scsi: qedf: sanitise uaccess
Subsystems (2)
cover scsi: qedf: sanitise uaccess
qedf driver, debugfs part of it specifically, touches __user pointers directly for printing out info to userspace via sprintf(), which may cause crash like this:
BUG: unable to handle kernel paging request at 00007ffd1d6b43a0 IP: [<ffffffffaa7a882a>] string.isra.7+0x6a/0xf0 Oops: 0003 [#1] SMP Call Trace: [<ffffffffaa7a9f31>] vsnprintf+0x201/0x6a0 [<ffffffffaa7aa556>] sprintf+0x56/0x80 [<ffffffffc04227ed>] qedf_dbg_stop_io_on_error_cmd_read+0x6d/0x90 [qedf] [<ffffffffaa65bb2f>] vfs_read+0x9f/0x170 [<ffffffffaa65cb82>] SyS_pread64+0x92/0xc0
Avoid this by preparing the info in a kernel buffer first, either allocated on stack for small printouts, or via vmalloc() for big ones, and then copying it to the userspace properly.
Previous submission is an RFC: [1]. There are no code changes since then. The RFC prefix is dropped. The Tested-by tag from Laurence is added.
There's similar submission from Saurav [2], but we agreed I could nack it and proceed with my one.
[1] https://lore.kernel.org/linux-scsi/[email protected]/
[2] https://lore.kernel.org/linux-scsi/[email protected]/
Oleksandr Natalenko (3):
scsi: qedf: do not touch __user pointer in
qedf_dbg_stop_io_on_error_cmd_read() directly
scsi: qedf: do not touch __user pointer in qedf_dbg_debug_cmd_read()
directly
scsi: qedf: do not touch __user pointer in qedf_dbg_fp_int_cmd_read()
directly
drivers/scsi/qedf/qedf_dbg.h | 2 ++ drivers/scsi/qedf/qedf_debugfs.c | 35 +++++++++++++++++++------------- 2 files changed, 23 insertions(+), 14 deletions(-)
Signature · 1 line
2.41.0
PATCH 1/3 scsi: qedf: do not touch __user pointer in qedf_dbg_stop_io_on_error_cmd_read() directly
The qedf_dbg_stop_io_on_error_cmd_read() function invokes sprintf() directly on a __user pointer, which may crash the kernel.
Avoid doing that by using a small on-stack buffer for sprintf() and then calling simple_read_from_buffer() which does a proper copy_to_user() call.
--- drivers/scsi/qedf/qedf_debugfs.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/scsi/qedf/qedf_debugfs.c +3 -4
@@ -185,18 +185,17 @@ qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) { int cnt; +char cbuf[7 ];struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;struct qedf_ctx *qedf =container_of (qedf_dbg, struct qedf_ctx, dbg_ctx);QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"entered\n" ); - cnt =sprintf (buffer,"%s\n" , + cnt =sprintf (cbuf,"%s\n" , qedf->stop_io_on_error ?"true" :"false" ); - cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt); } static ssize_t
Signature · 1 line
2.41.0
RE: [PATCH 1/3] scsi: qedf: do not touch __user pointer in qedf_dbg_stop_io_on_error_cmd_read() directly
From: Oleksandr Natalenko
Quoted 10 lines
Sent: 28 July 2023 07:58
The
qedf_dbg_stop_io_on_error_cmd_read()function invokessprintf()directly on a__userpointer, which may crash the kernel.Avoid doing that by using a small on-stack buffer for
sprintf()and then callingsimple_read_from_buffer()which does a propercopy_to_user()call.Fixes 61d8658b4a ("scsi: qedf: Add QLogic FastLinQ offload FCoE driver framework.")
...
Quoted 8 lines
drivers/scsi/qedf/qedf_debugfs.c +0
-0@@ -185,18 +185,17 @@ qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) { int cnt;
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, · line 196
Quoted 10 lines
drivers/scsi/qedf/qedf_debugfs.c +2
-1+char cbuf[7 ];struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;struct qedf_ctx *qedf =container_of (qedf_dbg, struct qedf_ctx, dbg_ctx);QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"entered\n" ); - cnt =sprintf (buffer,"%s\n" , + cnt =sprintf (cbuf,"%s\n" , qedf->stop_io_on_error ?"true" :"false" );
You've made cbuf[] exactly just big enough. If anyone breathes on this code it could overflow. You really should use scnprintf() for safety.
