[PATCH V6 0/4] Fix zone revalidation memory allocation failures

[PATCH V6 0/4] Fix zone revalidation memory allocation failures

[Damien Le Moal]

This series addresses a reccuring problem with zone revalidation
failures observed during extensive testing with memory constrained
system and device hot-plugging.

The problem source is failure to allocate large memory areas with
alloc_pages() or kmalloc() in blk_revalidate_disk_zones() to store the
disk array of zones (struct blk_zone) or in sd_zbc_report_zones() for
the report zones command reply buffer.

The solution proposed here is to:
1) limit the number of zones to be reported with a single report zones
command execution, and
2) Use vmalloc to allocate large-ish arrays and buffers in place of
alloc_pages() or kmalloc().

With these changes, tests do not show any zone revalidation failures
while not impacting the time taken for a disk zone inspection during
device scan and revalidation.

Changes from v5:
* Remove the gfp_t argument from blkdev_report_zones() function and
  device report zones method, relying instead on
  memalloc_noio_save/restore() where GFP_NOIO was used.

Changes from v4:
* bio_copy_kern does not use the vmalloc buffer for IO so does not
  need flush/invalidate_kernel_vmap_range() of the buffer.

Changes from v3:
* Reworked use of flush_kernel_vmap_range() and
  invalidate_kernel_vmap_range() to contain the calls within bio.c,
  transparently to the user of bio_map_kern().
* Add similar support to bio_copy_kern().

Changes from v2:
* Move invalidate_kernel_vmap_range() of vmalloc-ed buffer to sd_zbc.c
  in patch 2, after completion of scsi_execute_req().
* In patch 2, add flush_kernel_vmap_range() before scsi_execute_req().

Changes from V1:
* Added call to invalidate_kernel_vmap_range() for vmalloc-ed buffers
  in patch 1.
* Fixed patch 2 compilation error with Sparc64 (kbuild robot)

Damien Le Moal (4):
  block: Allow mapping of vmalloc-ed buffers
  block: Kill gfp_t argument of blkdev_report_zones()
  sd_zbc: Fix report zones buffer allocation
  block: Limit zone array allocation size

 block/bio.c                    |  28 ++++++++-
 block/blk-zoned.c              |  67 +++++++++++---------
 drivers/block/null_blk.h       |   3 +-
 drivers/block/null_blk_zoned.c |   3 +-
 drivers/md/dm-flakey.c         |   5 +-
 drivers/md/dm-linear.c         |   5 +-
 drivers/md/dm-zoned-metadata.c |  16 +++--
 drivers/md/dm.c                |   6 +-
 drivers/scsi/sd.h              |   3 +-
 drivers/scsi/sd_zbc.c          | 108 +++++++++++++++++++++++----------
 fs/f2fs/super.c                |   4 +-
 include/linux/blkdev.h         |  10 ++-
 include/linux/device-mapper.h  |   3 +-
 13 files changed, 172 insertions(+), 89 deletions(-)

-- 
2.21.0

[PATCH V6 1/4] block: Allow mapping of vmalloc-ed buffers

[Damien Le Moal]

To allow the SCSI subsystem scsi_execute_req() function to issue
requests using large buffers that are better allocated with vmalloc()
rather than kmalloc(), modify bio_map_kern() to allow passing a buffer
allocated with vmalloc().

To do so, detect vmalloc-ed buffers using is_vmalloc_addr(). For
vmalloc-ed buffers, flush the buffer using flush_kernel_vmap_range(),
use vmalloc_to_page() instead of virt_to_page() to obtain the pages of
the buffer, and invalidate the buffer addresses with
invalidate_kernel_vmap_range() on completion of read BIOs. This last
point is executed using the function bio_invalidate_vmalloc_pages()
which is defined only if the architecture defines
ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE, that is, if the architecture
actually needs the invalidation done.

Fixes: 515ce6061312 ("scsi: sd_zbc: Fix sd_zbc_report_zones() buffer allocation")
Fixes: e76239a3748c ("block: add a report_zones method")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
---
 block/bio.c | 28 +++++++++++++++++++++++++++-
 1 file changed, 27 insertions(+), 1 deletion(-)

diff --git a/block/bio.c b/block/bio.c
index 2050bb4aacb5..3b6e35f73fd7 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -16,6 +16,7 @@
 #include <linux/workqueue.h>
 #include <linux/cgroup.h>
 #include <linux/blk-cgroup.h>
+#include <linux/highmem.h>
 
 #include <trace/events/block.h>
 #include "blk.h"
@@ -1479,8 +1480,22 @@ void bio_unmap_user(struct bio *bio)
 	bio_put(bio);
 }
 
+static void bio_invalidate_vmalloc_pages(struct bio *bio)
+{
+#ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
+	if (bio->bi_private && !op_is_write(bio_op(bio))) {
+		unsigned long i, len = 0;
+
+		for (i = 0; i < bio->bi_vcnt; i++)
+			len += bio->bi_io_vec[i].bv_len;
+		invalidate_kernel_vmap_range(bio->bi_private, len);
+	}
+#endif
+}
+
 static void bio_map_kern_endio(struct bio *bio)
 {
+	bio_invalidate_vmalloc_pages(bio);
 	bio_put(bio);
 }
 
