ext4: Add the journal checksum feature

The journal checksum feature adds two new flags i.e
JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and JBD2_FEATURE_COMPAT_CHECKSUM.

JBD2_FEATURE_CHECKSUM flag indicates that the commit block contains the
checksum for the blocks described by the descriptor blocks.
Due to checksums, writing of the commit record no longer needs to be
synchronous. Now commit record can be sent to disk without waiting for
descriptor blocks to be written to disk. This behavior is controlled
using JBD2_FEATURE_ASYNC_COMMIT flag. Older kernels/e2fsck should not be
able to recover the journal with _ASYNC_COMMIT hence it is made
incompat.
The commit header has been extended to hold the checksum along with the
type of the checksum.

For recovery in pass scan checksums are verified to ensure the sanity
and completeness(in case of _ASYNC_COMMIT) of every transaction.

Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Signed-off-by: Girish Shilamkar <girish@clusterfs.com>
Signed-off-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Signed-off-by: Mingming Cao <cmm@us.ibm.com>

3 files changed, 318 insertions, 57 deletions

diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
index 8749a86f4175..da8d0eb3b7b9 100644
--- a/fs/jbd2/commit.c
+++ b/fs/jbd2/commit.c
@@ -21,6 +21,7 @@
 #include <linux/mm.h>
 #include <linux/pagemap.h>
 #include <linux/jiffies.h>
+#include <linux/crc32.h>
 
 /*
  * Default IO end handler for temporary BJ_IO buffer_heads.
@@ -93,19 +94,23 @@ static int inverted_lock(journal_t *journal, struct buffer_head *bh)
 	return 1;
 }
 
-/* Done it all: now write the commit record.  We should have
+/*
+ * Done it all: now submit the commit record.  We should have
  * cleaned up our previous buffers by now, so if we are in abort
  * mode we can now just skip the rest of the journal write
  * entirely.
  *
  * Returns 1 if the journal needs to be aborted or 0 on success
  */
-static int journal_write_commit_record(journal_t *journal,
-					transaction_t *commit_transaction)
+static int journal_submit_commit_record(journal_t *journal,
+					transaction_t *commit_transaction,
+					struct buffer_head **cbh,
+					__u32 crc32_sum)
 {
 	struct journal_head *descriptor;
+	struct commit_header *tmp;
 	struct buffer_head *bh;
-	int i, ret;
+	int ret;
 	int barrier_done = 0;
 
 	if (is_journal_aborted(journal))
@@ -117,21 +122,33 @@ static int journal_write_commit_record(journal_t *journal,
 
 	bh = jh2bh(descriptor);
 
-	/* AKPM: buglet - add `i' to tmp! */
-	for (i = 0; i < bh->b_size; i += 512) {
-		journal_header_t *tmp = (journal_header_t*)bh->b_data;
-		tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
-		tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
-		tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
+	tmp = (struct commit_header *)bh->b_data;
+	tmp->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
+	tmp->h_blocktype = cpu_to_be32(JBD2_COMMIT_BLOCK);
+	tmp->h_sequence = cpu_to_be32(commit_transaction->t_tid);
+
+	if (JBD2_HAS_COMPAT_FEATURE(journal,
+				    JBD2_FEATURE_COMPAT_CHECKSUM)) {
+		tmp->h_chksum_type 	= JBD2_CRC32_CHKSUM;
+		tmp->h_chksum_size 	= JBD2_CRC32_CHKSUM_SIZE;
+		tmp->h_chksum[0] 	= cpu_to_be32(crc32_sum);
 	}
 
