Skip to contentSkip to conversation

ceph: fix OOB read in decode_lockers() via missing bounds check

PATCH In review Pavitra Jha 4 participants 8 messages May 23 – Jun 2

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ceph: fix OOB read in decode_lockers() via missing bounds check

ceph_start_decoding() accepts struct_len=0 as valid:
ceph_decode_need(p, end, 0, bad) always passes. When a malicious or compromised OSD sends a cls_lock_get_info_reply with struct_len=0, ceph_start_decoding() returns success with p == end, leaving zero bytes guaranteed for subsequent reads.

The immediately following bare ceph_decode_32(p) in decode_lockers() has no preceding bounds check. With p == end this is a 4-byte read past the validated buffer boundary. The garbage value is then passed to kzalloc_objs() as the locker count.

The sibling function decode_watchers() in osd_client.c already uses the safe variant ceph_decode_32_safe() after its own ceph_start_decoding() call. decode_lockers() is the only site using the bare variant, confirming an oversight.

Fix by replacing ceph_decode_32(p) with ceph_decode_32_safe(p, end, *num_lockers, err_inval), adding a new err_inval label that returns -EINVAL directly without attempting to free an uninitialized lockers pointer.

KASAN report (kernel 7.0.0-rc7, QEMU/x86_64, KASLR disabled):
==================================================================

  BUG: KASAN: slab-out-of-bounds in ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
  Read of size 4 at addr ffff88800a29b76e by task insmod/58

  CPU: 0 UID: 0 PID: 58 Comm: insmod Tainted: G           O        7.0.0-rc7-g9c2abf69da83-dirty #15 PREEMPT(lazy)
  Tainted: [O]=OOT_MODULE
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x4d/0x70
   print_report+0x170/0x4f3
   kasan_report+0xda/0x110
   ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Allocated by task 58:
   kasan_save_stack+0x30/0x50
   kasan_save_track+0x14/0x30
   __kasan_kmalloc+0x7f/0x90
   ceph_oob3_init+0x4d/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  The buggy address belongs to the object at ffff88800a29a000
   which belongs to the cache kmalloc-8k of size 8192
  The buggy address is located 5998 bytes inside of
   allocated 6000-byte region [ffff88800a29a000, ffff88800a29b770)

  Memory state around the buggy address:
   ffff88800a29b600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
   ffff88800a29b680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  >ffff88800a29b700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc fc
                                                               ^
   ffff88800a29b780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
  ==================================================================

num_lockers=0xccccaaaa (OOB garbage from KASAN redzone)

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition) without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")
---
 net/ceph/cls_lock_client.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
net/ceph/cls_lock_client.c +3 -1
index c6956f1df..78276273c 100644
@@ -299,7 +299,7 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 	if (ret)
 		return ret;
 
-	*num_lockers = ceph_decode_32(p);
+	ceph_decode_32_safe(p, end, *num_lockers, err_inval);
 	*lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO);
 	if (!*lockers)
 		return -ENOMEM;
@@ -320,6 +320,8 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 	*tag = s;
 	return 0;
 
+err_inval:
+	return -EINVAL;
 err_free_lockers:
 	ceph_free_lockers(*lockers, *num_lockers);
 	return ret;
Signature · 1 line
2.53.0
#1 lore raw

net/ceph/cls_lock_client.c static int decode_lockers(void **p, void *end, u8 *type, char **tag,

Quoted 90 lines · Pavitra Jha

ceph_start_decoding() accepts struct_len=0 as valid:
ceph_decode_need(p, end, 0, bad) always passes. When a malicious or compromised OSD sends a cls_lock_get_info_reply with struct_len=0, ceph_start_decoding() returns success with p == end, leaving zero bytes guaranteed for subsequent reads.

The immediately following bare ceph_decode_32(p) in decode_lockers() has no preceding bounds check. With p == end this is a 4-byte read past the validated buffer boundary. The garbage value is then passed to kzalloc_objs() as the locker count.

The sibling function decode_watchers() in osd_client.c already uses the safe variant ceph_decode_32_safe() after its own ceph_start_decoding() call. decode_lockers() is the only site using the bare variant, confirming an oversight.

Fix by replacing ceph_decode_32(p) with ceph_decode_32_safe(p, end, *num_lockers, err_inval), adding a new err_inval label that returns -EINVAL directly without attempting to free an uninitialized lockers pointer.