@@ -1501,6 +1516,8 @@ struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
 	unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 	unsigned long start = kaddr >> PAGE_SHIFT;
 	const int nr_pages = end - start;
+	bool is_vmalloc = is_vmalloc_addr(data);
+	struct page *page;
 	int offset, i;
 	struct bio *bio;
 
@@ -1508,6 +1525,11 @@ struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
 	if (!bio)
 		return ERR_PTR(-ENOMEM);
 
+	if (is_vmalloc) {
+		flush_kernel_vmap_range(data, len);
+		bio->bi_private = data;
+	}
+
 	offset = offset_in_page(kaddr);
 	for (i = 0; i < nr_pages; i++) {
 		unsigned int bytes = PAGE_SIZE - offset;
@@ -1518,7 +1540,11 @@ struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
 		if (bytes > len)
 			bytes = len;
 
-		if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
+		if (!is_vmalloc)
+			page = virt_to_page(data);
+		else
+			page = vmalloc_to_page(data);
+		if (bio_add_pc_page(q, bio, page, bytes,
 				    offset) < bytes) {
 			/* we don't support partial mappings */
 			bio_put(bio);
-- 
2.21.0

Re: [PATCH V6 1/4] block: Allow mapping of vmalloc-ed buffers

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH V6 1/4] block: Allow mapping of vmalloc-ed buffers

[Martin K. Petersen]

Damien,

> To allow the SCSI subsystem scsi_execute_req() function to issue
> requests using large buffers that are better allocated with vmalloc()
> rather than kmalloc(), modify bio_map_kern() to allow passing a buffer
> allocated with vmalloc().

Looks good to me.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen	Oracle Linux Engineering

[PATCH V6 2/4] block: Kill gfp_t argument of blkdev_report_zones()

[Damien Le Moal]

Only GFP_KERNEL and GFP_NOIO are used with blkdev_report_zones(). In
preparation of using vmalloc() for large report buffer and zone array
allocations used by this function, remove its "gfp_t gfp_mask" argument
and rely on the caller context to use memalloc_noio_save/restore() where
necessary (block layer zone revalidation and dm-zoned I/O error path).

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 block/blk-zoned.c              | 31 +++++++++++++++++++------------
 drivers/block/null_blk.h       |  3 +--
 drivers/block/null_blk_zoned.c |  3 +--
 drivers/md/dm-flakey.c         |  5 ++---
 drivers/md/dm-linear.c         |  5 ++---
 drivers/md/dm-zoned-metadata.c | 16 ++++++++++++----
 drivers/md/dm.c                |  6 ++----
 drivers/scsi/sd.h              |  3 +--
 drivers/scsi/sd_zbc.c          |  6 ++----
 fs/f2fs/super.c                |  4 +---
 include/linux/blkdev.h         |  5 ++---
 include/linux/device-mapper.h  |  3 +--
 12 files changed, 46 insertions(+), 44 deletions(-)

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index ae7e91bd0618..60dfc3f22350 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -14,6 +14,7 @@
 #include <linux/rbtree.h>
 #include <linux/blkdev.h>
 #include <linux/blk-mq.h>
+#include <linux/sched/mm.h>
 
 #include "blk.h"
 
@@ -117,8 +118,7 @@ static bool blkdev_report_zone(struct block_device *bdev, struct blk_zone *rep)
 }
 
 static int blk_report_zones(struct gendisk *disk, sector_t sector,
-			    struct blk_zone *zones, unsigned int *nr_zones,
-			    gfp_t gfp_mask)
+			    struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct request_queue *q = disk->queue;
 	unsigned int z = 0, n, nrz = *nr_zones;
@@ -127,8 +127,7 @@ static int blk_report_zones(struct gendisk *disk, sector_t sector,
 
 	while (z < nrz && sector < capacity) {
 		n = nrz - z;
-		ret = disk->fops->report_zones(disk, sector, &zones[z], &n,
-					       gfp_mask);
+		ret = disk->fops->report_zones(disk, sector, &zones[z], &n);
 		if (ret)
 			return ret;
 		if (!n)
@@ -149,17 +148,18 @@ static int blk_report_zones(struct gendisk *disk, sector_t sector,
  * @sector:	Sector from which to report zones
  * @zones:	Array of zone structures where to return the zones information
  * @nr_zones:	Number of zone structures in the zone array
- * @gfp_mask:	Memory allocation flags (for bio_alloc)
  *
  * Description:
  *    Get zone information starting from the zone containing @sector.
  *    The number of zone information reported may be less than the number
  *    requested by @nr_zones. The number of zones actually reported is
  *    returned in @nr_zones.
+ *    The caller must use memalloc_noXX_save/restore() calls to control
+ *    memory allocations done within this function (zone array and command
+ *    buffer allocation by the device driver).
  */
 int blkdev_report_zones(struct block_device *bdev, sector_t sector,
-			struct blk_zone *zones, unsigned int *nr_zones,
-			gfp_t gfp_mask)
+			struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct request_queue *q = bdev_get_queue(bdev);
 	unsigned int i, nrz;
@@ -184,7 +184,7 @@ int blkdev_report_zones(struct block_device *bdev, sector_t sector,
 	nrz = min(*nr_zones,
 		  __blkdev_nr_zones(q, bdev->bd_part->nr_sects - sector));
 	ret = blk_report_zones(bdev->bd_disk, get_start_sect(bdev) + sector,
-			       zones, &nrz, gfp_mask);
+			       zones, &nrz);
 	if (ret)
 		return ret;
 
