mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
@ 2026-06-09  0:59 Bryam Vargas
  2026-06-09  8:50 ` John Garry
  0 siblings, 1 reply; 6+ messages in thread
From: Bryam Vargas @ 2026-06-09  0:59 UTC (permalink / raw)
  To: Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, John Garry, David Disseldorp,
	linux-scsi, target-devel, linux-kernel

core_scsi3_emulate_pro_register_and_move() maps the PERSISTENT RESERVE OUT
parameter list with transport_kmap_data_sg() and parses the destination
TransportID with target_parse_pr_out_transport_id(). For an iSCSI
TransportID (FORMAT CODE 01b) iscsi_parse_pr_out_transport_id() returns
the ISID as a raw pointer into that mapped buffer (the bytes following the
",i,0x" separator).

The function then unmaps the buffer with transport_kunmap_data_sg() before
dereferencing iport_ptr in strcmp(), __core_scsi3_locate_pr_reg() and
core_scsi3_alloc_registration() (the last reads 8 bytes via
get_unaligned_be64() and copies the string with snprintf()). When the
parameter list spans more than one page (PARAMETER LIST LENGTH > 4096),
transport_kmap_data_sg() uses vmap() and transport_kunmap_data_sg() does
vunmap(), so the kernel virtual address backing iport_ptr is torn down on
all architectures and every subsequent dereference is a use-after-free of
the unmapped region.

initiator_str does not have this problem because the parser strscpy()s it
into a caller-owned buffer; iport_ptr is the only output left as a borrowed
alias. core_scsi3_decode_spec_i_port() consumes the same alias safely
because it unmaps only after all uses.

Move ownership of the ISID string into the parser: after lowercasing the
ISID in-place, allocate a private copy with kzalloc(PR_REG_ISID_LEN) and
return that instead of the raw buffer pointer. kzalloc() zero-fills the
allocation to PR_REG_ISID_LEN bytes so the fixed 8-byte
get_unaligned_be64() read in __core_scsi3_do_alloc_registration() stays
in-bounds and returns deterministic results even for an ISID shorter than
8 characters (plain kstrdup() would give an allocation shorter than 8 bytes
for a malformed short ISID, turning the be64 read into a heap out-of-bounds).

Both callers now own the returned pointer and must kfree() it when done.
core_scsi3_decode_spec_i_port() frees before each inner-loop parse
iteration (so that a failed ACL match that continues the search does not
leak the previous parse's allocation) and at the error and success exit
paths. core_scsi3_emulate_pro_register_and_move() frees at both the
success return and the error label.

Fixes: 4949314c7283 ("target: Allow control CDBs with data > 1 page")
Cc: stable@vger.kernel.org
Signed-off-by: Bryam Vargas <hexlabsecurity@proton.me>
---
v2 (review of v1 by John Garry):
 - Move the ISID copy into iscsi_parse_pr_out_transport_id() so the parser
   returns an owned allocation via *port_nexus_ptr; callers kfree() it.
   v1's stack-buffer approach (isid_buf[] + iport_ptr = isid_buf) is gone.
 - Use kzalloc(PR_REG_ISID_LEN) + strscpy_pad() rather than plain kstrdup():
   __core_scsi3_do_alloc_registration() reads the ISID with a fixed 8-byte
   get_unaligned_be64(isid); a malformed ISID shorter than 8 chars would
   give a kstrdup allocation smaller than 8 bytes, causing a heap OOB on
   that read.  kzalloc zero-fills to PR_REG_ISID_LEN (16) bytes.
 - core_scsi3_decode_spec_i_port(): add kfree(iport_ptr) before each
   inner-loop reset (line ~1574), at out_unmap:, and before return 0.

Class / impact: CWE-416 use-after-free (use-after-vunmap) in the LIO SCSI
target. Triggerable by an authenticated iSCSI initiator that is a current
Persistent Reservation registrant on the LUN: it sends PERSISTENT RESERVE
OUT / REGISTER AND MOVE with an iSCSI (FORMAT CODE 01b) TransportID and a
PARAMETER LIST LENGTH > 4096 so the parameter list spans >1 page and is
mapped with vmap(). After transport_kunmap_data_sg() vunmap()s that region,
the retained iport_ptr is dereferenced -> kernel read of an unmapped
vmalloc address (oops / DoS; memory-safety corruption confirmed by KASAN).
Primarily a remotely-reachable authenticated denial of service.

