57module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
  58MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
  59#endif
  60
  61EXPORT_SYMBOL(jbd2_journal_extend);
  62EXPORT_SYMBOL(jbd2_journal_stop);
  63EXPORT_SYMBOL(jbd2_journal_lock_updates);
  64EXPORT_SYMBOL(jbd2_journal_unlock_updates);
  65EXPORT_SYMBOL(jbd2_journal_get_write_access);
  66EXPORT_SYMBOL(jbd2_journal_get_create_access);
  67EXPORT_SYMBOL(jbd2_journal_get_undo_access);
  68EXPORT_SYMBOL(jbd2_journal_set_triggers);
  69EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
  70EXPORT_SYMBOL(jbd2_journal_forget);
  71#if 0
  72EXPORT_SYMBOL(journal_sync_buffer);
  73#endif
  74EXPORT_SYMBOL(jbd2_journal_flush);
  75EXPORT_SYMBOL(jbd2_journal_revoke);
  76
  77EXPORT_SYMBOL(jbd2_journal_init_dev);
  78EXPORT_SYMBOL(jbd2_journal_init_inode);
  79EXPORT_SYMBOL(jbd2_journal_check_used_features);
  80EXPORT_SYMBOL(jbd2_journal_check_available_features);
  81EXPORT_SYMBOL(jbd2_journal_set_features);
  82EXPORT_SYMBOL(jbd2_journal_load);
  83EXPORT_SYMBOL(jbd2_journal_destroy);
  84EXPORT_SYMBOL(jbd2_journal_abort);
  85EXPORT_SYMBOL(jbd2_journal_errno);
  86EXPORT_SYMBOL(jbd2_journal_ack_err);
  87EXPORT_SYMBOL(jbd2_journal_clear_err);
  88EXPORT_SYMBOL(jbd2_log_wait_commit);
  89EXPORT_SYMBOL(jbd2_log_start_commit);
  90EXPORT_SYMBOL(jbd2_journal_start_commit);
  91EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
  92EXPORT_SYMBOL(jbd2_journal_wipe);
  93EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
  94EXPORT_SYMBOL(jbd2_journal_invalidatepage);
  95EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
  96EXPORT_SYMBOL(jbd2_journal_force_commit);
  97EXPORT_SYMBOL(jbd2_journal_file_inode);
 
  98EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
  99EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
 100EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
 101EXPORT_SYMBOL(jbd2_inode_cache);
 102
 103static void __journal_abort_soft (journal_t *journal, int errno);
 104static int jbd2_journal_create_slab(size_t slab_size);
 105
 106#ifdef CONFIG_JBD2_DEBUG
 107void __jbd2_debug(int level, const char *file, const char *func,
 108		  unsigned int line, const char *fmt, ...)
 109{
 110	struct va_format vaf;
 111	va_list args;
 112
 113	if (level > jbd2_journal_enable_debug)
 114		return;
 115	va_start(args, fmt);
 116	vaf.fmt = fmt;
 117	vaf.va = &args;
 118	printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
 119	va_end(args);
 120}
 121EXPORT_SYMBOL(__jbd2_debug);
 122#endif
 123
 124/* Checksumming functions */
 125static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
 126{
 127	if (!jbd2_journal_has_csum_v2or3_feature(j))
 128		return 1;
 129
 130	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
 131}
 132
 133static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
 134{
 135	__u32 csum;
 136	__be32 old_csum;
 137
 138	old_csum = sb->s_checksum;
 139	sb->s_checksum = 0;
 140	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
 141	sb->s_checksum = old_csum;
 142
 143	return cpu_to_be32(csum);
 144}
 145
 146static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
 147{
 148	if (!jbd2_journal_has_csum_v2or3(j))
 149		return 1;
 150
 151	return sb->s_checksum == jbd2_superblock_csum(j, sb);
 152}
 153
 154static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
 155{
 156	if (!jbd2_journal_has_csum_v2or3(j))
 157		return;
 158
 159	sb->s_checksum = jbd2_superblock_csum(j, sb);
 160}
 161
 162/*
 163 * Helper function used to manage commit timeouts
 164 */
 165
 166static void commit_timeout(unsigned long __data)
 167{
 168	struct task_struct * p = (struct task_struct *) __data;
 169
 170	wake_up_process(p);
 171}
 172
 173/*
 174 * kjournald2: The main thread function used to manage a logging device
 175 * journal.
 176 *
 177 * This kernel thread is responsible for two things:
 178 *
 179 * 1) COMMIT:  Every so often we need to commit the current state of the
 180 *    filesystem to disk.  The journal thread is responsible for writing
 181 *    all of the metadata buffers to disk.
 182 *
 183 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
 184 *    of the data in that part of the log has been rewritten elsewhere on
 185 *    the disk.  Flushing these old buffers to reclaim space in the log is
 186 *    known as checkpointing, and this thread is responsible for that job.
 187 */
 188
 189static int kjournald2(void *arg)
 190{
 191	journal_t *journal = arg;
 192	transaction_t *transaction;
 193
 194	/*
 195	 * Set up an interval timer which can be used to trigger a commit wakeup
 196	 * after the commit interval expires
 197	 */
 198	setup_timer(&journal->j_commit_timer, commit_timeout,
 199			(unsigned long)current);
 200
 201	set_freezable();
 202
 203	/* Record that the journal thread is running */
 204	journal->j_task = current;
 205	wake_up(&journal->j_wait_done_commit);
 206
 207	/*
 
 
 
 
 
 
 
 
 208	 * And now, wait forever for commit wakeup events.
 209	 */
 210	write_lock(&journal->j_state_lock);
 211
 212loop:
 213	if (journal->j_flags & JBD2_UNMOUNT)
 214		goto end_loop;
 215
 216	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
 217		journal->j_commit_sequence, journal->j_commit_request);
 218
 219	if (journal->j_commit_sequence != journal->j_commit_request) {
 220		jbd_debug(1, "OK, requests differ\n");
 221		write_unlock(&journal->j_state_lock);
 222		del_timer_sync(&journal->j_commit_timer);
 223		jbd2_journal_commit_transaction(journal);
 224		write_lock(&journal->j_state_lock);
 225		goto loop;
 226	}
 227
 228	wake_up(&journal->j_wait_done_commit);
 229	if (freezing(current)) {
 230		/*
 231		 * The simpler the better. Flushing journal isn't a
 232		 * good idea, because that depends on threads that may
 233		 * be already stopped.
 234		 */
 235		jbd_debug(1, "Now suspending kjournald2\n");
 236		write_unlock(&journal->j_state_lock);
 237		try_to_freeze();
 238		write_lock(&journal->j_state_lock);
 239	} else {
 240		/*
 241		 * We assume on resume that commits are already there,
 242		 * so we don't sleep
 243		 */
 244		DEFINE_WAIT(wait);
 245		int should_sleep = 1;
 246
 247		prepare_to_wait(&journal->j_wait_commit, &wait,
 248				TASK_INTERRUPTIBLE);
 249		if (journal->j_commit_sequence != journal->j_commit_request)
 250			should_sleep = 0;
 251		transaction = journal->j_running_transaction;
 252		if (transaction && time_after_eq(jiffies,
 253						transaction->t_expires))
 254			should_sleep = 0;
 255		if (journal->j_flags & JBD2_UNMOUNT)
 256			should_sleep = 0;
 257		if (should_sleep) {
 258			write_unlock(&journal->j_state_lock);
 259			schedule();
 260			write_lock(&journal->j_state_lock);
 261		}
 262		finish_wait(&journal->j_wait_commit, &wait);
 263	}
 264
 265	jbd_debug(1, "kjournald2 wakes\n");
 266
 267	/*
 268	 * Were we woken up by a commit wakeup event?
 269	 */
 270	transaction = journal->j_running_transaction;
 271	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
 272		journal->j_commit_request = transaction->t_tid;
 273		jbd_debug(1, "woke because of timeout\n");
 274	}
 275	goto loop;
 276
 277end_loop:
 278	write_unlock(&journal->j_state_lock);
 279	del_timer_sync(&journal->j_commit_timer);
 280	journal->j_task = NULL;
 281	wake_up(&journal->j_wait_done_commit);
 282	jbd_debug(1, "Journal thread exiting.\n");
 
 283	return 0;
 284}
 285
 286static int jbd2_journal_start_thread(journal_t *journal)
 287{
 288	struct task_struct *t;
 289
 290	t = kthread_run(kjournald2, journal, "jbd2/%s",
 291			journal->j_devname);
 292	if (IS_ERR(t))
 293		return PTR_ERR(t);
 294
 295	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
 296	return 0;
 297}
 298
 299static void journal_kill_thread(journal_t *journal)
 300{
 301	write_lock(&journal->j_state_lock);
 302	journal->j_flags |= JBD2_UNMOUNT;
 303
 304	while (journal->j_task) {
 305		write_unlock(&journal->j_state_lock);
 306		wake_up(&journal->j_wait_commit);
 307		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
 308		write_lock(&journal->j_state_lock);
 309	}
 310	write_unlock(&journal->j_state_lock);
 311}
 312
 313/*
 314 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
 315 *
 316 * Writes a metadata buffer to a given disk block.  The actual IO is not
 317 * performed but a new buffer_head is constructed which labels the data
 318 * to be written with the correct destination disk block.
 319 *
 320 * Any magic-number escaping which needs to be done will cause a
 321 * copy-out here.  If the buffer happens to start with the
 322 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
 323 * magic number is only written to the log for descripter blocks.  In
 324 * this case, we copy the data and replace the first word with 0, and we
 325 * return a result code which indicates that this buffer needs to be
 326 * marked as an escaped buffer in the corresponding log descriptor
 327 * block.  The missing word can then be restored when the block is read
 328 * during recovery.
 329 *
 330 * If the source buffer has already been modified by a new transaction
 331 * since we took the last commit snapshot, we use the frozen copy of
 332 * that data for IO. If we end up using the existing buffer_head's data
 333 * for the write, then we have to make sure nobody modifies it while the
 334 * IO is in progress. do_get_write_access() handles this.
 335 *
 336 * The function returns a pointer to the buffer_head to be used for IO.
 337 * 
 338 *
 339 * Return value:
 340 *  <0: Error
 341 * >=0: Finished OK
 342 *
 343 * On success:
 344 * Bit 0 set == escape performed on the data
 345 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
 346 */
 347
 348int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
 349				  struct journal_head  *jh_in,
 350				  struct buffer_head **bh_out,
 351				  sector_t blocknr)
 352{
 353	int need_copy_out = 0;
 354	int done_copy_out = 0;
 355	int do_escape = 0;
 356	char *mapped_data;
 357	struct buffer_head *new_bh;
 358	struct page *new_page;
 359	unsigned int new_offset;
 360	struct buffer_head *bh_in = jh2bh(jh_in);
 361	journal_t *journal = transaction->t_journal;
 362
 363	/*
 364	 * The buffer really shouldn't be locked: only the current committing
 365	 * transaction is allowed to write it, so nobody else is allowed
 366	 * to do any IO.
 367	 *
 368	 * akpm: except if we're journalling data, and write() output is
 369	 * also part of a shared mapping, and another thread has
 370	 * decided to launch a writepage() against this buffer.
 371	 */
 372	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 373
 374	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
 375
 376	/* keep subsequent assertions sane */
 377	atomic_set(&new_bh->b_count, 1);
 378
 379	jbd_lock_bh_state(bh_in);
 380repeat:
 381	/*
 382	 * If a new transaction has already done a buffer copy-out, then
 383	 * we use that version of the data for the commit.
 384	 */
 385	if (jh_in->b_frozen_data) {
 386		done_copy_out = 1;
 387		new_page = virt_to_page(jh_in->b_frozen_data);
 388		new_offset = offset_in_page(jh_in->b_frozen_data);
 389	} else {
 390		new_page = jh2bh(jh_in)->b_page;
 391		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
 392	}
 393
 394	mapped_data = kmap_atomic(new_page);
 395	/*
 396	 * Fire data frozen trigger if data already wasn't frozen.  Do this
 397	 * before checking for escaping, as the trigger may modify the magic
 398	 * offset.  If a copy-out happens afterwards, it will have the correct
 399	 * data in the buffer.
 400	 */
 401	if (!done_copy_out)
 402		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
 403					   jh_in->b_triggers);
 404
 405	/*
 406	 * Check for escaping
 407	 */
 408	if (*((__be32 *)(mapped_data + new_offset)) ==
 409				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 410		need_copy_out = 1;
 411		do_escape = 1;
 412	}
 413	kunmap_atomic(mapped_data);
 414
 415	/*
 416	 * Do we need to do a data copy?
 417	 */
 418	if (need_copy_out && !done_copy_out) {
 419		char *tmp;
 420
 421		jbd_unlock_bh_state(bh_in);
 422		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
 423		if (!tmp) {
 424			brelse(new_bh);
 425			return -ENOMEM;
 426		}
 427		jbd_lock_bh_state(bh_in);
 428		if (jh_in->b_frozen_data) {
 429			jbd2_free(tmp, bh_in->b_size);
 430			goto repeat;
 431		}
 432
 433		jh_in->b_frozen_data = tmp;
 434		mapped_data = kmap_atomic(new_page);
 435		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
 436		kunmap_atomic(mapped_data);
 437
 438		new_page = virt_to_page(tmp);
 439		new_offset = offset_in_page(tmp);
 440		done_copy_out = 1;
 441
 442		/*
 443		 * This isn't strictly necessary, as we're using frozen
 444		 * data for the escaping, but it keeps consistency with
 445		 * b_frozen_data usage.
 446		 */
 447		jh_in->b_frozen_triggers = jh_in->b_triggers;
 448	}
 449
 450	/*
 451	 * Did we need to do an escaping?  Now we've done all the
 452	 * copying, we can finally do so.
 453	 */
 454	if (do_escape) {
 455		mapped_data = kmap_atomic(new_page);
 456		*((unsigned int *)(mapped_data + new_offset)) = 0;
 457		kunmap_atomic(mapped_data);
 458	}
 459
 460	set_bh_page(new_bh, new_page, new_offset);
 461	new_bh->b_size = bh_in->b_size;
 462	new_bh->b_bdev = journal->j_dev;
 463	new_bh->b_blocknr = blocknr;
 464	new_bh->b_private = bh_in;
 465	set_buffer_mapped(new_bh);
 466	set_buffer_dirty(new_bh);
 467
 468	*bh_out = new_bh;
 469
 470	/*
 471	 * The to-be-written buffer needs to get moved to the io queue,
 472	 * and the original buffer whose contents we are shadowing or
 473	 * copying is moved to the transaction's shadow queue.
 474	 */
 475	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
 476	spin_lock(&journal->j_list_lock);
 477	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
 478	spin_unlock(&journal->j_list_lock);
 479	set_buffer_shadow(bh_in);
 480	jbd_unlock_bh_state(bh_in);
 481
 482	return do_escape | (done_copy_out << 1);
 483}
 484
 485/*
 486 * Allocation code for the journal file.  Manage the space left in the
 487 * journal, so that we can begin checkpointing when appropriate.
 488 */
 489
 490/*
 491 * Called with j_state_lock locked for writing.
 492 * Returns true if a transaction commit was started.
 493 */
 494int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 495{
 496	/* Return if the txn has already requested to be committed */
 497	if (journal->j_commit_request == target)
 498		return 0;
 499
 500	/*
 501	 * The only transaction we can possibly wait upon is the
 502	 * currently running transaction (if it exists).  Otherwise,
 503	 * the target tid must be an old one.
 504	 */
 505	if (journal->j_running_transaction &&
 506	    journal->j_running_transaction->t_tid == target) {
 507		/*
 508		 * We want a new commit: OK, mark the request and wakeup the
 509		 * commit thread.  We do _not_ do the commit ourselves.
 510		 */
 511
 512		journal->j_commit_request = target;
 513		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
 514			  journal->j_commit_request,
 515			  journal->j_commit_sequence);
 516		journal->j_running_transaction->t_requested = jiffies;
 517		wake_up(&journal->j_wait_commit);
 518		return 1;
 519	} else if (!tid_geq(journal->j_commit_request, target))
 520		/* This should never happen, but if it does, preserve
 521		   the evidence before kjournald goes into a loop and
 522		   increments j_commit_sequence beyond all recognition. */
 523		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
 524			  journal->j_commit_request,
 525			  journal->j_commit_sequence,
 526			  target, journal->j_running_transaction ? 
 527			  journal->j_running_transaction->t_tid : 0);
 528	return 0;
 529}
 530
 531int jbd2_log_start_commit(journal_t *journal, tid_t tid)
 532{
 533	int ret;
 534
 535	write_lock(&journal->j_state_lock);
 536	ret = __jbd2_log_start_commit(journal, tid);
 537	write_unlock(&journal->j_state_lock);
 538	return ret;
 539}
 540
 541/*
 542 * Force and wait any uncommitted transactions.  We can only force the running
 543 * transaction if we don't have an active handle, otherwise, we will deadlock.
 544 * Returns: <0 in case of error,
 545 *           0 if nothing to commit,
 546 *           1 if transaction was successfully committed.
 547 */
 548static int __jbd2_journal_force_commit(journal_t *journal)
 549{
 550	transaction_t *transaction = NULL;
 551	tid_t tid;
 552	int need_to_start = 0, ret = 0;
 553
 554	read_lock(&journal->j_state_lock);
 555	if (journal->j_running_transaction && !current->journal_info) {
 556		transaction = journal->j_running_transaction;
 557		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
 558			need_to_start = 1;
 559	} else if (journal->j_committing_transaction)
 560		transaction = journal->j_committing_transaction;
 561
 562	if (!transaction) {
 563		/* Nothing to commit */
 564		read_unlock(&journal->j_state_lock);
 565		return 0;
 566	}
 567	tid = transaction->t_tid;
 568	read_unlock(&journal->j_state_lock);
 569	if (need_to_start)
 570		jbd2_log_start_commit(journal, tid);
 571	ret = jbd2_log_wait_commit(journal, tid);
 572	if (!ret)
 573		ret = 1;
 574
 575	return ret;
 576}
 577
 578/**
 579 * Force and wait upon a commit if the calling process is not within
 580 * transaction.  This is used for forcing out undo-protected data which contains
 581 * bitmaps, when the fs is running out of space.
 582 *
 583 * @journal: journal to force
 584 * Returns true if progress was made.
 585 */
 586int jbd2_journal_force_commit_nested(journal_t *journal)
 587{
 588	int ret;
 589
 590	ret = __jbd2_journal_force_commit(journal);
 591	return ret > 0;
 592}
 593
 594/**
 595 * int journal_force_commit() - force any uncommitted transactions
 596 * @journal: journal to force
 597 *
 598 * Caller want unconditional commit. We can only force the running transaction
 599 * if we don't have an active handle, otherwise, we will deadlock.
 600 */
 601int jbd2_journal_force_commit(journal_t *journal)
 602{
 603	int ret;
 604
 605	J_ASSERT(!current->journal_info);
 606	ret = __jbd2_journal_force_commit(journal);
 607	if (ret > 0)
 608		ret = 0;
 609	return ret;
 610}
 611
 612/*
 613 * Start a commit of the current running transaction (if any).  Returns true
 614 * if a transaction is going to be committed (or is currently already
 615 * committing), and fills its tid in at *ptid
 616 */
 617int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
 618{
 619	int ret = 0;
 620
 621	write_lock(&journal->j_state_lock);
 622	if (journal->j_running_transaction) {
 623		tid_t tid = journal->j_running_transaction->t_tid;
 624
 625		__jbd2_log_start_commit(journal, tid);
 626		/* There's a running transaction and we've just made sure
 627		 * it's commit has been scheduled. */
 628		if (ptid)
 629			*ptid = tid;
 630		ret = 1;
 631	} else if (journal->j_committing_transaction) {
 632		/*
 633		 * If commit has been started, then we have to wait for
 634		 * completion of that transaction.
 635		 */
 636		if (ptid)
 637			*ptid = journal->j_committing_transaction->t_tid;
 638		ret = 1;
 639	}
 640	write_unlock(&journal->j_state_lock);
 641	return ret;
 642}
 643
 644/*
 645 * Return 1 if a given transaction has not yet sent barrier request
 646 * connected with a transaction commit. If 0 is returned, transaction
 647 * may or may not have sent the barrier. Used to avoid sending barrier
 648 * twice in common cases.
 649 */
 650int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
 651{
 652	int ret = 0;
 653	transaction_t *commit_trans;
 654
 655	if (!(journal->j_flags & JBD2_BARRIER))
 656		return 0;
 657	read_lock(&journal->j_state_lock);
 658	/* Transaction already committed? */
 659	if (tid_geq(journal->j_commit_sequence, tid))
 660		goto out;
 661	commit_trans = journal->j_committing_transaction;
 662	if (!commit_trans || commit_trans->t_tid != tid) {
 663		ret = 1;
 664		goto out;
 665	}
 666	/*
 667	 * Transaction is being committed and we already proceeded to
 668	 * submitting a flush to fs partition?
 669	 */
 670	if (journal->j_fs_dev != journal->j_dev) {
 671		if (!commit_trans->t_need_data_flush ||
 672		    commit_trans->t_state >= T_COMMIT_DFLUSH)
 673			goto out;
 674	} else {
 675		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
 676			goto out;
 677	}
 678	ret = 1;
 679out:
 680	read_unlock(&journal->j_state_lock);
 681	return ret;
 682}
 683EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
 684
 685/*
 686 * Wait for a specified commit to complete.
 687 * The caller may not hold the journal lock.
 688 */
 689int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
 690{
 691	int err = 0;
 692
 693	read_lock(&journal->j_state_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 694#ifdef CONFIG_JBD2_DEBUG
 695	if (!tid_geq(journal->j_commit_request, tid)) {
 696		printk(KERN_ERR
 697		       "%s: error: j_commit_request=%d, tid=%d\n",
 698		       __func__, journal->j_commit_request, tid);
 699	}
 700#endif
 701	while (tid_gt(tid, journal->j_commit_sequence)) {
 702		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
 703				  tid, journal->j_commit_sequence);
 704		read_unlock(&journal->j_state_lock);
 705		wake_up(&journal->j_wait_commit);
 706		wait_event(journal->j_wait_done_commit,
 707				!tid_gt(tid, journal->j_commit_sequence));
 708		read_lock(&journal->j_state_lock);
 709	}
 710	read_unlock(&journal->j_state_lock);
 711
 712	if (unlikely(is_journal_aborted(journal)))
 713		err = -EIO;
 714	return err;
 715}
 716
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 717/*
 718 * When this function returns the transaction corresponding to tid
 719 * will be completed.  If the transaction has currently running, start
 720 * committing that transaction before waiting for it to complete.  If
 721 * the transaction id is stale, it is by definition already completed,
 722 * so just return SUCCESS.
 723 */
 724int jbd2_complete_transaction(journal_t *journal, tid_t tid)
 725{
 726	int	need_to_wait = 1;
 727
 728	read_lock(&journal->j_state_lock);
 729	if (journal->j_running_transaction &&
 730	    journal->j_running_transaction->t_tid == tid) {
 731		if (journal->j_commit_request != tid) {
 732			/* transaction not yet started, so request it */
 733			read_unlock(&journal->j_state_lock);
 734			jbd2_log_start_commit(journal, tid);
 735			goto wait_commit;
 736		}
 737	} else if (!(journal->j_committing_transaction &&
 738		     journal->j_committing_transaction->t_tid == tid))
 739		need_to_wait = 0;
 740	read_unlock(&journal->j_state_lock);
 741	if (!need_to_wait)
 742		return 0;
 743wait_commit:
 744	return jbd2_log_wait_commit(journal, tid);
 745}
 746EXPORT_SYMBOL(jbd2_complete_transaction);
 747
 748/*
 749 * Log buffer allocation routines:
 750 */
 751
 752int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
 753{
 754	unsigned long blocknr;
 755
 756	write_lock(&journal->j_state_lock);
 757	J_ASSERT(journal->j_free > 1);
 758
 759	blocknr = journal->j_head;
 760	journal->j_head++;
 761	journal->j_free--;
 762	if (journal->j_head == journal->j_last)
 763		journal->j_head = journal->j_first;
 764	write_unlock(&journal->j_state_lock);
 765	return jbd2_journal_bmap(journal, blocknr, retp);
 766}
 767
 768/*
 769 * Conversion of logical to physical block numbers for the journal
 770 *
 771 * On external journals the journal blocks are identity-mapped, so
 772 * this is a no-op.  If needed, we can use j_blk_offset - everything is
 773 * ready.
 774 */
 775int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 776		 unsigned long long *retp)
 777{
 778	int err = 0;
 779	unsigned long long ret;
 780
 781	if (journal->j_inode) {
 782		ret = bmap(journal->j_inode, blocknr);
 783		if (ret)
 784			*retp = ret;
 785		else {
 786			printk(KERN_ALERT "%s: journal block not found "
 787					"at offset %lu on %s\n",
 788			       __func__, blocknr, journal->j_devname);
 789			err = -EIO;
 790			__journal_abort_soft(journal, err);
 791		}
 792	} else {
 793		*retp = blocknr; /* +journal->j_blk_offset */
 794	}
 795	return err;
 796}
 797
 798/*
 799 * We play buffer_head aliasing tricks to write data/metadata blocks to
 800 * the journal without copying their contents, but for journal
 801 * descriptor blocks we do need to generate bona fide buffers.
 802 *
 803 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
 804 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
 805 * But we don't bother doing that, so there will be coherency problems with
 806 * mmaps of blockdevs which hold live JBD-controlled filesystems.
 807 */
 808struct buffer_head *
 809jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
 810{
 811	journal_t *journal = transaction->t_journal;
 812	struct buffer_head *bh;
 813	unsigned long long blocknr;
 814	journal_header_t *header;
 815	int err;
 816
 817	err = jbd2_journal_next_log_block(journal, &blocknr);
 818
 819	if (err)
 820		return NULL;
 821
 822	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 823	if (!bh)
 824		return NULL;
 825	lock_buffer(bh);
 826	memset(bh->b_data, 0, journal->j_blocksize);
 827	header = (journal_header_t *)bh->b_data;
 828	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
 829	header->h_blocktype = cpu_to_be32(type);
 830	header->h_sequence = cpu_to_be32(transaction->t_tid);
 831	set_buffer_uptodate(bh);
 832	unlock_buffer(bh);
 833	BUFFER_TRACE(bh, "return this buffer");
 834	return bh;
 835}
 836
 837void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
 838{
 839	struct jbd2_journal_block_tail *tail;
 840	__u32 csum;
 841
 842	if (!jbd2_journal_has_csum_v2or3(j))
 843		return;
 844
 845	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
 846			sizeof(struct jbd2_journal_block_tail));
 847	tail->t_checksum = 0;
 848	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 849	tail->t_checksum = cpu_to_be32(csum);
 850}
 851
 852/*
 853 * Return tid of the oldest transaction in the journal and block in the journal
 854 * where the transaction starts.
 855 *
 856 * If the journal is now empty, return which will be the next transaction ID
 857 * we will write and where will that transaction start.
 858 *
 859 * The return value is 0 if journal tail cannot be pushed any further, 1 if
 860 * it can.
 861 */
 862int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
 863			      unsigned long *block)
 864{
 865	transaction_t *transaction;
 866	int ret;
 867
 868	read_lock(&journal->j_state_lock);
 869	spin_lock(&journal->j_list_lock);
 870	transaction = journal->j_checkpoint_transactions;
 871	if (transaction) {
 872		*tid = transaction->t_tid;
 873		*block = transaction->t_log_start;
 874	} else if ((transaction = journal->j_committing_transaction) != NULL) {
 875		*tid = transaction->t_tid;
 876		*block = transaction->t_log_start;
 877	} else if ((transaction = journal->j_running_transaction) != NULL) {
 878		*tid = transaction->t_tid;
 879		*block = journal->j_head;
 880	} else {
 881		*tid = journal->j_transaction_sequence;
 882		*block = journal->j_head;
 883	}
 884	ret = tid_gt(*tid, journal->j_tail_sequence);
 885	spin_unlock(&journal->j_list_lock);
 886	read_unlock(&journal->j_state_lock);
 887
 888	return ret;
 889}
 890
 891/*
 892 * Update information in journal structure and in on disk journal superblock
 893 * about log tail. This function does not check whether information passed in
 894 * really pushes log tail further. It's responsibility of the caller to make
 895 * sure provided log tail information is valid (e.g. by holding
 896 * j_checkpoint_mutex all the time between computing log tail and calling this
 897 * function as is the case with jbd2_cleanup_journal_tail()).
 898 *
 899 * Requires j_checkpoint_mutex
 900 */
 901int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 902{
 903	unsigned long freed;
 904	int ret;
 905
 906	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 907
 908	/*
 909	 * We cannot afford for write to remain in drive's caches since as
 910	 * soon as we update j_tail, next transaction can start reusing journal
 911	 * space and if we lose sb update during power failure we'd replay
 912	 * old transaction with possibly newly overwritten data.
 913	 */
 914	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
 
