[PATCH v4] md/r5cache: handle alloc_page failure

[PATCH v4] md/r5cache: handle alloc_page failure

[Song Liu]

RMW of r5c write back cache uses an extra page to store old data for
prexor. handle_stripe_dirtying() allocates this page by calling
alloc_page(). However, alloc_page() may fail.

To handle alloc_page() failures, this patch adds an extra page to
disk_info. When alloc_page fails, handle_stripe() trys to use these
pages. When these pages are used by other stripe (R5C_EXTRA_PAGE_IN_USE),
the stripe is added to delayed_list.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 drivers/md/raid5-cache.c | 30 +++++++++++++++++
 drivers/md/raid5.c       | 84 +++++++++++++++++++++++++++++++++++++++++-------
 drivers/md/raid5.h       |  6 ++++
 3 files changed, 109 insertions(+), 11 deletions(-)

diff --git a/drivers/md/raid5-cache.c b/drivers/md/raid5-cache.c
index 8cb79fc..c395787 100644
--- a/drivers/md/raid5-cache.c
+++ b/drivers/md/raid5-cache.c
@@ -2334,15 +2334,45 @@ int r5c_try_caching_write(struct r5conf *conf,
  */
 void r5c_release_extra_page(struct stripe_head *sh)
 {
+	struct r5conf *conf = sh->raid_conf;
 	int i;
+	bool using_disk_info_extra_page;
+
+	using_disk_info_extra_page =
+		sh->dev[0].orig_page == conf->disks[0].extra_page;
 
 	for (i = sh->disks; i--; )
 		if (sh->dev[i].page != sh->dev[i].orig_page) {
 			struct page *p = sh->dev[i].orig_page;
 
 			sh->dev[i].orig_page = sh->dev[i].page;
+			if (!using_disk_info_extra_page)
+				put_page(p);
+		}
+
+	if (using_disk_info_extra_page) {
+		clear_bit(R5C_EXTRA_PAGE_IN_USE, &conf->cache_state);
+		md_wakeup_thread(conf->mddev->thread);
+	}
+}
+
+void r5c_use_extra_page(struct stripe_head *sh)
+{
+	struct r5conf *conf = sh->raid_conf;
+	int i;
+	struct r5dev *dev;
+	struct page *p;
+
+	for (i = sh->disks; i--; ) {
+		dev = &sh->dev[i];
+		if (dev->orig_page != dev->page) {
+			p = dev->orig_page;
+			dev->orig_page = dev->page;
 			put_page(p);
 		}
+		dev->orig_page = conf->disks[i].extra_page;
+		BUG_ON(!dev->orig_page);
+	}
 }
 
 /*
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index dbab8c7..5454c3b 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -876,6 +876,8 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
 
 	if (!test_bit(STRIPE_R5C_CACHING, &sh->state)) {
 		/* writing out phase */
+		if (s->waiting_extra_page)
+			return;
 		if (r5l_write_stripe(conf->log, sh) == 0)
 			return;
 	} else {  /* caching phase */
@@ -2007,6 +2009,7 @@ static struct stripe_head *alloc_stripe(struct kmem_cache *sc, gfp_t gfp,
 		INIT_LIST_HEAD(&sh->batch_list);
 		INIT_LIST_HEAD(&sh->lru);
 		INIT_LIST_HEAD(&sh->r5c);
+		INIT_LIST_HEAD(&sh->log_list);
 		atomic_set(&sh->count, 1);
 		sh->log_start = MaxSector;
 		for (i = 0; i < disks; i++) {
@@ -2255,8 +2258,27 @@ static int resize_stripes(struct r5conf *conf, int newsize)
 	if (ndisks) {
 		for (i=0; i<conf->raid_disks; i++)
 			ndisks[i] = conf->disks[i];
-		kfree(conf->disks);
-		conf->disks = ndisks;
+
+		/* allocate extra_page for ndisks */
+		for (i = 0; i < newsize; i++) {
+			ndisks[i].extra_page = alloc_page(GFP_NOIO);
+			if (!ndisks[i].extra_page)
+				err = -ENOMEM;
+		}
+
+		if (err) {
+			/* if any error, free extra_page for ndisks */
+			for (i = 0; i < newsize; i++)
+				if (ndisks[i].extra_page)
+					put_page(ndisks[i].extra_page);
+			kfree(ndisks);
+		} else {
+			/* if no error, free extra_page for old disks */
+			for (i = 0; i < conf->previous_raid_disks; i++)
+				put_page(ndisks[i].extra_page);
+			kfree(conf->disks);
+			conf->disks = ndisks;
+		}
 	} else
 		err = -ENOMEM;
 
