nvme: core: reject invalid LBA data size from Identify Namespace
PATCH 1/1 nvme: core: reject invalid LBA data size from Identify Namespace
nvme_update_ns_info_block() trusts id->lbaf[lbaf].ds from the controller and assigns it directly to ns->head->lba_shift without bounds checking. nvme_lba_to_sect() then does:
return lba << (head->lba_shift - SECTOR_SHIFT);
When called with lba = le64_to_cpu(id->nsze) to compute the device capacity, an attacker-controlled controller can choose ds < 9 or a combination of (ds, nsze) that makes the left shift overflow sector_t. The former is a C undefined behaviour that UBSAN reports as a BUG; the latter silently yields a bogus capacity that the block layer then trusts for bounds checking.
Validate ds against SECTOR_SHIFT and use check_shl_overflow() to compute capacity so that any (ds, nsze) combination that would overflow sector_t is rejected. The namespace is skipped with -EIO instead of crashing the kernel. This is reachable by a malicious NVMe device, a buggy firmware, or an attacker-controlled NVMe-oF target.
Stack trace (UBSAN, ds < 9 variant):
RIP: nvme_lba_to_sect drivers/nvme/host/nvme.h:699 [inline] RIP: nvme_update_ns_info_block.cold+0x5/0x7 Call Trace: nvme_update_ns_info+0x175/0xd90 drivers/nvme/host/core.c:2467 nvme_validate_ns drivers/nvme/host/core.c:4299 [inline] nvme_scan_ns drivers/nvme/host/core.c:4350 nvme_scan_ns_async+0xa5/0xe0 drivers/nvme/host/core.c:4383 async_run_entry_fn process_one_work worker_thread kthread
Found by Syzkaller.
--- drivers/nvme/host/core.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/nvme/host/core.c +11 -1
@@ -2407,9 +2407,19 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns, lim =queue_limits_start_update (ns->disk->queue); memflags =blk_mq_freeze_queue (ns->disk->queue); +if (id->lbaf[lbaf].ds <SECTOR_SHIFT || +check_shl_overflow (le64_to_cpu (id->nsze), + id->lbaf[lbaf].ds -SECTOR_SHIFT , + &capacity)) { +dev_warn_once (ns->ctrl->device, +"invalid LBA data size %u, skipping namespace\n" , + id->lbaf[lbaf].ds); + ret = -EIO ; +blk_mq_unfreeze_queue (ns->disk->queue, memflags); + goto out; + } ns->head->lba_shift = id->lbaf[lbaf].ds; ns->head->nuse =le64_to_cpu (id->nuse); - capacity =nvme_lba_to_sect (ns->head,le64_to_cpu (id->nsze));nvme_set_ctrl_limits (ns->ctrl, &lim, false);nvme_configure_metadata (ns->ctrl, ns->head, id, nvm, info);nvme_set_chunk_sectors (ns, id, &lim);
Signature · 1 line
2.43.0
Quoted 1 line · Chao Shi
Fixes 9419e71b8d67 ("nvme: move ns id info to struct nvme_ns_head")
This fixes tag is wrong. The above commit doesn't touch the code line in question.
Probably a5b1cd61820e ("nvme: move a few things out of nvme_update_disk_info") is the better choice here.
Keith Busch
maintainerQuoted 8 lines · Daniel Wagner
Quoted 1 line · Chao Shi
Fixes 9419e71b8d67 ("nvme: move ns id info to struct nvme_ns_head")This fixes tag is wrong. The above commit doesn't touch the code line in question.
Probably
a5b1cd61820e("nvme: move a few things out ofnvme_update_disk_info") is the better choice here.
We've never validated the 'ds' field though, so not sure anything introduced a regression here that warrants a fixing tag.
Hi Daniel,
Thanks for the catch on the Fixes tag.
Quoted 4 lines
This fixes tag is wrong. The above commit doesn't touch the code line in question.
