brintos

brintos / linux-shallow public Read only

0
0
Text · 65.2 KiB · b33664f Raw
2337 lines · c
1// SPDX-License-Identifier: GPL-2.02 3/*4 * fs/ext4/fast_commit.c5 *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 Commits17 * -----------------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. (See22 * struct ext4_fc_tl). Each TLV contains some delta that is replayed TLV by23 * TLV during the recovery phase. For the scenarios for which we currently24 * 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 unlink30 * - EXT4_FC_TAG_LINK		- records directory entry link31 * - EXT4_FC_TAG_CREAT		- records inode and directory entry creation32 *33 * (B) File specific data range updates:34 *35 * - EXT4_FC_TAG_ADD_RANGE	- records addition of new blocks to an inode36 * - EXT4_FC_TAG_DEL_RANGE	- records deletion of blocks from an inode37 *38 * (C) Inode metadata (mtime / ctime etc):39 *40 * - EXT4_FC_TAG_INODE		- record the inode that should be replayed41 *				  during recovery. Note that iblocks field is42 *				  not replayed and instead derived during43 *				  replay.44 * Commit Operation45 * ----------------46 * With fast commits, we maintain all the directory entry operations in the47 * order in which they are issued in an in-memory queue. This queue is flushed48 * to disk during the commit operation. We also maintain a list of inodes49 * that need to be committed during a fast commit in another in memory queue of50 * inodes. During the commit operation, we commit in the following order:51 *52 * [1] Lock inodes for any further data updates by setting COMMITTING state53 * [2] Submit data buffers of all the inodes54 * [3] Wait for [2] to complete55 * [4] Commit all the directory entry updates in the fast commit space56 * [5] Commit all the changed inode structures57 * [6] Write tail tag (this tag ensures the atomicity, please read the following58 *     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 an62 * update. If such an ongoing update is present, fast commit waits for it to63 * complete. The completion of such an update is marked by64 * ext4_fc_stop_update().65 *66 * Fast Commit Ineligibility67 * -------------------------68 *69 * Not all operations are supported by fast commits today (e.g extended70 * attributes). Fast commit ineligibility is marked by calling71 * ext4_fc_mark_ineligible(): This makes next fast commit operation to fall back72 * to full commit.73 *74 * Atomicity of commits75 * --------------------76 * In order to guarantee atomicity during the commit operation, fast commit77 * uses "EXT4_FC_TAG_TAIL" tag that marks a fast commit as complete. Tail78 * tag contains CRC of the contents and TID of the transaction after which79 * this fast commit should be applied. Recovery code replays fast commit80 * logs only if there's at least 1 valid tail present. For every fast commit81 * operation, there is 1 tail. This means, we may end up with multiple tails82 * in the fast commit space. Here's an example:83 *84 * - Create a new file A and remove existing file B85 * - fsync()86 * - Append contents to file A87 * - Truncate file A88 * - 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 Idempotence97 * ------------------------------98 *99 * Fast commits tags are idempotent in nature provided the recovery code follows100 * certain rules. The guiding principle that the commit path follows while101 * committing is that it stores the result of a particular operation instead of102 * 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 this106 * operation as a procedure "rename a to b", we store the resulting file system107 * state as a "series" of outcomes:108 *109 * - Link dirent b to inode 10110 * - Unlink dirent a111 * - Inode <10> with valid refcount112 *113 * Now when recovery code runs, it needs "enforce" this state on the file114 * 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 fast117 * commits make it idempotent. Consider following sequence of operations:118 *119 *     rm A;    mv B A;    read A120 *  (x)     (y)        (z)121 *122 * (x), (y) and (z) are the points at which we can crash. If we store this123 * sequence of operations as is then the replay is not idempotent. Let's say124 * while in replay, we crash at (z). During the second replay, file A (which was125 * 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 of127 * operations is not idempotent. However, as mentioned above, instead of storing128 * the procedure fast commits store the outcome of each procedure. Thus the fast129 * 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 to132 * 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 second138 * replay, we will remove file A (inode 11). But we will create it back and make139 * it point to inode 11. We won't find B, so we'll just skip that step. At this140 * point, the refcount for inode 11 is not reliable, but that gets fixed by the141 * replay of last inode 11 tag. Crashes at points (w), (x) and (y) get handled142 * similarly. Thus, by converting a non-idempotent procedure into a series of143 * idempotent outcomes, fast commits ensured idempotence during the replay.144 *145 * TODOs146 * -----147 *148 * 0) Fast commit replay path hardening: Fast commit replay code should use149 *    journal handles to make sure all the updates it does during the replay150 *    path are atomic. With that if we crash during fast commit replay, after151 *    trying to do recovery again, we will find a file system where fast commit152 *    area is invalid (because new full commit would be found). In order to deal153 *    with that, fast commit replay code should ensure that the "FC_REPLAY"154 *    superblock state is persisted before starting the replay, so that after155 *    the crash, fast commit recovery code can look at that flag and perform156 *    fast commit recovery even if that area is invalidated by later full157 *    commits.158 *159 * 1) Fast commit's commit path locks the entire file system during fast160 *    commit. This has significant performance penalty. Instead of that, we161 *    should use ext4_fc_start/stop_update functions to start inode level162 *    updates from ext4_journal_start/stop. Once we do that we can drop file163 *    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#else220	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#endif225	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 update240 *241 * This function is called by the high level call VFS callbacks before242 * performing any inode update. This function blocks if there's an ongoing243 * 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 committed282 * 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 FC310	 * 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 latest335 * ineligible transaction tid. This means until the recorded336 * 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	bool has_transaction = true;343	bool is_ineligible;344 345	if (ext4_fc_disabled(sb))346		return;347 348	if (handle && !IS_ERR(handle))349		tid = handle->h_transaction->t_tid;350	else {351		read_lock(&sbi->s_journal->j_state_lock);352		if (sbi->s_journal->j_running_transaction)353			tid = sbi->s_journal->j_running_transaction->t_tid;354		else355			has_transaction = false;356		read_unlock(&sbi->s_journal->j_state_lock);357	}358	spin_lock(&sbi->s_fc_lock);359	is_ineligible = ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);360	if (has_transaction &&361	    (!is_ineligible ||362	     (is_ineligible && tid_gt(tid, sbi->s_fc_ineligible_tid))))363		sbi->s_fc_ineligible_tid = tid;364	ext4_set_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);365	spin_unlock(&sbi->s_fc_lock);366	WARN_ON(reason >= EXT4_FC_REASON_MAX);367	sbi->s_fc_stats.fc_ineligible_reason_count[reason]++;368}369 370/*371 * Generic fast commit tracking function. If this is the first time this we are372 * called after a full commit, we initialize fast commit fields and then call373 * __fc_track_fn() with update = 0. If we have already been called after a full374 * commit, we pass update = 1. Based on that, the track function can determine375 * if it needs to track a field for the first time or if it needs to just376 * update the previously tracked value.377 *378 * If enqueue is set, this function enqueues the inode in fast commit list.379 */380static int ext4_fc_track_template(381	handle_t *handle, struct inode *inode,382	int (*__fc_track_fn)(handle_t *handle, struct inode *, void *, bool),383	void *args, int enqueue)384{385	bool update = false;386	struct ext4_inode_info *ei = EXT4_I(inode);387	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);388	tid_t tid = 0;389	int ret;390 391	tid = handle->h_transaction->t_tid;392	mutex_lock(&ei->i_fc_lock);393	if (tid == ei->i_sync_tid) {394		update = true;395	} else {396		ext4_fc_reset_inode(inode);397		ei->i_sync_tid = tid;398	}399	ret = __fc_track_fn(handle, inode, args, update);400	mutex_unlock(&ei->i_fc_lock);401 402	if (!enqueue)403		return ret;404 405	spin_lock(&sbi->s_fc_lock);406	if (list_empty(&EXT4_I(inode)->i_fc_list))407		list_add_tail(&EXT4_I(inode)->i_fc_list,408				(sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||409				 sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?410				&sbi->s_fc_q[FC_Q_STAGING] :411				&sbi->s_fc_q[FC_Q_MAIN]);412	spin_unlock(&sbi->s_fc_lock);413 414	return ret;415}416 417struct __track_dentry_update_args {418	struct dentry *dentry;419	int op;420};421 422/* __track_fn for directory entry updates. Called with ei->i_fc_lock. */423static int __track_dentry_update(handle_t *handle, struct inode *inode,424				 void *arg, bool update)425{426	struct ext4_fc_dentry_update *node;427	struct ext4_inode_info *ei = EXT4_I(inode);428	struct __track_dentry_update_args *dentry_update =429		(struct __track_dentry_update_args *)arg;430	struct dentry *dentry = dentry_update->dentry;431	struct inode *dir = dentry->d_parent->d_inode;432	struct super_block *sb = inode->i_sb;433	struct ext4_sb_info *sbi = EXT4_SB(sb);434 435	mutex_unlock(&ei->i_fc_lock);436 437	if (IS_ENCRYPTED(dir)) {438		ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_ENCRYPTED_FILENAME,439					handle);440		mutex_lock(&ei->i_fc_lock);441		return -EOPNOTSUPP;442	}443 444	node = kmem_cache_alloc(ext4_fc_dentry_cachep, GFP_NOFS);445	if (!node) {446		ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, handle);447		mutex_lock(&ei->i_fc_lock);448		return -ENOMEM;449	}450 451	node->fcd_op = dentry_update->op;452	node->fcd_parent = dir->i_ino;453	node->fcd_ino = inode->i_ino;454	if (dentry->d_name.len > DNAME_INLINE_LEN) {455		node->fcd_name.name = kmalloc(dentry->d_name.len, GFP_NOFS);456		if (!node->fcd_name.name) {457			kmem_cache_free(ext4_fc_dentry_cachep, node);458			ext4_fc_mark_ineligible(sb, EXT4_FC_REASON_NOMEM, handle);459			mutex_lock(&ei->i_fc_lock);460			return -ENOMEM;461		}462		memcpy((u8 *)node->fcd_name.name, dentry->d_name.name,463			dentry->d_name.len);464	} else {465		memcpy(node->fcd_iname, dentry->d_name.name,466			dentry->d_name.len);467		node->fcd_name.name = node->fcd_iname;468	}469	node->fcd_name.len = dentry->d_name.len;470	INIT_LIST_HEAD(&node->fcd_dilist);471	spin_lock(&sbi->s_fc_lock);472	if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||473		sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)474		list_add_tail(&node->fcd_list,475				&sbi->s_fc_dentry_q[FC_Q_STAGING]);476	else477		list_add_tail(&node->fcd_list, &sbi->s_fc_dentry_q[FC_Q_MAIN]);478 479	/*480	 * This helps us keep a track of all fc_dentry updates which is part of481	 * this ext4 inode. So in case the inode is getting unlinked, before482	 * even we get a chance to fsync, we could remove all fc_dentry483	 * references while evicting the inode in ext4_fc_del().484	 * Also with this, we don't need to loop over all the inodes in485	 * sbi->s_fc_q to get the corresponding inode in486	 * ext4_fc_commit_dentry_updates().487	 */488	if (dentry_update->op == EXT4_FC_TAG_CREAT) {489		WARN_ON(!list_empty(&ei->i_fc_dilist));490		list_add_tail(&node->fcd_dilist, &ei->i_fc_dilist);491	}492	spin_unlock(&sbi->s_fc_lock);493	mutex_lock(&ei->i_fc_lock);494 495	return 0;496}497 498void __ext4_fc_track_unlink(handle_t *handle,499		struct inode *inode, struct dentry *dentry)500{501	struct __track_dentry_update_args args;502	int ret;503 504	args.dentry = dentry;505	args.op = EXT4_FC_TAG_UNLINK;506 507	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,508					(void *)&args, 0);509	trace_ext4_fc_track_unlink(handle, inode, dentry, ret);510}511 512void ext4_fc_track_unlink(handle_t *handle, struct dentry *dentry)513{514	struct inode *inode = d_inode(dentry);515 516	if (ext4_fc_disabled(inode->i_sb))517		return;518 519	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))520		return;521 522	__ext4_fc_track_unlink(handle, inode, dentry);523}524 525void __ext4_fc_track_link(handle_t *handle,526	struct inode *inode, struct dentry *dentry)527{528	struct __track_dentry_update_args args;529	int ret;530 531	args.dentry = dentry;532	args.op = EXT4_FC_TAG_LINK;533 534	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,535					(void *)&args, 0);536	trace_ext4_fc_track_link(handle, inode, dentry, ret);537}538 539void ext4_fc_track_link(handle_t *handle, struct dentry *dentry)540{541	struct inode *inode = d_inode(dentry);542 543	if (ext4_fc_disabled(inode->i_sb))544		return;545 546	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))547		return;548 549	__ext4_fc_track_link(handle, inode, dentry);550}551 552void __ext4_fc_track_create(handle_t *handle, struct inode *inode,553			  struct dentry *dentry)554{555	struct __track_dentry_update_args args;556	int ret;557 558	args.dentry = dentry;559	args.op = EXT4_FC_TAG_CREAT;560 561	ret = ext4_fc_track_template(handle, inode, __track_dentry_update,562					(void *)&args, 0);563	trace_ext4_fc_track_create(handle, inode, dentry, ret);564}565 566void ext4_fc_track_create(handle_t *handle, struct dentry *dentry)567{568	struct inode *inode = d_inode(dentry);569 570	if (ext4_fc_disabled(inode->i_sb))571		return;572 573	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))574		return;575 576	__ext4_fc_track_create(handle, inode, dentry);577}578 579/* __track_fn for inode tracking */580static int __track_inode(handle_t *handle, struct inode *inode, void *arg,581			 bool update)582{583	if (update)584		return -EEXIST;585 586	EXT4_I(inode)->i_fc_lblk_len = 0;587 588	return 0;589}590 591void ext4_fc_track_inode(handle_t *handle, struct inode *inode)592{593	int ret;594 595	if (S_ISDIR(inode->i_mode))596		return;597 598	if (ext4_fc_disabled(inode->i_sb))599		return;600 601	if (ext4_should_journal_data(inode)) {602		ext4_fc_mark_ineligible(inode->i_sb,603					EXT4_FC_REASON_INODE_JOURNAL_DATA, handle);604		return;605	}606 607	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))608		return;609 610	ret = ext4_fc_track_template(handle, inode, __track_inode, NULL, 1);611	trace_ext4_fc_track_inode(handle, inode, ret);612}613 614struct __track_range_args {615	ext4_lblk_t start, end;616};617 618/* __track_fn for tracking data updates */619static int __track_range(handle_t *handle, struct inode *inode, void *arg,620			 bool update)621{622	struct ext4_inode_info *ei = EXT4_I(inode);623	ext4_lblk_t oldstart;624	struct __track_range_args *__arg =625		(struct __track_range_args *)arg;626 627	if (inode->i_ino < EXT4_FIRST_INO(inode->i_sb)) {628		ext4_debug("Special inode %ld being modified\n", inode->i_ino);629		return -ECANCELED;630	}631 632	oldstart = ei->i_fc_lblk_start;633 634	if (update && ei->i_fc_lblk_len > 0) {635		ei->i_fc_lblk_start = min(ei->i_fc_lblk_start, __arg->start);636		ei->i_fc_lblk_len =637			max(oldstart + ei->i_fc_lblk_len - 1, __arg->end) -638				ei->i_fc_lblk_start + 1;639	} else {640		ei->i_fc_lblk_start = __arg->start;641		ei->i_fc_lblk_len = __arg->end - __arg->start + 1;642	}643 644	return 0;645}646 647void ext4_fc_track_range(handle_t *handle, struct inode *inode, ext4_lblk_t start,648			 ext4_lblk_t end)649{650	struct __track_range_args args;651	int ret;652 653	if (S_ISDIR(inode->i_mode))654		return;655 656	if (ext4_fc_disabled(inode->i_sb))657		return;658 659	if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_INELIGIBLE))660		return;661 662	if (ext4_has_inline_data(inode)) {663		ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR,664					handle);665		return;666	}667 668	args.start = start;669	args.end = end;670 671	ret = ext4_fc_track_template(handle, inode,  __track_range, &args, 1);672 673	trace_ext4_fc_track_range(handle, inode, start, end, ret);674}675 676static void ext4_fc_submit_bh(struct super_block *sb, bool is_tail)677{678	blk_opf_t write_flags = REQ_SYNC;679	struct buffer_head *bh = EXT4_SB(sb)->s_fc_bh;680 681	/* Add REQ_FUA | REQ_PREFLUSH only its tail */682	if (test_opt(sb, BARRIER) && is_tail)683		write_flags |= REQ_FUA | REQ_PREFLUSH;684	lock_buffer(bh);685	