 915	if (ret)
 916		goto out;
 917
 918	write_lock(&journal->j_state_lock);
 919	freed = block - journal->j_tail;
 920	if (block < journal->j_tail)
 921		freed += journal->j_last - journal->j_first;
 922
 923	trace_jbd2_update_log_tail(journal, tid, block, freed);
 924	jbd_debug(1,
 925		  "Cleaning journal tail from %d to %d (offset %lu), "
 926		  "freeing %lu\n",
 927		  journal->j_tail_sequence, tid, block, freed);
 928
 929	journal->j_free += freed;
 930	journal->j_tail_sequence = tid;
 931	journal->j_tail = block;
 932	write_unlock(&journal->j_state_lock);
 933
 934out:
 935	return ret;
 936}
 937
 938/*
 939 * This is a variaon of __jbd2_update_log_tail which checks for validity of
 940 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
 941 * with other threads updating log tail.
 942 */
 943void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 944{
 945	mutex_lock(&journal->j_checkpoint_mutex);
 946	if (tid_gt(tid, journal->j_tail_sequence))
 947		__jbd2_update_log_tail(journal, tid, block);
 948	mutex_unlock(&journal->j_checkpoint_mutex);
 949}
 950
 951struct jbd2_stats_proc_session {
 952	journal_t *journal;
 953	struct transaction_stats_s *stats;
 954	int start;
 955	int max;
 956};
 957
 958static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
 959{
 960	return *pos ? NULL : SEQ_START_TOKEN;
 961}
 962
 963static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
 964{
 965	return NULL;
 966}
 967
 968static int jbd2_seq_info_show(struct seq_file *seq, void *v)
 969{
 970	struct jbd2_stats_proc_session *s = seq->private;
 971
 972	if (v != SEQ_START_TOKEN)
 973		return 0;
 974	seq_printf(seq, "%lu transactions (%lu requested), "
 975		   "each up to %u blocks\n",
 976		   s->stats->ts_tid, s->stats->ts_requested,
 977		   s->journal->j_max_transaction_buffers);
 978	if (s->stats->ts_tid == 0)
 979		return 0;
 980	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
 981	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
 982	seq_printf(seq, "  %ums request delay\n",
 983	    (s->stats->ts_requested == 0) ? 0 :
 984	    jiffies_to_msecs(s->stats->run.rs_request_delay /
 985			     s->stats->ts_requested));
 986	seq_printf(seq, "  %ums running transaction\n",
 987	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
 988	seq_printf(seq, "  %ums transaction was being locked\n",
 989	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
 990	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
 991	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
 992	seq_printf(seq, "  %ums logging transaction\n",
 993	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
 994	seq_printf(seq, "  %lluus average transaction commit time\n",
 995		   div_u64(s->journal->j_average_commit_time, 1000));
 996	seq_printf(seq, "  %lu handles per transaction\n",
 997	    s->stats->run.rs_handle_count / s->stats->ts_tid);
 998	seq_printf(seq, "  %lu blocks per transaction\n",
 999	    s->stats->run.rs_blocks / s->stats->ts_tid);
1000	seq_printf(seq, "  %lu logged blocks per transaction\n",
1001	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1002	return 0;
1003}
1004
1005static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1006{
1007}
1008
1009static const struct seq_operations jbd2_seq_info_ops = {
1010	.start  = jbd2_seq_info_start,
1011	.next   = jbd2_seq_info_next,
1012	.stop   = jbd2_seq_info_stop,
1013	.show   = jbd2_seq_info_show,
1014};
1015
1016static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1017{
1018	journal_t *journal = PDE_DATA(inode);
1019	struct jbd2_stats_proc_session *s;
1020	int rc, size;
1021
1022	s = kmalloc(sizeof(*s), GFP_KERNEL);
1023	if (s == NULL)
1024		return -ENOMEM;
1025	size = sizeof(struct transaction_stats_s);
1026	s->stats = kmalloc(size, GFP_KERNEL);
1027	if (s->stats == NULL) {
1028		kfree(s);
1029		return -ENOMEM;
1030	}
1031	spin_lock(&journal->j_history_lock);
1032	memcpy(s->stats, &journal->j_stats, size);
1033	s->journal = journal;
1034	spin_unlock(&journal->j_history_lock);
1035
1036	rc = seq_open(file, &jbd2_seq_info_ops);
1037	if (rc == 0) {
1038		struct seq_file *m = file->private_data;
1039		m->private = s;
1040	} else {
1041		kfree(s->stats);
1042		kfree(s);
1043	}
1044	return rc;
1045
1046}
1047
1048static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1049{
1050	struct seq_file *seq = file->private_data;
1051	struct jbd2_stats_proc_session *s = seq->private;
1052	kfree(s->stats);
1053	kfree(s);
1054	return seq_release(inode, file);
1055}
1056
1057static const struct file_operations jbd2_seq_info_fops = {
1058	.owner		= THIS_MODULE,
1059	.open           = jbd2_seq_info_open,
1060	.read           = seq_read,
1061	.llseek         = seq_lseek,
1062	.release        = jbd2_seq_info_release,
1063};
1064
1065static struct proc_dir_entry *proc_jbd2_stats;
1066
1067static void jbd2_stats_proc_init(journal_t *journal)
1068{
1069	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1070	if (journal->j_proc_entry) {
1071		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1072				 &jbd2_seq_info_fops, journal);
1073	}
1074}
1075
1076static void jbd2_stats_proc_exit(journal_t *journal)
1077{
1078	remove_proc_entry("info", journal->j_proc_entry);
1079	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1080}
1081
1082/*
1083 * Management for journal control blocks: functions to create and
1084 * destroy journal_t structures, and to initialise and read existing
1085 * journal blocks from disk.  */
1086
1087/* First: create and setup a journal_t object in memory.  We initialise
1088 * very few fields yet: that has to wait until we have created the
1089 * journal structures from from scratch, or loaded them from disk. */
1090
1091static journal_t * journal_init_common (void)
 
 
1092{
 
1093	journal_t *journal;
1094	int err;
 
 
1095
1096	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1097	if (!journal)
1098		return NULL;
1099
1100	init_waitqueue_head(&journal->j_wait_transaction_locked);
1101	init_waitqueue_head(&journal->j_wait_done_commit);
1102	init_waitqueue_head(&journal->j_wait_commit);
1103	init_waitqueue_head(&journal->j_wait_updates);
1104	init_waitqueue_head(&journal->j_wait_reserved);
1105	mutex_init(&journal->j_barrier);
1106	mutex_init(&journal->j_checkpoint_mutex);
1107	spin_lock_init(&journal->j_revoke_lock);
1108	spin_lock_init(&journal->j_list_lock);
1109	rwlock_init(&journal->j_state_lock);
1110
1111	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1112	journal->j_min_batch_time = 0;
1113	journal->j_max_batch_time = 15000; /* 15ms */
1114	atomic_set(&journal->j_reserved_credits, 0);
1115
1116	/* The journal is marked for error until we succeed with recovery! */
1117	journal->j_flags = JBD2_ABORT;
1118
1119	/* Set up a default-sized revoke table for the new mount. */
1120	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1121	if (err) {
1122		kfree(journal);
1123		return NULL;
1124	}
1125
1126	spin_lock_init(&journal->j_history_lock);
1127
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1128	return journal;
 
 
 
 
 
 
1129}
1130
1131/* jbd2_journal_init_dev and jbd2_journal_init_inode:
1132 *
1133 * Create a journal structure assigned some fixed set of disk blocks to
1134 * the journal.  We don't actually touch those disk blocks yet, but we
1135 * need to set up all of the mapping information to tell the journaling
1136 * system where the journal blocks are.
1137 *
1138 */
1139
1140/**
1141 *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1142 *  @bdev: Block device on which to create the journal
1143 *  @fs_dev: Device which hold journalled filesystem for this journal.
1144 *  @start: Block nr Start of journal.
1145 *  @len:  Length of the journal in blocks.
1146 *  @blocksize: blocksize of journalling device
1147 *
1148 *  Returns: a newly created journal_t *
1149 *
1150 *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1151 *  range of blocks on an arbitrary block device.
1152 *
1153 */
1154journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1155			struct block_device *fs_dev,
1156			unsigned long long start, int len, int blocksize)
1157{
1158	journal_t *journal = journal_init_common();
1159	struct buffer_head *bh;
1160	int n;
1161
 
1162	if (!journal)
1163		return NULL;
1164
1165	/* journal descriptor can store up to n blocks -bzzz */
1166	journal->j_blocksize = blocksize;
1167	journal->j_dev = bdev;
1168	journal->j_fs_dev = fs_dev;
1169	journal->j_blk_offset = start;
1170	journal->j_maxlen = len;
1171	bdevname(journal->j_dev, journal->j_devname);
1172	strreplace(journal->j_devname, '/', '!');
1173	jbd2_stats_proc_init(journal);
1174	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1175	journal->j_wbufsize = n;
1176	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1177	if (!journal->j_wbuf) {
1178		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1179			__func__);
1180		goto out_err;
1181	}
1182
1183	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1184	if (!bh) {
1185		printk(KERN_ERR
1186		       "%s: Cannot get buffer for journal superblock\n",
1187		       __func__);
1188		goto out_err;
1189	}
1190	journal->j_sb_buffer = bh;
1191	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1192
1193	return journal;
1194out_err:
1195	kfree(journal->j_wbuf);
1196	jbd2_stats_proc_exit(journal);
1197	kfree(journal);
1198	return NULL;
1199}
1200
1201/**
1202 *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1203 *  @inode: An inode to create the journal in
1204 *
1205 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1206 * the journal.  The inode must exist already, must support bmap() and
1207 * must have all data blocks preallocated.
1208 */
1209journal_t * jbd2_journal_init_inode (struct inode *inode)
1210{
1211	struct buffer_head *bh;
1212	journal_t *journal = journal_init_common();
1213	char *p;
1214	int err;
1215	int n;
1216	unsigned long long blocknr;
1217
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1218	if (!journal)
1219		return NULL;
1220
1221	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1222	journal->j_inode = inode;
1223	bdevname(journal->j_dev, journal->j_devname);
1224	p = strreplace(journal->j_devname, '/', '!');
1225	sprintf(p, "-%lu", journal->j_inode->i_ino);
1226	jbd_debug(1,
1227		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1228		  journal, inode->i_sb->s_id, inode->i_ino,
1229		  (long long) inode->i_size,
1230		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1231
1232	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1233	journal->j_blocksize = inode->i_sb->s_blocksize;
1234	jbd2_stats_proc_init(journal);
1235
1236	/* journal descriptor can store up to n blocks -bzzz */
1237	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1238	journal->j_wbufsize = n;
1239	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1240	if (!journal->j_wbuf) {
1241		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1242			__func__);
1243		goto out_err;
1244	}
1245
1246	err = jbd2_journal_bmap(journal, 0, &blocknr);
1247	/* If that failed, give up */
1248	if (err) {
1249		printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1250		       __func__);
1251		goto out_err;
1252	}
1253
1254	bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1255	if (!bh) {
1256		printk(KERN_ERR
1257		       "%s: Cannot get buffer for journal superblock\n",
1258		       __func__);
1259		goto out_err;
1260	}
1261	journal->j_sb_buffer = bh;
1262	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1263
1264	return journal;
1265out_err:
1266	kfree(journal->j_wbuf);
1267	jbd2_stats_proc_exit(journal);
1268	kfree(journal);
1269	return NULL;
1270}
1271
1272/*
1273 * If the journal init or create aborts, we need to mark the journal
1274 * superblock as being NULL to prevent the journal destroy from writing
1275 * back a bogus superblock.
1276 */
1277static void journal_fail_superblock (journal_t *journal)
1278{
1279	struct buffer_head *bh = journal->j_sb_buffer;
1280	brelse(bh);
1281	journal->j_sb_buffer = NULL;
1282}
1283
1284/*
1285 * Given a journal_t structure, initialise the various fields for
1286 * startup of a new journaling session.  We use this both when creating
1287 * a journal, and after recovering an old journal to reset it for
1288 * subsequent use.
1289 */
1290
1291static int journal_reset(journal_t *journal)
1292{
1293	journal_superblock_t *sb = journal->j_superblock;
1294	unsigned long long first, last;
1295
1296	first = be32_to_cpu(sb->s_first);
1297	last = be32_to_cpu(sb->s_maxlen);
1298	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1299		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1300		       first, last);
1301		journal_fail_superblock(journal);
1302		return -EINVAL;
1303	}
1304
1305	journal->j_first = first;
1306	journal->j_last = last;
1307
1308	journal->j_head = first;
1309	journal->j_tail = first;
1310	journal->j_free = last - first;
1311
1312	journal->j_tail_sequence = journal->j_transaction_sequence;
1313	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1314	journal->j_commit_request = journal->j_commit_sequence;
1315
1316	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1317
1318	/*
1319	 * As a special case, if the on-disk copy is already marked as needing
1320	 * no recovery (s_start == 0), then we can safely defer the superblock
1321	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1322	 * attempting a write to a potential-readonly device.
1323	 */
1324	if (sb->s_start == 0) {
1325		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1326			"(start %ld, seq %d, errno %d)\n",
1327			journal->j_tail, journal->j_tail_sequence,
1328			journal->j_errno);
1329		journal->j_flags |= JBD2_FLUSHED;
1330	} else {
1331		/* Lock here to make assertions happy... */
1332		mutex_lock(&journal->j_checkpoint_mutex);
1333		/*
1334		 * Update log tail information. We use WRITE_FUA since new
1335		 * transaction will start reusing journal space and so we
1336		 * must make sure information about current log tail is on
1337		 * disk before that.
1338		 */
1339		jbd2_journal_update_sb_log_tail(journal,
1340						journal->j_tail_sequence,
1341						journal->j_tail,
1342						WRITE_FUA);
1343		mutex_unlock(&journal->j_checkpoint_mutex);
1344	}
1345	return jbd2_journal_start_thread(journal);
1346}
1347
1348static int jbd2_write_superblock(journal_t *journal, int write_op)
1349{
1350	struct buffer_head *bh = journal->j_sb_buffer;
1351	journal_superblock_t *sb = journal->j_superblock;
1352	int ret;
1353
1354	trace_jbd2_write_superblock(journal, write_op);
1355	if (!(journal->j_flags & JBD2_BARRIER))
1356		write_op &= ~(REQ_FUA | REQ_FLUSH);
1357	lock_buffer(bh);
1358	if (buffer_write_io_error(bh)) {
1359		/*
1360		 * Oh, dear.  A previous attempt to write the journal
1361		 * superblock failed.  This could happen because the
1362		 * USB device was yanked out.  Or it could happen to
1363		 * be a transient write error and maybe the block will
1364		 * be remapped.  Nothing we can do but to retry the
1365		 * write and hope for the best.
1366		 */
1367		printk(KERN_ERR "JBD2: previous I/O error detected "
1368		       "for journal superblock update for %s.\n",
1369		       journal->j_devname);
1370		clear_buffer_write_io_error(bh);
1371		set_buffer_uptodate(bh);
1372	}
1373	jbd2_superblock_csum_set(journal, sb);
1374	get_bh(bh);
1375	bh->b_end_io = end_buffer_write_sync;
1376	ret = submit_bh(write_op, bh);
1377	wait_on_buffer(bh);
1378	if (buffer_write_io_error(bh)) {
1379		clear_buffer_write_io_error(bh);
1380		set_buffer_uptodate(bh);
1381		ret = -EIO;
1382	}
1383	if (ret) {
1384		printk(KERN_ERR "JBD2: Error %d detected when updating "
1385		       "journal superblock for %s.\n", ret,
1386		       journal->j_devname);
1387		jbd2_journal_abort(journal, ret);
1388	}
1389
1390	return ret;
1391}
1392
1393/**
1394 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1395 * @journal: The journal to update.
1396 * @tail_tid: TID of the new transaction at the tail of the log
1397 * @tail_block: The first block of the transaction at the tail of the log
1398 * @write_op: With which operation should we write the journal sb
1399 *
1400 * Update a journal's superblock information about log tail and write it to
1401 * disk, waiting for the IO to complete.
1402 */
1403int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1404				     unsigned long tail_block, int write_op)
1405{
1406	journal_superblock_t *sb = journal->j_superblock;
1407	int ret;
1408
 
