1// SPDX-License-Identifier: GPL-2.0
   2
   3/*
   4 * fs/ext4/fast_commit.c
   5 *
   6 * Written by Harshad Shirwadkar <harshadshirwadkar@gmail.com>
   7 *
   8 * Ext4 fast commits routines.
   9 */
  10#include "ext4.h"
  11#include "ext4_jbd2.h"
  12#include "ext4_extents.h"
  13#include "mballoc.h"
  14
  15/*
  16 * Ext4 Fast Commits
  17 * -----------------
  18 *
  19 * Ext4 fast commits implement fine grained journalling for Ext4.
  20 *
  21 * Fast commits are organized as a log of tag-length-value (TLV) structs. (See
  22 * struct ext4_fc_tl). Each TLV contains some delta that is replayed TLV by
  23 * TLV during the recovery phase. For the scenarios for which we currently
  24 * don't have replay code, fast commit falls back to full commits.
  25 * Fast commits record delta in one of the following three categories.
  26 *
  27 * (A) Directory entry updates:
  28 *
  29 * - EXT4_FC_TAG_UNLINK		- records directory entry unlink
  30 * - EXT4_FC_TAG_LINK		- records directory entry link
  31 * - EXT4_FC_TAG_CREAT		- records inode and directory entry creation
  32 *
  33 * (B) File specific data range updates:
  34 *
  35 * - EXT4_FC_TAG_ADD_RANGE	- records addition of new blocks to an inode
  36 * - EXT4_FC_TAG_DEL_RANGE	- records deletion of blocks from an inode
  37 *
  38 * (C) Inode metadata (mtime / ctime etc):
  39 *
  40 * - EXT4_FC_TAG_INODE		- record the inode that should be replayed
  41 *				  during recovery. Note that iblocks field is
  42 *				  not replayed and instead derived during
  43 *				  replay.
  44 * Commit Operation
  45 * ----------------
  46 * With fast commits, we maintain all the directory entry operations in the
  47 * order in which they are issued in an in-memory queue. This queue is flushed
  48 * to disk during the commit operation. We also maintain a list of inodes
  49 * that need to be committed during a fast commit in another in memory queue of
  50 * inodes. During the commit operation, we commit in the following order:
  51 *
  52 * [1] Lock inodes for any further data updates by setting COMMITTING state
  53 * [2] Submit data buffers of all the inodes
  54 * [3] Wait for [2] to complete
  55 * [4] Commit all the directory entry updates in the fast commit space
  56 * [5] Commit all the changed inode structures
  57 * [6] Write tail tag (this tag ensures the atomicity, please read the following
  58 *     section for more details).
  59 * [7] Wait for [4], [5] and [6] to complete.
  60 *
  61 * All the inode updates must call ext4_fc_start_update() before starting an
  62 * update. If such an ongoing update is present, fast commit waits for it to
  63 * complete. The completion of such an update is marked by
  64 * ext4_fc_stop_update().
  65 *
  66 * Fast Commit Ineligibility
  67 * -------------------------
  68 *
  69 * Not all operations are supported by fast commits today (e.g extended
  70 * attributes). Fast commit ineligibility is marked by calling
  71 * ext4_fc_mark_ineligible(): This makes next fast commit operation to fall back
  72 * to full commit.
  73 *
  74 * Atomicity of commits
  75 * --------------------
  76 * In order to guarantee atomicity during the commit operation, fast commit
  77 * uses "EXT4_FC_TAG_TAIL" tag that marks a fast commit as complete. Tail
  78 * tag contains CRC of the contents and TID of the transaction after which
  79 * this fast commit should be applied. Recovery code replays fast commit
  80 * logs only if there's at least 1 valid tail present. For every fast commit
  81 * operation, there is 1 tail. This means, we may end up with multiple tails
  82 * in the fast commit space. Here's an example:
  83 *
  84 * - Create a new file A and remove existing file B
  85 * - fsync()
  86 * - Append contents to file A
  87 * - Truncate file A
  88 * - fsync()
  89 *
  90 * The fast commit space at the end of above operations would look like this:
  91 *      [HEAD] [CREAT A] [UNLINK B] [TAIL] [ADD_RANGE A] [DEL_RANGE A] [TAIL]
  92 *             |<---  Fast Commit 1   --->|<---      Fast Commit 2     ---->|
  93 *
  94 * Replay code should thus check for all the valid tails in the FC area.
  95 *
  96 * Fast Commit Replay Idempotence
  97 * ------------------------------
  98 *
  99 * Fast commits tags are idempotent in nature provided the recovery code follows
 100 * certain rules. The guiding principle that the commit path follows while
 101 * committing is that it stores the result of a particular operation instead of
 102 * storing the procedure.
 103 *
 104 * Let's consider this rename operation: 'mv /a /b'. Let's assume dirent '/a'
 105 * was associated with inode 10. During fast commit, instead of storing this
 106 * operation as a procedure "rename a to b", we store the resulting file system
 107 * state as a "series" of outcomes:
 108 *
 109 * - Link dirent b to inode 10
 110 * - Unlink dirent a
 111 * - Inode <10> with valid refcount
 112 *
 113 * Now when recovery code runs, it needs "enforce" this state on the file
 114 * system. This is what guarantees idempotence of fast commit replay.
 115 *
 116 * Let's take an example of a procedure that is not idempotent and see how fast
 117 * commits make it idempotent. Consider following sequence of operations:
 118 *
 119 *     rm A;    mv B A;    read A
 120 *  (x)     (y)        (z)
 121 *
 122 * (x), (y) and (z) are the points at which we can crash. If we store this
 123 * sequence of operations as is then the replay is not idempotent. Let's say
 124 * while in replay, we crash at (z). During the second replay, file A (which was
 125 * actually created as a result of "mv B A" operation) would get deleted. Thus,
 126 * file named A would be absent when we try to read A. So, this sequence of
 127 * operations is not idempotent. However, as mentioned above, instead of storing
 128 * the procedure fast commits store the outcome of each procedure. Thus the fast
 129 * commit log for above procedure would be as follows:
 130 *
 131 * (Let's assume dirent A was linked to inode 10 and dirent B was linked to
 132 * inode 11 before the replay)
 133 *
 134 *    [Unlink A]   [Link A to inode 11]   [Unlink B]   [Inode 11]
 135 * (w)          (x)                    (y)          (z)
 136 *
 137 * If we crash at (z), we will have file A linked to inode 11. During the second
 138 * replay, we will remove file A (inode 11). But we will create it back and make
 139 * it point to inode 11. We won't find B, so we'll just skip that step. At this
 140 * point, the refcount for inode 11 is not reliable, but that gets fixed by the
 141 * replay of last inode 11 tag. Crashes at points (w), (x) and (y) get handled
 142 * similarly. Thus, by converting a non-idempotent procedure into a series of
 143 * idempotent outcomes, fast commits ensured idempotence during the replay.
 144 *
 145 * TODOs
 146 * -----
 147 *
 148 * 0) Fast commit replay path hardening: Fast commit replay code should use
 149 *    journal handles to make sure all the updates it does during the replay
 150 *    path are atomic. With that if we crash during fast commit replay, after
 151 *    trying to do recovery again, we will find a file system where fast commit
 152 *    area is invalid (because new full commit would be found). In order to deal
 153 *    with that, fast commit replay code should ensure that the "FC_REPLAY"
 154 *    superblock state is persisted before starting the replay, so that after
 155 *    the crash, fast commit recovery code can look at that flag and perform
 156 *    fast commit recovery even if that area is invalidated by later full
 157 *    commits.
 158 *
 159 * 1) Fast commit's commit path locks the entire file system during fast
 160 *    commit. This has significant performance penalty. Instead of that, we
 161 *    should use ext4_fc_start/stop_update functions to start inode level
 162 *    updates from ext4_journal_start/stop. Once we do that we can drop file
 163 *    system locking during commit path.
 164 *
 165 * 2) Handle more ineligible cases.
 166 */
 167
 168#include <trace/events/ext4.h>
 169static struct kmem_cache *ext4_fc_dentry_cachep;
 170
 171static void ext4_end_buffer_io_sync(struct buffer_head *bh, int uptodate)
 172{
 173	BUFFER_TRACE(bh, "");
 174	if (uptodate) {
 175		ext4_debug("%s: Block %lld up-to-date",
 176			   __func__, bh->b_blocknr);
 177		set_buffer_uptodate(bh);
 178	} else {
 179		ext4_debug("%s: Block %lld not up-to-date",
 180			   __func__, bh->b_blocknr);
 181		clear_buffer_uptodate(bh);
 182	}
 183
 184	unlock_buffer(bh);
 185}
 186
 187static inline void ext4_fc_reset_inode(struct inode *inode)
 188{
 189	struct ext4_inode_info *ei = EXT4_I(inode);
 190
 191	ei->i_fc_lblk_start = 0;
 192	ei->i_fc_lblk_len = 0;
 193}
 194
 195void ext4_fc_init_inode(struct inode *inode)
 196{
 197	struct ext4_inode_info *ei = EXT4_I(inode);
 198
 199	ext4_fc_reset_inode(inode);
 200	ext4_clear_inode_state(inode, EXT4_STATE_FC_COMMITTING);
 201	INIT_LIST_HEAD(&ei->i_fc_list);
 202	INIT_LIST_HEAD(&ei->i_fc_dilist);
 203	init_waitqueue_head(&ei->i_fc_wait);
 204	atomic_set(&ei->i_fc_updates, 0);
 205}
 206
 207/* This function must be called with sbi->s_fc_lock held. */
 208static void ext4_fc_wait_committing_inode(struct inode *inode)
 209__releases(&EXT4_SB(inode->i_sb)->s_fc_lock)
 210{
 211	wait_queue_head_t *wq;
 212	struct ext4_inode_info *ei = EXT4_I(inode);
 213
 214#if (BITS_PER_LONG < 64)
 215	DEFINE_WAIT_BIT(wait, &ei->i_state_flags,
 216			EXT4_STATE_FC_COMMITTING);
 217	wq = bit_waitqueue(&ei->i_state_flags,
 218				EXT4_STATE_FC_COMMITTING);
 219#else
 220	DEFINE_WAIT_BIT(wait, &ei->i_flags,
 221			EXT4_STATE_FC_COMMITTING);
 222	wq = bit_waitqueue(&ei->i_flags,
 223				EXT4_STATE_FC_COMMITTING);
 224#endif
 225	lockdep_assert_held(&EXT4_SB(inode->i_sb)->s_fc_lock);
 226	prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
 227	spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
 228	schedule();
 229	finish_wait(wq, &wait.wq_entry);
 230}
 231
 232static bool ext4_fc_disabled(struct super_block *sb)
 233{
 234	return (!test_opt2(sb, JOURNAL_FAST_COMMIT) ||
 235		(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY));
 236}
 237
 238/*
 239 * Inform Ext4's fast about start of an inode update
 240 *
 241 * This function is called by the high level call VFS callbacks before
 242 * performing any inode update. This function blocks if there's an ongoing
 243 * fast commit on the inode in question.
 244 */
 245void ext4_fc_start_update(struct inode *inode)
 246{
 247	struct ext4_inode_info *ei = EXT4_I(inode);
 248
 249	if (ext4_fc_disabled(inode->i_sb))
 250		return;
 251
 252restart:
 253	spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
 254	if (list_empty(&ei->i_fc_list))
 255		goto out;
 256
 257	if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
 258		ext4_fc_wait_committing_inode(inode);
 259		goto restart;
 260	}
 261out:
 262	atomic_inc(&ei->i_fc_updates);
 263	spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
 264}
 265
 266/*
 267 * Stop inode update and wake up waiting fast commits if any.
 268 */
 269void ext4_fc_stop_update(struct inode *inode)
 270{
 271	struct ext4_inode_info *ei = EXT4_I(inode);
 272
 273	if (ext4_fc_disabled(inode->i_sb))
 274		return;
 275
 276	if (atomic_dec_and_test(&ei->i_fc_updates))
 277		wake_up_all(&ei->i_fc_wait);
 278}
 279
 280/*
 281 * Remove inode from fast commit list. If the inode is being committed
 282 * we wait until inode commit is done.
 283 */
 284void ext4_fc_del(struct inode *inode)
 285{
 286	struct ext4_inode_info *ei = EXT4_I(inode);
 287	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 288	struct ext4_fc_dentry_update *fc_dentry;
 289
 290	if (ext4_fc_disabled(inode->i_sb))
 291		return;
 292
 293restart:
 294	spin_lock(&EXT4_SB(inode->i_sb)->s_fc_lock);
 295	if (list_empty(&ei->i_fc_list) && list_empty(&ei->i_fc_dilist)) {
 296		spin_unlock(&EXT4_SB(inode->i_sb)->s_fc_lock);
 297		return;
 298	}
 299
 300	if (ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING)) {
 301		ext4_fc_wait_committing_inode(inode);
 302		goto restart;
 303	}
 304
 305	if (!list_empty(&ei->i_fc_list))
 306		list_del_init(&ei->i_fc_list);
 307
 308	/*
 309	 * Since this inode is getting removed, let's also remove all FC
 310	 * dentry create references, since it is not needed to log it anyways.
 311	 */
 312	if (list_empty(&ei->i_fc_dilist)) {
 313		spin_unlock(&sbi->s_fc_lock);
 314		return;
 315	}
 316
 317	fc_dentry = list_first_entry(&ei->i_fc_dilist, struct ext4_fc_dentry_update, fcd_dilist);
 318	WARN_ON(fc_dentry->fcd_op != EXT4_FC_TAG_CREAT);
 319	list_del_init(&fc_dentry->fcd_list);
 320	list_del_init(&fc_dentry->fcd_dilist);
 321
 322	WARN_ON(!list_empty(&ei->i_fc_dilist));
 323	spin_unlock(&sbi->s_fc_lock);
 324
 325	if (fc_dentry->fcd_name.name &&
 326		fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
 327		kfree(fc_dentry->fcd_name.name);
 328	kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
 329
 330	return;
 331}
 332
 333/*
 334 * Mark file system as fast commit ineligible, and record latest
 335 * ineligible transaction tid. This means until the recorded
 336 * transaction, commit operation would result in a full jbd2 commit.
 337 */
 338void ext4_fc_mark_ineligible(struct super_block *sb, int reason, handle_t *handle)
 339{
 340	struct ext4_sb_info *sbi = EXT4_SB(sb);
 341	tid_t tid;
 
