On 2018/04/16 18:04, Bart Van Assche wrote:
> Since SCSI scanning occurs asynchronously, since sd_revalidate_disk()
> is called from sd_probe_async() and since sd_revalidate_disk() calls
> sd_zbc_read_zones() it can happen that sd_zbc_read_zones() is called
> concurrently with blkdev_report_zones() and/or blkdev_reset_zones().
> That can cause these functions to fail with -EIO because
> sd_zbc_read_zones() e.g. sets q->nr_zones to zero before restoring it
> to the actual value, even if no drive characteristics have changed.
> Avoid that this can happen by making the following changes:
> - Protect the code that updates zone information with blk_queue_enter()
> and blk_queue_exit().
> - Modify sd_zbc_setup_seq_zones_bitmap() and sd_zbc_setup() such that
> these functions do not modify struct scsi_disk before all zone
> information has been obtained.
>
> Note: since commit 055f6e18e08f ("block: Make q_usage_counter also
> track legacy requests"; kernel v4.15) the request queue freezing
> mechanism also affects legacy request queues.
>
> Fixes: 89d947561077 ("sd: Implement support for ZBC devices")
> Signed-off-by: Bart Van Assche <[email protected]>
> Cc: Jens Axboe <[email protected]>
> Cc: Damien Le Moal <[email protected]>
> Cc: Christoph Hellwig <[email protected]>
> Cc: Hannes Reinecke <[email protected]>
> Cc: [email protected] # v4.10
> ---
> drivers/scsi/sd_zbc.c | 140
> +++++++++++++++++++++++++++++--------------------
> include/linux/blkdev.h | 5 ++
> 2 files changed, 87 insertions(+), 58 deletions(-)
>
> diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
> index 2d0c06f7db3e..323e3dc4bc59 100644
> --- a/drivers/scsi/sd_zbc.c
> +++ b/drivers/scsi/sd_zbc.c
> @@ -390,8 +390,10 @@ static int sd_zbc_check_capacity(struct scsi_disk *sdkp,
> unsigned char *buf)
> *
> * Check that all zones of the device are equal. The last zone can however
> * be smaller. The zone size must also be a power of two number of LBAs.
> + *
> + * Returns the zone size in bytes upon success or an error code upon failure.
> */
> -static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
> +static s64 sd_zbc_check_zone_size(struct scsi_disk *sdkp)
> {
> u64 zone_blocks = 0;
> sector_t block = 0;
> @@ -402,8 +404,6 @@ static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
> int ret;
> u8 same;
>
> - sdkp->zone_blocks = 0;
> -
> /* Get a buffer */
> buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
> if (!buf)
> @@ -435,16 +435,17 @@ static int sd_zbc_check_zone_size(struct scsi_disk
> *sdkp)
>
> /* Parse zone descriptors */
> while (rec < buf + buf_len) {
> - zone_blocks = get_unaligned_be64(&rec[8]);
> - if (sdkp->zone_blocks == 0) {
> - sdkp->zone_blocks = zone_blocks;
> - } else if (zone_blocks != sdkp->zone_blocks &&
> - (block + zone_blocks < sdkp->capacity
> - || zone_blocks > sdkp->zone_blocks)) {
> - zone_blocks = 0;
> + u64 this_zone_blocks = get_unaligned_be64(&rec[8]);
> +
> + if (zone_blocks == 0) {
> + zone_blocks = this_zone_blocks;
> + } else if (this_zone_blocks != zone_blocks &&
> + (block + this_zone_blocks < sdkp->capacity
> + || this_zone_blocks > zone_blocks)) {
> + this_zone_blocks = 0;
> goto out;
> }
> - block += zone_blocks;
> + block += this_zone_blocks;
> rec += 64;
> }
>
> @@ -457,8 +458,6 @@ static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
>
> } while (block < sdkp->capacity);
>
> - zone_blocks = sdkp->zone_blocks;
> -
> out:
> if (!zone_blocks) {
> if (sdkp->first_scan)
> @@ -478,8 +477,7 @@ static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
> "Zone size too large\n");
> ret = -ENODEV;
> } else {
> - sdkp->zone_blocks = zone_blocks;
> - sdkp->zone_shift = ilog2(zone_blocks);
> + ret = zone_blocks;
> }
>
> out_free:
> @@ -490,15 +488,14 @@ static int sd_zbc_check_zone_size(struct scsi_disk
> *sdkp)
>
> /**
> * sd_zbc_alloc_zone_bitmap - Allocate a zone bitmap (one bit per zone).
> - * @sdkp: The disk of the bitmap
> + * @nr_zones: Number of zones to allocate space for.
> + * @numa_node: NUMA node to allocate the memory from.