 
 
1409	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1410	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1411		  tail_block, tail_tid);
1412
1413	sb->s_sequence = cpu_to_be32(tail_tid);
1414	sb->s_start    = cpu_to_be32(tail_block);
1415
1416	ret = jbd2_write_superblock(journal, write_op);
1417	if (ret)
1418		goto out;
1419
1420	/* Log is no longer empty */
1421	write_lock(&journal->j_state_lock);
1422	WARN_ON(!sb->s_sequence);
1423	journal->j_flags &= ~JBD2_FLUSHED;
1424	write_unlock(&journal->j_state_lock);
1425
1426out:
1427	return ret;
1428}
1429
1430/**
1431 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1432 * @journal: The journal to update.
1433 * @write_op: With which operation should we write the journal sb
1434 *
1435 * Update a journal's dynamic superblock fields to show that journal is empty.
1436 * Write updated superblock to disk waiting for IO to complete.
1437 */
1438static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1439{
1440	journal_superblock_t *sb = journal->j_superblock;
1441
1442	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1443	read_lock(&journal->j_state_lock);
1444	/* Is it already empty? */
1445	if (sb->s_start == 0) {
1446		read_unlock(&journal->j_state_lock);
1447		return;
1448	}
1449	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1450		  journal->j_tail_sequence);
1451
1452	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1453	sb->s_start    = cpu_to_be32(0);
1454	read_unlock(&journal->j_state_lock);
1455
1456	jbd2_write_superblock(journal, write_op);
1457
1458	/* Log is no longer empty */
1459	write_lock(&journal->j_state_lock);
1460	journal->j_flags |= JBD2_FLUSHED;
1461	write_unlock(&journal->j_state_lock);
1462}
1463
1464
1465/**
1466 * jbd2_journal_update_sb_errno() - Update error in the journal.
1467 * @journal: The journal to update.
1468 *
1469 * Update a journal's errno.  Write updated superblock to disk waiting for IO
1470 * to complete.
1471 */
1472void jbd2_journal_update_sb_errno(journal_t *journal)
1473{
1474	journal_superblock_t *sb = journal->j_superblock;
 
1475
1476	read_lock(&journal->j_state_lock);
1477	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1478		  journal->j_errno);
1479	sb->s_errno    = cpu_to_be32(journal->j_errno);
1480	read_unlock(&journal->j_state_lock);
 
 
 
 
1481
1482	jbd2_write_superblock(journal, WRITE_FUA);
1483}
1484EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1485
1486/*
1487 * Read the superblock for a given journal, performing initial
1488 * validation of the format.
1489 */
1490static int journal_get_superblock(journal_t *journal)
1491{
1492	struct buffer_head *bh;
1493	journal_superblock_t *sb;
1494	int err = -EIO;
1495
1496	bh = journal->j_sb_buffer;
1497
1498	J_ASSERT(bh != NULL);
1499	if (!buffer_uptodate(bh)) {
1500		ll_rw_block(READ, 1, &bh);
1501		wait_on_buffer(bh);
1502		if (!buffer_uptodate(bh)) {
1503			printk(KERN_ERR
1504				"JBD2: IO error reading journal superblock\n");
1505			goto out;
1506		}
1507	}
1508
1509	if (buffer_verified(bh))
1510		return 0;
1511
1512	sb = journal->j_superblock;
1513
1514	err = -EINVAL;
1515
1516	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1517	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1518		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1519		goto out;
1520	}
1521
1522	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1523	case JBD2_SUPERBLOCK_V1:
1524		journal->j_format_version = 1;
1525		break;
1526	case JBD2_SUPERBLOCK_V2:
1527		journal->j_format_version = 2;
1528		break;
1529	default:
1530		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1531		goto out;
1532	}
1533
1534	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1535		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1536	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1537		printk(KERN_WARNING "JBD2: journal file too short\n");
1538		goto out;
1539	}
1540
1541	if (be32_to_cpu(sb->s_first) == 0 ||
1542	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1543		printk(KERN_WARNING
1544			"JBD2: Invalid start block of journal: %u\n",
1545			be32_to_cpu(sb->s_first));
1546		goto out;
1547	}
1548
1549	if (jbd2_has_feature_csum2(journal) &&
1550	    jbd2_has_feature_csum3(journal)) {
1551		/* Can't have checksum v2 and v3 at the same time! */
1552		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1553		       "at the same time!\n");
1554		goto out;
1555	}
1556
1557	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1558	    jbd2_has_feature_checksum(journal)) {
1559		/* Can't have checksum v1 and v2 on at the same time! */
1560		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1561		       "at the same time!\n");
1562		goto out;
1563	}
1564
1565	if (!jbd2_verify_csum_type(journal, sb)) {
1566		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1567		goto out;
1568	}
1569
1570	/* Load the checksum driver */
1571	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1572		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1573		if (IS_ERR(journal->j_chksum_driver)) {
1574			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1575			err = PTR_ERR(journal->j_chksum_driver);
1576			journal->j_chksum_driver = NULL;
1577			goto out;
1578		}
1579	}
1580
1581	/* Check superblock checksum */
1582	if (!jbd2_superblock_csum_verify(journal, sb)) {
1583		printk(KERN_ERR "JBD2: journal checksum error\n");
1584		err = -EFSBADCRC;
1585		goto out;
1586	}
1587
1588	/* Precompute checksum seed for all metadata */
1589	if (jbd2_journal_has_csum_v2or3(journal))
1590		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1591						   sizeof(sb->s_uuid));
1592
1593	set_buffer_verified(bh);
1594
1595	return 0;
1596
1597out:
1598	journal_fail_superblock(journal);
1599	return err;
1600}
1601
1602/*
1603 * Load the on-disk journal superblock and read the key fields into the
1604 * journal_t.
1605 */
1606
1607static int load_superblock(journal_t *journal)
1608{
1609	int err;
1610	journal_superblock_t *sb;
1611
1612	err = journal_get_superblock(journal);
1613	if (err)
1614		return err;
1615
1616	sb = journal->j_superblock;
1617
1618	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1619	journal->j_tail = be32_to_cpu(sb->s_start);
1620	journal->j_first = be32_to_cpu(sb->s_first);
1621	journal->j_last = be32_to_cpu(sb->s_maxlen);
1622	journal->j_errno = be32_to_cpu(sb->s_errno);
1623
1624	return 0;
1625}
1626
1627
1628/**
1629 * int jbd2_journal_load() - Read journal from disk.
1630 * @journal: Journal to act on.
1631 *
1632 * Given a journal_t structure which tells us which disk blocks contain
1633 * a journal, read the journal from disk to initialise the in-memory
1634 * structures.
1635 */
1636int jbd2_journal_load(journal_t *journal)
1637{
1638	int err;
1639	journal_superblock_t *sb;
1640
1641	err = load_superblock(journal);
1642	if (err)
1643		return err;
1644
1645	sb = journal->j_superblock;
1646	/* If this is a V2 superblock, then we have to check the
1647	 * features flags on it. */
1648
1649	if (journal->j_format_version >= 2) {
1650		if ((sb->s_feature_ro_compat &
1651		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1652		    (sb->s_feature_incompat &
1653		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1654			printk(KERN_WARNING
1655				"JBD2: Unrecognised features on journal\n");
1656			return -EINVAL;
1657		}
1658	}
1659
1660	/*
1661	 * Create a slab for this blocksize
1662	 */
1663	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1664	if (err)
1665		return err;
1666
1667	/* Let the recovery code check whether it needs to recover any
1668	 * data from the journal. */
1669	if (jbd2_journal_recover(journal))
1670		goto recovery_error;
1671
1672	if (journal->j_failed_commit) {
1673		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1674		       "is corrupt.\n", journal->j_failed_commit,
1675		       journal->j_devname);
1676		return -EFSCORRUPTED;
1677	}
1678
1679	/* OK, we've finished with the dynamic journal bits:
1680	 * reinitialise the dynamic contents of the superblock in memory
1681	 * and reset them on disk. */
1682	if (journal_reset(journal))
1683		goto recovery_error;
1684
1685	journal->j_flags &= ~JBD2_ABORT;
1686	journal->j_flags |= JBD2_LOADED;
1687	return 0;
1688
1689recovery_error:
1690	printk(KERN_WARNING "JBD2: recovery failed\n");
1691	return -EIO;
1692}
1693
1694/**
1695 * void jbd2_journal_destroy() - Release a journal_t structure.
1696 * @journal: Journal to act on.
1697 *
1698 * Release a journal_t structure once it is no longer in use by the
1699 * journaled object.
1700 * Return <0 if we couldn't clean up the journal.
1701 */
1702int jbd2_journal_destroy(journal_t *journal)
1703{
1704	int err = 0;
1705
1706	/* Wait for the commit thread to wake up and die. */
1707	journal_kill_thread(journal);
1708
1709	/* Force a final log commit */
1710	if (journal->j_running_transaction)
1711		jbd2_journal_commit_transaction(journal);
1712
1713	/* Force any old transactions to disk */
1714
1715	/* Totally anal locking here... */
1716	spin_lock(&journal->j_list_lock);
1717	while (journal->j_checkpoint_transactions != NULL) {
1718		spin_unlock(&journal->j_list_lock);
1719		mutex_lock(&journal->j_checkpoint_mutex);
1720		err = jbd2_log_do_checkpoint(journal);
1721		mutex_unlock(&journal->j_checkpoint_mutex);
1722		/*
1723		 * If checkpointing failed, just free the buffers to avoid
1724		 * looping forever
1725		 */
1726		if (err) {
1727			jbd2_journal_destroy_checkpoint(journal);
1728			spin_lock(&journal->j_list_lock);
1729			break;
1730		}
1731		spin_lock(&journal->j_list_lock);
1732	}
1733
1734	J_ASSERT(journal->j_running_transaction == NULL);
1735	J_ASSERT(journal->j_committing_transaction == NULL);
1736	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1737	spin_unlock(&journal->j_list_lock);
1738
1739	if (journal->j_sb_buffer) {
1740		if (!is_journal_aborted(journal)) {
1741			mutex_lock(&journal->j_checkpoint_mutex);
1742
1743			write_lock(&journal->j_state_lock);
1744			journal->j_tail_sequence =
1745				++journal->j_transaction_sequence;
1746			write_unlock(&journal->j_state_lock);
1747
1748			jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
 
1749			mutex_unlock(&journal->j_checkpoint_mutex);
1750		} else
1751			err = -EIO;
1752		brelse(journal->j_sb_buffer);
1753	}
1754
1755	if (journal->j_proc_entry)
1756		jbd2_stats_proc_exit(journal);
1757	iput(journal->j_inode);
1758	if (journal->j_revoke)
1759		jbd2_journal_destroy_revoke(journal);
1760	if (journal->j_chksum_driver)
1761		crypto_free_shash(journal->j_chksum_driver);
1762	kfree(journal->j_wbuf);
1763	kfree(journal);
1764
1765	return err;
1766}
1767
1768
1769/**
1770 *int jbd2_journal_check_used_features () - Check if features specified are used.
1771 * @journal: Journal to check.
1772 * @compat: bitmask of compatible features
1773 * @ro: bitmask of features that force read-only mount
1774 * @incompat: bitmask of incompatible features
1775 *
1776 * Check whether the journal uses all of a given set of
1777 * features.  Return true (non-zero) if it does.
1778 **/
1779
1780int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1781				 unsigned long ro, unsigned long incompat)
1782{
1783	journal_superblock_t *sb;
1784
1785	if (!compat && !ro && !incompat)
1786		return 1;
1787	/* Load journal superblock if it is not loaded yet. */
1788	if (journal->j_format_version == 0 &&
1789	    journal_get_superblock(journal) != 0)
1790		return 0;
1791	if (journal->j_format_version == 1)
1792		return 0;
1793
1794	sb = journal->j_superblock;
1795
1796	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1797	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1798	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1799		return 1;
1800
1801	return 0;
1802}
1803
1804/**
1805 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1806 * @journal: Journal to check.
1807 * @compat: bitmask of compatible features
1808 * @ro: bitmask of features that force read-only mount
1809 * @incompat: bitmask of incompatible features
1810 *
1811 * Check whether the journaling code supports the use of
1812 * all of a given set of features on this journal.  Return true
1813 * (non-zero) if it can. */
1814
1815int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1816				      unsigned long ro, unsigned long incompat)
1817{
1818	if (!compat && !ro && !incompat)
1819		return 1;
1820
1821	/* We can support any known requested features iff the
1822	 * superblock is in version 2.  Otherwise we fail to support any
1823	 * extended sb features. */
1824
1825	if (journal->j_format_version != 2)
1826		return 0;
1827
1828	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1829	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1830	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1831		return 1;
1832
1833	return 0;
1834}
1835
1836/**
1837 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1838 * @journal: Journal to act on.
1839 * @compat: bitmask of compatible features
1840 * @ro: bitmask of features that force read-only mount
1841 * @incompat: bitmask of incompatible features
1842 *
1843 * Mark a given journal feature as present on the
1844 * superblock.  Returns true if the requested features could be set.
1845 *
1846 */
1847
1848int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1849			  unsigned long ro, unsigned long incompat)
1850{
1851#define INCOMPAT_FEATURE_ON(f) \
1852		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1853#define COMPAT_FEATURE_ON(f) \
1854		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1855	journal_superblock_t *sb;
1856
1857	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1858		return 1;
1859
1860	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1861		return 0;
1862
1863	/* If enabling v2 checksums, turn on v3 instead */
1864	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1865		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1866		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1867	}
1868
1869	/* Asking for checksumming v3 and v1?  Only give them v3. */
1870	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1871	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1872		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1873
1874	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1875		  compat, ro, incompat);
1876
1877	sb = journal->j_superblock;
1878
1879	/* If enabling v3 checksums, update superblock */
1880	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1881		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1882		sb->s_feature_compat &=
1883			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1884
1885		/* Load the checksum driver */
1886		if (journal->j_chksum_driver == NULL) {
1887			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1888								      0, 0);
1889			if (IS_ERR(journal->j_chksum_driver)) {
1890				printk(KERN_ERR "JBD2: Cannot load crc32c "
1891				       "driver.\n");
1892				journal->j_chksum_driver = NULL;
1893				return 0;
1894			}
1895
1896			/* Precompute checksum seed for all metadata */
1897			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1898							   sb->s_uuid,
1899							   sizeof(sb->s_uuid));
1900		}
1901	}
1902
1903	/* If enabling v1 checksums, downgrade superblock */
1904	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1905		sb->s_feature_incompat &=
1906			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1907				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
1908
1909	sb->s_feature_compat    |= cpu_to_be32(compat);
1910	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1911	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1912
1913	return 1;
1914#undef COMPAT_FEATURE_ON
1915#undef INCOMPAT_FEATURE_ON
1916}
1917
1918/*
1919 * jbd2_journal_clear_features () - Clear a given journal feature in the
1920 * 				    superblock
1921 * @journal: Journal to act on.
1922 * @compat: bitmask of compatible features
1923 * @ro: bitmask of features that force read-only mount
1924 * @incompat: bitmask of incompatible features
1925 *
1926 * Clear a given journal feature as present on the
1927 * superblock.
1928 */
1929void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1930				unsigned long ro, unsigned long incompat)
1931{
1932	journal_superblock_t *sb;
1933
1934	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1935		  compat, ro, incompat);
1936
1937	sb = journal->j_superblock;
1938
1939	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1940	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1941	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1942}
1943EXPORT_SYMBOL(jbd2_journal_clear_features);
1944
1945/**
1946 * int jbd2_journal_flush () - Flush journal
1947 * @journal: Journal to act on.
1948 *
1949 * Flush all data for a given journal to disk and empty the journal.
1950 * Filesystems can use this when remounting readonly to ensure that
1951 * recovery does not need to happen on remount.
1952 */
1953
1954int jbd2_journal_flush(journal_t *journal)
1955{
1956	int err = 0;
1957	transaction_t *transaction = NULL;
1958
1959	write_lock(&journal->j_state_lock);
1960
1961	/* Force everything buffered to the log... */
1962	if (journal->j_running_transaction) {
1963		transaction = journal->j_running_transaction;
1964		__jbd2_log_start_commit(journal, transaction->t_tid);
1965	} else if (journal->j_committing_transaction)
1966		transaction = journal->j_committing_transaction;
1967
1968	/* Wait for the log commit to complete... */
1969	if (transaction) {
1970		tid_t tid = transaction->t_tid;
1971
1972		write_unlock(&journal->j_state_lock);
1973		jbd2_log_wait_commit(journal, tid);
1974	} else {
1975		write_unlock(&journal->j_state_lock);
1976	}
1977
1978	/* ...and flush everything in the log out to disk. */
1979	spin_lock(&journal->j_list_lock);
1980	while (!err && journal->j_checkpoint_transactions != NULL) {
1981		spin_unlock(&journal->j_list_lock);
1982		mutex_lock(&journal->j_checkpoint_mutex);
1983		err = jbd2_log_do_checkpoint(journal);
1984		mutex_unlock(&journal->j_checkpoint_mutex);
1985		spin_lock(&journal->j_list_lock);
1986	}
1987	spin_unlock(&journal->j_list_lock);
1988
1989	if (is_journal_aborted(journal))
1990		return -EIO;
1991
1992	mutex_lock(&journal->j_checkpoint_mutex);
1993	if (!err) {
1994		err = jbd2_cleanup_journal_tail(journal);
1995		if (err < 0) {
1996			mutex_unlock(&journal->j_checkpoint_mutex);
1997			goto out;
1998		}
1999		err = 0;
2000	}
2001
2002	/* Finally, mark the journal as really needing no recovery.
2003	 * This sets s_start==0 in the underlying superblock, which is
2004	 * the magic code for a fully-recovered superblock.  Any future
2005	 * commits of data to the journal will restore the current
2006	 * s_start value. */
2007	jbd2_mark_journal_empty(journal, WRITE_FUA);
2008	mutex_unlock(&journal->j_checkpoint_mutex);
2009	write_lock(&journal->j_state_lock);
2010	J_ASSERT(!journal->j_running_transaction);
2011	J_ASSERT(!journal->j_committing_transaction);
2012	J_ASSERT(!journal->j_checkpoint_transactions);
2013	J_ASSERT(journal->j_head == journal->j_tail);
2014	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2015	write_unlock(&journal->j_state_lock);
2016out:
2017	return err;
2018}
2019
2020/**
2021 * int jbd2_journal_wipe() - Wipe journal contents
2022 * @journal: Journal to act on.
2023 * @write: flag (see below)
2024 *
2025 * Wipe out all of the contents of a journal, safely.  This will produce
2026 * a warning if the journal contains any valid recovery information.
2027 * Must be called between journal_init_*() and jbd2_journal_load().
2028 *
2029 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2030 * we merely suppress recovery.
2031 */
2032
2033int jbd2_journal_wipe(journal_t *journal, int write)
2034{
2035	int err = 0;
2036
2037	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2038
2039	err = load_superblock(journal);
2040	if (err)
2041		return err;
2042
2043	if (!journal->j_tail)
2044		goto no_recovery;
2045
2046	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2047		write ? "Clearing" : "Ignoring");
2048
2049	err = jbd2_journal_skip_recovery(journal);
2050	if (write) {
2051		/* Lock to make assertions happy... */
2052		mutex_lock(&journal->j_checkpoint_mutex);
2053		jbd2_mark_journal_empty(journal, WRITE_FUA);
2054		mutex_unlock(&journal->j_checkpoint_mutex);
2055	}
2056
2057 no_recovery:
2058	return err;
2059}
2060
2061/*
2062 * Journal abort has very specific semantics, which we describe
2063 * for journal abort.
2064 *
2065 * Two internal functions, which provide abort to the jbd layer
2066 * itself are here.
2067 */
2068
2069/*
2070 * Quick version for internal journal use (doesn't lock the journal).
2071 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2072 * and don't attempt to make any other journal updates.
2073 */
2074void __jbd2_journal_abort_hard(journal_t *journal)
2075{
2076	transaction_t *transaction;
2077
2078	if (journal->j_flags & JBD2_ABORT)
2079		return;
2080
2081	printk(KERN_ERR "Aborting journal on device %s.\n",
2082	       journal->j_devname);
2083
2084	write_lock(&journal->j_state_lock);
2085	journal->j_flags |= JBD2_ABORT;
2086	transaction = journal->j_running_transaction;
2087	if (transaction)
2088		__jbd2_log_start_commit(journal, transaction->t_tid);
2089	write_unlock(&journal->j_state_lock);
2090}
2091
2092/* Soft abort: record the abort error status in the journal superblock,
2093 * but don't do any other IO. */
2094static void __journal_abort_soft (journal_t *journal, int errno)
2095{
2096	if (journal->j_flags & JBD2_ABORT)
2097		return;
2098
2099	if (!journal->j_errno)
 