Affected: all maintained trees -- it became a destructive dangling
dereference with 4949314c7283 (v3.3, 2012), which introduced the
multi-page vmap() path. Verified present at mainline v7.1-rc6 and
stable v6.12.92.

Reproducer (authenticated iSCSI initiator, current PR reservation holder):
  1. PERSISTENT RESERVE OUT / REGISTER a key from the iSCSI nexus.
  2. PERSISTENT RESERVE OUT / REGISTER AND MOVE, FORMAT CODE 01b TransportID
     (IQN + ",i,0x" + 12-char ISID), RELATIVE TARGET PORT IDENTIFIER of an
     existing target port, with PARAMETER LIST LENGTH = 8192 (two pages ->
     vmap()/vunmap()), the inner ADDITIONAL LENGTH set so tid_len + 24 ==
     data_length, the remainder zero padding.

A/B verification (CONFIG_KASAN_VMALLOC=y, kasan.fault=report, x86-64,
6.12.90; reproduced with both a 64-bit and a 32-bit initiator):
  - Without this patch (8192-byte, two-page request):
      BUG: KASAN: vmalloc-out-of-bounds in strcmp+0xa7/0xb0
        strcmp
        core_scsi3_emulate_pro_register_and_move [target_core]
        ? remove_vm_area
        target_scsi3_emulate_pr_out [target_core]
        __target_execute_cmd / iscsit_execute_cmd / iscsi_target_rx_thread
      The buggy address belongs to a vmalloc virtual mapping
      BUG: unable to handle page fault for address ... (PTE 0)
  - Control (56/128-byte, single-page request): no report (kunmap is a
    no-op on 64-bit !HIGHMEM).
  - With this patch (same 8192-byte request): no report, command completes.

drivers/target/target_core_fabric_lib.c | 16 ++++++++++++++++
 drivers/target/target_core_pr.c         |  5 +++++
 2 files changed, 21 insertions(+)
diff --git a/drivers/target/target_core_fabric_lib.c b/drivers/target/target_core_fabric_lib.c
index 87c5d26a5089..b5ad45f072bd 100644
--- a/drivers/target/target_core_fabric_lib.c
+++ b/drivers/target/target_core_fabric_lib.c
@@ -19,6 +19,7 @@
 #include <linux/hex.h>
 #include <linux/kernel.h>
 #include <linux/string.h>
+#include <linux/slab.h>
 #include <linux/ctype.h>
 #include <linux/spinlock.h>
 #include <linux/export.h>
@@ -367,6 +368,21 @@ static bool iscsi_parse_pr_out_transport_id(
 			*p = tolower(*p);
 			p++;
 		}
+		/*
+		 * The loop above advanced p past the ISID; *port_nexus_ptr still
+		 * holds the ISID start.  Replace the borrowed buffer alias with an
+		 * owned heap copy so callers can safely use the ISID past the
+		 * buffer lifetime (e.g. after transport_kunmap_data_sg() in
+		 * register_and_move).  kzalloc() zero-fills to PR_REG_ISID_LEN
+		 * bytes so the 8-byte get_unaligned_be64() read in
+		 * __core_scsi3_do_alloc_registration() stays in-bounds even for
+		 * an ISID shorter than 8 characters.
+		 */
+		p = *port_nexus_ptr;
+		*port_nexus_ptr = kzalloc(PR_REG_ISID_LEN, GFP_KERNEL);
+		if (!*port_nexus_ptr)
+			return false;
+		strscpy_pad(*port_nexus_ptr, p, PR_REG_ISID_LEN);
 	} else
 		*port_nexus_ptr = NULL;

diff --git a/drivers/target/target_core_pr.c b/drivers/target/target_core_pr.c
index 11790f2c5d80..d7340c4fbd07 100644
--- a/drivers/target/target_core_pr.c
+++ b/drivers/target/target_core_pr.c
@@ -1571,6 +1571,7 @@ core_scsi3_decode_spec_i_port(
 				continue;
 			dest_rtpi = tmp_lun->lun_tpg->tpg_rtpi;

+			kfree(iport_ptr);
 			iport_ptr = NULL;
 			tid_found = target_parse_pr_out_transport_id(tmp_tpg,
 					ptr, &tid_len, &iport_ptr, i_str);
@@ -1808,9 +1809,11 @@ core_scsi3_decode_spec_i_port(
 		core_scsi3_tpg_undepend_item(dest_tpg);
 	}

