[PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Chaitanya Kulkarni]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=y, Size: 12778 bytes --]

Hi,

This patch-series is based on the original RFC patch series:-
https://www.spinics.net/lists/linux-block/msg47933.html.

I've designed a rough testcase based on the information present
in the mailing list archive for original RFC, it may need
some corrections from the author.

If anyone is interested, test results are at the end of this patch.

Following is the original cover-letter :-

Information about continuous extent placement may be useful
for some block devices. Say, distributed network filesystems,
which provide block device interface, may use this information
for better blocks placement over the nodes in their cluster,
and for better performance. Block devices, which map a file
on another filesystem (loop), may request the same length extent
on underlining filesystem for less fragmentation and for batching
allocation requests. Also, hypervisors like QEMU may use this
information for optimization of cluster allocations.

This patchset introduces REQ_OP_ASSIGN_RANGE, which is going
to be used for forwarding user's fallocate(0) requests into
block device internals. It rather similar to existing
REQ_OP_DISCARD, REQ_OP_WRITE_ZEROES, etc. The corresponding
exported primitive is called blkdev_issue_assign_range().
See [1/3] for the details.

Patch [2/3] teaches loop driver to handle REQ_OP_ASSIGN_RANGE
requests by calling fallocate(0).

Patch [3/3] makes ext4 to notify a block device about fallocate(0).

Here is a simple test I did:
https://gist.github.com/tkhai/5b788651cdb74c1dbff3500745878856

I attached a file on ext4 to loop. Then, created ext4 partition
on loop device and started the test in the partition. Direct-io
is enabled on loop.

The test fallocates 4G file and writes from some offset with
given step, then it chooses another offset and repeats. After
the test all the blocks in the file become written.

The results shows that batching extents-assigning requests improves
the performance:

Before patchset: real ~ 1min 27sec
After patchset:  real ~ 1min 16sec (18% better)

Ordinary fallocate() before writes improves the performance
by batching the requests. These results just show, the same
is in case of forwarding extents information to underlining
filesystem.

Regards,
Chaitanya

Changes from RFC:-

1. Add missing plumbing for REQ_OP_ASSIGN_RANGE similar to write-zeores.
2. Add a prep patch to create a helper to submit payloadless bios.
3. Design a testcases around the description present in the
   cover-letter.

Chaitanya Kulkarni (1):
  block: create payloadless issue bio helper

Kirill Tkhai (3):
  block: Add support for REQ_OP_ASSIGN_RANGE
  loop: Forward REQ_OP_ASSIGN_RANGE into fallocate(0)
  ext4: Notify block device about alloc-assigned blk

 block/blk-core.c          |   5 ++
 block/blk-lib.c           | 115 +++++++++++++++++++++++++++++++-------
 block/blk-merge.c         |  21 +++++++
 block/blk-settings.c      |  19 +++++++
 block/blk-zoned.c         |   1 +
 block/bounce.c            |   1 +
 drivers/block/loop.c      |   5 ++
 fs/ext4/ext4.h            |   2 +
 fs/ext4/extents.c         |  12 +++-
 include/linux/bio.h       |   9 ++-
 include/linux/blk_types.h |   2 +
 include/linux/blkdev.h    |  34 +++++++++++
 12 files changed, 201 insertions(+), 25 deletions(-)

1. Setup :-
-----------
# git log --oneline -5 
c64a4c781915 (HEAD -> req-op-assign-range) ext4: Notify block device about alloc-assigned blk
000cbc6720a4 loop: Forward REQ_OP_ASSIGN_RANGE into fallocate(0)
89ceed8cac80 block: Add support for REQ_OP_ASSIGN_RANGE
a798743e87e7 block: create payloadless issue bio helper
b53df2e7442c (tag: block-5.6-2020-03-13) block: Fix partition support for host aware zoned block devices

# cat /proc/kallsyms | grep -i blkdev_issue_assign_range
ffffffffa3264a80 T blkdev_issue_assign_range
ffffffffa4027184 r __ksymtab_blkdev_issue_assign_range
ffffffffa40524be r __kstrtabns_blkdev_issue_assign_range
ffffffffa405a8eb r __kstrtab_blkdev_issue_assign_range

2. Test program, will be moved to blktest once code is upstream :-
-----------------
#define _GNU_SOURCE
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <fcntl.h>
#include <errno.h>

#define BLOCK_SIZE 4096
#define STEP (BLOCK_SIZE * 16)
#define SIZE (1024 * 1024 * 1024ULL)

int main(int argc, char *argv[])
{
	int fd, step, ret = 0;
	unsigned long i;
	void *buf;

	if (posix_memalign(&buf, BLOCK_SIZE, SIZE)) {
		perror("alloc");
		exit(1);
	}

	fd = open("/mnt/loop0/file.img", O_RDWR | O_CREAT | O_DIRECT);
	if (fd < 0) {
		perror("open");
		exit(1);
	}

	if (ftruncate(fd, SIZE)) {
		perror("ftruncate");
		exit(1);
	}

	ret = fallocate(fd, 0, 0, SIZE);
	if (ret) {
		perror("fallocate");
		exit(1);
	}
	
	for (step = STEP - BLOCK_SIZE; step >= 0; step -= BLOCK_SIZE) {
		printf("step=%u\n", step);
		for (i = step; i < SIZE; i += STEP) {
			errno = 0;
			if (pwrite(fd, buf, BLOCK_SIZE, i) != BLOCK_SIZE) {
				perror("pwrite");
				exit(1);
			}
		}

		if (fsync(fd)) {
			perror("fsync");
			exit(1);
		}
	}
	return 0;
}

3. Test script, will be moved to blktests once code is upstream :-
------------------------------------------------------------------
# cat req_op_assign_test.sh 
#!/bin/bash -x

NULLB_FILE="/mnt/backend/data"
NULLB_MNT="/mnt/backend"
LOOP_MNT="/mnt/loop0"

delete_loop()
{
	umount ${LOOP_MNT}
	losetup -D
	sleep 3
}

delete_nullb()
{
	umount ${NULLB_MNT}
	echo 1 > config/nullb/nullb0/power
	rmdir config/nullb/nullb0
	sleep 3
}

unload_modules()
{
	rmmod drivers/block/loop.ko
	rmmod fs/ext4/ext4.ko
	rmmod drivers/block/null_blk.ko
	lsmod | grep -e ext4 -e loop -e null_blk
}

unload()
{
	delete_loop
	delete_nullb
	unload_modules
}

load_ext4()
{
	make -j $(nproc) M=fs/ext4 modules
	local src=fs/ext4/
	local dest=/lib/modules/`uname -r`/kernel/fs/ext4
	\cp ${src}/ext4.ko ${dest}/

	modprobe mbcache
	modprobe jbd2
	sleep 1
	insmod fs/ext4/ext4.ko
	sleep 1
}

load_nullb()
{
	local src=drivers/block/
	local dest=/lib/modules/`uname -r`/kernel/drivers/block
	\cp ${src}/null_blk.ko ${dest}/

	modprobe null_blk nr_devices=0
	sleep 1

	mkdir config/nullb/nullb0
	tree config/nullb/nullb0

	echo 1 > config/nullb/nullb0/memory_backed
	echo 512 > config/nullb/nullb0/blocksize 

	# 20 GB
	echo 20480 > config/nullb/nullb0/size 
	echo 1 > config/nullb/nullb0/power
	sleep 2
	IDX=`cat config/nullb/nullb0/index`
	lsblk | grep null${IDX}
	sleep 1

	mkfs.ext4 /dev/nullb0 
	mount /dev/nullb0 ${NULLB_MNT}
	sleep 1
	mount | grep nullb

	# 10 GB
	dd if=/dev/zero of=${NULLB_FILE} count=2621440 bs=4096
}

load_loop()
{
	local src=drivers/block/
	local dest=/lib/modules/`uname -r`/kernel/drivers/block
	\cp ${src}/loop.ko ${dest}/

	insmod drivers/block/loop.ko max_loop=1
	sleep 3
	/root/util-linux/losetup --direct-io=off /dev/loop0 ${NULLB_FILE}
	sleep 3
	/root/util-linux/losetup
	ls -l /dev/loop*
	dmesg -c 
	mkfs.ext4 /dev/loop0
	mount /dev/loop0 ${LOOP_MNT}
	mount | grep loop0
}

load()
{
	make -j $(nproc) M=drivers/block modules

	load_ext4
	load_nullb
	load_loop
	sleep 1
	sync
	sync
	sync
}

unload
load
time ./test

4. Test Results :-
------------------

# ./req_op_assign_test.sh 
+ NULLB_FILE=/mnt/backend/data
+ NULLB_MNT=/mnt/backend
+ LOOP_MNT=/mnt/loop0
+ unload
+ delete_loop
+ umount /mnt/loop0
+ losetup -D
+ sleep 3
+ delete_nullb
+ umount /mnt/backend
+ echo 1
+ rmdir config/nullb/nullb0
+ sleep 3
+ unload_modules
+ rmmod drivers/block/loop.ko
+ rmmod fs/ext4/ext4.ko
+ rmmod drivers/block/null_blk.ko
+ lsmod
+ grep -e ext4 -e loop -e null_blk
+ load
++ nproc
+ make -j 32 M=drivers/block modules
  CC [M]  drivers/block/loop.o
  MODPOST 11 modules
  CC [M]  drivers/block/loop.mod.o
  LD [M]  drivers/block/loop.ko
+ load_ext4
++ nproc
+ make -j 32 M=fs/ext4 modules
  CC [M]  fs/ext4/balloc.o
  CC [M]  fs/ext4/bitmap.o
  CC [M]  fs/ext4/block_validity.o
  CC [M]  fs/ext4/dir.o
  CC [M]  fs/ext4/ext4_jbd2.o
  CC [M]  fs/ext4/extents.o
  CC [M]  fs/ext4/extents_status.o
  CC [M]  fs/ext4/file.o
  CC [M]  fs/ext4/fsmap.o
  CC [M]  fs/ext4/fsync.o
  CC [M]  fs/ext4/hash.o
  CC [M]  fs/ext4/ialloc.o
  CC [M]  fs/ext4/indirect.o
  CC [M]  fs/ext4/inline.o
  CC [M]  fs/ext4/inode.o
  CC [M]  fs/ext4/ioctl.o
  CC [M]  fs/ext4/mballoc.o
  CC [M]  fs/ext4/migrate.o
  CC [M]  fs/ext4/mmp.o
  CC [M]  fs/ext4/move_extent.o
  CC [M]  fs/ext4/namei.o
  CC [M]  fs/ext4/page-io.o
  CC [M]  fs/ext4/readpage.o
  CC [M]  fs/ext4/resize.o
  CC [M]  fs/ext4/super.o
  CC [M]  fs/ext4/symlink.o
  CC [M]  fs/ext4/sysfs.o
  CC [M]  fs/ext4/xattr.o
  CC [M]  fs/ext4/xattr_trusted.o
  CC [M]  fs/ext4/xattr_user.o
  CC [M]  fs/ext4/acl.o
  CC [M]  fs/ext4/xattr_security.o
  LD [M]  fs/ext4/ext4.o
  MODPOST 1 modules
  LD [M]  fs/ext4/ext4.ko
+ local src=fs/ext4/
++ uname -r
+ local dest=/lib/modules/5.6.0-rc3lbk+/kernel/fs/ext4
+ cp fs/ext4//ext4.ko /lib/modules/5.6.0-rc3lbk+/kernel/fs/ext4/
+ modprobe mbcache
+ modprobe jbd2
+ sleep 1
+ insmod fs/ext4/ext4.ko
+ sleep 1
+ load_nullb
+ local src=drivers/block/
++ uname -r
+ local dest=/lib/modules/5.6.0-rc3lbk+/kernel/drivers/block
+ cp drivers/block//null_blk.ko /lib/modules/5.6.0-rc3lbk+/kernel/drivers/block/
+ modprobe null_blk nr_devices=0
+ sleep 1
+ mkdir config/nullb/nullb0
+ tree config/nullb/nullb0
config/nullb/nullb0
├── badblocks
├── blocking
├── blocksize
├── cache_size
├── completion_nsec
├── discard
├── home_node
├── hw_queue_depth
├── index
├── irqmode
├── mbps
├── memory_backed
├── power
├── queue_mode
├── size
├── submit_queues
├── use_per_node_hctx
├── zoned
├── zone_nr_conv
└── zone_size

0 directories, 20 files
+ echo 1
+ echo 512
+ echo 20480
+ echo 1
+ sleep 2
++ cat config/nullb/nullb0/index
+ IDX=0
+ lsblk
+ grep null0
+ sleep 1
+ mkfs.ext4 /dev/nullb0
mke2fs 1.42.9 (28-Dec-2013)
Filesystem label=
OS type: Linux
Block size=4096 (log=2)
Fragment size=4096 (log=2)
Stride=0 blocks, Stripe width=0 blocks
1310720 inodes, 5242880 blocks
262144 blocks (5.00%) reserved for the super user
First data block=0
Maximum filesystem blocks=2153775104
160 block groups
32768 blocks per group, 32768 fragments per group
8192 inodes per group
Superblock backups stored on blocks: 
	32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208, 
	4096000

Allocating group tables: done                            
Writing inode tables: done                            
Creating journal (32768 blocks): done
Writing superblocks and filesystem accounting information: done   

+ mount /dev/nullb0 /mnt/backend
+ sleep 1
+ mount
+ grep nullb
/dev/nullb0 on /mnt/backend type ext4 (rw,relatime,seclabel)
+ dd if=/dev/zero of=/mnt/backend/data count=2621440 bs=4096
2621440+0 records in
2621440+0 records out
10737418240 bytes (11 GB) copied, 27.4579 s, 391 MB/s
+ load_loop
+ local src=drivers/block/
++ uname -r
+ local dest=/lib/modules/5.6.0-rc3lbk+/kernel/drivers/block
+ cp drivers/block//loop.ko /lib/modules/5.6.0-rc3lbk+/kernel/drivers/block/
+ insmod drivers/block/loop.ko max_loop=1
+ sleep 3
+ /root/util-linux/losetup --direct-io=off /dev/loop0 /mnt/backend/data
+ sleep 3
+ /root/util-linux/losetup
NAME       SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE         DIO LOG-SEC
/dev/loop0         0      0         0  0 /mnt/backend/data   0     512
+ ls -l /dev/loop0 /dev/loop-control
brw-rw----. 1 root disk  7,   0 Mar 29 10:28 /dev/loop0
crw-rw----. 1 root disk 10, 237 Mar 29 10:28 /dev/loop-control
+ dmesg -c
[42963.967060] null_blk: module loaded
[42968.419481] EXT4-fs (nullb0): mounted filesystem with ordered data mode. Opts: (null)
[42996.928141] loop: module loaded
+ mkfs.ext4 /dev/loop0
mke2fs 1.42.9 (28-Dec-2013)
Discarding device blocks: done                            
Filesystem label=
OS type: Linux
Block size=4096 (log=2)
Fragment size=4096 (log=2)
Stride=0 blocks, Stripe width=0 blocks
655360 inodes, 2621440 blocks
131072 blocks (5.00%) reserved for the super user
First data block=0
Maximum filesystem blocks=2151677952
80 block groups
32768 blocks per group, 32768 fragments per group
8192 inodes per group
Superblock backups stored on blocks: 
	32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632

Allocating group tables: done                            
Writing inode tables: done                            
Creating journal (32768 blocks): done
Writing superblocks and filesystem accounting information: done 

+ mount /dev/loop0 /mnt/loop0
+ mount
+ grep loop0
/dev/loop0 on /mnt/loop0 type ext4 (rw,relatime,seclabel)
+ sleep 1
+ sync
+ sync
+ sync
+ ./test
step=61440
step=57344
step=53248
step=49152
step=45056
step=40960
step=36864
step=32768
step=28672
step=24576
step=20480
step=16384
step=12288
step=8192
step=4096
step=0

real	9m34.472s
user	0m0.062s
sys	0m5.783s

-- 
2.22.0

[PATCH 1/4] block: create payloadless issue bio helper

[Chaitanya Kulkarni]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=y, Size: 2560 bytes --]

This is a prep-patch that creates a helper to submit payloadless bio
with all the required arguments. This is needed to avoid the code
repetition in blk-lib,c so that new payloadless ops can use it.  

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
---
 block/blk-lib.c | 51 +++++++++++++++++++++++++++++--------------------
 1 file changed, 30 insertions(+), 21 deletions(-)

diff --git a/block/blk-lib.c b/block/blk-lib.c
index 5f2c429d4378..8e53e393703c 100644
--- a/block/blk-lib.c
+++ b/block/blk-lib.c
@@ -209,13 +209,40 @@ int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
 }
 EXPORT_SYMBOL(blkdev_issue_write_same);
 
+static void __blkdev_issue_payloadless(struct block_device *bdev,
+		unsigned op, sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
+		struct bio **biop, unsigned bio_opf, unsigned int max_sectors)
+{
+	struct bio *bio = *biop;
+
+	while (nr_sects) {
+		bio = blk_next_bio(bio, 0, gfp_mask);
+		bio->bi_iter.bi_sector = sector;
+		bio_set_dev(bio, bdev);
+		bio->bi_opf = op;
+		bio->bi_opf |= bio_opf;
+
+		if (nr_sects > max_sectors) {
+			bio->bi_iter.bi_size = max_sectors << 9;
+			nr_sects -= max_sectors;
+			sector += max_sectors;
+		} else {
+			bio->bi_iter.bi_size = nr_sects << 9;
+			nr_sects = 0;
+		}
+		cond_resched();
+	}
+
+	*biop = bio;
+}
+
 static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 		sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
 		struct bio **biop, unsigned flags)
 {
-	struct bio *bio = *biop;
 	unsigned int max_write_zeroes_sectors;
 	struct request_queue *q = bdev_get_queue(bdev);
+	unsigned int unmap = (flags & BLKDEV_ZERO_NOUNMAP) ? REQ_NOUNMAP : 0;
 
 	if (!q)
 		return -ENXIO;
@@ -229,26 +256,8 @@ static int __blkdev_issue_write_zeroes(struct block_device *bdev,
 	if (max_write_zeroes_sectors == 0)
 		return -EOPNOTSUPP;
 
-	while (nr_sects) {
-		bio = blk_next_bio(bio, 0, gfp_mask);
-		bio->bi_iter.bi_sector = sector;
-		bio_set_dev(bio, bdev);
-		bio->bi_opf = REQ_OP_WRITE_ZEROES;
-		if (flags & BLKDEV_ZERO_NOUNMAP)
-			bio->bi_opf |= REQ_NOUNMAP;
-
-		if (nr_sects > max_write_zeroes_sectors) {
-			bio->bi_iter.bi_size = max_write_zeroes_sectors << 9;
-			nr_sects -= max_write_zeroes_sectors;
-			sector += max_write_zeroes_sectors;
-		} else {
-			bio->bi_iter.bi_size = nr_sects << 9;
-			nr_sects = 0;
-		}
-		cond_resched();
-	}
-
-	*biop = bio;
+	__blkdev_issue_payloadless(bdev, REQ_OP_WRITE_ZEROES, sector, nr_sects,
+		       gfp_mask, biop, unmap, max_write_zeroes_sectors);
 	return 0;
 }
 
-- 
2.22.0

[PATCH 2/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Chaitanya Kulkarni]

From: Kirill Tkhai <ktkhai@virtuozzo.com>

This operation allows to notify a device about the fact, that some
sectors range was chosen by a filesystem as a single extent, and
the device should try its best to reflect that (keep the range as a
single hunk in its internals, or represent the range as minimal set of
hunks). Speaking directly, the operation is for forwarding fallocate(0)
requests into an essence, on which the device is based.

This may be useful for some distributed network filesystems, providing
block device interface, for optimization of their blocks placement over
the cluster nodes.

Also, block devices mapping a file (like loop) are users of that, since
this allows to allocate more continuous extents and since this batches
blocks allocation requests. In addition, hypervisors like QEMU may use
this for better blocks placement.