@@ -305,9 +305,7 @@ int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
 	if (!zones)
 		return -ENOMEM;
 
-	ret = blkdev_report_zones(bdev, rep.sector,
-				  zones, &rep.nr_zones,
-				  GFP_KERNEL);
+	ret = blkdev_report_zones(bdev, rep.sector, zones, &rep.nr_zones);
 	if (ret)
 		goto out;
 
@@ -415,6 +413,7 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 	unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
 	unsigned int i, rep_nr_zones = 0, z = 0, nrz;
 	struct blk_zone *zones = NULL;
+	unsigned int noio_flag;
 	sector_t sector = 0;
 	int ret = 0;
 
@@ -427,6 +426,12 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 		return 0;
 	}
 
+	/*
+	 * Ensure that all memory allocations in this context are done as
+	 * if GFP_NOIO was specified.
+	 */
+	noio_flag = memalloc_noio_save();
+
 	if (!blk_queue_is_zoned(q) || !nr_zones) {
 		nr_zones = 0;
 		goto update;
@@ -449,7 +454,7 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 
 	while (z < nr_zones) {
 		nrz = min(nr_zones - z, rep_nr_zones);
-		ret = blk_report_zones(disk, sector, zones, &nrz, GFP_NOIO);
+		ret = blk_report_zones(disk, sector, zones, &nrz);
 		if (ret)
 			goto out;
 		if (!nrz)
@@ -480,6 +485,8 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 	blk_mq_unfreeze_queue(q);
 
 out:
+	memalloc_noio_restore(noio_flag);
+
 	free_pages((unsigned long)zones,
 		   get_order(rep_nr_zones * sizeof(struct blk_zone)));
 	kfree(seq_zones_wlock);
diff --git a/drivers/block/null_blk.h b/drivers/block/null_blk.h
index 34b22d6523ba..4b9bbe3bb5a1 100644
--- a/drivers/block/null_blk.h
+++ b/drivers/block/null_blk.h
@@ -89,8 +89,7 @@ struct nullb {
 int null_zone_init(struct nullb_device *dev);
 void null_zone_exit(struct nullb_device *dev);
 int null_zone_report(struct gendisk *disk, sector_t sector,
-		     struct blk_zone *zones, unsigned int *nr_zones,
-		     gfp_t gfp_mask);
+		     struct blk_zone *zones, unsigned int *nr_zones);
 void null_zone_write(struct nullb_cmd *cmd, sector_t sector,
 			unsigned int nr_sectors);
 void null_zone_reset(struct nullb_cmd *cmd, sector_t sector);
diff --git a/drivers/block/null_blk_zoned.c b/drivers/block/null_blk_zoned.c
index fca0c97ff1aa..cb28d93f2bd1 100644
--- a/drivers/block/null_blk_zoned.c
+++ b/drivers/block/null_blk_zoned.c
@@ -67,8 +67,7 @@ void null_zone_exit(struct nullb_device *dev)
 }
 
 int null_zone_report(struct gendisk *disk, sector_t sector,
-		     struct blk_zone *zones, unsigned int *nr_zones,
-		     gfp_t gfp_mask)
+		     struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct nullb *nullb = disk->private_data;
 	struct nullb_device *dev = nullb->dev;
diff --git a/drivers/md/dm-flakey.c b/drivers/md/dm-flakey.c
index a9bc518156f2..2900fbde89b3 100644
--- a/drivers/md/dm-flakey.c
+++ b/drivers/md/dm-flakey.c
@@ -461,15 +461,14 @@ static int flakey_prepare_ioctl(struct dm_target *ti, struct block_device **bdev
 
 #ifdef CONFIG_BLK_DEV_ZONED
 static int flakey_report_zones(struct dm_target *ti, sector_t sector,
-			       struct blk_zone *zones, unsigned int *nr_zones,
-			       gfp_t gfp_mask)
+			       struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct flakey_c *fc = ti->private;
 	int ret;
 
 	/* Do report and remap it */
 	ret = blkdev_report_zones(fc->dev->bdev, flakey_map_sector(ti, sector),
-				  zones, nr_zones, gfp_mask);
+				  zones, nr_zones);
 	if (ret != 0)
 		return ret;
 
diff --git a/drivers/md/dm-linear.c b/drivers/md/dm-linear.c
index ad980a38fb1e..ecefe6703736 100644
--- a/drivers/md/dm-linear.c
+++ b/drivers/md/dm-linear.c
@@ -137,15 +137,14 @@ static int linear_prepare_ioctl(struct dm_target *ti, struct block_device **bdev
 
 #ifdef CONFIG_BLK_DEV_ZONED
 static int linear_report_zones(struct dm_target *ti, sector_t sector,
-			       struct blk_zone *zones, unsigned int *nr_zones,
-			       gfp_t gfp_mask)
+			       struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct linear_c *lc = (struct linear_c *) ti->private;
 	int ret;
 
 	/* Do report and remap it */
 	ret = blkdev_report_zones(lc->dev->bdev, linear_map_sector(ti, sector),
-				  zones, nr_zones, gfp_mask);
+				  zones, nr_zones);
 	if (ret != 0)
 		return ret;
 
diff --git a/drivers/md/dm-zoned-metadata.c b/drivers/md/dm-zoned-metadata.c
index d8334cd45d7c..9faf3e49c7af 100644
--- a/drivers/md/dm-zoned-metadata.c
+++ b/drivers/md/dm-zoned-metadata.c
@@ -8,6 +8,7 @@
 