KASAN report (kernel 7.0.0-rc7, QEMU/x86_64, KASLR disabled):
==================================================================

  BUG: KASAN: slab-out-of-bounds in ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
  Read of size 4 at addr ffff88800a29b76e by task insmod/58

  CPU: 0 UID: 0 PID: 58 Comm: insmod Tainted: G           O        7.0.0-rc7-g9c2abf69da83-dirty #15 PREEMPT(lazy)
  Tainted: [O]=OOT_MODULE
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x4d/0x70
   print_report+0x170/0x4f3
   kasan_report+0xda/0x110
   ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Allocated by task 58:
   kasan_save_stack+0x30/0x50
   kasan_save_track+0x14/0x30
   __kasan_kmalloc+0x7f/0x90
   ceph_oob3_init+0x4d/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  The buggy address belongs to the object at ffff88800a29a000
   which belongs to the cache kmalloc-8k of size 8192
  The buggy address is located 5998 bytes inside of
   allocated 6000-byte region [ffff88800a29a000, ffff88800a29b770)

  Memory state around the buggy address:
   ffff88800a29b600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
   ffff88800a29b680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  >ffff88800a29b700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc fc
                                                               ^
   ffff88800a29b780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
  ==================================================================

num_lockers=0xccccaaaa (OOB garbage from KASAN redzone)

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition) without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")
---
 net/ceph/cls_lock_client.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
net/ceph/cls_lock_client.c +0 -0
index c6956f1df..78276273c 100644
@@ -299,7 +299,7 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,

As far as I can see, decode_locker() has the same issue. Have you sent the patch for this?

Quoted 9 lines
net/ceph/cls_lock_client.c +1 -1
 	if (ret)
 		return ret;
 
-	*num_lockers = ceph_decode_32(p);
+	ceph_decode_32_safe(p, end, *num_lockers, err_inval);
 	*lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO);
 	if (!*lockers)
 		return -ENOMEM;

net/ceph/cls_lock_client.c static int decode_lockers(void **p, void *end, u8 *type, char **tag, · line 327

Quoted 8 lines
net/ceph/cls_lock_client.c +2 -0
@@ -320,6 +320,8 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 	*tag = s;
 	return 0;
 
+err_inval:
+	return -EINVAL;
 err_free_lockers:
 	ceph_free_lockers(*lockers, *num_lockers);
 	return ret;

I am still not completely sure about -EINVAL here too. Maybe, -ENOMEM could be more correct.

What about this [1]?

*type = ceph_decode_8(p);

Should we rework it too?

Thanks,
Slava.

[1]
https://elixir.bootlin.com/linux/v7.1-rc4/source/net/ceph/cls_lock_client.c#L313

#2 lore raw

PATCH v2 1/1 ceph: fix bare ceph_decode_8 OOB in decode_lockers()

decode_lockers() in cls_lock_client.c contains a bare ceph_decode_8(p) call after the decode_locker() loop that has no preceding bounds check.

If a malicious or compromised OSD sends a cls_lock_get_info_reply where num_lockers is crafted such that the decode_locker() loop advances p exactly to end (or if num_lockers=0 and p is already at end after ceph_start_decoding() accepts struct_len=0), the subsequent bare ceph_decode_8(p) reads one byte past the validated buffer boundary.

The result is passed directly into *type, which is subsequently used as a lock type discriminator by callers. An OSD-controlled one-byte OOB read at this position gives an attacker influence over the lock type field with no further preconditions.

The safe variant ceph_decode_8_safe() already exists and is used consistently throughout the codebase. This site is the only remaining bare ceph_decode_8() in the decode_lockers() post-loop path.

The goto target is err_free_lockers (not err_inval) because *lockers is already allocated at this point and must be freed on any decode failure.

v1 of this series fixed the bare ceph_decode_32() before kzalloc_objs() and added the err_inval label. This v2 addresses the second bare decode identified by Viacheslav Dubeyko's review.

Regarding the -EINVAL choice (raised in review): -EINVAL is correct for the err_inval path. The failure is structural malformation of OSD-supplied data, not a memory shortage. -ENOMEM would misrepresent the failure class to callers and to stable@ backporters triaging error paths.

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition) without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")

---
v2: Replace bare *type = ceph_decode_8(p) with ceph_decode_8_safe(),
goto err_free_lockers to correctly free *lockers on failure. Address Viacheslav Dubeyko's review question about this site and clarify -EINVAL rationale.

