]> git.baikalelectronics.ru Git - kernel.git/commitdiff
scsi: sd: sd_zbc: Hide gap zones
authorDamien Le Moal <damien.lemoal@opensource.wdc.com>
Thu, 21 Apr 2022 18:30:20 +0000 (11:30 -0700)
committerMartin K. Petersen <martin.petersen@oracle.com>
Tue, 26 Apr 2022 03:23:05 +0000 (23:23 -0400)
ZBC-2 allows host-managed disks to report gap zones. This allow zoned disks
to report an offset between data zone starts that is a power of two even if
the number of logical blocks with data per zone is not a power of two.

Another new feature in ZBC-2 is support for constant zone starting LBA
offsets. For zoned disks that report a constant zone starting LBA offset,
hide the gap zones from the block layer. Report the offset between data
zone starts as zone size and report the number of logical blocks with data
per zone as the zone capacity.

Link: https://lore.kernel.org/r/20220421183023.3462291-7-bvanassche@acm.org
Acked-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
[ bvanassche: Reworked this patch ]
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/sd.h
drivers/scsi/sd_zbc.c
include/scsi/scsi_proto.h

index 47434f905b0a7b08dca3caf09da28f273e20d29d..d249933ba69e01bc6c5160b7051d3946871ceafb 100644 (file)
@@ -99,6 +99,11 @@ struct scsi_disk {
        u32             zones_optimal_open;
        u32             zones_optimal_nonseq;
        u32             zones_max_open;
+       /*
+        * Either zero or a power of two. If not zero it means that the offset
+        * between zone starting LBAs is constant.
+        */
+       u32             zone_starting_lba_gran;
        u32             *zones_wp_offset;
        spinlock_t      zones_wp_offset_lock;
        u32             *rev_wp_offset;
index c53e166362b9d38b61c8bcbba6dd6e6b8953efee..5b9fad70aa88a4f1f7a9ea67312d98eb17eb19de 100644 (file)
@@ -50,6 +50,12 @@ static unsigned int sd_zbc_get_zone_wp_offset(struct blk_zone *zone)
        }
 }
 
+/* Whether or not a SCSI zone descriptor describes a gap zone. */
+static bool sd_zbc_is_gap_zone(const u8 buf[64])
+{
+       return (buf[0] & 0xf) == ZBC_ZONE_TYPE_GAP;
+}
+
 /**
  * sd_zbc_parse_report - Parse a SCSI zone descriptor
  * @sdkp: SCSI disk pointer.
@@ -69,8 +75,12 @@ static int sd_zbc_parse_report(struct scsi_disk *sdkp, const u8 buf[64],
 {
        struct scsi_device *sdp = sdkp->device;
        struct blk_zone zone = { 0 };
+       sector_t start_lba, gran;
        int ret;
 
+       if (WARN_ON_ONCE(sd_zbc_is_gap_zone(buf)))
+               return -EINVAL;
+
        zone.type = buf[0] & 0x0f;
        zone.cond = (buf[1] >> 4) & 0xf;
        if (buf[1] & 0x01)
@@ -78,9 +88,27 @@ static int sd_zbc_parse_report(struct scsi_disk *sdkp, const u8 buf[64],
        if (buf[1] & 0x02)
                zone.non_seq = 1;
 
-       zone.len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
-       zone.capacity = zone.len;
-       zone.start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
+       start_lba = get_unaligned_be64(&buf[16]);
+       zone.start = logical_to_sectors(sdp, start_lba);
+       zone.capacity = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
+       zone.len = zone.capacity;
+       if (sdkp->zone_starting_lba_gran) {
+               gran = logical_to_sectors(sdp, sdkp->zone_starting_lba_gran);
+               if (zone.len > gran) {
+                       sd_printk(KERN_ERR, sdkp,
+                                 "Invalid zone at LBA %llu with capacity %llu and length %llu; granularity = %llu\n",
+                                 start_lba,
+                                 sectors_to_logical(sdp, zone.capacity),
+                                 sectors_to_logical(sdp, zone.len),
+                                 sectors_to_logical(sdp, gran));
+                       return -EINVAL;
+               }
+               /*
+                * Use the starting LBA granularity instead of the zone length
+                * obtained from the REPORT ZONES command.
+                */
+               zone.len = gran;
+       }
        if (zone.cond == ZBC_ZONE_COND_FULL)
                zone.wp = zone.start + zone.len;
        else
@@ -227,6 +255,7 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
        sector_t lba = sectors_to_logical(sdkp->device, sector);
        unsigned int nr, i;
        unsigned char *buf;
+       u64 zone_length, start_lba;
        size_t offset, buflen = 0;
        int zone_idx = 0;
        int ret;
@@ -255,14 +284,36 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 
                for (i = 0; i < nr && zone_idx < nr_zones; i++) {
                        offset += 64;
+                       start_lba = get_unaligned_be64(&buf[offset + 16]);
+                       zone_length = get_unaligned_be64(&buf[offset + 8]);
+                       if ((zone_idx == 0 &&
+                           (lba < start_lba ||
+                            lba >= start_lba + zone_length)) ||
+                           (zone_idx > 0 && start_lba != lba) ||
+                           start_lba + zone_length < start_lba) {
+                               sd_printk(KERN_ERR, sdkp,
+                                         "Zone %d at LBA %llu is invalid: %llu + %llu\n",
+                                         zone_idx, lba, start_lba, zone_length);
+                               ret = -EINVAL;
+                               goto out;
+                       }
+                       lba = start_lba + zone_length;
+                       if (sd_zbc_is_gap_zone(&buf[offset])) {
+                               if (sdkp->zone_starting_lba_gran)
+                                       continue;
+                               sd_printk(KERN_ERR, sdkp,
+                                         "Gap zone without constant LBA offsets\n");
+                               ret = -EINVAL;
+                               goto out;
+                       }
+
                        ret = sd_zbc_parse_report(sdkp, buf + offset, zone_idx,
                                                  cb, data);
                        if (ret)
                                goto out;
+
                        zone_idx++;
                }
-
-               lba += sdkp->zone_info.zone_blocks * i;
        }
 