This patch adds a new blkdev_issue_assign_range() primitive, which is
rather similar to existing blkdev_issue_{*} api. Also, a new queue
limit.max_assign_range_sectors is added.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
---
 block/blk-core.c          |  5 +++
 block/blk-lib.c           | 64 +++++++++++++++++++++++++++++++++++++++
 block/blk-merge.c         | 21 +++++++++++++
 block/blk-settings.c      | 19 ++++++++++++
 block/blk-zoned.c         |  1 +
 block/bounce.c            |  1 +
 include/linux/bio.h       |  9 ++++--
 include/linux/blk_types.h |  2 ++
 include/linux/blkdev.h    | 34 +++++++++++++++++++++
 9 files changed, 153 insertions(+), 3 deletions(-)

diff --git a/block/blk-core.c b/block/blk-core.c
index 60dc9552ef8d..25165fa8fe46 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -137,6 +137,7 @@ static const char *const blk_op_name[] = {
 	REQ_OP_NAME(ZONE_FINISH),
 	REQ_OP_NAME(WRITE_SAME),
 	REQ_OP_NAME(WRITE_ZEROES),
+	REQ_OP_NAME(ASSIGN_RANGE),
 	REQ_OP_NAME(SCSI_IN),
 	REQ_OP_NAME(SCSI_OUT),
 	REQ_OP_NAME(DRV_IN),
@@ -952,6 +953,10 @@ generic_make_request_checks(struct bio *bio)
 		if (!q->limits.max_write_zeroes_sectors)
 			goto not_supported;
 		break;
+	case REQ_OP_ASSIGN_RANGE:
+		if (!q->limits.max_assign_range_sectors)
+			goto not_supported;
+		break;
 	default:
 		break;
 	}
diff --git a/block/blk-lib.c b/block/blk-lib.c
index 8e53e393703c..16dc9dbf6c79 100644
--- a/block/blk-lib.c
+++ b/block/blk-lib.c
@@ -414,3 +414,67 @@ int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
 	return ret;
 }
 EXPORT_SYMBOL(blkdev_issue_zeroout);
+
+static int __blkdev_issue_assign_range(struct block_device *bdev,
+		sector_t sector, sector_t nr_sects, gfp_t gfp_mask,
+		struct bio **biop)
+{
+	unsigned int max_assign_range_sectors;
+	struct request_queue *q = bdev_get_queue(bdev);
+
+	if (!q)
+		return -ENXIO;
+
+	if (bdev_read_only(bdev))
+		return -EPERM;
+
+	max_assign_range_sectors = bdev_assign_range_sectors(bdev);
+
+	if (max_assign_range_sectors == 0)
+		return -EOPNOTSUPP;
+
+	__blkdev_issue_payloadless(bdev, REQ_OP_ASSIGN_RANGE, sector, nr_sects,
+			gfp_mask, biop, 0, max_assign_range_sectors);
+	return 0;
+}
+
+/**
+ * __blkdev_issue_assign_range - generate number of assign range bios
+ * @bdev:	blockdev to issue
+ * @sector:	start sector
+ * @nr_sects:	number of sectors to write
+ * @gfp_mask:	memory allocation flags (for bio_alloc)
+ * @biop:	pointer to anchor bio
+ *
+ * Description:
+ *  Assign a block range for batched allocation requests. Useful in stacking
+ *  block device on the top of the file system.
+ *
+ */
+int blkdev_issue_assign_range(struct block_device *bdev, sector_t sector,
+			sector_t nr_sects, gfp_t gfp_mask)
+{
+	int ret = 0;
+	sector_t bs_mask;
+	struct blk_plug plug;
+	struct bio *bio = NULL;
+
+	if (bdev_assign_range_sectors(bdev) == 0)
+		return 0;
+
+	bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1;
+	if ((sector | nr_sects) & bs_mask)
+		return -EINVAL;
+
+	blk_start_plug(&plug);
+	ret = __blkdev_issue_assign_range(bdev, sector, nr_sects,
+					  gfp_mask, &bio);
+	if (ret == 0 && bio) {
+		ret = submit_bio_wait(bio);
+		bio_put(bio);
+	}
+	blk_finish_plug(&plug);
+
+	return ret;
+}
+EXPORT_SYMBOL(blkdev_issue_assign_range);
diff --git a/block/blk-merge.c b/block/blk-merge.c
index 1534ed736363..441d1620de03 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -116,6 +116,22 @@ static struct bio *blk_bio_write_zeroes_split(struct request_queue *q,
 	return bio_split(bio, q->limits.max_write_zeroes_sectors, GFP_NOIO, bs);
 }
 
+static struct bio *blk_bio_assign_range_split(struct request_queue *q,
+					      struct bio *bio,
+					      struct bio_set *bs,
+					      unsigned *nsegs)
+{
+	*nsegs = 0;
+
+	if (!q->limits.max_assign_range_sectors)
+		return NULL;
+
+	if (bio_sectors(bio) <= q->limits.max_assign_range_sectors)
+		return NULL;
+
+	return bio_split(bio, q->limits.max_assign_range_sectors, GFP_NOIO, bs);
+}
+
 static struct bio *blk_bio_write_same_split(struct request_queue *q,
 					    struct bio *bio,
 					    struct bio_set *bs,
@@ -308,6 +324,10 @@ void __blk_queue_split(struct request_queue *q, struct bio **bio,
 		split = blk_bio_write_zeroes_split(q, *bio, &q->bio_split,
 				nr_segs);
 		break;
+	case REQ_OP_ASSIGN_RANGE:
+		split = blk_bio_assign_range_split(q, *bio, &q->bio_split,
+				nr_segs);
+		break;
 	case REQ_OP_WRITE_SAME:
 		split = blk_bio_write_same_split(q, *bio, &q->bio_split,
 				nr_segs);
@@ -386,6 +406,7 @@ unsigned int blk_recalc_rq_segments(struct request *rq)
 	case REQ_OP_DISCARD:
 	case REQ_OP_SECURE_ERASE:
 	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_ASSIGN_RANGE:
 		return 0;
 	case REQ_OP_WRITE_SAME:
 		return 1;
diff --git a/block/blk-settings.c b/block/blk-settings.c
index c8eda2e7b91e..6beee0585580 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -48,6 +48,7 @@ void blk_set_default_limits(struct queue_limits *lim)
 	lim->chunk_sectors = 0;
 	lim->max_write_same_sectors = 0;
 	lim->max_write_zeroes_sectors = 0;
+	lim->max_assign_range_sectors = 0;
 	lim->max_discard_sectors = 0;
 	lim->max_hw_discard_sectors = 0;
 	lim->discard_granularity = 0;
@@ -83,6 +84,7 @@ void blk_set_stacking_limits(struct queue_limits *lim)
 	lim->max_dev_sectors = UINT_MAX;
 	lim->max_write_same_sectors = UINT_MAX;
 	lim->max_write_zeroes_sectors = UINT_MAX;
+	lim->max_assign_range_sectors = UINT_MAX;
 }
 EXPORT_SYMBOL(blk_set_stacking_limits);
 
@@ -257,6 +259,21 @@ void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
 }
 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
 
+/**
+ * blk_queue_max_assign_range_sectors - set max sectors for a single
+ *                                      assign_range
+ *
+ * @q:  the request queue for the device
+ * @max_assign_range_sectors: maximum number of sectors to assign range per
+ *                            command
+ **/
+void blk_queue_max_assign_range_sectors(struct request_queue *q,
+		unsigned int max_assign_range_sectors)
+{
+	q->limits.max_assign_range_sectors = max_assign_range_sectors;
+}
+EXPORT_SYMBOL(blk_queue_max_assign_range_sectors);
+
 /**
  * blk_queue_max_segments - set max hw segments for a request for this queue
  * @q:  the request queue for the device
@@ -506,6 +523,8 @@ int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
 					b->max_write_same_sectors);
 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
 					b->max_write_zeroes_sectors);
+	t->max_assign_range_sectors = min(t->max_assign_range_sectors,
+				    b->max_assign_range_sectors);
 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
 
 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index 05741c6f618b..14b1fbed40f6 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -41,6 +41,7 @@ bool blk_req_needs_zone_write_lock(struct request *rq)
 
 	switch (req_op(rq)) {
 	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_ASSIGN_RANGE:
 	case REQ_OP_WRITE_SAME:
 	case REQ_OP_WRITE:
 		return blk_rq_zone_is_seq(rq);
diff --git a/block/bounce.c b/block/bounce.c
index f8ed677a1bf7..0eeb20b290ec 100644
--- a/block/bounce.c
+++ b/block/bounce.c
@@ -257,6 +257,7 @@ static struct bio *bounce_clone_bio(struct bio *bio_src, gfp_t gfp_mask,
 	case REQ_OP_DISCARD:
 	case REQ_OP_SECURE_ERASE:
 	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_ASSIGN_RANGE:
 		break;
 	case REQ_OP_WRITE_SAME:
 		bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
diff --git a/include/linux/bio.h b/include/linux/bio.h
index 853d92ceee64..8617abfc6f78 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -64,7 +64,8 @@ static inline bool bio_has_data(struct bio *bio)
 	    bio->bi_iter.bi_size &&
 	    bio_op(bio) != REQ_OP_DISCARD &&
 	    bio_op(bio) != REQ_OP_SECURE_ERASE &&
-	    bio_op(bio) != REQ_OP_WRITE_ZEROES)
+	    bio_op(bio) != REQ_OP_WRITE_ZEROES &&
+	    bio_op(bio) != REQ_OP_ASSIGN_RANGE)
 		return true;
 
 	return false;
@@ -75,7 +76,8 @@ static inline bool bio_no_advance_iter(struct bio *bio)
 	return bio_op(bio) == REQ_OP_DISCARD ||
 	       bio_op(bio) == REQ_OP_SECURE_ERASE ||
 	       bio_op(bio) == REQ_OP_WRITE_SAME ||
-	       bio_op(bio) == REQ_OP_WRITE_ZEROES;
+	       bio_op(bio) == REQ_OP_WRITE_ZEROES ||
+	       bio_op(bio) == REQ_OP_ASSIGN_RANGE;
 }
 
 static inline bool bio_mergeable(struct bio *bio)
@@ -178,7 +180,7 @@ static inline unsigned bio_segments(struct bio *bio)
 	struct bvec_iter iter;
 
 	/*
-	 * We special case discard/write same/write zeroes, because they
+	 * We special case discard/write same/write zeroes/assign range, because
 	 * interpret bi_size differently:
 	 */
 
@@ -186,6 +188,7 @@ static inline unsigned bio_segments(struct bio *bio)
 	case REQ_OP_DISCARD:
 	case REQ_OP_SECURE_ERASE:
 	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_ASSIGN_RANGE:
 		return 0;
 	case REQ_OP_WRITE_SAME:
 		return 1;
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 70254ae11769..bef450026044 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -296,6 +296,8 @@ enum req_opf {
 	REQ_OP_ZONE_CLOSE	= 11,
 	/* Transition a zone to full */
 	REQ_OP_ZONE_FINISH	= 12,
+	/* Assign a sector range */
+	REQ_OP_ASSIGN_RANGE	= 15,
 
 	/* SCSI passthrough using struct scsi_request */
 	REQ_OP_SCSI_IN		= 32,
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index f629d40c645c..3a63c14e2cbc 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -336,6 +336,7 @@ struct queue_limits {
 	unsigned int		max_hw_discard_sectors;
 	unsigned int		max_write_same_sectors;
 	unsigned int		max_write_zeroes_sectors;
+	unsigned int		max_assign_range_sectors;
 	unsigned int		discard_granularity;
 	unsigned int		discard_alignment;
 
@@ -747,6 +748,9 @@ static inline bool rq_mergeable(struct request *rq)
 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
 		return false;
 
+	if (req_op(rq) == REQ_OP_ASSIGN_RANGE)
+		return false;
+
 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
 		return false;
 	if (rq->rq_flags & RQF_NOMERGE_FLAGS)
@@ -1000,6 +1004,10 @@ static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
 	if (unlikely(op == REQ_OP_WRITE_ZEROES))
 		return q->limits.max_write_zeroes_sectors;
 
+	if (unlikely(op == REQ_OP_ASSIGN_RANGE))
+		return min(q->limits.max_assign_range_sectors,
+			   UINT_MAX >> SECTOR_SHIFT);
+
 	return q->limits.max_sectors;
 }
 
@@ -1077,6 +1085,8 @@ extern void blk_queue_max_write_same_sectors(struct request_queue *q,
 		unsigned int max_write_same_sectors);
 extern void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
 		unsigned int max_write_same_sectors);
+extern void blk_queue_max_assign_range_sectors(struct request_queue *q,
+		unsigned int max_assign_range_sectors);
 extern void blk_queue_logical_block_size(struct request_queue *, unsigned int);
 extern void blk_queue_physical_block_size(struct request_queue *, unsigned int);
 extern void blk_queue_alignment_offset(struct request_queue *q,
@@ -1246,6 +1256,20 @@ static inline int sb_issue_zeroout(struct super_block *sb, sector_t block,
 				    gfp_mask, 0);
 }
 
+extern int blkdev_issue_assign_range(struct block_device *bdev, sector_t sector,
+		sector_t nr_sects, gfp_t gfp_mask);
+
+static inline int sb_issue_assign_range(struct super_block *sb, sector_t block,
+		sector_t nr_blocks, gfp_t gfp_mask)
+{
+	return blkdev_issue_assign_range(sb->s_bdev,
+					 block << (sb->s_blocksize_bits -
+						   SECTOR_SHIFT),
+					 nr_blocks << (sb->s_blocksize_bits -
+						       SECTOR_SHIFT),
+					 gfp_mask);
+}
+
 extern int blk_verify_command(unsigned char *cmd, fmode_t mode);
 
 enum blk_default_limits {
@@ -1427,6 +1451,16 @@ static inline unsigned int bdev_write_zeroes_sectors(struct block_device *bdev)
 	return 0;
 }
 
+static inline unsigned int bdev_assign_range_sectors(struct block_device *bdev)
+{
+	struct request_queue *q = bdev_get_queue(bdev);
+
+	if (q)
+		return q->limits.max_assign_range_sectors;
+
+	return 0;
+}
+
 static inline enum blk_zoned_model bdev_zoned_model(struct block_device *bdev)
 {
 	struct request_queue *q = bdev_get_queue(bdev);
-- 
2.22.0

[PATCH 3/4] loop: Forward REQ_OP_ASSIGN_RANGE into fallocate(0)

[Chaitanya Kulkarni]

From: Kirill Tkhai <ktkhai@virtuozzo.com>

From: Kirill Tkhai <ktkhai@virtuozzo.com>

Send fallocate(0) request into underlining filesystem after upper
filesystem sent REQ_OP_ASSIGN_RANGE request to block device.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
[Use blk_queue_max_assign_range_sectors() from newly updated previous
 patch.]
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
---
 drivers/block/loop.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 739b372a5112..0a28db66c485 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -609,6 +609,8 @@ static int do_req_filebacked(struct loop_device *lo, struct request *rq)
 				FALLOC_FL_PUNCH_HOLE);
 	case REQ_OP_DISCARD:
 		return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
+	case REQ_OP_ASSIGN_RANGE:
+		return lo_fallocate(lo, rq, pos, 0);
 	case REQ_OP_WRITE:
 		if (lo->transfer)
 			return lo_write_transfer(lo, rq, pos);
@@ -876,6 +878,7 @@ static void loop_config_discard(struct loop_device *lo)
 		q->limits.discard_granularity = 0;
 		q->limits.discard_alignment = 0;
 		blk_queue_max_discard_sectors(q, 0);
+		blk_queue_max_assign_range_sectors(q, 0);
 		blk_queue_max_write_zeroes_sectors(q, 0);
 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
 		return;
@@ -886,6 +889,7 @@ static void loop_config_discard(struct loop_device *lo)
 
 	blk_queue_max_discard_sectors(q, UINT_MAX >> 9);
 	blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9);
+	blk_queue_max_assign_range_sectors(q, UINT_MAX >> 9);
 	blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
 }
 
@@ -1917,6 +1921,7 @@ static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
 	case REQ_OP_FLUSH:
 	case REQ_OP_DISCARD:
 	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_ASSIGN_RANGE:
 		cmd->use_aio = false;
 		break;
 	default:
-- 
2.22.0

[PATCH 4/4] ext4: Notify block device about alloc-assigned blk

[Chaitanya Kulkarni]

From: Kirill Tkhai <ktkhai@virtuozzo.com>

From: Kirill Tkhai <ktkhai@virtuozzo.com>

Call sb_issue_assign_range() after extent range was allocated on user
request. Hopeful, this helps block device to maintain its internals in
the best way, if this is appliable.

Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
---
 fs/ext4/ext4.h    |  2 ++
 fs/ext4/extents.c | 12 +++++++++++-
 2 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 61b37a052052..0d0fa9904147 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -622,6 +622,8 @@ enum {
 	 * allows jbd2 to avoid submitting data before commit. */
 #define EXT4_GET_BLOCKS_IO_SUBMIT		0x0400
 
+#define EXT4_GET_BLOCKS_SUBMIT_ALLOC		0x0800
+
 /*
  * The bit position of these flags must not overlap with any of the
  * EXT4_GET_BLOCKS_*.  They are used by ext4_find_extent(),
diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index 954013d6076b..598b700c4d4c 100644
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4449,6 +4449,14 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
 		ar.len = allocated;
 
 got_allocated_blocks:
+	if ((flags & EXT4_GET_BLOCKS_SUBMIT_ALLOC) &&
+			inode->i_fop->fallocate) {
+		err = sb_issue_assign_range(inode->i_sb, newblock,
+				EXT4_C2B(sbi, allocated_clusters), GFP_NOFS);
+		if (err)
+			goto free_on_err;
+	}
+
 	/* try to insert new extent into found leaf and return */
 	ext4_ext_store_pblock(&newex, newblock + offset);
 	newex.ee_len = cpu_to_le16(ar.len);
@@ -4466,6 +4474,7 @@ int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
 		err = ext4_ext_insert_extent(handle, inode, &path,
 					     &newex, flags);
 
+free_on_err:
 	if (err && free_on_err) {
 		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
 			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
@@ -4733,7 +4742,8 @@ static long ext4_zero_range(struct file *file, loff_t offset,
 			goto out_mutex;
 	}
 
-	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
+	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT |
+		EXT4_GET_BLOCKS_SUBMIT_ALLOC;
 	if (mode & FALLOC_FL_KEEP_SIZE)
 		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
 
-- 
2.22.0

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Konstantin Khlebnikov]

On 29/03/2020 20.47, Chaitanya Kulkarni wrote:
> Hi,
> 
> This patch-series is based on the original RFC patch series:-
> https://www.spinics.net/lists/linux-block/msg47933.html.
> 
> I've designed a rough testcase based on the information present
> in the mailing list archive for original RFC, it may need
> some corrections from the author.
> 
> If anyone is interested, test results are at the end of this patch.
> 
> Following is the original cover-letter :-
> 
> Information about continuous extent placement may be useful
> for some block devices. Say, distributed network filesystems,
> which provide block device interface, may use this information
> for better blocks placement over the nodes in their cluster,
> and for better performance. Block devices, which map a file
> on another filesystem (loop), may request the same length extent
> on underlining filesystem for less fragmentation and for batching
> allocation requests. Also, hypervisors like QEMU may use this
> information for optimization of cluster allocations.
> 
> This patchset introduces REQ_OP_ASSIGN_RANGE, which is going
> to be used for forwarding user's fallocate(0) requests into
> block device internals. It rather similar to existing
> REQ_OP_DISCARD, REQ_OP_WRITE_ZEROES, etc. The corresponding
> exported primitive is called blkdev_issue_assign_range().

What exact semantics of that?

It may/must preserve present data or may/must discard them, or may fill range with random garbage?

Obviously I prefer weakest one - may discard data, may return garbage, may do nothing.

I.e. lower layer could reuse blocks without zeroing, for encrypted storage this is even safe.

So, this works as third type of dicasrd in addtion to REQ_OP_DISCARD and REQ_OP_SECURE_ERASE.