---
 net/ceph/cls_lock_client.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ceph/cls_lock_client.c +1 -1
index 4f27b3d15..c9183a348 100644
@@ -314,7 +314,7 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 			goto err_free_lockers;
 	}
 
-	*type = ceph_decode_8(p);
+	ceph_decode_8_safe(p, end, *type, err_free_lockers);
 	s = ceph_extract_encoded_string(p, end, NULL, GFP_NOIO);
 	if (IS_ERR(s)) {
 		ret = PTR_ERR(s);
Signature · 1 line
2.53.0
#3 lore raw

Re: [PATCH v2] ceph: fix bare ceph_decode_8 OOB in decode_lockers()

Quoted 52 lines · Pavitra Jha

decode_lockers() in cls_lock_client.c contains a bare ceph_decode_8(p) call after the decode_locker() loop that has no preceding bounds check.

If a malicious or compromised OSD sends a cls_lock_get_info_reply where num_lockers is crafted such that the decode_locker() loop advances p exactly to end (or if num_lockers=0 and p is already at end after ceph_start_decoding() accepts struct_len=0), the subsequent bare ceph_decode_8(p) reads one byte past the validated buffer boundary.

The result is passed directly into *type, which is subsequently used as a lock type discriminator by callers. An OSD-controlled one-byte OOB read at this position gives an attacker influence over the lock type field with no further preconditions.

The safe variant ceph_decode_8_safe() already exists and is used consistently throughout the codebase. This site is the only remaining bare ceph_decode_8() in the decode_lockers() post-loop path.

The goto target is err_free_lockers (not err_inval) because *lockers is already allocated at this point and must be freed on any decode failure.

v1 of this series fixed the bare ceph_decode_32() before kzalloc_objs() and added the err_inval label. This v2 addresses the second bare decode identified by Viacheslav Dubeyko's review.

Regarding the -EINVAL choice (raised in review): -EINVAL is correct for the err_inval path. The failure is structural malformation of OSD-supplied data, not a memory shortage. -ENOMEM would misrepresent the failure class to callers and to stable@ backporters triaging error paths.

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition) without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")

---
v2: Replace bare *type = ceph_decode_8(p) with ceph_decode_8_safe(),
goto err_free_lockers to correctly free *lockers on failure. Address Viacheslav Dubeyko's review question about this site and clarify -EINVAL rationale.

---
 net/ceph/cls_lock_client.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/ceph/cls_lock_client.c +0 -0
index 4f27b3d15..c9183a348 100644

net/ceph/cls_lock_client.c static int decode_lockers(void **p, void *end, u8 *type, char **tag, · line 320

Quoted 8 lines
net/ceph/cls_lock_client.c +1 -1
@@ -314,7 +314,7 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 			goto err_free_lockers;
 	}
 
-	*type = ceph_decode_8(p);
+	ceph_decode_8_safe(p, end, *type, err_free_lockers);
 	s = ceph_extract_encoded_string(p, end, NULL, GFP_NOIO);
 	if (IS_ERR(s)) {
 		ret = PTR_ERR(s);

Is it correct patch? Because, initial patch contained this:

net/ceph/cls_lock_client.c

Quoted 3 lines
net/ceph/cls_lock_client.c +1 -1
-	*num_lockers = ceph_decode_32(p);
+	ceph_decode_32_safe(p, end, *num_lockers, err_inval);
 	*lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO);

And I expected to see both modifications. I am slightly confused, frankly speaking.

Thanks,
Slava.

build bot
#4 lore raw

Re: [PATCH v2] ceph: fix bare ceph_decode_8 OOB in decode_lockers()

In reply to Pavitra Jha
BCC: [email protected] CC: [email protected] In-Reply-To: <[email protected]> References: <[email protected]> TO: Pavitra Jha…

BCC: [email protected]
CC: [email protected]
In-Reply-To: <[email protected]>
References: <[email protected]>
TO: Pavitra Jha <[email protected]>
TO: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: [email protected]
CC: Pavitra Jha <[email protected]>

Hi Pavitra,

kernel test robot noticed the following build warnings:

[auto build test WARNING on ceph-client/testing]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Pavitra-Jha/ceph-fix-bare-ceph_decode_8-OOB-in-decode_lockers/20260528-212749
base:   https://github.com/ceph/ceph-client.git testing
patch link:    https://lore.kernel.org/r/20260528132521.843004-1-jhapavitra98%40gmail.com
patch subject: [PATCH v2] ceph: fix bare ceph_decode_8 OOB in decode_lockers()