 
 342
 343	if (ext4_fc_disabled(sb))
 344		return;
 345
 346	ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
 347	if (handle && !IS_ERR(handle))
 348		tid = handle->h_transaction->t_tid;
 349	else {
 350		read_lock(&sbi->s_journal->j_state_lock);
 351		tid = sbi->s_journal->j_running_transaction ?
 352				sbi->s_journal->j_running_transaction->t_tid : 0;
 
 
 353		read_unlock(&sbi->s_journal->j_state_lock);
 354	}
 355	spin_lock(&sbi->s_fc_lock);
 356	if (sbi->s_fc_ineligible_tid < tid)
 
 357		sbi->s_fc_ineligible_tid = tid;
 
 358	spin_unlock(&sbi->s_fc_lock);
 359	WARN_ON(reason >= EXT4_FC_REASON_MAX);
 360	sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;
 361}
 362
 363/*
 364 * Generic fast commit tracking function. If this is the first time this we are
 365 * called after a full commit, we initialize fast commit fields and then call
 366 * __fc_track_fn() with update = 0. If we have already been called after a full
 367 * commit, we pass update = 1. Based on that, the track function can determine
 368 * if it needs to track a field for the first time or if it needs to just
 369 * update the previously tracked value.
 370 *
 371 * If enqueue is set, this function enqueues the inode in fast commit list.
 372 */
 373static int ext4_fc_track_template(
 374	handle_t *handle, struct inode *inode,
 375	int (*__fc_track_fn)(struct inode *, void *, bool),
 376	void *args, int enqueue)
 377{
 378	bool update = false;
 379	struct ext4_inode_info *ei = EXT4_I(inode);
 380	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 381	tid_t tid = 0;
 382	int ret;
 383
 384	tid = handle->h_transaction->t_tid;
 385	mutex_lock(&ei->i_fc_lock);
 386	if (tid == ei->i_sync_tid) {
 387		update = true;
 388	} else {
 389		ext4_fc_reset_inode(inode);
 390		ei->i_sync_tid = tid;
 391	}
 392	ret = __fc_track_fn(inode, args, update);
 393	mutex_unlock(&ei->i_fc_lock);
 394
 395	if (!enqueue)
 396		return ret;
 397
 398	spin_lock(&sbi->s_fc_lock);
 399	if (list_empty(&EXT4_I(inode)->i_fc_list))
 400		list_add_tail(&EXT4_I(inode)->i_fc_list,
 401				(sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
 402				 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?
 403				&sbi->s_fc_q[FC_Q_STAGING] :
 404				&sbi->s_fc_q[FC_Q_MAIN]);
 405	spin_unlock(&sbi->s_fc_lock);
 406
 407	return ret;
 408}
 409
 410struct __track_dentry_update_args {
 411	struct dentry *dentry;
 412	int op;
 413};
 414
 415/* __track_fn for directory entry updates. Called with ei->i_fc_lock. */
 416static int __track_dentry_update(struct inode *inode, void *arg, bool update)
 
 417{
 418	struct ext4_fc_dentry_update *node;
 419	struct ext4_inode_info *ei = EXT4_I(inode);
 420	struct __track_dentry_update_args *dentry_update =
 421		(struct __track_dentry_update_args *)arg;
 422	struct dentry *dentry = dentry_update->dentry;
 423	struct inode *dir = dentry->d_parent->d_inode;
 424	struct super_block *sb = inode->i_sb;
 425	struct ext4_sb_info *sbi = EXT4_SB(sb);
 426
 427	mutex_unlock(&ei->i_fc_lock);
 428
 429	if (IS_ENCRYPTED(dir)) {
 430		ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_ENCRYPTED_FILENAME,
 431					NULL);
 432		mutex_lock(&ei->i_fc_lock);
 433		return -EOPNOTSUPP;
 434	}
 435
 436	node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);
 437	if (!node) {
 438		ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
 439		mutex_lock(&ei->i_fc_lock);
 440		return -ENOMEM;
 441	}
 442
 443	node->fcd_op = dentry_update->op;
 444	node->fcd_parent = dir->i_ino;
 445	node->fcd_ino = inode->i_ino;
 446	if (dentry->d_name.len > DNAME_INLINE_LEN) {
 447		node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);
 448		if (!node->fcd_name.name) {
 449			kmem_cache_free(ext4_fc_dentry_cachep, node);
 450			ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, NULL);
 451			mutex_lock(&ei->i_fc_lock);
 452			return -ENOMEM;
 453		}
 454		memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,
 455			dentry->d_name.len);
 456	} else {
 457		memcpy(node->fcd_iname, dentry->d_name.name,
 458			dentry->d_name.len);
 459		node->fcd_name.name = node->fcd_iname;
 460	}
 461	node->fcd_name.len = dentry->d_name.len;
 462	INIT_LIST_HEAD(&node->fcd_dilist);
 463	spin_lock(&sbi->s_fc_lock);
 464	if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
 465		sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)
 466		list_add_tail(&node->fcd_list,
 467				&sbi->s_fc_dentry_q[FC_Q_STAGING]);
 468	else
 469		list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);
 470
 471	/*
 472	 * This helps us keep a track of all fc_dentry updates which is part of
 473	 * this ext4 inode. So in case the inode is getting unlinked, before
 474	 * even we get a chance to fsync, we could remove all fc_dentry
 475	 * references while evicting the inode in ext4_fc_del().
 476	 * Also with this, we don't need to loop over all the inodes in
 477	 * sbi->s_fc_q to get the corresponding inode in
 478	 * ext4_fc_commit_dentry_updates().
 479	 */
 480	if (dentry_update->op == EXT4_FC_TAG_CREAT) {
 481		WARN_ON(!list_empty(&ei->i_fc_dilist));
 482		list_add_tail(&node->fcd_dilist, &ei->i_fc_dilist);
 483	}
 484	spin_unlock(&sbi->s_fc_lock);
 485	mutex_lock(&ei->i_fc_lock);
 486
 487	return 0;
 488}
 489
 490void __ext4_fc_track_unlink(handle_t *handle,
 491		struct inode *inode, struct dentry *dentry)
 492{
 493	struct __track_dentry_update_args args;
 494	int ret;
 495
 496	args.dentry = dentry;
 497	args.op = EXT4_FC_TAG_UNLINK;
 498
 499	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
 500					(void *)&args, 0);
 501	trace_ext4_fc_track_unlink(handle, inode, dentry, ret);
 502}
 503
 504void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)
 505{
 506	struct inode *inode = d_inode(dentry);
 507
 508	if (ext4_fc_disabled(inode->i_sb))
 509		return;
 510
 511	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
 512		return;
 513
 514	__ext4_fc_track_unlink(handle, inode, dentry);
 515}
 516
 517void __ext4_fc_track_link(handle_t *handle,
 518	struct inode *inode, struct dentry *dentry)
 519{
 520	struct __track_dentry_update_args args;
 521	int ret;
 522
 523	args.dentry = dentry;
 524	args.op = EXT4_FC_TAG_LINK;
 525
 526	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
 527					(void *)&args, 0);
 528	trace_ext4_fc_track_link(handle, inode, dentry, ret);
 529}
 530
 531void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)
 532{
 533	struct inode *inode = d_inode(dentry);
 534
 535	if (ext4_fc_disabled(inode->i_sb))
 536		return;
 537
 538	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
 539		return;
 540
 541	__ext4_fc_track_link(handle, inode, dentry);
 542}
 543
 544void __ext4_fc_track_create(handle_t *handle, struct inode *inode,
 545			  struct dentry *dentry)
 546{
 547	struct __track_dentry_update_args args;
 548	int ret;
 549
 550	args.dentry = dentry;
 551	args.op = EXT4_FC_TAG_CREAT;
 552
 553	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,
 554					(void *)&args, 0);
 555	trace_ext4_fc_track_create(handle, inode, dentry, ret);
 556}
 557
 558void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)
 559{
 560	struct inode *inode = d_inode(dentry);
 561
 562	if (ext4_fc_disabled(inode->i_sb))
 563		return;
 564
 565	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
 566		return;
 567
 568	__ext4_fc_track_create(handle, inode, dentry);
 569}
 570
 571/* __track_fn for inode tracking */
 572static int __track_inode(struct inode *inode, void *arg, bool update)
 