set_buffer_dirty(bh);686	set_buffer_uptodate(bh);687	bh->b_end_io = ext4_end_buffer_io_sync;688	submit_bh(REQ_OP_WRITE | write_flags, bh);689	EXT4_SB(sb)->s_fc_bh = NULL;690}691 692/* Ext4 commit path routines */693 694/*695 * Allocate len bytes on a fast commit buffer.696 *697 * During the commit time this function is used to manage fast commit698 * block space. We don't split a fast commit log onto different699 * blocks. So this function makes sure that if there's not enough space700 * on the current block, the remaining space in the current block is701 * marked as unused by adding EXT4_FC_TAG_PAD tag. In that case,702 * new block is from jbd2 and CRC is updated to reflect the padding703 * we added.704 */705static u8 *ext4_fc_reserve_space(struct super_block *sb, int len, u32 *crc)706{707	struct ext4_fc_tl tl;708	struct ext4_sb_info *sbi = EXT4_SB(sb);709	struct buffer_head *bh;710	int bsize = sbi->s_journal->j_blocksize;711	int ret, off = sbi->s_fc_bytes % bsize;712	int remaining;713	u8 *dst;714 715	/*716	 * If 'len' is too long to fit in any block alongside a PAD tlv, then we717	 * cannot fulfill the request.718	 */719	if (len > bsize - EXT4_FC_TAG_BASE_LEN)720		return NULL;721 722	if (!sbi->s_fc_bh) {723		ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);724		if (ret)725			return NULL;726		sbi->s_fc_bh = bh;727	}728	dst = sbi->s_fc_bh->b_data + off;729 730	/*731	 * Allocate the bytes in the current block if we can do so while still732	 * leaving enough space for a PAD tlv.733	 */734	remaining = bsize - EXT4_FC_TAG_BASE_LEN - off;735	if (len <= remaining) {736		sbi->s_fc_bytes += len;737		return dst;738	}739 740	/*741	 * Else, terminate the current block with a PAD tlv, then allocate a new742	 * block and allocate the bytes at the start of that new block.743	 */744 745	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_PAD);746	tl.fc_len = cpu_to_le16(remaining);747	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);748	memset(dst + EXT4_FC_TAG_BASE_LEN, 0, remaining);749	*crc = ext4_chksum(sbi, *crc, sbi->s_fc_bh->b_data, bsize);750 751	ext4_fc_submit_bh(sb, false);752 753	ret = jbd2_fc_get_buf(EXT4_SB(sb)->s_journal, &bh);754	if (ret)755		return NULL;756	sbi->s_fc_bh = bh;757	sbi->s_fc_bytes += bsize - off + len;758	return sbi->s_fc_bh->b_data;759}760 761/*762 * Complete a fast commit by writing tail tag.763 *764 * Writing tail tag marks the end of a fast commit. In order to guarantee765 * atomicity, after writing tail tag, even if there's space remaining766 * in the block, next commit shouldn't use it. That's why tail tag767 * has the length as that of the remaining space on the block.768 */769static int ext4_fc_write_tail(struct super_block *sb, u32 crc)770{771	struct ext4_sb_info *sbi = EXT4_SB(sb);772	struct ext4_fc_tl tl;773	struct ext4_fc_tail tail;774	int off, bsize = sbi->s_journal->j_blocksize;775	u8 *dst;776 777	/*778	 * ext4_fc_reserve_space takes care of allocating an extra block if779	 * there's no enough space on this block for accommodating this tail.780	 */781	dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + sizeof(tail), &crc);782	if (!dst)783		return -ENOSPC;784 785	off = sbi->s_fc_bytes % bsize;786 787	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_TAIL);788	tl.fc_len = cpu_to_le16(bsize - off + sizeof(struct ext4_fc_tail));789	sbi->s_fc_bytes = round_up(sbi->s_fc_bytes, bsize);790 791	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);792	dst += EXT4_FC_TAG_BASE_LEN;793	tail.fc_tid = cpu_to_le32(sbi->s_journal->j_running_transaction->t_tid);794	memcpy(dst, &tail.fc_tid, sizeof(tail.fc_tid));795	dst += sizeof(tail.fc_tid);796	crc = ext4_chksum(sbi, crc, sbi->s_fc_bh->b_data,797			  dst - (u8 *)sbi->s_fc_bh->b_data);798	tail.fc_crc = cpu_to_le32(crc);799	memcpy(dst, &tail.fc_crc, sizeof(tail.fc_crc));800	dst += sizeof(tail.fc_crc);801	memset(dst, 0, bsize - off); /* Don't leak uninitialized memory. */802 803	ext4_fc_submit_bh(sb, true);804 805	return 0;806}807 808/*809 * Adds tag, length, value and updates CRC. Returns true if tlv was added.810 * Returns false if there's not enough space.811 */812static bool ext4_fc_add_tlv(struct super_block *sb, u16 tag, u16 len, u8 *val,813			   u32 *crc)814{815	struct ext4_fc_tl tl;816	u8 *dst;817 818	dst = ext4_fc_reserve_space(sb, EXT4_FC_TAG_BASE_LEN + len, crc);819	if (!dst)820		return false;821 822	tl.fc_tag = cpu_to_le16(tag);823	tl.fc_len = cpu_to_le16(len);824 825	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);826	memcpy(dst + EXT4_FC_TAG_BASE_LEN, val, len);827 828	return true;829}830 831/* Same as above, but adds dentry tlv. */832static bool ext4_fc_add_dentry_tlv(struct super_block *sb, u32 *crc,833				   struct ext4_fc_dentry_update *fc_dentry)834{835	struct ext4_fc_dentry_info fcd;836	struct ext4_fc_tl tl;837	int dlen = fc_dentry->fcd_name.len;838	u8 *dst = ext4_fc_reserve_space(sb,839			EXT4_FC_TAG_BASE_LEN + sizeof(fcd) + dlen, crc);840 841	if (!dst)842		return false;843 844	fcd.fc_parent_ino = cpu_to_le32(fc_dentry->fcd_parent);845	fcd.fc_ino = cpu_to_le32(fc_dentry->fcd_ino);846	tl.fc_tag = cpu_to_le16(fc_dentry->fcd_op);847	tl.fc_len = cpu_to_le16(sizeof(fcd) + dlen);848	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);849	dst += EXT4_FC_TAG_BASE_LEN;850	memcpy(dst, &fcd, sizeof(fcd));851	dst += sizeof(fcd);852	memcpy(dst, fc_dentry->fcd_name.name, dlen);853 854	return true;855}856 857/*858 * Writes inode in the fast commit space under TLV with tag @tag.859 * Returns 0 on success, error on failure.860 */861static int ext4_fc_write_inode(struct inode *inode, u32 *crc)862{863	struct ext4_inode_info *ei = EXT4_I(inode);864	int inode_len = EXT4_GOOD_OLD_INODE_SIZE;865	int ret;866	struct ext4_iloc iloc;867	struct ext4_fc_inode fc_inode;868	struct ext4_fc_tl tl;869	u8 *dst;870 871	ret = ext4_get_inode_loc(inode, &iloc);872	if (ret)873		return ret;874 875	if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))876		inode_len = EXT4_INODE_SIZE(inode->i_sb);877	else if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE)878		inode_len += ei->i_extra_isize;879 880	fc_inode.fc_ino = cpu_to_le32(inode->i_ino);881	tl.fc_tag = cpu_to_le16(EXT4_FC_TAG_INODE);882	tl.fc_len = cpu_to_le16(inode_len + sizeof(fc_inode.fc_ino));883 884	ret = -ECANCELED;885	dst = ext4_fc_reserve_space(inode->i_sb,886		EXT4_FC_TAG_BASE_LEN + inode_len + sizeof(fc_inode.fc_ino), crc);887	if (!dst)888		goto err;889 890	memcpy(dst, &tl, EXT4_FC_TAG_BASE_LEN);891	dst += EXT4_FC_TAG_BASE_LEN;892	memcpy(dst, &fc_inode, sizeof(fc_inode));893	dst += sizeof(fc_inode);894	memcpy(dst, (u8 *)ext4_raw_inode(&iloc), inode_len);895	ret = 0;896err:897	brelse(iloc.bh);898	return ret;899}900 901/*902 * Writes updated data ranges for the inode in question. Updates CRC.903 * Returns 0 on success, error otherwise.904 */905static int ext4_fc_write_inode_data(struct inode *inode, u32 *crc)906{907	ext4_lblk_t old_blk_size, cur_lblk_off, new_blk_size;908	struct ext4_inode_info *ei = EXT4_I(inode);909	struct ext4_map_blocks map;910	struct ext4_fc_add_range fc_ext;911	struct ext4_fc_del_range lrange;912	struct ext4_extent *ex;913	int ret;914 915	mutex_lock(&ei->i_fc_lock);916	if (ei->i_fc_lblk_len == 0) {917		mutex_unlock(&ei->i_fc_lock);918		return 0;919	}920	old_blk_size = ei->i_fc_lblk_start;921	new_blk_size = ei->i_fc_lblk_start + ei->i_fc_lblk_len - 1;922	ei->i_fc_lblk_len = 0;923	mutex_unlock(&ei->i_fc_lock);924 925	cur_lblk_off = old_blk_size;926	ext4_debug("will try writing %d to %d for inode %ld\n",927		   cur_lblk_off, new_blk_size, inode->i_ino);928 929	while (cur_lblk_off <= new_blk_size) {930		map.m_lblk = cur_lblk_off;931		map.m_len = new_blk_size - cur_lblk_off + 1;932		ret = ext4_map_blocks(NULL, inode, &map, 0);933		if (ret < 0)934			return -ECANCELED;935 936		if (map.m_len == 0) {937			cur_lblk_off++;938			continue;939		}940 941		if (ret == 0) {942			lrange.fc_ino = cpu_to_le32(inode->i_ino);943			lrange.fc_lblk = cpu_to_le32(map.m_lblk);944			lrange.fc_len = cpu_to_le32(map.m_len);945			if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_DEL_RANGE,946					    sizeof(lrange), (u8 *)&lrange, crc))947				return -ENOSPC;948		} else {949			unsigned int max = (map.m_flags & EXT4_MAP_UNWRITTEN) ?950				EXT_UNWRITTEN_MAX_LEN : EXT_INIT_MAX_LEN;951 952			/* Limit the number of blocks in one extent */953			map.m_len = min(max, map.m_len);954 955			fc_ext.fc_ino = cpu_to_le32(inode->i_ino);956			ex = (struct