+	kfree(iport_ptr);
 	return 0;
 out_unmap:
 	transport_kunmap_data_sg(cmd);
+	kfree(iport_ptr);
 out:
 	/*
 	 * For the failure case, release everything from tid_dest_list
@@ -3532,10 +3535,12 @@ core_scsi3_emulate_pro_register_and_move(struct se_cmd *cmd, u64 res_key,
 	core_scsi3_update_and_write_aptpl(cmd->se_dev, aptpl);

 	core_scsi3_put_pr_reg(dest_pr_reg);
+	kfree(iport_ptr);
 	return 0;
 out:
 	if (buf)
 		transport_kunmap_data_sg(cmd);
+	kfree(iport_ptr);
 	if (dest_se_deve)
 		core_scsi3_lunacl_undepend_item(dest_se_deve);
 	if (dest_node_acl)

base-commit: 8e65320d91cdc3b241d4b94855c88459b91abf66
-- 
2.43.0


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
  2026-06-09  0:59 [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer Bryam Vargas
@ 2026-06-09  8:50 ` John Garry
  2026-06-09 11:36   ` James Bottomley
  0 siblings, 1 reply; 6+ messages in thread
From: John Garry @ 2026-06-09  8:50 UTC (permalink / raw)
  To: Bryam Vargas, Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, David Disseldorp, linux-scsi,
	target-devel, linux-kernel