> See [1/3] for the details.
> 
> Patch [2/3] teaches loop driver to handle REQ_OP_ASSIGN_RANGE
> requests by calling fallocate(0).
> 
> Patch [3/3] makes ext4 to notify a block device about fallocate(0).
> 
> Here is a simple test I did:
> https://gist.github.com/tkhai/5b788651cdb74c1dbff3500745878856
> 
> I attached a file on ext4 to loop. Then, created ext4 partition
> on loop device and started the test in the partition. Direct-io
> is enabled on loop.
> 
> The test fallocates 4G file and writes from some offset with
> given step, then it chooses another offset and repeats. After
> the test all the blocks in the file become written.
> 
> The results shows that batching extents-assigning requests improves
> the performance:
> 
> Before patchset: real ~ 1min 27sec
> After patchset:  real ~ 1min 16sec (18% better)
> 
> Ordinary fallocate() before writes improves the performance
> by batching the requests. These results just show, the same
> is in case of forwarding extents information to underlining
> filesystem.
> 
> Regards,
> Chaitanya
> 
> Changes from RFC:-
> 
> 1. Add missing plumbing for REQ_OP_ASSIGN_RANGE similar to write-zeores.
> 2. Add a prep patch to create a helper to submit payloadless bios.
> 3. Design a testcases around the description present in the
>     cover-letter.
> 
> Chaitanya Kulkarni (1):
>    block: create payloadless issue bio helper
> 
> Kirill Tkhai (3):
>    block: Add support for REQ_OP_ASSIGN_RANGE
>    loop: Forward REQ_OP_ASSIGN_RANGE into fallocate(0)
>    ext4: Notify block device about alloc-assigned blk
> 
>   block/blk-core.c          |   5 ++
>   block/blk-lib.c           | 115 +++++++++++++++++++++++++++++++-------
>   block/blk-merge.c         |  21 +++++++
>   block/blk-settings.c      |  19 +++++++
>   block/blk-zoned.c         |   1 +
>   block/bounce.c            |   1 +
>   drivers/block/loop.c      |   5 ++
>   fs/ext4/ext4.h            |   2 +
>   fs/ext4/extents.c         |  12 +++-
>   include/linux/bio.h       |   9 ++-
>   include/linux/blk_types.h |   2 +
>   include/linux/blkdev.h    |  34 +++++++++++
>   12 files changed, 201 insertions(+), 25 deletions(-)
> 
> 1. Setup :-
> -----------
> # git log --oneline -5
> c64a4c781915 (HEAD -> req-op-assign-range) ext4: Notify block device about alloc-assigned blk
> 000cbc6720a4 loop: Forward REQ_OP_ASSIGN_RANGE into fallocate(0)
> 89ceed8cac80 block: Add support for REQ_OP_ASSIGN_RANGE
> a798743e87e7 block: create payloadless issue bio helper
> b53df2e7442c (tag: block-5.6-2020-03-13) block: Fix partition support for host aware zoned block devices
> 
> # cat /proc/kallsyms | grep -i blkdev_issue_assign_range
> ffffffffa3264a80 T blkdev_issue_assign_range
> ffffffffa4027184 r __ksymtab_blkdev_issue_assign_range
> ffffffffa40524be r __kstrtabns_blkdev_issue_assign_range
> ffffffffa405a8eb r __kstrtab_blkdev_issue_assign_range
> 
> 2. Test program, will be moved to blktest once code is upstream :-
> -----------------
> #define _GNU_SOURCE
> #include <sys/types.h>
> #include <unistd.h>
> #include <stdlib.h>
> #include <stdio.h>
> #include <fcntl.h>
> #include <errno.h>
> 
> #define BLOCK_SIZE 4096
> #define STEP (BLOCK_SIZE * 16)
> #define SIZE (1024 * 1024 * 1024ULL)
> 
> int main(int argc, char *argv[])
> {
> 	int fd, step, ret = 0;
> 	unsigned long i;
> 	void *buf;
> 
> 	if (posix_memalign(&buf, BLOCK_SIZE, SIZE)) {
> 		perror("alloc");
> 		exit(1);
> 	}
> 
> 	fd = open("/mnt/loop0/file.img", O_RDWR | O_CREAT | O_DIRECT);
> 	if (fd < 0) {
> 		perror("open");
> 		exit(1);
> 	}
> 
> 	if (ftruncate(fd, SIZE)) {
> 		perror("ftruncate");
> 		exit(1);
> 	}
> 
> 	ret = fallocate(fd, 0, 0, SIZE);
> 	if (ret) {
> 		perror("fallocate");
> 		exit(1);
> 	}
> 	
> 	for (step = STEP - BLOCK_SIZE; step >= 0; step -= BLOCK_SIZE) {
> 		printf("step=%u\n", step);
> 		for (i = step; i < SIZE; i += STEP) {
> 			errno = 0;
> 			if (pwrite(fd, buf, BLOCK_SIZE, i) != BLOCK_SIZE) {
> 				perror("pwrite");
> 				exit(1);
> 			}
> 		}
> 
> 		if (fsync(fd)) {
> 			perror("fsync");
> 			exit(1);
> 		}
> 	}
> 	return 0;
> }
> 
> 3. Test script, will be moved to blktests once code is upstream :-
> ------------------------------------------------------------------
> # cat req_op_assign_test.sh
> #!/bin/bash -x
> 
> NULLB_FILE="/mnt/backend/data"
> NULLB_MNT="/mnt/backend"
> LOOP_MNT="/mnt/loop0"
> 
> delete_loop()
> {
> 	umount ${LOOP_MNT}
> 	losetup -D
> 	sleep 3
> }
> 
> delete_nullb()
> {
> 	umount ${NULLB_MNT}
> 	echo 1 > config/nullb/nullb0/power
> 	rmdir config/nullb/nullb0
> 	sleep 3
> }
> 
> unload_modules()
> {
> 	rmmod drivers/block/loop.ko
> 	rmmod fs/ext4/ext4.ko
> 	rmmod drivers/block/null_blk.ko
> 	lsmod | grep -e ext4 -e loop -e null_blk
> }
> 
> unload()
> {
> 	delete_loop
> 	delete_nullb
> 	unload_modules
> }
> 
> load_ext4()
> {
> 	make -j $(nproc) M=fs/ext4 modules
> 	local src=fs/ext4/
> 	local dest=/lib/modules/`uname -r`/kernel/fs/ext4
> 	\cp ${src}/ext4.ko ${dest}/
> 
> 	modprobe mbcache
> 	modprobe jbd2
> 	sleep 1
> 	insmod fs/ext4/ext4.ko
> 	sleep 1
> }
> 
> load_nullb()
> {
> 	local src=drivers/block/
> 	local dest=/lib/modules/`uname -r`/kernel/drivers/block
> 	\cp ${src}/null_blk.ko ${dest}/
> 
> 	modprobe null_blk nr_devices=0
> 	sleep 1
> 
> 	mkdir config/nullb/nullb0
> 	tree config/nullb/nullb0
> 
> 	echo 1 > config/nullb/nullb0/memory_backed
> 	echo 512 > config/nullb/nullb0/blocksize
> 
> 	# 20 GB
> 	echo 20480 > config/nullb/nullb0/size
> 	echo 1 > config/nullb/nullb0/power
> 	sleep 2
> 	IDX=`cat config/nullb/nullb0/index`
> 	lsblk | grep null${IDX}
> 	sleep 1
> 
> 	mkfs.ext4 /dev/nullb0
> 	mount /dev/nullb0 ${NULLB_MNT}
> 	sleep 1
> 	mount | grep nullb
> 
> 	# 10 GB
> 	dd if=/dev/zero of=${NULLB_FILE} count=2621440 bs=4096
> }
> 
> load_loop()
> {
> 	local src=drivers/block/
> 	local dest=/lib/modules/`uname -r`/kernel/drivers/block
> 	\cp ${src}/loop.ko ${dest}/
> 
> 	insmod drivers/block/loop.ko max_loop=1
> 	sleep 3
> 	/root/util-linux/losetup --direct-io=off /dev/loop0 ${NULLB_FILE}
> 	sleep 3
> 	/root/util-linux/losetup
> 	ls -l /dev/loop*
> 	dmesg -c
> 	mkfs.ext4 /dev/loop0
> 	mount /dev/loop0 ${LOOP_MNT}
> 	mount | grep loop0
> }
> 
> load()
> {
> 	make -j $(nproc) M=drivers/block modules
> 
> 	load_ext4
> 	load_nullb
> 	load_loop
> 	sleep 1
> 	sync
> 	sync
> 	sync
> }
> 
> unload
> load
> time ./test
> 
> 4. Test Results :-
> ------------------
> 
> # ./req_op_assign_test.sh
> + NULLB_FILE=/mnt/backend/data
> + NULLB_MNT=/mnt/backend
> + LOOP_MNT=/mnt/loop0
> + unload
> + delete_loop
> + umount /mnt/loop0
> + losetup -D
> + sleep 3
> + delete_nullb
> + umount /mnt/backend
> + echo 1
> + rmdir config/nullb/nullb0
> + sleep 3
> + unload_modules
> + rmmod drivers/block/loop.ko
> + rmmod fs/ext4/ext4.ko
> + rmmod drivers/block/null_blk.ko
> + lsmod
> + grep -e ext4 -e loop -e null_blk
> + load
> ++ nproc
> + make -j 32 M=drivers/block modules
>    CC [M]  drivers/block/loop.o
>    MODPOST 11 modules
>    CC [M]  drivers/block/loop.mod.o
>    LD [M]  drivers/block/loop.ko
> + load_ext4
> ++ nproc
> + make -j 32 M=fs/ext4 modules
>    CC [M]  fs/ext4/balloc.o
>    CC [M]  fs/ext4/bitmap.o
>    CC [M]  fs/ext4/block_validity.o
>    CC [M]  fs/ext4/dir.o
>    CC [M]  fs/ext4/ext4_jbd2.o
>    CC [M]  fs/ext4/extents.o
>    CC [M]  fs/ext4/extents_status.o
>    CC [M]  fs/ext4/file.o
>    CC [M]  fs/ext4/fsmap.o
>    CC [M]  fs/ext4/fsync.o
>    CC [M]  fs/ext4/hash.o
>    CC [M]  fs/ext4/ialloc.o
>    CC [M]  fs/ext4/indirect.o
>    CC [M]  fs/ext4/inline.o
>    CC [M]  fs/ext4/inode.o
>    CC [M]  fs/ext4/ioctl.o
>    CC [M]  fs/ext4/mballoc.o
>    CC [M]  fs/ext4/migrate.o
>    CC [M]  fs/ext4/mmp.o
>    CC [M]  fs/ext4/move_extent.o
>    CC [M]  fs/ext4/namei.o
>    CC [M]  fs/ext4/page-io.o
>    CC [M]  fs/ext4/readpage.o
>    CC [M]  fs/ext4/resize.o
>    CC [M]  fs/ext4/super.o
>    CC [M]  fs/ext4/symlink.o
>    CC [M]  fs/ext4/sysfs.o
>    CC [M]  fs/ext4/xattr.o
>    CC [M]  fs/ext4/xattr_trusted.o
>    CC [M]  fs/ext4/xattr_user.o
>    CC [M]  fs/ext4/acl.o
>    CC [M]  fs/ext4/xattr_security.o
>    LD [M]  fs/ext4/ext4.o
>    MODPOST 1 modules
>    LD [M]  fs/ext4/ext4.ko
> + local src=fs/ext4/
> ++ uname -r
> + local dest=/lib/modules/5.6.0-rc3lbk+/kernel/fs/ext4
> + cp fs/ext4//ext4.ko /lib/modules/5.6.0-rc3lbk+/kernel/fs/ext4/
> + modprobe mbcache
> + modprobe jbd2
> + sleep 1
> + insmod fs/ext4/ext4.ko
> + sleep 1
> + load_nullb
> + local src=drivers/block/
> ++ uname -r
> + local dest=/lib/modules/5.6.0-rc3lbk+/kernel/drivers/block
> + cp drivers/block//null_blk.ko /lib/modules/5.6.0-rc3lbk+/kernel/drivers/block/
> + modprobe null_blk nr_devices=0
> + sleep 1
> + mkdir config/nullb/nullb0
> + tree config/nullb/nullb0
> config/nullb/nullb0
> ├── badblocks
> ├── blocking
> ├── blocksize
> ├── cache_size
> ├── completion_nsec
> ├── discard
> ├── home_node
> ├── hw_queue_depth
> ├── index
> ├── irqmode
> ├── mbps
> ├── memory_backed
> ├── power
> ├── queue_mode
> ├── size
> ├── submit_queues
> ├── use_per_node_hctx
> ├── zoned
> ├── zone_nr_conv
> └── zone_size
> 
> 0 directories, 20 files
> + echo 1
> + echo 512
> + echo 20480
> + echo 1
> + sleep 2
> ++ cat config/nullb/nullb0/index
> + IDX=0
> + lsblk
> + grep null0
> + sleep 1
> + mkfs.ext4 /dev/nullb0
> mke2fs 1.42.9 (28-Dec-2013)
> Filesystem label=
> OS type: Linux
> Block size=4096 (log=2)
> Fragment size=4096 (log=2)
> Stride=0 blocks, Stripe width=0 blocks
> 1310720 inodes, 5242880 blocks
> 262144 blocks (5.00%) reserved for the super user
> First data block=0
> Maximum filesystem blocks=2153775104
> 160 block groups
> 32768 blocks per group, 32768 fragments per group
> 8192 inodes per group
> Superblock backups stored on blocks:
> 	32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
> 	4096000
> 
> Allocating group tables: done
> Writing inode tables: done
> Creating journal (32768 blocks): done
> Writing superblocks and filesystem accounting information: done
> 
> + mount /dev/nullb0 /mnt/backend
> + sleep 1
> + mount
> + grep nullb
> /dev/nullb0 on /mnt/backend type ext4 (rw,relatime,seclabel)
> + dd if=/dev/zero of=/mnt/backend/data count=2621440 bs=4096
> 2621440+0 records in
> 2621440+0 records out
> 10737418240 bytes (11 GB) copied, 27.4579 s, 391 MB/s
> + load_loop
> + local src=drivers/block/
> ++ uname -r
> + local dest=/lib/modules/5.6.0-rc3lbk+/kernel/drivers/block
> + cp drivers/block//loop.ko /lib/modules/5.6.0-rc3lbk+/kernel/drivers/block/
> + insmod drivers/block/loop.ko max_loop=1
> + sleep 3
> + /root/util-linux/losetup --direct-io=off /dev/loop0 /mnt/backend/data
> + sleep 3
> + /root/util-linux/losetup
> NAME       SIZELIMIT OFFSET AUTOCLEAR RO BACK-FILE         DIO LOG-SEC
> /dev/loop0         0      0         0  0 /mnt/backend/data   0     512
> + ls -l /dev/loop0 /dev/loop-control
> brw-rw----. 1 root disk  7,   0 Mar 29 10:28 /dev/loop0
> crw-rw----. 1 root disk 10, 237 Mar 29 10:28 /dev/loop-control
> + dmesg -c
> [42963.967060] null_blk: module loaded
> [42968.419481] EXT4-fs (nullb0): mounted filesystem with ordered data mode. Opts: (null)
> [42996.928141] loop: module loaded
> + mkfs.ext4 /dev/loop0
> mke2fs 1.42.9 (28-Dec-2013)
> Discarding device blocks: done
> Filesystem label=
> OS type: Linux
> Block size=4096 (log=2)
> Fragment size=4096 (log=2)
> Stride=0 blocks, Stripe width=0 blocks
> 655360 inodes, 2621440 blocks
> 131072 blocks (5.00%) reserved for the super user
> First data block=0
> Maximum filesystem blocks=2151677952
> 80 block groups
> 32768 blocks per group, 32768 fragments per group
> 8192 inodes per group
> Superblock backups stored on blocks:
> 	32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632
> 
> Allocating group tables: done
> Writing inode tables: done
> Creating journal (32768 blocks): done
> Writing superblocks and filesystem accounting information: done
> 
> + mount /dev/loop0 /mnt/loop0
> + mount
> + grep loop0
> /dev/loop0 on /mnt/loop0 type ext4 (rw,relatime,seclabel)
> + sleep 1
> + sync
> + sync
> + sync
> + ./test
> step=61440
> step=57344
> step=53248
> step=49152
> step=45056
> step=40960
> step=36864
> step=32768
> step=28672
> step=24576
> step=20480
> step=16384
> step=12288
> step=8192
> step=4096
> step=0
> 
> real	9m34.472s
> user	0m0.062s
> sys	0m5.783s
> 

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Konstantin,

>> The corresponding exported primitive is called
>> blkdev_issue_assign_range().
>
> What exact semantics of that?

REQ_OP_ALLOCATE will be used to compel a device to allocate a block
range. What a given block contains after successful allocation is
undefined (depends on the device implementation).

For block allocation with deterministic zeroing, one must keep using
REQ_OP_WRITE_ZEROES with the NOUNMAP flag set.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Konstantin Khlebnikov]

On 02/04/2020 05.29, Martin K. Petersen wrote:
> 
> Konstantin,
> 
>>> The corresponding exported primitive is called
>>> blkdev_issue_assign_range().
>>
>> What exact semantics of that?
> 
> REQ_OP_ALLOCATE will be used to compel a device to allocate a block
> range. What a given block contains after successful allocation is
> undefined (depends on the device implementation).

Ok. Then REQ_OP_ALLOCATE should be accounted as discard rather than write.
That's decided by helper op_is_discard() which is used only by statistics.
It seems REQ_OP_SECURE_ERASE also should be accounted in this way.

> 
> For block allocation with deterministic zeroing, one must keep using
> REQ_OP_WRITE_ZEROES with the NOUNMAP flag set.
> 

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Dave Chinner]

On Sun, Mar 29, 2020 at 10:47:10AM -0700, Chaitanya Kulkarni wrote:
> Hi,
> 
> This patch-series is based on the original RFC patch series:-
> https://www.spinics.net/lists/linux-block/msg47933.html.
> 
> I've designed a rough testcase based on the information present
> in the mailing list archive for original RFC, it may need
> some corrections from the author.
> 
> If anyone is interested, test results are at the end of this patch.
> 
> Following is the original cover-letter :-
> 
> Information about continuous extent placement may be useful
> for some block devices. Say, distributed network filesystems,
> which provide block device interface, may use this information
> for better blocks placement over the nodes in their cluster,
> and for better performance. Block devices, which map a file
> on another filesystem (loop), may request the same length extent
> on underlining filesystem for less fragmentation and for batching
> allocation requests. Also, hypervisors like QEMU may use this
> information for optimization of cluster allocations.
> 
> This patchset introduces REQ_OP_ASSIGN_RANGE, which is going
> to be used for forwarding user's fallocate(0) requests into
> block device internals. It rather similar to existing
> REQ_OP_DISCARD, REQ_OP_WRITE_ZEROES, etc. The corresponding
> exported primitive is called blkdev_issue_assign_range().
> See [1/3] for the details.
> 
> Patch [2/3] teaches loop driver to handle REQ_OP_ASSIGN_RANGE
> requests by calling fallocate(0).
> 
> Patch [3/3] makes ext4 to notify a block device about fallocate(0).

Ok, so ext4 has a very limited max allocation size for an extent, so
I expect this won't cause huge latency problems. However, what
happens when we use XFS, have a 64kB block size, and fallocate() is
allocating disk space in continguous 100GB extents and passing those
down to the block device?

How does this get split by dm devices? Are raid stripes going to
dice this into separate stripe unit sized bios, so instead of single
large requests we end up with hundreds or thousands or tiny
allocation requests being issued?

I know that for the loop device, it is going to serialise all IO to
the backing file while fallocate is run on it. Hence if you have
concurrent IO running, any REQ_OP_ASSIGN_RANGE is going to cause an
significant, measurable latency hit to all those IOs in flight.

How are we expecting hardware to behave here? Is this a queued
command in the scsi/nvme/sata protocols? Or is this, for the moment,
just a special snowflake that we can't actually use in production
because the hardware just can't handle what we throw at it?

IOWs, what sort of latency issues is this operation going to cause
on real hardware? Is this going to be like discard? i.e. where we
end up not using it at all because so few devices actually handle
the massive stream of operations the filesystem will end up sending
the device(s) in the course of normal operations?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Hi Dave!

> Ok, so ext4 has a very limited max allocation size for an extent, so
> I expect this won't cause huge latency problems. However, what
> happens when we use XFS, have a 64kB block size, and fallocate() is
> allocating disk space in continguous 100GB extents and passing those
> down to the block device?

Depends on the device.

> How does this get split by dm devices? Are raid stripes going to dice
> this into separate stripe unit sized bios, so instead of single large
> requests we end up with hundreds or thousands or tiny allocation
> requests being issued?

There is nothing special about this operation. It needs to be handled
the same way as all other splits. I.e. ideally coalesced at the bottom
of the stack so we can issue larger, contiguous commands to the
hardware.

> How are we expecting hardware to behave here? Is this a queued
> command in the scsi/nvme/sata protocols? Or is this, for the moment,
> just a special snowflake that we can't actually use in production
> because the hardware just can't handle what we throw at it?

For now it's SCSI and queued. Only found in high-end thinly provisioned
storage arrays and not in your average SSD.

The performance expectation for REQ_OP_ALLOCATE is that it is faster
than a write to the same block range since the device potentially needs
to do less work. I.e. the device simply needs to decrement the free
space and mark the LBAs reserved in a map. It doesn't need to write all
the blocks to zero them. If you want zeroed blocks, use
REQ_OP_WRITE_ZEROES.

> IOWs, what sort of latency issues is this operation going to cause
> on real hardware? Is this going to be like discard? i.e. where we
> end up not using it at all because so few devices actually handle
> the massive stream of operations the filesystem will end up sending
> the device(s) in the course of normal operations?

The intended use case, from a SCSI perspective, is that on a thinly
provisioned device you can use this operation to preallocate blocks so
that future writes to the LBAs in question will not fail due to the
device being out of space. I.e. you would use this to pin down block
ranges where you can not tolerate write failures. The advantage over
writing the blocks individually is that dedup won't apply and that the
device doesn't actually have to go write all the individual blocks.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Dave Chinner]

On Thu, Apr 02, 2020 at 09:34:43PM -0400, Martin K. Petersen wrote:
> 
> Hi Dave!
> 
> > Ok, so ext4 has a very limited max allocation size for an extent, so
> > I expect this won't cause huge latency problems. However, what
> > happens when we use XFS, have a 64kB block size, and fallocate() is
> > allocating disk space in continguous 100GB extents and passing those
> > down to the block device?
> 
> Depends on the device.

Great. :(

> > How does this get split by dm devices? Are raid stripes going to dice
> > this into separate stripe unit sized bios, so instead of single large
> > requests we end up with hundreds or thousands or tiny allocation
> > requests being issued?
> 
> There is nothing special about this operation. It needs to be handled
> the same way as all other splits. I.e. ideally coalesced at the bottom
> of the stack so we can issue larger, contiguous commands to the
> hardware.
> 
> > How are we expecting hardware to behave here? Is this a queued
> > command in the scsi/nvme/sata protocols? Or is this, for the moment,
> > just a special snowflake that we can't actually use in production
> > because the hardware just can't handle what we throw at it?
> 
> For now it's SCSI and queued. Only found in high-end thinly provisioned
> storage arrays and not in your average SSD.

So it's a special snowflake :)

> The performance expectation for REQ_OP_ALLOCATE is that it is faster
> than a write to the same block range since the device potentially needs
> to do less work. I.e. the device simply needs to decrement the free
> space and mark the LBAs reserved in a map. It doesn't need to write all
> the blocks to zero them. If you want zeroed blocks, use
> REQ_OP_WRITE_ZEROES.

I suspect that the implications of wiring filesystems directly up to
this hasn't been thought through entirely....

> > IOWs, what sort of latency issues is this operation going to cause
> > on real hardware? Is this going to be like discard? i.e. where we
> > end up not using it at all because so few devices actually handle
> > the massive stream of operations the filesystem will end up sending
> > the device(s) in the course of normal operations?
> 
> The intended use case, from a SCSI perspective, is that on a thinly
> provisioned device you can use this operation to preallocate blocks so
> that future writes to the LBAs in question will not fail due to the
> device being out of space. I.e. you would use this to pin down block
> ranges where you can not tolerate write failures. The advantage over
> writing the blocks individually is that dedup won't apply and that the
> device doesn't actually have to go write all the individual blocks.