 #include <linux/module.h>
 #include <linux/crc32.h>
+#include <linux/sched/mm.h>
 
 #define	DM_MSG_PREFIX		"zoned metadata"
 
@@ -1162,8 +1163,7 @@ static int dmz_init_zones(struct dmz_metadata *zmd)
 	while (sector < dev->capacity) {
 		/* Get zone information */
 		nr_blkz = DMZ_REPORT_NR_ZONES;
-		ret = blkdev_report_zones(dev->bdev, sector, blkz,
-					  &nr_blkz, GFP_KERNEL);
+		ret = blkdev_report_zones(dev->bdev, sector, blkz, &nr_blkz);
 		if (ret) {
 			dmz_dev_err(dev, "Report zones failed %d", ret);
 			goto out;
@@ -1201,12 +1201,20 @@ static int dmz_init_zones(struct dmz_metadata *zmd)
 static int dmz_update_zone(struct dmz_metadata *zmd, struct dm_zone *zone)
 {
 	unsigned int nr_blkz = 1;
+	unsigned int noio_flag;
 	struct blk_zone blkz;
 	int ret;
 
-	/* Get zone information from disk */
+	/*
+	 * Get zone information from disk. Since blkdev_report_zones() uses
+	 * GFP_KERNEL by default for memory allocations, set the per-task
+	 * PF_MEMALLOC_NOIO flag so that all allocations are done as if
+	 * GFP_NOIO was specified.
+	 */
+	noio_flag = memalloc_noio_save();
 	ret = blkdev_report_zones(zmd->dev->bdev, dmz_start_sect(zmd, zone),
-				  &blkz, &nr_blkz, GFP_NOIO);
+				  &blkz, &nr_blkz);
+	memalloc_noio_restore(noio_flag);
 	if (!nr_blkz)
 		ret = -EIO;
 	if (ret) {
diff --git a/drivers/md/dm.c b/drivers/md/dm.c
index 5475081dcbd6..61f1152b74e9 100644
--- a/drivers/md/dm.c
+++ b/drivers/md/dm.c
@@ -441,8 +441,7 @@ static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 }
 
 static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
-			       struct blk_zone *zones, unsigned int *nr_zones,
-			       gfp_t gfp_mask)
+			       struct blk_zone *zones, unsigned int *nr_zones)
 {
 #ifdef CONFIG_BLK_DEV_ZONED
 	struct mapped_device *md = disk->private_data;
@@ -480,8 +479,7 @@ static int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
 	 * So there is no need to loop here trying to fill the entire array
 	 * of zones.
 	 */
-	ret = tgt->type->report_zones(tgt, sector, zones,
-				      nr_zones, gfp_mask);
+	ret = tgt->type->report_zones(tgt, sector, zones, nr_zones);
 
 out:
 	dm_put_live_table(md, srcu_idx);
diff --git a/drivers/scsi/sd.h b/drivers/scsi/sd.h
index 5796ace76225..38c50946fc42 100644
--- a/drivers/scsi/sd.h
+++ b/drivers/scsi/sd.h
@@ -213,8 +213,7 @@ extern blk_status_t sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd);
 extern void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
 			    struct scsi_sense_hdr *sshdr);
 extern int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
-			       struct blk_zone *zones, unsigned int *nr_zones,
-			       gfp_t gfp_mask);
+			       struct blk_zone *zones, unsigned int *nr_zones);
 
 #else /* CONFIG_BLK_DEV_ZONED */
 
diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
index 7334024b64f1..ec3764c8f3f1 100644
--- a/drivers/scsi/sd_zbc.c
+++ b/drivers/scsi/sd_zbc.c
@@ -109,13 +109,11 @@ static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
  * @sector: Start 512B sector of the report
  * @zones: Array of zone descriptors
  * @nr_zones: Number of descriptors in the array
- * @gfp_mask: Memory allocation mask
  *
  * Execute a report zones command on the target disk.
  */
 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
-			struct blk_zone *zones, unsigned int *nr_zones,
-			gfp_t gfp_mask)
+			struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct scsi_disk *sdkp = scsi_disk(disk);
 	unsigned int i, buflen, nrz = *nr_zones;
@@ -134,7 +132,7 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 	 */
 	buflen = min(queue_max_hw_sectors(disk->queue) << 9,
 		     roundup((nrz + 1) * 64, 512));
-	buf = kmalloc(buflen, gfp_mask);
+	buf = kmalloc(buflen, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
 