> */
> -static inline unsigned long *sd_zbc_alloc_zone_bitmap(struct scsi_disk *sdkp)
> +static inline unsigned long *
> +sd_zbc_alloc_zone_bitmap(u32 nr_zones, int numa_node)
> {
> - struct request_queue *q = sdkp->disk->queue;
> -
> - return kzalloc_node(BITS_TO_LONGS(sdkp->nr_zones)
> - * sizeof(unsigned long),
> - GFP_KERNEL, q->node);
> + return kzalloc_node(BITS_TO_LONGS(nr_zones) * sizeof(unsigned long),
> + GFP_KERNEL, numa_node);
> }
>
> /**
> @@ -506,6 +503,7 @@ static inline unsigned long
> *sd_zbc_alloc_zone_bitmap(struct scsi_disk *sdkp)
> * @sdkp: disk used
> * @buf: report reply buffer
> * @buflen: length of @buf
> + * @zone_shift: logarithm base 2 of the number of blocks in a zone
> * @seq_zones_bitmap: bitmap of sequential zones to set
> *
> * Parse reported zone descriptors in @buf to identify sequential zones and
> @@ -515,7 +513,7 @@ static inline unsigned long
> *sd_zbc_alloc_zone_bitmap(struct scsi_disk *sdkp)
> * Return the LBA after the last zone reported.
> */
> static sector_t sd_zbc_get_seq_zones(struct scsi_disk *sdkp, unsigned char
> *buf,
> - unsigned int buflen,
> + unsigned int buflen, u32 zone_shift,
> unsigned long *seq_zones_bitmap)
> {
> sector_t lba, next_lba = sdkp->capacity;
> @@ -534,7 +532,7 @@ static sector_t sd_zbc_get_seq_zones(struct scsi_disk
> *sdkp, unsigned char *buf,
> if (type != ZBC_ZONE_TYPE_CONV &&
> cond != ZBC_ZONE_COND_READONLY &&
> cond != ZBC_ZONE_COND_OFFLINE)
> - set_bit(lba >> sdkp->zone_shift, seq_zones_bitmap);
> + set_bit(lba >> zone_shift, seq_zones_bitmap);
> next_lba = lba + get_unaligned_be64(&rec[8]);
> rec += 64;
> }
> @@ -543,12 +541,16 @@ static sector_t sd_zbc_get_seq_zones(struct scsi_disk
> *sdkp, unsigned char *buf,
> }
>
> /**
> - * sd_zbc_setup_seq_zones_bitmap - Initialize the disk seq zone bitmap.
> + * sd_zbc_setup_seq_zones_bitmap - Initialize a seq zone bitmap.
> * @sdkp: target disk
> + * @zone_shift: logarithm base 2 of the number of blocks in a zone
> + * @nr_zones: number of zones to set up a seq zone bitmap for
> *
> * Allocate a zone bitmap and initialize it by identifying sequential zones.
> */
> -static int sd_zbc_setup_seq_zones_bitmap(struct scsi_disk *sdkp)
> +static unsigned long *
> +sd_zbc_setup_seq_zones_bitmap(struct scsi_disk *sdkp, u32 zone_shift,
> + u32 nr_zones)
> {
> struct request_queue *q = sdkp->disk->queue;
> unsigned long *seq_zones_bitmap;
> @@ -556,9 +558,9 @@ static int sd_zbc_setup_seq_zones_bitmap(struct scsi_disk
> *sdkp)
> unsigned char *buf;
> int ret = -ENOMEM;
>
> - seq_zones_bitmap = sd_zbc_alloc_zone_bitmap(sdkp);
> + seq_zones_bitmap = sd_zbc_alloc_zone_bitmap(nr_zones, q->node);
> if (!seq_zones_bitmap)
> - return -ENOMEM;
> + return ERR_PTR(-ENOMEM);
>
> buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
> if (!buf)
> @@ -569,7 +571,7 @@ static int sd_zbc_setup_seq_zones_bitmap(struct scsi_disk
> *sdkp)
> if (ret)
> goto out;
> lba = sd_zbc_get_seq_zones(sdkp, buf, SD_ZBC_BUF_SIZE,
> - seq_zones_bitmap);
> + zone_shift, seq_zones_bitmap);
> }
>
> if (lba != sdkp->capacity) {
> @@ -581,12 +583,9 @@ static int sd_zbc_setup_seq_zones_bitmap(struct
> scsi_disk *sdkp)
> kfree(buf);
> if (ret) {
> kfree(seq_zones_bitmap);
> - return ret;
> + return ERR_PTR(ret);
> }
> -
> - q->seq_zones_bitmap = seq_zones_bitmap;
> -
> - return 0;
> + return seq_zones_bitmap;
> }
>
> static void sd_zbc_cleanup(struct scsi_disk *sdkp)
> @@ -602,44 +601,64 @@ static void sd_zbc_cleanup(struct scsi_disk *sdkp)
> q->nr_zones = 0;
> }
>
> -static int sd_zbc_setup(struct scsi_disk *sdkp)
> +static int sd_zbc_setup(struct scsi_disk *sdkp, u32 zone_blocks)
> {
> struct request_queue *q = sdkp->disk->queue;
> + u32 zone_shift = ilog2(zone_blocks);
> + u32 nr_zones;
> int ret;
>
> - /* READ16/WRITE16 is mandatory for ZBC disks */
> - sdkp->device->use_16_for_rw = 1;
> - sdkp->device->use_10_for_rw = 0;
> -
> /* chunk_sectors indicates the zone size */
> - blk_queue_chunk_sectors(sdkp->disk->queue,
> - logical_to_sectors(sdkp->device, sdkp->zone_blocks));
> - sdkp->nr_zones =
> - round_up(sdkp->capacity, sdkp->zone_blocks) >> sdkp->zone_shift;
> + blk_queue_chunk_sectors(q,
> + logical_to_sectors(sdkp->device, zone_blocks));
> + nr_zones = round_up(sdkp->capacity, zone_blocks) >> zone_shift;
>
> /*
> * Initialize the device request queue information if the number
> * of zones changed.