 
2100		journal->j_errno = errno;
2101
 
 
 
 
 
 
 
 
 
2102	__jbd2_journal_abort_hard(journal);
2103
2104	if (errno) {
2105		jbd2_journal_update_sb_errno(journal);
2106		write_lock(&journal->j_state_lock);
2107		journal->j_flags |= JBD2_REC_ERR;
2108		write_unlock(&journal->j_state_lock);
2109	}
2110}
2111
2112/**
2113 * void jbd2_journal_abort () - Shutdown the journal immediately.
2114 * @journal: the journal to shutdown.
2115 * @errno:   an error number to record in the journal indicating
2116 *           the reason for the shutdown.
2117 *
2118 * Perform a complete, immediate shutdown of the ENTIRE
2119 * journal (not of a single transaction).  This operation cannot be
2120 * undone without closing and reopening the journal.
2121 *
2122 * The jbd2_journal_abort function is intended to support higher level error
2123 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2124 * mode.
2125 *
2126 * Journal abort has very specific semantics.  Any existing dirty,
2127 * unjournaled buffers in the main filesystem will still be written to
2128 * disk by bdflush, but the journaling mechanism will be suspended
2129 * immediately and no further transaction commits will be honoured.
2130 *
2131 * Any dirty, journaled buffers will be written back to disk without
2132 * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2133 * filesystem, but we _do_ attempt to leave as much data as possible
2134 * behind for fsck to use for cleanup.
2135 *
2136 * Any attempt to get a new transaction handle on a journal which is in
2137 * ABORT state will just result in an -EROFS error return.  A
2138 * jbd2_journal_stop on an existing handle will return -EIO if we have
2139 * entered abort state during the update.
2140 *
2141 * Recursive transactions are not disturbed by journal abort until the
2142 * final jbd2_journal_stop, which will receive the -EIO error.
2143 *
2144 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2145 * which will be recorded (if possible) in the journal superblock.  This
2146 * allows a client to record failure conditions in the middle of a
2147 * transaction without having to complete the transaction to record the
2148 * failure to disk.  ext3_error, for example, now uses this
2149 * functionality.
2150 *
2151 * Errors which originate from within the journaling layer will NOT
2152 * supply an errno; a null errno implies that absolutely no further
2153 * writes are done to the journal (unless there are any already in
2154 * progress).
2155 *
2156 */
2157
2158void jbd2_journal_abort(journal_t *journal, int errno)
2159{
2160	__journal_abort_soft(journal, errno);
2161}
2162
2163/**
2164 * int jbd2_journal_errno () - returns the journal's error state.
2165 * @journal: journal to examine.
2166 *
2167 * This is the errno number set with jbd2_journal_abort(), the last
2168 * time the journal was mounted - if the journal was stopped
2169 * without calling abort this will be 0.
2170 *
2171 * If the journal has been aborted on this mount time -EROFS will
2172 * be returned.
2173 */
2174int jbd2_journal_errno(journal_t *journal)
2175{
2176	int err;
2177
2178	read_lock(&journal->j_state_lock);
2179	if (journal->j_flags & JBD2_ABORT)
2180		err = -EROFS;
2181	else
2182		err = journal->j_errno;
2183	read_unlock(&journal->j_state_lock);
2184	return err;
2185}
2186
2187/**
2188 * int jbd2_journal_clear_err () - clears the journal's error state
2189 * @journal: journal to act on.
2190 *
2191 * An error must be cleared or acked to take a FS out of readonly
2192 * mode.
2193 */
2194int jbd2_journal_clear_err(journal_t *journal)
2195{
2196	int err = 0;
2197
2198	write_lock(&journal->j_state_lock);
2199	if (journal->j_flags & JBD2_ABORT)
2200		err = -EROFS;
2201	else
2202		journal->j_errno = 0;
2203	write_unlock(&journal->j_state_lock);
2204	return err;
2205}
2206
2207/**
2208 * void jbd2_journal_ack_err() - Ack journal err.
2209 * @journal: journal to act on.
2210 *
2211 * An error must be cleared or acked to take a FS out of readonly
2212 * mode.
2213 */
2214void jbd2_journal_ack_err(journal_t *journal)
2215{
2216	write_lock(&journal->j_state_lock);
2217	if (journal->j_errno)
2218		journal->j_flags |= JBD2_ACK_ERR;
2219	write_unlock(&journal->j_state_lock);
2220}
2221
2222int jbd2_journal_blocks_per_page(struct inode *inode)
2223{
2224	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2225}
2226
2227/*
2228 * helper functions to deal with 32 or 64bit block numbers.
2229 */
2230size_t journal_tag_bytes(journal_t *journal)
2231{
2232	size_t sz;
2233
2234	if (jbd2_has_feature_csum3(journal))
2235		return sizeof(journal_block_tag3_t);
2236
2237	sz = sizeof(journal_block_tag_t);
2238
2239	if (jbd2_has_feature_csum2(journal))
2240		sz += sizeof(__u16);
2241
2242	if (jbd2_has_feature_64bit(journal))
2243		return sz;
2244	else
2245		return sz - sizeof(__u32);
2246}
2247
2248/*
2249 * JBD memory management
2250 *
2251 * These functions are used to allocate block-sized chunks of memory
2252 * used for making copies of buffer_head data.  Very often it will be
2253 * page-sized chunks of data, but sometimes it will be in
2254 * sub-page-size chunks.  (For example, 16k pages on Power systems
2255 * with a 4k block file system.)  For blocks smaller than a page, we
2256 * use a SLAB allocator.  There are slab caches for each block size,
2257 * which are allocated at mount time, if necessary, and we only free
2258 * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2259 * this reason we don't need to a mutex to protect access to
2260 * jbd2_slab[] allocating or releasing memory; only in
2261 * jbd2_journal_create_slab().
2262 */
2263#define JBD2_MAX_SLABS 8
2264static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2265
2266static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2267	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2268	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2269};
2270
2271
2272static void jbd2_journal_destroy_slabs(void)
2273{
2274	int i;
2275
2276	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2277		if (jbd2_slab[i])
2278			kmem_cache_destroy(jbd2_slab[i]);
2279		jbd2_slab[i] = NULL;
2280	}
2281}
2282
2283static int jbd2_journal_create_slab(size_t size)
2284{
2285	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2286	int i = order_base_2(size) - 10;
2287	size_t slab_size;
2288
2289	if (size == PAGE_SIZE)
2290		return 0;
2291
2292	if (i >= JBD2_MAX_SLABS)
2293		return -EINVAL;
2294
2295	if (unlikely(i < 0))
2296		i = 0;
2297	mutex_lock(&jbd2_slab_create_mutex);
2298	if (jbd2_slab[i]) {
2299		mutex_unlock(&jbd2_slab_create_mutex);
2300		return 0;	/* Already created */
2301	}
2302
2303	slab_size = 1 << (i+10);
2304	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2305					 slab_size, 0, NULL);
2306	mutex_unlock(&jbd2_slab_create_mutex);
2307	if (!jbd2_slab[i]) {
2308		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2309		return -ENOMEM;
2310	}
2311	return 0;
2312}
2313
2314static struct kmem_cache *get_slab(size_t size)
2315{
2316	int i = order_base_2(size) - 10;
2317
2318	BUG_ON(i >= JBD2_MAX_SLABS);
2319	if (unlikely(i < 0))
2320		i = 0;
2321	BUG_ON(jbd2_slab[i] == NULL);
2322	return jbd2_slab[i];
2323}
2324
2325void *jbd2_alloc(size_t size, gfp_t flags)
2326{
2327	void *ptr;
2328
2329	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2330
2331	flags |= __GFP_REPEAT;
2332	if (size == PAGE_SIZE)
2333		ptr = (void *)__get_free_pages(flags, 0);
2334	else if (size > PAGE_SIZE) {
2335		int order = get_order(size);
2336
2337		if (order < 3)
2338			ptr = (void *)__get_free_pages(flags, order);
2339		else
2340			ptr = vmalloc(size);
2341	} else
2342		ptr = kmem_cache_alloc(get_slab(size), flags);
 
 
2343
2344	/* Check alignment; SLUB has gotten this wrong in the past,
2345	 * and this can lead to user data corruption! */
2346	BUG_ON(((unsigned long) ptr) & (size-1));
2347
2348	return ptr;
2349}
2350
2351void jbd2_free(void *ptr, size_t size)
2352{
2353	if (size == PAGE_SIZE) {
2354		free_pages((unsigned long)ptr, 0);
2355		return;
2356	}
2357	if (size > PAGE_SIZE) {
2358		int order = get_order(size);
2359
2360		if (order < 3)
2361			free_pages((unsigned long)ptr, order);
2362		else
2363			vfree(ptr);
2364		return;
2365	}
2366	kmem_cache_free(get_slab(size), ptr);
2367};
2368
2369/*
2370 * Journal_head storage management
2371 */
2372static struct kmem_cache *jbd2_journal_head_cache;
2373#ifdef CONFIG_JBD2_DEBUG
2374static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2375#endif
2376
2377static int jbd2_journal_init_journal_head_cache(void)
2378{
2379	int retval;
2380
2381	J_ASSERT(jbd2_journal_head_cache == NULL);
2382	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2383				sizeof(struct journal_head),
2384				0,		/* offset */
2385				SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2386				NULL);		/* ctor */
2387	retval = 0;
2388	if (!jbd2_journal_head_cache) {
2389		retval = -ENOMEM;
2390		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2391	}
2392	return retval;
2393}
2394
2395static void jbd2_journal_destroy_journal_head_cache(void)
2396{
2397	if (jbd2_journal_head_cache) {
2398		kmem_cache_destroy(jbd2_journal_head_cache);
2399		jbd2_journal_head_cache = NULL;
2400	}
2401}
2402
2403/*
2404 * journal_head splicing and dicing
2405 */
2406static struct journal_head *journal_alloc_journal_head(void)
2407{
2408	struct journal_head *ret;
2409
2410#ifdef CONFIG_JBD2_DEBUG
2411	atomic_inc(&nr_journal_heads);
2412#endif
2413	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2414	if (!ret) {
2415		jbd_debug(1, "out of memory for journal_head\n");
2416		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2417		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2418				GFP_NOFS | __GFP_NOFAIL);
2419	}
2420	return ret;
2421}
2422
2423static void journal_free_journal_head(struct journal_head *jh)
2424{
2425#ifdef CONFIG_JBD2_DEBUG
2426	atomic_dec(&nr_journal_heads);
2427	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2428#endif
2429	kmem_cache_free(jbd2_journal_head_cache, jh);
2430}
2431
2432/*
2433 * A journal_head is attached to a buffer_head whenever JBD has an
2434 * interest in the buffer.
2435 *
2436 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2437 * is set.  This bit is tested in core kernel code where we need to take
2438 * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2439 * there.
2440 *
2441 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2442 *
2443 * When a buffer has its BH_JBD bit set it is immune from being released by
2444 * core kernel code, mainly via ->b_count.
2445 *
2446 * A journal_head is detached from its buffer_head when the journal_head's
2447 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2448 * transaction (b_cp_transaction) hold their references to b_jcount.
2449 *
2450 * Various places in the kernel want to attach a journal_head to a buffer_head
2451 * _before_ attaching the journal_head to a transaction.  To protect the
2452 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2453 * journal_head's b_jcount refcount by one.  The caller must call
2454 * jbd2_journal_put_journal_head() to undo this.
2455 *
2456 * So the typical usage would be:
2457 *
2458 *	(Attach a journal_head if needed.  Increments b_jcount)
2459 *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2460 *	...
2461 *      (Get another reference for transaction)
2462 *	jbd2_journal_grab_journal_head(bh);
2463 *	jh->b_transaction = xxx;
2464 *	(Put original reference)
2465 *	jbd2_journal_put_journal_head(jh);
2466 */
2467
2468/*
2469 * Give a buffer_head a journal_head.
2470 *
2471 * May sleep.
2472 */
2473struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2474{
2475	struct journal_head *jh;
2476	struct journal_head *new_jh = NULL;
2477
2478repeat:
2479	if (!buffer_jbd(bh))
2480		new_jh = journal_alloc_journal_head();
2481
2482	jbd_lock_bh_journal_head(bh);
2483	if (buffer_jbd(bh)) {
2484		jh = bh2jh(bh);
2485	} else {
2486		J_ASSERT_BH(bh,
2487			(atomic_read(&bh->b_count) > 0) ||
2488			(bh->b_page && bh->b_page->mapping));
2489
2490		if (!new_jh) {
2491			jbd_unlock_bh_journal_head(bh);
2492			goto repeat;
2493		}
2494
2495		jh = new_jh;
2496		new_jh = NULL;		/* We consumed it */
2497		set_buffer_jbd(bh);
2498		bh->b_private = jh;
2499		jh->b_bh = bh;
2500		get_bh(bh);
2501		BUFFER_TRACE(bh, "added journal_head");
2502	}
2503	jh->b_jcount++;
2504	jbd_unlock_bh_journal_head(bh);
2505	if (new_jh)
2506		journal_free_journal_head(new_jh);
2507	return bh->b_private;
2508}
2509
2510/*
2511 * Grab a ref against this buffer_head's journal_head.  If it ended up not
2512 * having a journal_head, return NULL
2513 */
2514struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2515{
2516	struct journal_head *jh = NULL;
2517
2518	jbd_lock_bh_journal_head(bh);
2519	if (buffer_jbd(bh)) {
2520		jh = bh2jh(bh);
2521		jh->b_jcount++;
2522	}
2523	jbd_unlock_bh_journal_head(bh);
2524	return jh;
2525}
2526
2527static void __journal_remove_journal_head(struct buffer_head *bh)
2528{
2529	struct journal_head *jh = bh2jh(bh);
2530
2531	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2532	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2533	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2534	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2535	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2536	J_ASSERT_BH(bh, buffer_jbd(bh));
2537	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2538	BUFFER_TRACE(bh, "remove journal_head");
2539	if (jh->b_frozen_data) {
2540		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2541		jbd2_free(jh->b_frozen_data, bh->b_size);
2542	}
2543	if (jh->b_committed_data) {
2544		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2545		jbd2_free(jh->b_committed_data, bh->b_size);
2546	}
2547	bh->b_private = NULL;
2548	jh->b_bh = NULL;	/* debug, really */
2549	clear_buffer_jbd(bh);
2550	journal_free_journal_head(jh);
2551}
2552
2553/*
2554 * Drop a reference on the passed journal_head.  If it fell to zero then
2555 * release the journal_head from the buffer_head.
2556 */
2557void jbd2_journal_put_journal_head(struct journal_head *jh)
2558{
2559	struct buffer_head *bh = jh2bh(jh);
2560
2561	jbd_lock_bh_journal_head(bh);
2562	J_ASSERT_JH(jh, jh->b_jcount > 0);
2563	--jh->b_jcount;
2564	if (!jh->b_jcount) {
2565		__journal_remove_journal_head(bh);
2566		jbd_unlock_bh_journal_head(bh);
2567		__brelse(bh);
2568	} else
2569		jbd_unlock_bh_journal_head(bh);
2570}
2571
2572/*
2573 * Initialize jbd inode head
2574 */
2575void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2576{
2577	jinode->i_transaction = NULL;
2578	jinode->i_next_transaction = NULL;
2579	jinode->i_vfs_inode = inode;
2580	jinode->i_flags = 0;
2581	INIT_LIST_HEAD(&jinode->i_list);
2582}
2583
2584/*
2585 * Function to be called before we start removing inode from memory (i.e.,
2586 * clear_inode() is a fine place to be called from). It removes inode from
2587 * transaction's lists.
2588 */
2589void jbd2_journal_release_jbd_inode(journal_t *journal,
2590				    struct jbd2_inode *jinode)
2591{
2592	if (!journal)
2593		return;
2594restart:
2595	spin_lock(&journal->j_list_lock);
2596	/* Is commit writing out inode - we have to wait */
2597	if (jinode->i_flags & JI_COMMIT_RUNNING) {
2598		wait_queue_head_t *wq;
2599		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2600		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2601		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2602		spin_unlock(&journal->j_list_lock);
2603		schedule();
2604		finish_wait(wq, &wait.wait);
2605		goto restart;
2606	}
2607
2608	if (jinode->i_transaction) {
2609		list_del(&jinode->i_list);
2610		jinode->i_transaction = NULL;
2611	}
2612	spin_unlock(&journal->j_list_lock);
2613}
2614
2615
2616#ifdef CONFIG_PROC_FS
2617
2618#define JBD2_STATS_PROC_NAME "fs/jbd2"
2619
2620static void __init jbd2_create_jbd_stats_proc_entry(void)
2621{
2622	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2623}
2624
2625static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2626{
2627	if (proc_jbd2_stats)
2628		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2629}
2630
2631#else
2632
2633#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2634#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2635
2636#endif
2637
2638struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2639
2640static int __init jbd2_journal_init_handle_cache(void)
2641{
2642	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2643	if (jbd2_handle_cache == NULL) {
2644		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2645		return -ENOMEM;
2646	}
2647	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2648	if (jbd2_inode_cache == NULL) {
2649		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2650		kmem_cache_destroy(jbd2_handle_cache);
2651		return -ENOMEM;
2652	}
2653	return 0;
2654}
2655
2656static void jbd2_journal_destroy_handle_cache(void)
2657{
2658	if (jbd2_handle_cache)
2659		kmem_cache_destroy(jbd2_handle_cache);
2660	if (jbd2_inode_cache)
2661		kmem_cache_destroy(jbd2_inode_cache);
2662
2663}
2664
2665/*
2666 * Module startup and shutdown
2667 */
2668
2669static int __init journal_init_caches(void)
2670{
2671	int ret;
2672
2673	ret = jbd2_journal_init_revoke_caches();
2674	if (ret == 0)
2675		ret = jbd2_journal_init_journal_head_cache();
2676	if (ret == 0)
2677		ret = jbd2_journal_init_handle_cache();
2678	if (ret == 0)
2679		ret = jbd2_journal_init_transaction_cache();
2680	return ret;
2681}
2682
2683static void jbd2_journal_destroy_caches(void)
2684{
2685	jbd2_journal_destroy_revoke_caches();
2686	jbd2_journal_destroy_journal_head_cache();
2687	jbd2_journal_destroy_handle_cache();
2688	jbd2_journal_destroy_transaction_cache();
2689	jbd2_journal_destroy_slabs();
2690}
2691
2692static int __init journal_init(void)
2693{
2694	int ret;
2695
2696	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2697
2698	ret = journal_init_caches();
2699	if (ret == 0) {
2700		jbd2_create_jbd_stats_proc_entry();
2701	} else {
2702		jbd2_journal_destroy_caches();
2703	}
2704	return ret;
2705}
2706
2707static void __exit journal_exit(void)
2708{
2709#ifdef CONFIG_JBD2_DEBUG
2710	int n = atomic_read(&nr_journal_heads);
2711	if (n)
2712		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2713#endif
2714	jbd2_remove_jbd_stats_proc_entry();
2715	jbd2_journal_destroy_caches();
2716}
2717
2718MODULE_LICENSE("GPL");
2719module_init(journal_init);
2720module_exit(journal_exit);
2721
v4.17
   1// SPDX-License-Identifier: GPL-2.0+
   2/*
   3 * linux/fs/jbd2/journal.c
   4 *
   5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
   6 *
   7 * Copyright 1998 Red Hat corp --- All Rights Reserved
   8 *
 
 
 