 573{
 574	if (update)
 575		return -EEXIST;
 576
 577	EXT4_I(inode)->i_fc_lblk_len = 0;
 578
 579	return 0;
 580}
 581
 582void ext4_fc_track_inode(handle_t *handle, struct inode *inode)
 583{
 584	int ret;
 585
 586	if (S_ISDIR(inode->i_mode))
 587		return;
 588
 589	if (ext4_fc_disabled(inode->i_sb))
 590		return;
 591
 592	if (ext4_should_journal_data(inode)) {
 593		ext4_fc_mark_ineligible(inode->i_sb,
 594					EXT4_FC_REASON_INODE_JOURNAL_DATA, handle);
 595		return;
 596	}
 597
 598	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
 599		return;
 600
 601	ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);
 602	trace_ext4_fc_track_inode(handle, inode, ret);
 603}
 604
 605struct __track_range_args {
 606	ext4_lblk_t start, end;
 607};
 608
 609/* __track_fn for tracking data updates */
 610static int __track_range(struct inode *inode, void *arg, bool update)
 
 611{
 612	struct ext4_inode_info *ei = EXT4_I(inode);
 613	ext4_lblk_t oldstart;
 614	struct __track_range_args *__arg =
 615		(struct __track_range_args *)arg;
 616
 617	if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {
 618		ext4_debug("Special inode %ld being modified\n", inode->i_ino);
 619		return -ECANCELED;
 620	}
 621
 622	oldstart = ei->i_fc_lblk_start;
 623
 624	if (update && ei->i_fc_lblk_len > 0) {
 625		ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);
 626		ei->i_fc_lblk_len =
 627			max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -
 628				ei->i_fc_lblk_start + 1;
 629	} else {
 630		ei->i_fc_lblk_start = __arg->start;
 631		ei->i_fc_lblk_len = __arg->end - __arg->start + 1;
 632	}
 633
 634	return 0;
 635}
 636
 637void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,
 638			 ext4_lblk_t end)
 639{
 640	struct __track_range_args args;
 641	int ret;
 642
 643	if (S_ISDIR(inode->i_mode))
 644		return;
 645
 646	if (ext4_fc_disabled(inode->i_sb))
 647		return;
 648
 649	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))
 650		return;
 651
 
 
 
 
 