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 6b959bbb336a..4e91ba6c8a2e 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -2841,9 +2841,7 @@ static int init_blkz_info(struct f2fs_sb_info *sbi, int devi)
 	while (zones && sector < nr_sectors) {
 
 		nr_zones = F2FS_REPORT_NR_ZONES;
-		err = blkdev_report_zones(bdev, sector,
-					  zones, &nr_zones,
-					  GFP_KERNEL);
+		err = blkdev_report_zones(bdev, sector, zones, &nr_zones);
 		if (err)
 			break;
 		if (!nr_zones) {
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 592669bcc536..472ba74ca968 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -347,7 +347,7 @@ struct queue_limits {
 extern unsigned int blkdev_nr_zones(struct block_device *bdev);
 extern int blkdev_report_zones(struct block_device *bdev,
 			       sector_t sector, struct blk_zone *zones,
-			       unsigned int *nr_zones, gfp_t gfp_mask);
+			       unsigned int *nr_zones);
 extern int blkdev_reset_zones(struct block_device *bdev, sector_t sectors,
 			      sector_t nr_sectors, gfp_t gfp_mask);
 extern int blk_revalidate_disk_zones(struct gendisk *disk);
@@ -1684,8 +1684,7 @@ struct block_device_operations {
 	/* this callback is with swap_lock and sometimes page table lock held */
 	void (*swap_slot_free_notify) (struct block_device *, unsigned long);
 	int (*report_zones)(struct gendisk *, sector_t sector,
-			    struct blk_zone *zones, unsigned int *nr_zones,
-			    gfp_t gfp_mask);
+			    struct blk_zone *zones, unsigned int *nr_zones);
 	struct module *owner;
 	const struct pr_ops *pr_ops;
 };
diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h
index e1f51d607cc5..3b470cb03b66 100644
--- a/include/linux/device-mapper.h
+++ b/include/linux/device-mapper.h
@@ -95,8 +95,7 @@ typedef int (*dm_prepare_ioctl_fn) (struct dm_target *ti, struct block_device **
 
 typedef int (*dm_report_zones_fn) (struct dm_target *ti, sector_t sector,
 				   struct blk_zone *zones,
-				   unsigned int *nr_zones,
-				   gfp_t gfp_mask);
+				   unsigned int *nr_zones);
 
 /*
  * These iteration functions are typically used to check (and combine)
-- 
2.21.0

Re: [PATCH V6 2/4] block: Kill gfp_t argument of blkdev_report_zones()

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH V6 2/4] block: Kill gfp_t argument of blkdev_report_zones()

[Martin K. Petersen]

Damien,

> Only GFP_KERNEL and GFP_NOIO are used with blkdev_report_zones(). In
> preparation of using vmalloc() for large report buffer and zone array
> allocations used by this function, remove its "gfp_t gfp_mask" argument
> and rely on the caller context to use memalloc_noio_save/restore() where
> necessary (block layer zone revalidation and dm-zoned I/O error path).

LGTM.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen	Oracle Linux Engineering

[PATCH V6 3/4] sd_zbc: Fix report zones buffer allocation

[Damien Le Moal]

During disk scan and revalidation done with sd_revalidate(), the zones
of a zoned disk are checked using the helper function
blk_revalidate_disk_zones() if a configuration change is detected
(change in the number of zones or zone size). The function
blk_revalidate_disk_zones() issues report_zones calls that are very
large, that is, to obtain zone information for all zones of the disk
with a single command. The size of the report zones command buffer
necessary for such large request generally is lower than the disk
max_hw_sectors and KMALLOC_MAX_SIZE (4MB) and succeeds on boot (no
memory fragmentation), but often fail at run time (e.g. hot-plug
event). This causes the disk revalidation to fail and the disk
capacity to be changed to 0.

This problem can be avoided by using vmalloc() instead of kmalloc() for
the buffer allocation. To limit the amount of memory to be allocated,
this patch also introduces the arbitrary SD_ZBC_REPORT_MAX_ZONES
maximum number of zones to report with a single report zones command.
This limit may be lowered further to satisfy the disk max_hw_sectors
limit. Finally, to ensure that the vmalloc-ed buffer can always be
mapped in a request, the buffer size is further limited to at most
queue_max_segments() pages, allowing successful mapping of the buffer
even in the worst case scenario where none of the buffer pages are
contiguous.

Fixes: 515ce6061312 ("scsi: sd_zbc: Fix sd_zbc_report_zones() buffer allocation")
Fixes: e76239a3748c ("block: add a report_zones method")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 drivers/scsi/sd_zbc.c | 104 ++++++++++++++++++++++++++++++------------
 1 file changed, 75 insertions(+), 29 deletions(-)

diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
index ec3764c8f3f1..db16c19e05c4 100644
--- a/drivers/scsi/sd_zbc.c
+++ b/drivers/scsi/sd_zbc.c
@@ -9,6 +9,8 @@
  */
 
 #include <linux/blkdev.h>
+#include <linux/vmalloc.h>
+#include <linux/sched/mm.h>
 
 #include <asm/unaligned.h>
 
@@ -50,7 +52,7 @@ static void sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
 /**
  * sd_zbc_do_report_zones - Issue a REPORT ZONES scsi command.
  * @sdkp: The target disk
- * @buf: Buffer to use for the reply
+ * @buf: vmalloc-ed buffer to use for the reply
  * @buflen: the buffer size
  * @lba: Start LBA of the report
  * @partial: Do partial report
@@ -79,7 +81,6 @@ static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
 	put_unaligned_be32(buflen, &cmd[10]);
 	if (partial)
 		cmd[14] = ZBC_REPORT_ZONE_PARTIAL;
-	memset(buf, 0, buflen);
 
 	result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
 				  buf, buflen, &sshdr,
@@ -103,6 +104,53 @@ static int sd_zbc_do_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
 	return 0;
 }
 