-	JBUFFER_TRACE(descriptor, "write commit block");
+	JBUFFER_TRACE(descriptor, "submit commit block");
+	lock_buffer(bh);
+
 	set_buffer_dirty(bh);
-	if (journal->j_flags & JBD2_BARRIER) {
+	set_buffer_uptodate(bh);
+	bh->b_end_io = journal_end_buffer_io_sync;
+
+	if (journal->j_flags & JBD2_BARRIER &&
+		!JBD2_HAS_COMPAT_FEATURE(journal,
+					 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
 		set_buffer_ordered(bh);
 		barrier_done = 1;
 	}
-	ret = sync_dirty_buffer(bh);
+	ret = submit_bh(WRITE, bh);
+
 	/* is it possible for another commit to fail at roughly
 	 * the same time as this one?  If so, we don't want to
 	 * trust the barrier flag in the super, but instead want
@@ -152,14 +169,72 @@ static int journal_write_commit_record(journal_t *journal,
 		clear_buffer_ordered(bh);
 		set_buffer_uptodate(bh);
 		set_buffer_dirty(bh);
-		ret = sync_dirty_buffer(bh);
+		ret = submit_bh(WRITE, bh);
 	}
-	put_bh(bh);		/* One for getblk() */
-	jbd2_journal_put_journal_head(descriptor);
+	*cbh = bh;
+	return ret;
+}
+
+/*
+ * This function along with journal_submit_commit_record
+ * allows to write the commit record asynchronously.
+ */
+static int journal_wait_on_commit_record(struct buffer_head *bh)
+{
+	int ret = 0;
+
+	clear_buffer_dirty(bh);
+	wait_on_buffer(bh);
 
-	return (ret == -EIO);
+	if (unlikely(!buffer_uptodate(bh)))
+		ret = -EIO;
+	put_bh(bh);            /* One for getblk() */
+	jbd2_journal_put_journal_head(bh2jh(bh));
+
+	return ret;
 }
 
+/*
+ * Wait for all submitted IO to complete.
+ */
+static int journal_wait_on_locked_list(journal_t *journal,
+				       transaction_t *commit_transaction)
+{
+	int ret = 0;
+	struct journal_head *jh;
+
+	while (commit_transaction->t_locked_list) {
+		struct buffer_head *bh;
+
+		jh = commit_transaction->t_locked_list->b_tprev;
+		bh = jh2bh(jh);
+		get_bh(bh);
+		if (buffer_locked(bh)) {
+			spin_unlock(&journal->j_list_lock);
+			wait_on_buffer(bh);
+			if (unlikely(!buffer_uptodate(bh)))
+				ret = -EIO;
+			spin_lock(&journal->j_list_lock);
+		}
+		if (!inverted_lock(journal, bh)) {
+			put_bh(bh);
+			spin_lock(&journal->j_list_lock);
+			continue;
+		}
+		if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
+			__jbd2_journal_unfile_buffer(jh);
+			jbd_unlock_bh_state(bh);
+			jbd2_journal_remove_journal_head(bh);
+			put_bh(bh);
+		} else {
+			jbd_unlock_bh_state(bh);
+		}
+		put_bh(bh);
+		cond_resched_lock(&journal->j_list_lock);
+	}
+	return ret;
+  }
+
 static void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
 {
 	int i;
@@ -275,7 +350,21 @@ write_out_data:
 	journal_do_submit_data(wbuf, bufs);
 }
 
-static inline void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
+static __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
+{
+	struct page *page = bh->b_page;
+	char *addr;
+	__u32 checksum;
+
+	addr = kmap_atomic(page, KM_USER0);
+	checksum = crc32_be(crc32_sum,
+		(void *)(addr + offset_in_page(bh->b_data)), bh->b_size);
+	kunmap_atomic(addr, KM_USER0);
+
+	return checksum;
+}
+
+static void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
 				   unsigned long long block)
 {
 	tag->t_blocknr = cpu_to_be32(block & (u32)~0);
@@ -307,6 +396,8 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	int tag_flag;
 	int i;
 	int tag_bytes = journal_tag_bytes(journal);
+	struct buffer_head *cbh = NULL; /* For transactional checksums */
+	__u32 crc32_sum = ~0;
 
 	/*
 	 * First job: lock down the current transaction and wait for
@@ -451,38 +542,15 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	journal_submit_data_buffers(journal, commit_transaction);
 