 
 652	args.start = start;
 653	args.end = end;
 654
 655	ret = ext4_fc_track_template(handle, inode,  __track_range, &args, 1);
 656
 657	trace_ext4_fc_track_range(handle, inode, start, end, ret);
 658}
 659
 660static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)
 661{
 662	blk_opf_t write_flags = REQ_SYNC;
 663	struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;
 664
 665	/* Add REQ_FUA | REQ_PREFLUSH only its tail */
 666	if (test_opt(sb, BARRIER) && is_tail)
 667		write_flags |= REQ_FUA | REQ_PREFLUSH;
 668	lock_buffer(bh);
 669	set_buffer_dirty(bh);
 670	set_buffer_uptodate(bh);
 671	bh->b_end_io = ext4_end_buffer_io_sync;
 672	submit_bh(REQ_OP_WRITE | write_flags, bh);
 673	EXT4_SB(sb)->s_fc_bh = NULL;
 674}
 675
 676/* Ext4 commit path routines */
 677
 678/*
 679 * Allocate len bytes on a fast commit buffer.
 680 *
 681 * During the commit time this function is used to manage fast commit
 682 * block space. We don't split a fast commit log onto different
 683 * blocks. So this function makes sure that if there's not enough space
 684 * on the current block, the remaining space in the current block is
 685 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,
 686 * new block is from jbd2 and CRC is updated to reflect the padding
 687 * we added.
 688 */
 689static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)
 690{
 691	struct ext4_fc_tl tl;
 692	struct ext4_sb_info *sbi = EXT4_SB(sb);
 693	struct buffer_head *bh;
 694	int bsize = sbi->s_journal->j_blocksize;
 695	int ret, off = sbi->s_fc_bytes % bsize;
 696	int remaining;
 697	u8 *dst;
 698
 699	/*
 700	 * If 'len' is too long to fit in any block alongside a PAD tlv, then we
 701	 * cannot fulfill the request.
 702	 */
 703	if (len > bsize - EXT4_FC_TAG_BASE_LEN)
 704		return NULL;
 705
 706	if (!sbi->s_fc_bh) {
 707		ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
 708		if (ret)
 709			return NULL;
 710		sbi->s_fc_bh = bh;
 711	}
 712	dst = sbi->s_fc_bh->b_data + off;
 713
 714	/*
 715	 * Allocate the bytes in the current block if we can do so while still
 716	 * leaving enough space for a PAD tlv.
 717	 */
 718	remaining = bsize - EXT4_FC_TAG_BASE_LEN - off;
 719	if (len <= remaining) {
 720		sbi->s_fc_bytes += len;
 721		return dst;
 722	}
 723
 724	/*
 725	 * Else, terminate the current block with a PAD tlv, then allocate a new
 726	 * block and allocate the bytes at the start of that new block.
 727	 */
 728
 729	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);
 730	tl.fc_len = cpu_to_le16(remaining);
 731	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
 732	memset(dst + EXT4_FC_TAG_BASE_LEN, 0, remaining);
 733	*crc = ext4_chksum(sbi, *crc, sbi->s_fc_bh->b_data, bsize);
 734
 735	ext4_fc_submit_bh(sb, false);
 736
 737	ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);
 738	if (ret)
 739		return NULL;
 740	sbi->s_fc_bh = bh;
 741	sbi->s_fc_bytes += bsize - off + len;
 742	return sbi->s_fc_bh->b_data;
 743}
 744
 745/*
 746 * Complete a fast commit by writing tail tag.
 747 *
 748 * Writing tail tag marks the end of a fast commit. In order to guarantee
 749 * atomicity, after writing tail tag, even if there's space remaining
 750 * in the block, next commit shouldn't use it. That's why tail tag
 751 * has the length as that of the remaining space on the block.
 752 */
 753static int ext4_fc_write_tail(struct super_block *sb, u32 crc)
 754{
 755	struct ext4_sb_info *sbi = EXT4_SB(sb);
 756	struct ext4_fc_tl tl;
 757	struct ext4_fc_tail tail;
 758	int off, bsize = sbi->s_journal->j_blocksize;
 759	u8 *dst;
 760
 761	/*
 762	 * ext4_fc_reserve_space takes care of allocating an extra block if
 763	 * there's no enough space on this block for accommodating this tail.
 764	 */
 765	dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + sizeof(tail), &crc);
 766	if (!dst)
 767		return -ENOSPC;
 768
 769	off = sbi->s_fc_bytes % bsize;
 770
 771	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);
 772	tl.fc_len = cpu_to_le16(bsize - off + sizeof(struct ext4_fc_tail));
 773	sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);
 774
 775	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
 776	dst += EXT4_FC_TAG_BASE_LEN;
 777	tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);
 778	memcpy(dst, &tail.fc_tid, sizeof(tail.fc_tid));
 779	dst += sizeof(tail.fc_tid);
 780	crc = ext4_chksum(sbi, crc, sbi->s_fc_bh->b_data,
 781			  dst - (u8 *)sbi->s_fc_bh->b_data);
 782	tail.fc_crc = cpu_to_le32(crc);
 783	memcpy(dst, &tail.fc_crc, sizeof(tail.fc_crc));
 784	dst += sizeof(tail.fc_crc);
 785	memset(dst, 0, bsize - off); /* Don't leak uninitialized memory. */
 786
 787	ext4_fc_submit_bh(sb, true);
 788
 789	return 0;
 790}
 791
 792/*
 793 * Adds tag, length, value and updates CRC. Returns true if tlv was added.
 794 * Returns false if there's not enough space.
 795 */
 796static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,
 797			   u32 *crc)
 798{
 799	struct ext4_fc_tl tl;
 800	u8 *dst;
 801
 802	dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + len, crc);
 803	if (!dst)
 804		return false;
 805
 806	tl.fc_tag = cpu_to_le16(tag);
 807	tl.fc_len = cpu_to_le16(len);
 808
 809	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
 810	memcpy(dst + EXT4_FC_TAG_BASE_LEN, val, len);
 811
 812	return true;
 813}
 814
 815/* Same as above, but adds dentry tlv. */
 816static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u32 *crc,
 817				   struct ext4_fc_dentry_update *fc_dentry)
 818{
 819	struct ext4_fc_dentry_info fcd;
 820	struct ext4_fc_tl tl;
 821	int dlen = fc_dentry->fcd_name.len;
 822	u8 *dst = ext4_fc_reserve_space(sb,
 823			EXT4_FC_TAG_BASE_LEN + sizeof(fcd) + dlen, crc);
 824
 825	if (!dst)
 826		return false;
 827
 828	fcd.fc_parent_ino = cpu_to_le32(fc_dentry->fcd_parent);
 829	fcd.fc_ino = cpu_to_le32(fc_dentry->fcd_ino);
 830	tl.fc_tag = cpu_to_le16(fc_dentry->fcd_op);
 831	tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);
 832	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
 833	dst += EXT4_FC_TAG_BASE_LEN;
 834	memcpy(dst, &fcd, sizeof(fcd));
 835	dst += sizeof(fcd);
 836	memcpy(dst, fc_dentry->fcd_name.name, dlen);
 837
 838	return true;
 839}
 840
 841/*
 842 * Writes inode in the fast commit space under TLV with tag @tag.
 843 * Returns 0 on success, error on failure.
 844 */
 845static int ext4_fc_write_inode(struct inode *inode, u32 *crc)
 846{
 847	struct ext4_inode_info *ei = EXT4_I(inode);
 848	int inode_len = EXT4_GOOD_OLD_INODE_SIZE;
 849	int ret;
 850	struct ext4_iloc iloc;
 851	struct ext4_fc_inode fc_inode;
 852	struct ext4_fc_tl tl;
 853	u8 *dst;
 854
 855	ret = ext4_get_inode_loc(inode, &iloc);
 856	if (ret)
 857		return ret;
 858
 859	if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
 860		inode_len = EXT4_INODE_SIZE(inode->i_sb);
 861	else if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)
 862		inode_len += ei->i_extra_isize;
 863
 864	fc_inode.fc_ino = cpu_to_le32(inode->i_ino);
 865	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);
 866	tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));
 867
 868	ret = -ECANCELED;
 869	dst = ext4_fc_reserve_space(inode->i_sb,
 870		EXT4_FC_TAG_BASE_LEN + inode_len + sizeof(fc_inode.fc_ino), crc);
 871	if (!dst)
 872		goto err;
 873
 874	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);
 875	dst += EXT4_FC_TAG_BASE_LEN;
 876	memcpy(dst, &fc_inode, sizeof(fc_inode));
 877	dst += sizeof(fc_inode);
 878	memcpy(dst, (u8 *)ext4_raw_inode(&iloc), inode_len);
 879	ret = 0;
 880err:
 881	brelse(iloc.bh);
 882	return ret;
 883}
 884
 885/*
 886 * Writes updated data ranges for the inode in question. Updates CRC.
 887 * Returns 0 on success, error otherwise.
 888 */
 889static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)
 890{
 891	ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;
 892	struct ext4_inode_info *ei = EXT4_I(inode);
 893	struct ext4_map_blocks map;
 894	struct ext4_fc_add_range fc_ext;
 895	struct ext4_fc_del_range lrange;
 896	struct ext4_extent *ex;
 897	int ret;
 898
 899	mutex_lock(&ei->i_fc_lock);
 900	if (ei->i_fc_lblk_len == 0) {
 901		mutex_unlock(&ei->i_fc_lock);
 902		return 0;
 903	}
 904	old_blk_size = ei->i_fc_lblk_start;
 905	new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;
 906	ei->i_fc_lblk_len = 0;
 907	mutex_unlock(&ei->i_fc_lock);
 908
 909	cur_lblk_off = old_blk_size;
 910	ext4_debug("will try writing %d to %d for inode %ld\n",
 911		   cur_lblk_off, new_blk_size, inode->i_ino);
 912
 913	while (cur_lblk_off <= new_blk_size) {
 914		map.m_lblk = cur_lblk_off;
 915		map.m_len = new_blk_size - cur_lblk_off + 1;
 916		ret = ext4_map_blocks(NULL, inode, &map, 0);
 917		if (ret < 0)
 918			return -ECANCELED;
 919
 920		if (map.m_len == 0) {
 921			cur_lblk_off++;
 922			continue;
 923		}
 924
 925		if (ret == 0) {
 926			lrange.fc_ino = cpu_to_le32(inode->i_ino);
 927			lrange.fc_lblk = cpu_to_le32(map.m_lblk);
 928			lrange.fc_len = cpu_to_le32(map.m_len);
 929			if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,
 930					    sizeof(lrange), (u8 *)&lrange, crc))
 931				return -ENOSPC;
 932		} else {
 933			unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?
 934				EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;
 935
 936			/* Limit the number of blocks in one extent */
 937			map.m_len = min(max, map.m_len);
 938
 939			fc_ext.fc_ino = cpu_to_le32(inode->i_ino);
 940			ex = (struct ext4_extent *)&fc_ext.fc_ex;
 941			ex->ee_block = cpu_to_le32(map.m_lblk);
 942			ex->ee_len = cpu_to_le16(map.m_len);
 943			ext4_ext_store_pblock(ex, map.m_pblk);
 944			if (map.m_flags & EXT4_MAP_UNWRITTEN)
 945				ext4_ext_mark_unwritten(ex);
 946			else
 947				ext4_ext_mark_initialized(ex);
 948			if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,
 949					    sizeof(fc_ext), (u8 *)&fc_ext, crc))
 950				return -ENOSPC;
 951		}
 952
 953		cur_lblk_off += map.m_len;
 954	}
 955
 956	return 0;
 957}
 958
 959
 960/* Submit data for all the fast commit inodes */
 961static int ext4_fc_submit_inode_data_all(journal_t *journal)
 962{
 963	struct super_block *sb = journal->j_private;
 964	struct ext4_sb_info *sbi = EXT4_SB(sb);
 965	struct ext4_inode_info *ei;
 966	int ret = 0;
 967
 968	spin_lock(&sbi->s_fc_lock);
 969	list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
 970		ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
 971		while (atomic_read(&ei->i_fc_updates)) {
 972			DEFINE_WAIT(wait);
 973
 974			prepare_to_wait(&ei->i_fc_wait, &wait,
 975						TASK_UNINTERRUPTIBLE);
 976			if (atomic_read(&ei->i_fc_updates)) {
 977				spin_unlock(&sbi->s_fc_lock);
 978				schedule();
 979				spin_lock(&sbi->s_fc_lock);
 980			}
 981			finish_wait(&ei->i_fc_wait, &wait);
 982		}
 983		spin_unlock(&sbi->s_fc_lock);
 984		ret = jbd2_submit_inode_data(journal, ei->jinode);
 985		if (ret)
 986			return ret;
 987		spin_lock(&sbi->s_fc_lock);
 988	}