.... because when backed by thinp storage, plumbing user level
fallocate() straight through from the filesystem introduces a
trivial, user level storage DOS vector....

i.e. a user can just fallocate a bunch of files and, because the
filesystem can do that instantly, can also run the back end array
out of space almost instantly. Storage admins are going to love
this!

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Dave,

> .... because when backed by thinp storage, plumbing user level
> fallocate() straight through from the filesystem introduces a
> trivial, user level storage DOS vector....
>
> i.e. a user can just fallocate a bunch of files and, because the
> filesystem can do that instantly, can also run the back end array
> out of space almost instantly. Storage admins are going to love
> this!

In the standards space, the allocation concept was mainly aimed at
protecting filesystem internals against out-of-space conditions on
devices that dedup identical blocks and where simply zeroing the blocks
therefore is ineffective.

So far we have mainly been talking about fallocate on block devices. How
XFS decides to enforce space allocation policy and potentially leverage
this plumbing is entirely up to you.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Dave Chinner]

On Thu, Apr 02, 2020 at 08:45:30PM -0700, Martin K. Petersen wrote:
> 
> Dave,
> 
> > .... because when backed by thinp storage, plumbing user level
> > fallocate() straight through from the filesystem introduces a
> > trivial, user level storage DOS vector....
> >
> > i.e. a user can just fallocate a bunch of files and, because the
> > filesystem can do that instantly, can also run the back end array
> > out of space almost instantly. Storage admins are going to love
> > this!
> 
> In the standards space, the allocation concept was mainly aimed at
> protecting filesystem internals against out-of-space conditions on
> devices that dedup identical blocks and where simply zeroing the blocks
> therefore is ineffective.

Um, so we're supposed to use space allocation before overwriting
existing metadata in the filesystem? So that the underlying storage
can reserve space for it before we write it? Which would mean we
have to issue a space allocation before we dirty the metadata, which
means before we dirty any metadata in a transaction. Which means
we'll basically have to redesign the filesystems from the ground up,
yes?

> So far we have mainly been talking about fallocate on block devices.

You might be talking about filesystem metadata and block devices,
but this patchset ends up connecting ext4's user data fallocate() to
the block device, thereby allowing users to reserve space directly
in the underlying block device and directly exposing this issue to
userspace.

I can only go on what is presented to me in patches - this patchset
nothing to do with filesystem metadata nor preventing ENOSPC issues
with internal filesystem updates.

XFS is no different to ext4 or btrfs here - the filesystem doesn't
matter because all of them can fallocate() terabytes of space in
a second or two these days....

> How XFS decides to enforce space allocation policy and potentially
> leverage this plumbing is entirely up to you.

Do I understand this correctly? i.e. that it is the filesystem's
responsibility to prevent users from preallocating more space than
exists in an underlying storage pool that has been intentionally
hidden from the filesystem so it can be underprovisioned?

IOWs, I'm struggling to understand exactly how the "standards space"
think filesystems are supposed to be using this feature whilst also
preventing unprivileged exhaustion of a underprovisioned storage
pool they know nothing about.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Hi Dave!

>> In the standards space, the allocation concept was mainly aimed at
>> protecting filesystem internals against out-of-space conditions on
>> devices that dedup identical blocks and where simply zeroing the blocks
>> therefore is ineffective.

> Um, so we're supposed to use space allocation before overwriting
> existing metadata in the filesystem?

Not before overwriting, no. Once you have allocated an LBA it remains
allocated until you discard it.

> So that the underlying storage can reserve space for it before we
> write it? Which would mean we have to issue a space allocation before
> we dirty the metadata, which means before we dirty any metadata in a
> transaction. Which means we'll basically have to redesign the
> filesystems from the ground up, yes?

My understanding is that this facility was aimed at filesystems that do
not dynamically allocate metadata. The intent was that mkfs would
preallocate the metadata LBA ranges, not the filesystem. For filesystems
that allocate metadata dynamically, then yes, an additional step is
required if you want to pin the LBAs.

> You might be talking about filesystem metadata and block devices,
> but this patchset ends up connecting ext4's user data fallocate() to
> the block device, thereby allowing users to reserve space directly
> in the underlying block device and directly exposing this issue to
> userspace.

I missed that Chaitanya's repost of this series included the ext4 patch.
Sorry!

>> How XFS decides to enforce space allocation policy and potentially
>> leverage this plumbing is entirely up to you.
>
> Do I understand this correctly? i.e. that it is the filesystem's
> responsibility to prevent users from preallocating more space than
> exists in an underlying storage pool that has been intentionally
> hidden from the filesystem so it can be underprovisioned?

No. But as an administrative policy it is useful to prevent runaway
applications from writing a petabyte of random garbage to media. My
point was that it is up to you and the other filesystem developers to
decide how you want to leverage the low-level allocation capability and
how you want to provide it to processes. And whether CAP_SYS_ADMIN,
ulimit, or something else is the appropriate policy interface for this.

In terms of thin provisioning and space management there are various
thresholds that may be reported by the device. In past discussions there
haven't been much interest in getting these exposed. It is also unclear
to me whether it is actually beneficial to send low space warnings to
hundreds or thousands of hosts attached to an array. In many cases the
individual server admins are not even the right audience. The most
common notification mechanism is a message to the storage array admin
saying "click here to buy more disk".

If you feel there is merit in having the kernel emit the threshold
warnings you could use as a feedback mechanism, I can absolutely look
into that.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Dave Chinner]

On Wed, Apr 08, 2020 at 12:10:12AM -0400, Martin K. Petersen wrote:
> 
> Hi Dave!
> 
> >> In the standards space, the allocation concept was mainly aimed at
> >> protecting filesystem internals against out-of-space conditions on
> >> devices that dedup identical blocks and where simply zeroing the blocks
> >> therefore is ineffective.
> 
> > Um, so we're supposed to use space allocation before overwriting
> > existing metadata in the filesystem?
> 
> Not before overwriting, no. Once you have allocated an LBA it remains
> allocated until you discard it.

That is not a consistent argument. If the data has been deduped and
we overwrite, the storage array has to allocate new physical space
for an overwrite to an existing LBA. i.e. deduped data has multiple
LBAs pointing to the same physical storage. Any overwrite of an LBA
that maps to mulitply referenced physical storage requires the
storage array to allocate new physical space for that overwrite.

i.e. allocation is not determined by whether the LBA has been
written to, "pinned" or not - it's whether the act of writing to
that LBA requires the storage to allocate new space to allow the
write to proceed.

That's my point here - one particular shared data overwrite case is
being special cased by preallocation (avoiding dedupe of zero filled
data) to prevent ENOSPC, ignoring all the other cases where we
overwrite shared non-zero data and will also require new physical
space for the new data. In all those cases, the storage has to take
the same action - allocation on overwrite - and so all of them are
susceptible to ENOSPC.

> > So that the underlying storage can reserve space for it before we
> > write it? Which would mean we have to issue a space allocation before
> > we dirty the metadata, which means before we dirty any metadata in a
> > transaction. Which means we'll basically have to redesign the
> > filesystems from the ground up, yes?
> 
> My understanding is that this facility was aimed at filesystems that do
> not dynamically allocate metadata. The intent was that mkfs would
> preallocate the metadata LBA ranges, not the filesystem. For filesystems
> that allocate metadata dynamically, then yes, an additional step is
> required if you want to pin the LBAs.

Ok, so you are confirming what I thought: it's almost completely
useless to us.

i.e. this requires issuing IO to "reserve" space whilst preserving
data before every metadata object goes from clean to dirty in
memory.  But the problem with that is we don't know how much
metadata we are going to dirty in any specific operation. Worse is
that we don't know exactly *what* metadata we will modify until we
walk structures and do lookups, which often happen after we've
dirtied other structures. An ENOSPC from a space reservation at that
point is fatal to the filesystem anyway, so there's no point in even
trying to do this.  Like I said, functionality like this cannot be
retrofitted to existing filesysetms.

IOWs, this is pretty much useless functionality for the filesystem
layer, and if the only use is for some mythical filesystem with
completely static metadata then the standards space really jumped
the shark on this one....

> > You might be talking about filesystem metadata and block devices,
> > but this patchset ends up connecting ext4's user data fallocate() to
> > the block device, thereby allowing users to reserve space directly
> > in the underlying block device and directly exposing this issue to
> > userspace.
> 
> I missed that Chaitanya's repost of this series included the ext4 patch.
> Sorry!
> 
> >> How XFS decides to enforce space allocation policy and potentially
> >> leverage this plumbing is entirely up to you.
> >
> > Do I understand this correctly? i.e. that it is the filesystem's
> > responsibility to prevent users from preallocating more space than
> > exists in an underlying storage pool that has been intentionally
> > hidden from the filesystem so it can be underprovisioned?
> 
> No. But as an administrative policy it is useful to prevent runaway
> applications from writing a petabyte of random garbage to media. My
> point was that it is up to you and the other filesystem developers to
> decide how you want to leverage the low-level allocation capability and
> how you want to provide it to processes. And whether CAP_SYS_ADMIN,
> ulimit, or something else is the appropriate policy interface for this.

My cynical translation: the storage standards space haven't given
any thought to how it can be used and/or administered in the real
world. Pass the buck - let the filesystem people work that out.

What I'm hearing is that this wasn't designed for typical filesystem
use, it wasn't designed for typical user application use, and how to
prevent abuse wasn't thought about at all.

That sounds like a big fat NACK to me....

> In terms of thin provisioning and space management there are various
> thresholds that may be reported by the device. In past discussions there
> haven't been much interest in getting these exposed. It is also unclear
> to me whether it is actually beneficial to send low space warnings to
> hundreds or thousands of hosts attached to an array. In many cases the
> individual server admins are not even the right audience. The most
> common notification mechanism is a message to the storage array admin
> saying "click here to buy more disk".

Notifications are not relevant to preallocation functionality at all.

-Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Dave,

>> Not before overwriting, no. Once you have allocated an LBA it remains
>> allocated until you discard it.

> Ok, so you are confirming what I thought: it's almost completely
> useless to us.
>
> i.e. this requires issuing IO to "reserve" space whilst preserving
> data before every metadata object goes from clean to dirty in memory.

You can only reserve the space prior to writing a block for the first
time. Once an LBA has been written ("Mapped" in the SCSI state machine),
it remains allocated until it is explicitly deallocated (via a
discard/Unmap operation).

This part of the SCSI spec was written eons ago under the assumption
that when a physical resource backing a given LBA had been established,
you could write the block over and over without having to allocate new
space.

This used to be true, but obviously the introduction of de-duplication
blew a major hole in that. I have been perusing the spec over and over
trying to understand how block provisioning state transitions are
defined when dedup is in the picture. However, much is left unexplained.

As a result, I reached out to various folks. Including the people who
worked on this feature in the standards way back. And the response that
I get from them is that allocation operation got irreparably broken when
support for de-duplication was added to the spec. Nobody attempted to
fix the state transitions since most vendors only cared about
deallocation. Consequently specifying the exact behavior of the
allocation operation in the context of dedup fell by the wayside.

The recommendation I got was that we should not rely on this feature
despite it being advertised as supported by the storage. I looked at
whether it was feasible to support it on non-dedup devices only, but it
does not look like it's worthwhile to pursue. And as a result there is
no need for block layer allocation operation to have parity with
SCSI. Although we may want to keep NVMe in mind when defining the
semantics.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Martin K. Petersen]

Chaitanya,

> This patchset introduces REQ_OP_ASSIGN_RANGE, which is going
> to be used for forwarding user's fallocate(0) requests into
> block device internals.

s/assign_range/allocate/g

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 0/4] block: Add support for REQ_OP_ASSIGN_RANGE

[Chaitanya Kulkarni]

On 3/31/20 7:31 PM, Martin K. Petersen wrote:
> Chaitanya,
>
>> This patchset introduces REQ_OP_ASSIGN_RANGE, which is going
>> to be used for forwarding user's fallocate(0) requests into
>> block device internals.
> s/assign_range/allocate/g
>
Okay, will send out the V2.

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Martin K. Petersen]

Christoph,

> The whole series looks good to me:
>
> Reviewed-by: Christoph Hellwig <hch@lst.de>
>
> I hope we can get this in 5.8 to help with the btrfs in 5.9.

Yep, I think this looks good.

I suspect this series going to clash with my sd revalidate surgery. I
may have to stick that in a postmerge branch based on Jens' tree.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Christoph Hellwig]

On Tue, May 12, 2020 at 09:01:18AM -0700, Martin K. Petersen wrote:
> I suspect this series going to clash with my sd revalidate surgery. I
> may have to stick that in a postmerge branch based on Jens' tree.

Where is that series?  I don't remember any changes in that area.

> 
> -- 
> Martin K. Petersen	Oracle Linux Engineering
---end quoted text---

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Martin K. Petersen]

Christoph,

> Where is that series?  I don't remember any changes in that area.

Haven't posted it yet. Still working on a few patches that address
validating reported values for devices that change parameters in flight.

The first part of the series is here:

https://git.kernel.org/pub/scm/linux/kernel/git/mkp/linux.git/log/?h=5.8/sd-revalidate

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Johannes Thumshirn]

On 12/05/2020 18:15, Martin K. Petersen wrote:
> 
> Christoph,
> 
>> Where is that series?  I don't remember any changes in that area.
> 
> Haven't posted it yet. Still working on a few patches that address
> validating reported values for devices that change parameters in flight.
> 
> The first part of the series is here:
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/mkp/linux.git/log/?h=5.8/sd-revalidate
> 

Just did a quick skim in the gitweb and I can't see anything that will clash. So I think we're good.

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Martin K. Petersen]

Johannes,

> Just did a quick skim in the gitweb and I can't see anything that will
> clash. So I think we're good.

The most intrusive pieces are impending. But no worries! Staging the
merge is my problem. Your series is good to go as far as I'm concerned.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 2/2] block: improve discard bio alignment in __blkdev_issue_discard()

[Martin K. Petersen]

Hi Coly!

> This patch improves discard bio split for address and size alignment
> in __blkdev_issue_discard(). The aligned discard bio may help
> underlying device controller to perform better discard and internal
> garbage collection, and avoid unnecessary internal fragment.

If the aim is to guarantee that all discard requests, except for head
and tail, are aligned multiples of the discard_granularity, you also
need to take the discard_alignment queue limit and the partition offset
into consideration.

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH 2/2] block: improve discard bio alignment in __blkdev_issue_discard()

[Coly Li]

On 2020/7/14 00:47, Martin K. Petersen wrote:
> 
> Hi Coly!
> 
>> This patch improves discard bio split for address and size alignment
>> in __blkdev_issue_discard(). The aligned discard bio may help
>> underlying device controller to perform better discard and internal
>> garbage collection, and avoid unnecessary internal fragment.
> 

Hi Martin,

> If the aim is to guarantee that all discard requests, except for head
> and tail, are aligned multiples of the discard_granularity, you also
> need to take the discard_alignment queue limit and the partition offset
> into consideration.
> 

The discard_granularity was considered and my though is,
discard_alignment normally is multiples of discard granularity, if the
discard bio bi_sector and bi_size are aligned to discard granularity,
when underlying driver splits its discard bio by discard_alignment, the
split bio bi_sector and bi_size can still be aligned to discard
granularity. Another reason I don't handle discard_alignment alignment
in __blkdev_issue_discard() is performance. Handling discard_alignment
bio split in driver's split loop may call blk_next_bio() much less in
__blkdev_issue_discard(), and be more friendly for memory and cache
footprint.

For the partition offset, my original idea was to suggest the partition
or dm target starts on offset 2048 sectors. But your opinion sound
better, if considering partition or target offset maybe users
misconfigured the partition offset may also gain the benefit of discard
alignment.

Let me try to improve a v3 patch to handle the partition offset too.

Thanks for the cool idea :-)

Coly Li

[PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Johannes Thumshirn]

The upcoming NVMe ZNS Specification will define a new type of write
command for zoned block devices, zone append.

When when writing to a zoned block device using zone append, the start
sector of the write is pointing at the start LBA of the zone to write to.
Upon completion the block device will respond with the position the data
has been placed in the zone. This from a high level perspective can be
seen like a file system's block allocator, where the user writes to a
file and the file-system takes care of the data placement on the device.

In order to fully exploit the new zone append command in file-systems and
other interfaces above the block layer, we choose to emulate zone append
in SCSI and null_blk. This way we can have a single write path for both
file-systems and other interfaces above the block-layer, like io_uring on
zoned block devices, without having to care too much about the underlying
characteristics of the device itself.

The emulation works by providing a cache of each zone's write pointer, so
zone append issued to the disk can be translated to a write with a
starting LBA of the write pointer. This LBA is used as input zone number
for the write pointer lookup in the zone write pointer offset cache and
the cached offset is then added to the LBA to get the actual position to
write the data. In SCSI we then turn the REQ_OP_ZONE_APPEND request into a
WRITE(16) command. Upon successful completion of the WRITE(16), the cache
will be updated to the new write pointer location and the written sector
will be noted in the request. On error the cache entry will be marked as
invalid and on the next write an update of the write pointer will be
scheduled, before issuing the actual write.

In order to reduce memory consumption, the only cached item is the offset
of the write pointer from the start of the zone, everything else can be
calculated. On an example drive with 52156 zones, the additional memory
consumption of the cache is thus 52156 * 4 = 208624 Bytes or 51 4k Byte
pages. The performance impact is neglectable for a spinning drive.

For null_blk the emulation is way simpler, as null_blk's zoned block
device emulation support already caches the write pointer position, so we
only need to report the position back to the upper layers. Additional
caching is not needed here.

Furthermore we have converted zonefs to run use ZONE_APPEND for synchronous
direct I/Os. Asynchronous I/O still uses the normal path via iomap.

Performance testing with zonefs sync writes on a 14 TB SMR drive and nullblk
shows good results. On the SMR drive we're not regressing (the performance
improvement is within noise), on nullblk we could drastically improve specific
workloads:

* nullblk:

Single Thread Multiple Zones
				kIOPS	MiB/s	MB/s	% delta
mq-deadline REQ_OP_WRITE	10.1	631	662
mq-deadline REQ_OP_ZONE_APPEND	13.2	828	868	+31.12
none REQ_OP_ZONE_APPEND		15.6	978	1026	+54.98


Multiple Threads Multiple Zones
				kIOPS	MiB/s	MB/s	% delta
mq-deadline REQ_OP_WRITE	10.2	640	671
mq-deadline REQ_OP_ZONE_APPEND	10.4	650	681	+1.49
none REQ_OP_ZONE_APPEND		16.9	1058	1109	+65.28

* 14 TB SMR drive

Single Thread Multiple Zones
				IOPS	MiB/s	MB/s	% delta
mq-deadline REQ_OP_WRITE	797	49.9	52.3
mq-deadline REQ_OP_ZONE_APPEND	806	50.4	52.9	+1.15

Multiple Threads Multiple Zones
				kIOPS	MiB/s	MB/s	% delta
mq-deadline REQ_OP_WRITE	745	46.6	48.9
mq-deadline REQ_OP_ZONE_APPEND	768	48	50.3	+2.86

The %-delta is against the baseline of REQ_OP_WRITE using mq-deadline as I/O
scheduler.

The series is based on Jens' for-5.8/block branch with HEAD:
ae979182ebb3 ("bdi: fix up for "remove the name field in struct backing_dev_info"")

As Christoph asked for a branch I pushed it to a git repo at:
git://git.kernel.org/pub/scm/linux/kernel/git/jth/linux.git zone-append.v11
https://git.kernel.org/pub/scm/linux/kernel/git/jth/linux.git/log/?h=zone-append.v11

Changes to v10:
- Added Reviews from Hannes
- Added Performance Numbers to cover letter

Changes to v9:
- Renamed zone_wp_ofst to zone_wp_offset (Hannes/Martin)
- Colledted Reviews
- Dropped already merged patches

Changes to v8:
- Added kerneldoc for bio_add_hw_page (Hannes)
- Simplified calculation of zone-boundary cross checking (Bart)
- Added safety nets for max_appen_sectors setting
- Added Reviews from Hannes
- Added Damien's Ack on the zonefs change

Changes to v7:
- Rebased on Jens' for-5.8/block
- Fixed up stray whitespace change (Bart)
- Added Reviews from Bart and Christoph

Changes to v6:
- Added Daniel's Reviewed-by's
- Addressed Christoph's comment on whitespace changes in 4/11
- Renamed driver_cb in 6/11
- Fixed lines over 80 characters in 8/11
- Damien simplified sd_zbc_revalidate_zones() in 8/11

Changes to v5:
- Added patch to fix the memleak on failed scsi command setup
- Added prep patch from Christoph for bio_add_hw_page
- Added Christoph's suggestions for adding append pages to bios
- Fixed compile warning with !CONFIG_BLK_DEV_ZONED
- Damien re-worked revalidate zone
- Added Christoph's suggestions for rescanning write pointers to update cache

Changes to v4:
- Added page merging for zone-append bios (Christoph)
- Removed different locking schmes for zone management operations (Christoph)
- Changed wp_ofst assignment from blk_revalidate_zones (Christoph)
- Smaller nitpicks (Christoph)
- Documented my changes to Keith's patch so it's clear where I messed up so he
  doesn't get blamed
- Added Damien as a Co-developer to the sd emulation patch as he wrote as much
  code for it as I did (if not more)

Changes since v3:
- Remove impact of zone-append from bio_full() and bio_add_page()
  fast-path (Christoph)
- All of the zone write pointer offset caching is handled in SCSI now
  (Christoph) 
- Drop null_blk pathces that damien sent separately (Christoph)
- Use EXPORT_SYMBOL_GPL for new exports (Christoph)	

Changes since v2:
- Remove iomap implementation and directly issue zone-appends from within
  zonefs (Christoph)
- Drop already merged patch
- Rebase onto new for-next branch

Changes since v1:
- Too much to mention, treat as a completely new series.