ext4_extent *)&fc_ext.fc_ex;957			ex->ee_block = cpu_to_le32(map.m_lblk);958			ex->ee_len = cpu_to_le16(map.m_len);959			ext4_ext_store_pblock(ex, map.m_pblk);960			if (map.m_flags & EXT4_MAP_UNWRITTEN)961				ext4_ext_mark_unwritten(ex);962			else963				ext4_ext_mark_initialized(ex);964			if (!ext4_fc_add_tlv(inode->i_sb, EXT4_FC_TAG_ADD_RANGE,965					    sizeof(fc_ext), (u8 *)&fc_ext, crc))966				return -ENOSPC;967		}968 969		cur_lblk_off += map.m_len;970	}971 972	return 0;973}974 975 976/* Submit data for all the fast commit inodes */977static int ext4_fc_submit_inode_data_all(journal_t *journal)978{979	struct super_block *sb = journal->j_private;980	struct ext4_sb_info *sbi = EXT4_SB(sb);981	struct ext4_inode_info *ei;982	int ret = 0;983 984	spin_lock(&sbi->s_fc_lock);985	list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {986		ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);987		while (atomic_read(&ei->i_fc_updates)) {988			DEFINE_WAIT(wait);989 990			prepare_to_wait(&ei->i_fc_wait, &wait,991						TASK_UNINTERRUPTIBLE);992			if (atomic_read(&ei->i_fc_updates)) {993				spin_unlock(&sbi->s_fc_lock);994				schedule();995				spin_lock(&sbi->s_fc_lock);996			}997			finish_wait(&ei->i_fc_wait, &wait);998		}999		spin_unlock(&sbi->s_fc_lock);1000		ret = jbd2_submit_inode_data(journal, ei->jinode);1001		if (ret)1002			return ret;1003		spin_lock(&sbi->s_fc_lock);1004	}1005	spin_unlock(&sbi->s_fc_lock);1006 1007	return ret;1008}1009 1010/* Wait for completion of data for all the fast commit inodes */1011static int ext4_fc_wait_inode_data_all(journal_t *journal)1012{1013	struct super_block *sb = journal->j_private;1014	struct ext4_sb_info *sbi = EXT4_SB(sb);1015	struct ext4_inode_info *pos, *n;1016	int ret = 0;1017 1018	spin_lock(&sbi->s_fc_lock);1019	list_for_each_entry_safe(pos, n, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {1020		if (!ext4_test_inode_state(&pos->vfs_inode,1021					   EXT4_STATE_FC_COMMITTING))1022			continue;1023		spin_unlock(&sbi->s_fc_lock);1024 1025		ret = jbd2_wait_inode_data(journal, pos->jinode);1026		if (ret)1027			return ret;1028		spin_lock(&sbi->s_fc_lock);1029	}1030	spin_unlock(&sbi->s_fc_lock);1031 1032	return 0;1033}1034 1035/* Commit all the directory entry updates */1036static int ext4_fc_commit_dentry_updates(journal_t *journal, u32 *crc)1037__acquires(&sbi->s_fc_lock)1038__releases(&sbi->s_fc_lock)1039{1040	struct super_block *sb = journal->j_private;1041	struct ext4_sb_info *sbi = EXT4_SB(sb);1042	struct ext4_fc_dentry_update *fc_dentry, *fc_dentry_n;1043	struct inode *inode;1044	struct ext4_inode_info *ei;1045	int ret;1046 1047	if (list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN]))1048		return 0;1049	list_for_each_entry_safe(fc_dentry, fc_dentry_n,1050				 &sbi->s_fc_dentry_q[FC_Q_MAIN], fcd_list) {1051		if (fc_dentry->fcd_op != EXT4_FC_TAG_CREAT) {1052			spin_unlock(&sbi->s_fc_lock);1053			if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {1054				ret = -ENOSPC;1055				goto lock_and_exit;1056			}1057			spin_lock(&sbi->s_fc_lock);1058			continue;1059		}1060		/*1061		 * With fcd_dilist we need not loop in sbi->s_fc_q to get the1062		 * corresponding inode pointer1063		 */1064		WARN_ON(list_empty(&fc_dentry->fcd_dilist));1065		ei = list_first_entry(&fc_dentry->fcd_dilist,1066				struct ext4_inode_info, i_fc_dilist);1067		inode = &ei->vfs_inode;1068		WARN_ON(inode->i_ino != fc_dentry->fcd_ino);1069 1070		spin_unlock(&sbi->s_fc_lock);1071 1072		/*1073		 * We first write the inode and then the create dirent. This1074		 * allows the recovery code to create an unnamed inode first1075		 * and then link it to a directory entry. This allows us1076		 * to use namei.c routines almost as is and simplifies1077		 * the recovery code.1078		 */1079		ret = ext4_fc_write_inode(inode, crc);1080		if (ret)1081			goto lock_and_exit;1082 1083		ret = ext4_fc_write_inode_data(inode, crc);1084		if (ret)1085			goto lock_and_exit;1086 1087		if (!ext4_fc_add_dentry_tlv(sb, crc, fc_dentry)) {1088			ret = -ENOSPC;1089			goto lock_and_exit;1090		}1091 1092		spin_lock(&sbi->s_fc_lock);1093	}1094	return 0;1095lock_and_exit:1096	spin_lock(&sbi->s_fc_lock);1097	return ret;1098}1099 1100static int ext4_fc_perform_commit(journal_t *journal)1101{1102	struct super_block *sb = journal->j_private;1103	struct ext4_sb_info *sbi = EXT4_SB(sb);1104	struct ext4_inode_info *iter;1105	struct ext4_fc_head head;1106	struct inode *inode;1107	struct blk_plug plug;1108	int ret = 0;1109	u32 crc = 0;1110 1111	ret = ext4_fc_submit_inode_data_all(journal);1112	if (ret)1113		return ret;1114 1115	ret = ext4_fc_wait_inode_data_all(journal);1116	if (ret)1117		return ret;1118 1119	/*1120	 * If file system device is different from journal device, issue a cache1121	 * flush before we start writing fast commit blocks.1122	 */1123	if (journal->j_fs_dev != journal->j_dev)1124		blkdev_issue_flush(journal->j_fs_dev);1125 1126	blk_start_plug(&plug);1127	if (sbi->s_fc_bytes == 0) {1128		/*1129		 * Add a head tag only if this is the first fast commit1130		 * in this TID.1131		 */1132		head.fc_features = cpu_to_le32(EXT4_FC_SUPPORTED_FEATURES);1133		head.fc_tid = cpu_to_le32(1134			sbi->s_journal->j_running_transaction->t_tid);1135		if (!ext4_fc_add_tlv(sb, EXT4_FC_TAG_HEAD, sizeof(head),1136			(u8 *)&head, &crc)) {1137			ret = -ENOSPC;1138			goto out;1139		}1140	}1141 1142	spin_lock(&sbi->s_fc_lock);1143	ret = ext4_fc_commit_dentry_updates(journal, &crc);1144	if (ret) {1145		spin_unlock(&sbi->s_fc_lock);1146		goto out;1147	}1148 1149	list_for_each_entry(iter, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {1150		inode = &iter->vfs_inode;1151		if (!ext4_test_inode_state(inode, EXT4_STATE_FC_COMMITTING))1152			continue;1153 1154		spin_unlock(&sbi->s_fc_lock);1155		ret = ext4_fc_write_inode_data(inode, &crc);1156		if (ret)1157			goto out;1158		ret = ext4_fc_write_inode(inode, &crc);1159		if (ret)1160			goto out;1161		spin_lock(&sbi->s_fc_lock);1162	}1163	spin_unlock(&sbi->s_fc_lock);1164 1165	ret = ext4_fc_write_tail(sb, crc);1166 1167out:1168	blk_finish_plug(&plug);1169	return ret;1170}1171 1172static void ext4_fc_update_stats(struct super_block *sb, int status,1173				 u64 commit_time, int nblks, tid_t commit_tid)1174{1175	struct ext4_fc_stats *stats = &EXT4_SB(sb)->s_fc_stats;1176 1177	ext4_debug("Fast commit ended with status = %d for tid %u",1178			status, commit_tid);1179	if (status == EXT4_FC_STATUS_OK) {1180		stats->fc_num_commits++;1181		stats->fc_numblks += nblks;1182		if (likely(stats->s_fc_avg_commit_time))1183			stats->s_fc_avg_commit_time =1184				(commit_time +1185				 stats->s_fc_avg_commit_time * 3) / 4;1186		else1187			stats->s_fc_avg_commit_time = commit_time;1188	} else if (status == EXT4_FC_STATUS_FAILED ||1189		   status == EXT4_FC_STATUS_INELIGIBLE) {1190		if (status == EXT4_FC_STATUS_FAILED)1191			stats->fc_failed_commits++;1192		stats->fc_ineligible_commits++;1193	} else {1194		stats->fc_skipped_commits++;1195	}1196	trace_ext4_fc_commit_stop(sb, nblks, status, commit_tid);1197}1198 1199/*1200 * The main commit entry point. Performs a fast commit for transaction1201 * commit_tid if needed. If it's not possible to perform a fast commit1202 * due to various reasons, we fall back to full commit. Returns 01203 * on success, error otherwise.1204 */1205int ext4_fc_commit(journal_t *journal, tid_t commit_tid)1206{1207	struct super_block *sb = journal->j_private;1208	struct ext4_sb_info *sbi = EXT4_SB(sb);1209	int nblks = 0, ret, bsize = journal->j_blocksize;1210	int subtid = atomic_read(&sbi->s_fc_subtid);1211	int status = EXT4_FC_STATUS_OK, fc_bufs_before = 0;1212	ktime_t start_time, commit_time;1213 1214	if (!test_opt2(sb, JOURNAL_FAST_COMMIT))1215		return jbd2_complete_transaction(journal, commit_tid);1216 1217	trace_ext4_fc_commit_start(sb, commit_tid);1218 1219	start_time = ktime_get();1220 1221restart_fc:1222	ret = jbd2_fc_begin_commit(journal, commit_tid);1223	if (ret == -EALREADY) {1224		/* There was an ongoing commit, check if we need to restart */1225		if (atomic_read(&sbi->s_fc_subtid) <= subtid &&1226		    tid_gt(commit_tid, journal->j_commit_sequence))1227			goto restart_fc;1228		ext4_fc_update_stats(sb, EXT4_FC_STATUS_SKIPPED, 0, 0,1229				commit_tid);1230		return 0;1231	} else if (ret) {1232		/*1233		 * Commit couldn't start. Just update stats and perform a1234		 * full commit.1235		 */1236		ext4_fc_update_stats(sb, EXT4_FC_STATUS_FAILED, 0, 