        ret = zone_idx;
@@ -579,6 +630,7 @@ unsigned int sd_zbc_complete(struct scsi_cmnd *cmd, unsigned int good_bytes,
 static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
                                              unsigned char *buf)
 {
+       u64 zone_starting_lba_gran;
 
        if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
                sd_printk(KERN_NOTICE, sdkp,
@@ -600,6 +652,29 @@ static int sd_zbc_check_zoned_characteristics(struct scsi_disk *sdkp,
        sdkp->zones_optimal_open = 0;
        sdkp->zones_optimal_nonseq = 0;
        sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
+       /* Check zone alignment method */
+       switch (buf[23] & 0xf) {
+       case 0:
+       case ZBC_CONSTANT_ZONE_LENGTH:
+               /* Use zone length */
+               break;
+       case ZBC_CONSTANT_ZONE_START_OFFSET:
+               zone_starting_lba_gran = get_unaligned_be64(&buf[24]);
+               if (zone_starting_lba_gran == 0 ||
+                   !is_power_of_2(zone_starting_lba_gran) ||
+                   logical_to_sectors(sdkp->device, zone_starting_lba_gran) >
+                   UINT_MAX) {
+                       sd_printk(KERN_ERR, sdkp,
+                                 "Invalid zone starting LBA granularity %llu\n",
+                                 zone_starting_lba_gran);
+                       return -ENODEV;
+               }
+               sdkp->zone_starting_lba_gran = zone_starting_lba_gran;
+               break;
+       default:
+               sd_printk(KERN_ERR, sdkp, "Invalid zone alignment method\n");
+               return -ENODEV;
+       }
 
        /*
         * Check for unconstrained reads: host-managed devices with
@@ -654,14 +729,18 @@ static int sd_zbc_check_capacity(struct scsi_disk *sdkp, unsigned char *buf,
                }
        }
 
-       /* Get the size of the first reported zone */
-       rec = buf + 64;
-       zone_blocks = get_unaligned_be64(&rec[8]);
-       if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
-               if (sdkp->first_scan)
-                       sd_printk(KERN_NOTICE, sdkp,
-                                 "Zone size too large\n");
-               return -EFBIG;
+       if (sdkp->zone_starting_lba_gran == 0) {
+               /* Get the size of the first reported zone */
+               rec = buf + 64;
+               zone_blocks = get_unaligned_be64(&rec[8]);
+               if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
+                       if (sdkp->first_scan)
+                               sd_printk(KERN_NOTICE, sdkp,
+                                         "Zone size too large\n");
+                       return -EFBIG;
+               }
+       } else {
+               zone_blocks = sdkp->zone_starting_lba_gran;
        }
 
        if (!is_power_of_2(zone_blocks)) {
index f017843a8124fe75c0ab99d3d3045873c75086f5..c03e35fc382cde37fd9f268bd7b04f3903632bc2 100644 (file)
@@ -307,7 +307,9 @@ enum zbc_zone_type {
        ZBC_ZONE_TYPE_CONV              = 0x1,
        ZBC_ZONE_TYPE_SEQWRITE_REQ      = 0x2,
        ZBC_ZONE_TYPE_SEQWRITE_PREF     = 0x3,
-       /* 0x4 to 0xf are reserved */
+       ZBC_ZONE_TYPE_SEQ_OR_BEFORE_REQ = 0x4,
+       ZBC_ZONE_TYPE_GAP               = 0x5,
+       /* 0x6 to 0xf are reserved */
 };
 
 /* Zone conditions of REPORT ZONES zone descriptors */
@@ -323,6 +325,11 @@ enum zbc_zone_cond {
        ZBC_ZONE_COND_OFFLINE           = 0xf,
 };
 
+enum zbc_zone_alignment_method {
+       ZBC_CONSTANT_ZONE_LENGTH        = 0x1,
+       ZBC_CONSTANT_ZONE_START_OFFSET  = 0x8,
+};
+
 /* Version descriptor values for INQUIRY */
 enum scsi_version_descriptor {
        SCSI_VERSION_DESCRIPTOR_FCP4    = 0x0a40,