On 09/06/2026 01:59, Bryam Vargas wrote:
> core_scsi3_emulate_pro_register_and_move() maps the PERSISTENT RESERVE OUT
> parameter list with transport_kmap_data_sg() and parses the destination
> TransportID with target_parse_pr_out_transport_id(). For an iSCSI
> TransportID (FORMAT CODE 01b) iscsi_parse_pr_out_transport_id() returns
> the ISID as a raw pointer into that mapped buffer (the bytes following the
> ",i,0x" separator).
> 
> The function then unmaps the buffer with transport_kunmap_data_sg() before
> dereferencing iport_ptr in strcmp(), __core_scsi3_locate_pr_reg() and
> core_scsi3_alloc_registration() (the last reads 8 bytes via
> get_unaligned_be64() and copies the string with snprintf()). When the
> parameter list spans more than one page (PARAMETER LIST LENGTH > 4096),
> transport_kmap_data_sg() uses vmap() and transport_kunmap_data_sg() does
> vunmap(), so the kernel virtual address backing iport_ptr is torn down on
> all architectures and every subsequent dereference is a use-after-free of
> the unmapped region.
> 
> initiator_str does not have this problem because the parser strscpy()s it
> into a caller-owned buffer; iport_ptr is the only output left as a borrowed
> alias. core_scsi3_decode_spec_i_port() consumes the same alias safely
> because it unmaps only after all uses.
> 
> Move ownership of the ISID string into the parser: after lowercasing the
> ISID in-place, allocate a private copy with kzalloc(PR_REG_ISID_LEN) and
> return that instead of the raw buffer pointer. kzalloc() zero-fills the
> allocation to PR_REG_ISID_LEN bytes so the fixed 8-byte
> get_unaligned_be64() read in __core_scsi3_do_alloc_registration() stays
> in-bounds and returns deterministic results even for an ISID shorter than
> 8 characters (plain kstrdup() would give an allocation shorter than 8 bytes
> for a malformed short ISID, turning the be64 read into a heap out-of-bounds).
> 
> Both callers now own the returned pointer and must kfree() it when done.
> core_scsi3_decode_spec_i_port() frees before each inner-loop parse
> iteration (so that a failed ACL match that continues the search does not
> leak the previous parse's allocation) and at the error and success exit
> paths. core_scsi3_emulate_pro_register_and_move() frees at both the
> success return and the error label.
> 
> Fixes: 4949314c7283 ("target: Allow control CDBs with data > 1 page")
> Cc: stable@vger.kernel.org
> Signed-off-by: Bryam Vargas <hexlabsecurity@proton.me>
> ---
> v2 (review of v1 by John Garry):
>   - Move the ISID copy into iscsi_parse_pr_out_transport_id() so the parser
>     returns an owned allocation via *port_nexus_ptr; callers kfree() it.
>     v1's stack-buffer approach (isid_buf[] + iport_ptr = isid_buf) is gone.
>   - Use kzalloc(PR_REG_ISID_LEN) + strscpy_pad() rather than plain kstrdup():
>     __core_scsi3_do_alloc_registration() reads the ISID with a fixed 8-byte
>     get_unaligned_be64(isid); a malformed ISID shorter than 8 chars would
>     give a kstrdup allocation smaller than 8 bytes, causing a heap OOB on
>     that read.  kzalloc zero-fills to PR_REG_ISID_LEN (16) bytes.
>   - core_scsi3_decode_spec_i_port(): add kfree(iport_ptr) before each
>     inner-loop reset (line ~1574), at out_unmap:, and before return 0.
> 
> Class / impact: CWE-416 use-after-free (use-after-vunmap) in the LIO SCSI
> target. Triggerable by an authenticated iSCSI initiator that is a current
> Persistent Reservation registrant on the LUN: it sends PERSISTENT RESERVE
> OUT / REGISTER AND MOVE with an iSCSI (FORMAT CODE 01b) TransportID and a
> PARAMETER LIST LENGTH > 4096 so the parameter list spans >1 page and is
> mapped with vmap(). After transport_kunmap_data_sg() vunmap()s that region,
> the retained iport_ptr is dereferenced -> kernel read of an unmapped
> vmalloc address (oops / DoS; memory-safety corruption confirmed by KASAN).
> Primarily a remotely-reachable authenticated denial of service.
> 
> Affected: all maintained trees -- it became a destructive dangling
> dereference with 4949314c7283 (v3.3, 2012), which introduced the
> multi-page vmap() path. Verified present at mainline v7.1-rc6 and
> stable v6.12.92.
> 
> Reproducer (authenticated iSCSI initiator, current PR reservation holder):
>    1. PERSISTENT RESERVE OUT / REGISTER a key from the iSCSI nexus.
>    2. PERSISTENT RESERVE OUT / REGISTER AND MOVE, FORMAT CODE 01b TransportID
>       (IQN + ",i,0x" + 12-char ISID), RELATIVE TARGET PORT IDENTIFIER of an
>       existing target port, with PARAMETER LIST LENGTH = 8192 (two pages ->
>       vmap()/vunmap()), the inner ADDITIONAL LENGTH set so tid_len + 24 ==
>       data_length, the remainder zero padding.
> 
> A/B verification (CONFIG_KASAN_VMALLOC=y, kasan.fault=report, x86-64,
> 6.12.90; reproduced with both a 64-bit and a 32-bit initiator):
>    - Without this patch (8192-byte, two-page request):
>        BUG: KASAN: vmalloc-out-of-bounds in strcmp+0xa7/0xb0
>          strcmp
>          core_scsi3_emulate_pro_register_and_move [target_core]
>          ? remove_vm_area
>          target_scsi3_emulate_pr_out [target_core]
>          __target_execute_cmd / iscsit_execute_cmd / iscsi_target_rx_thread
>        The buggy address belongs to a vmalloc virtual mapping
>        BUG: unable to handle page fault for address ... (PTE 0)
>    - Control (56/128-byte, single-page request): no report (kunmap is a
>      no-op on 64-bit !HIGHMEM).
>    - With this patch (same 8192-byte request): no report, command completes.
> 
> drivers/target/target_core_fabric_lib.c | 16 ++++++++++++++++
>   drivers/target/target_core_pr.c         |  5 +++++
>   2 files changed, 21 insertions(+)
> diff --git a/drivers/target/target_core_fabric_lib.c b/drivers/target/target_core_fabric_lib.c
> index 87c5d26a5089..b5ad45f072bd 100644
> --- a/drivers/target/target_core_fabric_lib.c
> +++ b/drivers/target/target_core_fabric_lib.c
> @@ -19,6 +19,7 @@
>   #include <linux/hex.h>
>   #include <linux/kernel.h>
>   #include <linux/string.h>
> +#include <linux/slab.h>
>   #include <linux/ctype.h>
>   #include <linux/spinlock.h>
>   #include <linux/export.h>
> @@ -367,6 +368,21 @@ static bool iscsi_parse_pr_out_transport_id(
>   			*p = tolower(*p);
>   			p++;
>   		}
> +		/*
> +		 * The loop above advanced p past the ISID; *port_nexus_ptr still
> +		 * holds the ISID start.  Replace the borrowed buffer alias with an
> +		 * owned heap copy so callers can safely use the ISID past the
> +		 * buffer lifetime (e.g. after transport_kunmap_data_sg() in
> +		 * register_and_move).  kzalloc() zero-fills to PR_REG_ISID_LEN
> +		 * bytes so the 8-byte get_unaligned_be64() read in
> +		 * __core_scsi3_do_alloc_registration() stays in-bounds even for
> +		 * an ISID shorter than 8 characters.
> +		 */
> +		p = *port_nexus_ptr;
> +		*port_nexus_ptr = kzalloc(PR_REG_ISID_LEN, GFP_KERNEL);
> +		if (!*port_nexus_ptr)
> +			return false;
> +		strscpy_pad(*port_nexus_ptr, p, PR_REG_ISID_LEN);
 >   	} else>   		*port_nexus_ptr = NULL;
> 
> diff --git a/drivers/target/target_core_pr.c b/drivers/target/target_core_pr.c
> index 11790f2c5d80..d7340c4fbd07 100644
> --- a/drivers/target/target_core_pr.c
> +++ b/drivers/target/target_core_pr.c
> @@ -1571,6 +1571,7 @@ core_scsi3_decode_spec_i_port(
>   				continue;
>   			dest_rtpi = tmp_lun->lun_tpg->tpg_rtpi;
> 
> +			kfree(iport_ptr);
>   			iport_ptr = NULL;
>   			tid_found = target_parse_pr_out_transport_id(tmp_tpg,
>   					ptr, &tid_len, &iport_ptr, i_str);
> @@ -1808,9 +1809,11 @@ core_scsi3_decode_spec_i_port(
>   		core_scsi3_tpg_undepend_item(dest_tpg);
>   	}
> 
> +	kfree(iport_ptr);
>   	return 0;
>   out_unmap:
>   	transport_kunmap_data_sg(cmd);
> +	kfree(iport_ptr);