@@ -3580,10 +3602,10 @@ static void handle_stripe_clean_event(struct r5conf *conf,
 		break_stripe_batch_list(head_sh, STRIPE_EXPAND_SYNC_FLAGS);
 }
 
-static void handle_stripe_dirtying(struct r5conf *conf,
-				   struct stripe_head *sh,
-				   struct stripe_head_state *s,
-				   int disks)
+static int handle_stripe_dirtying(struct r5conf *conf,
+				  struct stripe_head *sh,
+				  struct stripe_head_state *s,
+				  int disks)
 {
 	int rmw = 0, rcw = 0, i;
 	sector_t recovery_cp = conf->mddev->recovery_cp;
@@ -3649,12 +3671,32 @@ static void handle_stripe_dirtying(struct r5conf *conf,
 			    dev->page == dev->orig_page &&
 			    !test_bit(R5_LOCKED, &sh->dev[sh->pd_idx].flags)) {
 				/* alloc page for prexor */
-				dev->orig_page = alloc_page(GFP_NOIO);
+				struct page *p = alloc_page(GFP_NOIO);
+
+				if (p) {
+					dev->orig_page = p;
+					continue;
+				}
+
+				/*
+				 * alloc_page() failed, try use
+				 * disk_info->extra_page
+				 */
+				if (!test_and_set_bit(R5C_EXTRA_PAGE_IN_USE,
+						      &conf->cache_state)) {
+					r5c_use_extra_page(sh);
+					break;
+				}
 
-				/* will handle failure in a later patch*/
-				BUG_ON(!dev->orig_page);
+				/* extra_page in use, add to delayed_list */
+				set_bit(STRIPE_DELAYED, &sh->state);
+				s->waiting_extra_page = 1;
+				return -EAGAIN;
 			}
+		}
 
+		for (i = disks; i--; ) {
+			struct r5dev *dev = &sh->dev[i];
 			if ((dev->towrite ||
 			     i == sh->pd_idx || i == sh->qd_idx ||
 			     test_bit(R5_InJournal, &dev->flags)) &&
@@ -3730,6 +3772,7 @@ static void handle_stripe_dirtying(struct r5conf *conf,
 	    (s->locked == 0 && (rcw == 0 || rmw == 0) &&
 	     !test_bit(STRIPE_BIT_DELAY, &sh->state)))
 		schedule_reconstruction(sh, s, rcw == 0, 0);
+	return 0;
 }
 
 static void handle_parity_checks5(struct r5conf *conf, struct stripe_head *sh,
@@ -4545,8 +4588,12 @@ static void handle_stripe(struct stripe_head *sh)
 			if (ret == -EAGAIN ||
 			    /* stripe under reclaim: !caching && injournal */
 			    (!test_bit(STRIPE_R5C_CACHING, &sh->state) &&
-			     s.injournal > 0))
-				handle_stripe_dirtying(conf, sh, &s, disks);
+			     s.injournal > 0)) {
+				ret = handle_stripe_dirtying(conf, sh, &s,
+							     disks);
+				if (ret == -EAGAIN)
+					goto finish;
+			}
 		}
 	}
 
@@ -6458,6 +6505,8 @@ static void raid5_free_percpu(struct r5conf *conf)
 
 static void free_conf(struct r5conf *conf)
 {
+	int i;
+
 	if (conf->log)
 		r5l_exit_log(conf->log);
 	if (conf->shrinker.nr_deferred)
@@ -6466,6 +6515,12 @@ static void free_conf(struct r5conf *conf)
 	free_thread_groups(conf);
 	shrink_stripes(conf);
 	raid5_free_percpu(conf);
+	for (i = 0; i < conf->raid_disks; i++)
+		if (conf->disks[i].extra_page) {
+			put_page(conf->disks[i].extra_page);
+			conf->disks[i].extra_page = NULL;
+		}
+
 	kfree(conf->disks);
 	kfree(conf->stripe_hashtbl);
 	kfree(conf);
@@ -6612,9 +6667,16 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 
 	conf->disks = kzalloc(max_disks * sizeof(struct disk_info),
 			      GFP_KERNEL);
+
 	if (!conf->disks)
 		goto abort;
 
