Merge branch 'ext3-latency-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4

* 'ext3-latency-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: ext3: Add replace-on-rename hueristics for data=writeback mode ext3: Add replace-on-truncate hueristics for data=writeback mode ext3: Use WRITE_SYNC for commits which are caused by fsync() block_write_full_page: Use synchronous writes for WBC_SYNC_ALL writebacks

Merge branch 'ext3-latency-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
* 'ext3-latency-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: ext3: Add replace-on-rename hueristics for data=writeback mode ext3: Add replace-on-truncate hueristics for data=writeback mode ext3: Use WRITE_SYNC for commits which are caused by fsync() block_write_full_page: Use synchronous writes for WBC_SYNC_ALL writebacks
Linus Torvalds
2 parents 18b34b9546 e7c8f5079e
Showing 8 changed files Side-by-side Diff
fs/buffer.c
fs/ext3/file.c
fs/ext3/inode.c
fs/ext3/namei.c
fs/jbd/commit.c
fs/jbd/transaction.c
include/linux/ext3_fs.h
include/linux/jbd.h
@@ -1595,6 +1595,7 @@
 	struct buffer_head *bh, *head;
 	const unsigned blocksize = 1 << inode->i_blkbits;
 	int nr_underway = 0;
+	int write_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
  
 	BUG_ON(!PageLocked(page));
  
@@ -1686,7 +1687,7 @@
 	do {
 		struct buffer_head *next = bh->b_this_page;
 		if (buffer_async_write(bh)) {
-			submit_bh(WRITE, bh);
+			submit_bh(write_op, bh);
 			nr_underway++;
 		}
 		bh = next;
@@ -1740,7 +1741,7 @@
 		struct buffer_head *next = bh->b_this_page;
 		if (buffer_async_write(bh)) {
 			clear_buffer_dirty(bh);
-			submit_bh(WRITE, bh);
+			submit_bh(write_op, bh);
 			nr_underway++;
 		}
 		bh = next;
@@ -33,6 +33,10 @@
  */
 static int ext3_release_file (struct inode * inode, struct file * filp)
 {
+	if (EXT3_I(inode)->i_state & EXT3_STATE_FLUSH_ON_CLOSE) {
+		filemap_flush(inode->i_mapping);
+		EXT3_I(inode)->i_state &= ~EXT3_STATE_FLUSH_ON_CLOSE;
+	}
 	/* if we are the last writer on the inode, drop the block reservation */
 	if ((filp->f_mode & FMODE_WRITE) &&
 			(atomic_read(&inode->i_writecount) == 1))
@@ -2363,6 +2363,9 @@
 	if (!ext3_can_truncate(inode))
 		return;
  
+	if (inode->i_size == 0 && ext3_should_writeback_data(inode))
+		ei->i_state |= EXT3_STATE_FLUSH_ON_CLOSE;
+
 	/*
 	 * We have to lock the EOF page here, because lock_page() nests
 	 * outside journal_start().
@@ -2274,7 +2274,7 @@
 	struct inode * old_inode, * new_inode;
 	struct buffer_head * old_bh, * new_bh, * dir_bh;
 	struct ext3_dir_entry_2 * old_de, * new_de;
-	int retval;
+	int retval, flush_file = 0;
  
 	old_bh = new_bh = dir_bh = NULL;
  
@@ -2410,6 +2410,8 @@
 		ext3_mark_inode_dirty(handle, new_inode);
 		if (!new_inode->i_nlink)
 			ext3_orphan_add(handle, new_inode);
+		if (ext3_should_writeback_data(new_inode))
+			flush_file = 1;
 	}
 	retval = 0;
  
@@ -2418,6 +2420,8 @@
 	brelse (old_bh);
 	brelse (new_bh);
 	ext3_journal_stop(handle);
+	if (retval == 0 && flush_file)
+		filemap_flush(old_inode->i_mapping);
 	return retval;
 }
  
@@ -20,6 +20,7 @@
 #include <linux/slab.h>
 #include <linux/mm.h>
 #include <linux/pagemap.h>
+#include <linux/bio.h>
  
 /*
  * Default IO end handler for temporary BJ_IO buffer_heads.
  