 
   9 * Generic filesystem journal-writing code; part of the ext2fs
  10 * journaling system.
  11 *
  12 * This file manages journals: areas of disk reserved for logging
  13 * transactional updates.  This includes the kernel journaling thread
  14 * which is responsible for scheduling updates to the log.
  15 *
  16 * We do not actually manage the physical storage of the journal in this
  17 * file: that is left to a per-journal policy function, which allows us
  18 * to store the journal within a filesystem-specified area for ext2
  19 * journaling (ext2 can use a reserved inode for storing the log).
  20 */
  21
  22#include <linux/module.h>
  23#include <linux/time.h>
  24#include <linux/fs.h>
  25#include <linux/jbd2.h>
  26#include <linux/errno.h>
  27#include <linux/slab.h>
  28#include <linux/init.h>
  29#include <linux/mm.h>
  30#include <linux/freezer.h>
  31#include <linux/pagemap.h>
  32#include <linux/kthread.h>
  33#include <linux/poison.h>
  34#include <linux/proc_fs.h>
  35#include <linux/seq_file.h>
  36#include <linux/math64.h>
  37#include <linux/hash.h>
  38#include <linux/log2.h>
  39#include <linux/vmalloc.h>
  40#include <linux/backing-dev.h>
  41#include <linux/bitops.h>
  42#include <linux/ratelimit.h>
  43#include <linux/sched/mm.h>
  44
  45#define CREATE_TRACE_POINTS
  46#include <trace/events/jbd2.h>
  47
  48#include <linux/uaccess.h>
  49#include <asm/page.h>
  50
  51#ifdef CONFIG_JBD2_DEBUG
  52ushort jbd2_journal_enable_debug __read_mostly;
  53EXPORT_SYMBOL(jbd2_journal_enable_debug);
  54
  55module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
  56MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
  57#endif
  58
  59EXPORT_SYMBOL(jbd2_journal_extend);
  60EXPORT_SYMBOL(jbd2_journal_stop);
  61EXPORT_SYMBOL(jbd2_journal_lock_updates);
  62EXPORT_SYMBOL(jbd2_journal_unlock_updates);
  63EXPORT_SYMBOL(jbd2_journal_get_write_access);
  64EXPORT_SYMBOL(jbd2_journal_get_create_access);
  65EXPORT_SYMBOL(jbd2_journal_get_undo_access);
  66EXPORT_SYMBOL(jbd2_journal_set_triggers);
  67EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
  68EXPORT_SYMBOL(jbd2_journal_forget);
  69#if 0
  70EXPORT_SYMBOL(journal_sync_buffer);
  71#endif
  72EXPORT_SYMBOL(jbd2_journal_flush);
  73EXPORT_SYMBOL(jbd2_journal_revoke);
  74
  75EXPORT_SYMBOL(jbd2_journal_init_dev);
  76EXPORT_SYMBOL(jbd2_journal_init_inode);
  77EXPORT_SYMBOL(jbd2_journal_check_used_features);
  78EXPORT_SYMBOL(jbd2_journal_check_available_features);
  79EXPORT_SYMBOL(jbd2_journal_set_features);
  80EXPORT_SYMBOL(jbd2_journal_load);
  81EXPORT_SYMBOL(jbd2_journal_destroy);
  82EXPORT_SYMBOL(jbd2_journal_abort);
  83EXPORT_SYMBOL(jbd2_journal_errno);
  84EXPORT_SYMBOL(jbd2_journal_ack_err);
  85EXPORT_SYMBOL(jbd2_journal_clear_err);
  86EXPORT_SYMBOL(jbd2_log_wait_commit);
  87EXPORT_SYMBOL(jbd2_log_start_commit);
  88EXPORT_SYMBOL(jbd2_journal_start_commit);
  89EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
  90EXPORT_SYMBOL(jbd2_journal_wipe);
  91EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
  92EXPORT_SYMBOL(jbd2_journal_invalidatepage);
  93EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
  94EXPORT_SYMBOL(jbd2_journal_force_commit);
  95EXPORT_SYMBOL(jbd2_journal_inode_add_write);
  96EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
  97EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
  98EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
  99EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
 100EXPORT_SYMBOL(jbd2_inode_cache);
 101
 102static void __journal_abort_soft (journal_t *journal, int errno);
 103static int jbd2_journal_create_slab(size_t slab_size);
 104
 105#ifdef CONFIG_JBD2_DEBUG
 106void __jbd2_debug(int level, const char *file, const char *func,
 107		  unsigned int line, const char *fmt, ...)
 108{
 109	struct va_format vaf;
 110	va_list args;
 111
 112	if (level > jbd2_journal_enable_debug)
 113		return;
 114	va_start(args, fmt);
 115	vaf.fmt = fmt;
 116	vaf.va = &args;
 117	printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
 118	va_end(args);
 119}
 120EXPORT_SYMBOL(__jbd2_debug);
 121#endif
 122
 123/* Checksumming functions */
 124static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
 125{
 126	if (!jbd2_journal_has_csum_v2or3_feature(j))
 127		return 1;
 128
 129	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
 130}
 131
 132static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
 133{
 134	__u32 csum;
 135	__be32 old_csum;
 136
 137	old_csum = sb->s_checksum;
 138	sb->s_checksum = 0;
 139	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
 140	sb->s_checksum = old_csum;
 141
 142	return cpu_to_be32(csum);
 143}
 144
 145static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
 146{
 147	if (!jbd2_journal_has_csum_v2or3(j))
 148		return 1;
 149
 150	return sb->s_checksum == jbd2_superblock_csum(j, sb);
 151}
 152
 153static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
 154{
 155	if (!jbd2_journal_has_csum_v2or3(j))
 156		return;
 157
 158	sb->s_checksum = jbd2_superblock_csum(j, sb);
 159}
 160
 161/*
 162 * Helper function used to manage commit timeouts
 163 */
 164
 165static void commit_timeout(struct timer_list *t)
 166{
 167	journal_t *journal = from_timer(journal, t, j_commit_timer);
 168
 169	wake_up_process(journal->j_task);
 170}
 171
 172/*
 173 * kjournald2: The main thread function used to manage a logging device
 174 * journal.
 175 *
 176 * This kernel thread is responsible for two things:
 177 *
 178 * 1) COMMIT:  Every so often we need to commit the current state of the
 179 *    filesystem to disk.  The journal thread is responsible for writing
 180 *    all of the metadata buffers to disk.
 181 *
 182 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
 183 *    of the data in that part of the log has been rewritten elsewhere on
 184 *    the disk.  Flushing these old buffers to reclaim space in the log is
 185 *    known as checkpointing, and this thread is responsible for that job.
 186 */
 187
 188static int kjournald2(void *arg)
 189{
 190	journal_t *journal = arg;
 191	transaction_t *transaction;
 192
 193	/*
 194	 * Set up an interval timer which can be used to trigger a commit wakeup
 195	 * after the commit interval expires
 196	 */
 197	timer_setup(&journal->j_commit_timer, commit_timeout, 0);
 
 198
 199	set_freezable();
 200
 201	/* Record that the journal thread is running */
 202	journal->j_task = current;
 203	wake_up(&journal->j_wait_done_commit);
 204
 205	/*
 206	 * Make sure that no allocations from this kernel thread will ever
 207	 * recurse to the fs layer because we are responsible for the
 208	 * transaction commit and any fs involvement might get stuck waiting for
 209	 * the trasn. commit.
 210	 */
 211	memalloc_nofs_save();
 212
 213	/*
 214	 * And now, wait forever for commit wakeup events.
 215	 */
 216	write_lock(&journal->j_state_lock);
 217
 218loop:
 219	if (journal->j_flags & JBD2_UNMOUNT)
 220		goto end_loop;
 221
 222	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
 223		journal->j_commit_sequence, journal->j_commit_request);
 224
 225	if (journal->j_commit_sequence != journal->j_commit_request) {
 226		jbd_debug(1, "OK, requests differ\n");
 227		write_unlock(&journal->j_state_lock);
 228		del_timer_sync(&journal->j_commit_timer);
 229		jbd2_journal_commit_transaction(journal);
 230		write_lock(&journal->j_state_lock);
 231		goto loop;
 232	}
 233
 234	wake_up(&journal->j_wait_done_commit);
 235	if (freezing(current)) {
 236		/*
 237		 * The simpler the better. Flushing journal isn't a
 238		 * good idea, because that depends on threads that may
 239		 * be already stopped.
 240		 */
 241		jbd_debug(1, "Now suspending kjournald2\n");
 242		write_unlock(&journal->j_state_lock);
 243		try_to_freeze();
 244		write_lock(&journal->j_state_lock);
 245	} else {
 246		/*
 247		 * We assume on resume that commits are already there,
 248		 * so we don't sleep
 249		 */
 250		DEFINE_WAIT(wait);
 251		int should_sleep = 1;
 252
 253		prepare_to_wait(&journal->j_wait_commit, &wait,
 254				TASK_INTERRUPTIBLE);
 255		if (journal->j_commit_sequence != journal->j_commit_request)
 256			should_sleep = 0;
 257		transaction = journal->j_running_transaction;
 258		if (transaction && time_after_eq(jiffies,
 259						transaction->t_expires))
 260			should_sleep = 0;
 261		if (journal->j_flags & JBD2_UNMOUNT)
 262			should_sleep = 0;
 263		if (should_sleep) {
 264			write_unlock(&journal->j_state_lock);
 265			schedule();
 266			write_lock(&journal->j_state_lock);
 267		}
 268		finish_wait(&journal->j_wait_commit, &wait);
 269	}
 270
 271	jbd_debug(1, "kjournald2 wakes\n");
 272
 273	/*
 274	 * Were we woken up by a commit wakeup event?
 275	 */
 276	transaction = journal->j_running_transaction;
 277	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
 278		journal->j_commit_request = transaction->t_tid;
 279		jbd_debug(1, "woke because of timeout\n");
 280	}
 281	goto loop;
 282
 283end_loop:
 
