Merge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm

* git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm:
  dm kcopyd: return client directly and not through a pointer
  dm kcopyd: reserve fewer pages
  dm io: use fixed initial mempool size
  dm kcopyd: alloc pages from the main page allocator
  dm kcopyd: add gfp parm to alloc_pl
  dm kcopyd: remove superfluous page allocation spinlock
  dm kcopyd: preallocate sub jobs to avoid deadlock
  dm kcopyd: avoid pointless job splitting
  dm mpath: do not fail paths after integrity errors
  dm table: reject devices without request fns
  dm table: allow targets to support discards internally

8 files changed, 136 insertions, 120 deletions

diff --git a/drivers/md/dm-io.c b/drivers/md/dm-io.c
index 76a5af00a26b..2067288f61f9 100644
--- a/drivers/md/dm-io.c
+++ b/drivers/md/dm-io.c
@@ -19,6 +19,8 @@
 #define DM_MSG_PREFIX "io"
 
 #define DM_IO_MAX_REGIONS	BITS_PER_LONG
+#define MIN_IOS		16
+#define MIN_BIOS	16
 
 struct dm_io_client {
 	mempool_t *pool;
@@ -41,33 +43,21 @@ struct io {
 static struct kmem_cache *_dm_io_cache;
 
 /*
- * io contexts are only dynamically allocated for asynchronous
- * io.  Since async io is likely to be the majority of io we'll
- * have the same number of io contexts as bios! (FIXME: must reduce this).
- */
-
-static unsigned int pages_to_ios(unsigned int pages)
-{
-	return 4 * pages;	/* too many ? */
-}
-
-/*
  * Create a client with mempool and bioset.
  */
-struct dm_io_client *dm_io_client_create(unsigned num_pages)
+struct dm_io_client *dm_io_client_create(void)
 {
-	unsigned ios = pages_to_ios(num_pages);
 	struct dm_io_client *client;
 
 	client = kmalloc(sizeof(*client), GFP_KERNEL);
 	if (!client)
 		return ERR_PTR(-ENOMEM);
 
-	client->pool = mempool_create_slab_pool(ios, _dm_io_cache);
+	client->pool = mempool_create_slab_pool(MIN_IOS, _dm_io_cache);
 	if (!client->pool)
 		goto bad;
 
-	client->bios = bioset_create(16, 0);
+	client->bios = bioset_create(MIN_BIOS, 0);
 	if (!client->bios)
 		goto bad;
 
@@ -81,13 +71,6 @@ struct dm_io_client *dm_io_client_create(unsigned num_pages)
 }
 EXPORT_SYMBOL(dm_io_client_create);
 
-int dm_io_client_resize(unsigned num_pages, struct dm_io_client *client)
-{
-	return mempool_resize(client->pool, pages_to_ios(num_pages),
-			      GFP_KERNEL);
-}
-EXPORT_SYMBOL(dm_io_client_resize);
-
 void dm_io_client_destroy(struct dm_io_client *client)
 {
 	mempool_destroy(client->pool);
diff --git a/drivers/md/dm-kcopyd.c b/drivers/md/dm-kcopyd.c
index 1bb73a13ca40..819e37eaaeba 100644
--- a/drivers/md/dm-kcopyd.c
+++ b/drivers/md/dm-kcopyd.c
@@ -27,15 +27,19 @@
 
 #include "dm.h"
 
+#define SUB_JOB_SIZE	128
+#define SPLIT_COUNT	8
+#define MIN_JOBS	8
+#define RESERVE_PAGES	(DIV_ROUND_UP(SUB_JOB_SIZE << SECTOR_SHIFT, PAGE_SIZE))
+
 /*-----------------------------------------------------------------
  * Each kcopyd client has its own little pool of preallocated
  * pages for kcopyd io.
  *---------------------------------------------------------------*/
 struct dm_kcopyd_client {
-	spinlock_t lock;
 	struct page_list *pages;
-	unsigned int nr_pages;
-	unsigned int nr_free_pages;
+	unsigned nr_reserved_pages;
+	unsigned nr_free_pages;
 
 	struct dm_io_client *io_client;
 
@@ -67,15 +71,18 @@ static void wake(struct dm_kcopyd_client *kc)
 	queue_work(kc->kcopyd_wq, &kc->kcopyd_work);
 }
 
-static struct page_list *alloc_pl(void)
+/*
+ * Obtain one page for the use of kcopyd.
+ */
+static struct page_list *alloc_pl(gfp_t gfp)
 {
 	struct page_list *pl;
 
