Skip to contentSkip to conversation

dm-integrity: Fix flush with external metadata device

PATCH In review Lukas Straub 3 participants 8 messages Jan 4, 2021 – Jan 10, 2021 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dm-integrity: Fix flush with external metadata device

With an external metadata device, flush requests aren't passed down to the data device.

Fix this by issuing flush in the right places: In integrity_commit when not in journal mode, in do_journal_write after writing the contents of the journal to the disk and in dm_integrity_postsuspend.

---
 drivers/md/dm-integrity.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/md/dm-integrity.c +8 -0
index 5a7a1b90e671..a26ed65869f6 100644
@@ -2196,6 +2196,8 @@ static void integrity_commit(struct work_struct *w)
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
 		dm_integrity_flush_buffers(ic);
+		if (ic->meta_dev)
+			blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
 		goto release_flush_bios;
 	}
 
@@ -2410,6 +2412,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
 	wait_for_completion_io(&comp.comp);
 
 	dm_integrity_flush_buffers(ic);
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2949,6 +2954,9 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 #endif
 	}
 
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 
 	ic->journal_uptodate = true;
Signature · 1 line
2.20.1

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

#1 lore raw
Quoted 36 lines · On Sun, 20 Dec 2020 14:02:22 +0100 Lukas Straub <[email protected]> wrote:

With an external metadata device, flush requests aren't passed down to the data device.

Fix this by issuing flush in the right places: In integrity_commit when not in journal mode, in do_journal_write after writing the contents of the journal to the disk and in dm_integrity_postsuspend.

---
 drivers/md/dm-integrity.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/md/dm-integrity.c +5 -0
index 5a7a1b90e671..a26ed65869f6 100644
@@ -2196,6 +2196,8 @@ static void integrity_commit(struct work_struct *w)
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
 		dm_integrity_flush_buffers(ic);
+		if (ic->meta_dev)
+			blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
 		goto release_flush_bios;
 	}
 
@@ -2410,6 +2412,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
 	wait_for_completion_io(&comp.comp);
 
 	dm_integrity_flush_buffers(ic);
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 }
 
 static void integrity_writer(struct work_struct *w)

drivers/md/dm-integrity.c static void dm_integrity_postsuspend(struct dm_target *ti) · line 2962

Quoted 9 lines
drivers/md/dm-integrity.c +3 -0
@@ -2949,6 +2954,9 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 #endif
 	}
 
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 
 	ic->journal_uptodate = true;

Ping...

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

#2 lore raw
In reply to Lukas Straub
Quoted 47 lines · Lukas Straub

With an external metadata device, flush requests aren't passed down to the data device.

Fix this by issuing flush in the right places: In integrity_commit when not in journal mode, in do_journal_write after writing the contents of the journal to the disk and in dm_integrity_postsuspend.

---
 drivers/md/dm-integrity.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/md/dm-integrity.c +8 -0
index 5a7a1b90e671..a26ed65869f6 100644
@@ -2196,6 +2196,8 @@ static void integrity_commit(struct work_struct *w)
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
 		dm_integrity_flush_buffers(ic);
+		if (ic->meta_dev)
+			blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
 		goto release_flush_bios;
 	}
 
@@ -2410,6 +2412,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
 	wait_for_completion_io(&comp.comp);
 
 	dm_integrity_flush_buffers(ic);
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2949,6 +2954,9 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 #endif
 	}
 
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 
 	ic->journal_uptodate = true;
Signature · 1 line
2.20.1

Seems like a pretty bad oversight... but shouldn't you also make sure to flush the data device _before_ the metadata is flushed?

Mike

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

maintainer
#3 lore raw
Quoted 55 lines · Mike Snitzer
Quoted 47 lines · Lukas Straub

With an external metadata device, flush requests aren't passed down to the data device.

Fix this by issuing flush in the right places: In integrity_commit when not in journal mode, in do_journal_write after writing the contents of the journal to the disk and in dm_integrity_postsuspend.