Christoph Hellwig (1):
  block: rename __bio_add_pc_page to bio_add_hw_page

Damien Le Moal (2):
  block: Modify revalidate zones
  null_blk: Support REQ_OP_ZONE_APPEND

Johannes Thumshirn (6):
  block: provide fallbacks for blk_queue_zone_is_seq and
    blk_queue_zone_no
  block: introduce blk_req_zone_write_trylock
  scsi: sd_zbc: factor out sanity checks for zoned commands
  scsi: sd_zbc: emulate ZONE_APPEND commands
  block: export bio_release_pages and bio_iov_iter_get_pages
  zonefs: use REQ_OP_ZONE_APPEND for sync DIO

Keith Busch (1):
  block: Introduce REQ_OP_ZONE_APPEND

 block/bio.c                    | 129 ++++++++---
 block/blk-core.c               |  52 +++++
 block/blk-map.c                |   5 +-
 block/blk-mq.c                 |  27 +++
 block/blk-settings.c           |  31 +++
 block/blk-sysfs.c              |  13 ++
 block/blk-zoned.c              |  23 +-
 block/blk.h                    |   4 +-
 drivers/block/null_blk_zoned.c |  37 ++-
 drivers/scsi/scsi_lib.c        |   1 +
 drivers/scsi/sd.c              |  16 +-
 drivers/scsi/sd.h              |  43 +++-
 drivers/scsi/sd_zbc.c          | 399 ++++++++++++++++++++++++++++++---
 fs/zonefs/super.c              |  80 ++++++-
 include/linux/blk_types.h      |  14 ++
 include/linux/blkdev.h         |  25 ++-
 16 files changed, 807 insertions(+), 92 deletions(-)

-- 
2.24.1

[PATCH v11 01/10] block: provide fallbacks for blk_queue_zone_is_seq and blk_queue_zone_no

[Johannes Thumshirn]

blk_queue_zone_is_seq() and blk_queue_zone_no() have not been called with
CONFIG_BLK_DEV_ZONED disabled until now.

The introduction of REQ_OP_ZONE_APPEND will change this, so we need to
provide noop fallbacks for the !CONFIG_BLK_DEV_ZONED case.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 include/linux/blkdev.h | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index f00bd4042295..91c6e413bf6b 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -721,6 +721,16 @@ static inline unsigned int blk_queue_nr_zones(struct request_queue *q)
 {
 	return 0;
 }
+static inline bool blk_queue_zone_is_seq(struct request_queue *q,
+					 sector_t sector)
+{
+	return false;
+}
+static inline unsigned int blk_queue_zone_no(struct request_queue *q,
+					     sector_t sector)
+{
+	return 0;
+}
 #endif /* CONFIG_BLK_DEV_ZONED */
 
 static inline bool rq_is_sync(struct request *rq)
-- 
2.24.1

[PATCH v11 02/10] block: rename __bio_add_pc_page to bio_add_hw_page

[Johannes Thumshirn]

From: Christoph Hellwig <hch@lst.de>

Rename __bio_add_pc_page() to bio_add_hw_page() and explicitly pass in a
max_sectors argument.

This max_sectors argument can be used to specify constraints from the
hardware.

Signed-off-by: Christoph Hellwig <hch@lst.de>
[ jth: rebased and made public for blk-map.c ]
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 block/bio.c     | 65 ++++++++++++++++++++++++++++++-------------------
 block/blk-map.c |  5 ++--
 block/blk.h     |  4 +--
 3 files changed, 45 insertions(+), 29 deletions(-)

diff --git a/block/bio.c b/block/bio.c
index 21cbaa6a1c20..aad0a6dad4f9 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -748,9 +748,14 @@ static inline bool page_is_mergeable(const struct bio_vec *bv,
 	return true;
 }
 
-static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio,
-		struct page *page, unsigned len, unsigned offset,
-		bool *same_page)
+/*
+ * Try to merge a page into a segment, while obeying the hardware segment
+ * size limit.  This is not for normal read/write bios, but for passthrough
+ * or Zone Append operations that we can't split.
+ */
+static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
+				 struct page *page, unsigned len,
+				 unsigned offset, bool *same_page)
 {
 	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
 	unsigned long mask = queue_segment_boundary(q);
@@ -765,38 +770,32 @@ static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio,
 }
 
 /**
- *	__bio_add_pc_page	- attempt to add page to passthrough bio
- *	@q: the target queue
- *	@bio: destination bio
- *	@page: page to add
- *	@len: vec entry length
- *	@offset: vec entry offset
- *	@same_page: return if the merge happen inside the same page
- *
- *	Attempt to add a page to the bio_vec maplist. This can fail for a
- *	number of reasons, such as the bio being full or target block device
- *	limitations. The target block device must allow bio's up to PAGE_SIZE,
- *	so it is always possible to add a single page to an empty bio.
+ * bio_add_hw_page - attempt to add a page to a bio with hw constraints
+ * @q: the target queue
+ * @bio: destination bio
+ * @page: page to add
+ * @len: vec entry length
+ * @offset: vec entry offset
+ * @max_sectors: maximum number of sectors that can be added
+ * @same_page: return if the segment has been merged inside the same page
  *
- *	This should only be used by passthrough bios.
+ * Add a page to a bio while respecting the hardware max_sectors, max_segment
+ * and gap limitations.
  */
-int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
+int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 		struct page *page, unsigned int len, unsigned int offset,
-		bool *same_page)
+		unsigned int max_sectors, bool *same_page)
 {
 	struct bio_vec *bvec;
 
-	/*
-	 * cloned bio must not modify vec list
-	 */
-	if (unlikely(bio_flagged(bio, BIO_CLONED)))
+	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
 		return 0;
 
-	if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
+	if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
 		return 0;
 
 	if (bio->bi_vcnt > 0) {
-		if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page))
+		if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
 			return len;
 
 		/*
@@ -823,11 +822,27 @@ int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
 	return len;
 }
 
+/**
+ * bio_add_pc_page	- attempt to add page to passthrough bio
+ * @q: the target queue
+ * @bio: destination bio
+ * @page: page to add
+ * @len: vec entry length
+ * @offset: vec entry offset
+ *
+ * Attempt to add a page to the bio_vec maplist. This can fail for a
+ * number of reasons, such as the bio being full or target block device
+ * limitations. The target block device must allow bio's up to PAGE_SIZE,
+ * so it is always possible to add a single page to an empty bio.
+ *
+ * This should only be used by passthrough bios.
+ */
 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
 		struct page *page, unsigned int len, unsigned int offset)
 {
 	bool same_page = false;
-	return __bio_add_pc_page(q, bio, page, len, offset, &same_page);
+	return bio_add_hw_page(q, bio, page, len, offset,
+			queue_max_hw_sectors(q), &same_page);
 }
 EXPORT_SYMBOL(bio_add_pc_page);
 
diff --git a/block/blk-map.c b/block/blk-map.c
index b6fa343fea9f..e3e4ac48db45 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -257,6 +257,7 @@ static struct bio *bio_copy_user_iov(struct request_queue *q,
 static struct bio *bio_map_user_iov(struct request_queue *q,
 		struct iov_iter *iter, gfp_t gfp_mask)
 {
+	unsigned int max_sectors = queue_max_hw_sectors(q);
 	int j;
 	struct bio *bio;
 	int ret;
@@ -294,8 +295,8 @@ static struct bio *bio_map_user_iov(struct request_queue *q,
 				if (n > bytes)
 					n = bytes;
 
-				if (!__bio_add_pc_page(q, bio, page, n, offs,
-						&same_page)) {
+				if (!bio_add_hw_page(q, bio, page, n, offs,
+						     max_sectors, &same_page)) {
 					if (same_page)
 						put_page(page);
 					break;
diff --git a/block/blk.h b/block/blk.h
index 73bd3b1c6938..1ae3279df712 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -453,8 +453,8 @@ static inline void part_nr_sects_write(struct hd_struct *part, sector_t size)
 
 struct request_queue *__blk_alloc_queue(int node_id);
 
-int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
+int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 		struct page *page, unsigned int len, unsigned int offset,
-		bool *same_page);
+		unsigned int max_sectors, bool *same_page);
 
 #endif /* BLK_INTERNAL_H */
-- 
2.24.1

[PATCH v11 03/10] block: Introduce REQ_OP_ZONE_APPEND

[Johannes Thumshirn]

From: Keith Busch <kbusch@kernel.org>

Define REQ_OP_ZONE_APPEND to append-write sectors to a zone of a zoned
block device. This is a no-merge write operation.

A zone append write BIO must:
* Target a zoned block device
* Have a sector position indicating the start sector of the target zone
* The target zone must be a sequential write zone
* The BIO must not cross a zone boundary
* The BIO size must not be split to ensure that a single range of LBAs
  is written with a single command.

Implement these checks in generic_make_request_checks() using the
helper function blk_check_zone_append(). To avoid write append BIO
splitting, introduce the new max_zone_append_sectors queue limit
attribute and ensure that a BIO size is always lower than this limit.
Export this new limit through sysfs and check these limits in bio_full().

Also when a LLDD can't dispatch a request to a specific zone, it
will return BLK_STS_ZONE_RESOURCE indicating this request needs to
be delayed, e.g.  because the zone it will be dispatched to is still
write-locked. If this happens set the request aside in a local list
to continue trying dispatching requests such as READ requests or a
WRITE/ZONE_APPEND requests targetting other zones. This way we can
still keep a high queue depth without starving other requests even if
one request can't be served due to zone write-locking.

Finally, make sure that the bio sector position indicates the actual
write position as indicated by the device on completion.

Signed-off-by: Keith Busch <kbusch@kernel.org>
[ jth: added zone-append specific add_page and merge_page helpers ]
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 block/bio.c               | 62 ++++++++++++++++++++++++++++++++++++---
 block/blk-core.c          | 52 ++++++++++++++++++++++++++++++++
 block/blk-mq.c            | 27 +++++++++++++++++
 block/blk-settings.c      | 31 ++++++++++++++++++++
 block/blk-sysfs.c         | 13 ++++++++
 drivers/scsi/scsi_lib.c   |  1 +
 include/linux/blk_types.h | 14 +++++++++
 include/linux/blkdev.h    | 11 +++++++
 8 files changed, 207 insertions(+), 4 deletions(-)

diff --git a/block/bio.c b/block/bio.c
index aad0a6dad4f9..3aa3c4ce2e5e 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -1025,6 +1025,50 @@ static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
 	return 0;
 }
 
+static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
+{
+	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
+	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
+	struct request_queue *q = bio->bi_disk->queue;
+	unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
+	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
+	struct page **pages = (struct page **)bv;
+	ssize_t size, left;
+	unsigned len, i;
+	size_t offset;
+
+	if (WARN_ON_ONCE(!max_append_sectors))
+		return 0;
+
+	/*
+	 * Move page array up in the allocated memory for the bio vecs as far as
+	 * possible so that we can start filling biovecs from the beginning
+	 * without overwriting the temporary page array.
+	 */
+	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
+	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
+
+	size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
+	if (unlikely(size <= 0))
+		return size ? size : -EFAULT;
+
+	for (left = size, i = 0; left > 0; left -= len, i++) {
+		struct page *page = pages[i];
+		bool same_page = false;
+
+		len = min_t(size_t, PAGE_SIZE - offset, left);
+		if (bio_add_hw_page(q, bio, page, len, offset,
+				max_append_sectors, &same_page) != len)
+			return -EINVAL;
+		if (same_page)
+			put_page(page);
+		offset = 0;
+	}
+
+	iov_iter_advance(iter, size);
+	return 0;
+}
+
 /**
  * bio_iov_iter_get_pages - add user or kernel pages to a bio
  * @bio: bio to add pages to
@@ -1054,10 +1098,16 @@ int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
 		return -EINVAL;
 
 	do {
-		if (is_bvec)
-			ret = __bio_iov_bvec_add_pages(bio, iter);
-		else
-			ret = __bio_iov_iter_get_pages(bio, iter);
+		if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
+			if (WARN_ON_ONCE(is_bvec))
+				return -EINVAL;
+			ret = __bio_iov_append_get_pages(bio, iter);
+		} else {
+			if (is_bvec)
+				ret = __bio_iov_bvec_add_pages(bio, iter);
+			else
+				ret = __bio_iov_iter_get_pages(bio, iter);
+		}
 	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
 
 	if (is_bvec)
@@ -1460,6 +1510,10 @@ struct bio *bio_split(struct bio *bio, int sectors,
 	BUG_ON(sectors <= 0);
 	BUG_ON(sectors >= bio_sectors(bio));
 
+	/* Zone append commands cannot be split */
+	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
+		return NULL;
+
 	split = bio_clone_fast(bio, gfp, bs);
 	if (!split)
 		return NULL;
diff --git a/block/blk-core.c b/block/blk-core.c
index 538cbc725620..0f95fdfd3827 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -135,6 +135,7 @@ static const char *const blk_op_name[] = {
 	REQ_OP_NAME(ZONE_OPEN),
 	REQ_OP_NAME(ZONE_CLOSE),
 	REQ_OP_NAME(ZONE_FINISH),
+	REQ_OP_NAME(ZONE_APPEND),
 	REQ_OP_NAME(WRITE_SAME),
 	REQ_OP_NAME(WRITE_ZEROES),
 	REQ_OP_NAME(SCSI_IN),
@@ -240,6 +241,17 @@ static void req_bio_endio(struct request *rq, struct bio *bio,
 
 	bio_advance(bio, nbytes);
 
+	if (req_op(rq) == REQ_OP_ZONE_APPEND && error == BLK_STS_OK) {
+		/*
+		 * Partial zone append completions cannot be supported as the
+		 * BIO fragments may end up not being written sequentially.
+		 */
+		if (bio->bi_iter.bi_size)
+			bio->bi_status = BLK_STS_IOERR;
+		else
+			bio->bi_iter.bi_sector = rq->__sector;
+	}
+
 	/* don't actually finish bio if it's part of flush sequence */
 	if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
 		bio_endio(bio);
@@ -887,6 +899,41 @@ static inline int blk_partition_remap(struct bio *bio)
 	return ret;
 }
 
+/*
+ * Check write append to a zoned block device.
+ */
+static inline blk_status_t blk_check_zone_append(struct request_queue *q,
+						 struct bio *bio)
+{
+	sector_t pos = bio->bi_iter.bi_sector;
+	int nr_sectors = bio_sectors(bio);
+
+	/* Only applicable to zoned block devices */
+	if (!blk_queue_is_zoned(q))
+		return BLK_STS_NOTSUPP;
+
+	/* The bio sector must point to the start of a sequential zone */
+	if (pos & (blk_queue_zone_sectors(q) - 1) ||
+	    !blk_queue_zone_is_seq(q, pos))
+		return BLK_STS_IOERR;
+
+	/*
+	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
+	 * split and could result in non-contiguous sectors being written in
+	 * different zones.
+	 */
+	if (nr_sectors > q->limits.chunk_sectors)
+		return BLK_STS_IOERR;
+
+	/* Make sure the BIO is small enough and will not get split */
+	if (nr_sectors > q->limits.max_zone_append_sectors)
+		return BLK_STS_IOERR;
+
+	bio->bi_opf |= REQ_NOMERGE;
+
+	return BLK_STS_OK;
+}
+
 static noinline_for_stack bool
 generic_make_request_checks(struct bio *bio)
 {
@@ -959,6 +1006,11 @@ generic_make_request_checks(struct bio *bio)
 		if (!q->limits.max_write_same_sectors)
 			goto not_supported;
 		break;
+	case REQ_OP_ZONE_APPEND:
+		status = blk_check_zone_append(q, bio);
+		if (status != BLK_STS_OK)
+			goto end_io;
+		break;
 	case REQ_OP_ZONE_RESET:
 	case REQ_OP_ZONE_OPEN:
 	case REQ_OP_ZONE_CLOSE:
diff --git a/block/blk-mq.c b/block/blk-mq.c
index bc34d6b572b6..d743f0af5fb8 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1183,6 +1183,19 @@ static void blk_mq_handle_dev_resource(struct request *rq,
 	__blk_mq_requeue_request(rq);
 }
 
+static void blk_mq_handle_zone_resource(struct request *rq,
+					struct list_head *zone_list)
+{
+	/*
+	 * If we end up here it is because we cannot dispatch a request to a
+	 * specific zone due to LLD level zone-write locking or other zone
+	 * related resource not being available. In this case, set the request
+	 * aside in zone_list for retrying it later.
+	 */
+	list_add(&rq->queuelist, zone_list);
+	__blk_mq_requeue_request(rq);
+}
+
 /*
  * Returns true if we did some work AND can potentially do more.
  */
@@ -1195,6 +1208,7 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,
 	int errors, queued;
 	blk_status_t ret = BLK_STS_OK;
 	bool no_budget_avail = false;
+	LIST_HEAD(zone_list);
 
 	if (list_empty(list))
 		return false;
@@ -1256,6 +1270,16 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,
 		if (ret == BLK_STS_RESOURCE || ret == BLK_STS_DEV_RESOURCE) {
 			blk_mq_handle_dev_resource(rq, list);
 			break;
+		} else if (ret == BLK_STS_ZONE_RESOURCE) {
+			/*
+			 * Move the request to zone_list and keep going through
+			 * the dispatch list to find more requests the drive can
+			 * accept.
+			 */
+			blk_mq_handle_zone_resource(rq, &zone_list);
+			if (list_empty(list))
+				break;
+			continue;
 		}
 
 		if (unlikely(ret != BLK_STS_OK)) {
@@ -1267,6 +1291,9 @@ bool blk_mq_dispatch_rq_list(struct request_queue *q, struct list_head *list,
 		queued++;
 	} while (!list_empty(list));
 
+	if (!list_empty(&zone_list))
+		list_splice_tail_init(&zone_list, list);
+
 	hctx->dispatched[queued_to_index(queued)]++;
 
 	/*
diff --git a/block/blk-settings.c b/block/blk-settings.c
index 2ab1967b9716..9a2c23cd9700 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -48,6 +48,7 @@ void blk_set_default_limits(struct queue_limits *lim)
 	lim->chunk_sectors = 0;
 	lim->max_write_same_sectors = 0;
 	lim->max_write_zeroes_sectors = 0;
+	lim->max_zone_append_sectors = 0;
 	lim->max_discard_sectors = 0;
 	lim->max_hw_discard_sectors = 0;
 	lim->discard_granularity = 0;
@@ -83,6 +84,7 @@ void blk_set_stacking_limits(struct queue_limits *lim)
 	lim->max_dev_sectors = UINT_MAX;
 	lim->max_write_same_sectors = UINT_MAX;
 	lim->max_write_zeroes_sectors = UINT_MAX;
+	lim->max_zone_append_sectors = UINT_MAX;
 }
 EXPORT_SYMBOL(blk_set_stacking_limits);
 
@@ -221,6 +223,33 @@ void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
 }
 EXPORT_SYMBOL(blk_queue_max_write_zeroes_sectors);
 
+/**
+ * blk_queue_max_zone_append_sectors - set max sectors for a single zone append
+ * @q:  the request queue for the device
+ * @max_zone_append_sectors: maximum number of sectors to write per command
+ **/
+void blk_queue_max_zone_append_sectors(struct request_queue *q,
+		unsigned int max_zone_append_sectors)
+{
+	unsigned int max_sectors;
+
+	if (WARN_ON(!blk_queue_is_zoned(q)))
+		return;
+
+	max_sectors = min(q->limits.max_hw_sectors, max_zone_append_sectors);
+	max_sectors = min(q->limits.chunk_sectors, max_sectors);
+
+	/*
+	 * Signal eventual driver bugs resulting in the max_zone_append sectors limit
+	 * being 0 due to a 0 argument, the chunk_sectors limit (zone size) not set,
+	 * or the max_hw_sectors limit not set.
+	 */
+	WARN_ON(!max_sectors);
+
+	q->limits.max_zone_append_sectors = max_sectors;
+}
+EXPORT_SYMBOL_GPL(blk_queue_max_zone_append_sectors);
+
 /**
  * blk_queue_max_segments - set max hw segments for a request for this queue
  * @q:  the request queue for the device
@@ -470,6 +499,8 @@ int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
 					b->max_write_same_sectors);
 	t->max_write_zeroes_sectors = min(t->max_write_zeroes_sectors,
 					b->max_write_zeroes_sectors);
+	t->max_zone_append_sectors = min(t->max_zone_append_sectors,
+					b->max_zone_append_sectors);
 	t->bounce_pfn = min_not_zero(t->bounce_pfn, b->bounce_pfn);
 
 	t->seg_boundary_mask = min_not_zero(t->seg_boundary_mask,
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index fca9b158f4a0..02643e149d5e 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -218,6 +218,13 @@ static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 		(unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 }
 
+static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
+{
+	unsigned long long max_sectors = q->limits.max_zone_append_sectors;
+
+	return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
+}
+
 static ssize_t
 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 {
@@ -639,6 +646,11 @@ static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 	.show = queue_write_zeroes_max_show,
 };
 