 989	spin_unlock(&sbi->s_fc_lock);
 990
 991	return ret;
 992}
 993
 994/* Wait for completion of data for all the fast commit inodes */
 995static int ext4_fc_wait_inode_data_all(journal_t *journal)
 996{
 997	struct super_block *sb = journal->j_private;
 998	struct ext4_sb_info *sbi = EXT4_SB(sb);
 999	struct ext4_inode_info *pos, *n;
1000	int ret = 0;
1001
1002	spin_lock(&sbi->s_fc_lock);
1003	list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
1004		if (!ext4_test_inode_state(&pos->vfs_inode,
1005					   EXT4_STATE_FC_COMMITTING))
1006			continue;
1007		spin_unlock(&sbi->s_fc_lock);
1008
1009		ret = jbd2_wait_inode_data(journal, pos->jinode);
1010		if (ret)
1011			return ret;
1012		spin_lock(&sbi->s_fc_lock);
1013	}
1014	spin_unlock(&sbi->s_fc_lock);
1015
1016	return 0;
1017}
1018
1019/* Commit all the directory entry updates */
1020static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)
1021__acquires(&sbi->s_fc_lock)
1022__releases(&sbi->s_fc_lock)
1023{
1024	struct super_block *sb = journal->j_private;
1025	struct ext4_sb_info *sbi = EXT4_SB(sb);
1026	struct ext4_fc_dentry_update *fc_dentry, *fc_dentry_n;
1027	struct inode *inode;
1028	struct ext4_inode_info *ei;
1029	int ret;
1030
1031	if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))
1032		return 0;
1033	list_for_each_entry_safe(fc_dentry, fc_dentry_n,
1034				 &sbi->s_fc_dentry_q[FC_Q_MAIN], fcd_list) {
1035		if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {
1036			spin_unlock(&sbi->s_fc_lock);
1037			if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
1038				ret = -ENOSPC;
1039				goto lock_and_exit;
1040			}
1041			spin_lock(&sbi->s_fc_lock);
1042			continue;
1043		}
1044		/*
1045		 * With fcd_dilist we need not loop in sbi->s_fc_q to get the
1046		 * corresponding inode pointer
1047		 */
1048		WARN_ON(list_empty(&fc_dentry->fcd_dilist));
1049		ei = list_first_entry(&fc_dentry->fcd_dilist,
1050				struct ext4_inode_info, i_fc_dilist);
1051		inode = &ei->vfs_inode;
1052		WARN_ON(inode->i_ino != fc_dentry->fcd_ino);
1053
1054		spin_unlock(&sbi->s_fc_lock);
1055
1056		/*
1057		 * We first write the inode and then the create dirent. This
1058		 * allows the recovery code to create an unnamed inode first
1059		 * and then link it to a directory entry. This allows us
1060		 * to use namei.c routines almost as is and simplifies
1061		 * the recovery code.
1062		 */
1063		ret = ext4_fc_write_inode(inode, crc);
1064		if (ret)
1065			goto lock_and_exit;
1066
1067		ret = ext4_fc_write_inode_data(inode, crc);
1068		if (ret)
1069			goto lock_and_exit;
1070
1071		if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {
1072			ret = -ENOSPC;
1073			goto lock_and_exit;
1074		}
1075
1076		spin_lock(&sbi->s_fc_lock);
1077	}
1078	return 0;
1079lock_and_exit:
1080	spin_lock(&sbi->s_fc_lock);
1081	return ret;
1082}
1083
1084static int ext4_fc_perform_commit(journal_t *journal)
1085{
1086	struct super_block *sb = journal->j_private;
1087	struct ext4_sb_info *sbi = EXT4_SB(sb);
1088	struct ext4_inode_info *iter;
1089	struct ext4_fc_head head;
1090	struct inode *inode;
1091	struct blk_plug plug;
1092	int ret = 0;
1093	u32 crc = 0;
1094
1095	ret = ext4_fc_submit_inode_data_all(journal);
1096	if (ret)
1097		return ret;
1098
1099	ret = ext4_fc_wait_inode_data_all(journal);
1100	if (ret)
1101		return ret;
1102
1103	/*
1104	 * If file system device is different from journal device, issue a cache
1105	 * flush before we start writing fast commit blocks.
1106	 */
1107	if (journal->j_fs_dev != journal->j_dev)
1108		blkdev_issue_flush(journal->j_fs_dev);
1109
1110	blk_start_plug(&plug);
1111	if (sbi->s_fc_bytes == 0) {
1112		/*
1113		 * Add a head tag only if this is the first fast commit
1114		 * in this TID.
1115		 */
1116		head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);
1117		head.fc_tid = cpu_to_le32(
1118			sbi->s_journal->j_running_transaction->t_tid);
1119		if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),
1120			(u8 *)&head, &crc)) {
1121			ret = -ENOSPC;
1122			goto out;
1123		}
1124	}
1125
1126	spin_lock(&sbi->s_fc_lock);
1127	ret = ext4_fc_commit_dentry_updates(journal, &crc);
1128	if (ret) {
1129		spin_unlock(&sbi->s_fc_lock);
1130		goto out;
1131	}
1132
1133	list_for_each_entry(iter, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
1134		inode = &iter->vfs_inode;
1135		if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))
1136			continue;
1137
1138		spin_unlock(&sbi->s_fc_lock);
1139		ret = ext4_fc_write_inode_data(inode, &crc);
1140		if (ret)
1141			goto out;
1142		ret = ext4_fc_write_inode(inode, &crc);
1143		if (ret)
1144			goto out;
1145		spin_lock(&sbi->s_fc_lock);
1146	}
1147	spin_unlock(&sbi->s_fc_lock);
1148
1149	ret = ext4_fc_write_tail(sb, crc);
1150
1151out:
1152	blk_finish_plug(&plug);
1153	return ret;
1154}
1155
1156static void ext4_fc_update_stats(struct super_block *sb, int status,
1157				 u64 commit_time, int nblks, tid_t commit_tid)
1158{
1159	struct ext4_fc_stats *stats = &EXT4_SB(sb)->s_fc_stats;
1160
1161	ext4_debug("Fast commit ended with status = %d for tid %u",
1162			status, commit_tid);
1163	if (status == EXT4_FC_STATUS_OK) {
1164		stats->fc_num_commits++;
1165		stats->fc_numblks += nblks;
1166		if (likely(stats->s_fc_avg_commit_time))
1167			stats->s_fc_avg_commit_time =
1168				(commit_time +
1169				 stats->s_fc_avg_commit_time * 3) / 4;
1170		else
1171			stats->s_fc_avg_commit_time = commit_time;
1172	} else if (status == EXT4_FC_STATUS_FAILED ||
1173		   status == EXT4_FC_STATUS_INELIGIBLE) {
1174		if (status == EXT4_FC_STATUS_FAILED)
1175			stats->fc_failed_commits++;
1176		stats->fc_ineligible_commits++;
1177	} else {
1178		stats->fc_skipped_commits++;
1179	}
1180	trace_ext4_fc_commit_stop(sb, nblks, status, commit_tid);
1181}
1182
1183/*
1184 * The main commit entry point. Performs a fast commit for transaction
1185 * commit_tid if needed. If it's not possible to perform a fast commit
1186 * due to various reasons, we fall back to full commit. Returns 0
1187 * on success, error otherwise.
1188 */
1189int ext4_fc_commit(journal_t *journal, tid_t commit_tid)
1190{
1191	struct super_block *sb = journal->j_private;
1192	struct ext4_sb_info *sbi = EXT4_SB(sb);
1193	int nblks = 0, ret, bsize = journal->j_blocksize;
1194	int subtid = atomic_read(&sbi->s_fc_subtid);
1195	int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;
1196	ktime_t start_time, commit_time;
1197
1198	if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
1199		return jbd2_complete_transaction(journal, commit_tid);
1200
1201	trace_ext4_fc_commit_start(sb, commit_tid);
1202
1203	start_time = ktime_get();
1204
1205restart_fc:
1206	ret = jbd2_fc_begin_commit(journal, commit_tid);
1207	if (ret == -EALREADY) {
1208		/* There was an ongoing commit, check if we need to restart */
1209		if (atomic_read(&sbi->s_fc_subtid) <= subtid &&
1210			commit_tid > journal->j_commit_sequence)
1211			goto restart_fc;
1212		ext4_fc_update_stats(sb, EXT4_FC_STATUS_SKIPPED, 0, 0,
1213				commit_tid);
1214		return 0;
1215	} else if (ret) {
1216		/*
1217		 * Commit couldn't start. Just update stats and perform a
1218		 * full commit.
1219		 */
1220		ext4_fc_update_stats(sb, EXT4_FC_STATUS_FAILED, 0, 0,
1221				commit_tid);
1222		return jbd2_complete_transaction(journal, commit_tid);
1223	}
1224
1225	/*
1226	 * After establishing journal barrier via jbd2_fc_begin_commit(), check
1227	 * if we are fast commit ineligible.
1228	 */
1229	if (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE)) {
1230		status = EXT4_FC_STATUS_INELIGIBLE;
1231		goto fallback;
1232	}
1233
1234	fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;
1235	ret = ext4_fc_perform_commit(journal);
1236	if (ret < 0) {
1237		status = EXT4_FC_STATUS_FAILED;
1238		goto fallback;
1239	}
1240	nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;
1241	ret = jbd2_fc_wait_bufs(journal, nblks);
1242	if (ret < 0) {
1243		status = EXT4_FC_STATUS_FAILED;
1244		goto fallback;
1245	}
1246	atomic_inc(&sbi->s_fc_subtid);
1247	ret = jbd2_fc_end_commit(journal);
1248	/*
1249	 * weight the commit time higher than the average time so we
1250	 * don't react too strongly to vast changes in the commit time
1251	 */
1252	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));
1253	ext4_fc_update_stats(sb, status, commit_time, nblks, commit_tid);
1254	return ret;
1255
1256fallback:
1257	ret = jbd2_fc_end_commit_fallback(journal);
1258	ext4_fc_update_stats(sb, status, 0, 0, commit_tid);
1259	return ret;
1260}
1261
1262/*
1263 * Fast commit cleanup routine. This is called after every fast commit and
1264 * full commit. full is true if we are called after a full commit.
1265 */
1266static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
1267{
1268	struct super_block *sb = journal->j_private;
1269	struct ext4_sb_info *sbi = EXT4_SB(sb);
1270	struct ext4_inode_info *iter, *iter_n;
1271	struct ext4_fc_dentry_update *fc_dentry;
1272
1273	if (full && sbi->s_fc_bh)
1274		sbi->s_fc_bh = NULL;
1275
1276	trace_ext4_fc_cleanup(journal, full, tid);
1277	jbd2_fc_release_bufs(journal);
1278
1279	spin_lock(&sbi->s_fc_lock);
1280	list_for_each_entry_safe(iter, iter_n, &sbi->s_fc_q[FC_Q_MAIN],
1281				 i_fc_list) {
1282		list_del_init(&iter->i_fc_list);
1283		ext4_clear_inode_state(&iter->vfs_inode,
1284				       EXT4_STATE_FC_COMMITTING);
1285		if (iter->i_sync_tid <= tid)
1286			ext4_fc_reset_inode(&iter->vfs_inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
1287		/* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
1288		smp_mb();
1289#if (BITS_PER_LONG < 64)
1290		wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);
1291#else
1292		wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);
1293#endif
1294	}
1295
1296	while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {
1297		fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],
1298					     struct ext4_fc_dentry_update,
1299					     fcd_list);
1300		list_del_init(&fc_dentry->fcd_list);
1301		list_del_init(&fc_dentry->fcd_dilist);
1302		spin_unlock(&sbi->s_fc_lock);
1303
1304		if (fc_dentry->fcd_name.name &&
1305			fc_dentry->fcd_name.len > DNAME_INLINE_LEN)
1306			kfree(fc_dentry->fcd_name.name);
1307		kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);
1308		spin_lock(&sbi->s_fc_lock);
1309	}
1310
1311	list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],
1312				&sbi->s_fc_dentry_q[FC_Q_MAIN]);
1313	list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
1314				&sbi->s_fc_q[FC_Q_MAIN]);
1315
1316	if (tid >= sbi->s_fc_ineligible_tid) {
1317		sbi->s_fc_ineligible_tid = 0;
1318		ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
1319	}
1320
1321	if (full)
1322		sbi->s_fc_bytes = 0;
1323	spin_unlock(&sbi->s_fc_lock);
1324	trace_ext4_fc_stats(sb);
1325}
1326
1327/* Ext4 Replay Path Routines */
1328
1329/* Helper struct for dentry replay routines */
1330struct dentry_info_args {
1331	int parent_ino, dname_len, ino, inode_len;
1332	char *dname;
1333};
1334
 