+	for (i = 0; i < conf->raid_disks; i++) {
+		conf->disks[i].extra_page = alloc_page(GFP_KERNEL);
+		if (!conf->disks[i].extra_page)
+			goto abort;
+	}
+
 	conf->mddev = mddev;
 
 	if ((conf->stripe_hashtbl = kzalloc(PAGE_SIZE, GFP_KERNEL)) == NULL)
diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
index d13fe45..ed8e136 100644
--- a/drivers/md/raid5.h
+++ b/drivers/md/raid5.h
@@ -276,6 +276,7 @@ struct stripe_head_state {
 	struct md_rdev *blocked_rdev;
 	int handle_bad_blocks;
 	int log_failed;
+	int waiting_extra_page;
 };
 
 /* Flags for struct r5dev.flags */
@@ -439,6 +440,7 @@ enum {
 
 struct disk_info {
 	struct md_rdev	*rdev, *replacement;
+	struct page	*extra_page; /* extra page to use in prexor */
 };
 
 /*
@@ -559,6 +561,9 @@ enum r5_cache_state {
 				 * only process stripes that are already
 				 * occupying the log
 				 */
+	R5C_EXTRA_PAGE_IN_USE,	/* a stripe is using disk_info.extra_page
+				 * for prexor
+				 */
 };
 
 struct r5conf {
@@ -765,6 +770,7 @@ extern void
 r5c_finish_stripe_write_out(struct r5conf *conf, struct stripe_head *sh,
 			    struct stripe_head_state *s);
 extern void r5c_release_extra_page(struct stripe_head *sh);
+extern void r5c_use_extra_page(struct stripe_head *sh);
 extern void r5l_wake_reclaim(struct r5l_log *log, sector_t space);
 extern void r5c_handle_cached_data_endio(struct r5conf *conf,
 	struct stripe_head *sh, int disks, struct bio_list *return_bi);
-- 
2.9.3

Re: [PATCH v4] md/r5cache: handle alloc_page failure

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 2517 bytes --]

On Thu, Nov 24 2016, Song Liu wrote:

> +void r5c_use_extra_page(struct stripe_head *sh)
> +{
> +	struct r5conf *conf = sh->raid_conf;
> +	int i;
> +	struct r5dev *dev;
> +	struct page *p;
> +
> +	for (i = sh->disks; i--; ) {
> +		dev = &sh->dev[i];
> +		if (dev->orig_page != dev->page) {
> +			p = dev->orig_page;
> +			dev->orig_page = dev->page;
>  			put_page(p);
>  		}
> +		dev->orig_page = conf->disks[i].extra_page;

It seems a bit pointless to assign to dev->orig_page twice.
Why not:

  if (dev->orig_page != dev->page)
      put_page(dev->orig_page);
  dev->orig_page = conf->......

??

> @@ -2255,8 +2258,27 @@ static int resize_stripes(struct r5conf *conf, int newsize)
>  	if (ndisks) {
>  		for (i=0; i<conf->raid_disks; i++)
>  			ndisks[i] = conf->disks[i];
> -		kfree(conf->disks);
> -		conf->disks = ndisks;
> +
> +		/* allocate extra_page for ndisks */
> +		for (i = 0; i < newsize; i++) {
> +			ndisks[i].extra_page = alloc_page(GFP_NOIO);
> +			if (!ndisks[i].extra_page)
> +				err = -ENOMEM;
> +		}
> +
> +		if (err) {
> +			/* if any error, free extra_page for ndisks */
> +			for (i = 0; i < newsize; i++)
> +				if (ndisks[i].extra_page)
> +					put_page(ndisks[i].extra_page);
> +			kfree(ndisks);
> +		} else {
> +			/* if no error, free extra_page for old disks */
> +			for (i = 0; i < conf->previous_raid_disks; i++)
> +				put_page(ndisks[i].extra_page);
> +			kfree(conf->disks);
> +			conf->disks = ndisks;
> +		}

This looks a bit odd too.  We never reduce conf->pool_size, so we never
need to free anything.

 for (i = conf->pool_size; i < newsize; i++)
     if ((ndisks[i].extra_page = alloc_page(GFP_NOIO)) == NULL)
         err = -ENOMEM;
 for (i = conf->pool_size; err == -ENOMEM && i < newsize; i++)
     if (ndisks[i].extra_page)
         put_page(ndisks[i].extra_page);


Maybe that it a little terse, but something like that would be better I
think.
Certainly you don't need to free ndisks.  If the allocation succeeds,
just use the new array whether other allocations succeed or fail.