+static struct queue_sysfs_entry queue_zone_append_max_entry = {
+	.attr = {.name = "zone_append_max_bytes", .mode = 0444 },
+	.show = queue_zone_append_max_show,
+};
+
 static struct queue_sysfs_entry queue_nonrot_entry = {
 	.attr = {.name = "rotational", .mode = 0644 },
 	.show = queue_show_nonrot,
@@ -749,6 +761,7 @@ static struct attribute *queue_attrs[] = {
 	&queue_discard_zeroes_data_entry.attr,
 	&queue_write_same_max_entry.attr,
 	&queue_write_zeroes_max_entry.attr,
+	&queue_zone_append_max_entry.attr,
 	&queue_nonrot_entry.attr,
 	&queue_zoned_entry.attr,
 	&queue_nr_zones_entry.attr,
diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index f0cb26b3da6a..af00e4a3f006 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1712,6 +1712,7 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
 	case BLK_STS_OK:
 		break;
 	case BLK_STS_RESOURCE:
+	case BLK_STS_ZONE_RESOURCE:
 		if (atomic_read(&sdev->device_busy) ||
 		    scsi_device_blocked(sdev))
 			ret = BLK_STS_DEV_RESOURCE;
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 90895d594e64..b90dca1fa430 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -63,6 +63,18 @@ typedef u8 __bitwise blk_status_t;
  */
 #define BLK_STS_DEV_RESOURCE	((__force blk_status_t)13)
 
+/*
+ * BLK_STS_ZONE_RESOURCE is returned from the driver to the block layer if zone
+ * related resources are unavailable, but the driver can guarantee the queue
+ * will be rerun in the future once the resources become available again.
+ *
+ * This is different from BLK_STS_DEV_RESOURCE in that it explicitly references
+ * a zone specific resource and IO to a different zone on the same device could
+ * still be served. Examples of that are zones that are write-locked, but a read
+ * to the same zone could be served.
+ */
+#define BLK_STS_ZONE_RESOURCE	((__force blk_status_t)14)
+
 /**
  * blk_path_error - returns true if error may be path related
  * @error: status the request was completed with
@@ -296,6 +308,8 @@ enum req_opf {
 	REQ_OP_ZONE_CLOSE	= 11,
 	/* Transition a zone to full */
 	REQ_OP_ZONE_FINISH	= 12,
+	/* write data at the current zone write pointer */
+	REQ_OP_ZONE_APPEND	= 13,
 
 	/* SCSI passthrough using struct scsi_request */
 	REQ_OP_SCSI_IN		= 32,
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 91c6e413bf6b..158641fbc7cd 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -331,6 +331,7 @@ struct queue_limits {
 	unsigned int		max_hw_discard_sectors;
 	unsigned int		max_write_same_sectors;
 	unsigned int		max_write_zeroes_sectors;
+	unsigned int		max_zone_append_sectors;
 	unsigned int		discard_granularity;
 	unsigned int		discard_alignment;
 
@@ -749,6 +750,9 @@ static inline bool rq_mergeable(struct request *rq)
 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
 		return false;
 
+	if (req_op(rq) == REQ_OP_ZONE_APPEND)
+		return false;
+
 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
 		return false;
 	if (rq->rq_flags & RQF_NOMERGE_FLAGS)
@@ -1083,6 +1087,8 @@ extern void blk_queue_max_write_same_sectors(struct request_queue *q,
 extern void blk_queue_max_write_zeroes_sectors(struct request_queue *q,
 		unsigned int max_write_same_sectors);
 extern void blk_queue_logical_block_size(struct request_queue *, unsigned int);
+extern void blk_queue_max_zone_append_sectors(struct request_queue *q,
+		unsigned int max_zone_append_sectors);
 extern void blk_queue_physical_block_size(struct request_queue *, unsigned int);
 extern void blk_queue_alignment_offset(struct request_queue *q,
 				       unsigned int alignment);
@@ -1300,6 +1306,11 @@ static inline unsigned int queue_max_segment_size(const struct request_queue *q)
 	return q->limits.max_segment_size;
 }
 
+static inline unsigned int queue_max_zone_append_sectors(const struct request_queue *q)
+{
+	return q->limits.max_zone_append_sectors;
+}
+
 static inline unsigned queue_logical_block_size(const struct request_queue *q)
 {
 	int retval = 512;
-- 
2.24.1

[PATCH v11 04/10] block: introduce blk_req_zone_write_trylock

[Johannes Thumshirn]

Introduce blk_req_zone_write_trylock(), which either grabs the write-lock
for a sequential zone or returns false, if the zone is already locked.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 block/blk-zoned.c      | 14 ++++++++++++++
 include/linux/blkdev.h |  1 +
 2 files changed, 15 insertions(+)

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index f87956e0dcaf..c822cfa7a102 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -82,6 +82,20 @@ bool blk_req_needs_zone_write_lock(struct request *rq)
 }
 EXPORT_SYMBOL_GPL(blk_req_needs_zone_write_lock);
 
+bool blk_req_zone_write_trylock(struct request *rq)
+{
+	unsigned int zno = blk_rq_zone_no(rq);
+
+	if (test_and_set_bit(zno, rq->q->seq_zones_wlock))
+		return false;
+
+	WARN_ON_ONCE(rq->rq_flags & RQF_ZONE_WRITE_LOCKED);
+	rq->rq_flags |= RQF_ZONE_WRITE_LOCKED;
+
+	return true;
+}
+EXPORT_SYMBOL_GPL(blk_req_zone_write_trylock);
+
 void __blk_req_zone_write_lock(struct request *rq)
 {
 	if (WARN_ON_ONCE(test_and_set_bit(blk_rq_zone_no(rq),
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 158641fbc7cd..d6e6ce3dc656 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1737,6 +1737,7 @@ extern int bdev_write_page(struct block_device *, sector_t, struct page *,
 
 #ifdef CONFIG_BLK_DEV_ZONED
 bool blk_req_needs_zone_write_lock(struct request *rq);
+bool blk_req_zone_write_trylock(struct request *rq);
 void __blk_req_zone_write_lock(struct request *rq);
 void __blk_req_zone_write_unlock(struct request *rq);
 
-- 
2.24.1

[PATCH v11 05/10] block: Modify revalidate zones

[Johannes Thumshirn]

From: Damien Le Moal <damien.lemoal@wdc.com>

Modify the interface of blk_revalidate_disk_zones() to add an optional
driver callback function that a driver can use to extend processing
done during zone revalidation. The callback, if defined, is executed
with the device request queue frozen, after all zones have been
inspected.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 block/blk-zoned.c              | 9 ++++++++-
 drivers/block/null_blk_zoned.c | 2 +-
 include/linux/blkdev.h         | 3 ++-
 3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
index c822cfa7a102..23831fa8701d 100644
--- a/block/blk-zoned.c
+++ b/block/blk-zoned.c
@@ -471,14 +471,19 @@ static int blk_revalidate_zone_cb(struct blk_zone *zone, unsigned int idx,
 /**
  * blk_revalidate_disk_zones - (re)allocate and initialize zone bitmaps
  * @disk:	Target disk
+ * @update_driver_data:	Callback to update driver data on the frozen disk
  *
  * Helper function for low-level device drivers to (re) allocate and initialize
  * a disk request queue zone bitmaps. This functions should normally be called
  * within the disk ->revalidate method for blk-mq based drivers.  For BIO based
  * drivers only q->nr_zones needs to be updated so that the sysfs exposed value
  * is correct.
+ * If the @update_driver_data callback function is not NULL, the callback is
+ * executed with the device request queue frozen after all zones have been
+ * checked.
  */
-int blk_revalidate_disk_zones(struct gendisk *disk)
+int blk_revalidate_disk_zones(struct gendisk *disk,
+			      void (*update_driver_data)(struct gendisk *disk))
 {
 	struct request_queue *q = disk->queue;
 	struct blk_revalidate_zone_args args = {
@@ -512,6 +517,8 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 		q->nr_zones = args.nr_zones;
 		swap(q->seq_zones_wlock, args.seq_zones_wlock);
 		swap(q->conv_zones_bitmap, args.conv_zones_bitmap);
+		if (update_driver_data)
+			update_driver_data(disk);
 		ret = 0;
 	} else {
 		pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
diff --git a/drivers/block/null_blk_zoned.c b/drivers/block/null_blk_zoned.c
index 9e4bcdad1a80..46641df2e58e 100644
--- a/drivers/block/null_blk_zoned.c
+++ b/drivers/block/null_blk_zoned.c
@@ -73,7 +73,7 @@ int null_register_zoned_dev(struct nullb *nullb)
 	struct request_queue *q = nullb->q;
 
 	if (queue_is_mq(q))
-		return blk_revalidate_disk_zones(nullb->disk);
+		return blk_revalidate_disk_zones(nullb->disk, NULL);
 
 	blk_queue_chunk_sectors(q, nullb->dev->zone_size_sects);
 	q->nr_zones = blkdev_nr_zones(nullb->disk);
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index d6e6ce3dc656..fd405dac8eb0 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -357,7 +357,8 @@ unsigned int blkdev_nr_zones(struct gendisk *disk);
 extern int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op,
 			    sector_t sectors, sector_t nr_sectors,
 			    gfp_t gfp_mask);
-extern int blk_revalidate_disk_zones(struct gendisk *disk);
+int blk_revalidate_disk_zones(struct gendisk *disk,
+			      void (*update_driver_data)(struct gendisk *disk));
 
 extern int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
 				     unsigned int cmd, unsigned long arg);
-- 
2.24.1

[PATCH v11 06/10] scsi: sd_zbc: factor out sanity checks for zoned commands

[Johannes Thumshirn]

Factor sanity checks for zoned commands from sd_zbc_setup_zone_mgmt_cmnd().

This will help with the introduction of an emulated ZONE_APPEND command.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
---
 drivers/scsi/sd_zbc.c | 36 +++++++++++++++++++++++++-----------
 1 file changed, 25 insertions(+), 11 deletions(-)

diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
index f45c22b09726..ee156fbf3780 100644
--- a/drivers/scsi/sd_zbc.c
+++ b/drivers/scsi/sd_zbc.c
@@ -209,6 +209,26 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 	return ret;
 }
 
+static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
+{
+	struct request *rq = cmd->request;
+	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
+	sector_t sector = blk_rq_pos(rq);
+
+	if (!sd_is_zoned(sdkp))
+		/* Not a zoned device */
+		return BLK_STS_IOERR;
+
+	if (sdkp->device->changed)
+		return BLK_STS_IOERR;
+
+	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
+		/* Unaligned request */
+		return BLK_STS_IOERR;
+
+	return BLK_STS_OK;
+}
+
 /**
  * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
  *			can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
@@ -223,20 +243,14 @@ blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
 					 unsigned char op, bool all)
 {
 	struct request *rq = cmd->request;
-	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 	sector_t sector = blk_rq_pos(rq);
+	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
 	sector_t block = sectors_to_logical(sdkp->device, sector);
+	blk_status_t ret;
 
-	if (!sd_is_zoned(sdkp))
-		/* Not a zoned device */
-		return BLK_STS_IOERR;
-
-	if (sdkp->device->changed)
-		return BLK_STS_IOERR;
-
-	if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
-		/* Unaligned request */
-		return BLK_STS_IOERR;
+	ret = sd_zbc_cmnd_checks(cmd);
+	if (ret != BLK_STS_OK)
+		return ret;
 
 	cmd->cmd_len = 16;
 	memset(cmd->cmnd, 0, cmd->cmd_len);
-- 
2.24.1

[PATCH v11 07/10] scsi: sd_zbc: emulate ZONE_APPEND commands

[Johannes Thumshirn]

Emulate ZONE_APPEND for SCSI disks using a regular WRITE(16) command
with a start LBA set to the target zone write pointer position.

In order to always know the write pointer position of a sequential write
zone, the write pointer of all zones is tracked using an array of 32bits
zone write pointer offset attached to the scsi disk structure. Each
entry of the array indicate a zone write pointer position relative to
the zone start sector. The write pointer offsets are maintained in sync
with the device as follows:
1) the write pointer offset of a zone is reset to 0 when a
   REQ_OP_ZONE_RESET command completes.
2) the write pointer offset of a zone is set to the zone size when a
   REQ_OP_ZONE_FINISH command completes.
3) the write pointer offset of a zone is incremented by the number of
   512B sectors written when a write, write same or a zone append
   command completes.
4) the write pointer offset of all zones is reset to 0 when a
   REQ_OP_ZONE_RESET_ALL command completes.

Since the block layer does not write lock zones for zone append
commands, to ensure a sequential ordering of the regular write commands
used for the emulation, the target zone of a zone append command is
locked when the function sd_zbc_prepare_zone_append() is called from
sd_setup_read_write_cmnd(). If the zone write lock cannot be obtained
(e.g. a zone append is in-flight or a regular write has already locked
the zone), the zone append command dispatching is delayed by returning
BLK_STS_ZONE_RESOURCE.

To avoid the need for write locking all zones for REQ_OP_ZONE_RESET_ALL
requests, use a spinlock to protect accesses and modifications of the
zone write pointer offsets. This spinlock is initialized from sd_probe()
using the new function sd_zbc_init().

Co-developed-by: Damien Le Moal <Damien.LeMoal@wdc.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 drivers/scsi/sd.c     |  16 +-
 drivers/scsi/sd.h     |  43 ++++-
 drivers/scsi/sd_zbc.c | 363 +++++++++++++++++++++++++++++++++++++++---
 3 files changed, 392 insertions(+), 30 deletions(-)

diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
index a793cb08d025..7b0383e42b4c 100644
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -1206,6 +1206,12 @@ static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
 		}
 	}
 
+	if (req_op(rq) == REQ_OP_ZONE_APPEND) {
+		ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
+		if (ret)
+			return ret;
+	}
+
 	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
 	dix = scsi_prot_sg_count(cmd);
 	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
@@ -1287,6 +1293,7 @@ static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
 		return sd_setup_flush_cmnd(cmd);
 	case REQ_OP_READ:
 	case REQ_OP_WRITE:
+	case REQ_OP_ZONE_APPEND:
 		return sd_setup_read_write_cmnd(cmd);
 	case REQ_OP_ZONE_RESET:
 		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
@@ -2055,7 +2062,7 @@ static int sd_done(struct scsi_cmnd *SCpnt)
 
  out:
 	if (sd_is_zoned(sdkp))
-		sd_zbc_complete(SCpnt, good_bytes, &sshdr);
+		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
 
 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
 					   "sd_done: completed %d of %d bytes\n",
@@ -3372,6 +3379,10 @@ static int sd_probe(struct device *dev)
 	sdkp->first_scan = 1;
 	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
 
+	error = sd_zbc_init_disk(sdkp);
+	if (error)
+		goto out_free_index;
+
 	sd_revalidate_disk(gd);
 
 	gd->flags = GENHD_FL_EXT_DEVT;
@@ -3409,6 +3420,7 @@ static int sd_probe(struct device *dev)
  out_put:
 	put_disk(gd);
  out_free:
+	sd_zbc_release_disk(sdkp);
 	kfree(sdkp);
  out:
 	scsi_autopm_put_device(sdp);
@@ -3485,6 +3497,8 @@ static void scsi_disk_release(struct device *dev)
 	put_disk(disk);
 	put_device(&sdkp->device->sdev_gendev);
 
+	sd_zbc_release_disk(sdkp);
+
 	kfree(sdkp);
 }
 
diff --git a/drivers/scsi/sd.h b/drivers/scsi/sd.h
index 50fff0bf8c8e..3a74f4b45134 100644
--- a/drivers/scsi/sd.h
+++ b/drivers/scsi/sd.h
@@ -79,6 +79,12 @@ struct scsi_disk {
 	u32		zones_optimal_open;
 	u32		zones_optimal_nonseq;
 	u32		zones_max_open;
+	u32		*zones_wp_offset;
+	spinlock_t	zones_wp_offset_lock;
+	u32		*rev_wp_offset;
+	struct mutex	rev_mutex;
+	struct work_struct zone_wp_offset_work;
+	char		*zone_wp_update_buf;
 #endif
 	atomic_t	openers;
 	sector_t	capacity;	/* size in logical blocks */
@@ -207,17 +213,35 @@ static inline int sd_is_zoned(struct scsi_disk *sdkp)
 
 #ifdef CONFIG_BLK_DEV_ZONED
 
+int sd_zbc_init_disk(struct scsi_disk *sdkp);
+void sd_zbc_release_disk(struct scsi_disk *sdkp);
 extern int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buffer);
 extern void sd_zbc_print_zones(struct scsi_disk *sdkp);
 blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
 					 unsigned char op, bool all);
-extern void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
-			    struct scsi_sense_hdr *sshdr);
+unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
+			     struct scsi_sense_hdr *sshdr);
 int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 		unsigned int nr_zones, report_zones_cb cb, void *data);
 
+blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
+				        unsigned int nr_blocks);
+
 #else /* CONFIG_BLK_DEV_ZONED */
 
+static inline int sd_zbc_init(void)
+{
+	return 0;
+}
+
+static inline int sd_zbc_init_disk(struct scsi_disk *sdkp)
+{
+	return 0;
+}
+
+static inline void sd_zbc_exit(void) {}
+static inline void sd_zbc_release_disk(struct scsi_disk *sdkp) {}
+
 static inline int sd_zbc_read_zones(struct scsi_disk *sdkp,
 				    unsigned char *buf)
 {
@@ -233,9 +257,18 @@ static inline blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
 	return BLK_STS_TARGET;
 }
 
-static inline void sd_zbc_complete(struct scsi_cmnd *cmd,
-				   unsigned int good_bytes,
-				   struct scsi_sense_hdr *sshdr) {}
+static inline unsigned int sd_zbc_complete(struct scsi_cmnd *cmd,
+			unsigned int good_bytes, struct scsi_sense_hdr *sshdr)
+{
+	return 0;
+}
+
+static inline blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd,
+						      sector_t *lba,
+						      unsigned int nr_blocks)
+{
+	return BLK_STS_TARGET;
+}
 
 #define sd_zbc_report_zones NULL
 
diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
index ee156fbf3780..bb87fbba2a09 100644
--- a/drivers/scsi/sd_zbc.c
+++ b/drivers/scsi/sd_zbc.c
@@ -11,6 +11,7 @@
 #include <linux/blkdev.h>
 #include <linux/vmalloc.h>
 #include <linux/sched/mm.h>
+#include <linux/mutex.h>
 
 #include <asm/unaligned.h>
 
@@ -19,11 +20,36 @@
 
 #include "sd.h"
 
+static unsigned int sd_zbc_get_zone_wp_offset(struct blk_zone *zone)
+{
+	if (zone->type == ZBC_ZONE_TYPE_CONV)
+		return 0;
+
+	switch (zone->cond) {
+	case BLK_ZONE_COND_IMP_OPEN:
+	case BLK_ZONE_COND_EXP_OPEN:
+	case BLK_ZONE_COND_CLOSED:
+		return zone->wp - zone->start;
+	case BLK_ZONE_COND_FULL:
+		return zone->len;
+	case BLK_ZONE_COND_EMPTY:
+	case BLK_ZONE_COND_OFFLINE:
+	case BLK_ZONE_COND_READONLY:
+	default:
+		/*
+		 * Offline and read-only zones do not have a valid
+		 * write pointer. Use 0 as for an empty zone.
+		 */
+		return 0;
+	}
+}
+
 static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
 			       unsigned int idx, report_zones_cb cb, void *data)
 {
 	struct scsi_device *sdp = sdkp->device;
 	struct blk_zone zone = { 0 };
+	int ret;
 
 	zone.type = buf[0] & 0x0f;
 	zone.cond = (buf[1] >> 4) & 0xf;
@@ -39,7 +65,14 @@ static int sd_zbc_parse_report(struct scsi_disk *sdkp, u8 *buf,
 	    zone.cond == ZBC_ZONE_COND_FULL)
 		zone.wp = zone.start + zone.len;
 
-	return cb(&zone, idx, data);
+	ret = cb(&zone, idx, data);
+	if (ret)
+		return ret;
+
+	if (sdkp->rev_wp_offset)
+		sdkp->rev_wp_offset[idx] = sd_zbc_get_zone_wp_offset(&zone);
+
+	return 0;
 }
 
 /**
@@ -229,6 +262,116 @@ static blk_status_t sd_zbc_cmnd_checks(struct scsi_cmnd *cmd)
 	return BLK_STS_OK;
 }
 