sorry for suggesting this change, but this is worse than what you had 
originally, as we have so many paths to call kfree() [which means more 
bugs possible]

it's hard to make good suggestions for this target code as the functions 
are so large and complex.

Is there any reason why we can't just keep the transport_kmap_data_sg() 
in place for (much) longer, i.e. until at the out: label? We already 
handle the the unmap properly there. I do notice that there would be 
regions which we keep spinlocks held when this mapping is in place, but 
I am not sure if that makes an difference

>   out:
>   	/*
>   	 * For the failure case, release everything from tid_dest_list
> @@ -3532,10 +3535,12 @@ core_scsi3_emulate_pro_register_and_move(struct se_cmd *cmd, u64 res_key,
>   	core_scsi3_update_and_write_aptpl(cmd->se_dev, aptpl);
> 
>   	core_scsi3_put_pr_reg(dest_pr_reg);
> +	kfree(iport_ptr);
>   	return 0;
>   out:
>   	if (buf)
>   		transport_kunmap_data_sg(cmd);
> +	kfree(iport_ptr);
>   	if (dest_se_deve)
>   		core_scsi3_lunacl_undepend_item(dest_se_deve);
>   	if (dest_node_acl)
> 
> base-commit: 8e65320d91cdc3b241d4b94855c88459b91abf66


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
  2026-06-09  8:50 ` John Garry
@ 2026-06-09 11:36   ` James Bottomley
  2026-06-09 12:14     ` John Garry
  0 siblings, 1 reply; 6+ messages in thread
From: James Bottomley @ 2026-06-09 11:36 UTC (permalink / raw)
  To: John Garry, Bryam Vargas, Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, David Disseldorp, linux-scsi,
	target-devel, linux-kernel

On Tue, 2026-06-09 at 09:50 +0100, John Garry wrote:
> > @@ -1808,9 +1809,11 @@ core_scsi3_decode_spec_i_port(
> >    		core_scsi3_tpg_undepend_item(dest_tpg);
> >    	}
> > 
> > +	kfree(iport_ptr);
> >    	return 0;
> >    out_unmap:
> >    	transport_kunmap_data_sg(cmd);
> > +	kfree(iport_ptr);
> 
> sorry for suggesting this change, but this is worse than what you had
> originally, as we have so many paths to call kfree() [which means
> more bugs possible]
> 
> it's hard to make good suggestions for this target code as the
> functions are so large and complex.

Given that it's allocated in a function called by the routine but never
actually retained by anything what about defining it as

chat *iport_ptr __free(kfree) = NULL;

?  That way we don't need to care about freeing it in the error legs. 
Even if the value were retained under some conditions, you just signal
retention by setting iport_ptr to NULL to prevent the kfree on exit.

Regards,

James


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
  2026-06-09 11:36   ` James Bottomley
@ 2026-06-09 12:14     ` John Garry
  2026-06-09 12:27       ` James Bottomley
  0 siblings, 1 reply; 6+ messages in thread
From: John Garry @ 2026-06-09 12:14 UTC (permalink / raw)
  To: James Bottomley, Bryam Vargas, Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, David Disseldorp, linux-scsi,
	target-devel, linux-kernel

On 09/06/2026 12:36, James Bottomley wrote:
> On Tue, 2026-06-09 at 09:50 +0100, John Garry wrote:
>>> @@ -1808,9 +1809,11 @@ core_scsi3_decode_spec_i_port(
>>>     		core_scsi3_tpg_undepend_item(dest_tpg);
>>>     	}
>>>
>>> +	kfree(iport_ptr);
>>>     	return 0;
>>>     out_unmap:
>>>     	transport_kunmap_data_sg(cmd);
>>> +	kfree(iport_ptr);
>>
>> sorry for suggesting this change, but this is worse than what you had
>> originally, as we have so many paths to call kfree() [which means
>> more bugs possible]
>>
>> it's hard to make good suggestions for this target code as the
>> functions are so large and complex.
> 
> Given that it's allocated in a function called by the routine but never
> actually retained by anything what about defining it as
> 
> chat *iport_ptr __free(kfree) = NULL;
> 
> ?  That way we don't need to care about freeing it in the error legs.

It's not just freed in the error legs, as 
core_scsi3_decode_spec_i_port() calls target_parse_pr_out() in a loop

while (tpdl > 0) {

...

	spin_lock(&dev->se_port_lock);
	list_for_each_entry(tmp_lun, &dev->dev_sep_list, lun_dev_link) {
		...

		kfree(iport_ptr);
		iport_ptr = NULL;
		tid_found = target_parse_pr_out_transport_id(tmp_tpg,
				ptr, &tid_len, &iport_ptr, i_str);
		if (!tid_found)
			continue;

....

	kfree(iport_ptr);
out_unmap:
	kfree(iport_ptr);
out:
	...

If __free(kfree) attr can still handle this sort of flow, then ok.

I still think that it would be simpler to keep the mapping in place for 
longer (to avoid all of this), if possible.

Thanks,
John

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
  2026-06-09 12:14     ` John Garry
@ 2026-06-09 12:27       ` James Bottomley
  2026-06-10  4:22         ` Bryam Vargas
  0 siblings, 1 reply; 6+ messages in thread
From: James Bottomley @ 2026-06-09 12:27 UTC (permalink / raw)
  To: John Garry, Bryam Vargas, Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, David Disseldorp, linux-scsi,
	target-devel, linux-kernel

On Tue, 2026-06-09 at 13:14 +0100, John Garry wrote:
> On 09/06/2026 12:36, James Bottomley wrote:
> > On Tue, 2026-06-09 at 09:50 +0100, John Garry wrote:
> > > > @@ -1808,9 +1809,11 @@ core_scsi3_decode_spec_i_port(
> > > >     		core_scsi3_tpg_undepend_item(dest_tpg);
> > > >     	}
> > > > 
> > > > +	kfree(iport_ptr);
> > > >     	return 0;
> > > >     out_unmap:
> > > >     	transport_kunmap_data_sg(cmd);
> > > > +	kfree(iport_ptr);
> > > 
> > > sorry for suggesting this change, but this is worse than what you
> > > had
> > > originally, as we have so many paths to call kfree() [which means
> > > more bugs possible]
> > > 
> > > it's hard to make good suggestions for this target code as the
> > > functions are so large and complex.
> > 
> > Given that it's allocated in a function called by the routine but
> > never
> > actually retained by anything what about defining it as
> > 
> > chat *iport_ptr __free(kfree) = NULL;
> > 
> > ?  That way we don't need to care about freeing it in the error
> > legs.
> 
> It's not just freed in the error legs, as 
> core_scsi3_decode_spec_i_port() calls target_parse_pr_out() in a loop
> 
> while (tpdl > 0) {
> 
> ...
> 
> 	spin_lock(&dev->se_port_lock);
> 	list_for_each_entry(tmp_lun, &dev->dev_sep_list,
> lun_dev_link) {
> 		...
> 
> 		kfree(iport_ptr);
> 		iport_ptr = NULL;
> 		tid_found =
> target_parse_pr_out_transport_id(tmp_tpg,
> 				ptr, &tid_len, &iport_ptr, i_str);
> 		if (!tid_found)
> 			continue;
> 
> ....
> 
> 	kfree(iport_ptr);
> out_unmap:
> 	kfree(iport_ptr);
> out:
> 	...

You'd simply remove those since the pointer would be kfree'd on any
function return.

> If __free(kfree) attr can still handle this sort of flow, then ok.

It can, but there's nuance.  If you do an early free (or a transfer of
ownership) you have to set the original pointer to NULL to avoid kfree
being called on it on return.

> I still think that it would be simpler to keep the mapping in place
> for longer (to avoid all of this), if possible.

I was just looking at the multiple error legs in the patched code and
thinking it's an ideal candidate for a __free() attribute.  If there's
a way of rearranging the code to keep it longer that you think is more
readable, by all means do that instead.

Regards,

James


^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer
  2026-06-09 12:27       ` James Bottomley
@ 2026-06-10  4:22         ` Bryam Vargas
  0 siblings, 0 replies; 6+ messages in thread
From: Bryam Vargas @ 2026-06-10  4:22 UTC (permalink / raw)
  To: Martin K . Petersen
  Cc: Mike Christie, Maurizio Lombardi, John Garry, James Bottomley,
	David Disseldorp, linux-scsi, target-devel, linux-kernel

On 2026-06-09, James Bottomley wrote:
> I was just looking at the multiple error legs in the patched code and
> thinking it's an ideal candidate for a __free() attribute.  If there's
> a way of rearranging the code to keep it longer that you think is more
> readable, by all means do that instead.

I'll keep the mapping in place, which turns out to be the simplest of the
three options: it needs no ownership transfer at all, so neither kfree()
nor __free() comes into it.

The only reason v2 had an iport_ptr to free was that v2 made the parser
allocate.  If the PR-OUT parameter list instead stays mapped, iport_ptr
can remain what it has always been -- a borrowed alias into that buffer --
and v3 is just:

  - drop the early transport_kunmap_data_sg() / buf = NULL after the parse;
  - unmap once on the success path, right before "return 0";
  - the existing "if (buf) transport_kunmap_data_sg(cmd)" at out: now
    covers every error leg, because buf is no longer cleared early.

That is -2/+1 lines (plus a comment) in
core_scsi3_emulate_pro_register_and_move() only.
target_core_fabric_lib.c is untouched, and core_scsi3_decode_spec_i_port()
needs no change -- it already uses iport_ptr before its own unmap, so the
loop/free question there does not arise.

On 2026-06-09, John Garry wrote:
> I do notice that there would be regions which we keep spinlocks held
> when this mapping is in place, but I am not sure if that makes a
> difference

It does not.  Holding an already-established kmap/vmap mapping across a
spinlock is fine; what is not allowed is calling kmap()/kunmap() (or
vmap()/vunmap(), which may sleep) while atomic.  In this version both the
map and the unmap happen outside any lock -- the map before
dev_reservation_lock is taken, the unmap after it is dropped (and at out:
after all locks are released).  The only thing done under
dev_reservation_lock is __core_scsi3_locate_pr_reg() reading the ISID
through the mapping, which is an ordinary load.

I re-ran the KASAN A/B for this form on the same CONFIG_KASAN_VMALLOC rig
(6.12.90, 64-bit and 32-bit initiator): unpatched, an 8192-byte (two-page)
REGISTER AND MOVE reports vmalloc-out-of-bounds in strcmp() and the rx
kthread dies; with the patch the same request completes with no report.

v3 follows as a separate posting.

Thanks,
Bryam


^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2026-06-10  4:22 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-06-09  0:59 [PATCH v2] scsi: target: copy iSCSI ISID before unmapping the PR OUT buffer Bryam Vargas
2026-06-09  8:50 ` John Garry
2026-06-09 11:36   ` James Bottomley
2026-06-09 12:14     ` John Garry
2026-06-09 12:27       ` James Bottomley
2026-06-10  4:22         ` Bryam Vargas

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®