 	/*
-	 * Wait for all previously submitted IO to complete.
+	 * Wait for all previously submitted IO to complete if commit
+	 * record is to be written synchronously.
 	 */
 	spin_lock(&journal->j_list_lock);
-	while (commit_transaction->t_locked_list) {
-		struct buffer_head *bh;
+	if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
+		JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT))
+		err = journal_wait_on_locked_list(journal,
+						commit_transaction);
 
-		jh = commit_transaction->t_locked_list->b_tprev;
-		bh = jh2bh(jh);
-		get_bh(bh);
-		if (buffer_locked(bh)) {
-			spin_unlock(&journal->j_list_lock);
-			wait_on_buffer(bh);
-			if (unlikely(!buffer_uptodate(bh)))
-				err = -EIO;
-			spin_lock(&journal->j_list_lock);
-		}
-		if (!inverted_lock(journal, bh)) {
-			put_bh(bh);
-			spin_lock(&journal->j_list_lock);
-			continue;
-		}
-		if (buffer_jbd(bh) && jh->b_jlist == BJ_Locked) {
-			__jbd2_journal_unfile_buffer(jh);
-			jbd_unlock_bh_state(bh);
-			jbd2_journal_remove_journal_head(bh);
-			put_bh(bh);
-		} else {
-			jbd_unlock_bh_state(bh);
-		}
-		put_bh(bh);
-		cond_resched_lock(&journal->j_list_lock);
-	}
 	spin_unlock(&journal->j_list_lock);
 
 	if (err)
@@ -656,6 +724,15 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 start_journal_io:
 			for (i = 0; i < bufs; i++) {
 				struct buffer_head *bh = wbuf[i];
+				/*
+				 * Compute checksum.
+				 */
+				if (JBD2_HAS_COMPAT_FEATURE(journal,
+					JBD2_FEATURE_COMPAT_CHECKSUM)) {
+					crc32_sum =
+					    jbd2_checksum_data(crc32_sum, bh);
+				}
+
 				lock_buffer(bh);
 				clear_buffer_dirty(bh);
 				set_buffer_uptodate(bh);
@@ -672,6 +749,23 @@ start_journal_io:
 		}
 	}
 
+	/* Done it all: now write the commit record asynchronously. */
+
+	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
+		JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
+		err = journal_submit_commit_record(journal, commit_transaction,
+						 &cbh, crc32_sum);
+		if (err)
+			__jbd2_journal_abort_hard(journal);
+
+		spin_lock(&journal->j_list_lock);
+		err = journal_wait_on_locked_list(journal,
+						commit_transaction);
+		spin_unlock(&journal->j_list_lock);
+		if (err)
+			__jbd2_journal_abort_hard(journal);
+	}
+
 	/* Lo and behold: we have just managed to send a transaction to
            the log.  Before we can commit it, wait for the IO so far to
            complete.  Control buffers being written are on the
@@ -771,8 +865,14 @@ wait_for_iobuf:
 
 	jbd_debug(3, "JBD: commit phase 6\n");
 
-	if (journal_write_commit_record(journal, commit_transaction))
-		err = -EIO;
+	if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
+		JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
+		err = journal_submit_commit_record(journal, commit_transaction,
+						&cbh, crc32_sum);
+		if (err)
+			__jbd2_journal_abort_hard(journal);
+	}
+	err = journal_wait_on_commit_record(cbh);
 
 	if (err)
 		jbd2_journal_abort(journal, err);
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 3667c91bc786..59ba2494dcaf 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -1578,6 +1578,32 @@ int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
 	return 1;
 }
 