+#define SD_ZBC_INVALID_WP_OFST	(~0u)
+#define SD_ZBC_UPDATING_WP_OFST	(SD_ZBC_INVALID_WP_OFST - 1)
+
+static int sd_zbc_update_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
+				    void *data)
+{
+	struct scsi_disk *sdkp = data;
+
+	lockdep_assert_held(&sdkp->zones_wp_offset_lock);
+
+	sdkp->zones_wp_offset[idx] = sd_zbc_get_zone_wp_offset(zone);
+
+	return 0;
+}
+
+static void sd_zbc_update_wp_offset_workfn(struct work_struct *work)
+{
+	struct scsi_disk *sdkp;
+	unsigned int zno;
+	int ret;
+
+	sdkp = container_of(work, struct scsi_disk, zone_wp_offset_work);
+
+	spin_lock_bh(&sdkp->zones_wp_offset_lock);
+	for (zno = 0; zno < sdkp->nr_zones; zno++) {
+		if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
+			continue;
+
+		spin_unlock_bh(&sdkp->zones_wp_offset_lock);
+		ret = sd_zbc_do_report_zones(sdkp, sdkp->zone_wp_update_buf,
+					     SD_BUF_SIZE,
+					     zno * sdkp->zone_blocks, true);
+		spin_lock_bh(&sdkp->zones_wp_offset_lock);
+		if (!ret)
+			sd_zbc_parse_report(sdkp, sdkp->zone_wp_update_buf + 64,
+					    zno, sd_zbc_update_wp_offset_cb,
+					    sdkp);
+	}
+	spin_unlock_bh(&sdkp->zones_wp_offset_lock);
+
+	scsi_device_put(sdkp->device);
+}
+
+/**
+ * sd_zbc_prepare_zone_append() - Prepare an emulated ZONE_APPEND command.
+ * @cmd: the command to setup
+ * @lba: the LBA to patch
+ * @nr_blocks: the number of LBAs to be written
+ *
+ * Called from sd_setup_read_write_cmnd() for REQ_OP_ZONE_APPEND.
+ * @sd_zbc_prepare_zone_append() handles the necessary zone wrote locking and
+ * patching of the lba for an emulated ZONE_APPEND command.
+ *
+ * In case the cached write pointer offset is %SD_ZBC_INVALID_WP_OFST it will
+ * schedule a REPORT ZONES command and return BLK_STS_IOERR.
+ */
+blk_status_t sd_zbc_prepare_zone_append(struct scsi_cmnd *cmd, sector_t *lba,
+					unsigned int nr_blocks)
+{
+	struct request *rq = cmd->request;
+	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
+	unsigned int wp_offset, zno = blk_rq_zone_no(rq);
+	blk_status_t ret;
+
+	ret = sd_zbc_cmnd_checks(cmd);
+	if (ret != BLK_STS_OK)
+		return ret;
+
+	if (!blk_rq_zone_is_seq(rq))
+		return BLK_STS_IOERR;
+
+	/* Unlock of the write lock will happen in sd_zbc_complete() */
+	if (!blk_req_zone_write_trylock(rq))
+		return BLK_STS_ZONE_RESOURCE;
+
+	spin_lock_bh(&sdkp->zones_wp_offset_lock);
+	wp_offset = sdkp->zones_wp_offset[zno];
+	switch (wp_offset) {
+	case SD_ZBC_INVALID_WP_OFST:
+		/*
+		 * We are about to schedule work to update a zone write pointer
+		 * offset, which will cause the zone append command to be
+		 * requeued. So make sure that the scsi device does not go away
+		 * while the work is being processed.
+		 */
+		if (scsi_device_get(sdkp->device)) {
+			ret = BLK_STS_IOERR;
+			break;
+		}
+		sdkp->zones_wp_offset[zno] = SD_ZBC_UPDATING_WP_OFST;
+		schedule_work(&sdkp->zone_wp_offset_work);
+		fallthrough;
+	case SD_ZBC_UPDATING_WP_OFST:
+		ret = BLK_STS_DEV_RESOURCE;
+		break;
+	default:
+		wp_offset = sectors_to_logical(sdkp->device, wp_offset);
+		if (wp_offset + nr_blocks > sdkp->zone_blocks) {
+			ret = BLK_STS_IOERR;
+			break;
+		}
+
+		*lba += wp_offset;
+	}
+	spin_unlock_bh(&sdkp->zones_wp_offset_lock);
+	if (ret)
+		blk_req_zone_write_unlock(rq);
+	return ret;
+}
+
 /**
  * sd_zbc_setup_zone_mgmt_cmnd - Prepare a zone ZBC_OUT command. The operations
  *			can be RESET WRITE POINTER, OPEN, CLOSE or FINISH.
@@ -269,16 +412,105 @@ blk_status_t sd_zbc_setup_zone_mgmt_cmnd(struct scsi_cmnd *cmd,
 	return BLK_STS_OK;
 }
 
+static bool sd_zbc_need_zone_wp_update(struct request *rq)
+{
+	switch (req_op(rq)) {
+	case REQ_OP_ZONE_APPEND:
+	case REQ_OP_ZONE_FINISH:
+	case REQ_OP_ZONE_RESET:
+	case REQ_OP_ZONE_RESET_ALL:
+		return true;
+	case REQ_OP_WRITE:
+	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_WRITE_SAME:
+		return blk_rq_zone_is_seq(rq);
+	default:
+		return false;
+	}
+}
+
+/**
+ * sd_zbc_zone_wp_update - Update cached zone write pointer upon cmd completion
+ * @cmd: Completed command
+ * @good_bytes: Command reply bytes
+ *
+ * Called from sd_zbc_complete() to handle the update of the cached zone write
+ * pointer value in case an update is needed.
+ */
+static unsigned int sd_zbc_zone_wp_update(struct scsi_cmnd *cmd,
+					  unsigned int good_bytes)
+{
+	int result = cmd->result;
+	struct request *rq = cmd->request;
+	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
+	unsigned int zno = blk_rq_zone_no(rq);
+	enum req_opf op = req_op(rq);
+
+	/*
+	 * If we got an error for a command that needs updating the write
+	 * pointer offset cache, we must mark the zone wp offset entry as
+	 * invalid to force an update from disk the next time a zone append
+	 * command is issued.
+	 */
+	spin_lock_bh(&sdkp->zones_wp_offset_lock);
+
+	if (result && op != REQ_OP_ZONE_RESET_ALL) {
+		if (op == REQ_OP_ZONE_APPEND) {
+			/* Force complete completion (no retry) */
+			good_bytes = 0;
+			scsi_set_resid(cmd, blk_rq_bytes(rq));
+		}
+
+		/*
+		 * Force an update of the zone write pointer offset on
+		 * the next zone append access.
+		 */
+		if (sdkp->zones_wp_offset[zno] != SD_ZBC_UPDATING_WP_OFST)
+			sdkp->zones_wp_offset[zno] = SD_ZBC_INVALID_WP_OFST;
+		goto unlock_wp_offset;
+	}
+
+	switch (op) {
+	case REQ_OP_ZONE_APPEND:
+		rq->__sector += sdkp->zones_wp_offset[zno];
+		fallthrough;
+	case REQ_OP_WRITE_ZEROES:
+	case REQ_OP_WRITE_SAME:
+	case REQ_OP_WRITE:
+		if (sdkp->zones_wp_offset[zno] < sd_zbc_zone_sectors(sdkp))
+			sdkp->zones_wp_offset[zno] +=
+						good_bytes >> SECTOR_SHIFT;
+		break;
+	case REQ_OP_ZONE_RESET:
+		sdkp->zones_wp_offset[zno] = 0;
+		break;
+	case REQ_OP_ZONE_FINISH:
+		sdkp->zones_wp_offset[zno] = sd_zbc_zone_sectors(sdkp);
+		break;
+	case REQ_OP_ZONE_RESET_ALL:
+		memset(sdkp->zones_wp_offset, 0,
+		       sdkp->nr_zones * sizeof(unsigned int));
+		break;
+	default:
+		break;
+	}
+
+unlock_wp_offset:
+	spin_unlock_bh(&sdkp->zones_wp_offset_lock);
+
+	return good_bytes;
+}
+
 /**
  * sd_zbc_complete - ZBC command post processing.
  * @cmd: Completed command
  * @good_bytes: Command reply bytes
  * @sshdr: command sense header
  *
- * Called from sd_done(). Process report zones reply and handle reset zone
- * and write commands errors.
+ * Called from sd_done() to handle zone commands errors and updates to the
+ * device queue zone write pointer offset cahce.
  */
-void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
+unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
 		     struct scsi_sense_hdr *sshdr)
 {
 	int result = cmd->result;
@@ -294,7 +526,13 @@ void sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
 		 * so be quiet about the error.
 		 */
 		rq->rq_flags |= RQF_QUIET;
-	}
+	} else if (sd_zbc_need_zone_wp_update(rq))
+		good_bytes = sd_zbc_zone_wp_update(cmd, good_bytes);
+
+	if (req_op(rq) == REQ_OP_ZONE_APPEND)
+		blk_req_zone_write_unlock(rq);
+
+	return good_bytes;
 }
 
 /**
@@ -396,11 +634,67 @@ static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
 	return 0;
 }
 
+static void sd_zbc_revalidate_zones_cb(struct gendisk *disk)
+{
+	struct scsi_disk *sdkp = scsi_disk(disk);
+
+	swap(sdkp->zones_wp_offset, sdkp->rev_wp_offset);
+}
+
+static int sd_zbc_revalidate_zones(struct scsi_disk *sdkp,
+				   u32 zone_blocks,
+				   unsigned int nr_zones)
+{
+	struct gendisk *disk = sdkp->disk;
+	int ret = 0;
+
+	/*
+	 * Make sure revalidate zones are serialized to ensure exclusive
+	 * updates of the scsi disk data.
+	 */
+	mutex_lock(&sdkp->rev_mutex);
+
+	/*
+	 * Revalidate the disk zones to update the device request queue zone
+	 * bitmaps and the zone write pointer offset array. Do this only once
+	 * the device capacity is set on the second revalidate execution for
+	 * disk scan or if something changed when executing a normal revalidate.
+	 */
+	if (sdkp->first_scan) {
+		sdkp->zone_blocks = zone_blocks;
+		sdkp->nr_zones = nr_zones;
+		goto unlock;
+	}
+
+	if (sdkp->zone_blocks == zone_blocks &&
+	    sdkp->nr_zones == nr_zones &&
+	    disk->queue->nr_zones == nr_zones)
+		goto unlock;
+
+	sdkp->rev_wp_offset = kvcalloc(nr_zones, sizeof(u32), GFP_NOIO);
+	if (!sdkp->rev_wp_offset) {
+		ret = -ENOMEM;
+		goto unlock;
+	}
+
+	ret = blk_revalidate_disk_zones(disk, sd_zbc_revalidate_zones_cb);
+
+	kvfree(sdkp->rev_wp_offset);
+	sdkp->rev_wp_offset = NULL;
+
+unlock:
+	mutex_unlock(&sdkp->rev_mutex);
+
+	return ret;
+}
+
 int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
 {
 	struct gendisk *disk = sdkp->disk;
+	struct request_queue *q = disk->queue;
 	unsigned int nr_zones;
 	u32 zone_blocks = 0;
+	u32 max_append;
 	int ret;
 
 	if (!sd_is_zoned(sdkp))
@@ -421,35 +715,31 @@ int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
 		goto err;
 
 	/* The drive satisfies the kernel restrictions: set it up */
-	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, sdkp->disk->queue);
-	blk_queue_required_elevator_features(sdkp->disk->queue,
-					     ELEVATOR_F_ZBD_SEQ_WRITE);
+	blk_queue_flag_set(QUEUE_FLAG_ZONE_RESETALL, q);
+	blk_queue_required_elevator_features(q, ELEVATOR_F_ZBD_SEQ_WRITE);
 	nr_zones = round_up(sdkp->capacity, zone_blocks) >> ilog2(zone_blocks);
 
 	/* READ16/WRITE16 is mandatory for ZBC disks */
 	sdkp->device->use_16_for_rw = 1;
 	sdkp->device->use_10_for_rw = 0;
 
+	ret = sd_zbc_revalidate_zones(sdkp, zone_blocks, nr_zones);
+	if (ret)
+		goto err;
+
 	/*
-	 * Revalidate the disk zone bitmaps once the block device capacity is
-	 * set on the second revalidate execution during disk scan and if
-	 * something changed when executing a normal revalidate.
+	 * On the first scan 'chunk_sectors' isn't setup yet, so calling
+	 * blk_queue_max_zone_append_sectors() will result in a WARN(). Defer
+	 * this setting to the second scan.
 	 */
-	if (sdkp->first_scan) {
-		sdkp->zone_blocks = zone_blocks;
-		sdkp->nr_zones = nr_zones;
+	if (sdkp->first_scan)
 		return 0;
-	}
 
-	if (sdkp->zone_blocks != zone_blocks ||
-	    sdkp->nr_zones != nr_zones ||
-	    disk->queue->nr_zones != nr_zones) {
-		ret = blk_revalidate_disk_zones(disk);
-		if (ret != 0)
-			goto err;
-		sdkp->zone_blocks = zone_blocks;
-		sdkp->nr_zones = nr_zones;
-	}
+	max_append = min_t(u32, logical_to_sectors(sdkp->device, zone_blocks),
+			   q->limits.max_segments << (PAGE_SHIFT - 9));
+	max_append = min_t(u32, max_append, queue_max_hw_sectors(q));
+
+	blk_queue_max_zone_append_sectors(q, max_append);
 
 	return 0;
 
@@ -475,3 +765,28 @@ void sd_zbc_print_zones(struct scsi_disk *sdkp)
 			  sdkp->nr_zones,
 			  sdkp->zone_blocks);
 }
+
+int sd_zbc_init_disk(struct scsi_disk *sdkp)
+{
+	if (!sd_is_zoned(sdkp))
+		return 0;
+
+	sdkp->zones_wp_offset = NULL;
+	spin_lock_init(&sdkp->zones_wp_offset_lock);
+	sdkp->rev_wp_offset = NULL;
+	mutex_init(&sdkp->rev_mutex);
+	INIT_WORK(&sdkp->zone_wp_offset_work, sd_zbc_update_wp_offset_workfn);
+	sdkp->zone_wp_update_buf = kzalloc(SD_BUF_SIZE, GFP_KERNEL);
+	if (!sdkp->zone_wp_update_buf)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void sd_zbc_release_disk(struct scsi_disk *sdkp)
+{
+	kvfree(sdkp->zones_wp_offset);
+	sdkp->zones_wp_offset = NULL;
+	kfree(sdkp->zone_wp_update_buf);
+	sdkp->zone_wp_update_buf = NULL;
+}
-- 
2.24.1

[PATCH v11 08/10] null_blk: Support REQ_OP_ZONE_APPEND

[Johannes Thumshirn]

From: Damien Le Moal <damien.lemoal@wdc.com>

Support REQ_OP_ZONE_APPEND requests for null_blk devices with zoned
mode enabled. Use the internally tracked zone write pointer position
as the actual write position and return it using the command request
__sector field in the case of an mq device and using the command BIO
sector in the case of a BIO device.

Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
---
 drivers/block/null_blk_zoned.c | 37 ++++++++++++++++++++++++++--------
 1 file changed, 29 insertions(+), 8 deletions(-)

diff --git a/drivers/block/null_blk_zoned.c b/drivers/block/null_blk_zoned.c
index 46641df2e58e..9c19f747f394 100644
--- a/drivers/block/null_blk_zoned.c
+++ b/drivers/block/null_blk_zoned.c
@@ -70,13 +70,20 @@ int null_init_zoned_dev(struct nullb_device *dev, struct request_queue *q)
 
 int null_register_zoned_dev(struct nullb *nullb)
 {
+	struct nullb_device *dev = nullb->dev;
 	struct request_queue *q = nullb->q;
 
-	if (queue_is_mq(q))
-		return blk_revalidate_disk_zones(nullb->disk, NULL);
+	if (queue_is_mq(q)) {
+		int ret = blk_revalidate_disk_zones(nullb->disk, NULL);
+
+		if (ret)
+			return ret;
+	} else {
+		blk_queue_chunk_sectors(q, dev->zone_size_sects);
+		q->nr_zones = blkdev_nr_zones(nullb->disk);
+	}
 
-	blk_queue_chunk_sectors(q, nullb->dev->zone_size_sects);
-	q->nr_zones = blkdev_nr_zones(nullb->disk);
+	blk_queue_max_zone_append_sectors(q, dev->zone_size_sects);
 
 	return 0;
 }
@@ -138,7 +145,7 @@ size_t null_zone_valid_read_len(struct nullb *nullb,
 }
 
 static blk_status_t null_zone_write(struct nullb_cmd *cmd, sector_t sector,
-		     unsigned int nr_sectors)
+				    unsigned int nr_sectors, bool append)
 {
 	struct nullb_device *dev = cmd->nq->dev;
 	unsigned int zno = null_zone_no(dev, sector);
@@ -158,9 +165,21 @@ static blk_status_t null_zone_write(struct nullb_cmd *cmd, sector_t sector,
 	case BLK_ZONE_COND_IMP_OPEN:
 	case BLK_ZONE_COND_EXP_OPEN:
 	case BLK_ZONE_COND_CLOSED:
-		/* Writes must be at the write pointer position */
-		if (sector != zone->wp)
+		/*
+		 * Regular writes must be at the write pointer position.
+		 * Zone append writes are automatically issued at the write
+		 * pointer and the position returned using the request or BIO
+		 * sector.
+		 */
+		if (append) {
+			sector = zone->wp;
+			if (cmd->bio)
+				cmd->bio->bi_iter.bi_sector = sector;
+			else
+				cmd->rq->__sector = sector;
+		} else if (sector != zone->wp) {
 			return BLK_STS_IOERR;
+		}
 
 		if (zone->cond != BLK_ZONE_COND_EXP_OPEN)
 			zone->cond = BLK_ZONE_COND_IMP_OPEN;
@@ -242,7 +261,9 @@ blk_status_t null_process_zoned_cmd(struct nullb_cmd *cmd, enum req_opf op,
 {
 	switch (op) {
 	case REQ_OP_WRITE:
-		return null_zone_write(cmd, sector, nr_sectors);
+		return null_zone_write(cmd, sector, nr_sectors, false);
+	case REQ_OP_ZONE_APPEND:
+		return null_zone_write(cmd, sector, nr_sectors, true);
 	case REQ_OP_ZONE_RESET:
 	case REQ_OP_ZONE_RESET_ALL:
 	case REQ_OP_ZONE_OPEN:
-- 
2.24.1

[PATCH v11 09/10] block: export bio_release_pages and bio_iov_iter_get_pages

[Johannes Thumshirn]

Export bio_release_pages and bio_iov_iter_get_pages, so they can be used
from modular code.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
---
 block/bio.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/block/bio.c b/block/bio.c
index 3aa3c4ce2e5e..e4c46e2bd5ba 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -951,6 +951,7 @@ void bio_release_pages(struct bio *bio, bool mark_dirty)
 		put_page(bvec->bv_page);
 	}
 }
+EXPORT_SYMBOL_GPL(bio_release_pages);
 
 static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter)
 {
@@ -1114,6 +1115,7 @@ int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
 		bio_set_flag(bio, BIO_NO_PAGE_REF);
 	return bio->bi_vcnt ? 0 : ret;
 }
+EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
 
 static void submit_bio_wait_endio(struct bio *bio)
 {
-- 
2.24.1

[PATCH v11 10/10] zonefs: use REQ_OP_ZONE_APPEND for sync DIO

[Johannes Thumshirn]

Synchronous direct I/O to a sequential write only zone can be issued using
the new REQ_OP_ZONE_APPEND request operation. As dispatching multiple
BIOs can potentially result in reordering, we cannot support asynchronous
IO via this interface.

We also can only dispatch up to queue_max_zone_append_sectors() via the
new zone-append method and have to return a short write back to user-space
in case an IO larger than queue_max_zone_append_sectors() has been issued.

Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Acked-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 fs/zonefs/super.c | 80 ++++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 72 insertions(+), 8 deletions(-)

diff --git a/fs/zonefs/super.c b/fs/zonefs/super.c
index 3ce9829a6936..0bf7009f50a2 100644
--- a/fs/zonefs/super.c
+++ b/fs/zonefs/super.c
@@ -20,6 +20,7 @@
 #include <linux/mman.h>
 #include <linux/sched/mm.h>
 #include <linux/crc32.h>
+#include <linux/task_io_accounting_ops.h>
 
 #include "zonefs.h"
 
@@ -596,6 +597,61 @@ static const struct iomap_dio_ops zonefs_write_dio_ops = {
 	.end_io			= zonefs_file_write_dio_end_io,
 };
 
+static ssize_t zonefs_file_dio_append(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct zonefs_inode_info *zi = ZONEFS_I(inode);
+	struct block_device *bdev = inode->i_sb->s_bdev;
+	unsigned int max;
+	struct bio *bio;
+	ssize_t size;
+	int nr_pages;
+	ssize_t ret;
+
+	nr_pages = iov_iter_npages(from, BIO_MAX_PAGES);
+	if (!nr_pages)
+		return 0;
+
+	max = queue_max_zone_append_sectors(bdev_get_queue(bdev));
+	max = ALIGN_DOWN(max << SECTOR_SHIFT, inode->i_sb->s_blocksize);
+	iov_iter_truncate(from, max);
+
+	bio = bio_alloc_bioset(GFP_NOFS, nr_pages, &fs_bio_set);
+	if (!bio)
+		return -ENOMEM;
+
+	bio_set_dev(bio, bdev);
+	bio->bi_iter.bi_sector = zi->i_zsector;
+	bio->bi_write_hint = iocb->ki_hint;
+	bio->bi_ioprio = iocb->ki_ioprio;
+	bio->bi_opf = REQ_OP_ZONE_APPEND | REQ_SYNC | REQ_IDLE;
+	if (iocb->ki_flags & IOCB_DSYNC)
+		bio->bi_opf |= REQ_FUA;
+
+	ret = bio_iov_iter_get_pages(bio, from);
+	if (unlikely(ret)) {
+		bio_io_error(bio);
+		return ret;
+	}
+	size = bio->bi_iter.bi_size;
+	task_io_account_write(ret);
+
+	if (iocb->ki_flags & IOCB_HIPRI)
+		bio_set_polled(bio, iocb);
+
+	ret = submit_bio_wait(bio);
+
+	bio_put(bio);
+
+	zonefs_file_write_dio_end_io(iocb, size, ret, 0);
+	if (ret >= 0) {
+		iocb->ki_pos += size;
+		return size;
+	}
+
+	return ret;
+}
+
 /*
  * Handle direct writes. For sequential zone files, this is the only possible
  * write path. For these files, check that the user is issuing writes
@@ -611,6 +667,8 @@ static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
 	struct inode *inode = file_inode(iocb->ki_filp);
 	struct zonefs_inode_info *zi = ZONEFS_I(inode);
 	struct super_block *sb = inode->i_sb;
+	bool sync = is_sync_kiocb(iocb);
+	bool append = false;
 	size_t count;
 	ssize_t ret;
 
@@ -619,7 +677,7 @@ static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
 	 * as this can cause write reordering (e.g. the first aio gets EAGAIN
 	 * on the inode lock but the second goes through but is now unaligned).
 	 */
-	if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !is_sync_kiocb(iocb) &&
+	if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !sync &&
 	    (iocb->ki_flags & IOCB_NOWAIT))
 		return -EOPNOTSUPP;
 
@@ -643,16 +701,22 @@ static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
 	}
 
 	/* Enforce sequential writes (append only) in sequential zones */
-	mutex_lock(&zi->i_truncate_mutex);
-	if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && iocb->ki_pos != zi->i_wpoffset) {
+	if (zi->i_ztype == ZONEFS_ZTYPE_SEQ) {
+		mutex_lock(&zi->i_truncate_mutex);
+		if (iocb->ki_pos != zi->i_wpoffset) {
+			mutex_unlock(&zi->i_truncate_mutex);
+			ret = -EINVAL;
+			goto inode_unlock;
+		}
 		mutex_unlock(&zi->i_truncate_mutex);
-		ret = -EINVAL;
-		goto inode_unlock;
+		append = sync;
 	}
-	mutex_unlock(&zi->i_truncate_mutex);
 
-	ret = iomap_dio_rw(iocb, from, &zonefs_iomap_ops,
-			   &zonefs_write_dio_ops, is_sync_kiocb(iocb));
+	if (append)
+		ret = zonefs_file_dio_append(iocb, from);
+	else
+		ret = iomap_dio_rw(iocb, from, &zonefs_iomap_ops,
+				   &zonefs_write_dio_ops, sync);
 	if (zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
 	    (ret > 0 || ret == -EIOCBQUEUED)) {
 		if (ret > 0)
-- 
2.24.1

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Christoph Hellwig]

The whole series looks good to me:

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

I hope we can get this in 5.8 to help with the btrfs in 5.9.