---
 drivers/md/dm-integrity.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/md/dm-integrity.c +8 -0
index 5a7a1b90e671..a26ed65869f6 100644
@@ -2196,6 +2196,8 @@ static void integrity_commit(struct work_struct *w)
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
 		dm_integrity_flush_buffers(ic);
+		if (ic->meta_dev)
+			blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
 		goto release_flush_bios;
 	}
 
@@ -2410,6 +2412,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
 	wait_for_completion_io(&comp.comp);
 
 	dm_integrity_flush_buffers(ic);
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2949,6 +2954,9 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 #endif
 	}
 
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 
 	ic->journal_uptodate = true;
Signature · 1 line
2.20.1

Seems like a pretty bad oversight... but shouldn't you also make sure to flush the data device _before_ the metadata is flushed?

Quoted 1 line

Mike

I think, ordering is not a problem.

A disk may flush its cache spontaneously anytime, so it doesn't matter in which order do we flush them. Similarly a dm-bufio buffer may be flushed anytime - if the machine is running out of memory and a dm-bufio shrinker is called.

I'll send another patch for this - I've created a patch that flushes the metadata device cache and data device cache in parallel, so that performance degradation is reduced.

My patch also doesn't use GFP_NOIO allocation - which can in theory deadlock if we are swapping on dm-integrity device.

Mikulas

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

maintainer
#4 lore raw

PATCH 1/1 dm-integrity: Fix flush with external metadata device

With external metadata device, flush requests are not passed down to the data device.

Fix this by submitting the flush request in dm_integrity_flush_buffers. In order to not degrade performance, we overlap the data device flush with the metadata device flush.

---
 drivers/md/dm-bufio.c     |    6 ++++
 drivers/md/dm-integrity.c |   60 +++++++++++++++++++++++++++++++++++++---------
 include/linux/dm-bufio.h  |    1 
 3 files changed, 56 insertions(+), 11 deletions(-)
drivers/md/dm-integrity.c +49 -11
Index: linux-2.6/drivers/md/dm-integrity.c
===================================================================
@@ -1379,12 +1379,52 @@ thorough_test:
 #undef MAY_BE_HASH
 }
 
-static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
+struct flush_request {
+	struct dm_io_request io_req;
+	struct dm_io_region io_reg;
+	struct dm_integrity_c *ic;
+	struct completion comp;
+};
+
+static void flush_notify(unsigned long error, void *fr_)
+{
+	struct flush_request *fr = fr_;
+	if (unlikely(error != 0))
+		dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
+	complete(&fr->comp);
+}
+
+static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
 {
 	int r;
+
+	struct flush_request fr;
+
+	if (!ic->meta_dev)
+		flush_data = false;
+	if (flush_data) {
+		fr.io_req.bi_op = REQ_OP_WRITE,
+		fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
+		fr.io_req.mem.type = DM_IO_KMEM,
+		fr.io_req.mem.ptr.addr = NULL,
+		fr.io_req.notify.fn = flush_notify,
+		fr.io_req.notify.context = &fr;
+		fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
+		fr.io_reg.bdev = ic->dev->bdev,
+		fr.io_reg.sector = 0,
+		fr.io_reg.count = 0,
+		fr.ic = ic;
+		init_completion(&fr.comp);
+		r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
+		BUG_ON(r);
+	}
+
 	r = dm_bufio_write_dirty_buffers(ic->bufio);
 	if (unlikely(r))
 		dm_integrity_io_error(ic, "writing tags", r);
+
+	if (flush_data)
+		wait_for_completion(&fr.comp);
 }
 
 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
@@ -2110,7 +2150,7 @@ offload_to_thread:
 
 	if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
 		integrity_metadata(&dio->work);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, false);
 
 		dio->in_flight = (atomic_t)ATOMIC_INIT(1);
 		dio->completion = NULL;
@@ -2195,7 +2235,7 @@ static void integrity_commit(struct work
 	flushes = bio_list_get(&ic->flush_bio_list);
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 		goto release_flush_bios;
 	}
 
@@ -2409,7 +2449,7 @@ skip_io:
 	complete_journal_op(&comp);
 	wait_for_completion_io(&comp.comp);
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, true);
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2451,7 +2491,7 @@ static void recalc_write_super(struct dm
 {
 	int r;
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, false);
 	if (dm_integrity_failed(ic))
 		return;
 
@@ -2654,7 +2694,7 @@ static void bitmap_flush_work(struct wor
 	unsigned long limit;
 	struct bio *bio;
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, false);
 
 	range.logical_sector = 0;
 	range.n_sectors = ic->provided_data_sectors;
@@ -2663,9 +2703,7 @@ static void bitmap_flush_work(struct wor
 	add_new_range_and_wait(ic, &range);
 	spin_unlock_irq(&ic->endio_wait.lock);
 
-	dm_integrity_flush_buffers(ic);
-	if (ic->meta_dev)
-		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+	dm_integrity_flush_buffers(ic, true);
 
 	limit = ic->provided_data_sectors;
 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
@@ -2934,11 +2972,11 @@ static void dm_integrity_postsuspend(str
 		if (ic->meta_dev)
 			queue_work(ic->writer_wq, &ic->writer_work);
 		drain_workqueue(ic->writer_wq);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 	}
 
 	if (ic->mode == 'B') {
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 #if 1
 		/* set to 0 to test bitmap replay code */
 		init_journal(ic, 0, ic->journal_sections, 0);
include/linux/dm-bufio.h +1 -0
Index: linux-2.6/include/linux/dm-bufio.h
===================================================================
@@ -150,6 +150,7 @@ void dm_bufio_set_minimum_buffers(struct
 
 unsigned dm_bufio_get_block_size(struct dm_bufio_client *c);
 sector_t dm_bufio_get_device_size(struct dm_bufio_client *c);
+struct dm_io_client *dm_bufio_get_dm_io_client(struct dm_bufio_client *c);
 sector_t dm_bufio_get_block_number(struct dm_buffer *b);
 void *dm_bufio_get_block_data(struct dm_buffer *b);
 void *dm_bufio_get_aux_data(struct dm_buffer *b);
drivers/md/dm-bufio.c +6 -0
Index: linux-2.6/drivers/md/dm-bufio.c
===================================================================
@@ -1534,6 +1534,12 @@ sector_t dm_bufio_get_device_size(struct
 }
 EXPORT_SYMBOL_GPL(dm_bufio_get_device_size);
 
+struct dm_io_client *dm_bufio_get_dm_io_client(struct dm_bufio_client *c)
+{
+	return c->dm_io;
+}
+EXPORT_SYMBOL_GPL(dm_bufio_get_dm_io_client);
+
 sector_t dm_bufio_get_block_number(struct dm_buffer *b)
 {
 	return b->block;

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

author Reviewed-by on v1 1/1
#5 lore raw
Quoted 10 lines · On Fri, 8 Jan 2021 11:15:56 -0500 (EST) Mikulas Patocka <[email protected]>…

With external metadata device, flush requests are not passed down to the data device.

Fix this by submitting the flush request in dm_integrity_flush_buffers. In order to not degrade performance, we overlap the data device flush with the metadata device flush.

Looks good to me.

Regards,
Lukas Straub

Quoted 164 lines
---
 drivers/md/dm-bufio.c     |    6 ++++
 drivers/md/dm-integrity.c |   60 +++++++++++++++++++++++++++++++++++++---------
 include/linux/dm-bufio.h  |    1 
 3 files changed, 56 insertions(+), 11 deletions(-)

Index: linux-2.6/drivers/md/dm-integrity.c
===================================================================

drivers/md/dm-integrity.c +49 -11
@@ -1379,12 +1379,52 @@ thorough_test:
 #undef MAY_BE_HASH
 }
 
-static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
+struct flush_request {
+	struct dm_io_request io_req;
+	struct dm_io_region io_reg;
+	struct dm_integrity_c *ic;
+	struct completion comp;
+};
+
+static void flush_notify(unsigned long error, void *fr_)
+{
+	struct flush_request *fr = fr_;
+	if (unlikely(error != 0))
+		dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
+	complete(&fr->comp);
+}
+
+static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
 {
 	int r;
+
+	struct flush_request fr;
+
+	if (!ic->meta_dev)
+		flush_data = false;
+	if (flush_data) {
+		fr.io_req.bi_op = REQ_OP_WRITE,
+		fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
+		fr.io_req.mem.type = DM_IO_KMEM,
+		fr.io_req.mem.ptr.addr = NULL,
+		fr.io_req.notify.fn = flush_notify,
+		fr.io_req.notify.context = &fr;
+		fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
+		fr.io_reg.bdev = ic->dev->bdev,
+		fr.io_reg.sector = 0,
+		fr.io_reg.count = 0,
+		fr.ic = ic;
+		init_completion(&fr.comp);
+		r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
+		BUG_ON(r);
+	}
+
 	r = dm_bufio_write_dirty_buffers(ic->bufio);
 	if (unlikely(r))
 		dm_integrity_io_error(ic, "writing tags", r);
+
+	if (flush_data)
+		wait_for_completion(&fr.comp);
 }
 
