Commit 68569684 authored by Andrew Morton's avatar Andrew Morton Committed by Linus Torvalds

[PATCH] ext3 journal commit I/O error fix

From: Hua Zhong <hzhong@cisco.com>

The current ext3 totally ignores I/O errors that happened during a
journal_force_commit time, causing user space to falsely believe it has
succeeded, which actually did not.

This patch  checks IO error during  journal_commit_transaction. and aborts
the journal when there is I/O error.

Originally I thought about reporting the error without doing aborting the
journal, but it probably needs a new flag. Aborting the journal seems to be
the easy way to  signal "hey sth is wrong..".
parent c20fb5f1
...@@ -72,6 +72,5 @@ int ext3_sync_file(struct file * file, struct dentry *dentry, int datasync) ...@@ -72,6 +72,5 @@ int ext3_sync_file(struct file * file, struct dentry *dentry, int datasync)
* (they were dirtied by commit). But that's OK - the blocks are * (they were dirtied by commit). But that's OK - the blocks are
* safe in-journal, which is all fsync() needs to ensure. * safe in-journal, which is all fsync() needs to ensure.
*/ */
ext3_force_commit(inode->i_sb); return ext3_force_commit(inode->i_sb);
return 0;
} }
...@@ -173,6 +173,7 @@ void journal_commit_transaction(journal_t *journal) ...@@ -173,6 +173,7 @@ void journal_commit_transaction(journal_t *journal)
* on the transaction lists. Data blocks go first. * on the transaction lists. Data blocks go first.
*/ */
err = 0;
/* /*
* Whenever we unlock the journal and sleep, things can get added * Whenever we unlock the journal and sleep, things can get added
* onto ->t_datalist, so we have to keep looping back to write_out_data * onto ->t_datalist, so we have to keep looping back to write_out_data
...@@ -251,9 +252,13 @@ void journal_commit_transaction(journal_t *journal) ...@@ -251,9 +252,13 @@ void journal_commit_transaction(journal_t *journal)
jh = jh->b_tprev; /* Wait on the last written */ jh = jh->b_tprev; /* Wait on the last written */
bh = jh2bh(jh); bh = jh2bh(jh);
if (buffer_locked(bh)) { if (buffer_locked(bh)) {
get_bh(bh);
spin_unlock(&journal_datalist_lock); spin_unlock(&journal_datalist_lock);
unlock_journal(journal); unlock_journal(journal);
wait_on_buffer(bh); wait_on_buffer(bh);
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
put_bh(bh);
/* the journal_head may have been removed now */ /* the journal_head may have been removed now */
lock_journal(journal); lock_journal(journal);
goto write_out_data; goto write_out_data;
...@@ -446,7 +451,10 @@ void journal_commit_transaction(journal_t *journal) ...@@ -446,7 +451,10 @@ void journal_commit_transaction(journal_t *journal)
jbd_debug(3, "JBD: commit phase 4\n"); jbd_debug(3, "JBD: commit phase 4\n");
/* akpm: these are BJ_IO, and journal_datalist_lock is not needed */ /*
* akpm: these are BJ_IO, and journal_datalist_lock is not needed.
* See __journal_try_to_free_buffer.
*/
wait_for_iobuf: wait_for_iobuf:
while (commit_transaction->t_iobuf_list != NULL) { while (commit_transaction->t_iobuf_list != NULL) {
struct buffer_head *bh; struct buffer_head *bh;
...@@ -455,6 +463,8 @@ void journal_commit_transaction(journal_t *journal) ...@@ -455,6 +463,8 @@ void journal_commit_transaction(journal_t *journal)
if (buffer_locked(bh)) { if (buffer_locked(bh)) {
unlock_journal(journal); unlock_journal(journal);
wait_on_buffer(bh); wait_on_buffer(bh);
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
lock_journal(journal); lock_journal(journal);
goto wait_for_iobuf; goto wait_for_iobuf;
} }
...@@ -516,6 +526,8 @@ void journal_commit_transaction(journal_t *journal) ...@@ -516,6 +526,8 @@ void journal_commit_transaction(journal_t *journal)
if (buffer_locked(bh)) { if (buffer_locked(bh)) {
unlock_journal(journal); unlock_journal(journal);
wait_on_buffer(bh); wait_on_buffer(bh);
if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
lock_journal(journal); lock_journal(journal);
goto wait_for_ctlbuf; goto wait_for_ctlbuf;
} }
...@@ -563,7 +575,9 @@ void journal_commit_transaction(journal_t *journal) ...@@ -563,7 +575,9 @@ void journal_commit_transaction(journal_t *journal)