@@ -171,14 +172,15 @@
 	return (ret == -EIO);
 }
  
-static void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
+static void journal_do_submit_data(struct buffer_head **wbuf, int bufs,
+				   int write_op)
 {
 	int i;
  
 	for (i = 0; i < bufs; i++) {
 		wbuf[i]->b_end_io = end_buffer_write_sync;
 		/* We use-up our safety reference in submit_bh() */
-		submit_bh(WRITE, wbuf[i]);
+		submit_bh(write_op, wbuf[i]);
 	}
 }
  
@@ -186,7 +188,8 @@
  *  Submit all the data buffers to disk
  */
 static int journal_submit_data_buffers(journal_t *journal,
-				transaction_t *commit_transaction)
+				       transaction_t *commit_transaction,
+				       int write_op)
 {
 	struct journal_head *jh;
 	struct buffer_head *bh;
@@ -225,7 +228,7 @@
 				BUFFER_TRACE(bh, "needs blocking lock");
 				spin_unlock(&journal->j_list_lock);
 				/* Write out all data to prevent deadlocks */
-				journal_do_submit_data(wbuf, bufs);
+				journal_do_submit_data(wbuf, bufs, write_op);
 				bufs = 0;
 				lock_buffer(bh);
 				spin_lock(&journal->j_list_lock);
@@ -256,7 +259,7 @@
 			jbd_unlock_bh_state(bh);
 			if (bufs == journal->j_wbufsize) {
 				spin_unlock(&journal->j_list_lock);
-				journal_do_submit_data(wbuf, bufs);
+				journal_do_submit_data(wbuf, bufs, write_op);
 				bufs = 0;
 				goto write_out_data;
 			}
@@ -286,7 +289,7 @@
 		}
 	}
 	spin_unlock(&journal->j_list_lock);
-	journal_do_submit_data(wbuf, bufs);
+	journal_do_submit_data(wbuf, bufs, write_op);
  
 	return err;
 }
@@ -315,6 +318,7 @@
 	int first_tag = 0;
 	int tag_flag;
 	int i;
+	int write_op = WRITE;
  
 	/*
 	 * First job: lock down the current transaction and wait for
@@ -347,6 +351,8 @@
 	spin_lock(&journal->j_state_lock);
 	commit_transaction->t_state = T_LOCKED;
  
+	if (commit_transaction->t_synchronous_commit)
+		write_op = WRITE_SYNC;
 	spin_lock(&commit_transaction->t_handle_lock);
 	while (commit_transaction->t_updates) {
 		DEFINE_WAIT(wait);
@@ -431,7 +437,8 @@
 	 * Now start flushing things to disk, in the order they appear
 	 * on the transaction lists.  Data blocks go first.
 	 */
-	err = journal_submit_data_buffers(journal, commit_transaction);
+	err = journal_submit_data_buffers(journal, commit_transaction,
+					  write_op);
  
 	/*
 	 * Wait for all previously submitted IO to complete.
@@ -660,7 +667,7 @@
 				clear_buffer_dirty(bh);
 				set_buffer_uptodate(bh);
 				bh->b_end_io = journal_end_buffer_io_sync;
-				submit_bh(WRITE, bh);
+				submit_bh(write_op, bh);
 			}
 			cond_resched();
  
@@ -1440,6 +1440,8 @@
 		}
 	}
  
+	if (handle->h_sync)
+		transaction->t_synchronous_commit = 1;
 	current->journal_info = NULL;
 	spin_lock(&journal->j_state_lock);
 	spin_lock(&transaction->t_handle_lock);
@@ -208,6 +208,7 @@
 #define EXT3_STATE_JDATA		0x00000001 /* journaled data exists */
 #define EXT3_STATE_NEW			0x00000002 /* inode is newly created */
 #define EXT3_STATE_XATTR		0x00000004 /* has in-inode xattrs */
+#define EXT3_STATE_FLUSH_ON_CLOSE	0x00000008
  
 /* Used to pass group descriptor data when online resize is done */
 struct ext3_new_group_input {
@@ -552,6 +552,11 @@
 	 */
 	int t_handle_count;
  
+	/*
+	 * This transaction is being forced and some process is
+	 * waiting for it to finish.
+	 */
+	int t_synchronous_commit:1;
 };
  
 /**
...	...	@@ -1595,6 +1595,7 @@
1595	1595	struct buffer_head bh, head;
1596	1596	const unsigned blocksize = 1 << inode->i_blkbits;
1597	1597	int nr_underway = 0;
	1598	+ int write_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
1598	1599
1599	1600	BUG_ON(!PageLocked(page));
1600	1601
...	...	@@ -1686,7 +1687,7 @@
1686	1687	do {
1687	1688	struct buffer_head *next = bh->b_this_page;
1688	1689	if (buffer_async_write(bh)) {
1689		- submit_bh(WRITE, bh);
	1690	+ submit_bh(write_op, bh);
1690	1691	nr_underway++;
1691	1692	}
1692	1693	bh = next;
...	...	@@ -1740,7 +1741,7 @@
1740	1741	struct buffer_head *next = bh->b_this_page;
1741	1742	if (buffer_async_write(bh)) {
1742	1743	clear_buffer_dirty(bh);
1743		- submit_bh(WRITE, bh);
	1744	+ submit_bh(write_op, bh);
1744	1745	nr_underway++;
1745	1746	}
1746	1747	bh = next;
...	...	@@ -33,6 +33,10 @@
33	33	*/
34	34	static int ext3_release_file (struct inode * inode, struct file * filp)
35	35	{
	36	+ if (EXT3_I(inode)->i_state & EXT3_STATE_FLUSH_ON_CLOSE) {
	37	+ filemap_flush(inode->i_mapping);
	38	+ EXT3_I(inode)->i_state &= ~EXT3_STATE_FLUSH_ON_CLOSE;
	39	+ }
36	40	/* if we are the last writer on the inode, drop the block reservation */
37	41	if ((filp->f_mode & FMODE_WRITE) &&
38	42	(atomic_read(&inode->i_writecount) == 1))
...	...	@@ -2363,6 +2363,9 @@
2363	2363	if (!ext3_can_truncate(inode))
2364	2364	return;
2365	2365
	2366	+ if (inode->i_size == 0 && ext3_should_writeback_data(inode))
	2367	+ ei->i_state \|= EXT3_STATE_FLUSH_ON_CLOSE;
	2368	+
2366	2369	/*
2367	2370	* We have to lock the EOF page here, because lock_page() nests
2368	2371	* outside journal_start().
...	...	@@ -2274,7 +2274,7 @@
2274	2274	struct inode * old_inode, * new_inode;
2275	2275	struct buffer_head * old_bh, * new_bh, * dir_bh;
2276	2276	struct ext3_dir_entry_2 * old_de, * new_de;
2277		- int retval;
	2277	+ int retval, flush_file = 0;
2278	2278
2279	2279	old_bh = new_bh = dir_bh = NULL;
2280	2280
...	...	@@ -2410,6 +2410,8 @@
2410	2410	ext3_mark_inode_dirty(handle, new_inode);
2411	2411	if (!new_inode->i_nlink)
2412	2412	ext3_orphan_add(handle, new_inode);
	2413	+ if (ext3_should_writeback_data(new_inode))
	2414	+ flush_file = 1;
2413	2415	}
2414	2416	retval = 0;
2415	2417
...	...	@@ -2418,6 +2420,8 @@
2418	2420	brelse (old_bh);
2419	2421	brelse (new_bh);
2420	2422	ext3_journal_stop(handle);
	2423	+ if (retval == 0 && flush_file)
	2424	+ filemap_flush(old_inode->i_mapping);
2421	2425	return retval;
2422	2426	}
2423	2427
...	...	@@ -20,6 +20,7 @@
20	20	#include <linux/slab.h>
21	21	#include <linux/mm.h>
22	22	#include <linux/pagemap.h>
	23	+#include <linux/bio.h>
23	24
24	25	/*
25	26	* Default IO end handler for temporary BJ_IO buffer_heads.
26	27
...	...	@@ -171,14 +172,15 @@
171	172	return (ret == -EIO);
172	173	}
173	174
174		-static void journal_do_submit_data(struct buffer_head **wbuf, int bufs)
	175	+static void journal_do_submit_data(struct buffer_head **wbuf, int bufs,
	176	+ int write_op)
175	177	{
176	178	int i;
177	179
178	180	for (i = 0; i < bufs; i++) {
179	181	wbuf[i]->b_end_io = end_buffer_write_sync;
180	182	/* We use-up our safety reference in submit_bh() */
181		- submit_bh(WRITE, wbuf[i]);
	183	+ submit_bh(write_op, wbuf[i]);
182	184	}
183	185	}
184	186
...	...	@@ -186,7 +188,8 @@
186	188	* Submit all the data buffers to disk