+/*
+ * Maximum number of zones to get with one report zones command.
+ */
+#define SD_ZBC_REPORT_MAX_ZONES		8192U
+
+/**
+ * Allocate a buffer for report zones reply.
+ * @sdkp: The target disk
+ * @nr_zones: Maximum number of zones to report
+ * @buflen: Size of the buffer allocated
+ *
+ * Try to allocate a reply buffer for the number of requested zones.
+ * The size of the buffer allocated may be smaller than requested to
+ * satify the device constraint (max_hw_sectors, max_segments, etc).
+ *
+ * Return the address of the allocated buffer and update @buflen with
+ * the size of the allocated buffer.
+ */
+static void *sd_zbc_alloc_report_buffer(struct scsi_disk *sdkp,
+					unsigned int nr_zones, size_t *buflen)
+{
+	struct request_queue *q = sdkp->disk->queue;
+	size_t bufsize;
+	void *buf;
+
+	/*
+	 * Report zone buffer size should be at most 64B times the number of
+	 * zones requested plus the 64B reply header, but should be at least
+	 * SECTOR_SIZE for ATA devices.
+	 * Make sure that this size does not exceed the hardware capabilities.
+	 * Furthermore, since the report zone command cannot be split, make
+	 * sure that the allocated buffer can always be mapped by limiting the
+	 * number of pages allocated to the HBA max segments limit.
+	 */
+	nr_zones = min(nr_zones, SD_ZBC_REPORT_MAX_ZONES);
+	bufsize = roundup((nr_zones + 1) * 64, 512);
+	bufsize = min_t(size_t, bufsize,
+			queue_max_hw_sectors(q) << SECTOR_SHIFT);
+	bufsize = min_t(size_t, bufsize, queue_max_segments(q) << PAGE_SHIFT);
+
+	buf = vzalloc(bufsize);
+	if (buf)
+		*buflen = bufsize;
+
+	return buf;
+}
+
 /**
  * sd_zbc_report_zones - Disk report zones operation.
  * @disk: The target disk
@@ -116,30 +164,23 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 			struct blk_zone *zones, unsigned int *nr_zones)
 {
 	struct scsi_disk *sdkp = scsi_disk(disk);
-	unsigned int i, buflen, nrz = *nr_zones;
+	unsigned int i, nrz = *nr_zones;
 	unsigned char *buf;
-	size_t offset = 0;
+	size_t buflen = 0, offset = 0;
 	int ret = 0;
 
 	if (!sd_is_zoned(sdkp))
 		/* Not a zoned device */
 		return -EOPNOTSUPP;
 
-	/*
-	 * Get a reply buffer for the number of requested zones plus a header,
-	 * without exceeding the device maximum command size. For ATA disks,
-	 * buffers must be aligned to 512B.
-	 */
-	buflen = min(queue_max_hw_sectors(disk->queue) << 9,
-		     roundup((nrz + 1) * 64, 512));
-	buf = kmalloc(buflen, GFP_KERNEL);
+	buf = sd_zbc_alloc_report_buffer(sdkp, nrz, &buflen);
 	if (!buf)
 		return -ENOMEM;
 
 	ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
 			sectors_to_logical(sdkp->device, sector), true);
 	if (ret)
-		goto out_free_buf;
+		goto out;
 
 	nrz = min(nrz, get_unaligned_be32(&buf[0]) / 64);
 	for (i = 0; i < nrz; i++) {
@@ -150,8 +191,8 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 
 	*nr_zones = nrz;
 
-out_free_buf:
-	kfree(buf);
+out:
+	kvfree(buf);
 
 	return ret;
 }
@@ -285,8 +326,6 @@ static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
 	return 0;
 }
 
-#define SD_ZBC_BUF_SIZE 131072U
-
 /**
  * sd_zbc_check_zones - Check the device capacity and zone sizes
  * @sdkp: Target disk
@@ -302,22 +341,28 @@ static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
  */
 static int sd_zbc_check_zones(struct scsi_disk *sdkp, u32 *zblocks)
 {
+	size_t bufsize, buflen;
+	unsigned int noio_flag;
 	u64 zone_blocks = 0;
 	sector_t max_lba, block = 0;
 	unsigned char *buf;
 	unsigned char *rec;
-	unsigned int buf_len;
-	unsigned int list_length;
 	int ret;
 	u8 same;
 
+	/* Do all memory allocations as if GFP_NOIO was specified */
+	noio_flag = memalloc_noio_save();
+
 	/* Get a buffer */
-	buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
-	if (!buf)
-		return -ENOMEM;
+	buf = sd_zbc_alloc_report_buffer(sdkp, SD_ZBC_REPORT_MAX_ZONES,
+					 &bufsize);
+	if (!buf) {
+		ret = -ENOMEM;
+		goto out;
+	}
 
 	/* Do a report zone to get max_lba and the same field */
-	ret = sd_zbc_do_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0, false);
+	ret = sd_zbc_do_report_zones(sdkp, buf, bufsize, 0, false);
 	if (ret)
 		goto out_free;
 
@@ -353,12 +398,12 @@ static int sd_zbc_check_zones(struct scsi_disk *sdkp, u32 *zblocks)
 	do {
 
 		/* Parse REPORT ZONES header */
-		list_length = get_unaligned_be32(&buf[0]) + 64;
+		buflen = min_t(size_t, get_unaligned_be32(&buf[0]) + 64,
+			       bufsize);
 		rec = buf + 64;
-		buf_len = min(list_length, SD_ZBC_BUF_SIZE);
 