 284	del_timer_sync(&journal->j_commit_timer);
 285	journal->j_task = NULL;
 286	wake_up(&journal->j_wait_done_commit);
 287	jbd_debug(1, "Journal thread exiting.\n");
 288	write_unlock(&journal->j_state_lock);
 289	return 0;
 290}
 291
 292static int jbd2_journal_start_thread(journal_t *journal)
 293{
 294	struct task_struct *t;
 295
 296	t = kthread_run(kjournald2, journal, "jbd2/%s",
 297			journal->j_devname);
 298	if (IS_ERR(t))
 299		return PTR_ERR(t);
 300
 301	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
 302	return 0;
 303}
 304
 305static void journal_kill_thread(journal_t *journal)
 306{
 307	write_lock(&journal->j_state_lock);
 308	journal->j_flags |= JBD2_UNMOUNT;
 309
 310	while (journal->j_task) {
 311		write_unlock(&journal->j_state_lock);
 312		wake_up(&journal->j_wait_commit);
 313		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
 314		write_lock(&journal->j_state_lock);
 315	}
 316	write_unlock(&journal->j_state_lock);
 317}
 318
 319/*
 320 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
 321 *
 322 * Writes a metadata buffer to a given disk block.  The actual IO is not
 323 * performed but a new buffer_head is constructed which labels the data
 324 * to be written with the correct destination disk block.
 325 *
 326 * Any magic-number escaping which needs to be done will cause a
 327 * copy-out here.  If the buffer happens to start with the
 328 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
 329 * magic number is only written to the log for descripter blocks.  In
 330 * this case, we copy the data and replace the first word with 0, and we
 331 * return a result code which indicates that this buffer needs to be
 332 * marked as an escaped buffer in the corresponding log descriptor
 333 * block.  The missing word can then be restored when the block is read
 334 * during recovery.
 335 *
 336 * If the source buffer has already been modified by a new transaction
 337 * since we took the last commit snapshot, we use the frozen copy of
 338 * that data for IO. If we end up using the existing buffer_head's data
 339 * for the write, then we have to make sure nobody modifies it while the
 340 * IO is in progress. do_get_write_access() handles this.
 341 *
 342 * The function returns a pointer to the buffer_head to be used for IO.
 343 * 
 344 *
 345 * Return value:
 346 *  <0: Error
 347 * >=0: Finished OK
 348 *
 349 * On success:
 350 * Bit 0 set == escape performed on the data
 351 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
 352 */
 353
 354int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
 355				  struct journal_head  *jh_in,
 356				  struct buffer_head **bh_out,
 357				  sector_t blocknr)
 358{
 359	int need_copy_out = 0;
 360	int done_copy_out = 0;
 361	int do_escape = 0;
 362	char *mapped_data;
 363	struct buffer_head *new_bh;
 364	struct page *new_page;
 365	unsigned int new_offset;
 366	struct buffer_head *bh_in = jh2bh(jh_in);
 367	journal_t *journal = transaction->t_journal;
 368
 369	/*
 370	 * The buffer really shouldn't be locked: only the current committing
 371	 * transaction is allowed to write it, so nobody else is allowed
 372	 * to do any IO.
 373	 *
 374	 * akpm: except if we're journalling data, and write() output is
 375	 * also part of a shared mapping, and another thread has
 376	 * decided to launch a writepage() against this buffer.
 377	 */
 378	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 379
 380	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
 381
 382	/* keep subsequent assertions sane */
 383	atomic_set(&new_bh->b_count, 1);
 384
 385	jbd_lock_bh_state(bh_in);
 386repeat:
 387	/*
 388	 * If a new transaction has already done a buffer copy-out, then
 389	 * we use that version of the data for the commit.
 390	 */
 391	if (jh_in->b_frozen_data) {
 392		done_copy_out = 1;
 393		new_page = virt_to_page(jh_in->b_frozen_data);
 394		new_offset = offset_in_page(jh_in->b_frozen_data);
 395	} else {
 396		new_page = jh2bh(jh_in)->b_page;
 397		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
 398	}
 399
 400	mapped_data = kmap_atomic(new_page);
 401	/*
 402	 * Fire data frozen trigger if data already wasn't frozen.  Do this
 403	 * before checking for escaping, as the trigger may modify the magic
 404	 * offset.  If a copy-out happens afterwards, it will have the correct
 405	 * data in the buffer.
 406	 */
 407	if (!done_copy_out)
 408		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
 409					   jh_in->b_triggers);
 410
 411	/*
 412	 * Check for escaping
 413	 */
 414	if (*((__be32 *)(mapped_data + new_offset)) ==
 415				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 416		need_copy_out = 1;
 417		do_escape = 1;
 418	}
 419	kunmap_atomic(mapped_data);
 420
 421	/*
 422	 * Do we need to do a data copy?
 423	 */
 424	if (need_copy_out && !done_copy_out) {
 425		char *tmp;
 426
 427		jbd_unlock_bh_state(bh_in);
 428		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
 429		if (!tmp) {
 430			brelse(new_bh);
 431			return -ENOMEM;
 432		}
 433		jbd_lock_bh_state(bh_in);
 434		if (jh_in->b_frozen_data) {
 435			jbd2_free(tmp, bh_in->b_size);
 436			goto repeat;
 437		}
 438
 439		jh_in->b_frozen_data = tmp;
 440		mapped_data = kmap_atomic(new_page);
 441		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
 442		kunmap_atomic(mapped_data);
 443
 444		new_page = virt_to_page(tmp);
 445		new_offset = offset_in_page(tmp);
 446		done_copy_out = 1;
 447
 448		/*
 449		 * This isn't strictly necessary, as we're using frozen
 450		 * data for the escaping, but it keeps consistency with
 451		 * b_frozen_data usage.
 452		 */
 453		jh_in->b_frozen_triggers = jh_in->b_triggers;
 454	}
 455
 456	/*
 457	 * Did we need to do an escaping?  Now we've done all the
 458	 * copying, we can finally do so.
 459	 */
 460	if (do_escape) {
 461		mapped_data = kmap_atomic(new_page);
 462		*((unsigned int *)(mapped_data + new_offset)) = 0;
 463		kunmap_atomic(mapped_data);
 464	}
 465
 466	set_bh_page(new_bh, new_page, new_offset);
 467	new_bh->b_size = bh_in->b_size;
 468	new_bh->b_bdev = journal->j_dev;
 469	new_bh->b_blocknr = blocknr;
 470	new_bh->b_private = bh_in;
 471	set_buffer_mapped(new_bh);
 472	set_buffer_dirty(new_bh);
 473
 474	*bh_out = new_bh;
 475
 476	/*
 477	 * The to-be-written buffer needs to get moved to the io queue,
 478	 * and the original buffer whose contents we are shadowing or
 479	 * copying is moved to the transaction's shadow queue.
 480	 */
 481	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
 482	spin_lock(&journal->j_list_lock);
 483	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
 484	spin_unlock(&journal->j_list_lock);
 485	set_buffer_shadow(bh_in);
 486	jbd_unlock_bh_state(bh_in);
 487
 488	return do_escape | (done_copy_out << 1);
 489}
 490
 491/*
 492 * Allocation code for the journal file.  Manage the space left in the
 493 * journal, so that we can begin checkpointing when appropriate.
 494 */
 495
 496/*
 497 * Called with j_state_lock locked for writing.
 498 * Returns true if a transaction commit was started.
 499 */
 500int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 501{
 502	/* Return if the txn has already requested to be committed */
 503	if (journal->j_commit_request == target)
 504		return 0;
 505
 506	/*
 507	 * The only transaction we can possibly wait upon is the
 508	 * currently running transaction (if it exists).  Otherwise,
 509	 * the target tid must be an old one.
 510	 */
 511	if (journal->j_running_transaction &&
 512	    journal->j_running_transaction->t_tid == target) {
 513		/*
 514		 * We want a new commit: OK, mark the request and wakeup the
 515		 * commit thread.  We do _not_ do the commit ourselves.
 516		 */
 517
 518		journal->j_commit_request = target;
 519		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
 520			  journal->j_commit_request,
 521			  journal->j_commit_sequence);
 522		journal->j_running_transaction->t_requested = jiffies;
 523		wake_up(&journal->j_wait_commit);
 524		return 1;
 525	} else if (!tid_geq(journal->j_commit_request, target))
 526		/* This should never happen, but if it does, preserve
 527		   the evidence before kjournald goes into a loop and
 528		   increments j_commit_sequence beyond all recognition. */
 529		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
 530			  journal->j_commit_request,
 531			  journal->j_commit_sequence,
 532			  target, journal->j_running_transaction ? 
 533			  journal->j_running_transaction->t_tid : 0);
 534	return 0;
 535}
 536
 537int jbd2_log_start_commit(journal_t *journal, tid_t tid)
 538{
 539	int ret;
 540
 541	write_lock(&journal->j_state_lock);
 542	ret = __jbd2_log_start_commit(journal, tid);
 543	write_unlock(&journal->j_state_lock);
 544	return ret;
 545}
 546
 547/*
 548 * Force and wait any uncommitted transactions.  We can only force the running
 549 * transaction if we don't have an active handle, otherwise, we will deadlock.
 550 * Returns: <0 in case of error,
 551 *           0 if nothing to commit,
 552 *           1 if transaction was successfully committed.
 553 */
 554static int __jbd2_journal_force_commit(journal_t *journal)
 555{
 556	transaction_t *transaction = NULL;
 557	tid_t tid;
 558	int need_to_start = 0, ret = 0;
 559
 560	read_lock(&journal->j_state_lock);
 561	if (journal->j_running_transaction && !current->journal_info) {
 562		transaction = journal->j_running_transaction;
 563		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
 564			need_to_start = 1;
 565	} else if (journal->j_committing_transaction)
 566		transaction = journal->j_committing_transaction;
 567
 568	if (!transaction) {
 569		/* Nothing to commit */
 570		read_unlock(&journal->j_state_lock);
 571		return 0;
 572	}
 573	tid = transaction->t_tid;
 574	read_unlock(&journal->j_state_lock);
 575	if (need_to_start)
 576		jbd2_log_start_commit(journal, tid);
 577	ret = jbd2_log_wait_commit(journal, tid);
 578	if (!ret)
 579		ret = 1;
 580
 581	return ret;
 582}
 583
 584/**
 585 * Force and wait upon a commit if the calling process is not within
 586 * transaction.  This is used for forcing out undo-protected data which contains
 587 * bitmaps, when the fs is running out of space.
 588 *
 589 * @journal: journal to force
 590 * Returns true if progress was made.
 591 */
 592int jbd2_journal_force_commit_nested(journal_t *journal)
 593{
 594	int ret;
 595
 596	ret = __jbd2_journal_force_commit(journal);
 597	return ret > 0;
 598}
 599
 600/**
 601 * int journal_force_commit() - force any uncommitted transactions
 602 * @journal: journal to force
 603 *
 604 * Caller want unconditional commit. We can only force the running transaction
 605 * if we don't have an active handle, otherwise, we will deadlock.
 606 */
 607int jbd2_journal_force_commit(journal_t *journal)
 608{
 609	int ret;
 610
 611	J_ASSERT(!current->journal_info);
 612	ret = __jbd2_journal_force_commit(journal);
 613	if (ret > 0)
 614		ret = 0;
 615	return ret;
 616}
 617
 618/*
 619 * Start a commit of the current running transaction (if any).  Returns true
 620 * if a transaction is going to be committed (or is currently already
 621 * committing), and fills its tid in at *ptid
 622 */
 623int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
 624{
 625	int ret = 0;
 626
 627	write_lock(&journal->j_state_lock);
 628	if (journal->j_running_transaction) {
 629		tid_t tid = journal->j_running_transaction->t_tid;
 630
 631		__jbd2_log_start_commit(journal, tid);
 632		/* There's a running transaction and we've just made sure
 633		 * it's commit has been scheduled. */
 634		if (ptid)
 635			*ptid = tid;
 636		ret = 1;
 637	} else if (journal->j_committing_transaction) {
 638		/*
 639		 * If commit has been started, then we have to wait for
 640		 * completion of that transaction.
 641		 */
 642		if (ptid)
 643			*ptid = journal->j_committing_transaction->t_tid;
 644		ret = 1;
 645	}
 646	write_unlock(&journal->j_state_lock);
 647	return ret;
 648}
 649
 650/*
 651 * Return 1 if a given transaction has not yet sent barrier request
 652 * connected with a transaction commit. If 0 is returned, transaction
 653 * may or may not have sent the barrier. Used to avoid sending barrier
 654 * twice in common cases.
 655 */
 656int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
 657{
 658	int ret = 0;
 659	transaction_t *commit_trans;
 660
 661	if (!(journal->j_flags & JBD2_BARRIER))
 662		return 0;
 663	read_lock(&journal->j_state_lock);
 664	/* Transaction already committed? */
 665	if (tid_geq(journal->j_commit_sequence, tid))
 666		goto out;
 667	commit_trans = journal->j_committing_transaction;
 668	if (!commit_trans || commit_trans->t_tid != tid) {
 669		ret = 1;
 670		goto out;
 671	}
 672	/*
 673	 * Transaction is being committed and we already proceeded to
 674	 * submitting a flush to fs partition?
 675	 */
 676	if (journal->j_fs_dev != journal->j_dev) {
 677		if (!commit_trans->t_need_data_flush ||
 678		    commit_trans->t_state >= T_COMMIT_DFLUSH)
 679			goto out;
 680	} else {
 681		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
 682			goto out;
 683	}
 684	ret = 1;
 685out:
 686	read_unlock(&journal->j_state_lock);
 687	return ret;
 688}
 689EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
 690
 691/*
 692 * Wait for a specified commit to complete.
 693 * The caller may not hold the journal lock.
 694 */
 695int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
 696{
 697	int err = 0;
 698
 699	read_lock(&journal->j_state_lock);
 700#ifdef CONFIG_PROVE_LOCKING
 701	/*
 702	 * Some callers make sure transaction is already committing and in that
 703	 * case we cannot block on open handles anymore. So don't warn in that
 704	 * case.
 705	 */
 706	if (tid_gt(tid, journal->j_commit_sequence) &&
 707	    (!journal->j_committing_transaction ||
 708	     journal->j_committing_transaction->t_tid != tid)) {
 709		read_unlock(&journal->j_state_lock);
 710		jbd2_might_wait_for_commit(journal);
 711		read_lock(&journal->j_state_lock);
 712	}
 713#endif
 714#ifdef CONFIG_JBD2_DEBUG
 715	if (!tid_geq(journal->j_commit_request, tid)) {
 716		printk(KERN_ERR
 717		       "%s: error: j_commit_request=%d, tid=%d\n",
 718		       __func__, journal->j_commit_request, tid);
 719	}
 720#endif
 721	while (tid_gt(tid, journal->j_commit_sequence)) {
 722		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
 723				  tid, journal->j_commit_sequence);
 724		read_unlock(&journal->j_state_lock);
 725		wake_up(&journal->j_wait_commit);
 726		wait_event(journal->j_wait_done_commit,
 727				!tid_gt(tid, journal->j_commit_sequence));
 728		read_lock(&journal->j_state_lock);
 729	}
 730	read_unlock(&journal->j_state_lock);
 731
 732	if (unlikely(is_journal_aborted(journal)))
 733		err = -EIO;
 734	return err;
 735}
 736
 737/* Return 1 when transaction with given tid has already committed. */
 738int jbd2_transaction_committed(journal_t *journal, tid_t tid)
 739{
 740	int ret = 1;
 741
 742	read_lock(&journal->j_state_lock);
 743	if (journal->j_running_transaction &&
 744	    journal->j_running_transaction->t_tid == tid)
 745		ret = 0;
 746	if (journal->j_committing_transaction &&
 747	    journal->j_committing_transaction->t_tid == tid)
 748		ret = 0;
 749	read_unlock(&journal->j_state_lock);
 750	return ret;
 751}
 752EXPORT_SYMBOL(jbd2_transaction_committed);
 753
 754/*
 755 * When this function returns the transaction corresponding to tid
 756 * will be completed.  If the transaction has currently running, start
 757 * committing that transaction before waiting for it to complete.  If
 758 * the transaction id is stale, it is by definition already completed,
 759 * so just return SUCCESS.
 760 */
 761int jbd2_complete_transaction(journal_t *journal, tid_t tid)
 762{
 763	int	need_to_wait = 1;
 764
 765	read_lock(&journal->j_state_lock);
 766	if (journal->j_running_transaction &&
 767	    journal->j_running_transaction->t_tid == tid) {
 768		if (journal->j_commit_request != tid) {
 769			/* transaction not yet started, so request it */
 770			read_unlock(&journal->j_state_lock);
 771			jbd2_log_start_commit(journal, tid);
 772			goto wait_commit;
 773		}
 774	} else if (!(journal->j_committing_transaction &&
 775		     journal->j_committing_transaction->t_tid == tid))
 776		need_to_wait = 0;
 777	read_unlock(&journal->j_state_lock);
 778	if (!need_to_wait)
 779		return 0;
 780wait_commit:
 781	return jbd2_log_wait_commit(journal, tid);
 782}
 783EXPORT_SYMBOL(jbd2_complete_transaction);
 784
 785/*
 786 * Log buffer allocation routines:
 787 */
 788
 789int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
 790{
 791	unsigned long blocknr;
 792
 793	write_lock(&journal->j_state_lock);
 794	J_ASSERT(journal->j_free > 1);
 795
 796	blocknr = journal->j_head;
 797	journal->j_head++;
 798	journal->j_free--;
 799	if (journal->j_head == journal->j_last)
 800		journal->j_head = journal->j_first;
 801	write_unlock(&journal->j_state_lock);
 802	return jbd2_journal_bmap(journal, blocknr, retp);
 803}
 804
 805/*
 806 * Conversion of logical to physical block numbers for the journal
 807 *
 808 * On external journals the journal blocks are identity-mapped, so
 809 * this is a no-op.  If needed, we can use j_blk_offset - everything is
 810 * ready.
 811 */
 812int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 813		 unsigned long long *retp)
 814{
 815	int err = 0;
 816	unsigned long long ret;
 817
 818	if (journal->j_inode) {
 819		ret = bmap(journal->j_inode, blocknr);
 820		if (ret)
 821			*retp = ret;
 822		else {
 823			printk(KERN_ALERT "%s: journal block not found "
 824					"at offset %lu on %s\n",
 825			       __func__, blocknr, journal->j_devname);
 826			err = -EIO;
 827			__journal_abort_soft(journal, err);
 828		}
 829	} else {
 830		*retp = blocknr; /* +journal->j_blk_offset */
 831	}
 832	return err;
 833}
 834
 835/*
 836 * We play buffer_head aliasing tricks to write data/metadata blocks to
 837 * the journal without copying their contents, but for journal
 838 * descriptor blocks we do need to generate bona fide buffers.
 839 *
 840 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
 841 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
 842 * But we don't bother doing that, so there will be coherency problems with
 843 * mmaps of blockdevs which hold live JBD-controlled filesystems.
 844 */
 845struct buffer_head *
 846jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
 847{
 848	journal_t *journal = transaction->t_journal;
 849	struct buffer_head *bh;
 850	unsigned long long blocknr;
 851	journal_header_t *header;
 852	int err;
 853
 854	err = jbd2_journal_next_log_block(journal, &blocknr);
 855
 856	if (err)
 857		return NULL;
 858
 859	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 860	if (!bh)
 861		return NULL;
 862	lock_buffer(bh);
 863	memset(bh->b_data, 0, journal->j_blocksize);
 864	header = (journal_header_t *)bh->b_data;
 865	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
 866	header->h_blocktype = cpu_to_be32(type);
 867	header->h_sequence = cpu_to_be32(transaction->t_tid);
 868	set_buffer_uptodate(bh);
 869	unlock_buffer(bh);
 870	BUFFER_TRACE(bh, "return this buffer");
 871	return bh;
 872}
 873
 874void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
 875{
 876	struct jbd2_journal_block_tail *tail;
 877	__u32 csum;
 878
 879	if (!jbd2_journal_has_csum_v2or3(j))
 880		return;
 881
 882	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
 883			sizeof(struct jbd2_journal_block_tail));
 884	tail->t_checksum = 0;
 885	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 886	tail->t_checksum = cpu_to_be32(csum);
 887}
 888
 889/*
 890 * Return tid of the oldest transaction in the journal and block in the journal
 891 * where the transaction starts.
 892 *
 893 * If the journal is now empty, return which will be the next transaction ID
 894 * we will write and where will that transaction start.
 895 *
 896 * The return value is 0 if journal tail cannot be pushed any further, 1 if
 897 * it can.
 898 */
 899int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
 900			      unsigned long *block)
 901{
 902	transaction_t *transaction;
 903	int ret;
 904
 905	read_lock(&journal->j_state_lock);
 906	spin_lock(&journal->j_list_lock);
 907	transaction = journal->j_checkpoint_transactions;
 908	if (transaction) {
 909		*tid = transaction->t_tid;
 910		*block = transaction->t_log_start;
 911	} else if ((transaction = journal->j_committing_transaction) != NULL) {
 912		*tid = transaction->t_tid;
 913		*block = transaction->t_log_start;
 914	} else if ((transaction = journal->j_running_transaction) != NULL) {
 915		*tid = transaction->t_tid;
 916		*block = journal->j_head;
 917	} else {
 918		*tid = journal->j_transaction_sequence;
 919		*block = journal->j_head;
 920	}
 921	ret = tid_gt(*tid, journal->j_tail_sequence);
 922	spin_unlock(&journal->j_list_lock);
 923	read_unlock(&journal->j_state_lock);
 924
 925	return ret;
 926}
 927
 928/*
 929 * Update information in journal structure and in on disk journal superblock
 930 * about log tail. This function does not check whether information passed in
 931 * really pushes log tail further. It's responsibility of the caller to make
 932 * sure provided log tail information is valid (e.g. by holding
 933 * j_checkpoint_mutex all the time between computing log tail and calling this
 934 * function as is the case with jbd2_cleanup_journal_tail()).
 935 *
 936 * Requires j_checkpoint_mutex
 937 */
 938int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 939{
 940	unsigned long freed;
 941	int ret;
 942
 943	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 944
 945	/*
 946	 * We cannot afford for write to remain in drive's caches since as
 947	 * soon as we update j_tail, next transaction can start reusing journal
 948	 * space and if we lose sb update during power failure we'd replay
 949	 * old transaction with possibly newly overwritten data.
 950	 */
 951	ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
 952					      REQ_SYNC | REQ_FUA);
 953	if (ret)
 954		goto out;
 955
 956	write_lock(&journal->j_state_lock);
 957	freed = block - journal->j_tail;
 958	if (block < journal->j_tail)
 959		freed += journal->j_last - journal->j_first;
 960
 961	trace_jbd2_update_log_tail(journal, tid, block, freed);
 962	jbd_debug(1,
 963		  "Cleaning journal tail from %d to %d (offset %lu), "
 964		  "freeing %lu\n",
 965		  journal->j_tail_sequence, tid, block, freed);
 966
 967	journal->j_free += freed;
 968	journal->j_tail_sequence = tid;
 969	journal->j_tail = block;
 970	write_unlock(&journal->j_state_lock);
 971
 972out:
 973	return ret;
 974}
 975
 976/*
 977 * This is a variation of __jbd2_update_log_tail which checks for validity of
 978 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
 979 * with other threads updating log tail.
 980 */
 981void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 982{
 983	mutex_lock_io(&journal->j_checkpoint_mutex);
 984	if (tid_gt(tid, journal->j_tail_sequence))
 985		__jbd2_update_log_tail(journal, tid, block);
 986	mutex_unlock(&journal->j_checkpoint_mutex);
 987}
 988
 989struct jbd2_stats_proc_session {
 990	journal_t *journal;
 991	struct transaction_stats_s *stats;
 992	int start;
 993	int max;
 994};
 995
 996static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
 997{
 998	return *pos ? NULL : SEQ_START_TOKEN;
 999}
1000
1001static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1002{
1003	return NULL;
1004}
1005
1006static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1007{
1008	struct jbd2_stats_proc_session *s = seq->private;
1009
1010	if (v != SEQ_START_TOKEN)
1011		return 0;
1012	seq_printf(seq, "%lu transactions (%lu requested), "
1013		   "each up to %u blocks\n",
1014		   s->stats->ts_tid, s->stats->ts_requested,
1015		   s->journal->j_max_transaction_buffers);
1016	if (s->stats->ts_tid == 0)
1017		return 0;
1018	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1019	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1020	seq_printf(seq, "  %ums request delay\n",
1021	    (s->stats->ts_requested == 0) ? 0 :
1022	    jiffies_to_msecs(s->stats->run.rs_request_delay /
1023			     s->stats->ts_requested));
1024	seq_printf(seq, "  %ums running transaction\n",
1025	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1026	seq_printf(seq, "  %ums transaction was being locked\n",
1027	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1028	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1029	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1030	seq_printf(seq, "  %ums logging transaction\n",
1031	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1032	seq_printf(seq, "  %lluus average transaction commit time\n",
1033		   div_u64(s->journal->j_average_commit_time, 1000));
1034	seq_printf(seq, "  %lu handles per transaction\n",
1035	    s->stats->run.rs_handle_count / s->stats->ts_tid);
1036	seq_printf(seq, "  %lu blocks per transaction\n",
1037	    s->stats->run.rs_blocks / s->stats->ts_tid);
1038	seq_printf(seq, "  %lu logged blocks per transaction\n",
1039	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1040	return 0;
1041}
1042
1043static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1044{
1045}
1046
1047static const struct seq_operations jbd2_seq_info_ops = {
1048	.start  = jbd2_seq_info_start,
1049	.next   = jbd2_seq_info_next,
1050	.stop   = jbd2_seq_info_stop,
1051	.show   = jbd2_seq_info_show,
1052};
1053
1054static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1055{
1056	journal_t *journal = PDE_DATA(inode);
1057	struct jbd2_stats_proc_session *s;
1058	int rc, size;
1059
1060	s = kmalloc(sizeof(*s), GFP_KERNEL);
1061	if (s == NULL)
1062		return -ENOMEM;
1063	size = sizeof(struct transaction_stats_s);
1064	s->stats = kmalloc(size, GFP_KERNEL);
1065	if (s->stats == NULL) {
1066		kfree(s);
1067		return -ENOMEM;
1068	}
1069	spin_lock(&journal->j_history_lock);
1070	memcpy(s->stats, &journal->j_stats, size);
1071	s->journal = journal;
1072	spin_unlock(&journal->j_history_lock);
1073
1074	rc = seq_open(file, &jbd2_seq_info_ops);
1075	if (rc == 0) {
1076		struct seq_file *m = file->private_data;
1077		m->private = s;
1078	} else {
1079		kfree(s->stats);
1080		kfree(s);
1081	}
1082	return rc;
1083
1084}
1085
1086static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1087{
1088	struct seq_file *seq = file->private_data;
1089	struct jbd2_stats_proc_session *s = seq->private;
1090	kfree(s->stats);
1091	kfree(s);
1092	return seq_release(inode, file);
1093}
1094
1095static const struct file_operations jbd2_seq_info_fops = {
1096	.owner		= THIS_MODULE,
1097	.open           = jbd2_seq_info_open,
1098	.read           = seq_read,
1099	.llseek         = seq_lseek,
1100	.release        = jbd2_seq_info_release,
1101};
1102
1103static struct proc_dir_entry *proc_jbd2_stats;
1104
1105static void jbd2_stats_proc_init(journal_t *journal)
1106{
1107	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1108	if (journal->j_proc_entry) {
1109		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1110				 &jbd2_seq_info_fops, journal);
1111	}
1112}
1113
1114static void jbd2_stats_proc_exit(journal_t *journal)
1115{
1116	remove_proc_entry("info", journal->j_proc_entry);
1117	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1118}
1119
1120/*
1121 * Management for journal control blocks: functions to create and
1122 * destroy journal_t structures, and to initialise and read existing
1123 * journal blocks from disk.  */
1124
1125/* First: create and setup a journal_t object in memory.  We initialise
1126 * very few fields yet: that has to wait until we have created the
1127 * journal structures from from scratch, or loaded them from disk. */
1128
1129static journal_t *journal_init_common(struct block_device *bdev,
1130			struct block_device *fs_dev,
1131			unsigned long long start, int len, int blocksize)
1132{
1133	static struct lock_class_key jbd2_trans_commit_key;
1134	journal_t *journal;
1135	int err;
1136	struct buffer_head *bh;
1137	int n;
1138
1139	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1140	if (!journal)
1141		return NULL;
1142
1143	init_waitqueue_head(&journal->j_wait_transaction_locked);
1144	init_waitqueue_head(&journal->j_wait_done_commit);
1145	init_waitqueue_head(&journal->j_wait_commit);
1146	init_waitqueue_head(&journal->j_wait_updates);
1147	init_waitqueue_head(&journal->j_wait_reserved);
1148	mutex_init(&journal->j_barrier);
1149	mutex_init(&journal->j_checkpoint_mutex);
1150	spin_lock_init(&journal->j_revoke_lock);
1151	spin_lock_init(&journal->j_list_lock);
1152	rwlock_init(&journal->j_state_lock);
1153
1154	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1155	journal->j_min_batch_time = 0;
1156	journal->j_max_batch_time = 15000; /* 15ms */
1157	atomic_set(&journal->j_reserved_credits, 0);
1158
1159	/* The journal is marked for error until we succeed with recovery! */
1160	journal->j_flags = JBD2_ABORT;
1161
1162	/* Set up a default-sized revoke table for the new mount. */
1163	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1164	if (err)
1165		goto err_cleanup;
 
 
1166
1167	spin_lock_init(&journal->j_history_lock);
1168
1169	lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1170			 &jbd2_trans_commit_key, 0);
1171
1172	/* journal descriptor can store up to n blocks -bzzz */
1173	journal->j_blocksize = blocksize;
1174	journal->j_dev = bdev;
1175	journal->j_fs_dev = fs_dev;
1176	journal->j_blk_offset = start;
1177	journal->j_maxlen = len;
1178	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1179	journal->j_wbufsize = n;
1180	journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1181					GFP_KERNEL);
1182	if (!journal->j_wbuf)
1183		goto err_cleanup;
1184
1185	bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1186	if (!bh) {
1187		pr_err("%s: Cannot get buffer for journal superblock\n",
1188			__func__);
1189		goto err_cleanup;
1190	}
1191	journal->j_sb_buffer = bh;
1192	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1193
1194	return journal;
1195
1196err_cleanup:
1197	kfree(journal->j_wbuf);
1198	jbd2_journal_destroy_revoke(journal);
1199	kfree(journal);
1200	return NULL;
1201}
1202
1203/* jbd2_journal_init_dev and jbd2_journal_init_inode:
1204 *
1205 * Create a journal structure assigned some fixed set of disk blocks to
1206 * the journal.  We don't actually touch those disk blocks yet, but we
1207 * need to set up all of the mapping information to tell the journaling
1208 * system where the journal blocks are.
1209 *
1210 */
1211
1212/**
1213 *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1214 *  @bdev: Block device on which to create the journal
1215 *  @fs_dev: Device which hold journalled filesystem for this journal.
1216 *  @start: Block nr Start of journal.
1217 *  @len:  Length of the journal in blocks.
1218 *  @blocksize: blocksize of journalling device
1219 *
1220 *  Returns: a newly created journal_t *
1221 *
1222 *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1223 *  range of blocks on an arbitrary block device.
1224 *
1225 */
1226journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1227			struct block_device *fs_dev,
1228			unsigned long long start, int len, int blocksize)
1229{
1230	journal_t *journal;
 
 
1231
1232	journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1233	if (!journal)
1234		return NULL;
1235
 
 
 
 
 
 
1236	bdevname(journal->j_dev, journal->j_devname);
1237	strreplace(journal->j_devname, '/', '!');
1238	jbd2_stats_proc_init(journal);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1239
1240	return journal;
 
 
 
 
 
1241}
1242
1243/**
1244 *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1245 *  @inode: An inode to create the journal in
1246 *
1247 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1248 * the journal.  The inode must exist already, must support bmap() and
1249 * must have all data blocks preallocated.
1250 */
1251journal_t *jbd2_journal_init_inode(struct inode *inode)
1252{
1253	journal_t *journal;
 
1254	char *p;
 
 
1255	unsigned long long blocknr;
1256
1257	blocknr = bmap(inode, 0);
1258	if (!blocknr) {
1259		pr_err("%s: Cannot locate journal superblock\n",
1260			__func__);
1261		return NULL;
1262	}
1263
1264	jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1265		  inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1266		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1267
1268	journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1269			blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1270			inode->i_sb->s_blocksize);
1271	if (!journal)
1272		return NULL;
1273
 
1274	journal->j_inode = inode;
1275	bdevname(journal->j_dev, journal->j_devname);
1276	p = strreplace(journal->j_devname, '/', '!');
1277	sprintf(p, "-%lu", journal->j_inode->i_ino);
 
 
 
 
 
 
 
 
1278	jbd2_stats_proc_init(journal);
1279
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1280	return journal;
 
 
 
 
 