struct buffer_head *bh = jh2bh(descriptor); struct buffer_head *bh = jh2bh(descriptor);
set_buffer_uptodate(bh); set_buffer_uptodate(bh);
sync_dirty_buffer(bh); sync_dirty_buffer(bh);
__brelse(bh); /* One for getblk() */ if (unlikely(!buffer_uptodate(bh)))
err = -EIO;
put_bh(bh); /* One for getblk() */
journal_unlock_journal_head(descriptor); journal_unlock_journal_head(descriptor);
} }
...@@ -574,6 +588,12 @@ void journal_commit_transaction(journal_t *journal) ...@@ -574,6 +588,12 @@ void journal_commit_transaction(journal_t *journal)
skip_commit: /* The journal should be unlocked by now. */ skip_commit: /* The journal should be unlocked by now. */
if (err) {
lock_journal(journal);
__journal_abort_hard(journal);
unlock_journal(journal);
}
/* Call any callbacks that had been registered for handles in this /* Call any callbacks that had been registered for handles in this
* transaction. It is up to the callback to free any allocated * transaction. It is up to the callback to free any allocated
* memory. * memory.
......
...@@ -580,8 +580,10 @@ tid_t log_start_commit (journal_t *journal, transaction_t *transaction) ...@@ -580,8 +580,10 @@ tid_t log_start_commit (journal_t *journal, transaction_t *transaction)
* Wait for a specified commit to complete. * Wait for a specified commit to complete.
* The caller may not hold the journal lock. * The caller may not hold the journal lock.
*/ */
void log_wait_commit (journal_t *journal, tid_t tid) int log_wait_commit (journal_t *journal, tid_t tid)
{ {
int err = 0;
lock_kernel(); lock_kernel();
#ifdef CONFIG_JBD_DEBUG #ifdef CONFIG_JBD_DEBUG
lock_journal(journal); lock_journal(journal);
...@@ -598,7 +600,14 @@ void log_wait_commit (journal_t *journal, tid_t tid) ...@@ -598,7 +600,14 @@ void log_wait_commit (journal_t *journal, tid_t tid)
wake_up(&journal->j_wait_commit); wake_up(&journal->j_wait_commit);
sleep_on(&journal->j_wait_done_commit); sleep_on(&journal->j_wait_done_commit);
} }
if (unlikely(is_journal_aborted(journal))) {
printk(KERN_EMERG "journal commit I/O error\n");
err = -EIO;
}
unlock_kernel(); unlock_kernel();
return err;
} }
/* /*
......
...@@ -1402,7 +1402,7 @@ int journal_stop(handle_t *handle) ...@@ -1402,7 +1402,7 @@ int journal_stop(handle_t *handle)
* to wait for the commit to complete. * to wait for the commit to complete.
*/ */
if (handle->h_sync && !(current->flags & PF_MEMALLOC)) if (handle->h_sync && !(current->flags & PF_MEMALLOC))
log_wait_commit(journal, tid); err = log_wait_commit(journal, tid);
} }
jbd_free_handle(handle); jbd_free_handle(handle);
return err; return err;
...@@ -1418,7 +1418,7 @@ int journal_stop(handle_t *handle) ...@@ -1418,7 +1418,7 @@ int journal_stop(handle_t *handle)
int journal_force_commit(journal_t *journal) int journal_force_commit(journal_t *journal)
{ {
handle_t *handle; handle_t *handle;
int ret = 0; int ret;
lock_kernel(); lock_kernel();
handle = journal_start(journal, 1); handle = journal_start(journal, 1);
...@@ -1427,7 +1427,7 @@ int journal_force_commit(journal_t *journal) ...@@ -1427,7 +1427,7 @@ int journal_force_commit(journal_t *journal)
goto out; goto out;
} }
handle->h_sync = 1; handle->h_sync = 1;
journal_stop(handle); ret = journal_stop(handle);
out: out:
unlock_kernel(); unlock_kernel();
return ret; return ret;
......
...@@ -855,7 +855,7 @@ extern void journal_brelse_array(struct buffer_head *b[], int n); ...@@ -855,7 +855,7 @@ extern void journal_brelse_array(struct buffer_head *b[], int n);
extern int log_space_left (journal_t *); /* Called with journal locked */ extern int log_space_left (journal_t *); /* Called with journal locked */
extern tid_t log_start_commit (journal_t *, transaction_t *); extern tid_t log_start_commit (journal_t *, transaction_t *);
extern void log_wait_commit (journal_t *, tid_t); extern int log_wait_commit (journal_t *, tid_t);
extern int log_do_checkpoint (journal_t *, int); extern int log_do_checkpoint (journal_t *, int);
extern void log_wait_for_space(journal_t *, int nblocks); extern void log_wait_for_space(journal_t *, int nblocks);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment