From: Jaegeuk Kim <jaegeuk@kernel.org>
To: Bart Van Assche <bvanassche@acm.org>
Cc: Jens Axboe <axboe@kernel.dk>, linux-block@vger.kernel.org
Subject: Re: [PATCH v2] loop: avoid EAGAIN, if offset or block_size are changed
Date: Tue, 26 Nov 2019 17:09:26 -0800 [thread overview]
Message-ID: <20191127010926.GA34613@jaegeuk-macbookpro.roam.corp.google.com> (raw)
In-Reply-To: <3ca36251-57c4-b62c-c029-77b643ddea77@acm.org>
On 11/26, Bart Van Assche wrote:
> On 11/26/19 4:04 PM, Jaegeuk Kim wrote:
> > Subject: [PATCH] loop: avoid EAGAIN, if offset or block_size are changed
> >
> > This patch tries to avoid EAGAIN due to nrpages!=0 that was originally trying
> > to drop stale pages resulting in wrong data access.
>
> Does this patch remove all code that returns EAGAIN from the code paths used
> for changing the offset and block size? If so, please make the commit
> message more affirmative.
>
> > if (lo->lo_offset != info->lo_offset ||
> > - lo->lo_sizelimit != info->lo_sizelimit) {
> > - sync_blockdev(lo->lo_device);
> > - kill_bdev(lo->lo_device);
> > - }
> > + lo->lo_sizelimit != info->lo_sizelimit)
> > + drop_caches = true;
>
> If the offset is changed and dirty pages are only flushed after the loop
> device offset has been changed, can that cause data to be written at a wrong
> LBA? In other words, I'd like to keep a sync_blockdev() call here.
>
> > + /* truncate stale pages cached by previous operations */
> > + if (!err && drop_caches) {
> > + sync_blockdev(lo->lo_device);
> > + invalidate_bdev(lo->lo_device);
> > + }
>
> Is the invalidate_bdev() call necessary here?
We need this to reload 4KB-sized buffer caches back.
How about this?
From ceef42dbf4ec74c34d58125a20cc11ef13e2e1c4 Mon Sep 17 00:00:00 2001
From: Jaegeuk Kim <jaegeuk@kernel.org>
Date: Fri, 17 May 2019 16:37:50 -0700
Subject: [PATCH] loop: avoid EAGAIN, if offset or block_size are changed
Previously, there was a bug where user could see stale buffer cache (e.g, 512B)
attached in the 4KB-sized pager cache, when the block size was changed from
512B to 4KB. That was fixed by:
commit 5db470e229e2 ("loop: drop caches if offset or block_size are changed")
But, there were some regression reports saying the fix returns EAGAIN easily.
So, this patch removes previously added EAGAIN condition, nrpages != 0.
Instead, it changes the flow like this:
- sync_blockdev()
- blk_mq_freeze_queue()
: change the loop configuration
- blk_mq_unfreeze_queue()
- sync_blockdev()
- invalidate_bdev()
After invalidating the buffer cache, we must see the full valid 4KB page.
Additional concern came from Bart in which we can lose some data when
changing the lo_offset. In that case, this patch adds:
- sync_blockdev()
- blk_set_queue_dying
- blk_mq_freeze_queue()
: change the loop configuration
- blk_mq_unfreeze_queue()
- blk_queue_flag_clear(QUEUE_FLAG_DYING);
- sync_blockdev()
- invalidate_bdev()
Report: https://bugs.chromium.org/p/chromium/issues/detail?id=938958#c38
Cc: <stable@vger.kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Cc: Bart Van Assche <bvanassche@acm.org>
Fixes: 5db470e229e2 ("loop: drop caches if offset or block_size are changed")
Reported-by: Gwendal Grignou <gwendal@chromium.org>
Reported-by: grygorii tertychnyi <gtertych@cisco.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
drivers/block/loop.c | 59 ++++++++++++++++++++++----------------------
1 file changed, 29 insertions(+), 30 deletions(-)
diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index f6f77eaa7217..9c1985de85e0 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -1232,6 +1232,8 @@ loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
kuid_t uid = current_uid();
struct block_device *bdev;
bool partscan = false;
+ bool drop_request = false;
+ bool drop_cache = false;
err = mutex_lock_killable(&loop_ctl_mutex);
if (err)
@@ -1251,11 +1253,16 @@ loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
goto out_unlock;
}
+ if (lo->lo_offset != info->lo_offset)
+ drop_request = true;
if (lo->lo_offset != info->lo_offset ||
- lo->lo_sizelimit != info->lo_sizelimit) {
- sync_blockdev(lo->lo_device);
- kill_bdev(lo->lo_device);
- }
+ lo->lo_sizelimit != info->lo_sizelimit)
+ drop_cache = true;
+
+ sync_blockdev(lo->lo_device);
