> On Mon, 21 Jan 2008 22:02:12 -0500 "Theodore Ts'o" <[email protected]> wrote:
> From: Girish Shilamkar <[email protected]>
>
> 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.
>
> ...
>
> +static inline __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
unneeded inlining.
> +{
> + 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;
> +}
Can this buffer actually be in highmem?
> static inline void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
> unsigned long long block)
More unnecessary inlining.
> +/*
> + * 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. Returns true if the requested features could be reset.
> + */
> +int 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);
> +
> + return 1;
> +}
> +EXPORT_SYMBOL(jbd2_journal_clear_features);
Kernel usually returns 0 on success. So we can return a useful errno on
failure.
> +/*
> + * 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 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)++;
unsigned <- unsigned long.
Are all the types appropriate in here?
> + wrap(journal, *next_log_block);
> + err = jread(&obh, journal, io_block);
> + if (err) {
> + printk(KERN_ERR "JBD: IO error %d recovering block "
> + "%u 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;
> @@ -419,9 +452,23 @@ 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()
> + brelse(bh);
> + break;
> + }
> + brelse(bh);
> + continue;
put_bh()
> + }
> next_log_block += count_tags(journal, bh);
> wrap(journal, next_log_block);
> brelse(bh);
> @@ -516,9 +563,96 @@ static int do_one_pass(journal_t *journal,
> continue;
>
> + brelse(bh);
etc
On Jan 23, 2008 14:07 -0800, Andrew Morton wrote:
> > +{
> > + 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;
> > +}
>
> Can this buffer actually be in highmem?
Yes, this was found during system testing. While ext3/4 will only allocate
buffer heads in lowmem, the jbd/jbd2 code can allocate buffers in highmem.
I was surprised about this also.
Please see the thread in ext4-devel:
[PATCH][RFC]JBD2: Fix journal checksum kernel oops on NUMA
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
On Wed, 2008-01-23 at 14:07 -0800, Andrew Morton wrote:
> > On Mon, 21 Jan 2008 22:02:12 -0500 "Theodore Ts'o" <[email protected]> wrote:
> > From: Girish Shilamkar <[email protected]>
> >
> > 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.
> >
> > ...
> >
> > +static inline __u32 jbd2_checksum_data(__u32 crc32_sum, struct buffer_head *bh)
>
> unneeded inlining.
>
> > +{
> > + 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;
> > +}
>
> Can this buffer actually be in highmem?
>
> > static inline void write_tag_block(int tag_bytes, journal_block_tag_t *tag,
> > unsigned long long block)
>
> More unnecessary inlining.
>
> > +/*
> > + * 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. Returns true if the requested features could be reset.
> > + */
> > +int 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);
> > +
> > + return 1;
> > +}
> > +EXPORT_SYMBOL(jbd2_journal_clear_features);
>
> Kernel usually returns 0 on success. So we can return a useful errno on
> failure.
>
> > +/*
> > + * 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 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)++;
>
> unsigned <- unsigned long.
>
> Are all the types appropriate in here?
>
> > + wrap(journal, *next_log_block);
> > + err = jread(&obh, journal, io_block);
> > + if (err) {
> > + printk(KERN_ERR "JBD: IO error %d recovering block "
> > + "%u 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;
> > @@ -419,9 +452,23 @@ 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()
>
> > + brelse(bh);
> > + break;
> > + }
> > + brelse(bh);
> > + continue;
>
> put_bh()
>
> > + }
> > next_log_block += count_tags(journal, bh);
> > wrap(journal, next_log_block);
> > brelse(bh);
> > @@ -516,9 +563,96 @@ static int do_one_pass(journal_t *journal,
> > continue;
> >
> > + brelse(bh);
>
> etc
>
Thanks, Updated patch below:
ext4: Add the journal checksum feature
From: Girish Shilamkar <[email protected]>
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 <[email protected]>
Signed-off-by: Girish Shilamkar <[email protected]>
Signed-off-by: Dave Kleikamp <[email protected]>
Signed-off-by: Mingming Cao <[email protected]>
---
Documentation/filesystems/ext4.txt | 10 +
fs/Kconfig | 1
fs/ext4/super.c | 25 ++++
fs/jbd2/commit.c | 198 +++++++++++++++++++++++++++----------
fs/jbd2/journal.c | 26 ++++
fs/jbd2/recovery.c | 151 ++++++++++++++++++++++++++--
include/linux/ext4_fs.h | 3
include/linux/jbd2.h | 36 +++++-
8 files changed, 388 insertions(+), 62 deletions(-)
Index: linux-2.6.24-rc8/Documentation/filesystems/ext4.txt
===================================================================
--- linux-2.6.24-rc8.orig/Documentation/filesystems/ext4.txt 2008-01-24 11:18:08.000000000 -0800
+++ linux-2.6.24-rc8/Documentation/filesystems/ext4.txt 2008-01-24 13:00:44.000000000 -0800
@@ -89,6 +89,16 @@ When mounting an ext4 filesystem, the fo
extents ext4 will use extents to address file data. The
file system will no longer be mountable by ext3.
