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
v4.10.11
   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 <linux/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_inode_add_write);
  98EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
  99EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
 100EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
 101EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
 102EXPORT_SYMBOL(jbd2_inode_cache);
 103
 104static void __journal_abort_soft (journal_t *journal, int errno);
 105static int jbd2_journal_create_slab(size_t slab_size);
 106
 107#ifdef CONFIG_JBD2_DEBUG
 108void __jbd2_debug(int level, const char *file, const char *func,
 109		  unsigned int line, const char *fmt, ...)
 110{
 111	struct va_format vaf;
 112	va_list args;
 113
 114	if (level > jbd2_journal_enable_debug)
 115		return;
 116	va_start(args, fmt);
 117	vaf.fmt = fmt;
 118	vaf.va = &args;
 119	printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
 120	va_end(args);
 121}
 122EXPORT_SYMBOL(__jbd2_debug);
 123#endif
 124
 125/* Checksumming functions */
 126static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
 127{
 128	if (!jbd2_journal_has_csum_v2or3_feature(j))
 129		return 1;
 130
 131	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
 132}
 133
 134static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
 135{
 136	__u32 csum;
 137	__be32 old_csum;
 138
 139	old_csum = sb->s_checksum;
 140	sb->s_checksum = 0;
 141	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
 142	sb->s_checksum = old_csum;
 143
 144	return cpu_to_be32(csum);
 145}
 146
 147static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
 148{
 149	if (!jbd2_journal_has_csum_v2or3(j))
 150		return 1;
 151
 152	return sb->s_checksum == jbd2_superblock_csum(j, sb);
 153}
 154
 155static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
 156{
 157	if (!jbd2_journal_has_csum_v2or3(j))
 158		return;
 159
 160	sb->s_checksum = jbd2_superblock_csum(j, sb);
 161}
 162
 163/*
 164 * Helper function used to manage commit timeouts
 165 */
 166
 167static void commit_timeout(unsigned long __data)
 168{
 169	struct task_struct * p = (struct task_struct *) __data;
 170
 171	wake_up_process(p);
 172}
 173
 174/*
 175 * kjournald2: The main thread function used to manage a logging device
 176 * journal.
 177 *
 178 * This kernel thread is responsible for two things:
 179 *
 180 * 1) COMMIT:  Every so often we need to commit the current state of the
 181 *    filesystem to disk.  The journal thread is responsible for writing
 182 *    all of the metadata buffers to disk.
 183 *
 184 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
 185 *    of the data in that part of the log has been rewritten elsewhere on
 186 *    the disk.  Flushing these old buffers to reclaim space in the log is
 187 *    known as checkpointing, and this thread is responsible for that job.
 188 */
 189
 190static int kjournald2(void *arg)
 191{
 192	journal_t *journal = arg;
 193	transaction_t *transaction;
 194
 195	/*
 196	 * Set up an interval timer which can be used to trigger a commit wakeup
 197	 * after the commit interval expires
 198	 */
 199	setup_timer(&journal->j_commit_timer, commit_timeout,
 200			(unsigned long)current);
 201
 202	set_freezable();
 203
 204	/* Record that the journal thread is running */
 205	journal->j_task = current;
 206	wake_up(&journal->j_wait_done_commit);
 207
 208	/*
 209	 * And now, wait forever for commit wakeup events.
 210	 */
 211	write_lock(&journal->j_state_lock);
 212
 213loop:
 214	if (journal->j_flags & JBD2_UNMOUNT)
 215		goto end_loop;
 216
 217	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
 218		journal->j_commit_sequence, journal->j_commit_request);
 219
 220	if (journal->j_commit_sequence != journal->j_commit_request) {
 221		jbd_debug(1, "OK, requests differ\n");
 222		write_unlock(&journal->j_state_lock);
 223		del_timer_sync(&journal->j_commit_timer);
 224		jbd2_journal_commit_transaction(journal);
 225		write_lock(&journal->j_state_lock);
 226		goto loop;
 227	}
 228
 229	wake_up(&journal->j_wait_done_commit);
 230	if (freezing(current)) {
 231		/*
 232		 * The simpler the better. Flushing journal isn't a
 233		 * good idea, because that depends on threads that may
 234		 * be already stopped.
 235		 */
 236		jbd_debug(1, "Now suspending kjournald2\n");
 237		write_unlock(&journal->j_state_lock);
 238		try_to_freeze();
 239		write_lock(&journal->j_state_lock);
 240	} else {
 241		/*
 242		 * We assume on resume that commits are already there,
 243		 * so we don't sleep
 244		 */
 245		DEFINE_WAIT(wait);
 246		int should_sleep = 1;
 247
 248		prepare_to_wait(&journal->j_wait_commit, &wait,
 249				TASK_INTERRUPTIBLE);
 250		if (journal->j_commit_sequence != journal->j_commit_request)
 251			should_sleep = 0;
 252		transaction = journal->j_running_transaction;
 253		if (transaction && time_after_eq(jiffies,
 254						transaction->t_expires))
 255			should_sleep = 0;
 256		if (journal->j_flags & JBD2_UNMOUNT)
 257			should_sleep = 0;
 258		if (should_sleep) {
 259			write_unlock(&journal->j_state_lock);
 260			schedule();
 261			write_lock(&journal->j_state_lock);
 262		}
 263		finish_wait(&journal->j_wait_commit, &wait);
 264	}
 265
 266	jbd_debug(1, "kjournald2 wakes\n");
 267
 268	/*
 269	 * Were we woken up by a commit wakeup event?
 270	 */
 271	transaction = journal->j_running_transaction;
 272	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
 273		journal->j_commit_request = transaction->t_tid;
 274		jbd_debug(1, "woke because of timeout\n");
 275	}
 276	goto loop;
 277
 278end_loop:
 279	write_unlock(&journal->j_state_lock);
 280	del_timer_sync(&journal->j_commit_timer);
 281	journal->j_task = NULL;
 282	wake_up(&journal->j_wait_done_commit);
 283	jbd_debug(1, "Journal thread exiting.\n");
 284	return 0;
 285}
 286
 287static int jbd2_journal_start_thread(journal_t *journal)
 288{
 289	struct task_struct *t;
 290
 291	t = kthread_run(kjournald2, journal, "jbd2/%s",
 292			journal->j_devname);
 293	if (IS_ERR(t))
 294		return PTR_ERR(t);
 295
 296	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
 297	return 0;
 298}
 299
 300static void journal_kill_thread(journal_t *journal)
 301{
 302	write_lock(&journal->j_state_lock);
 303	journal->j_flags |= JBD2_UNMOUNT;
 304
 305	while (journal->j_task) {
 306		write_unlock(&journal->j_state_lock);
 307		wake_up(&journal->j_wait_commit);
 308		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
 309		write_lock(&journal->j_state_lock);
 310	}
 311	write_unlock(&journal->j_state_lock);
 312}
 313
 314/*
 315 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
 316 *
 317 * Writes a metadata buffer to a given disk block.  The actual IO is not
 318 * performed but a new buffer_head is constructed which labels the data
 319 * to be written with the correct destination disk block.
 320 *
 321 * Any magic-number escaping which needs to be done will cause a
 322 * copy-out here.  If the buffer happens to start with the
 323 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
 324 * magic number is only written to the log for descripter blocks.  In
 325 * this case, we copy the data and replace the first word with 0, and we
 326 * return a result code which indicates that this buffer needs to be
 327 * marked as an escaped buffer in the corresponding log descriptor
 328 * block.  The missing word can then be restored when the block is read
 329 * during recovery.
 330 *
 331 * If the source buffer has already been modified by a new transaction
 332 * since we took the last commit snapshot, we use the frozen copy of
 333 * that data for IO. If we end up using the existing buffer_head's data
 334 * for the write, then we have to make sure nobody modifies it while the
 335 * IO is in progress. do_get_write_access() handles this.
 336 *
 337 * The function returns a pointer to the buffer_head to be used for IO.
 338 * 
 339 *
 340 * Return value:
 341 *  <0: Error
 342 * >=0: Finished OK
 343 *
 344 * On success:
 345 * Bit 0 set == escape performed on the data
 346 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
 347 */
 348
 349int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
 350				  struct journal_head  *jh_in,
 351				  struct buffer_head **bh_out,
 352				  sector_t blocknr)
 353{
 354	int need_copy_out = 0;
 355	int done_copy_out = 0;
 356	int do_escape = 0;
 357	char *mapped_data;
 358	struct buffer_head *new_bh;
 359	struct page *new_page;
 360	unsigned int new_offset;
 361	struct buffer_head *bh_in = jh2bh(jh_in);
 362	journal_t *journal = transaction->t_journal;
 363
 364	/*
 365	 * The buffer really shouldn't be locked: only the current committing
 366	 * transaction is allowed to write it, so nobody else is allowed
 367	 * to do any IO.
 368	 *
 369	 * akpm: except if we're journalling data, and write() output is
 370	 * also part of a shared mapping, and another thread has
 371	 * decided to launch a writepage() against this buffer.
 372	 */
 373	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
 374
 375	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
 376
 377	/* keep subsequent assertions sane */
 378	atomic_set(&new_bh->b_count, 1);
 379
 380	jbd_lock_bh_state(bh_in);
 381repeat:
 382	/*
 383	 * If a new transaction has already done a buffer copy-out, then
 384	 * we use that version of the data for the commit.
 385	 */
 386	if (jh_in->b_frozen_data) {
 387		done_copy_out = 1;
 388		new_page = virt_to_page(jh_in->b_frozen_data);
 389		new_offset = offset_in_page(jh_in->b_frozen_data);
 390	} else {
 391		new_page = jh2bh(jh_in)->b_page;
 392		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
 393	}
 394
 395	mapped_data = kmap_atomic(new_page);
 396	/*
 397	 * Fire data frozen trigger if data already wasn't frozen.  Do this
 398	 * before checking for escaping, as the trigger may modify the magic
 399	 * offset.  If a copy-out happens afterwards, it will have the correct
 400	 * data in the buffer.
 401	 */
 402	if (!done_copy_out)
 403		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
 404					   jh_in->b_triggers);
 405
 406	/*
 407	 * Check for escaping
 408	 */
 409	if (*((__be32 *)(mapped_data + new_offset)) ==
 410				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
 411		need_copy_out = 1;
 412		do_escape = 1;
 413	}
 414	kunmap_atomic(mapped_data);
 415
 416	/*
 417	 * Do we need to do a data copy?
 418	 */
 419	if (need_copy_out && !done_copy_out) {
 420		char *tmp;
 421
 422		jbd_unlock_bh_state(bh_in);
 423		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
 424		if (!tmp) {
 425			brelse(new_bh);
 426			return -ENOMEM;
 427		}
 428		jbd_lock_bh_state(bh_in);
 429		if (jh_in->b_frozen_data) {
 430			jbd2_free(tmp, bh_in->b_size);
 431			goto repeat;
 432		}
 433
 434		jh_in->b_frozen_data = tmp;
 435		mapped_data = kmap_atomic(new_page);
 436		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
 437		kunmap_atomic(mapped_data);
 438
 439		new_page = virt_to_page(tmp);
 440		new_offset = offset_in_page(tmp);
 441		done_copy_out = 1;
 442
 443		/*
 444		 * This isn't strictly necessary, as we're using frozen
 445		 * data for the escaping, but it keeps consistency with