Probablya5b1cd61820e("nvme: move a few things out ofnvme_update_disk_info") is the better choice here.
Keith subsequently pointed out that ds has never been validated upstream, so no single commit introduced a regression to fix. I dropped the Fixes tag entirely in v2 (and v3 keeps it dropped).
Thanks,
Chao
Quoted 9 lines · Daniel Wagner
Quoted 1 line · Chao Shi
Fixes 9419e71b8d67 ("nvme: move ns id info to struct nvme_ns_head")This fixes tag is wrong. The above commit doesn't touch the code line in question.
Probably
a5b1cd61820e("nvme: move a few things out ofnvme_update_disk_info") is the better choice here.
Quoted 50 lines · Chao Shi
nvme_update_ns_info_block()trusts id->lbaf[lbaf].ds from the controller and assigns it directly tons->head->lba_shiftwithout bounds checking.nvme_lba_to_sect()then does:return lba << (head->lba_shift - SECTOR_SHIFT);When called with lba =
le64_to_cpu(id->nsze)to compute the device capacity, an attacker-controlled controller can choose ds < 9 or a combination of (ds, nsze) that makes the left shift overflowsector_t. The former is a C undefined behaviour that UBSAN reports as a BUG; the latter silently yields a bogus capacity that the block layer then trusts for bounds checking.Validate ds against
SECTOR_SHIFTand usecheck_shl_overflow()to compute capacity so that any (ds, nsze) combination that would overflowsector_tis rejected. The namespace is skipped with-EIOinstead of crashing the kernel. This is reachable by a malicious NVMe device, a buggy firmware, or an attacker-controlled NVMe-oF target.Stack trace (UBSAN, ds < 9 variant):
RIP: nvme_lba_to_sect drivers/nvme/host/nvme.h:699 [inline] RIP: nvme_update_ns_info_block.cold+0x5/0x7 Call Trace: nvme_update_ns_info+0x175/0xd90 drivers/nvme/host/core.c:2467 nvme_validate_ns drivers/nvme/host/core.c:4299 [inline] nvme_scan_ns drivers/nvme/host/core.c:4350 nvme_scan_ns_async+0xa5/0xe0 drivers/nvme/host/core.c:4383 async_run_entry_fn process_one_work worker_thread kthreadFound by Syzkaller.
Fixes 9419e71b8d67 ("nvme: move ns id info to struct nvme_ns_head")Acked-by Sungwoo Kim <[email protected]>Acked-by Dave Tian <[email protected]>Acked-by Weidong Zhu <[email protected]>Signed-off-by Chao Shi <[email protected]>--- drivers/nvme/host/core.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)drivers/nvme/host/core.c +0
-0
drivers/nvme/host/core.c static int nvme_update_ns_info_block(struct nvme_ns *ns, · line 2416
Quoted 10 lines
drivers/nvme/host/core.c +7
-0@@ -2407,9 +2407,19 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns, lim =queue_limits_start_update (ns->disk->queue); memflags =blk_mq_freeze_queue (ns->disk->queue); +if (id->lbaf[lbaf].ds <SECTOR_SHIFT || +check_shl_overflow (le64_to_cpu (id->nsze), + id->lbaf[lbaf].ds -SECTOR_SHIFT , + &capacity)) { +dev_warn_once (ns->ctrl->device, +"invalid LBA data size %u, skipping namespace\n" , + id->lbaf[lbaf].ds);
Just a nit:
If I'm reading the NVMe spec correctly, ds == 0 has a special meaning:
'LBA format is not currently available.'
maybe we should use a different dev_warn() for ds == 0 ?
Maurizio
Hi Maurizio,
Thanks for the nit on v1.
Quoted 3 lines
If I'm reading the NVMe spec correctly, ds == 0 has a special
meaning: 'LBA format is not currently available.' maybe we should use a differentdev_warn()for ds == 0 ?
I added the separate warning in v2, but Keith pointed out on the v2 thread that the "LBA format not available" wording in the spec refers to broadcast identification. For an attached namespace, any unusable flbas index value (including 0) is invalid, so v3 merges both cases into a single check.
Apologies for the back-and-forth.
Also: v2 bounced to your old redhat.com address — using [email protected] from here on.
Thanks,
Chao
Quoted 71 lines · Maurizio Lombardi
Quoted 50 lines · Chao Shi
nvme_update_ns_info_block()trusts id->lbaf[lbaf].ds from the controller and assigns it directly tons->head->lba_shiftwithout bounds checking.nvme_lba_to_sect()then does:return lba << (head->lba_shift - SECTOR_SHIFT);When called with lba =
le64_to_cpu(id->nsze)to compute the device capacity, an attacker-controlled controller can choose ds < 9 or a combination of (ds, nsze) that makes the left shift overflowsector_t. The former is a C undefined behaviour that UBSAN reports as a BUG; the latter silently yields a bogus capacity that the block layer then trusts for bounds checking.Validate ds against
SECTOR_SHIFTand usecheck_shl_overflow()to compute capacity so that any (ds, nsze) combination that would overflowsector_tis rejected. The namespace is skipped with-EIOinstead of crashing the kernel. This is reachable by a malicious NVMe device, a buggy firmware, or an attacker-controlled NVMe-oF target.Stack trace (UBSAN, ds < 9 variant):
RIP: nvme_lba_to_sect drivers/nvme/host/nvme.h:699 [inline] RIP: nvme_update_ns_info_block.cold+0x5/0x7 Call Trace: nvme_update_ns_info+0x175/0xd90 drivers/nvme/host/core.c:2467 nvme_validate_ns drivers/nvme/host/core.c:4299 [inline] nvme_scan_ns drivers/nvme/host/core.c:4350 nvme_scan_ns_async+0xa5/0xe0 drivers/nvme/host/core.c:4383 async_run_entry_fn process_one_work worker_thread kthreadFound by Syzkaller.
Fixes 9419e71b8d67 ("nvme: move ns id info to struct nvme_ns_head")Acked-by Sungwoo Kim <[email protected]>Acked-by Dave Tian <[email protected]>Acked-by Weidong Zhu <[email protected]>Signed-off-by Chao Shi <[email protected]>--- drivers/nvme/host/core.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)drivers/nvme/host/core.c +0
-0drivers/nvme/host/core.c static int nvme_update_ns_info_block(struct nvme_ns *ns, · line 2416
Quoted 10 lines
drivers/nvme/host/core.c +7
-0@@ -2407,9 +2407,19 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns, lim =queue_limits_start_update (ns->disk->queue); memflags =blk_mq_freeze_queue (ns->disk->queue); +if (id->lbaf[lbaf].ds <SECTOR_SHIFT || +check_shl_overflow (le64_to_cpu (id->nsze), + id->lbaf[lbaf].ds -SECTOR_SHIFT , + &capacity)) { +dev_warn_once (ns->ctrl->device, +"invalid LBA data size %u, skipping namespace\n" , + id->lbaf[lbaf].ds);Just a nit:
If I'm reading the NVMe spec correctly, ds == 0 has a special meaning:
'LBA format is not currently available.'
maybe we should use a differentdev_warn()for ds == 0 ?Maurizio
Keith Busch
maintainerdrivers/nvme/host/core.c static int nvme_update_ns_info_block(struct nvme_ns *ns, · line 2417
Quoted 9 lines · Chao Shi
drivers/nvme/host/core.c +8
-0memflags =blk_mq_freeze_queue (ns->disk->queue); +if (id->lbaf[lbaf].ds <SECTOR_SHIFT || +check_shl_overflow (le64_to_cpu (id->nsze), + id->lbaf[lbaf].ds -SECTOR_SHIFT , + &capacity)) { +dev_warn_once (ns->ctrl->device, +"invalid LBA data size %u, skipping namespace\n" , + id->lbaf[lbaf].ds); + ret = -EIO ;
I think ENODEV is more appropriate errno.
drivers/nvme/host/core.c static int nvme_update_ns_info_block(struct nvme_ns *ns, · line 2419
Quoted 2 lines
drivers/nvme/host/core.c +2
-0+blk_mq_unfreeze_queue (ns->disk->queue, memflags); + goto out;
I don't see any particular reason why we shouldn't validate this value before starting the queue updates and freezing the queue, like we for the ncap field up higher. Doing that would make the error case much simpler. Case in point, you're missing the corresponding queue_limits_cancel_update() for this error case.
Quoted 10 lines
drivers/nvme/host/core.c +1
-1+ } ns->head->lba_shift = id->lbaf[lbaf].ds; ns->head->nuse =le64_to_cpu (id->nuse); - capacity =nvme_lba_to_sect (ns->head,le64_to_cpu (id->nsze));nvme_set_ctrl_limits (ns->ctrl, &lim, false);nvme_configure_metadata (ns->ctrl, ns->head, id, nvm, info);nvme_set_chunk_sectors (ns, id, &lim);Signature · 1 line
2.43.0
v2 cover nvme: core: reject invalid LBA data size from Identify Namespace
In reply to Chao ShiThanks for sharing comments! I did such changes in v2:
- Remove Fixes tag(Keith Busch: ds has never been validated)
- Change -EIO to -ENODEV (Keith Busch)
- Add missing "queue_limits_cancel_update()" in error path (Keith Busch)
- Add warning for ds == 0 (LBA format not available) (Maurizio Lombardi)
---
Quoted 1 line · Keith Busch
I think ENODEV is more appropriate errno.
Fixed in v2.
Quoted 2 lines
you're missing the corresponding
queue_limits_cancel_update()for this error case.
Added in both error paths in v2.
Quoted 1 line
This fixes tag is wrong.
Removed.
Quoted 1 line · Daniel Wagner
This fixes tag is wrong. Probably
a5b1cd61820eis the better choice here.
Keith pointed out that ds has never been validated so no commit introduced a regression — removed the Fixes tag entirely.
Quoted 1 line · Maurizio Lombardi
ds == 0 has a special meaning: 'LBA format is not currently available.'
Thanks,added dev_warn_once() for ds==0 in v2.
Chao Shi (1):
nvme: core: reject invalid LBA data size from Identify Namespace
drivers/nvme/host/core.c | 21 ++++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-)
Signature · 1 line
2.43.0
PATCH v2 1/1 nvme: core: reject invalid LBA data size from Identify Namespace
nvme_update_ns_info_block() trusts id->lbaf[lbaf].ds from the controller and assigns it directly to ns->head->lba_shift without bounds checking. nvme_lba_to_sect() then does:
return lba << (head->lba_shift - SECTOR_SHIFT);
When called with lba = le64_to_cpu(id->nsze) to compute the device capacity, an attacker-controlled controller can choose ds < 9 or a combination of (ds, nsze) that makes the left shift overflow sector_t. The former is a C undefined behaviour that UBSAN reports as a BUG; the latter silently yields a bogus capacity that the block layer then trusts for bounds checking.
Validate ds against SECTOR_SHIFT and use check_shl_overflow() to compute capacity so that any (ds, nsze) combination that would overflow sector_t is rejected. The namespace is skipped with -ENODEV instead of crashing the kernel. This is reachable by a malicious NVMe device, a buggy firmware, or an attacker-controlled NVMe-oF target.
Stack trace (UBSAN, ds < 9 variant):
RIP: nvme_lba_to_sect drivers/nvme/host/nvme.h:699 [inline] RIP: nvme_update_ns_info_block.cold+0x5/0x7 Call Trace: nvme_update_ns_info+0x175/0xd90 drivers/nvme/host/core.c:2467 nvme_validate_ns drivers/nvme/host/core.c:4299 [inline] nvme_scan_ns drivers/nvme/host/core.c:4350 nvme_scan_ns_async+0xa5/0xe0 drivers/nvme/host/core.c:4383 async_run_entry_fn process_one_work worker_thread kthread
Found by Syzkaller.
