[PATCH 5.15.y] Revert "scsi: sd: sd_zbc: Use logical blocks as unit when querying zones"

[Simon Liebold <simonlie@amazon.de>]

This reverts commit 9c78f3cf7785f3de5afb3d632f0d81c2cb0e9020, the 5.15.y
backport of upstream 43af5da09efb.

43af5da09efb is a cleanup ("slightly simplifies sd_zbc_report_zones()"),
not a fix. It entered 5.15.y only as a Stable-dep-of: 93dde0bf2f39
("scsi: scsi_debug: Fix REPORT ZONES alloc_len underflow OOB write"),
which touches only scsi_debug.c and does not actually depend on it.

On its own it zeroes the capacity of host-managed SMR drives at boot. It
changed the REPORT ZONES loop advance from

	sector += sd_zbc_zone_sectors(sdkp) * i;	/* sector_t, 64-bit */
to
	lba += sdkp->zone_blocks * i;			/* u32, wraps */

(zone_info.zone_blocks on 5.15.y after bffcb7d6a287). zone_blocks is
u32, so the product is computed in 32 bits and wraps before being
widened into the 64-bit lba. A drive returning >= 8192 zones in one
REPORT ZONES (~2 TiB+ at 256 MiB zones, i.e. any modern HM-SMR part)
overflows on the first iteration; the next batch is requested at a
wrapped LBA, the block layer rejects the backward jump, and sd truncates
the device to zero:

	sdX: Zone gap at sectors 18219532288..1039663104
	sdX: failed to revalidate zones

(524288 * 34751 = 18219532288, mod 2^32 = 1039663104.)

Mainline never shipped this: in the same series, patch 7 c976e588b34e
("scsi: sd: sd_zbc: Hide gap zones") rewrote the loop to advance per
descriptor from 64-bit device values. 5.15.y took patch 4 and the
intervening 628617be8968 but not c976e588b34e, keeping the broken
intermediate state.

Reverting is preferred over pulling in c976e588b34e: it is a ZBC-2
feature (gap zones, constant zone-starting-LBA granularity) adding ~130
lines across sd_zbc.c, sd.h and include/scsi/scsi_proto.h. It gains
nothing for the constant-zone-size HM-SMR drives 5.15.y serves, which
the restored sector_t advance already handles correctly.

The revert conflicted with bffcb7d6a287 (which renamed the field on the
advance line); resolved by keeping the restored "sector +=
sd_zbc_zone_sectors(sdkp) * i", which reads zone_info.zone_blocks via
logical_to_sectors() and is 64-bit. 93dde0bf2f39 stays in place and is
unaffected.

Signed-off-by: Simon Liebold <simonlie@amazon.de>
---
 drivers/scsi/sd_zbc.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/drivers/scsi/sd_zbc.c b/drivers/scsi/sd_zbc.c
index 90d0f8ddd179c..d84a9c1d76c5a 100644
--- a/drivers/scsi/sd_zbc.c
+++ b/drivers/scsi/sd_zbc.c
@@ -223,7 +223,7 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 			unsigned int nr_zones, report_zones_cb cb, void *data)
 {
 	struct scsi_disk *sdkp = scsi_disk(disk);
-	sector_t lba = sectors_to_logical(sdkp->device, sector);
+	sector_t capacity = logical_to_sectors(sdkp->device, sdkp->capacity);
 	unsigned int nr, i;
 	unsigned char *buf;
 	size_t offset, buflen = 0;
@@ -234,7 +234,7 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 		/* Not a zoned device */
 		return -EOPNOTSUPP;
 
-	if (!sdkp->capacity)
+	if (!capacity)
 		/* Device gone or invalid */
 		return -ENODEV;
 
@@ -242,8 +242,9 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 	if (!buf)
 		return -ENOMEM;
 
-	while (zone_idx < nr_zones && lba < sdkp->capacity) {
-		ret = sd_zbc_do_report_zones(sdkp, buf, buflen, lba, true);
+	while (zone_idx < nr_zones && sector < capacity) {
+		ret = sd_zbc_do_report_zones(sdkp, buf, buflen,
+				sectors_to_logical(sdkp->device, sector), true);
 		if (ret)
 			goto out;
 
@@ -261,7 +262,7 @@ int sd_zbc_report_zones(struct gendisk *disk, sector_t sector,
 			zone_idx++;
 		}
 
-		lba += sdkp->zone_info.zone_blocks * i;
+		sector += sd_zbc_zone_sectors(sdkp) * i;
 	}
 
 	ret = zone_idx;

base-commit: 0248c33e835ecbec3a591f93fdaae53f7b90a4d6
-- 
2.50.1