-	pl = kmalloc(sizeof(*pl), GFP_KERNEL);
+	pl = kmalloc(sizeof(*pl), gfp);
 	if (!pl)
 		return NULL;
 
-	pl->page = alloc_page(GFP_KERNEL);
+	pl->page = alloc_page(gfp);
 	if (!pl->page) {
 		kfree(pl);
 		return NULL;
@@ -90,41 +97,56 @@ static void free_pl(struct page_list *pl)
 	kfree(pl);
 }
 
-static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
-			    unsigned int nr, struct page_list **pages)
+/*
+ * Add the provided pages to a client's free page list, releasing
+ * back to the system any beyond the reserved_pages limit.
+ */
+static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
 {
-	struct page_list *pl;
-
-	spin_lock(&kc->lock);
-	if (kc->nr_free_pages < nr) {
-		spin_unlock(&kc->lock);
-		return -ENOMEM;
-	}
-
-	kc->nr_free_pages -= nr;
-	for (*pages = pl = kc->pages; --nr; pl = pl->next)
-		;
+	struct page_list *next;
 
-	kc->pages = pl->next;
-	pl->next = NULL;
+	do {
+		next = pl->next;
 
-	spin_unlock(&kc->lock);
+		if (kc->nr_free_pages >= kc->nr_reserved_pages)
+			free_pl(pl);
+		else {
+			pl->next = kc->pages;
+			kc->pages = pl;
+			kc->nr_free_pages++;
+		}
 
-	return 0;
+		pl = next;
+	} while (pl);
 }
 
-static void kcopyd_put_pages(struct dm_kcopyd_client *kc, struct page_list *pl)
+static int kcopyd_get_pages(struct dm_kcopyd_client *kc,
+			    unsigned int nr, struct page_list **pages)
 {
-	struct page_list *cursor;
+	struct page_list *pl;
+
+	*pages = NULL;
+
+	do {
+		pl = alloc_pl(__GFP_NOWARN | __GFP_NORETRY);
+		if (unlikely(!pl)) {
+			/* Use reserved pages */
+			pl = kc->pages;
+			if (unlikely(!pl))
+				goto out_of_memory;
+			kc->pages = pl->next;
+			kc->nr_free_pages--;
+		}
+		pl->next = *pages;
+		*pages = pl;
+	} while (--nr);
 
-	spin_lock(&kc->lock);
-	for (cursor = pl; cursor->next; cursor = cursor->next)
-		kc->nr_free_pages++;
+	return 0;
 
-	kc->nr_free_pages++;
-	cursor->next = kc->pages;
-	kc->pages = pl;
-	spin_unlock(&kc->lock);
+out_of_memory:
+	if (*pages)
+		kcopyd_put_pages(kc, *pages);
+	return -ENOMEM;
 }
 
 /*
@@ -141,13 +163,16 @@ static void drop_pages(struct page_list *pl)
 	}
 }
 
-static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
+/*
+ * Allocate and reserve nr_pages for the use of a specific client.
+ */
+static int client_reserve_pages(struct dm_kcopyd_client *kc, unsigned nr_pages)
 {
-	unsigned int i;
+	unsigned i;
 	struct page_list *pl = NULL, *next;
 
-	for (i = 0; i < nr; i++) {
-		next = alloc_pl();
+	for (i = 0; i < nr_pages; i++) {
+		next = alloc_pl(GFP_KERNEL);
 		if (!next) {
 			if (pl)
 				drop_pages(pl);
@@ -157,17 +182,18 @@ static int client_alloc_pages(struct dm_kcopyd_client *kc, unsigned int nr)
 		pl = next;
 	}
 
+	kc->nr_reserved_pages += nr_pages;
 	kcopyd_put_pages(kc, pl);
-	kc->nr_pages += nr;
+
 	return 0;
 }
 
 static void client_free_pages(struct dm_kcopyd_client *kc)
 {
-	BUG_ON(kc->nr_free_pages != kc->nr_pages);
+	BUG_ON(kc->nr_free_pages != kc->nr_reserved_pages);
 	drop_pages(kc->pages);
 	kc->pages = NULL;
-	kc->nr_free_pages = kc->nr_pages = 0;
+	kc->nr_free_pages = kc->nr_reserved_pages = 0;
 }
 
 /*-----------------------------------------------------------------
@@ -216,16 +242,17 @@ struct kcopyd_job {
 	struct mutex lock;
 	atomic_t sub_jobs;
 	sector_t progress;
-};
 