+journal_checksum Enable checksumming of the journal transactions.
+ This will allow the recovery code in e2fsck and the
+ kernel to detect corruption in the kernel. It is a
+ compatible change and will be ignored by older kernels.
+
+journal_async_commit Commit block can be written to disk without waiting
+ for descriptor blocks. If enabled older kernels cannot
+ mount the device. This will enable 'journal_checksum'
+ internally.
+
journal=update Update the ext4 file system's journal to the current
format.
Index: linux-2.6.24-rc8/fs/Kconfig
===================================================================
--- linux-2.6.24-rc8.orig/fs/Kconfig 2008-01-24 11:18:08.000000000 -0800
+++ linux-2.6.24-rc8/fs/Kconfig 2008-01-24 11:18:55.000000000 -0800
@@ -236,6 +236,7 @@ config JBD_DEBUG
config JBD2
tristate
+ select CRC32
help
This is a generic journaling layer for block devices that support
both 32-bit and 64-bit block numbers. It is currently used by
Index: linux-2.6.24-rc8/fs/ext4/super.c
===================================================================
--- linux-2.6.24-rc8.orig/fs/ext4/super.c 2008-01-24 11:18:52.000000000 -0800
+++ linux-2.6.24-rc8/fs/ext4/super.c 2008-01-24 13:00:45.000000000 -0800
@@ -869,6 +869,7 @@ enum {
Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
Opt_reservation, Opt_noreservation, Opt_noload, Opt_nobh, Opt_bh,
Opt_commit, Opt_journal_update, Opt_journal_inum, Opt_journal_dev,
+ Opt_journal_checksum, Opt_journal_async_commit,
Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
@@ -908,6 +909,8 @@ static match_table_t tokens = {
{Opt_journal_update, "journal=update"},
{Opt_journal_inum, "journal=%u"},
{Opt_journal_dev, "journal_dev=%u"},
+ {Opt_journal_checksum, "journal_checksum"},
+ {Opt_journal_async_commit, "journal_async_commit"},
{Opt_abort, "abort"},
{Opt_data_journal, "data=journal"},
{Opt_data_ordered, "data=ordered"},
@@ -1095,6 +1098,13 @@ static int parse_options (char *options,
return 0;
*journal_devnum = option;
break;
+ case Opt_journal_checksum:
+ set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
+ break;
+ case Opt_journal_async_commit:
+ set_opt(sbi->s_mount_opt, JOURNAL_ASYNC_COMMIT);
+ set_opt(sbi->s_mount_opt, JOURNAL_CHECKSUM);
+ break;
case Opt_noload:
set_opt (sbi->s_mount_opt, NOLOAD);
break;
@@ -2114,6 +2124,21 @@ static int ext4_fill_super (struct super
goto failed_mount4;
}
+ if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
+ jbd2_journal_set_features(sbi->s_journal,
+ JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+ } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
+ jbd2_journal_set_features(sbi->s_journal,
+ JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
+ jbd2_journal_clear_features(sbi->s_journal, 0, 0,
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+ } else {
+ jbd2_journal_clear_features(sbi->s_journal,
+ JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+ }
+
/* We have now updated the journal if required, so we can
* validate the data journaling mode. */
switch (test_opt(sb, DATA_FLAGS)) {
Index: linux-2.6.24-rc8/fs/jbd2/commit.c
===================================================================
--- linux-2.6.24-rc8.orig/fs/jbd2/commit.c 2008-01-24 11:18:54.000000000 -0800
+++ linux-2.6.24-rc8/fs/jbd2/commit.c 2008-01-24 13:02:43.000000000 -0800
@@ -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 *jour
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(j
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(j
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);
+
+ if (unlikely(!buffer_uptodate(bh)))
+ ret = -EIO;
+ put_bh(bh); /* One for getblk() */
+ jbd2_journal_put_journal_head(bh2jh(bh));
- return (ret == -EIO);
+ 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(jou
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(jou
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(jou
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);
Index: linux-2.6.24-rc8/fs/jbd2/journal.c
===================================================================
--- linux-2.6.24-rc8.orig/fs/jbd2/journal.c 2008-01-24 11:18:54.000000000 -0800
+++ linux-2.6.24-rc8/fs/jbd2/journal.c 2008-01-24 13:06:53.000000000 -0800
@@ -1578,6 +1578,32 @@ int jbd2_journal_set_features (journal_t
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.
+ */
+int 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.