:::::: branch date: 2 days ago
:::::: commit date: 2 days ago
config: um-randconfig-r073-20260530 (https://download.01.org/0day-ci/archive/20260531/[email protected]/config)
compiler: gcc-14 (Debian 14.2.0-19) 14.2.0
smatch: v0.5.0-9185-gbcc58b9c

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags

| Reported-by: kernel test robot <[email protected]>
| Reported-by: Dan Carpenter <[email protected]>
| Closes: https://lore.kernel.org/r/[email protected]/

smatch warnings:
net/ceph/cls_lock_client.c:313 decode_lockers() warn: missing error code 'ret'

vim +/ret +313 net/ceph/cls_lock_client.c

d4ed4a53056288 Douglas Fuller 2015-06-29  287  
d4ed4a53056288 Douglas Fuller 2015-06-29  288  static int decode_lockers(void **p, void *end, u8 *type, char **tag,
d4ed4a53056288 Douglas Fuller 2015-06-29  289  			  struct ceph_locker **lockers, u32 *num_lockers)
d4ed4a53056288 Douglas Fuller 2015-06-29  290  {
d4ed4a53056288 Douglas Fuller 2015-06-29  291  	u8 struct_v;
d4ed4a53056288 Douglas Fuller 2015-06-29  292  	u32 struct_len;
d4ed4a53056288 Douglas Fuller 2015-06-29  293  	char *s;
d4ed4a53056288 Douglas Fuller 2015-06-29  294  	int i;
d4ed4a53056288 Douglas Fuller 2015-06-29  295  	int ret;
d4ed4a53056288 Douglas Fuller 2015-06-29  296  
d4ed4a53056288 Douglas Fuller 2015-06-29  297  	ret = ceph_start_decoding(p, end, 1, "cls_lock_get_info_reply",
d4ed4a53056288 Douglas Fuller 2015-06-29  298  				  &struct_v, &struct_len);
d4ed4a53056288 Douglas Fuller 2015-06-29  299  	if (ret)
d4ed4a53056288 Douglas Fuller 2015-06-29  300  		return ret;
d4ed4a53056288 Douglas Fuller 2015-06-29  301  
d4ed4a53056288 Douglas Fuller 2015-06-29  302  	*num_lockers = ceph_decode_32(p);
69050f8d6d075d Kees Cook      2026-02-20  303  	*lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO);
d4ed4a53056288 Douglas Fuller 2015-06-29  304  	if (!*lockers)
d4ed4a53056288 Douglas Fuller 2015-06-29  305  		return -ENOMEM;
d4ed4a53056288 Douglas Fuller 2015-06-29  306  
d4ed4a53056288 Douglas Fuller 2015-06-29  307  	for (i = 0; i < *num_lockers; i++) {
d4ed4a53056288 Douglas Fuller 2015-06-29  308  		ret = decode_locker(p, end, *lockers + i);
d4ed4a53056288 Douglas Fuller 2015-06-29  309  		if (ret)
d4ed4a53056288 Douglas Fuller 2015-06-29  310  			goto err_free_lockers;
d4ed4a53056288 Douglas Fuller 2015-06-29  311  	}
d4ed4a53056288 Douglas Fuller 2015-06-29  312  
cff58e4599d8e1 Pavitra Jha    2026-05-28 @313  	ceph_decode_8_safe(p, end, *type, err_free_lockers);
d4ed4a53056288 Douglas Fuller 2015-06-29  314  	s = ceph_extract_encoded_string(p, end, NULL, GFP_NOIO);
d4ed4a53056288 Douglas Fuller 2015-06-29  315  	if (IS_ERR(s)) {
d4ed4a53056288 Douglas Fuller 2015-06-29  316  		ret = PTR_ERR(s);
d4ed4a53056288 Douglas Fuller 2015-06-29  317  		goto err_free_lockers;
d4ed4a53056288 Douglas Fuller 2015-06-29  318  	}
d4ed4a53056288 Douglas Fuller 2015-06-29  319  
d4ed4a53056288 Douglas Fuller 2015-06-29  320  	*tag = s;
d4ed4a53056288 Douglas Fuller 2015-06-29  321  	return 0;
d4ed4a53056288 Douglas Fuller 2015-06-29  322  
d4ed4a53056288 Douglas Fuller 2015-06-29  323  err_free_lockers:
d4ed4a53056288 Douglas Fuller 2015-06-29  324  	ceph_free_lockers(*lockers, *num_lockers);
d4ed4a53056288 Douglas Fuller 2015-06-29  325  	return ret;
d4ed4a53056288 Douglas Fuller 2015-06-29  326  }
d4ed4a53056288 Douglas Fuller 2015-06-29  327  