 		/* Parse zone descriptors */
-		while (rec < buf + buf_len) {
+		while (rec < buf + buflen) {
 			u64 this_zone_blocks = get_unaligned_be64(&rec[8]);
 
 			if (zone_blocks == 0) {
@@ -374,8 +419,8 @@ static int sd_zbc_check_zones(struct scsi_disk *sdkp, u32 *zblocks)
 		}
 
 		if (block < sdkp->capacity) {
-			ret = sd_zbc_do_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE,
-						     block, true);
+			ret = sd_zbc_do_report_zones(sdkp, buf, bufsize, block,
+						     true);
 			if (ret)
 				goto out_free;
 		}
@@ -406,7 +451,8 @@ static int sd_zbc_check_zones(struct scsi_disk *sdkp, u32 *zblocks)
 	}
 
 out_free:
-	kfree(buf);
+	memalloc_noio_restore(noio_flag);
+	kvfree(buf);
 
 	return ret;
 }
-- 
2.21.0

Re: [PATCH V6 3/4] sd_zbc: Fix report zones buffer allocation

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH V6 3/4] sd_zbc: Fix report zones buffer allocation

[Martin K. Petersen]

Damien,

> During disk scan and revalidation done with sd_revalidate(), the zones
> of a zoned disk are checked using the helper function
> blk_revalidate_disk_zones() if a configuration change is detected
> (change in the number of zones or zone size). The function
> blk_revalidate_disk_zones() issues report_zones calls that are very
> large, that is, to obtain zone information for all zones of the disk
> with a single command. The size of the report zones command buffer
> necessary for such large request generally is lower than the disk
> max_hw_sectors and KMALLOC_MAX_SIZE (4MB) and succeeds on boot (no
> memory fragmentation), but often fail at run time (e.g. hot-plug
> event). This causes the disk revalidation to fail and the disk
> capacity to be changed to 0.

Probably easiest to funnel this through block with the rest of the
series.

Acked-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH V6 3/4] sd_zbc: Fix report zones buffer allocation

[Damien Le Moal]

Martin,

On 2019/07/12 9:28, Martin K. Petersen wrote:
> 
> Damien,
> 
>> During disk scan and revalidation done with sd_revalidate(), the zones
>> of a zoned disk are checked using the helper function
>> blk_revalidate_disk_zones() if a configuration change is detected
>> (change in the number of zones or zone size). The function
>> blk_revalidate_disk_zones() issues report_zones calls that are very
>> large, that is, to obtain zone information for all zones of the disk
>> with a single command. The size of the report zones command buffer
>> necessary for such large request generally is lower than the disk
>> max_hw_sectors and KMALLOC_MAX_SIZE (4MB) and succeeds on boot (no
>> memory fragmentation), but often fail at run time (e.g. hot-plug
>> event). This causes the disk revalidation to fail and the disk
>> capacity to be changed to 0.
> 
> Probably easiest to funnel this through block with the rest of the
> series.

Sounds good to me.

Jens,

Could you please consider this series for 5.3 ? We have been using it in tests
since 2-3 weeks ago (5.2-rc6) without any problems. All revalidation problems
disappeared and with no side effects detected.

Thank you.

> Acked-by: Martin K. Petersen <martin.petersen@oracle.com>

Thanks !

-- 
Damien Le Moal
Western Digital Research

[PATCH V6 4/4] block: Limit zone array allocation size

[Damien Le Moal]

Limit the size of the struct blk_zone array used in
blk_revalidate_disk_zones() to avoid memory allocation failures leading
to disk revalidation failure. Also further reduce the likelyhood of
such failures by using kvcalloc() (that is vmalloc()) instead of
allocating contiguous pages with alloc_pages().

Fixes: 515ce6061312 ("scsi: sd_zbc: Fix sd_zbc_report_zones() buffer allocation")
Fixes: e76239a3748c ("block: add a report_zones method")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 block/blk-zoned.c      | 36 ++++++++++++++++++++----------------
 include/linux/blkdev.h |  5 +++++
 2 files changed, 25 insertions(+), 16 deletions(-)

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index 60dfc3f22350..79ad269b545d 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -14,6 +14,8 @@
 #include <linux/rbtree.h>
 #include <linux/blkdev.h>
 #include <linux/blk-mq.h>
+#include <linux/mm.h>
+#include <linux/vmalloc.h>
 #include <linux/sched/mm.h>
 
 #include "blk.h"
@@ -371,22 +373,25 @@ static inline unsigned long *blk_alloc_zone_bitmap(int node,
  * Allocate an array of struct blk_zone to get nr_zones zone information.
  * The allocated array may be smaller than nr_zones.
  */
-static struct blk_zone *blk_alloc_zones(int node, unsigned int *nr_zones)
+static struct blk_zone *blk_alloc_zones(unsigned int *nr_zones)
 {
-	size_t size = *nr_zones * sizeof(struct blk_zone);
-	struct page *page;
-	int order;
-
-	for (order = get_order(size); order >= 0; order--) {
-		page = alloc_pages_node(node, GFP_NOIO | __GFP_ZERO, order);
-		if (page) {
-			*nr_zones = min_t(unsigned int, *nr_zones,
-				(PAGE_SIZE << order) / sizeof(struct blk_zone));
-			return page_address(page);
-		}
+	struct blk_zone *zones;
+	size_t nrz = min(*nr_zones, BLK_ZONED_REPORT_MAX_ZONES);
+
+	/*
+	 * GFP_KERNEL here is meaningless as the caller task context has
+	 * the PF_MEMALLOC_NOIO flag set in blk_revalidate_disk_zones()
+	 * with memalloc_noio_save().
+	 */
+	zones = kvcalloc(nrz, sizeof(struct blk_zone), GFP_KERNEL);
+	if (!zones) {
+		*nr_zones = 0;
+		return NULL;
 	}
 