0,1237				commit_tid);1238		return jbd2_complete_transaction(journal, commit_tid);1239	}1240 1241	/*1242	 * After establishing journal barrier via jbd2_fc_begin_commit(), check1243	 * if we are fast commit ineligible.1244	 */1245	if (ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE)) {1246		status = EXT4_FC_STATUS_INELIGIBLE;1247		goto fallback;1248	}1249 1250	fc_bufs_before = (sbi->s_fc_bytes + bsize - 1) / bsize;1251	ret = ext4_fc_perform_commit(journal);1252	if (ret < 0) {1253		status = EXT4_FC_STATUS_FAILED;1254		goto fallback;1255	}1256	nblks = (sbi->s_fc_bytes + bsize - 1) / bsize - fc_bufs_before;1257	ret = jbd2_fc_wait_bufs(journal, nblks);1258	if (ret < 0) {1259		status = EXT4_FC_STATUS_FAILED;1260		goto fallback;1261	}1262	atomic_inc(&sbi->s_fc_subtid);1263	ret = jbd2_fc_end_commit(journal);1264	/*1265	 * weight the commit time higher than the average time so we1266	 * don't react too strongly to vast changes in the commit time1267	 */1268	commit_time = ktime_to_ns(ktime_sub(ktime_get(), start_time));1269	ext4_fc_update_stats(sb, status, commit_time, nblks, commit_tid);1270	return ret;1271 1272fallback:1273	ret = jbd2_fc_end_commit_fallback(journal);1274	ext4_fc_update_stats(sb, status, 0, 0, commit_tid);1275	return ret;1276}1277 1278/*1279 * Fast commit cleanup routine. This is called after every fast commit and1280 * full commit. full is true if we are called after a full commit.1281 */1282static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)1283{1284	struct super_block *sb = journal->j_private;1285	struct ext4_sb_info *sbi = EXT4_SB(sb);1286	struct ext4_inode_info *iter, *iter_n;1287	struct ext4_fc_dentry_update *fc_dentry;1288 1289	if (full && sbi->s_fc_bh)1290		sbi->s_fc_bh = NULL;1291 1292	trace_ext4_fc_cleanup(journal, full, tid);1293	jbd2_fc_release_bufs(journal);1294 1295	spin_lock(&sbi->s_fc_lock);1296	list_for_each_entry_safe(iter, iter_n, &sbi->s_fc_q[FC_Q_MAIN],1297				 i_fc_list) {1298		list_del_init(&iter->i_fc_list);1299		ext4_clear_inode_state(&iter->vfs_inode,1300				       EXT4_STATE_FC_COMMITTING);1301		if (tid_geq(tid, iter->i_sync_tid)) {1302			ext4_fc_reset_inode(&iter->vfs_inode);1303		} else if (full) {1304			/*1305			 * We are called after a full commit, inode has been1306			 * modified while the commit was running. Re-enqueue1307			 * the inode into STAGING, which will then be splice1308			 * back into MAIN. This cannot happen during1309			 * fastcommit because the journal is locked all the1310			 * time in that case (and tid doesn't increase so1311			 * tid check above isn't reliable).1312			 */1313			list_add_tail(&EXT4_I(&iter->vfs_inode)->i_fc_list,1314				      &sbi->s_fc_q[FC_Q_STAGING]);1315		}1316		/* Make sure EXT4_STATE_FC_COMMITTING bit is clear */1317		smp_mb();1318#if (BITS_PER_LONG < 64)1319		wake_up_bit(&iter->i_state_flags, EXT4_STATE_FC_COMMITTING);1320#else1321		wake_up_bit(&iter->i_flags, EXT4_STATE_FC_COMMITTING);1322#endif1323	}1324 1325	while (!list_empty(&sbi->s_fc_dentry_q[FC_Q_MAIN])) {1326		fc_dentry = list_first_entry(&sbi->s_fc_dentry_q[FC_Q_MAIN],1327					     struct ext4_fc_dentry_update,1328					     fcd_list);1329		list_del_init(&fc_dentry->fcd_list);1330		list_del_init(&fc_dentry->fcd_dilist);1331		spin_unlock(&sbi->s_fc_lock);1332 1333		if (fc_dentry->fcd_name.name &&1334			fc_dentry->fcd_name.len > DNAME_INLINE_LEN)1335			kfree(fc_dentry->fcd_name.name);1336		kmem_cache_free(ext4_fc_dentry_cachep, fc_dentry);1337		spin_lock(&sbi->s_fc_lock);1338	}1339 1340	list_splice_init(&sbi->s_fc_dentry_q[FC_Q_STAGING],1341				&sbi->s_fc_dentry_q[FC_Q_MAIN]);1342	list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],1343				&sbi->s_fc_q[FC_Q_MAIN]);1344 1345	if (tid_geq(tid, sbi->s_fc_ineligible_tid)) {1346		sbi->s_fc_ineligible_tid = 0;1347		ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);1348	}1349 1350	if (full)1351		sbi->s_fc_bytes = 0;1352	spin_unlock(&sbi->s_fc_lock);1353	trace_ext4_fc_stats(sb);1354}1355 1356/* Ext4 Replay Path Routines */1357 1358/* Helper struct for dentry replay routines */1359struct dentry_info_args {1360	int parent_ino, dname_len, ino, inode_len;1361	char *dname;1362};1363 1364/* Same as struct ext4_fc_tl, but uses native endianness fields */1365struct ext4_fc_tl_mem {1366	u16 fc_tag;1367	u16 fc_len;1368};1369 1370static inline void tl_to_darg(struct dentry_info_args *darg,1371			      struct ext4_fc_tl_mem *tl, u8 *val)1372{1373	struct ext4_fc_dentry_info fcd;1374 1375	memcpy(&fcd, val, sizeof(fcd));1376 1377	darg->parent_ino = le32_to_cpu(fcd.fc_parent_ino);1378	darg->ino = le32_to_cpu(fcd.fc_ino);1379	darg->dname = val + offsetof(struct ext4_fc_dentry_info, fc_dname);1380	darg->dname_len = tl->fc_len - sizeof(struct ext4_fc_dentry_info);1381}1382 1383static inline void ext4_fc_get_tl(struct ext4_fc_tl_mem *tl, u8 *val)1384{1385	struct ext4_fc_tl tl_disk;1386 1387	memcpy(&tl_disk, val, EXT4_FC_TAG_BASE_LEN);1388	tl->fc_len = le16_to_cpu(tl_disk.fc_len);1389	tl->fc_tag = le16_to_cpu(tl_disk.fc_tag);1390}1391 1392/* Unlink replay function */1393static int ext4_fc_replay_unlink(struct super_block *sb,1394				 struct ext4_fc_tl_mem *tl, u8 *val)1395{1396	struct inode *inode, *old_parent;1397	struct qstr entry;1398	struct dentry_info_args darg;1399	int ret = 0;1400 1401	tl_to_darg(&darg, tl, val);1402 1403	trace_ext4_fc_replay(sb, EXT4_FC_TAG_UNLINK, darg.ino,1404			darg.parent_ino, darg.dname_len);1405 1406	entry.name = darg.dname;1407	entry.len = darg.dname_len;1408	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);1409 1410	if (IS_ERR(inode)) {1411		ext4_debug("Inode %d not found", darg.ino);1412		return 0;1413	}1414 1415	old_parent = ext4_iget(sb, darg.parent_ino,1416				EXT4_IGET_NORMAL);1417	if (IS_ERR(old_parent)) {1418		ext4_debug("Dir with inode %d not found", darg.parent_ino);1419		iput(inode);1420		return 0;1421	}1422 1423	ret = __ext4_unlink(old_parent, &entry, inode, NULL);1424	/* -ENOENT ok coz it might not exist anymore. */1425	if (ret == -ENOENT)1426		ret = 0;1427	iput(old_parent);1428	iput(inode);1429	return ret;1430}1431 1432static int ext4_fc_replay_link_internal(struct super_block *sb,1433				struct dentry_info_args *darg,1434				struct inode *inode)1435{1436	struct inode *dir = NULL;1437	struct dentry *dentry_dir = NULL, *dentry_inode = NULL;1438	struct qstr qstr_dname = QSTR_INIT(darg->dname, darg->dname_len);1439	int ret = 0;1440 1441	dir = ext4_iget(sb, darg->parent_ino, EXT4_IGET_NORMAL);1442	if (IS_ERR(dir)) {1443		ext4_debug("Dir with inode %d not found.", darg->parent_ino);1444		dir = NULL;1445		goto out;1446	}1447 1448	dentry_dir = d_obtain_alias(dir);1449	if (IS_ERR(dentry_dir)) {1450		ext4_debug("Failed to obtain dentry");1451		dentry_dir = NULL;1452		goto out;1453	}1454 1455	dentry_inode = d_alloc(dentry_dir, &qstr_dname);1456	if (!dentry_inode) {1457		ext4_debug("Inode dentry not created.");1458		ret = -ENOMEM;1459		goto out;1460	}1461 1462	ret = __ext4_link(dir, inode, dentry_inode);1463	/*1464	 * It's possible that link already existed since data blocks1465	 * for the dir in question got persisted before we crashed OR1466	 * we replayed this tag and crashed before the entire replay1467	 * could complete.1468	 */1469	if (ret && ret != -EEXIST) {1470		ext4_debug("Failed to link\n");1471		goto out;1472	}1473 1474	ret = 0;1475out:1476	if (dentry_dir) {1477		d_drop(dentry_dir);1478		dput(dentry_dir);1479	} else if (dir) {1480		iput(dir);1481	}1482	if (dentry_inode) {1483		d_drop(dentry_inode);1484		dput(dentry_inode);1485	}1486 1487	return ret;1488}1489 1490/* Link replay function */1491static int ext4_fc_replay_link(struct super_block *sb,1492			       struct ext4_fc_tl_mem *tl, u8 *val)1493{1494	struct inode *inode;1495	struct dentry_info_args darg;1496	int ret = 0;1497 1498	tl_to_darg(&darg, tl, val);1499	trace_ext4_fc_replay(sb, EXT4_FC_TAG_LINK, darg.ino,1500			darg.parent_ino, darg.dname_len);1501 1502	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);1503	if (IS_ERR(inode)) {1504		ext4_debug("Inode not found.");1505		return 0;1506	}1507 1508	ret = ext4_fc_replay_link_internal(sb, &darg, inode);1509	iput(inode);1510	return ret;1511}1512 1513/*1514 * Record all the modified inodes during replay. We use this later to setup1515 * block bitmaps correctly.1516 */1517static int ext4_fc_record_modified_inode(struct super_block *sb, int ino)1518{1519	struct ext4_fc_replay_state *state;1520	int i;1521 1522	state = &EXT4_SB(sb)->s_fc_replay_state;1523	