+/*
+ * jbd2_journal_clear_features () - Clear a given journal feature in the
+ * 				    superblock
+ * @journal: Journal to act on.
+ * @compat: bitmask of compatible features
+ * @ro: bitmask of features that force read-only mount
+ * @incompat: bitmask of incompatible features
+ *
+ * Clear a given journal feature as present on the
+ * superblock.
+ */
+void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
+				unsigned long ro, unsigned long incompat)
+{
+	journal_superblock_t *sb;
+
+	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
+		  compat, ro, incompat);
+
+	sb = journal->j_superblock;
+
+	sb->s_feature_compat    &= ~cpu_to_be32(compat);
+	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
+	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
+}
+EXPORT_SYMBOL(jbd2_journal_clear_features);
 
 /**
  * int jbd2_journal_update_format () - Update on-disk journal structure.
diff --git a/fs/jbd2/recovery.c b/fs/jbd2/recovery.c
index d0ce627539ef..921680663fa2 100644
--- a/fs/jbd2/recovery.c
+++ b/fs/jbd2/recovery.c
@@ -21,6 +21,7 @@
 #include <linux/jbd2.h>
 #include <linux/errno.h>
 #include <linux/slab.h>
+#include <linux/crc32.h>
 #endif
 
 /*
@@ -316,6 +317,37 @@ static inline unsigned long long read_tag_block(int tag_bytes, journal_block_tag
 	return block;
 }
 
+/*
+ * calc_chksums calculates the checksums for the blocks described in the
+ * descriptor block.
+ */
+static int calc_chksums(journal_t *journal, struct buffer_head *bh,
+			unsigned long *next_log_block, __u32 *crc32_sum)
+{
+	int i, num_blks, err;
+	unsigned long io_block;
+	struct buffer_head *obh;
+
+	num_blks = count_tags(journal, bh);
+	/* Calculate checksum of the descriptor block. */
+	*crc32_sum = crc32_be(*crc32_sum, (void *)bh->b_data, bh->b_size);
+
+	for (i = 0; i < num_blks; i++) {
+		io_block = (*next_log_block)++;
+		wrap(journal, *next_log_block);
+		err = jread(&obh, journal, io_block);
+		if (err) {
+			printk(KERN_ERR "JBD: IO error %d recovering block "
+				"%lu in log\n", err, io_block);
+			return 1;
+		} else {
+			*crc32_sum = crc32_be(*crc32_sum, (void *)obh->b_data,
+				     obh->b_size);
+		}
+	}
+	return 0;
+}
+
 static int do_one_pass(journal_t *journal,
 			struct recovery_info *info, enum passtype pass)
 {
@@ -328,6 +360,7 @@ static int do_one_pass(journal_t *journal,
 	unsigned int		sequence;
 	int			blocktype;
 	int			tag_bytes = journal_tag_bytes(journal);
+	__u32			crc32_sum = ~0; /* Transactional Checksums */
 
 	/* Precompute the maximum metadata descriptors in a descriptor block */
 	int			MAX_BLOCKS_PER_DESC;
@@ -419,12 +452,26 @@ static int do_one_pass(journal_t *journal,
 		switch(blocktype) {
 		case JBD2_DESCRIPTOR_BLOCK:
 			/* If it is a valid descriptor block, replay it
-			 * in pass REPLAY; otherwise, just skip over the
-			 * blocks it describes. */
+			 * in pass REPLAY; if journal_checksums enabled, then
+			 * calculate checksums in PASS_SCAN, otherwise,
+			 * just skip over the blocks it describes. */
 			if (pass != PASS_REPLAY) {
+				if (pass == PASS_SCAN &&
+				    JBD2_HAS_COMPAT_FEATURE(journal,
+					    JBD2_FEATURE_COMPAT_CHECKSUM) &&
+				    !info->end_transaction) {
+					if (calc_chksums(journal, bh,
+							&next_log_block,
+							&crc32_sum)) {
+						put_bh(bh);
+						break;
+					}
+					put_bh(bh);
+					continue;
+				}
 				next_log_block += count_tags(journal, bh);
 				wrap(journal, next_log_block);
-				brelse(bh);
+				put_bh(bh);
 				continue;
 			}
 