 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
@@ -2110,7 +2150,7 @@ offload_to_thread:
 
 	if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
 		integrity_metadata(&dio->work);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, false);
 
 		dio->in_flight = (atomic_t)ATOMIC_INIT(1);
 		dio->completion = NULL;
@@ -2195,7 +2235,7 @@ static void integrity_commit(struct work
 	flushes = bio_list_get(&ic->flush_bio_list);
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 		goto release_flush_bios;
 	}
 
@@ -2409,7 +2449,7 @@ skip_io:
 	complete_journal_op(&comp);
 	wait_for_completion_io(&comp.comp);
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, true);
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2451,7 +2491,7 @@ static void recalc_write_super(struct dm
 {
 	int r;
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, false);
 	if (dm_integrity_failed(ic))
 		return;
 
@@ -2654,7 +2694,7 @@ static void bitmap_flush_work(struct wor
 	unsigned long limit;
 	struct bio *bio;
 
-	dm_integrity_flush_buffers(ic);
+	dm_integrity_flush_buffers(ic, false);
 
 	range.logical_sector = 0;
 	range.n_sectors = ic->provided_data_sectors;
@@ -2663,9 +2703,7 @@ static void bitmap_flush_work(struct wor
 	add_new_range_and_wait(ic, &range);
 	spin_unlock_irq(&ic->endio_wait.lock);
 
-	dm_integrity_flush_buffers(ic);
-	if (ic->meta_dev)
-		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+	dm_integrity_flush_buffers(ic, true);
 
 	limit = ic->provided_data_sectors;
 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
@@ -2934,11 +2972,11 @@ static void dm_integrity_postsuspend(str
 		if (ic->meta_dev)
 			queue_work(ic->writer_wq, &ic->writer_work);
 		drain_workqueue(ic->writer_wq);
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 	}
 
 	if (ic->mode == 'B') {
-		dm_integrity_flush_buffers(ic);
+		dm_integrity_flush_buffers(ic, true);
 #if 1
 		/* set to 0 to test bitmap replay code */
 		init_journal(ic, 0, ic->journal_sections, 0);

Index: linux-2.6/include/linux/dm-bufio.h
===================================================================

include/linux/dm-bufio.h +1 -0
@@ -150,6 +150,7 @@ void dm_bufio_set_minimum_buffers(struct
 
 unsigned dm_bufio_get_block_size(struct dm_bufio_client *c);
 sector_t dm_bufio_get_device_size(struct dm_bufio_client *c);
+struct dm_io_client *dm_bufio_get_dm_io_client(struct dm_bufio_client *c);
 sector_t dm_bufio_get_block_number(struct dm_buffer *b);
 void *dm_bufio_get_block_data(struct dm_buffer *b);
 void *dm_bufio_get_aux_data(struct dm_buffer *b);

Index: linux-2.6/drivers/md/dm-bufio.c
===================================================================

drivers/md/dm-bufio.c +6 -0
@@ -1534,6 +1534,12 @@ sector_t dm_bufio_get_device_size(struct
 }
 EXPORT_SYMBOL_GPL(dm_bufio_get_device_size);
 
+struct dm_io_client *dm_bufio_get_dm_io_client(struct dm_bufio_client *c)
+{
+	return c->dm_io;
+}
+EXPORT_SYMBOL_GPL(dm_bufio_get_dm_io_client);
+
 sector_t dm_bufio_get_block_number(struct dm_buffer *b)
 {
 	return b->block;

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

#6 lore raw
In reply to Mikulas Patocka
Quoted 68 lines · Mikulas Patocka
Quoted 55 lines · Mike Snitzer
Quoted 47 lines · Lukas Straub

With an external metadata device, flush requests aren't passed down to the data device.