for (i = 0; i < state->fc_modified_inodes_used; i++)1524		if (state->fc_modified_inodes[i] == ino)1525			return 0;1526	if (state->fc_modified_inodes_used == state->fc_modified_inodes_size) {1527		int *fc_modified_inodes;1528 1529		fc_modified_inodes = krealloc(state->fc_modified_inodes,1530				sizeof(int) * (state->fc_modified_inodes_size +1531				EXT4_FC_REPLAY_REALLOC_INCREMENT),1532				GFP_KERNEL);1533		if (!fc_modified_inodes)1534			return -ENOMEM;1535		state->fc_modified_inodes = fc_modified_inodes;1536		state->fc_modified_inodes_size +=1537			EXT4_FC_REPLAY_REALLOC_INCREMENT;1538	}1539	state->fc_modified_inodes[state->fc_modified_inodes_used++] = ino;1540	return 0;1541}1542 1543/*1544 * Inode replay function1545 */1546static int ext4_fc_replay_inode(struct super_block *sb,1547				struct ext4_fc_tl_mem *tl, u8 *val)1548{1549	struct ext4_fc_inode fc_inode;1550	struct ext4_inode *raw_inode;1551	struct ext4_inode *raw_fc_inode;1552	struct inode *inode = NULL;1553	struct ext4_iloc iloc;1554	int inode_len, ino, ret, tag = tl->fc_tag;1555	struct ext4_extent_header *eh;1556	size_t off_gen = offsetof(struct ext4_inode, i_generation);1557 1558	memcpy(&fc_inode, val, sizeof(fc_inode));1559 1560	ino = le32_to_cpu(fc_inode.fc_ino);1561	trace_ext4_fc_replay(sb, tag, ino, 0, 0);1562 1563	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);1564	if (!IS_ERR(inode)) {1565		ext4_ext_clear_bb(inode);1566		iput(inode);1567	}1568	inode = NULL;1569 1570	ret = ext4_fc_record_modified_inode(sb, ino);1571	if (ret)1572		goto out;1573 1574	raw_fc_inode = (struct ext4_inode *)1575		(val + offsetof(struct ext4_fc_inode, fc_raw_inode));1576	ret = ext4_get_fc_inode_loc(sb, ino, &iloc);1577	if (ret)1578		goto out;1579 1580	inode_len = tl->fc_len - sizeof(struct ext4_fc_inode);1581	raw_inode = ext4_raw_inode(&iloc);1582 1583	memcpy(raw_inode, raw_fc_inode, offsetof(struct ext4_inode, i_block));1584	memcpy((u8 *)raw_inode + off_gen, (u8 *)raw_fc_inode + off_gen,1585	       inode_len - off_gen);1586	if (le32_to_cpu(raw_inode->i_flags) & EXT4_EXTENTS_FL) {1587		eh = (struct ext4_extent_header *)(&raw_inode->i_block[0]);1588		if (eh->eh_magic != EXT4_EXT_MAGIC) {1589			memset(eh, 0, sizeof(*eh));1590			eh->eh_magic = EXT4_EXT_MAGIC;1591			eh->eh_max = cpu_to_le16(1592				(sizeof(raw_inode->i_block) -1593				 sizeof(struct ext4_extent_header))1594				 / sizeof(struct ext4_extent));1595		}1596	} else if (le32_to_cpu(raw_inode->i_flags) & EXT4_INLINE_DATA_FL) {1597		memcpy(raw_inode->i_block, raw_fc_inode->i_block,1598			sizeof(raw_inode->i_block));1599	}1600 1601	/* Immediately update the inode on disk. */1602	ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);1603	if (ret)1604		goto out;1605	ret = sync_dirty_buffer(iloc.bh);1606	if (ret)1607		goto out;1608	ret = ext4_mark_inode_used(sb, ino);1609	if (ret)1610		goto out;1611 1612	/* Given that we just wrote the inode on disk, this SHOULD succeed. */1613	inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);1614	if (IS_ERR(inode)) {1615		ext4_debug("Inode not found.");1616		return -EFSCORRUPTED;1617	}1618 1619	/*1620	 * Our allocator could have made different decisions than before1621	 * crashing. This should be fixed but until then, we calculate1622	 * the number of blocks the inode.1623	 */1624	if (!ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))1625		ext4_ext_replay_set_iblocks(inode);1626 1627	inode->i_generation = le32_to_cpu(ext4_raw_inode(&iloc)->i_generation);1628	ext4_reset_inode_seed(inode);1629 1630	ext4_inode_csum_set(inode, ext4_raw_inode(&iloc), EXT4_I(inode));1631	ret = ext4_handle_dirty_metadata(NULL, NULL, iloc.bh);1632	sync_dirty_buffer(iloc.bh);1633	brelse(iloc.bh);1634out:1635	iput(inode);1636	if (!ret)1637		blkdev_issue_flush(sb->s_bdev);1638 1639	return 0;1640}1641 1642/*1643 * Dentry create replay function.1644 *1645 * EXT4_FC_TAG_CREAT is preceded by EXT4_FC_TAG_INODE_FULL. Which means, the1646 * inode for which we are trying to create a dentry here, should already have1647 * been replayed before we start here.1648 */1649static int ext4_fc_replay_create(struct super_block *sb,1650				 struct ext4_fc_tl_mem *tl, u8 *val)1651{1652	int ret = 0;1653	struct inode *inode = NULL;1654	struct inode *dir = NULL;1655	struct dentry_info_args darg;1656 1657	tl_to_darg(&darg, tl, val);1658 1659	trace_ext4_fc_replay(sb, EXT4_FC_TAG_CREAT, darg.ino,1660			darg.parent_ino, darg.dname_len);1661 1662	/* This takes care of update group descriptor and other metadata */1663	ret = ext4_mark_inode_used(sb, darg.ino);1664	if (ret)1665		goto out;1666 1667	inode = ext4_iget(sb, darg.ino, EXT4_IGET_NORMAL);1668	if (IS_ERR(inode)) {1669		ext4_debug("inode %d not found.", darg.ino);1670		inode = NULL;1671		ret = -EINVAL;1672		goto out;1673	}1674 1675	if (S_ISDIR(inode->i_mode)) {1676		/*1677		 * If we are creating a directory, we need to make sure that the1678		 * dot and dot dot dirents are setup properly.1679		 */1680		dir = ext4_iget(sb, darg.parent_ino, EXT4_IGET_NORMAL);1681		if (IS_ERR(dir)) {1682			ext4_debug("Dir %d not found.", darg.ino);1683			goto out;1684		}1685		ret = ext4_init_new_dir(NULL, dir, inode);1686		iput(dir);1687		if (ret) {1688			ret = 0;1689			goto out;1690		}1691	}1692	ret = ext4_fc_replay_link_internal(sb, &darg, inode);1693	if (ret)1694		goto out;1695	set_nlink(inode, 1);1696	ext4_mark_inode_dirty(NULL, inode);1697out:1698	iput(inode);1699	return ret;1700}1701 1702/*1703 * Record physical disk regions which are in use as per fast commit area,1704 * and used by inodes during replay phase. Our simple replay phase1705 * allocator excludes these regions from allocation.1706 */1707int ext4_fc_record_regions(struct super_block *sb, int ino,1708		ext4_lblk_t lblk, ext4_fsblk_t pblk, int len, int replay)1709{1710	struct ext4_fc_replay_state *state;1711	struct ext4_fc_alloc_region *region;1712 1713	state = &EXT4_SB(sb)->s_fc_replay_state;1714	/*1715	 * during replay phase, the fc_regions_valid may not same as1716	 * fc_regions_used, update it when do new additions.1717	 */1718	if (replay && state->fc_regions_used != state->fc_regions_valid)1719		state->fc_regions_used = state->fc_regions_valid;1720	if (state->fc_regions_used == state->fc_regions_size) {1721		struct ext4_fc_alloc_region *fc_regions;1722 1723		fc_regions = krealloc(state->fc_regions,1724				      sizeof(struct ext4_fc_alloc_region) *1725				      (state->fc_regions_size +1726				       EXT4_FC_REPLAY_REALLOC_INCREMENT),1727				      GFP_KERNEL);1728		if (!fc_regions)1729			return -ENOMEM;1730		state->fc_regions_size +=1731			EXT4_FC_REPLAY_REALLOC_INCREMENT;1732		state->fc_regions = fc_regions;1733	}1734	region = &state->fc_regions[state->fc_regions_used++];1735	region->ino = ino;1736	region->lblk = lblk;1737	region->pblk = pblk;1738	region->len = len;1739 1740	if (replay)1741		state->fc_regions_valid++;1742 1743	return 0;1744}1745 1746/* Replay add range tag */1747static int ext4_fc_replay_add_range(struct super_block *sb,1748				    struct ext4_fc_tl_mem *tl, u8 *val)1749{1750	struct ext4_fc_add_range fc_add_ex;1751	struct ext4_extent newex, *ex;1752	struct inode *inode;1753	ext4_lblk_t start, cur;1754	int remaining, len;1755	ext4_fsblk_t start_pblk;1756	struct ext4_map_blocks map;1757	struct ext4_ext_path *path = NULL;1758	int ret;1759 1760	memcpy(&fc_add_ex, val, sizeof(fc_add_ex));1761	ex = (struct ext4_extent *)&fc_add_ex.fc_ex;1762 1763	trace_ext4_fc_replay(sb, EXT4_FC_TAG_ADD_RANGE,1764		le32_to_cpu(fc_add_ex.fc_ino), le32_to_cpu(ex->ee_block),1765		ext4_ext_get_actual_len(ex));1766 1767	inode = ext4_iget(sb, le32_to_cpu(fc_add_ex.fc_ino), EXT4_IGET_NORMAL);1768	if (IS_ERR(inode)) {1769		ext4_debug("Inode not found.");1770		return 0;1771	}1772 1773	ret = ext4_fc_record_modified_inode(sb, inode->i_ino);1774	if (ret)1775		goto out;1776 1777	start = le32_to_cpu(ex->ee_block);1778	start_pblk = ext4_ext_pblock(ex);1779	len = ext4_ext_get_actual_len(ex);1780 1781	cur = start;1782	remaining = len;1783	ext4_debug("ADD_RANGE, lblk %d, pblk %lld, len %d, unwritten %d, inode %ld\n",1784		  start, start_pblk, len, ext4_ext_is_unwritten(ex),1785		  inode->i_ino);1786 1787	while (remaining > 0) {1788		map.m_lblk = cur;1789		map.m_len = remaining;1790		map.m_pblk = 0;1791		ret = ext4_map_blocks(NULL, inode, &map, 0);1792 1793		if (ret < 0)1794			goto out;1795 1796		if (ret == 0) {1797			/* Range is not mapped */1798			path = ext4_find_extent(inode, cur, path, 