-/* FIXME: this should scale with the number of pages */
-#define MIN_JOBS 512
+	struct kcopyd_job *master_job;
+};
 
 static struct kmem_cache *_job_cache;
 
 int __init dm_kcopyd_init(void)
 {
-	_job_cache = KMEM_CACHE(kcopyd_job, 0);
+	_job_cache = kmem_cache_create("kcopyd_job",
+				sizeof(struct kcopyd_job) * (SPLIT_COUNT + 1),
+				__alignof__(struct kcopyd_job), 0, NULL);
 	if (!_job_cache)
 		return -ENOMEM;
 
@@ -299,7 +326,12 @@ static int run_complete_job(struct kcopyd_job *job)
 
 	if (job->pages)
 		kcopyd_put_pages(kc, job->pages);
-	mempool_free(job, kc->job_pool);
+	/*
+	 * If this is the master job, the sub jobs have already
+	 * completed so we can free everything.
+	 */
+	if (job->master_job == job)
+		mempool_free(job, kc->job_pool);
 	fn(read_err, write_err, context);
 
 	if (atomic_dec_and_test(&kc->nr_jobs))
@@ -460,14 +492,14 @@ static void dispatch_job(struct kcopyd_job *job)
 	wake(kc);
 }
 
-#define SUB_JOB_SIZE 128
 static void segment_complete(int read_err, unsigned long write_err,
 			     void *context)
 {
 	/* FIXME: tidy this function */
 	sector_t progress = 0;
 	sector_t count = 0;
-	struct kcopyd_job *job = (struct kcopyd_job *) context;
+	struct kcopyd_job *sub_job = (struct kcopyd_job *) context;
+	struct kcopyd_job *job = sub_job->master_job;
 	struct dm_kcopyd_client *kc = job->kc;
 
 	mutex_lock(&job->lock);
@@ -498,8 +530,6 @@ static void segment_complete(int read_err, unsigned long write_err,
 
 	if (count) {
 		int i;
-		struct kcopyd_job *sub_job = mempool_alloc(kc->job_pool,
-							   GFP_NOIO);
 
 		*sub_job = *job;
 		sub_job->source.sector += progress;
@@ -511,7 +541,7 @@ static void segment_complete(int read_err, unsigned long write_err,
 		}
 
 		sub_job->fn = segment_complete;
-		sub_job->context = job;
+		sub_job->context = sub_job;
 		dispatch_job(sub_job);
 
 	} else if (atomic_dec_and_test(&job->sub_jobs)) {
@@ -531,19 +561,19 @@ static void segment_complete(int read_err, unsigned long write_err,
 }
 
 /*
- * Create some little jobs that will do the move between
- * them.
+ * Create some sub jobs to share the work between them.
  */
-#define SPLIT_COUNT 8
-static void split_job(struct kcopyd_job *job)
+static void split_job(struct kcopyd_job *master_job)
 {
 	int i;
 
-	atomic_inc(&job->kc->nr_jobs);
+	atomic_inc(&master_job->kc->nr_jobs);
 
-	atomic_set(&job->sub_jobs, SPLIT_COUNT);
-	for (i = 0; i < SPLIT_COUNT; i++)
-		segment_complete(0, 0u, job);
+	atomic_set(&master_job->sub_jobs, SPLIT_COUNT);
+	for (i = 0; i < SPLIT_COUNT; i++) {
+		master_job[i + 1].master_job = master_job;
+		segment_complete(0, 0u, &master_job[i + 1]);
+	}
 }
 
 int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
@@ -553,7 +583,8 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 	struct kcopyd_job *job;
 
 	/*
-	 * Allocate a new job.
+	 * Allocate an array of jobs consisting of one master job
+	 * followed by SPLIT_COUNT sub jobs.
 	 */
 	job = mempool_alloc(kc->job_pool, GFP_NOIO);
 
@@ -577,10 +608,10 @@ int dm_kcopyd_copy(struct dm_kcopyd_client *kc, struct dm_io_region *from,
 
 	job->fn = fn;
 	job->context = context;
+	job->master_job = job;
 
-	if (job->source.count < SUB_JOB_SIZE)
+	if (job->source.count <= SUB_JOB_SIZE)
 		dispatch_job(job);
-
 	else {
 		mutex_init(&job->lock);
 		job->progress = 0;
@@ -606,17 +637,15 @@ int kcopyd_cancel(struct kcopyd_job *job, int block)
 /*-----------------------------------------------------------------
  * Client setup
  *---------------------------------------------------------------*/
-int dm_kcopyd_client_create(unsigned int nr_pages,
-			    struct dm_kcopyd_client **result)
+struct dm_kcopyd_client *dm_kcopyd_client_create(void)
 {
 	int r = -ENOMEM;
 	struct dm_kcopyd_client *kc;
 
