* [PATCH] enable bypass raid5 journal for full stripe writes
@ 2015-09-04 6:01 Song Liu
2015-09-30 8:03 ` Neil Brown
0 siblings, 1 reply; 2+ messages in thread
From: Song Liu @ 2015-09-04 6:01 UTC (permalink / raw)
To: linux-raid; +Cc: shli, neilb, hch, dan.j.williams, hch, Kernel-team, Song Liu
Summary:
To save life time of journal device, we can config the device to
bypass journal writes for full stripe write. This is configured by:
echo "yes" > /sys/block/mdX/md/r5l_bypass_full_stripe
and
echo "no" > /sys/block/mdX/md/r5l_bypass_full_stripe
For file system integrity, full stripe with REQ_FUA will still
write to journal first.
This patch applies on top of Shaohua's most recent patches:
http://marc.info/?l=linux-raid&m=144122700510667
Signed-off-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Shaohua Li <shli@fb.com>
---
drivers/md/raid5-cache.c | 20 +++++++++++++++++
drivers/md/raid5.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++
drivers/md/raid5.h | 2 ++
3 files changed, 78 insertions(+)
diff --git a/drivers/md/raid5-cache.c b/drivers/md/raid5-cache.c
index 410b85b..0c3ddc5 100644
--- a/drivers/md/raid5-cache.c
+++ b/drivers/md/raid5-cache.c
@@ -82,6 +82,8 @@ struct r5l_log {
struct list_head no_space_stripes; /* pending stripes, log has no space */
spinlock_t no_space_stripes_lock;
+
+ int bypass_full_stripe;
};
/*
@@ -438,6 +440,7 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
int meta_size;
int reserve;
int i;
+ int fua = 0;
if (!log)
return -EAGAIN;
@@ -453,6 +456,8 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
void *addr;
if (!test_bit(R5_Wantwrite, &sh->dev[i].flags))
continue;
+ if (test_bit(R5_WantFUA, &sh->dev[i].flags))
+ fua = 1;
write_disks++;
/* checksum is already calculated in last run */
if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
@@ -462,6 +467,10 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
addr, PAGE_SIZE);
kunmap_atomic(addr);
}
+
+ if (log->bypass_full_stripe && (write_disks == sh->disks) && (!fua))
+ return -EAGAIN; /* bypass journal device */
+
parity_pages = 1 + !!(sh->qd_idx >= 0);
data_pages = write_disks - parity_pages;
@@ -520,6 +529,16 @@ int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio)
return -EAGAIN;
}
+int r5l_get_bypass_full_stripe(struct r5l_log *log)
+{
+ return log->bypass_full_stripe;
+}
+
+void r5l_set_bypass_full_stripe(struct r5l_log *log, int val)
+{
+ log->bypass_full_stripe = val;
+}
+
/* This will run after log space is reclaimed */
static void r5l_run_no_space_stripes(struct r5l_log *log)
{
@@ -1105,6 +1124,7 @@ int r5l_init_log(struct r5conf *conf, struct md_rdev *rdev)
if (!log->io_kc)
goto io_kc;
+ log->bypass_full_stripe = 0;
log->reclaim_thread = md_register_thread(r5l_reclaim_thread,
log->rdev->mddev, "reclaim");
if (!log->reclaim_thread)
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 394cdf8..5781987 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -6223,6 +6223,61 @@ raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
raid5_show_group_thread_cnt,
raid5_store_group_thread_cnt);
+static ssize_t
+r5l_show_bypass_full_stripe(struct mddev *mddev, char *page)
+{
+ struct r5conf *conf;
+ int ret = 0;
+
+ spin_lock(&mddev->lock);
+ conf = mddev->private;
+ if (conf) {
+ if (conf->log)
+ ret = sprintf(page, "%s\n",
+ r5l_get_bypass_full_stripe(conf->log) ? "yes" : "no");
+ else
+ ret = sprintf(page, "n/a\n");
+ }
+ spin_unlock(&mddev->lock);
+ return ret;
+}
+
+static ssize_t
+r5l_store_bypass_full_stripe(struct mddev *mddev, const char *page, size_t len)
+{
+ struct r5conf *conf;
+ int err = 0;
+ int val;
+
+ if (strncmp(page, "yes", 3) == 0 &&
+ (page[3] == '\n' || page[3] == '\0'))
+ val = 1;
+ else if (strncmp(page, "no", 2) == 0 &&
+ (page[2] == '\n' || page[2] == '\0'))
+ val = 0;
+ else
+ return -EINVAL;
+
