Skip to contentSkip to conversation

pstore: fix two ramoops error-path bugs

PATCH Applied Andrea Parri 2 participants 7 messages Sep 21

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

cover pstore: fix two ramoops error-path bugs

Two bugs in the ramoops error paths, found by review and confirmed with test-only reproducers against an unfixed kernel:

- ramoops_init_prz() stores persistent_ram_new()'s ERR_PTR in the zone pointer and returns without clearing it; the probe error path then passes it to persistent_ram_free(), which dereferences it and oopses.

- persistent_ram_buffer_map() computes buffer_size = size - sizeof(struct persistent_ram_buffer) with no lower bound, so a zone no larger than the header underflows buffer_size to a huge value and defeats the bounds checks in the write paths.

The second patch adds a new failure return from persistent_ram_new() for undersized zones, which is exactly the error path that exposes the first
bug: a tiny zone now leaves an ERR_PTR behind on cleanup. Applied without the first patch, it turns the memory-corruption bug into a probe-time panic on the same input, so the ERR_PTR fix is sent first.

Andrea Parri (2):
pstore: don't leave an ERR_PTR in the ramoops zone pointer
pstore: reject persistent RAM zones too small for the buffer header

 fs/pstore/ram.c      | 1 +
 fs/pstore/ram_core.c | 6 ++++++
 2 files changed, 7 insertions(+)
Signature · 1 line
2.53.0
#1 lore raw

PATCH 1/2 pstore: don't leave an ERR_PTR in the ramoops zone pointer

ramoops_init_prz() stores the result of persistent_ram_new() in *prz and returns the error without clearing it when initialization fails. ramoops_probe() then cleans up through ramoops_free_przs(), which hands *prz to persistent_ram_free(); that treats the ERR_PTR as a valid zone and dereferences it, crashing the kernel.

Clear *prz before returning so that cleanup sees a NULL zone.

Fixes b5d38e9bf1b0c ("pstore/ram: Add console messages handling")
Assisted-by LLM
---
 fs/pstore/ram.c | 1 +
 1 file changed, 1 insertion(+)
fs/pstore/ram.c +1 -0
index 2eb0d4fa21869..aa056d0395a4d 100644
@@ -604,6 +604,7 @@ static int ramoops_init_prz(const char *name,
 	if (IS_ERR(*prz)) {
 		int err = PTR_ERR(*prz);
 
+		*prz = NULL;
 		dev_err(dev, "failed to request %s mem region (0x%zx@0x%llx): %d\n",
 			name, sz, (unsigned long long)*paddr, err);
 		return err;
Signature · 1 line
2.53.0
#2 lore raw

PATCH 2/2 pstore: reject persistent RAM zones too small for the buffer header

In reply to Andrea Parri

persistent_ram_buffer_map() computes

	prz->buffer_size = size - sizeof(struct persistent_ram_buffer);

without checking that size is large enough to hold the header. A ramoops zone configured with a tiny size (e.g. ramoops.pmsg_size=1) underflows buffer_size to a huge value, which defeats every bounds check in persistent_ram_write() and persistent_ram_write_user(); a write to /dev/pmsg0 then copies user data far past the end of the zone and corrupts kernel memory.

Reject zones that are not larger than the buffer header.

Fixes 404a6043385de ("staging: android: persistent_ram: handle reserving and mapping memory")
Assisted-by LLM
---
 fs/pstore/ram_core.c | 6 ++++++
 1 file changed, 6 insertions(+)
fs/pstore/ram_core.c +6 -0
index b4ec8009b8dc9..ae941777037b8 100644
@@ -505,6 +505,12 @@ static int persistent_ram_buffer_map(phys_addr_t start, phys_addr_t size,
 	prz->paddr = start;
 	prz->size = size;
 
+	if (size <= sizeof(struct persistent_ram_buffer)) {
+		pr_err("%s: persistent ram zone too small (%#llx bytes)\n",
+		       prz->label, (unsigned long long)size);
+		return -EINVAL;
+	}
+
 	if (pfn_valid(start >> PAGE_SHIFT))
 		prz->vaddr = persistent_ram_vmap(start, size, memtype);
 	else
Signature · 1 line
2.53.0
maintainer
#3 lore raw

Re: [PATCH 2/2] pstore: reject persistent RAM zones too small for the buffer header

Quoted 26 lines · Andrea Parri

persistent_ram_buffer_map() computes

	prz->buffer_size = size - sizeof(struct persistent_ram_buffer);

without checking that size is large enough to hold the header. A ramoops zone configured with a tiny size (e.g. ramoops.pmsg_size=1) underflows buffer_size to a huge value, which defeats every bounds check in persistent_ram_write() and persistent_ram_write_user(); a write to /dev/pmsg0 then copies user data far past the end of the zone and corrupts kernel memory.