1281}
1282
1283/*
1284 * If the journal init or create aborts, we need to mark the journal
1285 * superblock as being NULL to prevent the journal destroy from writing
1286 * back a bogus superblock.
1287 */
1288static void journal_fail_superblock (journal_t *journal)
1289{
1290	struct buffer_head *bh = journal->j_sb_buffer;
1291	brelse(bh);
1292	journal->j_sb_buffer = NULL;
1293}
1294
1295/*
1296 * Given a journal_t structure, initialise the various fields for
1297 * startup of a new journaling session.  We use this both when creating
1298 * a journal, and after recovering an old journal to reset it for
1299 * subsequent use.
1300 */
1301
1302static int journal_reset(journal_t *journal)
1303{
1304	journal_superblock_t *sb = journal->j_superblock;
1305	unsigned long long first, last;
1306
1307	first = be32_to_cpu(sb->s_first);
1308	last = be32_to_cpu(sb->s_maxlen);
1309	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1310		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1311		       first, last);
1312		journal_fail_superblock(journal);
1313		return -EINVAL;
1314	}
1315
1316	journal->j_first = first;
1317	journal->j_last = last;
1318
1319	journal->j_head = first;
1320	journal->j_tail = first;
1321	journal->j_free = last - first;
1322
1323	journal->j_tail_sequence = journal->j_transaction_sequence;
1324	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1325	journal->j_commit_request = journal->j_commit_sequence;
1326
1327	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1328
1329	/*
1330	 * As a special case, if the on-disk copy is already marked as needing
1331	 * no recovery (s_start == 0), then we can safely defer the superblock
1332	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1333	 * attempting a write to a potential-readonly device.
1334	 */
1335	if (sb->s_start == 0) {
1336		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1337			"(start %ld, seq %d, errno %d)\n",
1338			journal->j_tail, journal->j_tail_sequence,
1339			journal->j_errno);
1340		journal->j_flags |= JBD2_FLUSHED;
1341	} else {
1342		/* Lock here to make assertions happy... */
1343		mutex_lock_io(&journal->j_checkpoint_mutex);
1344		/*
1345		 * Update log tail information. We use REQ_FUA since new
1346		 * transaction will start reusing journal space and so we
1347		 * must make sure information about current log tail is on
1348		 * disk before that.
1349		 */
1350		jbd2_journal_update_sb_log_tail(journal,
1351						journal->j_tail_sequence,
1352						journal->j_tail,
1353						REQ_SYNC | REQ_FUA);
1354		mutex_unlock(&journal->j_checkpoint_mutex);
1355	}
1356	return jbd2_journal_start_thread(journal);
1357}
1358
1359static int jbd2_write_superblock(journal_t *journal, int write_flags)
1360{
1361	struct buffer_head *bh = journal->j_sb_buffer;
1362	journal_superblock_t *sb = journal->j_superblock;
1363	int ret;
1364
1365	trace_jbd2_write_superblock(journal, write_flags);
1366	if (!(journal->j_flags & JBD2_BARRIER))
1367		write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1368	lock_buffer(bh);
1369	if (buffer_write_io_error(bh)) {
1370		/*
1371		 * Oh, dear.  A previous attempt to write the journal
1372		 * superblock failed.  This could happen because the
1373		 * USB device was yanked out.  Or it could happen to
1374		 * be a transient write error and maybe the block will
1375		 * be remapped.  Nothing we can do but to retry the
1376		 * write and hope for the best.
1377		 */
1378		printk(KERN_ERR "JBD2: previous I/O error detected "
1379		       "for journal superblock update for %s.\n",
1380		       journal->j_devname);
1381		clear_buffer_write_io_error(bh);
1382		set_buffer_uptodate(bh);
1383	}
1384	jbd2_superblock_csum_set(journal, sb);
1385	get_bh(bh);
1386	bh->b_end_io = end_buffer_write_sync;
1387	ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1388	wait_on_buffer(bh);
1389	if (buffer_write_io_error(bh)) {
1390		clear_buffer_write_io_error(bh);
1391		set_buffer_uptodate(bh);
1392		ret = -EIO;
1393	}
1394	if (ret) {
1395		printk(KERN_ERR "JBD2: Error %d detected when updating "
1396		       "journal superblock for %s.\n", ret,
1397		       journal->j_devname);
1398		jbd2_journal_abort(journal, ret);
1399	}
1400
1401	return ret;
1402}
1403
1404/**
1405 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1406 * @journal: The journal to update.
1407 * @tail_tid: TID of the new transaction at the tail of the log
1408 * @tail_block: The first block of the transaction at the tail of the log
1409 * @write_op: With which operation should we write the journal sb
1410 *
1411 * Update a journal's superblock information about log tail and write it to
1412 * disk, waiting for the IO to complete.
1413 */
1414int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1415				     unsigned long tail_block, int write_op)
1416{
1417	journal_superblock_t *sb = journal->j_superblock;
1418	int ret;
1419
1420	if (is_journal_aborted(journal))
1421		return -EIO;
1422
1423	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1424	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1425		  tail_block, tail_tid);
1426
1427	sb->s_sequence = cpu_to_be32(tail_tid);
1428	sb->s_start    = cpu_to_be32(tail_block);
1429
1430	ret = jbd2_write_superblock(journal, write_op);
1431	if (ret)
1432		goto out;
1433
1434	/* Log is no longer empty */
1435	write_lock(&journal->j_state_lock);
1436	WARN_ON(!sb->s_sequence);
1437	journal->j_flags &= ~JBD2_FLUSHED;
1438	write_unlock(&journal->j_state_lock);
1439
1440out:
1441	return ret;
1442}
1443
1444/**
1445 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1446 * @journal: The journal to update.
1447 * @write_op: With which operation should we write the journal sb
1448 *
1449 * Update a journal's dynamic superblock fields to show that journal is empty.
1450 * Write updated superblock to disk waiting for IO to complete.
1451 */
1452static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1453{
1454	journal_superblock_t *sb = journal->j_superblock;
1455
1456	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1457	read_lock(&journal->j_state_lock);
1458	/* Is it already empty? */
1459	if (sb->s_start == 0) {
1460		read_unlock(&journal->j_state_lock);
1461		return;
1462	}
1463	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1464		  journal->j_tail_sequence);
1465
1466	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1467	sb->s_start    = cpu_to_be32(0);
1468	read_unlock(&journal->j_state_lock);
1469
1470	jbd2_write_superblock(journal, write_op);
1471
1472	/* Log is no longer empty */
1473	write_lock(&journal->j_state_lock);
1474	journal->j_flags |= JBD2_FLUSHED;
1475	write_unlock(&journal->j_state_lock);
1476}
1477
1478
1479/**
1480 * jbd2_journal_update_sb_errno() - Update error in the journal.
1481 * @journal: The journal to update.
1482 *
1483 * Update a journal's errno.  Write updated superblock to disk waiting for IO
1484 * to complete.
1485 */
1486void jbd2_journal_update_sb_errno(journal_t *journal)
1487{
1488	journal_superblock_t *sb = journal->j_superblock;
1489	int errcode;
1490
1491	read_lock(&journal->j_state_lock);
1492	errcode = journal->j_errno;
 
 
1493	read_unlock(&journal->j_state_lock);
1494	if (errcode == -ESHUTDOWN)
1495		errcode = 0;
1496	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1497	sb->s_errno    = cpu_to_be32(errcode);
1498
1499	jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1500}
1501EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1502
1503/*
1504 * Read the superblock for a given journal, performing initial
1505 * validation of the format.
1506 */
1507static int journal_get_superblock(journal_t *journal)
1508{
1509	struct buffer_head *bh;
1510	journal_superblock_t *sb;
1511	int err = -EIO;
1512
1513	bh = journal->j_sb_buffer;
1514
1515	J_ASSERT(bh != NULL);
1516	if (!buffer_uptodate(bh)) {
1517		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1518		wait_on_buffer(bh);
1519		if (!buffer_uptodate(bh)) {
1520			printk(KERN_ERR
1521				"JBD2: IO error reading journal superblock\n");
1522			goto out;
1523		}
1524	}
1525
1526	if (buffer_verified(bh))
1527		return 0;
1528
1529	sb = journal->j_superblock;
1530
1531	err = -EINVAL;
1532
1533	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1534	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1535		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1536		goto out;
1537	}
1538
1539	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1540	case JBD2_SUPERBLOCK_V1:
1541		journal->j_format_version = 1;
1542		break;
1543	case JBD2_SUPERBLOCK_V2:
1544		journal->j_format_version = 2;
1545		break;
1546	default:
1547		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1548		goto out;
1549	}
1550
1551	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1552		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1553	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1554		printk(KERN_WARNING "JBD2: journal file too short\n");
1555		goto out;
1556	}
1557
1558	if (be32_to_cpu(sb->s_first) == 0 ||
1559	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1560		printk(KERN_WARNING
1561			"JBD2: Invalid start block of journal: %u\n",
1562			be32_to_cpu(sb->s_first));
1563		goto out;
1564	}
1565
1566	if (jbd2_has_feature_csum2(journal) &&
1567	    jbd2_has_feature_csum3(journal)) {
1568		/* Can't have checksum v2 and v3 at the same time! */
1569		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1570		       "at the same time!\n");
1571		goto out;
1572	}
1573
1574	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1575	    jbd2_has_feature_checksum(journal)) {
1576		/* Can't have checksum v1 and v2 on at the same time! */
1577		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1578		       "at the same time!\n");
1579		goto out;
1580	}
1581
1582	if (!jbd2_verify_csum_type(journal, sb)) {
1583		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1584		goto out;
1585	}
1586
1587	/* Load the checksum driver */
1588	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1589		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1590		if (IS_ERR(journal->j_chksum_driver)) {
1591			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1592			err = PTR_ERR(journal->j_chksum_driver);
1593			journal->j_chksum_driver = NULL;
1594			goto out;
1595		}
1596	}
1597
1598	/* Check superblock checksum */
1599	if (!jbd2_superblock_csum_verify(journal, sb)) {
1600		printk(KERN_ERR "JBD2: journal checksum error\n");
1601		err = -EFSBADCRC;
1602		goto out;
1603	}
1604
1605	/* Precompute checksum seed for all metadata */
1606	if (jbd2_journal_has_csum_v2or3(journal))
1607		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1608						   sizeof(sb->s_uuid));
1609
1610	set_buffer_verified(bh);
1611
1612	return 0;
1613
1614out:
1615	journal_fail_superblock(journal);
1616	return err;
1617}
1618
1619/*
1620 * Load the on-disk journal superblock and read the key fields into the
1621 * journal_t.
1622 */
1623
1624static int load_superblock(journal_t *journal)
1625{
1626	int err;
1627	journal_superblock_t *sb;
1628
1629	err = journal_get_superblock(journal);
1630	if (err)
1631		return err;
1632
1633	sb = journal->j_superblock;
1634
1635	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1636	journal->j_tail = be32_to_cpu(sb->s_start);
1637	journal->j_first = be32_to_cpu(sb->s_first);
1638	journal->j_last = be32_to_cpu(sb->s_maxlen);
1639	journal->j_errno = be32_to_cpu(sb->s_errno);
1640
1641	return 0;
1642}
1643
1644
1645/**
1646 * int jbd2_journal_load() - Read journal from disk.
1647 * @journal: Journal to act on.
1648 *
1649 * Given a journal_t structure which tells us which disk blocks contain
1650 * a journal, read the journal from disk to initialise the in-memory
1651 * structures.
1652 */
1653int jbd2_journal_load(journal_t *journal)
1654{
1655	int err;
1656	journal_superblock_t *sb;
1657
1658	err = load_superblock(journal);
1659	if (err)
1660		return err;
1661
1662	sb = journal->j_superblock;
1663	/* If this is a V2 superblock, then we have to check the
1664	 * features flags on it. */
1665
1666	if (journal->j_format_version >= 2) {
1667		if ((sb->s_feature_ro_compat &
1668		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1669		    (sb->s_feature_incompat &
1670		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1671			printk(KERN_WARNING
1672				"JBD2: Unrecognised features on journal\n");
1673			return -EINVAL;
1674		}
1675	}
1676
1677	/*
1678	 * Create a slab for this blocksize
1679	 */
1680	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1681	if (err)
1682		return err;
1683
1684	/* Let the recovery code check whether it needs to recover any
1685	 * data from the journal. */
1686	if (jbd2_journal_recover(journal))
1687		goto recovery_error;
1688
1689	if (journal->j_failed_commit) {
1690		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1691		       "is corrupt.\n", journal->j_failed_commit,
1692		       journal->j_devname);
1693		return -EFSCORRUPTED;
1694	}
1695
1696	/* OK, we've finished with the dynamic journal bits:
1697	 * reinitialise the dynamic contents of the superblock in memory
1698	 * and reset them on disk. */
1699	if (journal_reset(journal))
1700		goto recovery_error;
1701
1702	journal->j_flags &= ~JBD2_ABORT;
1703	journal->j_flags |= JBD2_LOADED;
1704	return 0;
1705
1706recovery_error:
1707	printk(KERN_WARNING "JBD2: recovery failed\n");
1708	return -EIO;
1709}
1710
1711/**
1712 * void jbd2_journal_destroy() - Release a journal_t structure.
1713 * @journal: Journal to act on.
1714 *
1715 * Release a journal_t structure once it is no longer in use by the
1716 * journaled object.
1717 * Return <0 if we couldn't clean up the journal.
1718 */
1719int jbd2_journal_destroy(journal_t *journal)
1720{
1721	int err = 0;
1722
1723	/* Wait for the commit thread to wake up and die. */
1724	journal_kill_thread(journal);
1725
1726	/* Force a final log commit */
1727	if (journal->j_running_transaction)
1728		jbd2_journal_commit_transaction(journal);
1729
1730	/* Force any old transactions to disk */
1731
1732	/* Totally anal locking here... */
1733	spin_lock(&journal->j_list_lock);
1734	while (journal->j_checkpoint_transactions != NULL) {
1735		spin_unlock(&journal->j_list_lock);
1736		mutex_lock_io(&journal->j_checkpoint_mutex);
1737		err = jbd2_log_do_checkpoint(journal);
1738		mutex_unlock(&journal->j_checkpoint_mutex);
1739		/*
1740		 * If checkpointing failed, just free the buffers to avoid
1741		 * looping forever
1742		 */
1743		if (err) {
1744			jbd2_journal_destroy_checkpoint(journal);
1745			spin_lock(&journal->j_list_lock);
1746			break;
1747		}
1748		spin_lock(&journal->j_list_lock);
1749	}
1750
1751	J_ASSERT(journal->j_running_transaction == NULL);
1752	J_ASSERT(journal->j_committing_transaction == NULL);
1753	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1754	spin_unlock(&journal->j_list_lock);
1755
1756	if (journal->j_sb_buffer) {
1757		if (!is_journal_aborted(journal)) {
1758			mutex_lock_io(&journal->j_checkpoint_mutex);
1759
1760			write_lock(&journal->j_state_lock);
1761			journal->j_tail_sequence =
1762				++journal->j_transaction_sequence;
1763			write_unlock(&journal->j_state_lock);
1764
1765			jbd2_mark_journal_empty(journal,
1766					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
1767			mutex_unlock(&journal->j_checkpoint_mutex);
1768		} else
1769			err = -EIO;
1770		brelse(journal->j_sb_buffer);
1771	}
1772
1773	if (journal->j_proc_entry)
1774		jbd2_stats_proc_exit(journal);
1775	iput(journal->j_inode);
1776	if (journal->j_revoke)
1777		jbd2_journal_destroy_revoke(journal);
1778	if (journal->j_chksum_driver)
1779		crypto_free_shash(journal->j_chksum_driver);
1780	kfree(journal->j_wbuf);
1781	kfree(journal);
1782
1783	return err;
1784}
1785
1786
1787/**
1788 *int jbd2_journal_check_used_features () - Check if features specified are used.
1789 * @journal: Journal to check.
1790 * @compat: bitmask of compatible features
1791 * @ro: bitmask of features that force read-only mount
1792 * @incompat: bitmask of incompatible features
1793 *
1794 * Check whether the journal uses all of a given set of
1795 * features.  Return true (non-zero) if it does.
1796 **/
1797
1798int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1799				 unsigned long ro, unsigned long incompat)
1800{
1801	journal_superblock_t *sb;
1802
1803	if (!compat && !ro && !incompat)
1804		return 1;
1805	/* Load journal superblock if it is not loaded yet. */
1806	if (journal->j_format_version == 0 &&
1807	    journal_get_superblock(journal) != 0)
1808		return 0;
1809	if (journal->j_format_version == 1)
1810		return 0;
1811
1812	sb = journal->j_superblock;
1813
1814	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1815	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1816	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1817		return 1;
1818
1819	return 0;
1820}
1821
1822/**
1823 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1824 * @journal: Journal to check.
1825 * @compat: bitmask of compatible features
1826 * @ro: bitmask of features that force read-only mount
1827 * @incompat: bitmask of incompatible features
1828 *
1829 * Check whether the journaling code supports the use of
1830 * all of a given set of features on this journal.  Return true
1831 * (non-zero) if it can. */
1832
1833int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1834				      unsigned long ro, unsigned long incompat)
1835{
1836	if (!compat && !ro && !incompat)
1837		return 1;
1838
1839	/* We can support any known requested features iff the
1840	 * superblock is in version 2.  Otherwise we fail to support any
1841	 * extended sb features. */
1842
1843	if (journal->j_format_version != 2)
1844		return 0;
1845
1846	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1847	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1848	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1849		return 1;
1850
1851	return 0;
1852}
1853
1854/**
1855 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1856 * @journal: Journal to act on.
1857 * @compat: bitmask of compatible features
1858 * @ro: bitmask of features that force read-only mount
1859 * @incompat: bitmask of incompatible features
1860 *
1861 * Mark a given journal feature as present on the
1862 * superblock.  Returns true if the requested features could be set.
1863 *
1864 */
1865
1866int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1867			  unsigned long ro, unsigned long incompat)
1868{
1869#define INCOMPAT_FEATURE_ON(f) \
1870		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1871#define COMPAT_FEATURE_ON(f) \
1872		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1873	journal_superblock_t *sb;
1874
1875	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1876		return 1;
1877
1878	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1879		return 0;
1880
1881	/* If enabling v2 checksums, turn on v3 instead */
1882	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1883		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1884		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1885	}
1886
1887	/* Asking for checksumming v3 and v1?  Only give them v3. */
1888	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1889	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1890		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1891
1892	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1893		  compat, ro, incompat);
1894
1895	sb = journal->j_superblock;
1896
1897	/* If enabling v3 checksums, update superblock */
1898	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1899		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1900		sb->s_feature_compat &=
1901			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1902
1903		/* Load the checksum driver */
1904		if (journal->j_chksum_driver == NULL) {
1905			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1906								      0, 0);
1907			if (IS_ERR(journal->j_chksum_driver)) {
1908				printk(KERN_ERR "JBD2: Cannot load crc32c "
1909				       "driver.\n");
1910				journal->j_chksum_driver = NULL;
1911				return 0;
1912			}
1913
1914			/* Precompute checksum seed for all metadata */
1915			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1916							   sb->s_uuid,
1917							   sizeof(sb->s_uuid));
1918		}
1919	}
1920
1921	/* If enabling v1 checksums, downgrade superblock */
1922	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1923		sb->s_feature_incompat &=
1924			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1925				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
1926
1927	sb->s_feature_compat    |= cpu_to_be32(compat);
1928	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1929	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1930
1931	return 1;
1932#undef COMPAT_FEATURE_ON
1933#undef INCOMPAT_FEATURE_ON
1934}
1935
1936/*
1937 * jbd2_journal_clear_features () - Clear a given journal feature in the
1938 * 				    superblock
1939 * @journal: Journal to act on.
1940 * @compat: bitmask of compatible features
1941 * @ro: bitmask of features that force read-only mount
1942 * @incompat: bitmask of incompatible features
1943 *
1944 * Clear a given journal feature as present on the
1945 * superblock.
1946 */
1947void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1948				unsigned long ro, unsigned long incompat)
1949{
1950	journal_superblock_t *sb;
1951
1952	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1953		  compat, ro, incompat);
1954
1955	sb = journal->j_superblock;
1956
1957	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1958	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1959	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1960}
1961EXPORT_SYMBOL(jbd2_journal_clear_features);
1962
1963/**
1964 * int jbd2_journal_flush () - Flush journal
1965 * @journal: Journal to act on.
1966 *
1967 * Flush all data for a given journal to disk and empty the journal.
1968 * Filesystems can use this when remounting readonly to ensure that
1969 * recovery does not need to happen on remount.
1970 */
1971
1972int jbd2_journal_flush(journal_t *journal)
1973{
1974	int err = 0;
1975	transaction_t *transaction = NULL;
1976
1977	write_lock(&journal->j_state_lock);
1978
1979	/* Force everything buffered to the log... */
1980	if (journal->j_running_transaction) {
1981		transaction = journal->j_running_transaction;
1982		__jbd2_log_start_commit(journal, transaction->t_tid);
1983	} else if (journal->j_committing_transaction)
1984		transaction = journal->j_committing_transaction;
1985
1986	/* Wait for the log commit to complete... */
1987	if (transaction) {
1988		tid_t tid = transaction->t_tid;
1989
1990		write_unlock(&journal->j_state_lock);
1991		jbd2_log_wait_commit(journal, tid);
1992	} else {
1993		write_unlock(&journal->j_state_lock);
1994	}
1995
1996	/* ...and flush everything in the log out to disk. */
1997	spin_lock(&journal->j_list_lock);
1998	while (!err && journal->j_checkpoint_transactions != NULL) {
1999		spin_unlock(&journal->j_list_lock);
2000		mutex_lock_io(&journal->j_checkpoint_mutex);
2001		err = jbd2_log_do_checkpoint(journal);
2002		mutex_unlock(&journal->j_checkpoint_mutex);
2003		spin_lock(&journal->j_list_lock);
2004	}
2005	spin_unlock(&journal->j_list_lock);
2006
2007	if (is_journal_aborted(journal))
2008		return -EIO;
2009
2010	mutex_lock_io(&journal->j_checkpoint_mutex);
2011	if (!err) {
2012		err = jbd2_cleanup_journal_tail(journal);
2013		if (err < 0) {
2014			mutex_unlock(&journal->j_checkpoint_mutex);
2015			goto out;
2016		}
2017		err = 0;
2018	}
2019
2020	/* Finally, mark the journal as really needing no recovery.
2021	 * This sets s_start==0 in the underlying superblock, which is
2022	 * the magic code for a fully-recovered superblock.  Any future
2023	 * commits of data to the journal will restore the current
2024	 * s_start value. */
2025	jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2026	mutex_unlock(&journal->j_checkpoint_mutex);
2027	write_lock(&journal->j_state_lock);
2028	J_ASSERT(!journal->j_running_transaction);
2029	J_ASSERT(!journal->j_committing_transaction);
2030	J_ASSERT(!journal->j_checkpoint_transactions);
2031	J_ASSERT(journal->j_head == journal->j_tail);
2032	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2033	write_unlock(&journal->j_state_lock);
2034out:
2035	return err;
2036}
2037
2038/**
2039 * int jbd2_journal_wipe() - Wipe journal contents
2040 * @journal: Journal to act on.
2041 * @write: flag (see below)
2042 *
2043 * Wipe out all of the contents of a journal, safely.  This will produce
2044 * a warning if the journal contains any valid recovery information.
2045 * Must be called between journal_init_*() and jbd2_journal_load().
2046 *
2047 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2048 * we merely suppress recovery.
2049 */
2050
2051int jbd2_journal_wipe(journal_t *journal, int write)
2052{
2053	int err = 0;
2054
2055	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2056
2057	err = load_superblock(journal);
2058	if (err)
2059		return err;
2060
2061	if (!journal->j_tail)
2062		goto no_recovery;
2063
2064	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2065		write ? "Clearing" : "Ignoring");
2066
2067	err = jbd2_journal_skip_recovery(journal);
2068	if (write) {
2069		/* Lock to make assertions happy... */
2070		mutex_lock(&journal->j_checkpoint_mutex);
2071		jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2072		mutex_unlock(&journal->j_checkpoint_mutex);
2073	}
2074
2075 no_recovery:
2076	return err;
2077}
2078
2079/*
2080 * Journal abort has very specific semantics, which we describe
2081 * for journal abort.
2082 *
2083 * Two internal functions, which provide abort to the jbd layer
2084 * itself are here.
2085 */
2086
2087/*
2088 * Quick version for internal journal use (doesn't lock the journal).
2089 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2090 * and don't attempt to make any other journal updates.
2091 */
2092void __jbd2_journal_abort_hard(journal_t *journal)
2093{
2094	transaction_t *transaction;
2095
2096	if (journal->j_flags & JBD2_ABORT)
2097		return;
2098
2099	printk(KERN_ERR "Aborting journal on device %s.\n",
2100	       journal->j_devname);
2101
2102	write_lock(&journal->j_state_lock);
2103	journal->j_flags |= JBD2_ABORT;
2104	transaction = journal->j_running_transaction;
2105	if (transaction)
2106		__jbd2_log_start_commit(journal, transaction->t_tid);
2107	write_unlock(&journal->j_state_lock);
2108}
2109
2110/* Soft abort: record the abort error status in the journal superblock,
2111 * but don't do any other IO. */
2112static void __journal_abort_soft (journal_t *journal, int errno)
2113{
2114	int old_errno;
 