 
 
 
 
 
1335static inline void tl_to_darg(struct dentry_info_args *darg,
1336			      struct ext4_fc_tl *tl, u8 *val)
1337{
1338	struct ext4_fc_dentry_info fcd;
1339
1340	memcpy(&fcd, val, sizeof(fcd));
1341
1342	darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);
1343	darg->ino = le32_to_cpu(fcd.fc_ino);
1344	darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);
1345	darg->dname_len = tl->fc_len - sizeof(struct ext4_fc_dentry_info);
1346}
1347
1348static inline void ext4_fc_get_tl(struct ext4_fc_tl *tl, u8 *val)
1349{
1350	memcpy(tl, val, EXT4_FC_TAG_BASE_LEN);
1351	tl->fc_len = le16_to_cpu(tl->fc_len);
1352	tl->fc_tag = le16_to_cpu(tl->fc_tag);
 
 
1353}
1354
1355/* Unlink replay function */
1356static int ext4_fc_replay_unlink(struct super_block *sb, struct ext4_fc_tl *tl,
1357				 u8 *val)
1358{
1359	struct inode *inode, *old_parent;
1360	struct qstr entry;
1361	struct dentry_info_args darg;
1362	int ret = 0;
1363
1364	tl_to_darg(&darg, tl, val);
1365
1366	trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,
1367			darg.parent_ino, darg.dname_len);
1368
1369	entry.name = darg.dname;
1370	entry.len = darg.dname_len;
1371	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1372
1373	if (IS_ERR(inode)) {
1374		ext4_debug("Inode %d not found", darg.ino);
1375		return 0;
1376	}
1377
1378	old_parent = ext4_iget(sb, darg.parent_ino,
1379				EXT4_IGET_NORMAL);
1380	if (IS_ERR(old_parent)) {
1381		ext4_debug("Dir with inode %d not found", darg.parent_ino);
1382		iput(inode);
1383		return 0;
1384	}
1385
1386	ret = __ext4_unlink(old_parent, &entry, inode, NULL);
1387	/* -ENOENT ok coz it might not exist anymore. */
1388	if (ret == -ENOENT)
1389		ret = 0;
1390	iput(old_parent);
1391	iput(inode);
1392	return ret;
1393}
1394
1395static int ext4_fc_replay_link_internal(struct super_block *sb,
1396				struct dentry_info_args *darg,
1397				struct inode *inode)
1398{
1399	struct inode *dir = NULL;
1400	struct dentry *dentry_dir = NULL, *dentry_inode = NULL;
1401	struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);
1402	int ret = 0;
1403
1404	dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);
1405	if (IS_ERR(dir)) {
1406		ext4_debug("Dir with inode %d not found.", darg->parent_ino);
1407		dir = NULL;
1408		goto out;
1409	}
1410
1411	dentry_dir = d_obtain_alias(dir);
1412	if (IS_ERR(dentry_dir)) {
1413		ext4_debug("Failed to obtain dentry");
1414		dentry_dir = NULL;
1415		goto out;
1416	}
1417
1418	dentry_inode = d_alloc(dentry_dir, &qstr_dname);
1419	if (!dentry_inode) {
1420		ext4_debug("Inode dentry not created.");
1421		ret = -ENOMEM;
1422		goto out;
1423	}
1424
1425	ret = __ext4_link(dir, inode, dentry_inode);
1426	/*
1427	 * It's possible that link already existed since data blocks
1428	 * for the dir in question got persisted before we crashed OR
1429	 * we replayed this tag and crashed before the entire replay
1430	 * could complete.
1431	 */
1432	if (ret && ret != -EEXIST) {
1433		ext4_debug("Failed to link\n");
1434		goto out;
1435	}
1436
1437	ret = 0;
1438out:
1439	if (dentry_dir) {
1440		d_drop(dentry_dir);
1441		dput(dentry_dir);
1442	} else if (dir) {
1443		iput(dir);
1444	}
1445	if (dentry_inode) {
1446		d_drop(dentry_inode);
1447		dput(dentry_inode);
1448	}
1449
1450	return ret;
1451}
1452
1453/* Link replay function */
1454static int ext4_fc_replay_link(struct super_block *sb, struct ext4_fc_tl *tl,
1455			       u8 *val)
1456{
1457	struct inode *inode;
1458	struct dentry_info_args darg;
1459	int ret = 0;
1460
1461	tl_to_darg(&darg, tl, val);
1462	trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,
1463			darg.parent_ino, darg.dname_len);
1464
1465	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1466	if (IS_ERR(inode)) {
1467		ext4_debug("Inode not found.");
1468		return 0;
1469	}
1470
1471	ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1472	iput(inode);
1473	return ret;
1474}
1475
1476/*
1477 * Record all the modified inodes during replay. We use this later to setup
1478 * block bitmaps correctly.
1479 */
1480static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)
1481{
1482	struct ext4_fc_replay_state *state;
1483	int i;
1484
1485	state = &EXT4_SB(sb)->s_fc_replay_state;
1486	for (i = 0; i < state->fc_modified_inodes_used; i++)
1487		if (state->fc_modified_inodes[i] == ino)
1488			return 0;
1489	if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {
1490		int *fc_modified_inodes;
1491
1492		fc_modified_inodes = krealloc(state->fc_modified_inodes,
1493				sizeof(int) * (state->fc_modified_inodes_size +
1494				EXT4_FC_REPLAY_REALLOC_INCREMENT),
1495				GFP_KERNEL);
1496		if (!fc_modified_inodes)
1497			return -ENOMEM;
1498		state->fc_modified_inodes = fc_modified_inodes;
1499		state->fc_modified_inodes_size +=
1500			EXT4_FC_REPLAY_REALLOC_INCREMENT;
1501	}
1502	state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;
1503	return 0;
1504}
1505
1506/*
1507 * Inode replay function
1508 */
1509static int ext4_fc_replay_inode(struct super_block *sb, struct ext4_fc_tl *tl,
1510				u8 *val)
1511{
1512	struct ext4_fc_inode fc_inode;
1513	struct ext4_inode *raw_inode;
1514	struct ext4_inode *raw_fc_inode;
1515	struct inode *inode = NULL;
1516	struct ext4_iloc iloc;
1517	int inode_len, ino, ret, tag = tl->fc_tag;
1518	struct ext4_extent_header *eh;
1519	size_t off_gen = offsetof(struct ext4_inode, i_generation);
1520
1521	memcpy(&fc_inode, val, sizeof(fc_inode));
1522
1523	ino = le32_to_cpu(fc_inode.fc_ino);
1524	trace_ext4_fc_replay(sb, tag, ino, 0, 0);
1525
1526	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1527	if (!IS_ERR(inode)) {
1528		ext4_ext_clear_bb(inode);
1529		iput(inode);
1530	}
1531	inode = NULL;
1532
1533	ret = ext4_fc_record_modified_inode(sb, ino);
1534	if (ret)
1535		goto out;
1536
1537	raw_fc_inode = (struct ext4_inode *)
1538		(val + offsetof(struct ext4_fc_inode, fc_raw_inode));
1539	ret = ext4_get_fc_inode_loc(sb, ino, &iloc);
1540	if (ret)
1541		goto out;
1542
1543	inode_len = tl->fc_len - sizeof(struct ext4_fc_inode);
1544	raw_inode = ext4_raw_inode(&iloc);
1545
1546	memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));
1547	memcpy((u8 *)raw_inode + off_gen, (u8 *)raw_fc_inode + off_gen,
1548	       inode_len - off_gen);
1549	if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {
1550		eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);
1551		if (eh->eh_magic != EXT4_EXT_MAGIC) {
1552			memset(eh, 0, sizeof(*eh));
1553			eh->eh_magic = EXT4_EXT_MAGIC;
1554			eh->eh_max = cpu_to_le16(
1555				(sizeof(raw_inode->i_block) -
1556				 sizeof(struct ext4_extent_header))
1557				 / sizeof(struct ext4_extent));
1558		}
1559	} else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {
1560		memcpy(raw_inode->i_block, raw_fc_inode->i_block,
1561			sizeof(raw_inode->i_block));
1562	}
1563
1564	/* Immediately update the inode on disk. */
1565	ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1566	if (ret)
1567		goto out;
1568	ret = sync_dirty_buffer(iloc.bh);
1569	if (ret)
1570		goto out;
1571	ret = ext4_mark_inode_used(sb, ino);
1572	if (ret)
1573		goto out;
1574
1575	/* Given that we just wrote the inode on disk, this SHOULD succeed. */
1576	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1577	if (IS_ERR(inode)) {
1578		ext4_debug("Inode not found.");
1579		return -EFSCORRUPTED;
1580	}
1581
1582	/*
1583	 * Our allocator could have made different decisions than before
1584	 * crashing. This should be fixed but until then, we calculate
1585	 * the number of blocks the inode.
1586	 */
1587	if (!ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
1588		ext4_ext_replay_set_iblocks(inode);
1589
1590	inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);
1591	ext4_reset_inode_seed(inode);
1592
1593	ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));
1594	ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);
1595	sync_dirty_buffer(iloc.bh);
1596	brelse(iloc.bh);
1597out:
1598	iput(inode);
1599	if (!ret)
1600		blkdev_issue_flush(sb->s_bdev);
1601
1602	return 0;
1603}
1604
1605/*
1606 * Dentry create replay function.
1607 *
1608 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the
1609 * inode for which we are trying to create a dentry here, should already have
1610 * been replayed before we start here.
1611 */
1612static int ext4_fc_replay_create(struct super_block *sb, struct ext4_fc_tl *tl,
1613				 u8 *val)
1614{
1615	int ret = 0;
1616	struct inode *inode = NULL;
1617	struct inode *dir = NULL;
1618	struct dentry_info_args darg;
1619
1620	tl_to_darg(&darg, tl, val);
1621
1622	trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,
1623			darg.parent_ino, darg.dname_len);
1624
1625	/* This takes care of update group descriptor and other metadata */
1626	ret = ext4_mark_inode_used(sb, darg.ino);
1627	if (ret)
1628		goto out;
1629
1630	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);
1631	if (IS_ERR(inode)) {
1632		ext4_debug("inode %d not found.", darg.ino);
1633		inode = NULL;
1634		ret = -EINVAL;
1635		goto out;
1636	}
1637
1638	if (S_ISDIR(inode->i_mode)) {
1639		/*
1640		 * If we are creating a directory, we need to make sure that the
1641		 * dot and dot dot dirents are setup properly.
1642		 */
1643		dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);
1644		if (IS_ERR(dir)) {
1645			ext4_debug("Dir %d not found.", darg.ino);
1646			goto out;
1647		}
1648		ret = ext4_init_new_dir(NULL, dir, inode);
1649		iput(dir);
1650		if (ret) {
1651			ret = 0;
1652			goto out;
1653		}
1654	}
1655	ret = ext4_fc_replay_link_internal(sb, &darg, inode);
1656	if (ret)
1657		goto out;
1658	set_nlink(inode, 1);
1659	ext4_mark_inode_dirty(NULL, inode);
1660out:
1661	iput(inode);
1662	return ret;
1663}
1664
1665/*
1666 * Record physical disk regions which are in use as per fast commit area,
1667 * and used by inodes during replay phase. Our simple replay phase
1668 * allocator excludes these regions from allocation.
1669 */
1670int ext4_fc_record_regions(struct super_block *sb, int ino,
1671		ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)
1672{
1673	struct ext4_fc_replay_state *state;
1674	struct ext4_fc_alloc_region *region;
1675
1676	state = &EXT4_SB(sb)->s_fc_replay_state;
1677	/*
1678	 * during replay phase, the fc_regions_valid may not same as
1679	 * fc_regions_used, update it when do new additions.
1680	 */
1681	if (replay && state->fc_regions_used != state->fc_regions_valid)
1682		state->fc_regions_used = state->fc_regions_valid;
1683	if (state->fc_regions_used == state->fc_regions_size) {
1684		struct ext4_fc_alloc_region *fc_regions;
1685
1686		fc_regions = krealloc(state->fc_regions,
1687				      sizeof(struct ext4_fc_alloc_region) *
1688				      (state->fc_regions_size +
1689				       EXT4_FC_REPLAY_REALLOC_INCREMENT),
1690				      GFP_KERNEL);
1691		if (!fc_regions)
1692			return -ENOMEM;
1693		state->fc_regions_size +=
1694			EXT4_FC_REPLAY_REALLOC_INCREMENT;
1695		state->fc_regions = fc_regions;
1696	}
1697	region = &state->fc_regions[state->fc_regions_used++];
1698	region->ino = ino;
1699	region->lblk = lblk;
1700	region->pblk = pblk;
1701	region->len = len;
1702
1703	if (replay)
1704		state->fc_regions_valid++;
1705
1706	return 0;
1707}
1708
1709/* Replay add range tag */
1710static int ext4_fc_replay_add_range(struct super_block *sb,
1711				    struct ext4_fc_tl *tl, u8 *val)
1712{
1713	struct ext4_fc_add_range fc_add_ex;
1714	struct ext4_extent newex, *ex;
1715	struct inode *inode;
1716	ext4_lblk_t start, cur;
1717	int remaining, len;
1718	ext4_fsblk_t start_pblk;
1719	struct ext4_map_blocks map;
1720	struct ext4_ext_path *path = NULL;
1721	int ret;
1722
1723	memcpy(&fc_add_ex, val, sizeof(fc_add_ex));
1724	ex = (struct ext4_extent *)&fc_add_ex.fc_ex;
1725
1726	trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,
1727		le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),
1728		ext4_ext_get_actual_len(ex));
1729
1730	inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);
1731	if (IS_ERR(inode)) {
1732		ext4_debug("Inode not found.");
1733		return 0;
1734	}
1735
1736	ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1737	if (ret)
1738		goto out;
1739
1740	start = le32_to_cpu(ex->ee_block);
1741	start_pblk = ext4_ext_pblock(ex);
1742	len = ext4_ext_get_actual_len(ex);
1743
1744	cur = start;
1745	remaining = len;
1746	ext4_debug("ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",
1747		  start, start_pblk, len, ext4_ext_is_unwritten(ex),
1748		  inode->i_ino);
1749
1750	while (remaining > 0) {
1751		map.m_lblk = cur;
1752		map.m_len = remaining;
1753		map.m_pblk = 0;
1754		ret = ext4_map_blocks(NULL, inode, &map, 0);
1755
1756		if (ret < 0)
1757			goto out;
1758
1759		if (ret == 0) {
1760			/* Range is not mapped */
1761			path = ext4_find_extent(inode, cur, NULL, 0);
1762			if (IS_ERR(path))
1763				goto out;
1764			memset(&newex, 0, sizeof(newex));
1765			newex.ee_block = cpu_to_le32(cur);
1766			ext4_ext_store_pblock(
1767				&newex, start_pblk + cur - start);
1768			newex.ee_len = cpu_to_le16(map.m_len);
1769			if (ext4_ext_is_unwritten(ex))
1770				ext4_ext_mark_unwritten(&newex);
1771			down_write(&EXT4_I(inode)->i_data_sem);
1772			ret = ext4_ext_insert_extent(
1773				NULL, inode, &path, &newex, 0);
1774			up_write((&EXT4_I(inode)->i_data_sem));
1775			ext4_free_ext_path(path);
1776			if (ret)
1777				goto out;
1778			goto next;
1779		}
1780
1781		if (start_pblk + cur - start != map.m_pblk) {
1782			/*
1783			 * Logical to physical mapping changed. This can happen
1784			 * if this range was removed and then reallocated to
1785			 * map to new physical blocks during a fast commit.
1786			 */
1787			ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1788					ext4_ext_is_unwritten(ex),
1789					start_pblk + cur - start);
1790			if (ret)
1791				goto out;
1792			/*
1793			 * Mark the old blocks as free since they aren't used
1794			 * anymore. We maintain an array of all the modified
1795			 * inodes. In case these blocks are still used at either
1796			 * a different logical range in the same inode or in
1797			 * some different inode, we will mark them as allocated
1798			 * at the end of the FC replay using our array of
1799			 * modified inodes.
1800			 */
1801			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1802			goto next;
1803		}
1804
1805		/* Range is mapped and needs a state change */
1806		ext4_debug("Converting from %ld to %d %lld",
1807				map.m_flags & EXT4_MAP_UNWRITTEN,
1808			ext4_ext_is_unwritten(ex), map.m_pblk);
1809		ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,
1810					ext4_ext_is_unwritten(ex), map.m_pblk);
1811		if (ret)
1812			goto out;
1813		/*
1814		 * We may have split the extent tree while toggling the state.
1815		 * Try to shrink the extent tree now.
1816		 */
1817		ext4_ext_replay_shrink_inode(inode, start + len);
1818next:
1819		cur += map.m_len;
1820		remaining -= map.m_len;
1821	}
1822	ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>
1823					sb->s_blocksize_bits);
1824out:
 