0);1799			if (IS_ERR(path))1800				goto out;1801			memset(&newex, 0, sizeof(newex));1802			newex.ee_block = cpu_to_le32(cur);1803			ext4_ext_store_pblock(1804				&newex, start_pblk + cur - start);1805			newex.ee_len = cpu_to_le16(map.m_len);1806			if (ext4_ext_is_unwritten(ex))1807				ext4_ext_mark_unwritten(&newex);1808			down_write(&EXT4_I(inode)->i_data_sem);1809			path = ext4_ext_insert_extent(NULL, inode,1810						      path, &newex, 0);1811			up_write((&EXT4_I(inode)->i_data_sem));1812			if (IS_ERR(path))1813				goto out;1814			goto next;1815		}1816 1817		if (start_pblk + cur - start != map.m_pblk) {1818			/*1819			 * Logical to physical mapping changed. This can happen1820			 * if this range was removed and then reallocated to1821			 * map to new physical blocks during a fast commit.1822			 */1823			ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,1824					ext4_ext_is_unwritten(ex),1825					start_pblk + cur - start);1826			if (ret)1827				goto out;1828			/*1829			 * Mark the old blocks as free since they aren't used1830			 * anymore. We maintain an array of all the modified1831			 * inodes. In case these blocks are still used at either1832			 * a different logical range in the same inode or in1833			 * some different inode, we will mark them as allocated1834			 * at the end of the FC replay using our array of1835			 * modified inodes.1836			 */1837			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, false);1838			goto next;1839		}1840 1841		/* Range is mapped and needs a state change */1842		ext4_debug("Converting from %ld to %d %lld",1843				map.m_flags & EXT4_MAP_UNWRITTEN,1844			ext4_ext_is_unwritten(ex), map.m_pblk);1845		ret = ext4_ext_replay_update_ex(inode, cur, map.m_len,1846					ext4_ext_is_unwritten(ex), map.m_pblk);1847		if (ret)1848			goto out;1849		/*1850		 * We may have split the extent tree while toggling the state.1851		 * Try to shrink the extent tree now.1852		 */1853		ext4_ext_replay_shrink_inode(inode, start + len);1854next:1855		cur += map.m_len;1856		remaining -= map.m_len;1857	}1858	ext4_ext_replay_shrink_inode(inode, i_size_read(inode) >>1859					sb->s_blocksize_bits);1860out:1861	ext4_free_ext_path(path);1862	iput(inode);1863	return 0;1864}1865 1866/* Replay DEL_RANGE tag */1867static int1868ext4_fc_replay_del_range(struct super_block *sb,1869			 struct ext4_fc_tl_mem *tl, u8 *val)1870{1871	struct inode *inode;1872	struct ext4_fc_del_range lrange;1873	struct ext4_map_blocks map;1874	ext4_lblk_t cur, remaining;1875	int ret;1876 1877	memcpy(&lrange, val, sizeof(lrange));1878	cur = le32_to_cpu(lrange.fc_lblk);1879	remaining = le32_to_cpu(lrange.fc_len);1880 1881	trace_ext4_fc_replay(sb, EXT4_FC_TAG_DEL_RANGE,1882		le32_to_cpu(lrange.fc_ino), cur, remaining);1883 1884	inode = ext4_iget(sb, le32_to_cpu(lrange.fc_ino), EXT4_IGET_NORMAL);1885	if (IS_ERR(inode)) {1886		ext4_debug("Inode %d not found", le32_to_cpu(lrange.fc_ino));1887		return 0;1888	}1889 1890	ret = ext4_fc_record_modified_inode(sb, inode->i_ino);1891	if (ret)1892		goto out;1893 1894	ext4_debug("DEL_RANGE, inode %ld, lblk %d, len %d\n",1895			inode->i_ino, le32_to_cpu(lrange.fc_lblk),1896			le32_to_cpu(lrange.fc_len));1897	while (remaining > 0) {1898		map.m_lblk = cur;1899		map.m_len = remaining;1900 1901		ret = ext4_map_blocks(NULL, inode, &map, 0);1902		if (ret < 0)1903			goto out;1904		if (ret > 0) {1905			remaining -= ret;1906			cur += ret;1907			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, false);1908		} else {1909			remaining -= map.m_len;1910			cur += map.m_len;1911		}1912	}1913 1914	down_write(&EXT4_I(inode)->i_data_sem);1915	ret = ext4_ext_remove_space(inode, le32_to_cpu(lrange.fc_lblk),1916				le32_to_cpu(lrange.fc_lblk) +1917				le32_to_cpu(lrange.fc_len) - 1);1918	up_write(&EXT4_I(inode)->i_data_sem);1919	if (ret)1920		goto out;1921	ext4_ext_replay_shrink_inode(inode,1922		i_size_read(inode) >> sb->s_blocksize_bits);1923	ext4_mark_inode_dirty(NULL, inode);1924out:1925	iput(inode);1926	return 0;1927}1928 1929static void ext4_fc_set_bitmaps_and_counters(struct super_block *sb)1930{1931	struct ext4_fc_replay_state *state;1932	struct inode *inode;1933	struct ext4_ext_path *path = NULL;1934	struct ext4_map_blocks map;1935	int i, ret, j;1936	ext4_lblk_t cur, end;1937 1938	state = &EXT4_SB(sb)->s_fc_replay_state;1939	for (i = 0; i < state->fc_modified_inodes_used; i++) {1940		inode = ext4_iget(sb, state->fc_modified_inodes[i],1941			EXT4_IGET_NORMAL);1942		if (IS_ERR(inode)) {1943			ext4_debug("Inode %d not found.",1944				state->fc_modified_inodes[i]);1945			continue;1946		}1947		cur = 0;1948		end = EXT_MAX_BLOCKS;1949		if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA)) {1950			iput(inode);1951			continue;1952		}1953		while (cur < end) {1954			map.m_lblk = cur;1955			map.m_len = end - cur;1956 1957			ret = ext4_map_blocks(NULL, inode, &map, 0);1958			if (ret < 0)1959				break;1960 1961			if (ret > 0) {1962				path = ext4_find_extent(inode, map.m_lblk, path, 0);1963				if (!IS_ERR(path)) {1964					for (j = 0; j < path->p_depth; j++)1965						ext4_mb_mark_bb(inode->i_sb,1966							path[j].p_block, 1, true);1967				} else {1968					path = NULL;1969				}1970				cur += ret;1971				ext4_mb_mark_bb(inode->i_sb, map.m_pblk,1972							map.m_len, true);1973			} else {1974				cur = cur + (map.m_len ? map.m_len : 1);1975			}1976		}1977		iput(inode);1978	}1979 1980	ext4_free_ext_path(path);1981}1982 1983/*1984 * Check if block is in excluded regions for block allocation. The simple1985 * allocator that runs during replay phase is calls this function to see1986 * if it is okay to use a block.1987 */1988bool ext4_fc_replay_check_excluded(struct super_block *sb, ext4_fsblk_t blk)1989{1990	int i;1991	struct ext4_fc_replay_state *state;1992 1993	state = &EXT4_SB(sb)->s_fc_replay_state;1994	for (i = 0; i < state->fc_regions_valid; i++) {1995		if (state->fc_regions[i].ino == 0 ||1996			state->fc_regions[i].len == 0)1997			continue;1998		if (in_range(blk, state->fc_regions[i].pblk,1999					state->fc_regions[i].len))2000			return true;2001	}2002	return false;2003}2004 2005/* Cleanup function called after replay */2006void ext4_fc_replay_cleanup(struct super_block *sb)2007{2008	struct ext4_sb_info *sbi = EXT4_SB(sb);2009 2010	sbi->s_mount_state &= ~EXT4_FC_REPLAY;2011	kfree(sbi->s_fc_replay_state.fc_regions);2012	kfree(sbi->s_fc_replay_state.fc_modified_inodes);2013}2014 2015static bool ext4_fc_value_len_isvalid(struct ext4_sb_info *sbi,2016				      int tag, int len)2017{2018	switch (tag) {2019	case EXT4_FC_TAG_ADD_RANGE:2020		return len == sizeof(struct ext4_fc_add_range);2021	case EXT4_FC_TAG_DEL_RANGE:2022		return len == sizeof(struct ext4_fc_del_range);2023	case EXT4_FC_TAG_CREAT:2024	case EXT4_FC_TAG_LINK:2025	case EXT4_FC_TAG_UNLINK:2026		len -= sizeof(struct ext4_fc_dentry_info);2027		return len >= 1 && len <= EXT4_NAME_LEN;2028	case EXT4_FC_TAG_INODE:2029		len -= sizeof(struct ext4_fc_inode);2030		return len >= EXT4_GOOD_OLD_INODE_SIZE &&2031			len <= sbi->s_inode_size;2032	case EXT4_FC_TAG_PAD:2033		return true; /* padding can have any length */2034	case EXT4_FC_TAG_TAIL:2035		return len >= sizeof(struct ext4_fc_tail);2036	case EXT4_FC_TAG_HEAD:2037		return len == sizeof(struct ext4_fc_head);2038	}2039	return false;2040}2041 2042/*2043 * Recovery Scan phase handler2044 *2045 * This function is called during the scan phase and is responsible2046 * for doing following things:2047 * - Make sure the fast commit area has valid tags for replay2048 * - Count number of tags that need to be replayed by the replay handler2049 * - Verify CRC2050 * - Create a list of excluded blocks for allocation during replay phase2051 *2052 * This function returns JBD2_FC_REPLAY_CONTINUE to indicate that SCAN is2053 * incomplete and JBD2 should send more blocks. It returns JBD2_FC_REPLAY_STOP2054 * to indicate that scan has finished and JBD2 can now start replay phase.2055 * It returns a negative error to indicate that there was an error. At the end2056 * of a successful scan phase, sbi->s_fc_replay_state.fc_replay_num_tags is set2057 * to indicate the number of tags that need to replayed during the replay phase.2058 */2059static int ext4_fc_replay_scan(journal_t *journal,2060				struct buffer_head *bh, int off,2061				tid_t expected_tid)2062{2063	struct super_block *sb = journal->j_private;2064	struct ext4_sb_info *sbi = EXT4_SB(sb);2065	struct ext4_fc_replay_state *state;2066	int ret = JBD2_FC_REPLAY_CONTINUE;2067	struct ext4_fc_add_range ext;2068	struct ext4_fc_tl_mem tl;2069	struct ext4_fc_tail tail;2070	__u8 *start, *end, *cur, *val;2071	struct ext4_fc_head head;2072	struct ext4_extent *ex;2073 2074	state = &sbi->s_fc_replay_state;2075 2076	start = (u8 *)bh->b_data;2077	end = start + journal->j_blocksize;2078 2079	if (state->fc_replay_expected_off == 0) {2080		state->fc_cur_tag = 0;2081		state->fc_replay_num_tags = 0;2082		state->fc_crc = 0;2083		state->fc_regions = NULL;2084		state->fc_regions_valid = state->fc_regions_used =2085			state->fc_regions_size = 0;2086		/* Check if we can stop early */2087		if (le16_to_cpu(((struct ext4_fc_tl *)start)->fc_tag)2088			!