 446		 * b_frozen_data usage.
 447		 */
 448		jh_in->b_frozen_triggers = jh_in->b_triggers;
 449	}
 450
 451	/*
 452	 * Did we need to do an escaping?  Now we've done all the
 453	 * copying, we can finally do so.
 454	 */
 455	if (do_escape) {
 456		mapped_data = kmap_atomic(new_page);
 457		*((unsigned int *)(mapped_data + new_offset)) = 0;
 458		kunmap_atomic(mapped_data);
 459	}
 460
 461	set_bh_page(new_bh, new_page, new_offset);
 462	new_bh->b_size = bh_in->b_size;
 463	new_bh->b_bdev = journal->j_dev;
 464	new_bh->b_blocknr = blocknr;
 465	new_bh->b_private = bh_in;
 466	set_buffer_mapped(new_bh);
 467	set_buffer_dirty(new_bh);
 468
 469	*bh_out = new_bh;
 470
 471	/*
 472	 * The to-be-written buffer needs to get moved to the io queue,
 473	 * and the original buffer whose contents we are shadowing or
 474	 * copying is moved to the transaction's shadow queue.
 475	 */
 476	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
 477	spin_lock(&journal->j_list_lock);
 478	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
 479	spin_unlock(&journal->j_list_lock);
 480	set_buffer_shadow(bh_in);
 481	jbd_unlock_bh_state(bh_in);
 482
 483	return do_escape | (done_copy_out << 1);
 484}
 485
 486/*
 487 * Allocation code for the journal file.  Manage the space left in the
 488 * journal, so that we can begin checkpointing when appropriate.
 489 */
 490
 491/*
 492 * Called with j_state_lock locked for writing.
 493 * Returns true if a transaction commit was started.
 494 */
 495int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 496{
 497	/* Return if the txn has already requested to be committed */
 498	if (journal->j_commit_request == target)
 499		return 0;
 500
 501	/*
 502	 * The only transaction we can possibly wait upon is the
 503	 * currently running transaction (if it exists).  Otherwise,
 504	 * the target tid must be an old one.
 505	 */
 506	if (journal->j_running_transaction &&
 507	    journal->j_running_transaction->t_tid == target) {
 508		/*
 509		 * We want a new commit: OK, mark the request and wakeup the
 510		 * commit thread.  We do _not_ do the commit ourselves.
 511		 */
 512
 513		journal->j_commit_request = target;
 514		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
 515			  journal->j_commit_request,
 516			  journal->j_commit_sequence);
 517		journal->j_running_transaction->t_requested = jiffies;
 518		wake_up(&journal->j_wait_commit);
 519		return 1;
 520	} else if (!tid_geq(journal->j_commit_request, target))
 521		/* This should never happen, but if it does, preserve
 522		   the evidence before kjournald goes into a loop and
 523		   increments j_commit_sequence beyond all recognition. */
 524		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
 525			  journal->j_commit_request,
 526			  journal->j_commit_sequence,
 527			  target, journal->j_running_transaction ? 
 528			  journal->j_running_transaction->t_tid : 0);
 529	return 0;
 530}
 531
 532int jbd2_log_start_commit(journal_t *journal, tid_t tid)
 533{
 534	int ret;
 535
 536	write_lock(&journal->j_state_lock);
 537	ret = __jbd2_log_start_commit(journal, tid);
 538	write_unlock(&journal->j_state_lock);
 539	return ret;
 540}
 541
 542/*
 543 * Force and wait any uncommitted transactions.  We can only force the running
 544 * transaction if we don't have an active handle, otherwise, we will deadlock.
 545 * Returns: <0 in case of error,
 546 *           0 if nothing to commit,
 547 *           1 if transaction was successfully committed.
 548 */
 549static int __jbd2_journal_force_commit(journal_t *journal)
 550{
 551	transaction_t *transaction = NULL;
 552	tid_t tid;
 553	int need_to_start = 0, ret = 0;
 554
 555	read_lock(&journal->j_state_lock);
 556	if (journal->j_running_transaction && !current->journal_info) {
 557		transaction = journal->j_running_transaction;
 558		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
 559			need_to_start = 1;
 560	} else if (journal->j_committing_transaction)
 561		transaction = journal->j_committing_transaction;
 562
 563	if (!transaction) {
 564		/* Nothing to commit */
 565		read_unlock(&journal->j_state_lock);
 566		return 0;
 567	}
 568	tid = transaction->t_tid;
 569	read_unlock(&journal->j_state_lock);
 570	if (need_to_start)
 571		jbd2_log_start_commit(journal, tid);
 572	ret = jbd2_log_wait_commit(journal, tid);
 573	if (!ret)
 574		ret = 1;
 575
 576	return ret;
 577}
 578
 579/**
 580 * Force and wait upon a commit if the calling process is not within
 581 * transaction.  This is used for forcing out undo-protected data which contains
 582 * bitmaps, when the fs is running out of space.
 583 *
 584 * @journal: journal to force
 585 * Returns true if progress was made.
 586 */
 587int jbd2_journal_force_commit_nested(journal_t *journal)
 588{
 589	int ret;
 590
 591	ret = __jbd2_journal_force_commit(journal);
 592	return ret > 0;
 593}
 594
 595/**
 596 * int journal_force_commit() - force any uncommitted transactions
 597 * @journal: journal to force
 598 *
 599 * Caller want unconditional commit. We can only force the running transaction
 600 * if we don't have an active handle, otherwise, we will deadlock.
 601 */
 602int jbd2_journal_force_commit(journal_t *journal)
 603{
 604	int ret;
 605
 606	J_ASSERT(!current->journal_info);
 607	ret = __jbd2_journal_force_commit(journal);
 608	if (ret > 0)
 609		ret = 0;
 610	return ret;
 611}
 612
 613/*
 614 * Start a commit of the current running transaction (if any).  Returns true
 615 * if a transaction is going to be committed (or is currently already
 616 * committing), and fills its tid in at *ptid
 617 */
 618int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
 619{
 620	int ret = 0;
 621
 622	write_lock(&journal->j_state_lock);
 623	if (journal->j_running_transaction) {
 624		tid_t tid = journal->j_running_transaction->t_tid;
 625
 626		__jbd2_log_start_commit(journal, tid);
 627		/* There's a running transaction and we've just made sure
 628		 * it's commit has been scheduled. */
 629		if (ptid)
 630			*ptid = tid;
 631		ret = 1;
 632	} else if (journal->j_committing_transaction) {
 633		/*
 634		 * If commit has been started, then we have to wait for
 635		 * completion of that transaction.
 636		 */
 637		if (ptid)
 638			*ptid = journal->j_committing_transaction->t_tid;
 639		ret = 1;
 640	}
 641	write_unlock(&journal->j_state_lock);
 642	return ret;
 643}
 644
 645/*
 646 * Return 1 if a given transaction has not yet sent barrier request
 647 * connected with a transaction commit. If 0 is returned, transaction
 648 * may or may not have sent the barrier. Used to avoid sending barrier
 649 * twice in common cases.
 650 */
 651int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
 652{
 653	int ret = 0;
 654	transaction_t *commit_trans;
 655
 656	if (!(journal->j_flags & JBD2_BARRIER))
 657		return 0;
 658	read_lock(&journal->j_state_lock);
 659	/* Transaction already committed? */
 660	if (tid_geq(journal->j_commit_sequence, tid))
 661		goto out;
 662	commit_trans = journal->j_committing_transaction;
 663	if (!commit_trans || commit_trans->t_tid != tid) {
 664		ret = 1;
 665		goto out;
 666	}
 667	/*
 668	 * Transaction is being committed and we already proceeded to
 669	 * submitting a flush to fs partition?
 670	 */
 671	if (journal->j_fs_dev != journal->j_dev) {
 672		if (!commit_trans->t_need_data_flush ||
 673		    commit_trans->t_state >= T_COMMIT_DFLUSH)
 674			goto out;
 675	} else {
 676		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
 677			goto out;
 678	}
 679	ret = 1;
 680out:
 681	read_unlock(&journal->j_state_lock);
 682	return ret;
 683}
 684EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
 685
 686/*
 687 * Wait for a specified commit to complete.
 688 * The caller may not hold the journal lock.
 689 */
 690int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
 691{
 692	int err = 0;
 693
 694	jbd2_might_wait_for_commit(journal);
 695	read_lock(&journal->j_state_lock);
 696#ifdef CONFIG_JBD2_DEBUG
 697	if (!tid_geq(journal->j_commit_request, tid)) {
 698		printk(KERN_ERR
 699		       "%s: error: j_commit_request=%d, tid=%d\n",
 700		       __func__, journal->j_commit_request, tid);
 701	}
 702#endif
 703	while (tid_gt(tid, journal->j_commit_sequence)) {
 704		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
 705				  tid, journal->j_commit_sequence);
 706		read_unlock(&journal->j_state_lock);
 707		wake_up(&journal->j_wait_commit);
 708		wait_event(journal->j_wait_done_commit,
 709				!tid_gt(tid, journal->j_commit_sequence));
 710		read_lock(&journal->j_state_lock);
 711	}
 712	read_unlock(&journal->j_state_lock);
 713
 714	if (unlikely(is_journal_aborted(journal)))
 715		err = -EIO;
 716	return err;
 717}
 718
 719/*
 720 * When this function returns the transaction corresponding to tid
 721 * will be completed.  If the transaction has currently running, start
 722 * committing that transaction before waiting for it to complete.  If
 723 * the transaction id is stale, it is by definition already completed,
 724 * so just return SUCCESS.
 725 */
 726int jbd2_complete_transaction(journal_t *journal, tid_t tid)
 727{
 728	int	need_to_wait = 1;
 729
 730	read_lock(&journal->j_state_lock);
 731	if (journal->j_running_transaction &&
 732	    journal->j_running_transaction->t_tid == tid) {
 733		if (journal->j_commit_request != tid) {
 734			/* transaction not yet started, so request it */
 735			read_unlock(&journal->j_state_lock);
 736			jbd2_log_start_commit(journal, tid);
 737			goto wait_commit;
 738		}
 739	} else if (!(journal->j_committing_transaction &&
 740		     journal->j_committing_transaction->t_tid == tid))
 741		need_to_wait = 0;
 742	read_unlock(&journal->j_state_lock);
 743	if (!need_to_wait)
 744		return 0;
 745wait_commit:
 746	return jbd2_log_wait_commit(journal, tid);
 747}
 748EXPORT_SYMBOL(jbd2_complete_transaction);
 749
 750/*
 751 * Log buffer allocation routines:
 752 */
 753
 754int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
 755{
 756	unsigned long blocknr;
 757
 758	write_lock(&journal->j_state_lock);
 759	J_ASSERT(journal->j_free > 1);
 760
 761	blocknr = journal->j_head;
 762	journal->j_head++;
 763	journal->j_free--;
 764	if (journal->j_head == journal->j_last)
 765		journal->j_head = journal->j_first;
 766	write_unlock(&journal->j_state_lock);
 767	return jbd2_journal_bmap(journal, blocknr, retp);
 768}
 769
 770/*
 771 * Conversion of logical to physical block numbers for the journal
 772 *
 773 * On external journals the journal blocks are identity-mapped, so
 774 * this is a no-op.  If needed, we can use j_blk_offset - everything is
 775 * ready.
 776 */
 777int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
 778		 unsigned long long *retp)
 779{
 780	int err = 0;
 781	unsigned long long ret;
 782
 783	if (journal->j_inode) {
 784		ret = bmap(journal->j_inode, blocknr);
 785		if (ret)
 786			*retp = ret;
 787		else {
 788			printk(KERN_ALERT "%s: journal block not found "
 789					"at offset %lu on %s\n",
 790			       __func__, blocknr, journal->j_devname);
 791			err = -EIO;
 792			__journal_abort_soft(journal, err);
 793		}
 794	} else {
 795		*retp = blocknr; /* +journal->j_blk_offset */
 796	}
 797	return err;
 798}
 799
 800/*