Fix this by issuing flush in the right places: In integrity_commit when not in journal mode, in do_journal_write after writing the contents of the journal to the disk and in dm_integrity_postsuspend.

---
 drivers/md/dm-integrity.c | 8 ++++++++
 1 file changed, 8 insertions(+)
drivers/md/dm-integrity.c +8 -0
index 5a7a1b90e671..a26ed65869f6 100644
@@ -2196,6 +2196,8 @@ static void integrity_commit(struct work_struct *w)
 	if (unlikely(ic->mode != 'J')) {
 		spin_unlock_irq(&ic->endio_wait.lock);
 		dm_integrity_flush_buffers(ic);
+		if (ic->meta_dev)
+			blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
 		goto release_flush_bios;
 	}
 
@@ -2410,6 +2412,9 @@ static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
 	wait_for_completion_io(&comp.comp);
 
 	dm_integrity_flush_buffers(ic);
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 }
 
 static void integrity_writer(struct work_struct *w)
@@ -2949,6 +2954,9 @@ static void dm_integrity_postsuspend(struct dm_target *ti)
 #endif
 	}
 
+	if (ic->meta_dev)
+		blkdev_issue_flush(ic->dev->bdev, GFP_NOIO);
+
 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
 
 	ic->journal_uptodate = true;
Signature · 1 line
2.20.1

Seems like a pretty bad oversight... but shouldn't you also make sure to flush the data device _before_ the metadata is flushed?

Quoted 1 line

Mike

I think, ordering is not a problem.

A disk may flush its cache spontaneously anytime, so it doesn't matter in which order do we flush them. Similarly a dm-bufio buffer may be flushed anytime - if the machine is running out of memory and a dm-bufio shrinker is called.

Quoted 6 lines

I'll send another patch for this - I've created a patch that flushes the metadata device cache and data device cache in parallel, so that performance degradation is reduced.

My patch also doesn't use GFP_NOIO allocation - which can in theory deadlock if we are swapping on dm-integrity device.

OK, I see your patch, but my concern about ordering was more to do with crash consistency. What if metadata is updated but data isn't?

On the surface, your approach of issuing the flushes in parallel seems to expose us to inconsistency upon recovery from a crash. If that isn't the case please explain why not.

Thanks,
Mike

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel

maintainer
#7 lore raw
Quoted 16 lines · Mike Snitzer
Quoted 16 lines
Quoted 4 lines

Seems like a pretty bad oversight... but shouldn't you also make sure to flush the data device _before_ the metadata is flushed?

Mike

I think, ordering is not a problem.

A disk may flush its cache spontaneously anytime, so it doesn't matter in which order do we flush them. Similarly a dm-bufio buffer may be flushed anytime - if the machine is running out of memory and a dm-bufio shrinker is called.

I'll send another patch for this - I've created a patch that flushes the metadata device cache and data device cache in parallel, so that performance degradation is reduced.

Quoted 5 lines
Quoted 2 lines

My patch also doesn't use GFP_NOIO allocation - which can in theory deadlock if we are swapping on dm-integrity device.

OK, I see your patch, but my concern about ordering was more to do with crash consistency. What if metadata is updated but data isn't?

In journal mode: do_journal_write will copy data from the journal to appropriate places and then flushes both data and metadata device. If a crash happens during flush, the journal will be replayed on next reboot - the replay operation will overwrite both data and metadata from the journal.

In bitmap mode: bitmap_flush_work will first issue flush on both data and metadata device, and then clear the dirty bits in a bitmap. If a crash happens during the flush, the bits in the bitmap are still set and the checksums will be recalculated on next reboot.

Quoted 3 lines

On the surface, your approach of issuing the flushes in parallel seems to expose us to inconsistency upon recovery from a crash. If that isn't the case please explain why not.

The disk may flush its internal cache anytime. So, if you do "blkdev_issue_flush(disk1); blkdev_issue_flush(disk2);" there is no ordering guarantee at all. The firmware in disk2 may flush the cache spontaneously, before you called blkdev_issue_flush(disk1).

Quoted 2 lines

Thanks,
Mike

Mikulas

--
dm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/dm-devel