@@ -516,9 +563,96 @@ static int do_one_pass(journal_t *journal,
 			continue;
 
 		case JBD2_COMMIT_BLOCK:
-			/* Found an expected commit block: not much to
-			 * do other than move on to the next sequence
+			/*     How to differentiate between interrupted commit
+			 *               and journal corruption ?
+			 *
+			 * {nth transaction}
+			 *        Checksum Verification Failed
+			 *			 |
+			 *		 ____________________
+			 *		|		     |
+			 * 	async_commit             sync_commit
+			 *     		|                    |
+			 *		| GO TO NEXT    "Journal Corruption"
+			 *		| TRANSACTION
+			 *		|
+			 * {(n+1)th transanction}
+			 *		|
+			 * 	 _______|______________
+			 * 	|	 	      |
+			 * Commit block found	Commit block not found
+			 *      |		      |
+			 * "Journal Corruption"       |
+			 *		 _____________|_________
+			 *     		|	           	|
+			 *	nth trans corrupt	OR   nth trans
+			 *	and (n+1)th interrupted     interrupted
+			 *	before commit block
+			 *      could reach the disk.
+			 *	(Cannot find the difference in above
+			 *	 mentioned conditions. Hence assume
+			 *	 "Interrupted Commit".)
+			 */
+
+			/* Found an expected commit block: if checksums
+			 * are present verify them in PASS_SCAN; else not
+			 * much to do other than move on to the next sequence
 			 * number. */
+			if (pass == PASS_SCAN &&
+			    JBD2_HAS_COMPAT_FEATURE(journal,
+				    JBD2_FEATURE_COMPAT_CHECKSUM)) {
+				int chksum_err, chksum_seen;
+				struct commit_header *cbh =
+					(struct commit_header *)bh->b_data;
+				unsigned found_chksum =
+					be32_to_cpu(cbh->h_chksum[0]);
+
+				chksum_err = chksum_seen = 0;
+
+				if (info->end_transaction) {
+					printk(KERN_ERR "JBD: Transaction %u "
+						"found to be corrupt.\n",
+						next_commit_ID - 1);
+					brelse(bh);
+					break;
+				}
+
+				if (crc32_sum == found_chksum &&
+				    cbh->h_chksum_type == JBD2_CRC32_CHKSUM &&
+				    cbh->h_chksum_size ==
+						JBD2_CRC32_CHKSUM_SIZE)
+				       chksum_seen = 1;
+				else if (!(cbh->h_chksum_type == 0 &&
+					     cbh->h_chksum_size == 0 &&
+					     found_chksum == 0 &&
+					     !chksum_seen))
+				/*
+				 * If fs is mounted using an old kernel and then
+				 * kernel with journal_chksum is used then we
+				 * get a situation where the journal flag has
+				 * checksum flag set but checksums are not
+				 * present i.e chksum = 0, in the individual
+				 * commit blocks.
+				 * Hence to avoid checksum failures, in this
+				 * situation, this extra check is added.
+				 */
+						chksum_err = 1;
+
+				if (chksum_err) {
+					info->end_transaction = next_commit_ID;
+
+					if (!JBD2_HAS_COMPAT_FEATURE(journal,
+					   JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)){
+						printk(KERN_ERR
+						       "JBD: Transaction %u "
+						       "found to be corrupt.\n",
+						       next_commit_ID);
+						brelse(bh);
+						break;
+					}
+				}
+				crc32_sum = ~0;
+			}
 			brelse(bh);
 			next_commit_ID++;
 			continue;
@@ -554,9 +688,10 @@ static int do_one_pass(journal_t *journal,
 	 * transaction marks the end of the valid log.
 	 */
 
-	if (pass == PASS_SCAN)
-		info->end_transaction = next_commit_ID;
-	else {
+	if (pass == PASS_SCAN) {
+		if (!info->end_transaction)
+			info->end_transaction = next_commit_ID;
+	} else {
 		/* It's really bad news if different passes end up at
 		 * different places (but possible due to IO errors). */
 		if (info->end_transaction != next_commit_ID) {