 801 * We play buffer_head aliasing tricks to write data/metadata blocks to
 802 * the journal without copying their contents, but for journal
 803 * descriptor blocks we do need to generate bona fide buffers.
 804 *
 805 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
 806 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
 807 * But we don't bother doing that, so there will be coherency problems with
 808 * mmaps of blockdevs which hold live JBD-controlled filesystems.
 809 */
 810struct buffer_head *
 811jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
 812{
 813	journal_t *journal = transaction->t_journal;
 814	struct buffer_head *bh;
 815	unsigned long long blocknr;
 816	journal_header_t *header;
 817	int err;
 818
 819	err = jbd2_journal_next_log_block(journal, &blocknr);
 820
 821	if (err)
 822		return NULL;
 823
 824	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
 825	if (!bh)
 826		return NULL;
 827	lock_buffer(bh);
 828	memset(bh->b_data, 0, journal->j_blocksize);
 829	header = (journal_header_t *)bh->b_data;
 830	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
 831	header->h_blocktype = cpu_to_be32(type);
 832	header->h_sequence = cpu_to_be32(transaction->t_tid);
 833	set_buffer_uptodate(bh);
 834	unlock_buffer(bh);
 835	BUFFER_TRACE(bh, "return this buffer");
 836	return bh;
 837}
 838
 839void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
 840{
 841	struct jbd2_journal_block_tail *tail;
 842	__u32 csum;
 843
 844	if (!jbd2_journal_has_csum_v2or3(j))
 845		return;
 846
 847	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
 848			sizeof(struct jbd2_journal_block_tail));
 849	tail->t_checksum = 0;
 850	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
 851	tail->t_checksum = cpu_to_be32(csum);
 852}
 853
 854/*
 855 * Return tid of the oldest transaction in the journal and block in the journal
 856 * where the transaction starts.
 857 *
 858 * If the journal is now empty, return which will be the next transaction ID
 859 * we will write and where will that transaction start.
 860 *
 861 * The return value is 0 if journal tail cannot be pushed any further, 1 if
 862 * it can.
 863 */
 864int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
 865			      unsigned long *block)
 866{
 867	transaction_t *transaction;
 868	int ret;
 869
 870	read_lock(&journal->j_state_lock);
 871	spin_lock(&journal->j_list_lock);
 872	transaction = journal->j_checkpoint_transactions;
 873	if (transaction) {
 874		*tid = transaction->t_tid;
 875		*block = transaction->t_log_start;
 876	} else if ((transaction = journal->j_committing_transaction) != NULL) {
 877		*tid = transaction->t_tid;
 878		*block = transaction->t_log_start;
 879	} else if ((transaction = journal->j_running_transaction) != NULL) {
 880		*tid = transaction->t_tid;
 881		*block = journal->j_head;
 882	} else {
 883		*tid = journal->j_transaction_sequence;
 884		*block = journal->j_head;
 885	}
 886	ret = tid_gt(*tid, journal->j_tail_sequence);
 887	spin_unlock(&journal->j_list_lock);
 888	read_unlock(&journal->j_state_lock);
 889
 890	return ret;
 891}
 892
 893/*
 894 * Update information in journal structure and in on disk journal superblock
 895 * about log tail. This function does not check whether information passed in
 896 * really pushes log tail further. It's responsibility of the caller to make
 897 * sure provided log tail information is valid (e.g. by holding
 898 * j_checkpoint_mutex all the time between computing log tail and calling this
 899 * function as is the case with jbd2_cleanup_journal_tail()).
 900 *
 901 * Requires j_checkpoint_mutex
 902 */
 903int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 904{
 905	unsigned long freed;
 906	int ret;
 907
 908	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 909
 910	/*
 911	 * We cannot afford for write to remain in drive's caches since as
 912	 * soon as we update j_tail, next transaction can start reusing journal
 913	 * space and if we lose sb update during power failure we'd replay
 914	 * old transaction with possibly newly overwritten data.
 915	 */
 916	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
 917	if (ret)
 918		goto out;
 919
 920	write_lock(&journal->j_state_lock);
 921	freed = block - journal->j_tail;
 922	if (block < journal->j_tail)
 923		freed += journal->j_last - journal->j_first;
 924
 925	trace_jbd2_update_log_tail(journal, tid, block, freed);
 926	jbd_debug(1,
 927		  "Cleaning journal tail from %d to %d (offset %lu), "
 928		  "freeing %lu\n",
 929		  journal->j_tail_sequence, tid, block, freed);
 930
 931	journal->j_free += freed;
 932	journal->j_tail_sequence = tid;
 933	journal->j_tail = block;
 934	write_unlock(&journal->j_state_lock);
 935
 936out:
 937	return ret;
 938}
 939
 940/*
 941 * This is a variaon of __jbd2_update_log_tail which checks for validity of
 942 * provided log tail and locks j_checkpoint_mutex. So it is safe against races
 943 * with other threads updating log tail.
 944 */
 945void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
 946{
 947	mutex_lock(&journal->j_checkpoint_mutex);
 948	if (tid_gt(tid, journal->j_tail_sequence))
 949		__jbd2_update_log_tail(journal, tid, block);
 950	mutex_unlock(&journal->j_checkpoint_mutex);
 951}
 952
 953struct jbd2_stats_proc_session {
 954	journal_t *journal;
 955	struct transaction_stats_s *stats;
 956	int start;
 957	int max;
 958};
 959
 960static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
 961{
 962	return *pos ? NULL : SEQ_START_TOKEN;
 963}
 964
 965static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
 966{
 967	return NULL;
 968}
 969
 970static int jbd2_seq_info_show(struct seq_file *seq, void *v)
 971{
 972	struct jbd2_stats_proc_session *s = seq->private;
 973
 974	if (v != SEQ_START_TOKEN)
 975		return 0;
 976	seq_printf(seq, "%lu transactions (%lu requested), "
 977		   "each up to %u blocks\n",
 978		   s->stats->ts_tid, s->stats->ts_requested,
 979		   s->journal->j_max_transaction_buffers);
 980	if (s->stats->ts_tid == 0)
 981		return 0;
 982	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
 983	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
 984	seq_printf(seq, "  %ums request delay\n",
 985	    (s->stats->ts_requested == 0) ? 0 :
 986	    jiffies_to_msecs(s->stats->run.rs_request_delay /
 987			     s->stats->ts_requested));
 988	seq_printf(seq, "  %ums running transaction\n",
 989	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
 990	seq_printf(seq, "  %ums transaction was being locked\n",
 991	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
 992	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
 993	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
 994	seq_printf(seq, "  %ums logging transaction\n",
 995	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
 996	seq_printf(seq, "  %lluus average transaction commit time\n",
 997		   div_u64(s->journal->j_average_commit_time, 1000));
 998	seq_printf(seq, "  %lu handles per transaction\n",
 999	    s->stats->run.rs_handle_count / s->stats->ts_tid);
1000	seq_printf(seq, "  %lu blocks per transaction\n",
1001	    s->stats->run.rs_blocks / s->stats->ts_tid);
1002	seq_printf(seq, "  %lu logged blocks per transaction\n",
1003	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1004	return 0;
1005}
1006
1007static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1008{
1009}
1010
1011static const struct seq_operations jbd2_seq_info_ops = {
1012	.start  = jbd2_seq_info_start,
1013	.next   = jbd2_seq_info_next,
1014	.stop   = jbd2_seq_info_stop,
1015	.show   = jbd2_seq_info_show,
1016};
1017
1018static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1019{
1020	journal_t *journal = PDE_DATA(inode);
1021	struct jbd2_stats_proc_session *s;
1022	int rc, size;
1023
1024	s = kmalloc(sizeof(*s), GFP_KERNEL);
1025	if (s == NULL)
1026		return -ENOMEM;
1027	size = sizeof(struct transaction_stats_s);
1028	s->stats = kmalloc(size, GFP_KERNEL);
1029	if (s->stats == NULL) {
1030		kfree(s);
1031		return -ENOMEM;
1032	}
1033	spin_lock(&journal->j_history_lock);
1034	memcpy(s->stats, &journal->j_stats, size);
1035	s->journal = journal;
1036	spin_unlock(&journal->j_history_lock);
1037
1038	rc = seq_open(file, &jbd2_seq_info_ops);
1039	if (rc == 0) {
1040		struct seq_file *m = file->private_data;
1041		m->private = s;
1042	} else {
1043		kfree(s->stats);
1044		kfree(s);
1045	}
1046	return rc;
1047
1048}
1049
1050static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1051{
1052	struct seq_file *seq = file->private_data;
1053	struct jbd2_stats_proc_session *s = seq->private;
1054	kfree(s->stats);
1055	kfree(s);
1056	return seq_release(inode, file);
1057}
1058
1059static const struct file_operations jbd2_seq_info_fops = {
1060	.owner		= THIS_MODULE,
1061	.open           = jbd2_seq_info_open,
1062	.read           = seq_read,
1063	.llseek         = seq_lseek,
1064	.release        = jbd2_seq_info_release,
1065};
1066
1067static struct proc_dir_entry *proc_jbd2_stats;
1068
1069static void jbd2_stats_proc_init(journal_t *journal)
1070{
1071	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1072	if (journal->j_proc_entry) {
1073		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1074				 &jbd2_seq_info_fops, journal);
1075	}
1076}
1077
1078static void jbd2_stats_proc_exit(journal_t *journal)
1079{
1080	remove_proc_entry("info", journal->j_proc_entry);
1081	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1082}
1083
1084/*
1085 * Management for journal control blocks: functions to create and
1086 * destroy journal_t structures, and to initialise and read existing
1087 * journal blocks from disk.  */
1088
1089/* First: create and setup a journal_t object in memory.  We initialise
1090 * very few fields yet: that has to wait until we have created the
1091 * journal structures from from scratch, or loaded them from disk. */
1092
1093static journal_t *journal_init_common(struct block_device *bdev,
1094			struct block_device *fs_dev,
1095			unsigned long long start, int len, int blocksize)
1096{
1097	static struct lock_class_key jbd2_trans_commit_key;
1098	journal_t *journal;
1099	int err;
1100	struct buffer_head *bh;
1101	int n;
1102
1103	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1104	if (!journal)
1105		return NULL;
1106
1107	init_waitqueue_head(&journal->j_wait_transaction_locked);
1108	init_waitqueue_head(&journal->j_wait_done_commit);
1109	init_waitqueue_head(&journal->j_wait_commit);
1110	init_waitqueue_head(&journal->j_wait_updates);
1111	init_waitqueue_head(&journal->j_wait_reserved);
1112	mutex_init(&journal->j_barrier);
1113	mutex_init(&journal->j_checkpoint_mutex);
1114	spin_lock_init(&journal->j_revoke_lock);
1115	spin_lock_init(&journal->j_list_lock);
1116	rwlock_init(&journal->j_state_lock);
1117
1118	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1119	journal->j_min_batch_time = 0;
1120	journal->j_max_batch_time = 15000; /* 15ms */
1121	atomic_set(&journal->j_reserved_credits, 0);
1122
1123	/* The journal is marked for error until we succeed with recovery! */
1124	journal->j_flags = JBD2_ABORT;
1125
1126	/* Set up a default-sized revoke table for the new mount. */
1127	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1128	if (err)
1129		goto err_cleanup;
 