--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

#5 lore raw

PATCH v3 1/1 ceph: fix two unsafe bare decodes in decode_lockers()

decode_lockers() in cls_lock_client.c contains two bare decode operations that allow a malicious or compromised OSD to trigger slab-out-of-bounds reads:

1. ceph_decode_32(p) at the num_lockers field has no preceding bounds check. ceph_start_decoding() accepts struct_len=0 as valid -- the internal ceph_decode_need(p, end, 0, bad) always passes -- so when an OSD sends struct_len=0, ceph_start_decoding() returns success with p == end. The immediately following bare ceph_decode_32(p) then reads 4 bytes past the validated buffer boundary. The garbage value is passed directly to kzalloc_objs() as the locker count.

The sibling function decode_watchers() in osd_client.c already uses ceph_decode_32_safe() after its own ceph_start_decoding() call. decode_lockers() was the only site using the bare variant.

2. ceph_decode_8(p) after the decode_locker() loop has no preceding bounds check. If an OSD crafts num_lockers such that the loop advances p exactly to end, the subsequent bare ceph_decode_8(p) reads one byte past the validated buffer boundary. The result is passed directly into *type, which is used as a lock type discriminator by callers, giving an OSD-controlled one-byte OOB read with direct influence over the lock type field.

Fix both by replacing bare operations with their safe variants:
ceph_decode_32(p) -> ceph_decode_32_safe(p, end, *num_lockers,
err_inval)
ceph_decode_8(p) -> ceph_decode_8_safe(p, end, *type,
err_free_lockers)

The goto targets differ intentionally:
err_inval: is a new label returning -EINVAL directly. It is used for the pre-allocation failure path where *lockers is not yet allocated and must not be passed to ceph_free_lockers().

err_free_lockers: is the existing label. It is used for the post-allocation failure path where *lockers is allocated and must be freed.

ret is set to -EINVAL before ceph_decode_8_safe() so that err_free_lockers returns the correct error code on bounds violation. Without this, err_free_lockers would return a stale ret value (0 from the successful decode_locker() loop), silently swallowing the error.

-EINVAL is correct for both failure paths. The data received from the OSD is structurally malformed. -ENOMEM would misrepresent the failure class to callers and to stable@ backporters triaging error paths.

KASAN report for bug 1 (kernel 7.0.0-rc7, QEMU/x86_64, KASLR disabled):
==================================================================

  BUG: KASAN: slab-out-of-bounds in ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
  Read of size 4 at addr ffff88800a29b76e by task insmod/58

  CPU: 0 UID: 0 PID: 58 Comm: insmod Tainted: G           O        7.0.0-rc7-g9c2abf69da83-dirty #15 PREEMPT(lazy)
  Tainted: [O]=OOT_MODULE
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x4d/0x70
   print_report+0x170/0x4f3
   kasan_report+0xda/0x110
   ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Allocated by task 58:
   kasan_save_stack+0x30/0x50
   kasan_save_track+0x14/0x30
   __kasan_kmalloc+0x7f/0x90
   ceph_oob3_init+0x4d/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  The buggy address belongs to the object at ffff88800a29a000
   which belongs to the cache kmalloc-8k of size 8192
  The buggy address is located 5998 bytes inside of
   allocated 6000-byte region [ffff88800a29a000, ffff88800a29b770)

  Memory state around the buggy address:
   ffff88800a29b600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
   ffff88800a29b680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  >ffff88800a29b700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc fc
                                                               ^
   ffff88800a29b780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
  ==================================================================

num_lockers=0xccccaaaa (OOB garbage from KASAN redzone)

Bug 2 (ceph_decode_8) follows from the identical precondition. A dedicated PoC is available on request.