-	return NULL;
+	*nr_zones = nrz;
+
+	return zones;
 }
 
 void blk_queue_free_zone_bitmaps(struct request_queue *q)
@@ -448,7 +453,7 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 
 	/* Get zone information and initialize seq_zones_bitmap */
 	rep_nr_zones = nr_zones;
-	zones = blk_alloc_zones(q->node, &rep_nr_zones);
+	zones = blk_alloc_zones(&rep_nr_zones);
 	if (!zones)
 		goto out;
 
@@ -487,8 +492,7 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 out:
 	memalloc_noio_restore(noio_flag);
 
-	free_pages((unsigned long)zones,
-		   get_order(rep_nr_zones * sizeof(struct blk_zone)));
+	kvfree(zones);
 	kfree(seq_zones_wlock);
 	kfree(seq_zones_bitmap);
 
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 472ba74ca968..5ace0bb77213 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -344,6 +344,11 @@ struct queue_limits {
 
 #ifdef CONFIG_BLK_DEV_ZONED
 
+/*
+ * Maximum number of zones to report with a single report zones command.
+ */
+#define BLK_ZONED_REPORT_MAX_ZONES	8192U
+
 extern unsigned int blkdev_nr_zones(struct block_device *bdev);
 extern int blkdev_report_zones(struct block_device *bdev,
 			       sector_t sector, struct blk_zone *zones,
-- 
2.21.0

Re: [PATCH V6 4/4] block: Limit zone array allocation size

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

Re: [PATCH V6 4/4] block: Limit zone array allocation size

[Martin K. Petersen]

Damien,

> Limit the size of the struct blk_zone array used in
> blk_revalidate_disk_zones() to avoid memory allocation failures
> leading to disk revalidation failure. Also further reduce the
> likelyhood of such failures by using kvcalloc() (that is vmalloc())
> instead of allocating contiguous pages with alloc_pages().

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH V6 0/4] Fix zone revalidation memory allocation failures

[Damien Le Moal]

On 2019/07/01 14:09, Damien Le Moal wrote:
> This series addresses a recuring problem with zone revalidation
> failures observed during extensive testing with memory constrained
> system and device hot-plugging.

Jens, Martin,

Any comment regarding this series ?

Best regards.

-- 
Damien Le Moal
Western Digital Research

Re: [PATCH V6 0/4] Fix zone revalidation memory allocation failures

[Jens Axboe]

On 7/7/19 8:02 PM, Damien Le Moal wrote:
> On 2019/07/01 14:09, Damien Le Moal wrote:
>> This series addresses a recuring problem with zone revalidation
>> failures observed during extensive testing with memory constrained
>> system and device hot-plugging.
> 
> Jens, Martin,
> 
> Any comment regarding this series ?

LGTM, I'll queue it up for this release.

-- 
Jens Axboe

Re: [PATCH V6 0/4] Fix zone revalidation memory allocation failures

[Jens Axboe]

On 7/11/19 8:05 PM, Jens Axboe wrote:
> On 7/7/19 8:02 PM, Damien Le Moal wrote:
>> On 2019/07/01 14:09, Damien Le Moal wrote:
>>> This series addresses a recuring problem with zone revalidation
>>> failures observed during extensive testing with memory constrained
>>> system and device hot-plugging.
>>
>> Jens, Martin,
>>
>> Any comment regarding this series ?
> 
> LGTM, I'll queue it up for this release.

This broke !CONFIG_BLK_DEV_ZONED builds for null_blk, btw. Please be
sure to test with zoned enabled and disabled in your builds when
changing things that affect both.

I fixed it up:

http://git.kernel.dk/cgit/linux-block/commit/?h=for-linus&id=e347946439ed70e3af0d0c330b36d5648e71727b

-- 
Jens Axboe

Re: [PATCH V6 0/4] Fix zone revalidation memory allocation failures

[Damien Le Moal]

On 2019/07/13 0:13, Jens Axboe wrote:
> On 7/11/19 8:05 PM, Jens Axboe wrote:
>> On 7/7/19 8:02 PM, Damien Le Moal wrote:
>>> On 2019/07/01 14:09, Damien Le Moal wrote:
>>>> This series addresses a recuring problem with zone revalidation
>>>> failures observed during extensive testing with memory constrained
>>>> system and device hot-plugging.
>>>
>>> Jens, Martin,
>>>
>>> Any comment regarding this series ?
>>
>> LGTM, I'll queue it up for this release.
> 
> This broke !CONFIG_BLK_DEV_ZONED builds for null_blk, btw. Please be
> sure to test with zoned enabled and disabled in your builds when
> changing things that affect both.

I always check the !CONFIG_BLK_DEV_ZONED case. But clearly I made a mistake
somewhere on this one and missed the problem. My apologies for that.

> 
> I fixed it up:
> 
> http://git.kernel.dk/cgit/linux-block/commit/?h=for-linus&id=e347946439ed70e3af0d0c330b36d5648e71727b

Thank you for this.


-- 
Damien Le Moal
Western Digital Research