 
1130
1131	spin_lock_init(&journal->j_history_lock);
1132
1133	lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1134			 &jbd2_trans_commit_key, 0);
1135
1136	/* journal descriptor can store up to n blocks -bzzz */
1137	journal->j_blocksize = blocksize;
1138	journal->j_dev = bdev;
1139	journal->j_fs_dev = fs_dev;
1140	journal->j_blk_offset = start;
1141	journal->j_maxlen = len;
1142	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1143	journal->j_wbufsize = n;
1144	journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1145					GFP_KERNEL);
1146	if (!journal->j_wbuf)
1147		goto err_cleanup;
1148
1149	bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1150	if (!bh) {
1151		pr_err("%s: Cannot get buffer for journal superblock\n",
1152			__func__);
1153		goto err_cleanup;
1154	}
1155	journal->j_sb_buffer = bh;
1156	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1157
1158	return journal;
1159
1160err_cleanup:
1161	kfree(journal->j_wbuf);
1162	jbd2_journal_destroy_revoke(journal);
1163	kfree(journal);
1164	return NULL;
1165}
1166
1167/* jbd2_journal_init_dev and jbd2_journal_init_inode:
1168 *
1169 * Create a journal structure assigned some fixed set of disk blocks to
1170 * the journal.  We don't actually touch those disk blocks yet, but we
1171 * need to set up all of the mapping information to tell the journaling
1172 * system where the journal blocks are.
1173 *
1174 */
1175
1176/**
1177 *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1178 *  @bdev: Block device on which to create the journal
1179 *  @fs_dev: Device which hold journalled filesystem for this journal.
1180 *  @start: Block nr Start of journal.
1181 *  @len:  Length of the journal in blocks.
1182 *  @blocksize: blocksize of journalling device
1183 *
1184 *  Returns: a newly created journal_t *
1185 *
1186 *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1187 *  range of blocks on an arbitrary block device.
1188 *
1189 */
1190journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1191			struct block_device *fs_dev,
1192			unsigned long long start, int len, int blocksize)
1193{
1194	journal_t *journal;
 