> */
> - if (sdkp->nr_zones != q->nr_zones) {
> -
> - sd_zbc_cleanup(sdkp);
> -
> - q->nr_zones = sdkp->nr_zones;
> - if (sdkp->nr_zones) {
> - q->seq_zones_wlock = sd_zbc_alloc_zone_bitmap(sdkp);
> - if (!q->seq_zones_wlock) {
> + if (nr_zones != sdkp->nr_zones || nr_zones != q->nr_zones) {
> + unsigned long *seq_zones_wlock = NULL, *seq_zones_bitmap = NULL;
> + size_t zone_bitmap_size;
> +
> + if (nr_zones) {
> + seq_zones_wlock = sd_zbc_alloc_zone_bitmap(nr_zones,
> + q->node);
> + if (!seq_zones_wlock) {
> ret = -ENOMEM;
> goto err;
> }
>
> - ret = sd_zbc_setup_seq_zones_bitmap(sdkp);
> - if (ret) {
> - sd_zbc_cleanup(sdkp);
> + seq_zones_bitmap = sd_zbc_setup_seq_zones_bitmap(sdkp,
> + zone_shift, nr_zones);
> + if (IS_ERR(seq_zones_bitmap)) {
> + ret = PTR_ERR(seq_zones_bitmap);
> + kfree(seq_zones_wlock);
> goto err;
> }
> }
> -
> + zone_bitmap_size = BITS_TO_LONGS(nr_zones) *
> + sizeof(unsigned long);
> + blk_mq_freeze_queue(q);
> + if (q->nr_zones != nr_zones) {
> + /* READ16/WRITE16 is mandatory for ZBC disks */
> + sdkp->device->use_16_for_rw = 1;
> + sdkp->device->use_10_for_rw = 0;
> +
> + sdkp->zone_blocks = zone_blocks;
> + sdkp->zone_shift = zone_shift;
> + sdkp->nr_zones = nr_zones;
> + q->nr_zones = nr_zones;
> + swap(q->seq_zones_wlock, seq_zones_wlock);
> + swap(q->seq_zones_bitmap, seq_zones_bitmap);
> + } else if (memcmp(q->seq_zones_bitmap, seq_zones_bitmap,
> + zone_bitmap_size) != 0) {
> + memcpy(q->seq_zones_bitmap, seq_zones_bitmap,
> + zone_bitmap_size);
> + }
> + blk_mq_unfreeze_queue(q);
> + kfree(seq_zones_wlock);
> + kfree(seq_zones_bitmap);
> }
>
> return 0;
> @@ -651,6 +670,7 @@ static int sd_zbc_setup(struct scsi_disk *sdkp)
>
> int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned char *buf)
> {
> + int64_t zone_blocks;
> int ret;
>
> if (!sd_is_zoned(sdkp))
> @@ -687,12 +707,16 @@ int sd_zbc_read_zones(struct scsi_disk *sdkp, unsigned
> char *buf)
> * Check zone size: only devices with a constant zone size (except
> * an eventual last runt zone) that is a power of 2 are supported.
> */
> - ret = sd_zbc_check_zone_size(sdkp);
> - if (ret)
> + zone_blocks = sd_zbc_check_zone_size(sdkp);
> + ret = -EFBIG;
> + if (zone_blocks != (u32)zone_blocks)
> + goto err;
> + ret = zone_blocks;
> + if (ret < 0)
> goto err;
>
> /* The drive satisfies the kernel restrictions: set it up */
> - ret = sd_zbc_setup(sdkp);
> + ret = sd_zbc_setup(sdkp, zone_blocks);
> if (ret)
> goto err;
>
> diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
> index 9af3e0f430bc..21e21f273a21 100644
> --- a/include/linux/blkdev.h
> +++ b/include/linux/blkdev.h
> @@ -605,6 +605,11 @@ struct request_queue {
> * initialized by the low level device driver (e.g. scsi/sd.c).
> * Stacking drivers (device mappers) may or may not initialize
> * these fields.
> + *
> + * Reads of this information must be protected with blk_queue_enter() /
> + * blk_queue_exit(). Modifying this information is only allowed while
> + * no requests are being processed. See also blk_mq_freeze_queue() and
> + * blk_mq_unfreeze_queue().
> */
> unsigned int nr_zones;
> unsigned long *seq_zones_bitmap;
>
Reviewed-by: Damien Le Moal <[email protected]>
Note: This will not apply as is for stable kernels 4.10 to 4.15 since these do
not have zone information attached to the request queue.
--
Damien Le Moal
Western Digital Research