+ mddev_suspend(mddev);
+ spin_lock(&mddev->lock);
+ conf = mddev->private;
+ if (conf) {
+ if (conf->log) {
+ r5l_set_bypass_full_stripe(conf->log, val);
+ } else
+ err = -EINVAL;
+ } else
+ err = -ENODEV;
+ spin_unlock(&mddev->lock);
+ mddev_resume(mddev);
+ return err ?: len;
+}
+
+static struct md_sysfs_entry
+r5l_bypass_full_stripe = __ATTR(r5l_bypass_full_stripe, S_IRUGO | S_IWUSR,
+ r5l_show_bypass_full_stripe,
+ r5l_store_bypass_full_stripe);
+
static struct attribute *raid5_attrs[] = {
&raid5_stripecache_size.attr,
&raid5_stripecache_active.attr,
@@ -6230,6 +6285,7 @@ static struct attribute *raid5_attrs[] = {
&raid5_group_thread_cnt.attr,
&raid5_skip_copy.attr,
&raid5_rmw_level.attr,
+ &r5l_bypass_full_stripe.attr,
NULL,
};
static struct attribute_group raid5_attrs_group = {
diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
index e6b9a40..c1f6935 100644
--- a/drivers/md/raid5.h
+++ b/drivers/md/raid5.h
@@ -630,4 +630,6 @@ extern void r5l_write_stripe_run(struct r5l_log *log);
extern void r5l_flush_stripe_to_raid(struct r5l_log *log);
extern void r5l_stripe_write_finished(struct stripe_head *sh);
extern int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio);
+extern int r5l_get_bypass_full_stripe(struct r5l_log *log);
+extern void r5l_set_bypass_full_stripe(struct r5l_log *log, int val);
#endif
--
1.8.1
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH] enable bypass raid5 journal for full stripe writes
2015-09-04 6:01 [PATCH] enable bypass raid5 journal for full stripe writes Song Liu
@ 2015-09-30 8:03 ` Neil Brown
0 siblings, 0 replies; 2+ messages in thread
From: Neil Brown @ 2015-09-30 8:03 UTC (permalink / raw)
To: linux-raid; +Cc: shli, hch, dan.j.williams, hch, Kernel-team, Song Liu
[-- Attachment #1: Type: text/plain, Size: 5895 bytes --]
Song Liu <songliubraving@fb.com> writes:
> Summary:
> To save life time of journal device, we can config the device to
> bypass journal writes for full stripe write. This is configured by:
>
> echo "yes" > /sys/block/mdX/md/r5l_bypass_full_stripe
>
> and
>
> echo "no" > /sys/block/mdX/md/r5l_bypass_full_stripe
>
> For file system integrity, full stripe with REQ_FUA will still
> write to journal first.
>
> This patch applies on top of Shaohua's most recent patches:
>
> http://marc.info/?l=linux-raid&m=144122700510667
This just re-introduces the write hole.
You have no guarantee that all of the writes will complete before a
crash, but some might.
So after a crash on a degraded array, you still get unreliable data.
NeilBrown
>
> Signed-off-by: Song Liu <songliubraving@fb.com>
> Reviewed-by: Shaohua Li <shli@fb.com>
> ---
> drivers/md/raid5-cache.c | 20 +++++++++++++++++
> drivers/md/raid5.c | 56 ++++++++++++++++++++++++++++++++++++++++++++++++
> drivers/md/raid5.h | 2 ++
> 3 files changed, 78 insertions(+)
>
> diff --git a/drivers/md/raid5-cache.c b/drivers/md/raid5-cache.c
> index 410b85b..0c3ddc5 100644
> --- a/drivers/md/raid5-cache.c
> +++ b/drivers/md/raid5-cache.c
> @@ -82,6 +82,8 @@ struct r5l_log {
>
> struct list_head no_space_stripes; /* pending stripes, log has no space */
> spinlock_t no_space_stripes_lock;
> +
> + int bypass_full_stripe;
> };
>
> /*
> @@ -438,6 +440,7 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
> int meta_size;
> int reserve;
> int i;
> + int fua = 0;
>
> if (!log)
> return -EAGAIN;
> @@ -453,6 +456,8 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
> void *addr;
> if (!test_bit(R5_Wantwrite, &sh->dev[i].flags))
> continue;
> + if (test_bit(R5_WantFUA, &sh->dev[i].flags))
> + fua = 1;
> write_disks++;
> /* checksum is already calculated in last run */
> if (test_bit(STRIPE_LOG_TRAPPED, &sh->state))
> @@ -462,6 +467,10 @@ int r5l_write_stripe(struct r5l_log *log, struct stripe_head *sh)
> addr, PAGE_SIZE);
> kunmap_atomic(addr);
> }
> +