 
1195
1196	journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1197	if (!journal)
1198		return NULL;
1199
 
 
 
 
 
 
1200	bdevname(journal->j_dev, journal->j_devname);
1201	strreplace(journal->j_devname, '/', '!');
1202	jbd2_stats_proc_init(journal);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1203
1204	return journal;
 
 
 
 
 
1205}
1206
1207/**
1208 *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1209 *  @inode: An inode to create the journal in
1210 *
1211 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1212 * the journal.  The inode must exist already, must support bmap() and
1213 * must have all data blocks preallocated.
1214 */
1215journal_t *jbd2_journal_init_inode(struct inode *inode)
1216{
1217	journal_t *journal;
 
1218	char *p;
 
 
1219	unsigned long long blocknr;
1220
1221	blocknr = bmap(inode, 0);
1222	if (!blocknr) {
1223		pr_err("%s: Cannot locate journal superblock\n",
1224			__func__);
1225		return NULL;
1226	}
1227
1228	jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1229		  inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1230		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1231
1232	journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1233			blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1234			inode->i_sb->s_blocksize);
1235	if (!journal)
1236		return NULL;
1237
 
1238	journal->j_inode = inode;
1239	bdevname(journal->j_dev, journal->j_devname);
1240	p = strreplace(journal->j_devname, '/', '!');
1241	sprintf(p, "-%lu", journal->j_inode->i_ino);
 
 
 
 
 
 
 
 
1242	jbd2_stats_proc_init(journal);
1243
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1244	return journal;
 
 
 
 
 