Re: [PATCH v11 00/10] Introduce Zone Append for writing to zoned block devices

[Jens Axboe]

On 5/12/20 2:55 AM, Johannes Thumshirn wrote:
> The upcoming NVMe ZNS Specification will define a new type of write
> command for zoned block devices, zone append.
> 
> When when writing to a zoned block device using zone append, the start
> sector of the write is pointing at the start LBA of the zone to write to.
> Upon completion the block device will respond with the position the data
> has been placed in the zone. This from a high level perspective can be
> seen like a file system's block allocator, where the user writes to a
> file and the file-system takes care of the data placement on the device.
> 
> In order to fully exploit the new zone append command in file-systems and
> other interfaces above the block layer, we choose to emulate zone append
> in SCSI and null_blk. This way we can have a single write path for both
> file-systems and other interfaces above the block-layer, like io_uring on
> zoned block devices, without having to care too much about the underlying
> characteristics of the device itself.
> 
> The emulation works by providing a cache of each zone's write pointer, so
> zone append issued to the disk can be translated to a write with a
> starting LBA of the write pointer. This LBA is used as input zone number
> for the write pointer lookup in the zone write pointer offset cache and
> the cached offset is then added to the LBA to get the actual position to
> write the data. In SCSI we then turn the REQ_OP_ZONE_APPEND request into a
> WRITE(16) command. Upon successful completion of the WRITE(16), the cache
> will be updated to the new write pointer location and the written sector
> will be noted in the request. On error the cache entry will be marked as
> invalid and on the next write an update of the write pointer will be
> scheduled, before issuing the actual write.
> 
> In order to reduce memory consumption, the only cached item is the offset
> of the write pointer from the start of the zone, everything else can be
> calculated. On an example drive with 52156 zones, the additional memory
> consumption of the cache is thus 52156 * 4 = 208624 Bytes or 51 4k Byte
> pages. The performance impact is neglectable for a spinning drive.
> 
> For null_blk the emulation is way simpler, as null_blk's zoned block
> device emulation support already caches the write pointer position, so we
> only need to report the position back to the upper layers. Additional
> caching is not needed here.
> 
> Furthermore we have converted zonefs to run use ZONE_APPEND for synchronous
> direct I/Os. Asynchronous I/O still uses the normal path via iomap.
> 
> Performance testing with zonefs sync writes on a 14 TB SMR drive and nullblk
> shows good results. On the SMR drive we're not regressing (the performance
> improvement is within noise), on nullblk we could drastically improve specific
> workloads:
> 
> * nullblk:
> 
> Single Thread Multiple Zones
> 				kIOPS	MiB/s	MB/s	% delta
> mq-deadline REQ_OP_WRITE	10.1	631	662
> mq-deadline REQ_OP_ZONE_APPEND	13.2	828	868	+31.12
> none REQ_OP_ZONE_APPEND		15.6	978	1026	+54.98
> 
> 
> Multiple Threads Multiple Zones
> 				kIOPS	MiB/s	MB/s	% delta
> mq-deadline REQ_OP_WRITE	10.2	640	671
> mq-deadline REQ_OP_ZONE_APPEND	10.4	650	681	+1.49
> none REQ_OP_ZONE_APPEND		16.9	1058	1109	+65.28
> 
> * 14 TB SMR drive
> 
> Single Thread Multiple Zones
> 				IOPS	MiB/s	MB/s	% delta
> mq-deadline REQ_OP_WRITE	797	49.9	52.3
> mq-deadline REQ_OP_ZONE_APPEND	806	50.4	52.9	+1.15
> 
> Multiple Threads Multiple Zones
> 				kIOPS	MiB/s	MB/s	% delta
> mq-deadline REQ_OP_WRITE	745	46.6	48.9
> mq-deadline REQ_OP_ZONE_APPEND	768	48	50.3	+2.86
> 
> The %-delta is against the baseline of REQ_OP_WRITE using mq-deadline as I/O
> scheduler.
> 
> The series is based on Jens' for-5.8/block branch with HEAD:
> ae979182ebb3 ("bdi: fix up for "remove the name field in struct backing_dev_info"")

Applied for 5.8, thanks.

-- 
Jens Axboe

[PATCH 0/2] two generic block layer fixes for 5.9

[Coly Li]

Hi Jens,

These two patches are posted for a while, and have reviewed by several
other developers. Could you please to take them for Linux v5.9 ?

Thanks in advance.

Coly Li
---

Coly Li (2):
  block: change REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL to be odd
    numbers
  block: improve discard bio alignment in __blkdev_issue_discard()

 block/blk-lib.c           | 25 +++++++++++++++++++++++--
 block/blk.h               | 14 ++++++++++++++
 include/linux/blk_types.h |  8 ++++----
 3 files changed, 41 insertions(+), 6 deletions(-)

-- 
2.26.2

[PATCH 1/2] block: change REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL to be odd numbers

[Coly Li]

Currently REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL are defined as
even numbers 6 and 8, such zone reset bios are treated as READ bios by
bio_data_dir(), which is obviously misleading.

The macro bio_data_dir() is defined in include/linux/bio.h as,
 55 #define bio_data_dir(bio) \
 56         (op_is_write(bio_op(bio)) ? WRITE : READ)

And op_is_write() is defined in include/linux/blk_types.h as,
397 static inline bool op_is_write(unsigned int op)
398 {
399         return (op & 1);
400 }

The convention of op_is_write() is when there is data transfer then the
op code should be odd number, and treat as a write op. bio_data_dir()
treats all bio direction as READ if op_is_write() reports false, and
WRITE if op_is_write() reports true.

Because REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL are even numbers,
although they don't transfer data but reporting them as READ bio by
bio_data_dir() is misleading and might be wrong. Because these two
commands will reset the writer pointers of the resetting zones, and all
content after the reset write pointer will be invalid and unaccessible,
obviously they are not READ bios in any means.

This patch changes REQ_OP_ZONE_RESET from 6 to 15, and changes
REQ_OP_ZONE_RESET_ALL from 8 to 17. Now bios with these two op code
can be treated as WRITE by bio_data_dir(). Although they don't transfer
data, now we keep them consistent with REQ_OP_DISCARD and
REQ_OP_WRITE_ZEROES with the ituition that they change on-media content
and should be WRITE request.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Jens Axboe <axboe@fb.com>
Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Cc: Keith Busch <kbusch@kernel.org>
Cc: Shaun Tancheff <shaun.tancheff@seagate.com>
---
 include/linux/blk_types.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index ccb895f911b1..447b46a0accf 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -300,12 +300,8 @@ enum req_opf {
 	REQ_OP_DISCARD		= 3,
 	/* securely erase sectors */
 	REQ_OP_SECURE_ERASE	= 5,
-	/* reset a zone write pointer */
-	REQ_OP_ZONE_RESET	= 6,
 	/* write the same sector many times */
 	REQ_OP_WRITE_SAME	= 7,
-	/* reset all the zone present on the device */
-	REQ_OP_ZONE_RESET_ALL	= 8,
 	/* write the zero filled sector many times */
 	REQ_OP_WRITE_ZEROES	= 9,
 	/* Open a zone */
@@ -316,6 +312,10 @@ enum req_opf {
 	REQ_OP_ZONE_FINISH	= 12,
 	/* write data at the current zone write pointer */
 	REQ_OP_ZONE_APPEND	= 13,
+	/* reset a zone write pointer */
+	REQ_OP_ZONE_RESET	= 15,
+	/* reset all the zone present on the device */
+	REQ_OP_ZONE_RESET_ALL	= 17,
 
 	/* SCSI passthrough using struct scsi_request */
 	REQ_OP_SCSI_IN		= 32,
-- 
2.26.2

Re: [PATCH 1/2] block: change REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL to be odd numbers

[Damien Le Moal]

On 2020/07/13 21:35, Coly Li wrote:
> Currently REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL are defined as
> even numbers 6 and 8, such zone reset bios are treated as READ bios by
> bio_data_dir(), which is obviously misleading.
> 
> The macro bio_data_dir() is defined in include/linux/bio.h as,
>  55 #define bio_data_dir(bio) \
>  56         (op_is_write(bio_op(bio)) ? WRITE : READ)
> 
> And op_is_write() is defined in include/linux/blk_types.h as,
> 397 static inline bool op_is_write(unsigned int op)
> 398 {
> 399         return (op & 1);
> 400 }
> 
> The convention of op_is_write() is when there is data transfer then the
> op code should be odd number, and treat as a write op. bio_data_dir()
> treats all bio direction as READ if op_is_write() reports false, and
> WRITE if op_is_write() reports true.
> 
> Because REQ_OP_ZONE_RESET and REQ_OP_ZONE_RESET_ALL are even numbers,
> although they don't transfer data but reporting them as READ bio by
> bio_data_dir() is misleading and might be wrong. Because these two
> commands will reset the writer pointers of the resetting zones, and all
> content after the reset write pointer will be invalid and unaccessible,
> obviously they are not READ bios in any means.
> 
> This patch changes REQ_OP_ZONE_RESET from 6 to 15, and changes
> REQ_OP_ZONE_RESET_ALL from 8 to 17. Now bios with these two op code
> can be treated as WRITE by bio_data_dir(). Although they don't transfer
> data, now we keep them consistent with REQ_OP_DISCARD and
> REQ_OP_WRITE_ZEROES with the ituition that they change on-media content

s/ituition/assumption

> and should be WRITE request.
> 
> Signed-off-by: Coly Li <colyli@suse.de>
> Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
> Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Hannes Reinecke <hare@suse.de>
> Cc: Jens Axboe <axboe@fb.com>
> Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
> Cc: Keith Busch <kbusch@kernel.org>
> Cc: Shaun Tancheff <shaun.tancheff@seagate.com>
> ---
>  include/linux/blk_types.h | 8 ++++----
>  1 file changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
> index ccb895f911b1..447b46a0accf 100644
> --- a/include/linux/blk_types.h
> +++ b/include/linux/blk_types.h
> @@ -300,12 +300,8 @@ enum req_opf {
>  	REQ_OP_DISCARD		= 3,
>  	/* securely erase sectors */
>  	REQ_OP_SECURE_ERASE	= 5,
> -	/* reset a zone write pointer */
> -	REQ_OP_ZONE_RESET	= 6,
>  	/* write the same sector many times */
>  	REQ_OP_WRITE_SAME	= 7,
> -	/* reset all the zone present on the device */
> -	REQ_OP_ZONE_RESET_ALL	= 8,
>  	/* write the zero filled sector many times */
>  	REQ_OP_WRITE_ZEROES	= 9,
>  	/* Open a zone */
> @@ -316,6 +312,10 @@ enum req_opf {
>  	REQ_OP_ZONE_FINISH	= 12,
>  	/* write data at the current zone write pointer */
>  	REQ_OP_ZONE_APPEND	= 13,
> +	/* reset a zone write pointer */
> +	REQ_OP_ZONE_RESET	= 15,
> +	/* reset all the zone present on the device */
> +	REQ_OP_ZONE_RESET_ALL	= 17,
>  
>  	/* SCSI passthrough using struct scsi_request */
>  	REQ_OP_SCSI_IN		= 32,
> 

Looks good to me. Zone reset is very similar to a discard, albeit stronger (zone
reset is not a hint). So defining these ops as having the same data dir makes sense.

Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>

-- 
Damien Le Moal
Western Digital Research

[PATCH 2/2] block: improve discard bio alignment in __blkdev_issue_discard()

[Coly Li]

This patch improves discard bio split for address and size alignment in
__blkdev_issue_discard(). The aligned discard bio may help underlying
device controller to perform better discard and internal garbage
collection, and avoid unnecessary internal fragment.

Current discard bio split algorithm in __blkdev_issue_discard() may have
non-discarded fregment on device even the discard bio LBA and size are
both aligned to device's discard granularity size.

Here is the example steps on how to reproduce the above problem.
- On a VMWare ESXi 6.5 update3 installation, create a 51GB virtual disk
  with thin mode and give it to a Linux virtual machine.
- Inside the Linux virtual machine, if the 50GB virtual disk shows up as
  /dev/sdb, fill data into the first 50GB by,
        # dd if=/dev/zero of=/dev/sdb bs=4096 count=13107200
- Discard the 50GB range from offset 0 on /dev/sdb,
        # blkdiscard /dev/sdb -o 0 -l 53687091200
- Observe the underlying mapping status of the device
        # sg_get_lba_status /dev/sdb -m 1048 --lba=0
  descriptor LBA: 0x0000000000000000  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000000000800  blocks: 16773120  deallocated
  descriptor LBA: 0x0000000000fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000001000000  blocks: 8386560  deallocated
  descriptor LBA: 0x00000000017ff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000001800000  blocks: 8386560  deallocated
  descriptor LBA: 0x0000000001fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000002000000  blocks: 8386560  deallocated
  descriptor LBA: 0x00000000027ff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000002800000  blocks: 8386560  deallocated
  descriptor LBA: 0x0000000002fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000003000000  blocks: 8386560  deallocated
  descriptor LBA: 0x00000000037ff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000003800000  blocks: 8386560  deallocated
  descriptor LBA: 0x0000000003fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000004000000  blocks: 8386560  deallocated
  descriptor LBA: 0x00000000047ff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000004800000  blocks: 8386560  deallocated
  descriptor LBA: 0x0000000004fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000005000000  blocks: 8386560  deallocated
  descriptor LBA: 0x00000000057ff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000005800000  blocks: 8386560  deallocated
  descriptor LBA: 0x0000000005fff800  blocks: 2048  mapped (or unknown)
  descriptor LBA: 0x0000000006000000  blocks: 6291456  deallocated
  descriptor LBA: 0x0000000006600000  blocks: 0  deallocated

Although the discard bio starts at LBA 0 and has 50<<30 bytes size which
are perfect aligned to the discard granularity, from the above list
these are many 1MB (2048 sectors) internal fragments exist unexpectedly.

The problem is in __blkdev_issue_discard(), an improper algorithm causes
an improper bio size which is not aligned.

 25 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 26                 sector_t nr_sects, gfp_t gfp_mask, int flags,
 27                 struct bio **biop)
 28 {
 29         struct request_queue *q = bdev_get_queue(bdev);
   [snipped]
 56
 57         while (nr_sects) {
 58                 sector_t req_sects = min_t(sector_t, nr_sects,
 59                                 bio_allowed_max_sectors(q));
 60
 61                 WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
 62
 63                 bio = blk_next_bio(bio, 0, gfp_mask);
 64                 bio->bi_iter.bi_sector = sector;
 65                 bio_set_dev(bio, bdev);
 66                 bio_set_op_attrs(bio, op, 0);
 67
 68                 bio->bi_iter.bi_size = req_sects << 9;
 69                 sector += req_sects;
 70                 nr_sects -= req_sects;
   [snipped]
 79         }
 80
 81         *biop = bio;
 82         return 0;
 83 }
 84 EXPORT_SYMBOL(__blkdev_issue_discard);

At line 58-59, to discard a 50GB range, req_sects is set as return value
of bio_allowed_max_sectors(q), which is 8388607 sectors. In the above
case, the discard granularity is 2048 sectors, although the start LBA
and discard length are aligned to discard granularity, req_sects never
has chance to be aligned to discard granularity. This is why there are
some still-mapped 2048 sectors fragment in every 4 or 8 GB range.

If req_sects at line 58 is set to a value aligned to discard_granularity
and close to UNIT_MAX, then all consequent split bios inside device
driver are (almostly) aligned to discard_granularity of the device
queue. The 2048 sectors still-mapped fragment will disappear.

This patch introduces bio_aligned_discard_max_sectors() to return the
the value which is aligned to q->limits.discard_granularity and closest
to UINT_MAX. Then this patch replaces bio_allowed_max_sectors() with
this new routine to decide a more proper split bio length.

But we still need to handle the situation when discard start LBA is not
aligned to q->limits.discard_granularity, otherwise even the length is
aligned, current code may still leave 2048 fragment around every 4GB
range. Therefore, to calculate req_sects, firstly the start LBA of
discard range is checked, if it is not aligned to discard granularity,
the first split location should make sure following bio has bi_sector
aligned to discard granularity. Then there won't be still-mapped
fragment in the middle of the discard range.

The above is how this patch improves discard bio alignment in
__blkdev_issue_discard(). Now with this patch, after discard with same
command line mentiond previously, sg_get_lba_status returns,
descriptor LBA: 0x0000000000000000  blocks: 106954752  deallocated
descriptor LBA: 0x0000000006600000  blocks: 0  deallocated

We an see there is no 2048 sectors segment anymore, everything is clean.

Reported-and-tested-by: Acshai Manoj <acshai.manoj@microfocus.com>
Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Xiao Ni <xni@redhat.com>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Enzo Matsumiya <ematsumiya@suse.com>
Cc: Jens Axboe <axboe@kernel.dk>
---
 block/blk-lib.c | 25 +++++++++++++++++++++++--
 block/blk.h     | 14 ++++++++++++++
 2 files changed, 37 insertions(+), 2 deletions(-)

diff --git a/block/blk-lib.c b/block/blk-lib.c
index 5f2c429d4378..7bffdee63a20 100644
--- a/block/blk-lib.c
+++ b/block/blk-lib.c
@@ -55,8 +55,29 @@ int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 		return -EINVAL;
 
 	while (nr_sects) {
-		sector_t req_sects = min_t(sector_t, nr_sects,
-				bio_allowed_max_sectors(q));
+		sector_t granularity_aligned_lba;
+		sector_t req_sects;
+
+		granularity_aligned_lba = round_up(sector,
+				q->limits.discard_granularity >> SECTOR_SHIFT);
+
+		/*
+		 * Check whether the discard bio starts at a discard_granularity
+		 * aligned LBA,
+		 * - If no: set (granularity_aligned_lba - sector) to bi_size of
+		 *   the first split bio, then the second bio will start at a
+		 *   discard_granularity aligned LBA.
+		 * - If yes: use bio_aligned_discard_max_sectors() as the max
+		 *   possible bi_size of the first split bio. Then when this bio
+		 *   is split in device drive, the split ones are very probably
+		 *   to be aligned to discard_granularity of the device's queue.
+		 */
+		if (granularity_aligned_lba == sector)
+			req_sects = min_t(sector_t, nr_sects,
+					  bio_aligned_discard_max_sectors(q));
+		else
+			req_sects = min_t(sector_t, nr_sects,
+					  granularity_aligned_lba - sector);
 
 		WARN_ON_ONCE((req_sects << 9) > UINT_MAX);
 
diff --git a/block/blk.h b/block/blk.h
index b5d1f0fc6547..a80738581f84 100644
--- a/block/blk.h
+++ b/block/blk.h
@@ -281,6 +281,20 @@ static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
 }
 
+/*
+ * The max bio size which is aligned to q->limits.discard_granularity. This
+ * is a hint to split large discard bio in generic block layer, then if device
+ * driver needs to split the discard bio into smaller ones, their bi_size can
+ * be very probably and easily aligned to discard_granularity of the device's
+ * queue.
+ */
+static inline unsigned int bio_aligned_discard_max_sectors(
+					struct request_queue *q)
+{
+	return round_down(UINT_MAX, q->limits.discard_granularity) >>
+			SECTOR_SHIFT;
+}
+
 /*
  * Internal io_context interface
  */
-- 
2.26.2