 	kc = kmalloc(sizeof(*kc), GFP_KERNEL);
 	if (!kc)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
-	spin_lock_init(&kc->lock);
 	spin_lock_init(&kc->job_lock);
 	INIT_LIST_HEAD(&kc->complete_jobs);
 	INIT_LIST_HEAD(&kc->io_jobs);
@@ -633,12 +662,12 @@ int dm_kcopyd_client_create(unsigned int nr_pages,
 		goto bad_workqueue;
 
 	kc->pages = NULL;
-	kc->nr_pages = kc->nr_free_pages = 0;
-	r = client_alloc_pages(kc, nr_pages);
+	kc->nr_reserved_pages = kc->nr_free_pages = 0;
+	r = client_reserve_pages(kc, RESERVE_PAGES);
 	if (r)
 		goto bad_client_pages;
 
-	kc->io_client = dm_io_client_create(nr_pages);
+	kc->io_client = dm_io_client_create();
 	if (IS_ERR(kc->io_client)) {
 		r = PTR_ERR(kc->io_client);
 		goto bad_io_client;
@@ -647,8 +676,7 @@ int dm_kcopyd_client_create(unsigned int nr_pages,
 	init_waitqueue_head(&kc->destroyq);
 	atomic_set(&kc->nr_jobs, 0);
 
-	*result = kc;
-	return 0;
+	return kc;
 
 bad_io_client:
 	client_free_pages(kc);
@@ -659,7 +687,7 @@ bad_workqueue:
 bad_slab:
 	kfree(kc);
 
-	return r;
+	return ERR_PTR(r);
 }
 EXPORT_SYMBOL(dm_kcopyd_client_create);
 
diff --git a/drivers/md/dm-log.c b/drivers/md/dm-log.c
index a1f321889676..948e3f4925bf 100644
--- a/drivers/md/dm-log.c
+++ b/drivers/md/dm-log.c
@@ -449,8 +449,7 @@ static int create_log_context(struct dm_dirty_log *log, struct dm_target *ti,
 
 		lc->io_req.mem.type = DM_IO_VMA;
 		lc->io_req.notify.fn = NULL;
-		lc->io_req.client = dm_io_client_create(dm_div_up(buf_size,
-								   PAGE_SIZE));
+		lc->io_req.client = dm_io_client_create();
 		if (IS_ERR(lc->io_req.client)) {
 			r = PTR_ERR(lc->io_req.client);
 			DMWARN("couldn't allocate disk io client");
diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
index a550a057d991..aa4e570c2cb5 100644
--- a/drivers/md/dm-mpath.c
+++ b/drivers/md/dm-mpath.c
@@ -1290,7 +1290,7 @@ static int do_end_io(struct multipath *m, struct request *clone,
 	if (!error && !clone->errors)
 		return 0;	/* I/O complete */
 
-	if (error == -EOPNOTSUPP || error == -EREMOTEIO)
+	if (error == -EOPNOTSUPP || error == -EREMOTEIO || error == -EILSEQ)
 		return error;
 
 	if (mpio->pgpath)
diff --git a/drivers/md/dm-raid1.c b/drivers/md/dm-raid1.c
index 976ad4688afc..9bfd057be686 100644
--- a/drivers/md/dm-raid1.c
+++ b/drivers/md/dm-raid1.c
@@ -22,8 +22,6 @@
 #define DM_MSG_PREFIX "raid1"
 
 #define MAX_RECOVERY 1	/* Maximum number of regions recovered in parallel. */
-#define DM_IO_PAGES 64
-#define DM_KCOPYD_PAGES 64
 
 #define DM_RAID1_HANDLE_ERRORS 0x01
 #define errors_handled(p)	((p)->features & DM_RAID1_HANDLE_ERRORS)
@@ -887,7 +885,7 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
 		return NULL;
 	}
 
-	ms->io_client = dm_io_client_create(DM_IO_PAGES);
+	ms->io_client = dm_io_client_create();
 	if (IS_ERR(ms->io_client)) {
 		ti->error = "Error creating dm_io client";
 		mempool_destroy(ms->read_record_pool);
@@ -1117,9 +1115,11 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto err_destroy_wq;
 	}
 
-	r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client);
-	if (r)
+	ms->kcopyd_client = dm_kcopyd_client_create();
+	if (IS_ERR(ms->kcopyd_client)) {
+		r = PTR_ERR(ms->kcopyd_client);
 		goto err_destroy_wq;
+	}
 
 	wakeup_mirrord(ms);
 	return 0;
diff --git a/drivers/md/dm-snap-persistent.c b/drivers/md/dm-snap-persistent.c
index 95891dfcbca0..135c2f1fdbfc 100644
--- a/drivers/md/dm-snap-persistent.c
+++ b/drivers/md/dm-snap-persistent.c
@@ -154,11 +154,6 @@ struct pstore {
 	struct workqueue_struct *metadata_wq;
 };
 