> + if (log->bypass_full_stripe && (write_disks == sh->disks) && (!fua))
> + return -EAGAIN; /* bypass journal device */
> +
> parity_pages = 1 + !!(sh->qd_idx >= 0);
> data_pages = write_disks - parity_pages;
>
> @@ -520,6 +529,16 @@ int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio)
> return -EAGAIN;
> }
>
> +int r5l_get_bypass_full_stripe(struct r5l_log *log)
> +{
> + return log->bypass_full_stripe;
> +}
> +
> +void r5l_set_bypass_full_stripe(struct r5l_log *log, int val)
> +{
> + log->bypass_full_stripe = val;
> +}
> +
> /* This will run after log space is reclaimed */
> static void r5l_run_no_space_stripes(struct r5l_log *log)
> {
> @@ -1105,6 +1124,7 @@ int r5l_init_log(struct r5conf *conf, struct md_rdev *rdev)
> if (!log->io_kc)
> goto io_kc;
>
> + log->bypass_full_stripe = 0;
> log->reclaim_thread = md_register_thread(r5l_reclaim_thread,
> log->rdev->mddev, "reclaim");
> if (!log->reclaim_thread)
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index 394cdf8..5781987 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -6223,6 +6223,61 @@ raid5_group_thread_cnt = __ATTR(group_thread_cnt, S_IRUGO | S_IWUSR,
> raid5_show_group_thread_cnt,
> raid5_store_group_thread_cnt);
>
> +static ssize_t
> +r5l_show_bypass_full_stripe(struct mddev *mddev, char *page)
> +{
> + struct r5conf *conf;
> + int ret = 0;
> +
> + spin_lock(&mddev->lock);
> + conf = mddev->private;
> + if (conf) {
> + if (conf->log)
> + ret = sprintf(page, "%s\n",
> + r5l_get_bypass_full_stripe(conf->log) ? "yes" : "no");
> + else
> + ret = sprintf(page, "n/a\n");
> + }
> + spin_unlock(&mddev->lock);
> + return ret;
> +}
> +
> +static ssize_t
> +r5l_store_bypass_full_stripe(struct mddev *mddev, const char *page, size_t len)
> +{
> + struct r5conf *conf;
> + int err = 0;
> + int val;
> +
> + if (strncmp(page, "yes", 3) == 0 &&
> + (page[3] == '\n' || page[3] == '\0'))
> + val = 1;
> + else if (strncmp(page, "no", 2) == 0 &&
> + (page[2] == '\n' || page[2] == '\0'))
> + val = 0;
> + else
> + return -EINVAL;
> +
> + mddev_suspend(mddev);
> + spin_lock(&mddev->lock);
> + conf = mddev->private;
> + if (conf) {
> + if (conf->log) {
> + r5l_set_bypass_full_stripe(conf->log, val);
> + } else
> + err = -EINVAL;
> + } else
> + err = -ENODEV;
> + spin_unlock(&mddev->lock);
> + mddev_resume(mddev);
> + return err ?: len;
> +}
> +
> +static struct md_sysfs_entry
> +r5l_bypass_full_stripe = __ATTR(r5l_bypass_full_stripe, S_IRUGO | S_IWUSR,
> + r5l_show_bypass_full_stripe,
> + r5l_store_bypass_full_stripe);
> +
> static struct attribute *raid5_attrs[] = {
> &raid5_stripecache_size.attr,
> &raid5_stripecache_active.attr,
> @@ -6230,6 +6285,7 @@ static struct attribute *raid5_attrs[] = {
> &raid5_group_thread_cnt.attr,
> &raid5_skip_copy.attr,
> &raid5_rmw_level.attr,
> + &r5l_bypass_full_stripe.attr,
> NULL,
> };
> static struct attribute_group raid5_attrs_group = {
> diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
> index e6b9a40..c1f6935 100644
> --- a/drivers/md/raid5.h
> +++ b/drivers/md/raid5.h
> @@ -630,4 +630,6 @@ extern void r5l_write_stripe_run(struct r5l_log *log);
> extern void r5l_flush_stripe_to_raid(struct r5l_log *log);
> extern void r5l_stripe_write_finished(struct stripe_head *sh);
> extern int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio);
> +extern int r5l_get_bypass_full_stripe(struct r5l_log *log);
> +extern void r5l_set_bypass_full_stripe(struct r5l_log *log, int val);
> #endif
> --
> 1.8.1
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 818 bytes --]
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2015-09-30 8:03 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2015-09-04 6:01 [PATCH] enable bypass raid5 journal for full stripe writes Song Liu
2015-09-30 8:03 ` Neil Brown
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).