> @@ -6466,6 +6515,12 @@ static void free_conf(struct r5conf *conf)
>  	free_thread_groups(conf);
>  	shrink_stripes(conf);
>  	raid5_free_percpu(conf);
> +	for (i = 0; i < conf->raid_disks; i++)
> +		if (conf->disks[i].extra_page) {
> +			put_page(conf->disks[i].extra_page);
> +			conf->disks[i].extra_page = NULL;

There is no point setting extra_page to NULL, as the whole array is
freed on the next line.


Apart from those few little things, it looks good.  Thanks.

NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 800 bytes --]

Re: [PATCH v4] md/r5cache: handle alloc_page failure

[Song Liu]

> On Nov 23, 2016, at 10:18 PM, NeilBrown <neilb@suse.com> wrote:
> 
> On Thu, Nov 24 2016, Song Liu wrote:
> 
>> +void r5c_use_extra_page(struct stripe_head *sh)
>> +{
>> +	struct r5conf *conf = sh->raid_conf;
>> +	int i;
>> +	struct r5dev *dev;
>> +	struct page *p;
>> +
>> +	for (i = sh->disks; i--; ) {
>> +		dev = &sh->dev[i];
>> +		if (dev->orig_page != dev->page) {
>> +			p = dev->orig_page;
>> +			dev->orig_page = dev->page;
>> 			put_page(p);
>> 		}
>> +		dev->orig_page = conf->disks[i].extra_page;
> 
> It seems a bit pointless to assign to dev->orig_page twice.
> Why not:
> 
>  if (dev->orig_page != dev->page)
>      put_page(dev->orig_page);
>  dev->orig_page = conf->......
> 
> ??
> 
>> @@ -2255,8 +2258,27 @@ static int resize_stripes(struct r5conf *conf, int newsize)
>> 	if (ndisks) {
>> 		for (i=0; i<conf->raid_disks; i++)
>> 			ndisks[i] = conf->disks[i];
>> -		kfree(conf->disks);
>> -		conf->disks = ndisks;
>> +
>> +		/* allocate extra_page for ndisks */
>> +		for (i = 0; i < newsize; i++) {
>> +			ndisks[i].extra_page = alloc_page(GFP_NOIO);
>> +			if (!ndisks[i].extra_page)
>> +				err = -ENOMEM;
>> +		}
>> +
>> +		if (err) {
>> +			/* if any error, free extra_page for ndisks */
>> +			for (i = 0; i < newsize; i++)
>> +				if (ndisks[i].extra_page)
>> +					put_page(ndisks[i].extra_page);
>> +			kfree(ndisks);
>> +		} else {
>> +			/* if no error, free extra_page for old disks */
>> +			for (i = 0; i < conf->previous_raid_disks; i++)
>> +				put_page(ndisks[i].extra_page);
>> +			kfree(conf->disks);
>> +			conf->disks = ndisks;
>> +		}
> 
> This looks a bit odd too.  We never reduce conf->pool_size, so we never
> need to free anything.
> 
> for (i = conf->pool_size; i < newsize; i++)
>     if ((ndisks[i].extra_page = alloc_page(GFP_NOIO)) == NULL)
>         err = -ENOMEM;
> for (i = conf->pool_size; err == -ENOMEM && i < newsize; i++)
>     if (ndisks[i].extra_page)
>         put_page(ndisks[i].extra_page);
> 
> 
> Maybe that it a little terse, but something like that would be better I
> think.
> Certainly you don't need to free ndisks.  If the allocation succeeds,
> just use the new array whether other allocations succeed or fail.
> 
>> @@ -6466,6 +6515,12 @@ static void free_conf(struct r5conf *conf)
>> 	free_thread_groups(conf);
>> 	shrink_stripes(conf);
>> 	raid5_free_percpu(conf);
>> +	for (i = 0; i < conf->raid_disks; i++)
>> +		if (conf->disks[i].extra_page) {
>> +			put_page(conf->disks[i].extra_page);
>> +			conf->disks[i].extra_page = NULL;
> 
> There is no point setting extra_page to NULL, as the whole array is
> freed on the next line.
> 
> 
> Apart from those few little things, it looks good.  Thanks.
> 
> NeilBrown

Thanks for these feedback! New patch coming.  

Song