1825	iput(inode);
1826	return 0;
1827}
1828
1829/* Replay DEL_RANGE tag */
1830static int
1831ext4_fc_replay_del_range(struct super_block *sb, struct ext4_fc_tl *tl,
1832			 u8 *val)
1833{
1834	struct inode *inode;
1835	struct ext4_fc_del_range lrange;
1836	struct ext4_map_blocks map;
1837	ext4_lblk_t cur, remaining;
1838	int ret;
1839
1840	memcpy(&lrange, val, sizeof(lrange));
1841	cur = le32_to_cpu(lrange.fc_lblk);
1842	remaining = le32_to_cpu(lrange.fc_len);
1843
1844	trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,
1845		le32_to_cpu(lrange.fc_ino), cur, remaining);
1846
1847	inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);
1848	if (IS_ERR(inode)) {
1849		ext4_debug("Inode %d not found", le32_to_cpu(lrange.fc_ino));
1850		return 0;
1851	}
1852
1853	ret = ext4_fc_record_modified_inode(sb, inode->i_ino);
1854	if (ret)
1855		goto out;
1856
1857	ext4_debug("DEL_RANGE, inode %ld, lblk %d, len %d\n",
1858			inode->i_ino, le32_to_cpu(lrange.fc_lblk),
1859			le32_to_cpu(lrange.fc_len));
1860	while (remaining > 0) {
1861		map.m_lblk = cur;
1862		map.m_len = remaining;
1863
1864		ret = ext4_map_blocks(NULL, inode, &map, 0);
1865		if (ret < 0)
1866			goto out;
1867		if (ret > 0) {
1868			remaining -= ret;
1869			cur += ret;
1870			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
1871		} else {
1872			remaining -= map.m_len;
1873			cur += map.m_len;
1874		}
1875	}
1876
1877	down_write(&EXT4_I(inode)->i_data_sem);
1878	ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),
1879				le32_to_cpu(lrange.fc_lblk) +
1880				le32_to_cpu(lrange.fc_len) - 1);
1881	up_write(&EXT4_I(inode)->i_data_sem);
1882	if (ret)
1883		goto out;
1884	ext4_ext_replay_shrink_inode(inode,
1885		i_size_read(inode) >> sb->s_blocksize_bits);
1886	ext4_mark_inode_dirty(NULL, inode);
1887out:
1888	iput(inode);
1889	return 0;
1890}
1891
1892static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)
1893{
1894	struct ext4_fc_replay_state *state;
1895	struct inode *inode;
1896	struct ext4_ext_path *path = NULL;
1897	struct ext4_map_blocks map;
1898	int i, ret, j;
1899	ext4_lblk_t cur, end;
1900
1901	state = &EXT4_SB(sb)->s_fc_replay_state;
1902	for (i = 0; i < state->fc_modified_inodes_used; i++) {
1903		inode = ext4_iget(sb, state->fc_modified_inodes[i],
1904			EXT4_IGET_NORMAL);
1905		if (IS_ERR(inode)) {
1906			ext4_debug("Inode %d not found.",
1907				state->fc_modified_inodes[i]);
1908			continue;
1909		}
1910		cur = 0;
1911		end = EXT_MAX_BLOCKS;
1912		if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) {
1913			iput(inode);
1914			continue;
1915		}
1916		while (cur < end) {
1917			map.m_lblk = cur;
1918			map.m_len = end - cur;
1919
1920			ret = ext4_map_blocks(NULL, inode, &map, 0);
1921			if (ret < 0)
1922				break;
1923
1924			if (ret > 0) {
1925				path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
1926				if (!IS_ERR(path)) {
1927					for (j = 0; j < path->p_depth; j++)
1928						ext4_mb_mark_bb(inode->i_sb,
1929							path[j].p_block, 1, 1);
1930					ext4_free_ext_path(path);
 
1931				}
1932				cur += ret;
1933				ext4_mb_mark_bb(inode->i_sb, map.m_pblk,
1934							map.m_len, 1);
1935			} else {
1936				cur = cur + (map.m_len ? map.m_len : 1);
1937			}
1938		}
1939		iput(inode);
1940	}
 