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, · line 198
Quoted 5 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-3- cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt);
Or just:
if (gedf->stop_on_error) return simple_read_from_buffer(buffer, count, ppos, "true\n", 5); return simple_read_from_buffer(buffer, count, ppos, "false\n", 6); David
Quoted 5 lines
drivers/scsi/qedf/qedf_debugfs.c +0
-1} static ssize_t --2.41.0
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Re: [PATCH 1/3] scsi: qedf: do not touch __user pointer in qedf_dbg_stop_io_on_error_cmd_read() directly
Hello.
Quoted 26 lines · On pátek 28. července 2023 17:23:25 CEST David Laight wrote:
From: Oleksandr Natalenko
Quoted 10 lines
Sent: 28 July 2023 07:58
The
qedf_dbg_stop_io_on_error_cmd_read()function invokessprintf()directly on a__userpointer, which may crash the kernel.Avoid doing that by using a small on-stack buffer for
sprintf()and then callingsimple_read_from_buffer()which does a propercopy_to_user()call.Fixes 61d8658b4a ("scsi: qedf: Add QLogic FastLinQ offload FCoE driver framework.")...
Quoted 14 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-0@@ -185,18 +185,17 @@ qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) { int cnt; +char cbuf[7 ];struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;struct qedf_ctx *qedf =container_of (qedf_dbg, struct qedf_ctx, dbg_ctx);
Quoted 8 lines
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, · line 196
Quoted 4 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-1QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"entered\n" ); - cnt =sprintf (buffer,"%s\n" , + cnt =sprintf (cbuf,"%s\n" , qedf->stop_io_on_error ?"true" :"false" );You've made cbuf[] exactly just big enough. If anyone breathes on this code it could overflow. You really should use
scnprintf()for safety.
OK, I'll do scnprintf(cbuf, sizeof(cbuf), ...) in the next version of the submission.
Thanks.
Quoted 25 lines
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_stop_io_on_error_cmd_read(struct file *filp, char __user *buffer, · line 198
Quoted 5 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-3- cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt);Or just:
if (gedf->stop_on_error) return simple_read_from_buffer(buffer, count, ppos, "true\n", 5); return simple_read_from_buffer(buffer, count, ppos, "false\n", 6); DavidQuoted 5 lines
drivers/scsi/qedf/qedf_debugfs.c +0
-1} static ssize_t --2.41.0
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Signature · 2 lines
Oleksandr Natalenko (post-factum) Principal Software Maintenance Engineer
PATCH 2/3 scsi: qedf: do not touch __user pointer in qedf_dbg_debug_cmd_read() directly
In reply to Oleksandr NatalenkoThe qedf_dbg_debug_cmd_read() function invokes sprintf() directly on a __user pointer, which may crash the kernel.
Avoid doing that by using a small on-stack buffer for sprintf() and then calling simple_read_from_buffer() which does a proper copy_to_user() call.
--- drivers/scsi/qedf/qedf_debugfs.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/scsi/qedf/qedf_debugfs.c +3 -4
@@ -138,15 +138,14 @@ qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) {int cnt; +char cbuf[35 ];struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"debug mask=0x%x\n" , qedf_debug); - cnt =sprintf (buffer,"debug mask = 0x%x\n" , qedf_debug); + cnt =sprintf (cbuf,"debug mask = 0x%x\n" , qedf_debug); - cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt); } static ssize_t
Signature · 1 line
2.41.0
RE: [PATCH 2/3] scsi: qedf: do not touch __user pointer in qedf_dbg_debug_cmd_read() directly
From: Oleksandr Natalenko
Quoted 4 lines
Sent: 28 July 2023 07:58
The
qedf_dbg_debug_cmd_read()function invokessprintf()directly on a__userpointer, which may crash the kernel.