--- drivers/nvme/host/core.c | 21 ++++++++++++++++++++- 1 file changed, 20 insertions(+), 1 deletion(-)
drivers/nvme/host/core.c +20 -1
@@ -2407,9 +2407,28 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns, lim =queue_limits_start_update (ns->disk->queue); memflags =blk_mq_freeze_queue (ns->disk->queue); +if (id->lbaf[lbaf].ds ==0 ) { +dev_warn_once (ns->ctrl->device, +"LBA format not available, skipping namespace\n" ); + ret = -ENODEV ; +queue_limits_cancel_update (ns->disk->queue); +blk_mq_unfreeze_queue (ns->disk->queue, memflags); + goto out; + } +if (id->lbaf[lbaf].ds <SECTOR_SHIFT || +check_shl_overflow (le64_to_cpu (id->nsze), + id->lbaf[lbaf].ds -SECTOR_SHIFT , + &capacity)) { +dev_warn_once (ns->ctrl->device, +"invalid LBA data size %u, skipping namespace\n" , + id->lbaf[lbaf].ds); + ret = -ENODEV ; +queue_limits_cancel_update (ns->disk->queue); +blk_mq_unfreeze_queue (ns->disk->queue, memflags); + goto out; + } ns->head->lba_shift = id->lbaf[lbaf].ds; ns->head->nuse =le64_to_cpu (id->nuse); - capacity =nvme_lba_to_sect (ns->head,le64_to_cpu (id->nsze));nvme_set_ctrl_limits (ns->ctrl, &lim, false);nvme_configure_metadata (ns->ctrl, ns->head, id, nvm, info);nvme_set_chunk_sectors (ns, id, &lim);
Signature · 1 line
2.43.0
Keith Busch
maintainerQuoted 8 lines · Chao Shi
On Mon, Apr 20, 2026 at 02:51:00PM -0700, Keith Busch wrote:
Fixed in v2.
Quoted 2 lines
you're missing the corresponding
queue_limits_cancel_update()for this error case.Added in both error paths in v2.
Not quite what I suggested. Instead, if you move up the validation checks, you don't have any cleanup in the error case to do.
And don't bother splitting the error cases. The message about the "0" LBADS is in reference to the the broadcast identification. It's never valid for a controller to return an unusable flbas index value for an attached namespace.
Hi Keith,
Thanks for the follow-up — I misread your v1 comment.
Quoted 2 lines
Not quite what I suggested. Instead, if you move up the validation checks, you don't have any cleanup in the error case to do.
Understood now. In v3 the validation is hoisted above queue_limits_start_update() and blk_mq_freeze_queue(), so the error path is a plain goto out with no cancel_update / unfreeze needed. (Please let me know if I things on the right track)
Quoted 4 lines
And don't bother splitting the error cases. The message about the "0" LBADS is in reference to the broadcast identification. It's never valid for a controller to return an unusable flbas index value for an attached namespace.
Collapsed into a single check in v3.
Sending v3 shortly.
Thanks,
Chao
Quoted 18 lines · Keith Busch
Quoted 8 lines · Chao Shi
On Mon, Apr 20, 2026 at 02:51:00PM -0700, Keith Busch wrote:
Fixed in v2.
Quoted 2 lines
you're missing the corresponding
queue_limits_cancel_update()for this error case.Added in both error paths in v2.
Not quite what I suggested. Instead, if you move up the validation checks, you don't have any cleanup in the error case to do.
And don't bother splitting the error cases. The message about the "0" LBADS is in reference to the the broadcast identification. It's never valid for a controller to return an unusable flbas index value for an attached namespace.