Reject zones that are not larger than the buffer header.

Fixes 404a6043385de ("staging: android: persistent_ram: handle reserving and mapping memory")
Assisted-by LLM
---
 fs/pstore/ram_core.c | 6 ++++++
 1 file changed, 6 insertions(+)
fs/pstore/ram_core.c +0 -0
index b4ec8009b8dc9..ae941777037b8 100644

fs/pstore/ram_core.c static int persistent_ram_buffer_map(phys_addr_t start, phys_addr_t size, · line 512

Quoted 8 lines
fs/pstore/ram_core.c +5 -0
@@ -505,6 +505,12 @@ static int persistent_ram_buffer_map(phys_addr_t start, phys_addr_t size,
 	prz->paddr = start;
 	prz->size = size;
 
+	if (size <= sizeof(struct persistent_ram_buffer)) {
+		pr_err("%s: persistent ram zone too small (%#llx bytes)\n",
+		       prz->label, (unsigned long long)size);
+		return -EINVAL;
+	}

Whoops. Luckily it's all system-owner or hardware defined (hardware probes, and boot params).

I wonder if it should return ENOSPC instead, though?

Quoted 7 lines
fs/pstore/ram_core.c +1 -0
+
 	if (pfn_valid(start >> PAGE_SHIFT))
 		prz->vaddr = persistent_ram_vmap(start, size, memtype);
 	else
Signature · 1 line
2.53.0

Thanks for finding this!

Signature · 1 line
Kees Cook
#4 lore raw

Re: [PATCH 2/2] pstore: reject persistent RAM zones too small for the buffer header

(Dropped [email protected] and [email protected] from Cc, both bounced.)

Quoted 1 line

I wonder if it should return ENOSPC instead, though?

Fair point: nothing branches on the specific errno, but ENOSPC matches the "doesn't fit" convention used elsewhere in ram.c, while EINVAL is otherwise reserved for malformed config values. I'll send a one-line patch on top of 1b30df8662ac switching it, unless you'd rather fold it in yourself.

Thanks!

Andrea

maintainer
#5 lore raw

Re: [PATCH 2/2] pstore: reject persistent RAM zones too small for the buffer header

Quoted 2 lines · Andrea Parri

(Dropped [email protected] and [email protected] from Cc, both bounced.)

AFAICT, Anton isn't involved any more. But, hey John, you need a .mailmap entry. :)

Quoted 7 lines
Quoted 1 line

I wonder if it should return ENOSPC instead, though?

Fair point: nothing branches on the specific errno, but ENOSPC matches the "doesn't fit" convention used elsewhere in ram.c, while EINVAL is otherwise reserved for malformed config values. I'll send a one-line patch on top of 1b30df8662ac switching it, unless you'd rather fold it in yourself.

I'm happy to fold it in. Thank you!

-Kees

Signature · 1 line
Kees Cook
maintainer Applied
#6 lore raw
In reply to Andrea Parri
Quoted 8 lines · Andrea Parri

Two bugs in the ramoops error paths, found by review and confirmed with test-only reproducers against an unfixed kernel:

- ramoops_init_prz() stores persistent_ram_new()'s ERR_PTR in the zone pointer and returns without clearing it; the probe error path then passes it to persistent_ram_free(), which dereferences it and oopses.

[...]

Applied to for-next/pstore, thanks!

[1/2] pstore: don't leave an ERR_PTR in the ramoops zone pointer

      https://git.kernel.org/kees/c/70fec428c845
[2/2] pstore: reject persistent RAM zones too small for the buffer header
      https://git.kernel.org/kees/c/1b30df8662ac

Take care,

Signature · 1 line
Kees Cook