^^^ will
Quoted 4 lines
Avoid doing that by using a small on-stack buffer for
sprintf()and then callingsimple_read_from_buffer()which does a propercopy_to_user()call.
...
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, · line 141
Quoted 9 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-0@@ -138,15 +138,14 @@ qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) {int cnt; +char cbuf[35 ];
Why 35?
I pick a multiple of 8 that if 'enough.
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, · line 146
Quoted 6 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-1struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"debug mask=0x%x\n" , qedf_debug); - cnt =sprintf (buffer,"debug mask = 0x%x\n" , qedf_debug); + cnt =sprintf (cbuf,"debug mask = 0x%x\n" , qedf_debug);
Use scnprintf() to be sure it doesn't overflow. Much safer if someone does a quick update or copies the code.
David
Quoted 10 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-4- cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt); } static ssize_t --2.41.0
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Re: [PATCH 2/3] scsi: qedf: do not touch __user pointer in qedf_dbg_debug_cmd_read() directly
Hello/
Quoted 6 lines · On pátek 28. července 2023 17:26:11 CEST David Laight wrote:
From: Oleksandr Natalenko
Quoted 4 lines
Sent: 28 July 2023 07:58
The
qedf_dbg_debug_cmd_read()function invokessprintf()directly on a__userpointer, which may crash the kernel.^^^ will
I don't think it is 100% guaranteed, but for sure this is not a correct behaviour.
Quoted 5 lines
Quoted 4 lines
Avoid doing that by using a small on-stack buffer for
sprintf()and then callingsimple_read_from_buffer()which does a propercopy_to_user()call....
Quoted 12 lines
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, · line 141
Quoted 9 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-0@@ -138,15 +138,14 @@ qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) {int cnt; +char cbuf[35 ];Why 35?
I pick a multiple of 8 that if 'enough.
OK, I overestimated this, it should have been 27, but I'll make it 32 to be a multiple of 8.
Thanks.
Quoted 28 lines
drivers/scsi/qedf/qedf_debugfs.c qedf_dbg_debug_cmd_read(struct file *filp, char __user *buffer, size_t count, · line 146
Quoted 6 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-1struct qedf_dbg_ctx *qedf_dbg = (struct qedf_dbg_ctx *)filp->private_data;QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"debug mask=0x%x\n" , qedf_debug); - cnt =sprintf (buffer,"debug mask = 0x%x\n" , qedf_debug); + cnt =sprintf (cbuf,"debug mask = 0x%x\n" , qedf_debug);Use
scnprintf()to be sure it doesn't overflow. Much safer if someone does a quick update or copies the code.David
Quoted 10 lines
drivers/scsi/qedf/qedf_debugfs.c +1
-4- cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; +return simple_read_from_buffer (buffer, count, ppos, cbuf, cnt); } static ssize_t --2.41.0
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)
Signature · 2 lines
Oleksandr Natalenko (post-factum) Principal Software Maintenance Engineer
PATCH 3/3 scsi: qedf: do not touch __user pointer in qedf_dbg_fp_int_cmd_read() directly
In reply to Oleksandr NatalenkoThe qedf_dbg_fp_int_cmd_read() function invokes sprintf() directly on a __user pointer, which may crash the kernel.
Avoid doing that by vmalloc()'ating a buffer for scnprintf() and then calling simple_read_from_buffer() which does a proper copy_to_user() call.