Index: linux-2.6.24-rc8/fs/jbd2/recovery.c
===================================================================
--- linux-2.6.24-rc8.orig/fs/jbd2/recovery.c 2008-01-24 11:18:08.000000000 -0800
+++ linux-2.6.24-rc8/fs/jbd2/recovery.c 2008-01-24 13:16:07.000000000 -0800
@@ -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_ta
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 *journa
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 *journa
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 *journa
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 *journa
* 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) {
Index: linux-2.6.24-rc8/include/linux/ext4_fs.h
===================================================================
--- linux-2.6.24-rc8.orig/include/linux/ext4_fs.h 2008-01-24 11:18:52.000000000 -0800
+++ linux-2.6.24-rc8/include/linux/ext4_fs.h 2008-01-24 13:00:45.000000000 -0800
@@ -467,7 +467,8 @@ do { \
#define EXT4_MOUNT_USRQUOTA 0x100000 /* "old" user quota */
#define EXT4_MOUNT_GRPQUOTA 0x200000 /* "old" group quota */
#define EXT4_MOUNT_EXTENTS 0x400000 /* Extents support */
-
+#define EXT4_MOUNT_JOURNAL_CHECKSUM 0x800000 /* Journal checksums */
+#define EXT4_MOUNT_JOURNAL_ASYNC_COMMIT 0x1000000 /* Journal Async Commit */
/* Compatibility, for having both ext2_fs.h and ext4_fs.h included at once */
#ifndef _LINUX_EXT2_FS_H
#define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt
Index: linux-2.6.24-rc8/include/linux/jbd2.h
===================================================================
--- linux-2.6.24-rc8.orig/include/linux/jbd2.h 2008-01-24 11:18:54.000000000 -0800
+++ linux-2.6.24-rc8/include/linux/jbd2.h 2008-01-24 11:46:16.000000000 -0800
@@ -149,6 +149,28 @@ typedef struct journal_header_s
__be32 h_sequence;
} journal_header_t;
+/*
+ * Checksum types.
+ */
+#define JBD2_CRC32_CHKSUM 1
+#define JBD2_MD5_CHKSUM 2
+#define JBD2_SHA1_CHKSUM 3
+
+#define JBD2_CRC32_CHKSUM_SIZE 4
+
+#define JBD2_CHECKSUM_BYTES (32 / sizeof(u32))
+/*
+ * Commit block header for storing transactional checksums:
+ */
+struct commit_header {
+ __be32 h_magic;
+ __be32 h_blocktype;
+ __be32 h_sequence;
+ unsigned char h_chksum_type;
+ unsigned char h_chksum_size;
+ unsigned char h_padding[2];
+ __be32 h_chksum[JBD2_CHECKSUM_BYTES];
+};
/*
* The block tag: used to describe a single buffer in the journal.
@@ -242,14 +264,18 @@ typedef struct journal_superblock_s
((j)->j_format_version >= 2 && \
((j)->j_superblock->s_feature_incompat & cpu_to_be32((mask))))
-#define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001
-#define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002
+#define JBD2_FEATURE_COMPAT_CHECKSUM 0x00000001
+
+#define JBD2_FEATURE_INCOMPAT_REVOKE 0x00000001
+#define JBD2_FEATURE_INCOMPAT_64BIT 0x00000002
+#define JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT 0x00000004
/* Features known to this kernel version: */
-#define JBD2_KNOWN_COMPAT_FEATURES 0
+#define JBD2_KNOWN_COMPAT_FEATURES JBD2_FEATURE_COMPAT_CHECKSUM
#define JBD2_KNOWN_ROCOMPAT_FEATURES 0
#define JBD2_KNOWN_INCOMPAT_FEATURES (JBD2_FEATURE_INCOMPAT_REVOKE | \
- JBD2_FEATURE_INCOMPAT_64BIT)
+ JBD2_FEATURE_INCOMPAT_64BIT | \
+ JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)
#ifdef __KERNEL__
@@ -997,6 +1023,8 @@ extern int jbd2_journal_check_availab
(journal_t *, unsigned long, unsigned long, unsigned long);
extern int jbd2_journal_set_features
(journal_t *, unsigned long, unsigned long, unsigned long);
+extern int jbd2_journal_clear_features
+ (journal_t *, unsigned long, unsigned long, unsigned long);
extern int jbd2_journal_create (journal_t *);
extern int jbd2_journal_load (journal_t *journal);
extern void jbd2_journal_destroy (journal_t *);
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi,
On Thu, 2008-01-24 at 13:24 -0800, Mingming Cao wrote:
> -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(j
>
> 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);
> +
get_bh() is missing here.
bh refcount is decremented in journal_wait_on_commit_record(), but it is
not incremented in journal_submit_commit_record().
Thanks to Johann Lombardi for pointing this out.
Comments.
> 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(j
> 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);
> +
> + if (unlikely(!buffer_uptodate(bh)))
> + ret = -EIO;
> + put_bh(bh); /* One for getblk() */
> + jbd2_journal_put_journal_head(bh2jh(bh));
>
> - return (ret == -EIO);
> + return ret;
> }
>
-Girish