187	189	*/
188	190	static int journal_submit_data_buffers(journal_t *journal,
189		- transaction_t *commit_transaction)
	191	+ transaction_t *commit_transaction,
	192	+ int write_op)
190	193	{
191	194	struct journal_head *jh;
192	195	struct buffer_head *bh;
...	...	@@ -225,7 +228,7 @@
225	228	BUFFER_TRACE(bh, "needs blocking lock");
226	229	spin_unlock(&journal->j_list_lock);
227	230	/* Write out all data to prevent deadlocks */
228		- journal_do_submit_data(wbuf, bufs);
	231	+ journal_do_submit_data(wbuf, bufs, write_op);
229	232	bufs = 0;
230	233	lock_buffer(bh);
231	234	spin_lock(&journal->j_list_lock);
...	...	@@ -256,7 +259,7 @@
256	259	jbd_unlock_bh_state(bh);
257	260	if (bufs == journal->j_wbufsize) {
258	261	spin_unlock(&journal->j_list_lock);
259		- journal_do_submit_data(wbuf, bufs);
	262	+ journal_do_submit_data(wbuf, bufs, write_op);
260	263	bufs = 0;
261	264	goto write_out_data;
262	265	}
...	...	@@ -286,7 +289,7 @@
286	289	}
287	290	}
288	291	spin_unlock(&journal->j_list_lock);
289		- journal_do_submit_data(wbuf, bufs);
	292	+ journal_do_submit_data(wbuf, bufs, write_op);
290	293
291	294	return err;
292	295	}
...	...	@@ -315,6 +318,7 @@
315	318	int first_tag = 0;
316	319	int tag_flag;
317	320	int i;
	321	+ int write_op = WRITE;
318	322
319	323	/*
320	324	* First job: lock down the current transaction and wait for
...	...	@@ -347,6 +351,8 @@
347	351	spin_lock(&journal->j_state_lock);
348	352	commit_transaction->t_state = T_LOCKED;
349	353
	354	+ if (commit_transaction->t_synchronous_commit)
	355	+ write_op = WRITE_SYNC;
350	356	spin_lock(&commit_transaction->t_handle_lock);
351	357	while (commit_transaction->t_updates) {
352	358	DEFINE_WAIT(wait);
...	...	@@ -431,7 +437,8 @@
431	437	* Now start flushing things to disk, in the order they appear
432	438	* on the transaction lists. Data blocks go first.
433	439	*/
434		- err = journal_submit_data_buffers(journal, commit_transaction);
	440	+ err = journal_submit_data_buffers(journal, commit_transaction,
	441	+ write_op);
435	442
436	443	/*
437	444	* Wait for all previously submitted IO to complete.
...	...	@@ -660,7 +667,7 @@
660	667	clear_buffer_dirty(bh);
661	668	set_buffer_uptodate(bh);
662	669	bh->b_end_io = journal_end_buffer_io_sync;
663		- submit_bh(WRITE, bh);
	670	+ submit_bh(write_op, bh);
664	671	}
665	672	cond_resched();
666	673
...	...	@@ -1440,6 +1440,8 @@
1440	1440	}
1441	1441	}
1442	1442
	1443	+ if (handle->h_sync)
	1444	+ transaction->t_synchronous_commit = 1;
1443	1445	current->journal_info = NULL;
1444	1446	spin_lock(&journal->j_state_lock);
1445	1447	spin_lock(&transaction->t_handle_lock);
...	...	@@ -208,6 +208,7 @@
208	208	#define EXT3_STATE_JDATA 0x00000001 /* journaled data exists */
209	209	#define EXT3_STATE_NEW 0x00000002 /* inode is newly created */
210	210	#define EXT3_STATE_XATTR 0x00000004 /* has in-inode xattrs */
	211	+#define EXT3_STATE_FLUSH_ON_CLOSE 0x00000008
211	212
212	213	/* Used to pass group descriptor data when online resize is done */
213	214	struct ext3_new_group_input {
...	...	@@ -552,6 +552,11 @@
552	552	*/
553	553	int t_handle_count;
554	554
	555	+ /*
	556	+ * This transaction is being forced and some process is
	557	+ * waiting for it to finish.
	558	+ */
	559	+ int t_synchronous_commit:1;
555	560	};
556	561
557	562	/**