--- drivers/scsi/qedf/qedf_dbg.h | 2 ++ drivers/scsi/qedf/qedf_debugfs.c | 21 +++++++++++++++------ 2 files changed, 17 insertions(+), 6 deletions(-)
drivers/scsi/qedf/qedf_dbg.h +2 -0
@@ -59,6 +59,8 @@ extern uint qedf_debug;#define QEDF_LOG_NOTICE 0x40000000/* Notice logs */ #define QEDF_LOG_WARN 0x80000000/* Warning logs */ +#define QEDF_DEBUGFS_LOG_LEN (2 * PAGE_SIZE) +/* Debug context structure */ struct qedf_dbg_ctx {unsigned int host_no;
drivers/scsi/qedf/qedf_debugfs.c +15 -6
@@ -8,6 +8,7 @@#include <linux/uaccess.h> #include <linux/debugfs.h> #include <linux/module.h> +#include <linux/vmalloc.h> #include "qedf.h" #include "qedf_dbg.h"
@@ -98,7 +99,9 @@ static ssize_t qedf_dbg_fp_int_cmd_read(struct file *filp, char __user *buffer, size_t count, loff_t *ppos) { + ssize_t ret;size_t cnt =0 ; +char *cbuf;int id;struct qedf_fastpath *fp =NULL ; struct qedf_dbg_ctx *qedf_dbg =
@@ -108,19 +111,25 @@ qedf_dbg_fp_int_cmd_read(struct file *filp, char __user *buffer, size_t count,QEDF_INFO (qedf_dbg,QEDF_LOG_DEBUGFS ,"entered\n" ); - cnt =sprintf (buffer,"\nFastpath I/O completions\n\n" ); + cbuf =vmalloc (QEDF_DEBUGFS_LOG_LEN ); +if (!cbuf) +return 0 ; + + cnt +=scnprintf (cbuf + cnt,QEDF_DEBUGFS_LOG_LEN - cnt,"\nFastpath I/O completions\n\n" );for (id =0 ; id < qedf->num_queues; id++) { fp = &(qedf->fp_array[id]);if (fp->sb_id ==QEDF_SB_ID_NULL )continue ; - cnt +=sprintf ((buffer + cnt),"#%d: %lu\n" , id, - fp->completions); + cnt +=scnprintf (cbuf + cnt,QEDF_DEBUGFS_LOG_LEN - cnt, +"#%d: %lu\n" , id, fp->completions); } - cnt =min_t (int, count, cnt - *ppos); - *ppos += cnt; -return cnt; + ret =simple_read_from_buffer (buffer, count, ppos, cbuf, cnt); + +vfree (cbuf); + +return ret; } static ssize_t
Signature · 1 line
2.41.0
Quoted 36 lines · On pátek 28. července 2023 8:58:16 CEST Oleksandr Natalenko wrote:
qedf driver, debugfs part of it specifically, touches
__userpointers directly for printing out info to userspace viasprintf(), which may cause crash like this:BUG: unable to handle kernel paging request at 00007ffd1d6b43a0 IP: [<ffffffffaa7a882a>] string.isra.7+0x6a/0xf0 Oops: 0003 [#1] SMP Call Trace: [<ffffffffaa7a9f31>] vsnprintf+0x201/0x6a0 [<ffffffffaa7aa556>] sprintf+0x56/0x80 [<ffffffffc04227ed>] qedf_dbg_stop_io_on_error_cmd_read+0x6d/0x90 [qedf] [<ffffffffaa65bb2f>] vfs_read+0x9f/0x170 [<ffffffffaa65cb82>] SyS_pread64+0x92/0xc0Avoid this by preparing the info in a kernel buffer first, either allocated on stack for small printouts, or via
vmalloc()for big ones, and then copying it to the userspace properly.Previous submission is an RFC: [1]. There are no code changes since then. The RFC prefix is dropped. The Tested-by tag from Laurence is added.
There's similar submission from Saurav [2], but we agreed I could nack it and proceed with my one.
[1] https://lore.kernel.org/linux-scsi/[email protected]/
[2] https://lore.kernel.org/linux-scsi/[email protected]/Oleksandr Natalenko (3):
scsi: qedf: do not touch__userpointer in
qedf_dbg_stop_io_on_error_cmd_read()directly
scsi: qedf: do not touch__userpointer inqedf_dbg_debug_cmd_read()
directly
scsi: qedf: do not touch__userpointer inqedf_dbg_fp_int_cmd_read()
directly
Quoted 5 lines
drivers/scsi/qedf/qedf_dbg.h | 2 ++ drivers/scsi/qedf/qedf_debugfs.c | 35 +++++++++++++++++++------------- 2 files changed, 23 insertions(+), 14 deletions(-)
Oops, I forgot to add:
as per [1].