1245}
1246
1247/*
1248 * If the journal init or create aborts, we need to mark the journal
1249 * superblock as being NULL to prevent the journal destroy from writing
1250 * back a bogus superblock.
1251 */
1252static void journal_fail_superblock (journal_t *journal)
1253{
1254	struct buffer_head *bh = journal->j_sb_buffer;
1255	brelse(bh);
1256	journal->j_sb_buffer = NULL;
1257}
1258
1259/*
1260 * Given a journal_t structure, initialise the various fields for
1261 * startup of a new journaling session.  We use this both when creating
1262 * a journal, and after recovering an old journal to reset it for
1263 * subsequent use.
1264 */
1265
1266static int journal_reset(journal_t *journal)
1267{
1268	journal_superblock_t *sb = journal->j_superblock;
1269	unsigned long long first, last;
1270
1271	first = be32_to_cpu(sb->s_first);
1272	last = be32_to_cpu(sb->s_maxlen);
1273	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1274		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1275		       first, last);
1276		journal_fail_superblock(journal);
1277		return -EINVAL;
1278	}
1279
1280	journal->j_first = first;
1281	journal->j_last = last;
1282
1283	journal->j_head = first;
1284	journal->j_tail = first;
1285	journal->j_free = last - first;
1286
1287	journal->j_tail_sequence = journal->j_transaction_sequence;
1288	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1289	journal->j_commit_request = journal->j_commit_sequence;
1290
1291	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1292
1293	/*
1294	 * As a special case, if the on-disk copy is already marked as needing
1295	 * no recovery (s_start == 0), then we can safely defer the superblock
1296	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1297	 * attempting a write to a potential-readonly device.
1298	 */
1299	if (sb->s_start == 0) {
1300		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1301			"(start %ld, seq %d, errno %d)\n",
1302			journal->j_tail, journal->j_tail_sequence,
1303			journal->j_errno);
1304		journal->j_flags |= JBD2_FLUSHED;
1305	} else {
1306		/* Lock here to make assertions happy... */
1307		mutex_lock(&journal->j_checkpoint_mutex);
1308		/*
1309		 * Update log tail information. We use REQ_FUA since new
1310		 * transaction will start reusing journal space and so we
1311		 * must make sure information about current log tail is on
1312		 * disk before that.
1313		 */
1314		jbd2_journal_update_sb_log_tail(journal,
1315						journal->j_tail_sequence,
1316						journal->j_tail,
1317						REQ_FUA);
1318		mutex_unlock(&journal->j_checkpoint_mutex);
1319	}
1320	return jbd2_journal_start_thread(journal);
1321}
1322
1323static int jbd2_write_superblock(journal_t *journal, int write_flags)
1324{
1325	struct buffer_head *bh = journal->j_sb_buffer;
1326	journal_superblock_t *sb = journal->j_superblock;
1327	int ret;
1328
1329	trace_jbd2_write_superblock(journal, write_flags);
1330	if (!(journal->j_flags & JBD2_BARRIER))
1331		write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1332	lock_buffer(bh);
1333	if (buffer_write_io_error(bh)) {
1334		/*
1335		 * Oh, dear.  A previous attempt to write the journal
1336		 * superblock failed.  This could happen because the
1337		 * USB device was yanked out.  Or it could happen to
1338		 * be a transient write error and maybe the block will
1339		 * be remapped.  Nothing we can do but to retry the
1340		 * write and hope for the best.
1341		 */
1342		printk(KERN_ERR "JBD2: previous I/O error detected "
1343		       "for journal superblock update for %s.\n",
1344		       journal->j_devname);
1345		clear_buffer_write_io_error(bh);
1346		set_buffer_uptodate(bh);
1347	}
1348	jbd2_superblock_csum_set(journal, sb);
1349	get_bh(bh);
1350	bh->b_end_io = end_buffer_write_sync;
1351	ret = submit_bh(REQ_OP_WRITE, write_flags, bh);
1352	wait_on_buffer(bh);
1353	if (buffer_write_io_error(bh)) {
1354		clear_buffer_write_io_error(bh);
1355		set_buffer_uptodate(bh);
1356		ret = -EIO;
1357	}
1358	if (ret) {
1359		printk(KERN_ERR "JBD2: Error %d detected when updating "
1360		       "journal superblock for %s.\n", ret,
1361		       journal->j_devname);
1362		jbd2_journal_abort(journal, ret);
1363	}
1364
1365	return ret;
1366}
1367
1368/**
1369 * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1370 * @journal: The journal to update.
1371 * @tail_tid: TID of the new transaction at the tail of the log
1372 * @tail_block: The first block of the transaction at the tail of the log
1373 * @write_op: With which operation should we write the journal sb
1374 *
1375 * Update a journal's superblock information about log tail and write it to
1376 * disk, waiting for the IO to complete.
1377 */
1378int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1379				     unsigned long tail_block, int write_op)
1380{
1381	journal_superblock_t *sb = journal->j_superblock;
1382	int ret;
1383
1384	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1385	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1386		  tail_block, tail_tid);
1387
1388	sb->s_sequence = cpu_to_be32(tail_tid);
1389	sb->s_start    = cpu_to_be32(tail_block);
1390
1391	ret = jbd2_write_superblock(journal, write_op);
1392	if (ret)
1393		goto out;
1394
1395	/* Log is no longer empty */
1396	write_lock(&journal->j_state_lock);
1397	WARN_ON(!sb->s_sequence);
1398	journal->j_flags &= ~JBD2_FLUSHED;
1399	write_unlock(&journal->j_state_lock);
1400
1401out:
1402	return ret;
1403}
1404
1405/**
1406 * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1407 * @journal: The journal to update.
1408 * @write_op: With which operation should we write the journal sb
1409 *
1410 * Update a journal's dynamic superblock fields to show that journal is empty.
1411 * Write updated superblock to disk waiting for IO to complete.
1412 */
1413static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1414{
1415	journal_superblock_t *sb = journal->j_superblock;
1416
1417	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1418	read_lock(&journal->j_state_lock);
1419	/* Is it already empty? */
1420	if (sb->s_start == 0) {
1421		read_unlock(&journal->j_state_lock);
1422		return;
1423	}
1424	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1425		  journal->j_tail_sequence);
1426
1427	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1428	sb->s_start    = cpu_to_be32(0);
1429	read_unlock(&journal->j_state_lock);
1430
1431	jbd2_write_superblock(journal, write_op);
1432
1433	/* Log is no longer empty */
1434	write_lock(&journal->j_state_lock);
1435	journal->j_flags |= JBD2_FLUSHED;
1436	write_unlock(&journal->j_state_lock);
1437}
1438
1439
1440/**
1441 * jbd2_journal_update_sb_errno() - Update error in the journal.
1442 * @journal: The journal to update.
1443 *
1444 * Update a journal's errno.  Write updated superblock to disk waiting for IO
1445 * to complete.
1446 */
1447void jbd2_journal_update_sb_errno(journal_t *journal)
1448{
1449	journal_superblock_t *sb = journal->j_superblock;
1450
1451	read_lock(&journal->j_state_lock);
1452	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1453		  journal->j_errno);
1454	sb->s_errno    = cpu_to_be32(journal->j_errno);
1455	read_unlock(&journal->j_state_lock);
1456
1457	jbd2_write_superblock(journal, REQ_FUA);
1458}
1459EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1460
1461/*
1462 * Read the superblock for a given journal, performing initial
1463 * validation of the format.
1464 */
1465static int journal_get_superblock(journal_t *journal)
1466{
1467	struct buffer_head *bh;
1468	journal_superblock_t *sb;
1469	int err = -EIO;
1470
1471	bh = journal->j_sb_buffer;
1472
1473	J_ASSERT(bh != NULL);
1474	if (!buffer_uptodate(bh)) {
1475		ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1476		wait_on_buffer(bh);
1477		if (!buffer_uptodate(bh)) {
1478			printk(KERN_ERR
1479				"JBD2: IO error reading journal superblock\n");
1480			goto out;
1481		}
1482	}
1483
1484	if (buffer_verified(bh))
1485		return 0;
1486
1487	sb = journal->j_superblock;
1488
1489	err = -EINVAL;
1490
1491	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1492	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1493		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1494		goto out;
1495	}
1496
1497	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1498	case JBD2_SUPERBLOCK_V1:
1499		journal->j_format_version = 1;
1500		break;
1501	case JBD2_SUPERBLOCK_V2:
1502		journal->j_format_version = 2;
1503		break;
1504	default:
1505		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1506		goto out;
1507	}
1508
1509	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1510		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1511	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1512		printk(KERN_WARNING "JBD2: journal file too short\n");
1513		goto out;
1514	}
1515
1516	if (be32_to_cpu(sb->s_first) == 0 ||
1517	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1518		printk(KERN_WARNING
1519			"JBD2: Invalid start block of journal: %u\n",
1520			be32_to_cpu(sb->s_first));
1521		goto out;
1522	}
1523
1524	if (jbd2_has_feature_csum2(journal) &&
1525	    jbd2_has_feature_csum3(journal)) {
1526		/* Can't have checksum v2 and v3 at the same time! */
1527		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1528		       "at the same time!\n");
1529		goto out;
1530	}
1531
1532	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1533	    jbd2_has_feature_checksum(journal)) {
1534		/* Can't have checksum v1 and v2 on at the same time! */
1535		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1536		       "at the same time!\n");
1537		goto out;
1538	}
1539
1540	if (!jbd2_verify_csum_type(journal, sb)) {
1541		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1542		goto out;
1543	}
1544
1545	/* Load the checksum driver */
1546	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1547		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1548		if (IS_ERR(journal->j_chksum_driver)) {
1549			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1550			err = PTR_ERR(journal->j_chksum_driver);
1551			journal->j_chksum_driver = NULL;
1552			goto out;
1553		}
1554	}
1555
1556	/* Check superblock checksum */
1557	if (!jbd2_superblock_csum_verify(journal, sb)) {
1558		printk(KERN_ERR "JBD2: journal checksum error\n");
1559		err = -EFSBADCRC;
1560		goto out;
1561	}
1562
1563	/* Precompute checksum seed for all metadata */
1564	if (jbd2_journal_has_csum_v2or3(journal))
1565		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1566						   sizeof(sb->s_uuid));
1567
1568	set_buffer_verified(bh);
1569
1570	return 0;
1571
1572out:
1573	journal_fail_superblock(journal);
1574	return err;
1575}
1576
1577/*
1578 * Load the on-disk journal superblock and read the key fields into the
1579 * journal_t.
1580 */
1581
1582static int load_superblock(journal_t *journal)
1583{
1584	int err;
1585	journal_superblock_t *sb;
1586
1587	err = journal_get_superblock(journal);
1588	if (err)
1589		return err;
1590
1591	sb = journal->j_superblock;
1592
1593	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1594	journal->j_tail = be32_to_cpu(sb->s_start);
1595	journal->j_first = be32_to_cpu(sb->s_first);
1596	journal->j_last = be32_to_cpu(sb->s_maxlen);
1597	journal->j_errno = be32_to_cpu(sb->s_errno);
1598
1599	return 0;
1600}
1601
1602
1603/**
1604 * int jbd2_journal_load() - Read journal from disk.
1605 * @journal: Journal to act on.
1606 *
1607 * Given a journal_t structure which tells us which disk blocks contain
1608 * a journal, read the journal from disk to initialise the in-memory
1609 * structures.
1610 */
1611int jbd2_journal_load(journal_t *journal)
1612{
1613	int err;
1614	journal_superblock_t *sb;
1615
1616	err = load_superblock(journal);
1617	if (err)
1618		return err;
1619
1620	sb = journal->j_superblock;
1621	/* If this is a V2 superblock, then we have to check the
1622	 * features flags on it. */
1623
1624	if (journal->j_format_version >= 2) {
1625		if ((sb->s_feature_ro_compat &
1626		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1627		    (sb->s_feature_incompat &
1628		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1629			printk(KERN_WARNING
1630				"JBD2: Unrecognised features on journal\n");
1631			return -EINVAL;
1632		}
1633	}
1634
1635	/*
1636	 * Create a slab for this blocksize
1637	 */
1638	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1639	if (err)
1640		return err;
1641
1642	/* Let the recovery code check whether it needs to recover any
1643	 * data from the journal. */
1644	if (jbd2_journal_recover(journal))
1645		goto recovery_error;
1646
1647	if (journal->j_failed_commit) {
1648		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1649		       "is corrupt.\n", journal->j_failed_commit,
1650		       journal->j_devname);
1651		return -EFSCORRUPTED;
1652	}
1653
1654	/* OK, we've finished with the dynamic journal bits:
1655	 * reinitialise the dynamic contents of the superblock in memory
1656	 * and reset them on disk. */
1657	if (journal_reset(journal))
1658		goto recovery_error;
1659
1660	journal->j_flags &= ~JBD2_ABORT;
1661	journal->j_flags |= JBD2_LOADED;
1662	return 0;
1663
1664recovery_error:
1665	printk(KERN_WARNING "JBD2: recovery failed\n");
1666	return -EIO;
1667}
1668
1669/**
1670 * void jbd2_journal_destroy() - Release a journal_t structure.
1671 * @journal: Journal to act on.
1672 *
1673 * Release a journal_t structure once it is no longer in use by the
1674 * journaled object.
1675 * Return <0 if we couldn't clean up the journal.
1676 */
1677int jbd2_journal_destroy(journal_t *journal)
1678{
1679	int err = 0;
1680
1681	/* Wait for the commit thread to wake up and die. */
1682	journal_kill_thread(journal);
1683
1684	/* Force a final log commit */
1685	if (journal->j_running_transaction)
1686		jbd2_journal_commit_transaction(journal);
1687
1688	/* Force any old transactions to disk */
1689
1690	/* Totally anal locking here... */
1691	spin_lock(&journal->j_list_lock);
1692	while (journal->j_checkpoint_transactions != NULL) {
1693		spin_unlock(&journal->j_list_lock);
1694		mutex_lock(&journal->j_checkpoint_mutex);
1695		err = jbd2_log_do_checkpoint(journal);
1696		mutex_unlock(&journal->j_checkpoint_mutex);
1697		/*
1698		 * If checkpointing failed, just free the buffers to avoid
1699		 * looping forever
1700		 */
1701		if (err) {
1702			jbd2_journal_destroy_checkpoint(journal);
1703			spin_lock(&journal->j_list_lock);
1704			break;
1705		}
1706		spin_lock(&journal->j_list_lock);
1707	}
1708
1709	J_ASSERT(journal->j_running_transaction == NULL);
1710	J_ASSERT(journal->j_committing_transaction == NULL);
1711	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1712	spin_unlock(&journal->j_list_lock);
1713
1714	if (journal->j_sb_buffer) {
1715		if (!is_journal_aborted(journal)) {
1716			mutex_lock(&journal->j_checkpoint_mutex);
1717
1718			write_lock(&journal->j_state_lock);
1719			journal->j_tail_sequence =
1720				++journal->j_transaction_sequence;
1721			write_unlock(&journal->j_state_lock);
1722
1723			jbd2_mark_journal_empty(journal,
1724					REQ_PREFLUSH | REQ_FUA);
1725			mutex_unlock(&journal->j_checkpoint_mutex);
1726		} else
1727			err = -EIO;
1728		brelse(journal->j_sb_buffer);
1729	}
1730
1731	if (journal->j_proc_entry)
1732		jbd2_stats_proc_exit(journal);
1733	iput(journal->j_inode);
1734	if (journal->j_revoke)
1735		jbd2_journal_destroy_revoke(journal);
1736	if (journal->j_chksum_driver)
1737		crypto_free_shash(journal->j_chksum_driver);
1738	kfree(journal->j_wbuf);
1739	kfree(journal);
1740
1741	return err;
1742}
1743
1744
1745/**
1746 *int jbd2_journal_check_used_features () - Check if features specified are used.
1747 * @journal: Journal to check.
1748 * @compat: bitmask of compatible features
1749 * @ro: bitmask of features that force read-only mount
1750 * @incompat: bitmask of incompatible features
1751 *
1752 * Check whether the journal uses all of a given set of
1753 * features.  Return true (non-zero) if it does.
1754 **/
1755
1756int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1757				 unsigned long ro, unsigned long incompat)
1758{
1759	journal_superblock_t *sb;
1760
1761	if (!compat && !ro && !incompat)
1762		return 1;
1763	/* Load journal superblock if it is not loaded yet. */
1764	if (journal->j_format_version == 0 &&
1765	    journal_get_superblock(journal) != 0)
1766		return 0;
1767	if (journal->j_format_version == 1)
1768		return 0;
1769
1770	sb = journal->j_superblock;
1771
1772	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1773	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1774	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1775		return 1;
1776
1777	return 0;
1778}
1779
1780/**
1781 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1782 * @journal: Journal to check.
1783 * @compat: bitmask of compatible features
1784 * @ro: bitmask of features that force read-only mount
1785 * @incompat: bitmask of incompatible features
1786 *
1787 * Check whether the journaling code supports the use of
1788 * all of a given set of features on this journal.  Return true
1789 * (non-zero) if it can. */
1790
1791int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1792				      unsigned long ro, unsigned long incompat)
1793{
1794	if (!compat && !ro && !incompat)
1795		return 1;
1796
1797	/* We can support any known requested features iff the
1798	 * superblock is in version 2.  Otherwise we fail to support any
1799	 * extended sb features. */
1800
1801	if (journal->j_format_version != 2)
1802		return 0;
1803
1804	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1805	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1806	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1807		return 1;
1808
1809	return 0;
1810}
1811
1812/**
1813 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1814 * @journal: Journal to act on.
1815 * @compat: bitmask of compatible features
1816 * @ro: bitmask of features that force read-only mount
1817 * @incompat: bitmask of incompatible features
1818 *
1819 * Mark a given journal feature as present on the
1820 * superblock.  Returns true if the requested features could be set.
1821 *
1822 */
1823
1824int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1825			  unsigned long ro, unsigned long incompat)
1826{
1827#define INCOMPAT_FEATURE_ON(f) \
1828		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1829#define COMPAT_FEATURE_ON(f) \
1830		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1831	journal_superblock_t *sb;
1832
1833	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1834		return 1;
1835
1836	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1837		return 0;
1838
1839	/* If enabling v2 checksums, turn on v3 instead */
1840	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1841		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1842		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1843	}
1844
1845	/* Asking for checksumming v3 and v1?  Only give them v3. */
1846	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1847	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1848		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1849
1850	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1851		  compat, ro, incompat);
1852
1853	sb = journal->j_superblock;
1854
1855	/* If enabling v3 checksums, update superblock */
1856	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1857		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1858		sb->s_feature_compat &=
1859			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1860
1861		/* Load the checksum driver */
1862		if (journal->j_chksum_driver == NULL) {
1863			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1864								      0, 0);
1865			if (IS_ERR(journal->j_chksum_driver)) {
1866				printk(KERN_ERR "JBD2: Cannot load crc32c "
1867				       "driver.\n");
1868				journal->j_chksum_driver = NULL;
1869				return 0;
1870			}
1871
1872			/* Precompute checksum seed for all metadata */
1873			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1874							   sb->s_uuid,
1875							   sizeof(sb->s_uuid));
1876		}
1877	}
1878
1879	/* If enabling v1 checksums, downgrade superblock */
1880	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1881		sb->s_feature_incompat &=
1882			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1883				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
1884
1885	sb->s_feature_compat    |= cpu_to_be32(compat);
1886	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1887	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1888
1889	return 1;
1890#undef COMPAT_FEATURE_ON
1891#undef INCOMPAT_FEATURE_ON
1892}
1893
1894/*
1895 * jbd2_journal_clear_features () - Clear a given journal feature in the
1896 * 				    superblock
1897 * @journal: Journal to act on.
1898 * @compat: bitmask of compatible features
1899 * @ro: bitmask of features that force read-only mount
1900 * @incompat: bitmask of incompatible features
1901 *
1902 * Clear a given journal feature as present on the
1903 * superblock.
1904 */
1905void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1906				unsigned long ro, unsigned long incompat)
1907{
1908	journal_superblock_t *sb;
1909
1910	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1911		  compat, ro, incompat);
1912
1913	sb = journal->j_superblock;
1914
1915	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1916	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1917	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1918}
1919EXPORT_SYMBOL(jbd2_journal_clear_features);
1920
1921/**
1922 * int jbd2_journal_flush () - Flush journal
1923 * @journal: Journal to act on.
1924 *
1925 * Flush all data for a given journal to disk and empty the journal.
1926 * Filesystems can use this when remounting readonly to ensure that
1927 * recovery does not need to happen on remount.
1928 */
1929
1930int jbd2_journal_flush(journal_t *journal)
1931{
1932	int err = 0;
1933	transaction_t *transaction = NULL;
1934
1935	write_lock(&journal->j_state_lock);
1936
1937	/* Force everything buffered to the log... */
1938	if (journal->j_running_transaction) {
1939		transaction = journal->j_running_transaction;
1940		__jbd2_log_start_commit(journal, transaction->t_tid);
1941	} else if (journal->j_committing_transaction)
1942		transaction = journal->j_committing_transaction;
1943
1944	/* Wait for the log commit to complete... */
1945	if (transaction) {
1946		tid_t tid = transaction->t_tid;
1947
1948		write_unlock(&journal->j_state_lock);
1949		jbd2_log_wait_commit(journal, tid);
1950	} else {
1951		write_unlock(&journal->j_state_lock);
1952	}
1953
1954	/* ...and flush everything in the log out to disk. */
1955	spin_lock(&journal->j_list_lock);
1956	while (!err && journal->j_checkpoint_transactions != NULL) {
1957		spin_unlock(&journal->j_list_lock);
1958		mutex_lock(&journal->j_checkpoint_mutex);
1959		err = jbd2_log_do_checkpoint(journal);
1960		mutex_unlock(&journal->j_checkpoint_mutex);
1961		spin_lock(&journal->j_list_lock);
1962	}
1963	spin_unlock(&journal->j_list_lock);
1964
1965	if (is_journal_aborted(journal))
1966		return -EIO;
1967
1968	mutex_lock(&journal->j_checkpoint_mutex);
1969	if (!err) {
1970		err = jbd2_cleanup_journal_tail(journal);
1971		if (err < 0) {
1972			mutex_unlock(&journal->j_checkpoint_mutex);
1973			goto out;
1974		}
1975		err = 0;
1976	}
1977
1978	/* Finally, mark the journal as really needing no recovery.
1979	 * This sets s_start==0 in the underlying superblock, which is
1980	 * the magic code for a fully-recovered superblock.  Any future
1981	 * commits of data to the journal will restore the current
1982	 * s_start value. */
1983	jbd2_mark_journal_empty(journal, REQ_FUA);
1984	mutex_unlock(&journal->j_checkpoint_mutex);
1985	write_lock(&journal->j_state_lock);
1986	J_ASSERT(!journal->j_running_transaction);
1987	J_ASSERT(!journal->j_committing_transaction);
1988	J_ASSERT(!journal->j_checkpoint_transactions);
1989	J_ASSERT(journal->j_head == journal->j_tail);
1990	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1991	write_unlock(&journal->j_state_lock);
1992out:
1993	return err;
1994}
1995
1996/**
1997 * int jbd2_journal_wipe() - Wipe journal contents
1998 * @journal: Journal to act on.
1999 * @write: flag (see below)
2000 *
2001 * Wipe out all of the contents of a journal, safely.  This will produce
2002 * a warning if the journal contains any valid recovery information.
2003 * Must be called between journal_init_*() and jbd2_journal_load().
2004 *
2005 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2006 * we merely suppress recovery.
2007 */
2008
2009int jbd2_journal_wipe(journal_t *journal, int write)
2010{
2011	int err = 0;
2012
2013	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2014
2015	err = load_superblock(journal);
2016	if (err)
2017		return err;
2018
2019	if (!journal->j_tail)
2020		goto no_recovery;
2021
2022	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2023		write ? "Clearing" : "Ignoring");
2024
2025	err = jbd2_journal_skip_recovery(journal);
2026	if (write) {
2027		/* Lock to make assertions happy... */
2028		mutex_lock(&journal->j_checkpoint_mutex);
2029		jbd2_mark_journal_empty(journal, REQ_FUA);
2030		mutex_unlock(&journal->j_checkpoint_mutex);
2031	}
2032
2033 no_recovery:
2034	return err;
2035}
2036
2037/*
2038 * Journal abort has very specific semantics, which we describe
2039 * for journal abort.
2040 *
2041 * Two internal functions, which provide abort to the jbd layer
2042 * itself are here.
2043 */
2044
2045/*
2046 * Quick version for internal journal use (doesn't lock the journal).
2047 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2048 * and don't attempt to make any other journal updates.
2049 */
2050void __jbd2_journal_abort_hard(journal_t *journal)
2051{
2052	transaction_t *transaction;
2053
2054	if (journal->j_flags & JBD2_ABORT)
2055		return;
2056
2057	printk(KERN_ERR "Aborting journal on device %s.\n",
2058	       journal->j_devname);
2059
2060	write_lock(&journal->j_state_lock);
2061	journal->j_flags |= JBD2_ABORT;
2062	transaction = journal->j_running_transaction;
2063	if (transaction)
2064		__jbd2_log_start_commit(journal, transaction->t_tid);
2065	write_unlock(&journal->j_state_lock);
2066}
2067
2068/* Soft abort: record the abort error status in the journal superblock,
2069 * but don't do any other IO. */
2070static void __journal_abort_soft (journal_t *journal, int errno)
2071{
2072	if (journal->j_flags & JBD2_ABORT)
2073		return;
2074
2075	if (!journal->j_errno)
2076		journal->j_errno = errno;
2077
2078	__jbd2_journal_abort_hard(journal);
2079
2080	if (errno) {
2081		jbd2_journal_update_sb_errno(journal);
2082		write_lock(&journal->j_state_lock);
2083		journal->j_flags |= JBD2_REC_ERR;
2084		write_unlock(&journal->j_state_lock);
2085	}
2086}
2087
2088/**
2089 * void jbd2_journal_abort () - Shutdown the journal immediately.
2090 * @journal: the journal to shutdown.
2091 * @errno:   an error number to record in the journal indicating
2092 *           the reason for the shutdown.
2093 *
2094 * Perform a complete, immediate shutdown of the ENTIRE
2095 * journal (not of a single transaction).  This operation cannot be
2096 * undone without closing and reopening the journal.
2097 *
2098 * The jbd2_journal_abort function is intended to support higher level error
2099 * recovery mechanisms such as the ext2/ext3 remount-readonly error
2100 * mode.
2101 *
2102 * Journal abort has very specific semantics.  Any existing dirty,
2103 * unjournaled buffers in the main filesystem will still be written to
2104 * disk by bdflush, but the journaling mechanism will be suspended
2105 * immediately and no further transaction commits will be honoured.
2106 *
2107 * Any dirty, journaled buffers will be written back to disk without
2108 * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2109 * filesystem, but we _do_ attempt to leave as much data as possible
2110 * behind for fsck to use for cleanup.
2111 *
2112 * Any attempt to get a new transaction handle on a journal which is in
2113 * ABORT state will just result in an -EROFS error return.  A
2114 * jbd2_journal_stop on an existing handle will return -EIO if we have
2115 * entered abort state during the update.
2116 *
2117 * Recursive transactions are not disturbed by journal abort until the
2118 * final jbd2_journal_stop, which will receive the -EIO error.
2119 *
2120 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2121 * which will be recorded (if possible) in the journal superblock.  This
2122 * allows a client to record failure conditions in the middle of a
2123 * transaction without having to complete the transaction to record the
2124 * failure to disk.  ext3_error, for example, now uses this
2125 * functionality.
2126 *
2127 * Errors which originate from within the journaling layer will NOT
2128 * supply an errno; a null errno implies that absolutely no further
2129 * writes are done to the journal (unless there are any already in
2130 * progress).
2131 *
2132 */
2133
2134void jbd2_journal_abort(journal_t *journal, int errno)
2135{
2136	__journal_abort_soft(journal, errno);
2137}
2138
2139/**
2140 * int jbd2_journal_errno () - returns the journal's error state.
2141 * @journal: journal to examine.
2142 *
2143 * This is the errno number set with jbd2_journal_abort(), the last
2144 * time the journal was mounted - if the journal was stopped
2145 * without calling abort this will be 0.
2146 *
2147 * If the journal has been aborted on this mount time -EROFS will
2148 * be returned.
2149 */
2150int jbd2_journal_errno(journal_t *journal)
2151{
2152	int err;
2153
2154	read_lock(&journal->j_state_lock);
2155	if (journal->j_flags & JBD2_ABORT)
2156		err = -EROFS;
2157	else
2158		err = journal->j_errno;
2159	read_unlock(&journal->j_state_lock);
2160	return err;
2161}
2162
2163/**
2164 * int jbd2_journal_clear_err () - clears the journal's error state
2165 * @journal: journal to act on.
2166 *
2167 * An error must be cleared or acked to take a FS out of readonly
2168 * mode.
2169 */
2170int jbd2_journal_clear_err(journal_t *journal)
2171{
2172	int err = 0;
2173
2174	write_lock(&journal->j_state_lock);
2175	if (journal->j_flags & JBD2_ABORT)
2176		err = -EROFS;
2177	else
2178		journal->j_errno = 0;
2179	write_unlock(&journal->j_state_lock);
2180	return err;
2181}
2182
2183/**
2184 * void jbd2_journal_ack_err() - Ack journal err.
2185 * @journal: journal to act on.
2186 *
2187 * An error must be cleared or acked to take a FS out of readonly
2188 * mode.
2189 */
2190void jbd2_journal_ack_err(journal_t *journal)
2191{
2192	write_lock(&journal->j_state_lock);
2193	if (journal->j_errno)
2194		journal->j_flags |= JBD2_ACK_ERR;
2195	write_unlock(&journal->j_state_lock);
2196}
2197
2198int jbd2_journal_blocks_per_page(struct inode *inode)
2199{
2200	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2201}
2202
2203/*
2204 * helper functions to deal with 32 or 64bit block numbers.
2205 */
2206size_t journal_tag_bytes(journal_t *journal)
2207{
2208	size_t sz;
2209
2210	if (jbd2_has_feature_csum3(journal))
2211		return sizeof(journal_block_tag3_t);
2212
2213	sz = sizeof(journal_block_tag_t);
2214
2215	if (jbd2_has_feature_csum2(journal))
2216		sz += sizeof(__u16);
2217
2218	if (jbd2_has_feature_64bit(journal))
2219		return sz;
2220	else
2221		return sz - sizeof(__u32);
2222}
2223
2224/*
2225 * JBD memory management
2226 *
2227 * These functions are used to allocate block-sized chunks of memory
2228 * used for making copies of buffer_head data.  Very often it will be
2229 * page-sized chunks of data, but sometimes it will be in
2230 * sub-page-size chunks.  (For example, 16k pages on Power systems
2231 * with a 4k block file system.)  For blocks smaller than a page, we
2232 * use a SLAB allocator.  There are slab caches for each block size,
2233 * which are allocated at mount time, if necessary, and we only free
2234 * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2235 * this reason we don't need to a mutex to protect access to
2236 * jbd2_slab[] allocating or releasing memory; only in
2237 * jbd2_journal_create_slab().
2238 */
2239#define JBD2_MAX_SLABS 8
2240static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2241
2242static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2243	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2244	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2245};
2246
2247
2248static void jbd2_journal_destroy_slabs(void)
2249{
2250	int i;
2251
2252	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2253		if (jbd2_slab[i])
2254			kmem_cache_destroy(jbd2_slab[i]);
2255		jbd2_slab[i] = NULL;
2256	}
2257}
2258
2259static int jbd2_journal_create_slab(size_t size)
2260{
2261	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2262	int i = order_base_2(size) - 10;
2263	size_t slab_size;
2264
2265	if (size == PAGE_SIZE)
2266		return 0;
2267
2268	if (i >= JBD2_MAX_SLABS)
2269		return -EINVAL;
2270
2271	if (unlikely(i < 0))
2272		i = 0;
2273	mutex_lock(&jbd2_slab_create_mutex);
2274	if (jbd2_slab[i]) {
2275		mutex_unlock(&jbd2_slab_create_mutex);
2276		return 0;	/* Already created */
2277	}
2278
2279	slab_size = 1 << (i+10);
2280	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2281					 slab_size, 0, NULL);
2282	mutex_unlock(&jbd2_slab_create_mutex);
2283	if (!jbd2_slab[i]) {
2284		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2285		return -ENOMEM;
2286	}
2287	return 0;
2288}
2289
2290static struct kmem_cache *get_slab(size_t size)
2291{
2292	int i = order_base_2(size) - 10;
2293
2294	BUG_ON(i >= JBD2_MAX_SLABS);
2295	if (unlikely(i < 0))
2296		i = 0;
2297	BUG_ON(jbd2_slab[i] == NULL);
2298	return jbd2_slab[i];
2299}
2300
2301void *jbd2_alloc(size_t size, gfp_t flags)
2302{
2303	void *ptr;
2304
2305	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2306
2307	if (size < PAGE_SIZE)
 