+
+ if (drop_request)
+ blk_set_queue_dying(lo->lo_queue);
/* I/O need to be drained during transfer transition */
blk_mq_freeze_queue(lo->lo_queue);
@@ -1285,14 +1292,6 @@ loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
if (lo->lo_offset != info->lo_offset ||
lo->lo_sizelimit != info->lo_sizelimit) {
- /* kill_bdev should have truncated all the pages */
- if (lo->lo_device->bd_inode->i_mapping->nrpages) {
- err = -EAGAIN;
- pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
- __func__, lo->lo_number, lo->lo_file_name,
- lo->lo_device->bd_inode->i_mapping->nrpages);
- goto out_unfreeze;
- }
if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit)) {
err = -EFBIG;
goto out_unfreeze;
@@ -1329,6 +1328,8 @@ loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
out_unfreeze:
blk_mq_unfreeze_queue(lo->lo_queue);
+ if (drop_request)
+ blk_queue_flag_clear(QUEUE_FLAG_DYING, lo->lo_queue);
if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) &&
!(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
@@ -1337,6 +1338,12 @@ loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
bdev = lo->lo_device;
partscan = true;
}
+
+ /* truncate stale pages cached by previous operations */
+ if (!err && drop_cache) {
+ sync_blockdev(lo->lo_device);
+ invalidate_bdev(lo->lo_device);
+ }
out_unlock:
mutex_unlock(&loop_ctl_mutex);
if (partscan)
@@ -1518,7 +1525,7 @@ static int loop_set_dio(struct loop_device *lo, unsigned long arg)
static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
{
- int err = 0;
+ bool drop_cache = false;
if (lo->lo_state != Lo_bound)
return -ENXIO;
@@ -1526,31 +1533,23 @@ static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg))
return -EINVAL;
- if (lo->lo_queue->limits.logical_block_size != arg) {
- sync_blockdev(lo->lo_device);
- kill_bdev(lo->lo_device);
- }
+ if (lo->lo_queue->limits.logical_block_size != arg)
+ drop_cache = true;
+ sync_blockdev(lo->lo_device);
blk_mq_freeze_queue(lo->lo_queue);
-
- /* kill_bdev should have truncated all the pages */
- if (lo->lo_queue->limits.logical_block_size != arg &&
- lo->lo_device->bd_inode->i_mapping->nrpages) {
- err = -EAGAIN;
- pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n",
- __func__, lo->lo_number, lo->lo_file_name,
- lo->lo_device->bd_inode->i_mapping->nrpages);
- goto out_unfreeze;
- }
-
blk_queue_logical_block_size(lo->lo_queue, arg);
blk_queue_physical_block_size(lo->lo_queue, arg);
blk_queue_io_min(lo->lo_queue, arg);
loop_update_dio(lo);
-out_unfreeze:
blk_mq_unfreeze_queue(lo->lo_queue);
- return err;
+ /* truncate stale pages cached by previous operations */
+ if (drop_cache) {
+ sync_blockdev(lo->lo_device);
+ invalidate_bdev(lo->lo_device);
+ }
+ return 0;
}
static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
--
2.19.0.605.g01d371f741-goog
next prev parent reply other threads:[~2019-11-27 1:09 UTC|newest]
Thread overview: 25+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-05-18 0:47 [PATCH] loop: avoid EAGAIN, if offset or block_size are changed Jaegeuk Kim
2019-05-18 0:53 ` [PATCH v2] " Jaegeuk Kim
2019-06-17 21:08 ` [f2fs-dev] " Jaegeuk Kim
2019-11-18 18:36 ` Andrew Norrie
2019-11-19 4:00 ` Greg KH
2019-11-19 23:40 ` [PATCH v2] " Bart Van Assche
2019-11-25 17:59 ` Jaegeuk Kim
2019-11-25 18:35 ` Bart Van Assche
2019-11-25 19:22 ` Jaegeuk Kim
2019-11-25 19:41 ` Bart Van Assche
2019-11-25 22:27 ` Bart Van Assche
2019-11-26 18:29 ` Jaegeuk Kim
2019-11-26 18:59 ` Bart Van Assche
2019-11-26 22:32 ` Jaegeuk Kim
2019-11-26 22:54 ` Bart Van Assche
2019-11-27 0:04 ` Jaegeuk Kim
2019-11-27 0:26 ` Bart Van Assche
2019-11-27 1:09 ` Jaegeuk Kim [this message]
2019-11-27 16:35 ` Bart Van Assche
2019-11-27 18:17 ` Jaegeuk Kim
2019-11-27 18:18 ` [f2fs-dev] [PATCH v3] " Jaegeuk Kim
2019-11-27 18:54 ` Bart Van Assche
2020-02-19 19:58 ` Andrew Norrie
2020-03-05 21:04 ` [PATCH] " Jan Kara
2019-06-10 21:49 [PATCH v2] loop: avoid EAGAIN, if offset or block size " Francesco Ruggeri
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20191127010926.GA34613@jaegeuk-macbookpro.roam.corp.google.com \
--to=jaegeuk@kernel.org \
--cc=axboe@kernel.dk \
--cc=bvanassche@acm.org \
--cc=linux-block@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).