My ask to the maintainer to add it if the submission is accepted, or let me know if I should do a v2 instead.
[1] https://lore.kernel.org/linux-scsi/[email protected]/
Signature · 2 lines
Oleksandr Natalenko (post-factum) Principal Software Maintenance Engineer
Quoted 70 lines
-----Original Message-----
From: Oleksandr Natalenko <[email protected]>
Sent: Friday, July 28, 2023 12:28 PM
To: [email protected]
Cc: [email protected]; Saurav Kashyap <[email protected]>; Johannes Thumshirn <[email protected]>; GR-QLogic-Storage- Upstream <[email protected]>; James E.J. Bottomley <[email protected]>; Martin K. Petersen <[email protected]>; Jozef Bacik <[email protected]>; Laurence Oberman <[email protected]>; Rob Evers <[email protected]>
Subject: [EXT] [PATCH 0/3] scsi: qedf: sanitise uaccessExternal Email
----------------------------------------------------------------------
qedf driver, debugfs part of it specifically, touches__userpointers directly for printing out info to userspace viasprintf(), which may cause crash like this:BUG: unable to handle kernel paging request at 00007ffd1d6b43a0 IP: [<ffffffffaa7a882a>] string.isra.7+0x6a/0xf0 Oops: 0003 [#1] SMP Call Trace: [<ffffffffaa7a9f31>] vsnprintf+0x201/0x6a0 [<ffffffffaa7aa556>] sprintf+0x56/0x80 [<ffffffffc04227ed>] qedf_dbg_stop_io_on_error_cmd_read+0x6d/0x90 [qedf] [<ffffffffaa65bb2f>] vfs_read+0x9f/0x170 [<ffffffffaa65cb82>] SyS_pread64+0x92/0xc0Avoid this by preparing the info in a kernel buffer first, either allocated on stack for small printouts, or via
vmalloc()for big ones, and then copying it to the userspace properly.Previous submission is an RFC: [1]. There are no code changes since then. The RFC prefix is dropped. The Tested-by tag from Laurence is added.
There's similar submission from Saurav [2], but we agreed I could nack it and proceed with my one.
[1] https://urldefense.proofpoint.com/v2/url?u=https-
3A__lore.kernel.org_linux-2Dscsi_20230724120241.40495-2D1-2Doleksandr- 40redhat.com_&d=DwIDAg&c=nKjWec2b6R0mOyPaz7xtfQ&r=ZHZbmY_LbM3 DUZK_BDO1OITP3ot_Vkb_5w- gas5TBMQ&m=I6CdYJNbvw1q9OWTYmMTCWzCcXFG7MelqlMZ_DmhEMDeW ViVj2b3_EadDZUUdzNT&s=joHzWsadjq2HUMSxGvJZMkLQULriRotQk2RYPgDW pWc&e=
[2] https://urldefense.proofpoint.com/v2/url?u=https-
3A__lore.kernel.org_linux-2Dscsi_20230726101236.11922-2D1-2Dskashyap- 40marvell.com_&d=DwIDAg&c=nKjWec2b6R0mOyPaz7xtfQ&r=ZHZbmY_LbM 3DUZK_BDO1OITP3ot_Vkb_5w- gas5TBMQ&m=I6CdYJNbvw1q9OWTYmMTCWzCcXFG7MelqlMZ_DmhEMDeW ViVj2b3_EadDZUUdzNT&s=q2LVoTVsEfj1rCnih48VaDUxOCOobRLhIaTatga1qm Q&e=Oleksandr Natalenko (3):
scsi: qedf: do not touch__userpointer in
qedf_dbg_stop_io_on_error_cmd_read()directly
scsi: qedf: do not touch__userpointer inqedf_dbg_debug_cmd_read()
directly
scsi: qedf: do not touch__userpointer inqedf_dbg_fp_int_cmd_read()
directlydrivers/scsi/qedf/qedf_dbg.h | 2 ++ drivers/scsi/qedf/qedf_debugfs.c | 35 +++++++++++++++++++------------- 2 files changed, 23 insertions(+), 14 deletions(-)--
2.41.0
Looks good,