= EXT4_FC_TAG_HEAD)2089			return 0;2090	}2091 2092	if (off != state->fc_replay_expected_off) {2093		ret = -EFSCORRUPTED;2094		goto out_err;2095	}2096 2097	state->fc_replay_expected_off++;2098	for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;2099	     cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {2100		ext4_fc_get_tl(&tl, cur);2101		val = cur + EXT4_FC_TAG_BASE_LEN;2102		if (tl.fc_len > end - val ||2103		    !ext4_fc_value_len_isvalid(sbi, tl.fc_tag, tl.fc_len)) {2104			ret = state->fc_replay_num_tags ?2105				JBD2_FC_REPLAY_STOP : -ECANCELED;2106			goto out_err;2107		}2108		ext4_debug("Scan phase, tag:%s, blk %lld\n",2109			   tag2str(tl.fc_tag), bh->b_blocknr);2110		switch (tl.fc_tag) {2111		case EXT4_FC_TAG_ADD_RANGE:2112			memcpy(&ext, val, sizeof(ext));2113			ex = (struct ext4_extent *)&ext.fc_ex;2114			ret = ext4_fc_record_regions(sb,2115				le32_to_cpu(ext.fc_ino),2116				le32_to_cpu(ex->ee_block), ext4_ext_pblock(ex),2117				ext4_ext_get_actual_len(ex), 0);2118			if (ret < 0)2119				break;2120			ret = JBD2_FC_REPLAY_CONTINUE;2121			fallthrough;2122		case EXT4_FC_TAG_DEL_RANGE:2123		case EXT4_FC_TAG_LINK:2124		case EXT4_FC_TAG_UNLINK:2125		case EXT4_FC_TAG_CREAT:2126		case EXT4_FC_TAG_INODE:2127		case EXT4_FC_TAG_PAD:2128			state->fc_cur_tag++;2129			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,2130				EXT4_FC_TAG_BASE_LEN + tl.fc_len);2131			break;2132		case EXT4_FC_TAG_TAIL:2133			state->fc_cur_tag++;2134			memcpy(&tail, val, sizeof(tail));2135			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,2136						EXT4_FC_TAG_BASE_LEN +2137						offsetof(struct ext4_fc_tail,2138						fc_crc));2139			if (le32_to_cpu(tail.fc_tid) == expected_tid &&2140				le32_to_cpu(tail.fc_crc) == state->fc_crc) {2141				state->fc_replay_num_tags = state->fc_cur_tag;2142				state->fc_regions_valid =2143					state->fc_regions_used;2144			} else {2145				ret = state->fc_replay_num_tags ?2146					JBD2_FC_REPLAY_STOP : -EFSBADCRC;2147			}2148			state->fc_crc = 0;2149			break;2150		case EXT4_FC_TAG_HEAD:2151			memcpy(&head, val, sizeof(head));2152			if (le32_to_cpu(head.fc_features) &2153				~EXT4_FC_SUPPORTED_FEATURES) {2154				ret = -EOPNOTSUPP;2155				break;2156			}2157			if (le32_to_cpu(head.fc_tid) != expected_tid) {2158				ret = JBD2_FC_REPLAY_STOP;2159				break;2160			}2161			state->fc_cur_tag++;2162			state->fc_crc = ext4_chksum(sbi, state->fc_crc, cur,2163				EXT4_FC_TAG_BASE_LEN + tl.fc_len);2164			break;2165		default:2166			ret = state->fc_replay_num_tags ?2167				JBD2_FC_REPLAY_STOP : -ECANCELED;2168		}2169		if (ret < 0 || ret == JBD2_FC_REPLAY_STOP)2170			break;2171	}2172 2173out_err:2174	trace_ext4_fc_replay_scan(sb, ret, off);2175	return ret;2176}2177 2178/*2179 * Main recovery path entry point.2180 * The meaning of return codes is similar as above.2181 */2182static int ext4_fc_replay(journal_t *journal, struct buffer_head *bh,2183				enum passtype pass, int off, tid_t expected_tid)2184{2185	struct super_block *sb = journal->j_private;2186	struct ext4_sb_info *sbi = EXT4_SB(sb);2187	struct ext4_fc_tl_mem tl;2188	__u8 *start, *end, *cur, *val;2189	int ret = JBD2_FC_REPLAY_CONTINUE;2190	struct ext4_fc_replay_state *state = &sbi->s_fc_replay_state;2191	struct ext4_fc_tail tail;2192 2193	if (pass == PASS_SCAN) {2194		state->fc_current_pass = PASS_SCAN;2195		return ext4_fc_replay_scan(journal, bh, off, expected_tid);2196	}2197 2198	if (state->fc_current_pass != pass) {2199		state->fc_current_pass = pass;2200		sbi->s_mount_state |= EXT4_FC_REPLAY;2201	}2202	if (!sbi->s_fc_replay_state.fc_replay_num_tags) {2203		ext4_debug("Replay stops\n");2204		ext4_fc_set_bitmaps_and_counters(sb);2205		return 0;2206	}2207 2208#ifdef CONFIG_EXT4_DEBUG2209	if (sbi->s_fc_debug_max_replay && off >= sbi->s_fc_debug_max_replay) {2210		pr_warn("Dropping fc block %d because max_replay set\n", off);2211		return JBD2_FC_REPLAY_STOP;2212	}2213#endif2214 2215	start = (u8 *)bh->b_data;2216	end = start + journal->j_blocksize;2217 2218	for (cur = start; cur <= end - EXT4_FC_TAG_BASE_LEN;2219	     cur = cur + EXT4_FC_TAG_BASE_LEN + tl.fc_len) {2220		ext4_fc_get_tl(&tl, cur);2221		val = cur + EXT4_FC_TAG_BASE_LEN;2222 2223		if (state->fc_replay_num_tags == 0) {2224			ret = JBD2_FC_REPLAY_STOP;2225			ext4_fc_set_bitmaps_and_counters(sb);2226			break;2227		}2228 2229		ext4_debug("Replay phase, tag:%s\n", tag2str(tl.fc_tag));2230		state->fc_replay_num_tags--;2231		switch (tl.fc_tag) {2232		case EXT4_FC_TAG_LINK:2233			ret = ext4_fc_replay_link(sb, &tl, val);2234			break;2235		case EXT4_FC_TAG_UNLINK:2236			ret = ext4_fc_replay_unlink(sb, &tl, val);2237			break;2238		case EXT4_FC_TAG_ADD_RANGE:2239			ret = ext4_fc_replay_add_range(sb, &tl, val);2240			break;2241		case EXT4_FC_TAG_CREAT:2242			ret = ext4_fc_replay_create(sb, &tl, val);2243			break;2244		case EXT4_FC_TAG_DEL_RANGE:2245			ret = ext4_fc_replay_del_range(sb, &tl, val);2246			break;2247		case EXT4_FC_TAG_INODE:2248			ret = ext4_fc_replay_inode(sb, &tl, val);2249			break;2250		case EXT4_FC_TAG_PAD:2251			trace_ext4_fc_replay(sb, EXT4_FC_TAG_PAD, 0,2252					     tl.fc_len, 0);2253			break;2254		case EXT4_FC_TAG_TAIL:2255			trace_ext4_fc_replay(sb, EXT4_FC_TAG_TAIL,2256					     0, tl.fc_len, 0);2257			memcpy(&tail, val, sizeof(tail));2258			WARN_ON(le32_to_cpu(tail.fc_tid) != expected_tid);2259			break;2260		case EXT4_FC_TAG_HEAD:2261			break;2262		default:2263			trace_ext4_fc_replay(sb, tl.fc_tag, 0, tl.fc_len, 0);2264			ret = -ECANCELED;2265			break;2266		}2267		if (ret < 0)2268			break;2269		ret = JBD2_FC_REPLAY_CONTINUE;2270	}2271	return ret;2272}2273 2274void ext4_fc_init(struct super_block *sb, journal_t *journal)2275{2276	/*2277	 * We set replay callback even if fast commit disabled because we may2278	 * could still have fast commit blocks that need to be replayed even if2279	 * fast commit has now been turned off.2280	 */2281	journal->j_fc_replay_callback = ext4_fc_replay;2282	if (!test_opt2(sb, JOURNAL_FAST_COMMIT))2283		return;2284	journal->j_fc_cleanup_callback = ext4_fc_cleanup;2285}2286 2287static const char * const fc_ineligible_reasons[] = {2288	[EXT4_FC_REASON_XATTR] = "Extended attributes changed",2289	[EXT4_FC_REASON_CROSS_RENAME] = "Cross rename",2290	[EXT4_FC_REASON_JOURNAL_FLAG_CHANGE] = "Journal flag changed",2291	[EXT4_FC_REASON_NOMEM] = "Insufficient memory",2292	[EXT4_FC_REASON_SWAP_BOOT] = "Swap boot",2293	[EXT4_FC_REASON_RESIZE] = "Resize",2294	[EXT4_FC_REASON_RENAME_DIR] = "Dir renamed",2295	[EXT4_FC_REASON_FALLOC_RANGE] = "Falloc range op",2296	[EXT4_FC_REASON_INODE_JOURNAL_DATA] = "Data journalling",2297	[EXT4_FC_REASON_ENCRYPTED_FILENAME] = "Encrypted filename",2298};2299 2300int ext4_fc_info_show(struct seq_file *seq, void *v)2301{2302	struct ext4_sb_info *sbi = EXT4_SB((struct super_block *)seq->private);2303	struct ext4_fc_stats *stats = &sbi->s_fc_stats;2304	int i;2305 2306	if (v != SEQ_START_TOKEN)2307		return 0;2308 2309	seq_printf(seq,2310		"fc stats:\n%ld commits\n%ld ineligible\n%ld numblks\n%lluus avg_commit_time\n",2311		   stats->fc_num_commits, stats->fc_ineligible_commits,2312		   stats->fc_numblks,2313		   div_u64(stats->s_fc_avg_commit_time, 1000));2314	seq_puts(seq, "Ineligible reasons:\n");2315	for (i = 0; i < EXT4_FC_REASON_MAX; i++)2316		seq_printf(seq, "\"%s\":\t%d\n", fc_ineligible_reasons[i],2317			stats->fc_ineligible_reason_count[i]);2318 2319	return 0;2320}2321 2322int __init ext4_fc_init_dentry_cache(void)2323{2324	ext4_fc_dentry_cachep = KMEM_CACHE(ext4_fc_dentry_update,2325					   SLAB_RECLAIM_ACCOUNT);2326 2327	if (ext4_fc_dentry_cachep == NULL)2328		return -ENOMEM;2329 2330	return 0;2331}2332 2333void ext4_fc_destroy_dentry_cache(void)2334{2335	kmem_cache_destroy(ext4_fc_dentry_cachep);2336}2337