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition) without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")

---
v3: Combine both fixes (ceph_decode_32 and ceph_decode_8) into a single
patch per Viacheslav Dubeyko's review. Set ret = -EINVAL before ceph_decode_8_safe() so err_free_lockers returns the correct error code, not stale ret (caught by Dan Carpenter / smatch). Clarify err_inval vs err_free_lockers goto selection rationale and -EINVAL justification.

---
 net/ceph/cls_lock_client.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
net/ceph/cls_lock_client.c +5 -2
index c6956f1df..4e6a6d3e4 100644
@@ -299,7 +299,7 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 	if (ret)
 		return ret;
 
-	*num_lockers = ceph_decode_32(p);
+	ceph_decode_32_safe(p, end, *num_lockers, err_inval);
 	*lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO);
 	if (!*lockers)
 		return -ENOMEM;
@@ -310,7 +310,8 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 			goto err_free_lockers;
 	}
 
-	*type = ceph_decode_8(p);
+	ret = -EINVAL;
+	ceph_decode_8_safe(p, end, *type, err_free_lockers);
 	s = ceph_extract_encoded_string(p, end, NULL, GFP_NOIO);
 	if (IS_ERR(s)) {
 		ret = PTR_ERR(s);
@@ -320,6 +321,8 @@ static int decode_lockers(void **p, void *end, u8 *type, char **tag,
 	*tag = s;
 	return 0;
 
+err_inval:
+	return -EINVAL;
 err_free_lockers:
 	ceph_free_lockers(*lockers, *num_lockers);
 	return ret;
Signature · 1 line
2.53.0
maintainer Reviewed-by on v3 1/1
#6 lore raw

Re: [PATCH v3] ceph: fix two unsafe bare decodes in decode_lockers()

Quoted 168 lines · Pavitra Jha

decode_lockers() in cls_lock_client.c contains two bare decode operations
that allow a malicious or compromised OSD to trigger slab-out-of- bounds
reads:

1. ceph_decode_32(p) at the num_lockers field has no preceding bounds check. ceph_start_decoding() accepts struct_len=0 as valid -- the internal ceph_decode_need(p, end, 0, bad) always passes -- so when an
   OSD sends struct_len=0, ceph_start_decoding() returns success with p == end. The immediately following bare ceph_decode_32(p) then reads
   4 bytes past the validated buffer boundary. The garbage value is passed directly to kzalloc_objs() as the locker count.

   The sibling function decode_watchers() in osd_client.c already uses
   ceph_decode_32_safe() after its own ceph_start_decoding() call. decode_lockers() was the only site using the bare variant.

2. ceph_decode_8(p) after the decode_locker() loop has no preceding bounds check. If an OSD crafts num_lockers such that the loop advances p exactly to end, the subsequent bare ceph_decode_8(p) reads
   one byte past the validated buffer boundary. The result is passed directly into *type, which is used as a lock type discriminator by callers, giving an OSD-controlled one-byte OOB read with direct influence over the lock type field.

Fix both by replacing bare operations with their safe variants:
  ceph_decode_32(p) -> ceph_decode_32_safe(p, end, *num_lockers, err_inval) ceph_decode_8(p)  -> ceph_decode_8_safe(p, end, *type, err_free_lockers)

The goto targets differ intentionally:
  err_inval: is a new label returning -EINVAL directly. It is used for
  the pre-allocation failure path where *lockers is not yet allocated and must not be passed to ceph_free_lockers().

  err_free_lockers: is the existing label. It is used for the post-allocation failure path where *lockers is allocated and must be freed.

ret is set to -EINVAL before ceph_decode_8_safe() so that err_free_lockers returns the correct error code on bounds violation. Without this, err_free_lockers would return a stale ret value (0 from the successful decode_locker() loop), silently swallowing the error.

-EINVAL is correct for both failure paths. The data received from the OSD is structurally malformed. -ENOMEM would misrepresent the failure class to callers and to stable@ backporters triaging error paths.