-static unsigned sectors_to_pages(unsigned sectors)
-{
-	return DIV_ROUND_UP(sectors, PAGE_SIZE >> 9);
-}
-
 static int alloc_area(struct pstore *ps)
 {
 	int r = -ENOMEM;
@@ -318,8 +313,7 @@ static int read_header(struct pstore *ps, int *new_snapshot)
 		chunk_size_supplied = 0;
 	}
 
-	ps->io_client = dm_io_client_create(sectors_to_pages(ps->store->
-							     chunk_size));
+	ps->io_client = dm_io_client_create();
 	if (IS_ERR(ps->io_client))
 		return PTR_ERR(ps->io_client);
 
@@ -368,11 +362,6 @@ static int read_header(struct pstore *ps, int *new_snapshot)
 		return r;
 	}
 
-	r = dm_io_client_resize(sectors_to_pages(ps->store->chunk_size),
-				ps->io_client);
-	if (r)
-		return r;
-
 	r = alloc_area(ps);
 	return r;
 
diff --git a/drivers/md/dm-snap.c b/drivers/md/dm-snap.c
index a2d330942cb2..9ecff5f3023a 100644
--- a/drivers/md/dm-snap.c
+++ b/drivers/md/dm-snap.c
@@ -40,11 +40,6 @@ static const char dm_snapshot_merge_target_name[] = "snapshot-merge";
 #define SNAPSHOT_COPY_PRIORITY 2
 
 /*
- * Reserve 1MB for each snapshot initially (with minimum of 1 page).
- */
-#define SNAPSHOT_PAGES (((1UL << 20) >> PAGE_SHIFT) ? : 1)
-
-/*
  * The size of the mempool used to track chunks in use.
  */
 #define MIN_IOS 256
@@ -1116,8 +1111,9 @@ static int snapshot_ctr(struct dm_target *ti, unsigned int argc, char **argv)
 		goto bad_hash_tables;
 	}
 
-	r = dm_kcopyd_client_create(SNAPSHOT_PAGES, &s->kcopyd_client);
-	if (r) {
+	s->kcopyd_client = dm_kcopyd_client_create();
+	if (IS_ERR(s->kcopyd_client)) {
+		r = PTR_ERR(s->kcopyd_client);
 		ti->error = "Could not create kcopyd client";
 		goto bad_kcopyd;
 	}
diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index cb8380c9767f..451c3bb176d2 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -362,6 +362,7 @@ static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 				  sector_t start, sector_t len, void *data)
 {
+	struct request_queue *q;
 	struct queue_limits *limits = data;
 	struct block_device *bdev = dev->bdev;
 	sector_t dev_size =
@@ -370,6 +371,22 @@ static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 		limits->logical_block_size >> SECTOR_SHIFT;
 	char b[BDEVNAME_SIZE];
 
+	/*
+	 * Some devices exist without request functions,
+	 * such as loop devices not yet bound to backing files.
+	 * Forbid the use of such devices.
+	 */
+	q = bdev_get_queue(bdev);
+	if (!q || !q->make_request_fn) {
+		DMWARN("%s: %s is not yet initialised: "
+		       "start=%llu, len=%llu, dev_size=%llu",
+		       dm_device_name(ti->table->md), bdevname(bdev, b),
+		       (unsigned long long)start,
+		       (unsigned long long)len,
+		       (unsigned long long)dev_size);
+		return 1;
+	}
+
 	if (!dev_size)
 		return 0;
 
@@ -1346,7 +1363,8 @@ bool dm_table_supports_discards(struct dm_table *t)
 		return 0;
 
 	/*
-	 * Ensure that at least one underlying device supports discards.
+	 * Unless any target used by the table set discards_supported,
+	 * require at least one underlying device to support discards.
 	 * t->devices includes internal dm devices such as mirror logs
 	 * so we need to use iterate_devices here, which targets
 	 * supporting discard must provide.
@@ -1354,6 +1372,9 @@ bool dm_table_supports_discards(struct dm_table *t)
 	while (i < dm_table_get_num_targets(t)) {
 		ti = dm_table_get_target(t, i++);
 
+		if (ti->discards_supported)
+			return 1;
+
 		if (ti->type->iterate_devices &&
 		    ti->type->iterate_devices(ti, device_discard_capable, NULL))
 			return 1;