 
1941}
1942
1943/*
1944 * Check if block is in excluded regions for block allocation. The simple
1945 * allocator that runs during replay phase is calls this function to see
1946 * if it is okay to use a block.
1947 */
1948bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)
1949{
1950	int i;
1951	struct ext4_fc_replay_state *state;
1952
1953	state = &EXT4_SB(sb)->s_fc_replay_state;
1954	for (i = 0; i < state->fc_regions_valid; i++) {
1955		if (state->fc_regions[i].ino == 0 ||
1956			state->fc_regions[i].len == 0)
1957			continue;
1958		if (in_range(blk, state->fc_regions[i].pblk,
1959					state->fc_regions[i].len))
1960			return true;
1961	}
1962	return false;
1963}
1964
1965/* Cleanup function called after replay */
1966void ext4_fc_replay_cleanup(struct super_block *sb)
1967{
1968	struct ext4_sb_info *sbi = EXT4_SB(sb);
1969
1970	sbi->s_mount_state &= ~EXT4_FC_REPLAY;
1971	kfree(sbi->s_fc_replay_state.fc_regions);
1972	kfree(sbi->s_fc_replay_state.fc_modified_inodes);
1973}
1974
1975static bool ext4_fc_value_len_isvalid(struct ext4_sb_info *sbi,
1976				      int tag, int len)
1977{
1978	switch (tag) {
1979	case EXT4_FC_TAG_ADD_RANGE:
1980		return len == sizeof(struct ext4_fc_add_range);
1981	case EXT4_FC_TAG_DEL_RANGE:
1982		return len == sizeof(struct ext4_fc_del_range);
1983	case EXT4_FC_TAG_CREAT:
1984	case EXT4_FC_TAG_LINK:
1985	case EXT4_FC_TAG_UNLINK:
1986		len -= sizeof(struct ext4_fc_dentry_info);
1987		return len >= 1 && len <= EXT4_NAME_LEN;
1988	case EXT4_FC_TAG_INODE:
1989		len -= sizeof(struct ext4_fc_inode);
1990		return len >= EXT4_GOOD_OLD_INODE_SIZE &&
1991			len <= sbi->s_inode_size;
1992	case EXT4_FC_TAG_PAD:
1993		return true; /* padding can have any length */
1994	case EXT4_FC_TAG_TAIL:
1995		return len >= sizeof(struct ext4_fc_tail);
1996	case EXT4_FC_TAG_HEAD:
1997		return len == sizeof(struct ext4_fc_head);
1998	}
1999	return false;
2000}
2001
2002/*
2003 * Recovery Scan phase handler
2004 *
2005 * This function is called during the scan phase and is responsible
2006 * for doing following things:
2007 * - Make sure the fast commit area has valid tags for replay
2008 * - Count number of tags that need to be replayed by the replay handler
2009 * - Verify CRC
2010 * - Create a list of excluded blocks for allocation during replay phase
2011 *
2012 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is
2013 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP
2014 * to indicate that scan has finished and JBD2 can now start replay phase.
2015 * It returns a negative error to indicate that there was an error. At the end
2016 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set
2017 * to indicate the number of tags that need to replayed during the replay phase.
2018 */
2019static int ext4_fc_replay_scan(journal_t *journal,
2020				struct buffer_head *bh, int off,
2021				tid_t expected_tid)
2022{
2023	struct super_block *sb = journal->j_private;
2024	struct ext4_sb_info *sbi = EXT4_SB(sb);
2025	struct ext4_fc_replay_state *state;
2026	int ret = JBD2_FC_REPLAY_CONTINUE;
2027	struct ext4_fc_add_range ext;
2028	struct ext4_fc_tl tl;
2029	struct ext4_fc_tail tail;
2030	__u8 *start, *end, *cur, *val;
2031	struct ext4_fc_head head;
2032	struct ext4_extent *ex;
2033
2034	state = &sbi->s_fc_replay_state;
2035
2036	start = (u8 *)bh->b_data;
2037	end = start + journal->j_blocksize;
2038
2039	if (state->fc_replay_expected_off == 0) {
2040		state->fc_cur_tag = 0;
2041		state->fc_replay_num_tags = 0;
2042		state->fc_crc = 0;
2043		state->fc_regions = NULL;
2044		state->fc_regions_valid = state->fc_regions_used =
2045			state->fc_regions_size = 0;
2046		/* Check if we can stop early */
2047		if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)
2048			!= EXT4_FC_TAG_HEAD)
2049			return 0;
2050	}
2051
2052	if (off != state->fc_replay_expected_off) {
2053		ret = -EFSCORRUPTED;
2054		goto out_err;
2055	}
2056
2057	state->fc_replay_expected_off++;
2058	for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
2059	     cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
2060		ext4_fc_get_tl(&tl, cur);
2061		val = cur + EXT4_FC_TAG_BASE_LEN;
2062		if (tl.fc_len > end - val ||
2063		    !ext4_fc_value_len_isvalid(sbi, tl.fc_tag, tl.fc_len)) {
2064			ret = state->fc_replay_num_tags ?
2065				JBD2_FC_REPLAY_STOP : -ECANCELED;
2066			goto out_err;
2067		}
2068		ext4_debug("Scan phase, tag:%s, blk %lld\n",
2069			   tag2str(tl.fc_tag), bh->b_blocknr);
2070		switch (tl.fc_tag) {
2071		case EXT4_FC_TAG_ADD_RANGE:
2072			memcpy(&ext, val, sizeof(ext));
2073			ex = (struct ext4_extent *)&ext.fc_ex;
2074			ret = ext4_fc_record_regions(sb,
2075				le32_to_cpu(ext.fc_ino),
2076				le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),
2077				ext4_ext_get_actual_len(ex), 0);
2078			if (ret < 0)
2079				break;
2080			ret = JBD2_FC_REPLAY_CONTINUE;
2081			fallthrough;
2082		case EXT4_FC_TAG_DEL_RANGE:
2083		case EXT4_FC_TAG_LINK:
2084		case EXT4_FC_TAG_UNLINK:
2085		case EXT4_FC_TAG_CREAT:
2086		case EXT4_FC_TAG_INODE:
2087		case EXT4_FC_TAG_PAD:
2088			state->fc_cur_tag++;
2089			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2090				EXT4_FC_TAG_BASE_LEN + tl.fc_len);
2091			break;
2092		case EXT4_FC_TAG_TAIL:
2093			state->fc_cur_tag++;
2094			memcpy(&tail, val, sizeof(tail));
2095			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2096						EXT4_FC_TAG_BASE_LEN +
2097						offsetof(struct ext4_fc_tail,
2098						fc_crc));
2099			if (le32_to_cpu(tail.fc_tid) == expected_tid &&
2100				le32_to_cpu(tail.fc_crc) == state->fc_crc) {
2101				state->fc_replay_num_tags = state->fc_cur_tag;
2102				state->fc_regions_valid =
2103					state->fc_regions_used;
2104			} else {
2105				ret = state->fc_replay_num_tags ?
2106					JBD2_FC_REPLAY_STOP : -EFSBADCRC;
2107			}
2108			state->fc_crc = 0;
2109			break;
2110		case EXT4_FC_TAG_HEAD:
2111			memcpy(&head, val, sizeof(head));
2112			if (le32_to_cpu(head.fc_features) &
2113				~EXT4_FC_SUPPORTED_FEATURES) {
2114				ret = -EOPNOTSUPP;
2115				break;
2116			}
2117			if (le32_to_cpu(head.fc_tid) != expected_tid) {
2118				ret = JBD2_FC_REPLAY_STOP;
2119				break;
2120			}
2121			state->fc_cur_tag++;
2122			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,
2123				EXT4_FC_TAG_BASE_LEN + tl.fc_len);
2124			break;
2125		default:
2126			ret = state->fc_replay_num_tags ?
2127				JBD2_FC_REPLAY_STOP : -ECANCELED;
2128		}
2129		if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)
2130			break;
2131	}
2132
2133out_err:
2134	trace_ext4_fc_replay_scan(sb, ret, off);
2135	return ret;
2136}
2137
2138/*
2139 * Main recovery path entry point.
2140 * The meaning of return codes is similar as above.
2141 */
2142static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,
2143				enum passtype pass, int off, tid_t expected_tid)
2144{
2145	struct super_block *sb = journal->j_private;
2146	struct ext4_sb_info *sbi = EXT4_SB(sb);
2147	struct ext4_fc_tl tl;
2148	__u8 *start, *end, *cur, *val;
2149	int ret = JBD2_FC_REPLAY_CONTINUE;
2150	struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;
2151	struct ext4_fc_tail tail;
2152
2153	if (pass == PASS_SCAN) {
2154		state->fc_current_pass = PASS_SCAN;
2155		return ext4_fc_replay_scan(journal, bh, off, expected_tid);
2156	}
2157
2158	if (state->fc_current_pass != pass) {
2159		state->fc_current_pass = pass;
2160		sbi->s_mount_state |= EXT4_FC_REPLAY;
2161	}
2162	if (!sbi->s_fc_replay_state.fc_replay_num_tags) {
2163		ext4_debug("Replay stops\n");
2164		ext4_fc_set_bitmaps_and_counters(sb);
2165		return 0;
2166	}
2167
2168#ifdef CONFIG_EXT4_DEBUG
2169	if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {
2170		pr_warn("Dropping fc block %d because max_replay set\n", off);
2171		return JBD2_FC_REPLAY_STOP;
2172	}
2173#endif
2174
2175	start = (u8 *)bh->b_data;
2176	end = start + journal->j_blocksize;
2177
2178	for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;
2179	     cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {
2180		ext4_fc_get_tl(&tl, cur);
2181		val = cur + EXT4_FC_TAG_BASE_LEN;
2182
2183		if (state->fc_replay_num_tags == 0) {
2184			ret = JBD2_FC_REPLAY_STOP;
2185			ext4_fc_set_bitmaps_and_counters(sb);
2186			break;
2187		}
2188
2189		ext4_debug("Replay phase, tag:%s\n", tag2str(tl.fc_tag));
2190		state->fc_replay_num_tags--;
2191		switch (tl.fc_tag) {
2192		case EXT4_FC_TAG_LINK:
2193			ret = ext4_fc_replay_link(sb, &tl, val);
2194			break;
2195		case EXT4_FC_TAG_UNLINK:
2196			ret = ext4_fc_replay_unlink(sb, &tl, val);
2197			break;
2198		case EXT4_FC_TAG_ADD_RANGE:
2199			ret = ext4_fc_replay_add_range(sb, &tl, val);
2200			break;
2201		case EXT4_FC_TAG_CREAT:
2202			ret = ext4_fc_replay_create(sb, &tl, val);
2203			break;
2204		case EXT4_FC_TAG_DEL_RANGE:
2205			ret = ext4_fc_replay_del_range(sb, &tl, val);
2206			break;
2207		case EXT4_FC_TAG_INODE:
2208			ret = ext4_fc_replay_inode(sb, &tl, val);
2209			break;
2210		case EXT4_FC_TAG_PAD:
2211			trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,
2212					     tl.fc_len, 0);
2213			break;
2214		case EXT4_FC_TAG_TAIL:
2215			trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL,
2216					     0, tl.fc_len, 0);
2217			memcpy(&tail, val, sizeof(tail));
2218			WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);
2219			break;
2220		case EXT4_FC_TAG_HEAD:
2221			break;
2222		default:
2223			trace_ext4_fc_replay(sb, tl.fc_tag, 0, tl.fc_len, 0);
2224			ret = -ECANCELED;
2225			break;
2226		}
2227		if (ret < 0)
2228			break;
2229		ret = JBD2_FC_REPLAY_CONTINUE;
2230	}
2231	return ret;
2232}
2233
2234void ext4_fc_init(struct super_block *sb, journal_t *journal)
2235{
2236	/*
2237	 * We set replay callback even if fast commit disabled because we may
2238	 * could still have fast commit blocks that need to be replayed even if
2239	 * fast commit has now been turned off.
2240	 */
2241	journal->j_fc_replay_callback = ext4_fc_replay;
2242	if (!test_opt2(sb, JOURNAL_FAST_COMMIT))
2243		return;
2244	journal->j_fc_cleanup_callback = ext4_fc_cleanup;
2245}
2246
2247static const char * const fc_ineligible_reasons[] = {
2248	[EXT4_FC_REASON_XATTR] = "Extended attributes changed",
2249	[EXT4_FC_REASON_CROSS_RENAME] = "Cross rename",
2250	[EXT4_FC_REASON_JOURNAL_FLAG_CHANGE] = "Journal flag changed",
2251	[EXT4_FC_REASON_NOMEM] = "Insufficient memory",
2252	[EXT4_FC_REASON_SWAP_BOOT] = "Swap boot",
2253	[EXT4_FC_REASON_RESIZE] = "Resize",
2254	[EXT4_FC_REASON_RENAME_DIR] = "Dir renamed",
2255	[EXT4_FC_REASON_FALLOC_RANGE] = "Falloc range op",
2256	[EXT4_FC_REASON_INODE_JOURNAL_DATA] = "Data journalling",
2257	[EXT4_FC_REASON_ENCRYPTED_FILENAME] = "Encrypted filename",
2258};
2259
2260int ext4_fc_info_show(struct seq_file *seq, void *v)
2261{
2262	struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);
2263	struct ext4_fc_stats *stats = &sbi->s_fc_stats;
2264	int i;
2265
2266	if (v != SEQ_START_TOKEN)
2267		return 0;
2268
2269	seq_printf(seq,
2270		"fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",
2271		   stats->fc_num_commits, stats->fc_ineligible_commits,
2272		   stats->fc_numblks,
2273		   div_u64(stats->s_fc_avg_commit_time, 1000));
2274	seq_puts(seq, "Ineligible reasons:\n");
2275	for (i = 0; i < EXT4_FC_REASON_MAX; i++)
2276		seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],
2277			stats->fc_ineligible_reason_count[i]);
2278
2279	return 0;
2280}
2281
2282int __init ext4_fc_init_dentry_cache(void)
2283{
2284	ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,
2285					   SLAB_RECLAIM_ACCOUNT);
2286
2287	if (ext4_fc_dentry_cachep == NULL)
2288		return -ENOMEM;
2289
2290	return 0;
2291}
2292
2293void ext4_fc_destroy_dentry_cache(void)
2294{
2295	kmem_cache_destroy(ext4_fc_dentry_cachep);
2296}