KASAN report for bug 1 (kernel 7.0.0-rc7, QEMU/x86_64, KASLR disabled):
  ==================================================================

  BUG: KASAN: slab-out-of-bounds in ceph_oob3_init+0x251/0xff0
[ceph_oob3_poc]
  Read of size 4 at addr ffff88800a29b76e by task insmod/58

  CPU: 0 UID: 0 PID: 58 Comm: insmod Tainted: G           O       
7.0.0-rc7-g9c2abf69da83-dirty #15 PREEMPT(lazy)
  Tainted: [O]=OOT_MODULE
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.17.0-
debian-1.17.0-1 04/01/2014
  Call Trace:
   <TASK>
   dump_stack_lvl+0x4d/0x70
   print_report+0x170/0x4f3
   kasan_report+0xda/0x110
   ceph_oob3_init+0x251/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  Allocated by task 58:
   kasan_save_stack+0x30/0x50
   kasan_save_track+0x14/0x30
   __kasan_kmalloc+0x7f/0x90
   ceph_oob3_init+0x4d/0xff0 [ceph_oob3_poc]
   do_one_initcall+0x9a/0x3a0
   do_init_module+0x27c/0x790
   load_module+0x4a9a/0x6350
   init_module_from_file+0x15c/0x180
   idempotent_init_module+0x21f/0x750
   __x64_sys_finit_module+0xba/0x120
   do_syscall_64+0xe2/0x570
   entry_SYSCALL_64_after_hwframe+0x77/0x7f

  The buggy address belongs to the object at ffff88800a29a000

   which belongs to the cache kmalloc-8k of size 8192 The buggy address is located 5998 bytes inside of allocated 6000-byte region [ffff88800a29a000, ffff88800a29b770)

  Memory state around the buggy address:
   ffff88800a29b600: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
   ffff88800a29b680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  >ffff88800a29b700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc fc

                                                               ^
   ffff88800a29b780: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
  ==================================================================

  num_lockers=0xccccaaaa (OOB garbage from KASAN redzone)

Bug 2 (ceph_decode_8) follows from the identical precondition. A dedicated PoC is available on request.

Attacker model: a malicious or compromised OSD in a multi-tenant Ceph deployment can trigger this against any kernel client that issues the lock.get_info class method (e.g. during RBD exclusive lock acquisition)
without any further privileges beyond OSD session establishment.

Fixes d4ed4a530562 ("libceph: support for lock.lock_info")

---
v3: Combine both fixes (ceph_decode_32 and ceph_decode_8) into a single
    patch per Viacheslav Dubeyko's review. Set ret = -EINVAL before ceph_decode_8_safe() so err_free_lockers returns the correct error
    code, not stale ret (caught by Dan Carpenter / smatch). Clarify err_inval vs err_free_lockers goto selection rationale and -EINVAL justification.
---
 net/ceph/cls_lock_client.c | 7 +++++--

 1 file changed, 5 insertions(+), 2 deletions(-)
net/ceph/cls_lock_client.c +0 -0
index c6956f1df..4e6a6d3e4 100644
@@ -299,7 +299,7 @@ static int decode_lockers(void **p, void *end, u8
 *type, char **tag,
  	if (ret)
  		return ret;

net/ceph/cls_lock_client.c +1 -1
-	*num_lockers = ceph_decode_32(p);
+	ceph_decode_32_safe(p, end, *num_lockers, err_inval);

  *lockers = kzalloc_objs(**lockers, *num_lockers, GFP_NOIO); if (!*lockers)
  return -ENOMEM;

net/ceph/cls_lock_client.c +0 -0
@@ -310,7 +310,8 @@ static int decode_lockers(void **p, void *end, u8
 *type, char **tag,
  			goto err_free_lockers;
  	}

net/ceph/cls_lock_client.c +2 -1
-	*type = ceph_decode_8(p);
+	ret = -EINVAL;
+	ceph_decode_8_safe(p, end, *type, err_free_lockers);

  s = ceph_extract_encoded_string(p, end, NULL, GFP_NOIO); if (IS_ERR(s)) {
  ret = PTR_ERR(s);

net/ceph/cls_lock_client.c +0 -0
@@ -320,6 +321,8 @@ static int decode_lockers(void **p, void *end, u8
 *type, char **tag,
  	*tag = s;
  	return 0;

net/ceph/cls_lock_client.c +2 -0
+err_inval:
+	return -EINVAL;
Quoted 3 lines

 err_free_lockers:
  ceph_free_lockers(*lockers, *num_lockers); return ret;

Looks good.

Thanks,
Slava.

#7 lore raw
In reply to Viacheslav Dubeyko

The follow-up patch titled:

"[PATCH v2] ceph: fix bare ceph_decode_8 OOB in decode_lockers()"

was intended as the continuation of this review thread and addresses the additional bare decode site pointed out by Slava here:

*type = ceph_decode_8(p);

I accidentally sent it as a separate thread instead of a reply to this discussion.

Thanks,
Pavitra