 
 
 
 
 
 
 
 
 
2308		ptr = kmem_cache_alloc(get_slab(size), flags);
2309	else
2310		ptr = (void *)__get_free_pages(flags, get_order(size));
2311
2312	/* Check alignment; SLUB has gotten this wrong in the past,
2313	 * and this can lead to user data corruption! */
2314	BUG_ON(((unsigned long) ptr) & (size-1));
2315
2316	return ptr;
2317}
2318
2319void jbd2_free(void *ptr, size_t size)
2320{
2321	if (size < PAGE_SIZE)
2322		kmem_cache_free(get_slab(size), ptr);
2323	else
2324		free_pages((unsigned long)ptr, get_order(size));
 
 
 
 
 
 
 
 
 
 
2325};
2326
2327/*
2328 * Journal_head storage management
2329 */
2330static struct kmem_cache *jbd2_journal_head_cache;
2331#ifdef CONFIG_JBD2_DEBUG
2332static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2333#endif
2334
2335static int jbd2_journal_init_journal_head_cache(void)
2336{
2337	int retval;
2338
2339	J_ASSERT(jbd2_journal_head_cache == NULL);
2340	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2341				sizeof(struct journal_head),
2342				0,		/* offset */
2343				SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2344				NULL);		/* ctor */
2345	retval = 0;
2346	if (!jbd2_journal_head_cache) {
2347		retval = -ENOMEM;
2348		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2349	}
2350	return retval;
2351}
2352
2353static void jbd2_journal_destroy_journal_head_cache(void)
2354{
2355	if (jbd2_journal_head_cache) {
2356		kmem_cache_destroy(jbd2_journal_head_cache);
2357		jbd2_journal_head_cache = NULL;
2358	}
2359}
2360
2361/*
2362 * journal_head splicing and dicing
2363 */
2364static struct journal_head *journal_alloc_journal_head(void)
2365{
2366	struct journal_head *ret;
2367
2368#ifdef CONFIG_JBD2_DEBUG
2369	atomic_inc(&nr_journal_heads);
2370#endif
2371	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2372	if (!ret) {
2373		jbd_debug(1, "out of memory for journal_head\n");
2374		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2375		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2376				GFP_NOFS | __GFP_NOFAIL);
2377	}
2378	return ret;
2379}
2380
2381static void journal_free_journal_head(struct journal_head *jh)
2382{
2383#ifdef CONFIG_JBD2_DEBUG
2384	atomic_dec(&nr_journal_heads);
2385	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2386#endif
2387	kmem_cache_free(jbd2_journal_head_cache, jh);
2388}
2389
2390/*
2391 * A journal_head is attached to a buffer_head whenever JBD has an
2392 * interest in the buffer.
2393 *
2394 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2395 * is set.  This bit is tested in core kernel code where we need to take
2396 * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2397 * there.
2398 *
2399 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2400 *
2401 * When a buffer has its BH_JBD bit set it is immune from being released by
2402 * core kernel code, mainly via ->b_count.
2403 *
2404 * A journal_head is detached from its buffer_head when the journal_head's
2405 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2406 * transaction (b_cp_transaction) hold their references to b_jcount.
2407 *
2408 * Various places in the kernel want to attach a journal_head to a buffer_head
2409 * _before_ attaching the journal_head to a transaction.  To protect the
2410 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2411 * journal_head's b_jcount refcount by one.  The caller must call
2412 * jbd2_journal_put_journal_head() to undo this.
2413 *
2414 * So the typical usage would be:
2415 *
2416 *	(Attach a journal_head if needed.  Increments b_jcount)
2417 *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2418 *	...
2419 *      (Get another reference for transaction)
2420 *	jbd2_journal_grab_journal_head(bh);
2421 *	jh->b_transaction = xxx;
2422 *	(Put original reference)
2423 *	jbd2_journal_put_journal_head(jh);
2424 */
2425
2426/*
2427 * Give a buffer_head a journal_head.
2428 *
2429 * May sleep.
2430 */
2431struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2432{
2433	struct journal_head *jh;
2434	struct journal_head *new_jh = NULL;
2435
2436repeat:
2437	if (!buffer_jbd(bh))
2438		new_jh = journal_alloc_journal_head();
2439
2440	jbd_lock_bh_journal_head(bh);
2441	if (buffer_jbd(bh)) {
2442		jh = bh2jh(bh);
2443	} else {
2444		J_ASSERT_BH(bh,
2445			(atomic_read(&bh->b_count) > 0) ||
2446			(bh->b_page && bh->b_page->mapping));
2447
2448		if (!new_jh) {
2449			jbd_unlock_bh_journal_head(bh);
2450			goto repeat;
2451		}
2452
2453		jh = new_jh;
2454		new_jh = NULL;		/* We consumed it */
2455		set_buffer_jbd(bh);
2456		bh->b_private = jh;
2457		jh->b_bh = bh;
2458		get_bh(bh);
2459		BUFFER_TRACE(bh, "added journal_head");
2460	}
2461	jh->b_jcount++;
2462	jbd_unlock_bh_journal_head(bh);
2463	if (new_jh)
2464		journal_free_journal_head(new_jh);
2465	return bh->b_private;
2466}
2467
2468/*
2469 * Grab a ref against this buffer_head's journal_head.  If it ended up not
2470 * having a journal_head, return NULL
2471 */
2472struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2473{
2474	struct journal_head *jh = NULL;
2475
2476	jbd_lock_bh_journal_head(bh);
2477	if (buffer_jbd(bh)) {
2478		jh = bh2jh(bh);
2479		jh->b_jcount++;
2480	}
2481	jbd_unlock_bh_journal_head(bh);
2482	return jh;
2483}
2484
2485static void __journal_remove_journal_head(struct buffer_head *bh)
2486{
2487	struct journal_head *jh = bh2jh(bh);
2488
2489	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2490	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2491	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2492	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2493	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2494	J_ASSERT_BH(bh, buffer_jbd(bh));
2495	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2496	BUFFER_TRACE(bh, "remove journal_head");
2497	if (jh->b_frozen_data) {
2498		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2499		jbd2_free(jh->b_frozen_data, bh->b_size);
2500	}
2501	if (jh->b_committed_data) {
2502		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2503		jbd2_free(jh->b_committed_data, bh->b_size);
2504	}
2505	bh->b_private = NULL;
2506	jh->b_bh = NULL;	/* debug, really */
2507	clear_buffer_jbd(bh);
2508	journal_free_journal_head(jh);
2509}
2510
2511/*
2512 * Drop a reference on the passed journal_head.  If it fell to zero then
2513 * release the journal_head from the buffer_head.
2514 */
2515void jbd2_journal_put_journal_head(struct journal_head *jh)
2516{
2517	struct buffer_head *bh = jh2bh(jh);
2518
2519	jbd_lock_bh_journal_head(bh);
2520	J_ASSERT_JH(jh, jh->b_jcount > 0);
2521	--jh->b_jcount;
2522	if (!jh->b_jcount) {
2523		__journal_remove_journal_head(bh);
2524		jbd_unlock_bh_journal_head(bh);
2525		__brelse(bh);
2526	} else
2527		jbd_unlock_bh_journal_head(bh);
2528}
2529
2530/*
2531 * Initialize jbd inode head
2532 */
2533void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2534{
2535	jinode->i_transaction = NULL;
2536	jinode->i_next_transaction = NULL;
2537	jinode->i_vfs_inode = inode;
2538	jinode->i_flags = 0;
2539	INIT_LIST_HEAD(&jinode->i_list);
2540}
2541
2542/*
2543 * Function to be called before we start removing inode from memory (i.e.,
2544 * clear_inode() is a fine place to be called from). It removes inode from
2545 * transaction's lists.
2546 */
2547void jbd2_journal_release_jbd_inode(journal_t *journal,
2548				    struct jbd2_inode *jinode)
2549{
2550	if (!journal)
2551		return;
2552restart:
2553	spin_lock(&journal->j_list_lock);
2554	/* Is commit writing out inode - we have to wait */
2555	if (jinode->i_flags & JI_COMMIT_RUNNING) {
2556		wait_queue_head_t *wq;
2557		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2558		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2559		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2560		spin_unlock(&journal->j_list_lock);
2561		schedule();
2562		finish_wait(wq, &wait.wait);
2563		goto restart;
2564	}
2565
2566	if (jinode->i_transaction) {
2567		list_del(&jinode->i_list);
2568		jinode->i_transaction = NULL;
2569	}
2570	spin_unlock(&journal->j_list_lock);
2571}
2572
2573
2574#ifdef CONFIG_PROC_FS
2575
2576#define JBD2_STATS_PROC_NAME "fs/jbd2"
2577
2578static void __init jbd2_create_jbd_stats_proc_entry(void)
2579{
2580	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2581}
2582
2583static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2584{
2585	if (proc_jbd2_stats)
2586		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2587}
2588
2589#else
2590
2591#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2592#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2593
2594#endif
2595
2596struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2597
2598static int __init jbd2_journal_init_handle_cache(void)
2599{
2600	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2601	if (jbd2_handle_cache == NULL) {
2602		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2603		return -ENOMEM;
2604	}
2605	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2606	if (jbd2_inode_cache == NULL) {
2607		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2608		kmem_cache_destroy(jbd2_handle_cache);
2609		return -ENOMEM;
2610	}
2611	return 0;
2612}
2613
2614static void jbd2_journal_destroy_handle_cache(void)
2615{
2616	if (jbd2_handle_cache)
2617		kmem_cache_destroy(jbd2_handle_cache);
2618	if (jbd2_inode_cache)
2619		kmem_cache_destroy(jbd2_inode_cache);
2620
2621}
2622
2623/*
2624 * Module startup and shutdown
2625 */
2626
2627static int __init journal_init_caches(void)
2628{
2629	int ret;
2630
2631	ret = jbd2_journal_init_revoke_caches();
2632	if (ret == 0)
2633		ret = jbd2_journal_init_journal_head_cache();
2634	if (ret == 0)
2635		ret = jbd2_journal_init_handle_cache();
2636	if (ret == 0)
2637		ret = jbd2_journal_init_transaction_cache();
2638	return ret;
2639}
2640
2641static void jbd2_journal_destroy_caches(void)
2642{
2643	jbd2_journal_destroy_revoke_caches();
2644	jbd2_journal_destroy_journal_head_cache();
2645	jbd2_journal_destroy_handle_cache();
2646	jbd2_journal_destroy_transaction_cache();
2647	jbd2_journal_destroy_slabs();
2648}
2649
2650static int __init journal_init(void)
2651{
2652	int ret;
2653
2654	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2655
2656	ret = journal_init_caches();
2657	if (ret == 0) {
2658		jbd2_create_jbd_stats_proc_entry();
2659	} else {
2660		jbd2_journal_destroy_caches();
2661	}
2662	return ret;
2663}
2664
2665static void __exit journal_exit(void)
2666{
2667#ifdef CONFIG_JBD2_DEBUG
2668	int n = atomic_read(&nr_journal_heads);
2669	if (n)
2670		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2671#endif
2672	jbd2_remove_jbd_stats_proc_entry();
2673	jbd2_journal_destroy_caches();
2674}
2675
2676MODULE_LICENSE("GPL");
2677module_init(journal_init);
2678module_exit(journal_exit);
2679