2115
2116	write_lock(&journal->j_state_lock);
2117	old_errno = journal->j_errno;
2118	if (!journal->j_errno || errno == -ESHUTDOWN)
2119		journal->j_errno = errno;
2120
2121	if (journal->j_flags & JBD2_ABORT) {
2122		write_unlock(&journal->j_state_lock);
2123		if (!old_errno && old_errno != -ESHUTDOWN &&
2124		    errno == -ESHUTDOWN)
2125			jbd2_journal_update_sb_errno(journal);
2126		return;
2127	}
2128	write_unlock(&journal->j_state_lock);
2129
2130	__jbd2_journal_abort_hard(journal);
2131
2132	if (errno) {
2133		jbd2_journal_update_sb_errno(journal);
2134		write_lock(&journal->j_state_lock);
2135		journal->j_flags |= JBD2_REC_ERR;
2136		write_unlock(&journal->j_state_lock);
2137	}
2138}
2139
2140/**
2141 * void jbd2_journal_abort () - Shutdown the journal immediately.
2142 * @journal: the journal to shutdown.
2143 * @errno:   an error number to record in the journal indicating
2144 *           the reason for the shutdown.
2145 *
2146 * Perform a complete, immediate shutdown of the ENTIRE
2147 * journal (not of a single transaction).  This operation cannot be
2148 * undone without closing and reopening the journal.
2149 *
2150 * The jbd2_journal_abort function is intended to support higher level error
2151 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2152 * mode.
2153 *
2154 * Journal abort has very specific semantics.  Any existing dirty,
2155 * unjournaled buffers in the main filesystem will still be written to
2156 * disk by bdflush, but the journaling mechanism will be suspended
2157 * immediately and no further transaction commits will be honoured.
2158 *
2159 * Any dirty, journaled buffers will be written back to disk without
2160 * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2161 * filesystem, but we _do_ attempt to leave as much data as possible
2162 * behind for fsck to use for cleanup.
2163 *
2164 * Any attempt to get a new transaction handle on a journal which is in
2165 * ABORT state will just result in an -EROFS error return.  A
2166 * jbd2_journal_stop on an existing handle will return -EIO if we have
2167 * entered abort state during the update.
2168 *
2169 * Recursive transactions are not disturbed by journal abort until the
2170 * final jbd2_journal_stop, which will receive the -EIO error.
2171 *
2172 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2173 * which will be recorded (if possible) in the journal superblock.  This
2174 * allows a client to record failure conditions in the middle of a
2175 * transaction without having to complete the transaction to record the
2176 * failure to disk.  ext3_error, for example, now uses this
2177 * functionality.
2178 *
2179 * Errors which originate from within the journaling layer will NOT
2180 * supply an errno; a null errno implies that absolutely no further
2181 * writes are done to the journal (unless there are any already in
2182 * progress).
2183 *
2184 */
2185
2186void jbd2_journal_abort(journal_t *journal, int errno)
2187{
2188	__journal_abort_soft(journal, errno);
2189}
2190
2191/**
2192 * int jbd2_journal_errno () - returns the journal's error state.
2193 * @journal: journal to examine.
2194 *
2195 * This is the errno number set with jbd2_journal_abort(), the last
2196 * time the journal was mounted - if the journal was stopped
2197 * without calling abort this will be 0.
2198 *
2199 * If the journal has been aborted on this mount time -EROFS will
2200 * be returned.
2201 */
2202int jbd2_journal_errno(journal_t *journal)
2203{
2204	int err;
2205
2206	read_lock(&journal->j_state_lock);
2207	if (journal->j_flags & JBD2_ABORT)
2208		err = -EROFS;
2209	else
2210		err = journal->j_errno;
2211	read_unlock(&journal->j_state_lock);
2212	return err;
2213}
2214
2215/**
2216 * int jbd2_journal_clear_err () - clears the journal's error state
2217 * @journal: journal to act on.
2218 *
2219 * An error must be cleared or acked to take a FS out of readonly
2220 * mode.
2221 */
2222int jbd2_journal_clear_err(journal_t *journal)
2223{
2224	int err = 0;
2225
2226	write_lock(&journal->j_state_lock);
2227	if (journal->j_flags & JBD2_ABORT)
2228		err = -EROFS;
2229	else
2230		journal->j_errno = 0;
2231	write_unlock(&journal->j_state_lock);
2232	return err;
2233}
2234
2235/**
2236 * void jbd2_journal_ack_err() - Ack journal err.
2237 * @journal: journal to act on.
2238 *
2239 * An error must be cleared or acked to take a FS out of readonly
2240 * mode.
2241 */
2242void jbd2_journal_ack_err(journal_t *journal)
2243{
2244	write_lock(&journal->j_state_lock);
2245	if (journal->j_errno)
2246		journal->j_flags |= JBD2_ACK_ERR;
2247	write_unlock(&journal->j_state_lock);
2248}
2249
2250int jbd2_journal_blocks_per_page(struct inode *inode)
2251{
2252	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2253}
2254
2255/*
2256 * helper functions to deal with 32 or 64bit block numbers.
2257 */
2258size_t journal_tag_bytes(journal_t *journal)
2259{
2260	size_t sz;
2261
2262	if (jbd2_has_feature_csum3(journal))
2263		return sizeof(journal_block_tag3_t);
2264
2265	sz = sizeof(journal_block_tag_t);
2266
2267	if (jbd2_has_feature_csum2(journal))
2268		sz += sizeof(__u16);
2269
2270	if (jbd2_has_feature_64bit(journal))
2271		return sz;
2272	else
2273		return sz - sizeof(__u32);
2274}
2275
2276/*
2277 * JBD memory management
2278 *
2279 * These functions are used to allocate block-sized chunks of memory
2280 * used for making copies of buffer_head data.  Very often it will be
2281 * page-sized chunks of data, but sometimes it will be in
2282 * sub-page-size chunks.  (For example, 16k pages on Power systems
2283 * with a 4k block file system.)  For blocks smaller than a page, we
2284 * use a SLAB allocator.  There are slab caches for each block size,
2285 * which are allocated at mount time, if necessary, and we only free
2286 * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2287 * this reason we don't need to a mutex to protect access to
2288 * jbd2_slab[] allocating or releasing memory; only in
2289 * jbd2_journal_create_slab().
2290 */
2291#define JBD2_MAX_SLABS 8
2292static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2293
2294static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2295	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2296	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2297};
2298
2299
2300static void jbd2_journal_destroy_slabs(void)
2301{
2302	int i;
2303
2304	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2305		if (jbd2_slab[i])
2306			kmem_cache_destroy(jbd2_slab[i]);
2307		jbd2_slab[i] = NULL;
2308	}
2309}
2310
2311static int jbd2_journal_create_slab(size_t size)
2312{
2313	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2314	int i = order_base_2(size) - 10;
2315	size_t slab_size;
2316
2317	if (size == PAGE_SIZE)
2318		return 0;
2319
2320	if (i >= JBD2_MAX_SLABS)
2321		return -EINVAL;
2322
2323	if (unlikely(i < 0))
2324		i = 0;
2325	mutex_lock(&jbd2_slab_create_mutex);
2326	if (jbd2_slab[i]) {
2327		mutex_unlock(&jbd2_slab_create_mutex);
2328		return 0;	/* Already created */
2329	}
2330
2331	slab_size = 1 << (i+10);
2332	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2333					 slab_size, 0, NULL);
2334	mutex_unlock(&jbd2_slab_create_mutex);
2335	if (!jbd2_slab[i]) {
2336		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2337		return -ENOMEM;
2338	}
2339	return 0;
2340}
2341
2342static struct kmem_cache *get_slab(size_t size)
2343{
2344	int i = order_base_2(size) - 10;
2345
2346	BUG_ON(i >= JBD2_MAX_SLABS);
2347	if (unlikely(i < 0))
2348		i = 0;
2349	BUG_ON(jbd2_slab[i] == NULL);
2350	return jbd2_slab[i];
2351}
2352
2353void *jbd2_alloc(size_t size, gfp_t flags)
2354{
2355	void *ptr;
2356
2357	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2358
2359	if (size < PAGE_SIZE)
 
 
 
 
 
 
 
 
 
 
2360		ptr = kmem_cache_alloc(get_slab(size), flags);
2361	else
2362		ptr = (void *)__get_free_pages(flags, get_order(size));
2363
2364	/* Check alignment; SLUB has gotten this wrong in the past,
2365	 * and this can lead to user data corruption! */
2366	BUG_ON(((unsigned long) ptr) & (size-1));
2367
2368	return ptr;
2369}
2370
2371void jbd2_free(void *ptr, size_t size)
2372{
2373	if (size < PAGE_SIZE)
2374		kmem_cache_free(get_slab(size), ptr);
2375	else
2376		free_pages((unsigned long)ptr, get_order(size));
 
 
 
 
 
 
 
 
 
 
2377};
2378
2379/*
2380 * Journal_head storage management
2381 */
2382static struct kmem_cache *jbd2_journal_head_cache;
2383#ifdef CONFIG_JBD2_DEBUG
2384static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2385#endif
2386
2387static int jbd2_journal_init_journal_head_cache(void)
2388{
2389	int retval;
2390
2391	J_ASSERT(jbd2_journal_head_cache == NULL);
2392	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2393				sizeof(struct journal_head),
2394				0,		/* offset */
2395				SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2396				NULL);		/* ctor */
2397	retval = 0;
2398	if (!jbd2_journal_head_cache) {
2399		retval = -ENOMEM;
2400		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2401	}
2402	return retval;
2403}
2404
2405static void jbd2_journal_destroy_journal_head_cache(void)
2406{
2407	if (jbd2_journal_head_cache) {
2408		kmem_cache_destroy(jbd2_journal_head_cache);
2409		jbd2_journal_head_cache = NULL;
2410	}
2411}
2412
2413/*
2414 * journal_head splicing and dicing
2415 */
2416static struct journal_head *journal_alloc_journal_head(void)
2417{
2418	struct journal_head *ret;
2419
2420#ifdef CONFIG_JBD2_DEBUG
2421	atomic_inc(&nr_journal_heads);
2422#endif
2423	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2424	if (!ret) {
2425		jbd_debug(1, "out of memory for journal_head\n");
2426		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2427		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2428				GFP_NOFS | __GFP_NOFAIL);
2429	}
2430	return ret;
2431}
2432
2433static void journal_free_journal_head(struct journal_head *jh)
2434{
2435#ifdef CONFIG_JBD2_DEBUG
2436	atomic_dec(&nr_journal_heads);
2437	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2438#endif
2439	kmem_cache_free(jbd2_journal_head_cache, jh);
2440}
2441
2442/*
2443 * A journal_head is attached to a buffer_head whenever JBD has an
2444 * interest in the buffer.
2445 *
2446 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2447 * is set.  This bit is tested in core kernel code where we need to take
2448 * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2449 * there.
2450 *
2451 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2452 *
2453 * When a buffer has its BH_JBD bit set it is immune from being released by
2454 * core kernel code, mainly via ->b_count.
2455 *
2456 * A journal_head is detached from its buffer_head when the journal_head's
2457 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2458 * transaction (b_cp_transaction) hold their references to b_jcount.
2459 *
2460 * Various places in the kernel want to attach a journal_head to a buffer_head
2461 * _before_ attaching the journal_head to a transaction.  To protect the
2462 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2463 * journal_head's b_jcount refcount by one.  The caller must call
2464 * jbd2_journal_put_journal_head() to undo this.
2465 *
2466 * So the typical usage would be:
2467 *
2468 *	(Attach a journal_head if needed.  Increments b_jcount)
2469 *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2470 *	...
2471 *      (Get another reference for transaction)
2472 *	jbd2_journal_grab_journal_head(bh);
2473 *	jh->b_transaction = xxx;
2474 *	(Put original reference)
2475 *	jbd2_journal_put_journal_head(jh);
2476 */
2477
2478/*
2479 * Give a buffer_head a journal_head.
2480 *
2481 * May sleep.
2482 */
2483struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2484{
2485	struct journal_head *jh;
2486	struct journal_head *new_jh = NULL;
2487
2488repeat:
2489	if (!buffer_jbd(bh))
2490		new_jh = journal_alloc_journal_head();
2491
2492	jbd_lock_bh_journal_head(bh);
2493	if (buffer_jbd(bh)) {
2494		jh = bh2jh(bh);
2495	} else {
2496		J_ASSERT_BH(bh,
2497			(atomic_read(&bh->b_count) > 0) ||
2498			(bh->b_page && bh->b_page->mapping));
2499
2500		if (!new_jh) {
2501			jbd_unlock_bh_journal_head(bh);
2502			goto repeat;
2503		}
2504
2505		jh = new_jh;
2506		new_jh = NULL;		/* We consumed it */
2507		set_buffer_jbd(bh);
2508		bh->b_private = jh;
2509		jh->b_bh = bh;
2510		get_bh(bh);
2511		BUFFER_TRACE(bh, "added journal_head");
2512	}
2513	jh->b_jcount++;
2514	jbd_unlock_bh_journal_head(bh);
2515	if (new_jh)
2516		journal_free_journal_head(new_jh);
2517	return bh->b_private;
2518}
2519
2520/*
2521 * Grab a ref against this buffer_head's journal_head.  If it ended up not
2522 * having a journal_head, return NULL
2523 */
2524struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2525{
2526	struct journal_head *jh = NULL;
2527
2528	jbd_lock_bh_journal_head(bh);
2529	if (buffer_jbd(bh)) {
2530		jh = bh2jh(bh);
2531		jh->b_jcount++;
2532	}
2533	jbd_unlock_bh_journal_head(bh);
2534	return jh;
2535}
2536
2537static void __journal_remove_journal_head(struct buffer_head *bh)
2538{
2539	struct journal_head *jh = bh2jh(bh);
2540
2541	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2542	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2543	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2544	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2545	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2546	J_ASSERT_BH(bh, buffer_jbd(bh));
2547	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2548	BUFFER_TRACE(bh, "remove journal_head");
2549	if (jh->b_frozen_data) {
2550		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2551		jbd2_free(jh->b_frozen_data, bh->b_size);
2552	}
2553	if (jh->b_committed_data) {
2554		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2555		jbd2_free(jh->b_committed_data, bh->b_size);
2556	}
2557	bh->b_private = NULL;
2558	jh->b_bh = NULL;	/* debug, really */
2559	clear_buffer_jbd(bh);
2560	journal_free_journal_head(jh);
2561}
2562
2563/*
2564 * Drop a reference on the passed journal_head.  If it fell to zero then
2565 * release the journal_head from the buffer_head.
2566 */
2567void jbd2_journal_put_journal_head(struct journal_head *jh)
2568{
2569	struct buffer_head *bh = jh2bh(jh);
2570
2571	jbd_lock_bh_journal_head(bh);
2572	J_ASSERT_JH(jh, jh->b_jcount > 0);
2573	--jh->b_jcount;
2574	if (!jh->b_jcount) {
2575		__journal_remove_journal_head(bh);
2576		jbd_unlock_bh_journal_head(bh);
2577		__brelse(bh);
2578	} else
2579		jbd_unlock_bh_journal_head(bh);
2580}
2581
2582/*
2583 * Initialize jbd inode head
2584 */
2585void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2586{
2587	jinode->i_transaction = NULL;
2588	jinode->i_next_transaction = NULL;
2589	jinode->i_vfs_inode = inode;
2590	jinode->i_flags = 0;
2591	INIT_LIST_HEAD(&jinode->i_list);
2592}
2593
2594/*
2595 * Function to be called before we start removing inode from memory (i.e.,
2596 * clear_inode() is a fine place to be called from). It removes inode from
2597 * transaction's lists.
2598 */
2599void jbd2_journal_release_jbd_inode(journal_t *journal,
2600				    struct jbd2_inode *jinode)
2601{
2602	if (!journal)
2603		return;
2604restart:
2605	spin_lock(&journal->j_list_lock);
2606	/* Is commit writing out inode - we have to wait */
2607	if (jinode->i_flags & JI_COMMIT_RUNNING) {
2608		wait_queue_head_t *wq;
2609		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2610		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2611		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
2612		spin_unlock(&journal->j_list_lock);
2613		schedule();
2614		finish_wait(wq, &wait.wq_entry);
2615		goto restart;
2616	}
2617
2618	if (jinode->i_transaction) {
2619		list_del(&jinode->i_list);
2620		jinode->i_transaction = NULL;
2621	}
2622	spin_unlock(&journal->j_list_lock);
2623}
2624
2625
2626#ifdef CONFIG_PROC_FS
2627
2628#define JBD2_STATS_PROC_NAME "fs/jbd2"
2629
2630static void __init jbd2_create_jbd_stats_proc_entry(void)
2631{
2632	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2633}
2634
2635static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2636{
2637	if (proc_jbd2_stats)
2638		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2639}
2640
2641#else
2642
2643#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2644#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2645
2646#endif
2647
2648struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2649
2650static int __init jbd2_journal_init_handle_cache(void)
2651{
2652	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2653	if (jbd2_handle_cache == NULL) {
2654		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2655		return -ENOMEM;
2656	}
2657	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2658	if (jbd2_inode_cache == NULL) {
2659		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2660		kmem_cache_destroy(jbd2_handle_cache);
2661		return -ENOMEM;
2662	}
2663	return 0;
2664}
2665
2666static void jbd2_journal_destroy_handle_cache(void)
2667{
2668	if (jbd2_handle_cache)
2669		kmem_cache_destroy(jbd2_handle_cache);
2670	if (jbd2_inode_cache)
2671		kmem_cache_destroy(jbd2_inode_cache);
2672
2673}
2674
2675/*
2676 * Module startup and shutdown
2677 */
2678
2679static int __init journal_init_caches(void)
2680{
2681	int ret;
2682
2683	ret = jbd2_journal_init_revoke_caches();
2684	if (ret == 0)
2685		ret = jbd2_journal_init_journal_head_cache();
2686	if (ret == 0)
2687		ret = jbd2_journal_init_handle_cache();
2688	if (ret == 0)
2689		ret = jbd2_journal_init_transaction_cache();
2690	return ret;
2691}
2692
2693static void jbd2_journal_destroy_caches(void)
2694{
2695	jbd2_journal_destroy_revoke_caches();
2696	jbd2_journal_destroy_journal_head_cache();
2697	jbd2_journal_destroy_handle_cache();
2698	jbd2_journal_destroy_transaction_cache();
2699	jbd2_journal_destroy_slabs();
2700}
2701
2702static int __init journal_init(void)
2703{
2704	int ret;
2705
2706	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2707
2708	ret = journal_init_caches();
2709	if (ret == 0) {
2710		jbd2_create_jbd_stats_proc_entry();
2711	} else {
2712		jbd2_journal_destroy_caches();
2713	}
2714	return ret;
2715}
2716
2717static void __exit journal_exit(void)
2718{
2719#ifdef CONFIG_JBD2_DEBUG
2720	int n = atomic_read(&nr_journal_heads);
2721	if (n)
2722		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2723#endif
2724	jbd2_remove_jbd_stats_proc_entry();
2725	jbd2_journal_destroy_caches();
2726}
2727
2728MODULE_LICENSE("GPL");
2729module_init(journal_init);
2730module_exit(journal_exit);
2731