i2ctransfer: Don't free memory which was never allocated
PATCH 1/3 i2ctransfer: Don't free memory which was never allocated
If an error occurs while msgs[] is been prepared for the transfer, we jump to the clean-up path. How many buffers need to be freed depends on the state. If we were parsing data, we should free up to nmsgs. However, if we were parsing descriptors, we should free up to nmsgs - 1 only. The code was unconditionally freeing up to nmsgs, potentially freeing a non-allocated buffer.
In most cases, it was not a problem, we would simply call free() on a NULL pointer and that's a no-op. However, if msgs[] was full then we would access memory beyond its end and call free() on a random pointer.
--- tools/i2ctransfer.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
tools/i2ctransfer.c +7 -1
@@ -364,7 +364,13 @@ int main(int argc, char *argv[]) err_out:close (file); -for (i =0 ; i <= nmsgs; i++) +/* +* If we were parsing data, the buffer for the last message was +* already allocated and nmsgs still points to it. +*/ +if (state ==PARSE_GET_DATA ) +free (msgs[nmsgs].buf); +for (i =0 ; i < nmsgs; i++)free (msgs[i].buf);exit (1 );
Signature · 2 lines
Jean Delvare SUSE L3 Support
PATCH 2/3 i2ctransfer: Prevent msgs[] overflow with many parameters
There's an off-by-one bug in the message count check to ensure that we do not process more messages than the kernel allows. nmsgs points to the index within msgs[] which would be used for the _next_ message. If this index is equal the maximum number of messages then we must stop already.
This closes bug #220112:
https://bugzilla.kernel.org/show_bug.cgi?id=220112
--- tools/i2ctransfer.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
tools/i2ctransfer.c +1 -1
@@ -193,7 +193,7 @@ int main(int argc, char *argv[])__u8 data, *buf;char *end; -if (nmsgs >I2C_RDRW_IOCTL_MAX_MSGS ) { +if (nmsgs ==I2C_RDRW_IOCTL_MAX_MSGS ) {fprintf (stderr,"Error: Too many messages (max: %d)\n" ,I2C_RDRW_IOCTL_MAX_MSGS ); goto err_out;
Signature · 2 lines
Jean Delvare SUSE L3 Support
Wolfram Sang
AppliedRe: [PATCH 2/3] i2ctransfer: Prevent msgs[] overflow with many parameters
Quoted 11 lines · Jean Delvare
There's an off-by-one bug in the message count check to ensure that we do not process more messages than the kernel allows. nmsgs points to the index within msgs[] which would be used for the _next_ message. If this index is equal the maximum number of messages then we must stop already.
This closes bug #220112:
https://bugzilla.kernel.org/show_bug.cgi?id=220112Fixes 9fc53a7fc669 ("i2c-tools: add new tool 'i2ctransfer'")Signed-off-by Jean Delvare <[email protected]>
Applied, thanks!
PATCH 3/3 i2ctransfer: Zero out memory passed to ioctl()
In reply to Jean DelvareValgrind complains that uninitialized memory may be passed to ioctl():
== Syscall param ioctl(I2C_RDWR) points to uninitialised byte(s)
== at 0x499382B: ioctl (in /lib64/libc.so.6)
== by 0x401957: main (i2ctransfer.c:343)
== Address 0x1ffefff94c is on thread 1's stack
== in frame #1, created by main (i2ctransfer.c:144)
==
== Syscall param ioctl(I2C_RDWR).msgs points to uninitialised byte(s)
== at 0x499382B: ioctl (in /lib64/libc.so.6)
== by 0x401957: main (i2ctransfer.c:343)
== Address 0x1ffefff956 is on thread 1's stack
== in frame #1, created by main (i2ctransfer.c:144)
Zero out the i2c_rdwr_ioctl_data struct as well as the msgs array to guarantee that no uninitialized memory will ever be passed to the kernel.
---
This one is not strictly needed, I can't see any actual bug. However making valgrind happy seems to be a sane goal, so that we can keep using it when debugging other issues without getting distracted. Wolfram, what do you think?
tools/i2ctransfer.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
tools/i2ctransfer.c +2 -2
@@ -149,8 +149,7 @@ int main(int argc, char *argv[])enum parse_state state =PARSE_GET_DESC ;unsigned int buf_idx =0 ; -for (i =0 ; i <I2C_RDRW_IOCTL_MAX_MSGS ; i++) - msgs[i].buf =NULL ; +memset (msgs,0 ,sizeof (msgs));/* handle (optional) flags first */ while ((opt =getopt (argc, argv,"abfhvVy" )) !=-1 ) {
@@ -334,6 +333,7 @@ int main(int argc, char *argv[])struct i2c_rdwr_ioctl_data rdwr;unsigned int print_flags =PRINT_READ_BUF ; +memset (&rdwr,0 ,sizeof (rdwr)); rdwr.msgs = msgs; rdwr.nmsgs = nmsgs; nmsgs_sent =ioctl (file,I2C_RDWR , &rdwr);
Signature · 2 lines
Jean Delvare SUSE L3 Support
Wolfram Sang
AppliedRe: [PATCH 3/3] i2ctransfer: Zero out memory passed to ioctl()
Quoted 15 lines · Jean Delvare
Valgrind complains that uninitialized memory may be passed to
ioctl():== Syscall param
ioctl(I2C_RDWR)points to uninitialised byte(s)
== at0x499382B: ioctl (in/lib64/libc.so.6)
== by0x401957: main (i2ctransfer.c:343)
== Address0x1ffefff94cis on thread 1's stack
== in frame #1, created by main (i2ctransfer.c:144)
==
== Syscall param ioctl(I2C_RDWR).msgs points to uninitialised byte(s)
== at0x499382B: ioctl (in/lib64/libc.so.6)
== by0x401957: main (i2ctransfer.c:343)
== Address0x1ffefff956is on thread 1's stack
== in frame #1, created by main (i2ctransfer.c:144)
Quoted 5 lines
Zero out the
i2c_rdwr_ioctl_datastruct as well as the msgs array to guarantee that no uninitialized memory will ever be passed to the kernel.Signed-off-by Jean Delvare <[email protected]>
Applied, thanks!
Quoted 4 lines
This one is not strictly needed, I can't see any actual bug. However making valgrind happy seems to be a sane goal, so that we can keep using it when debugging other issues without getting distracted. Wolfram, what do you think?
Sure, I agree. Also, I like the simpler code.
Wolfram Sang
AppliedQuoted 7 lines · Jean Delvare
If an error occurs while msgs[] is been prepared for the transfer, we jump to the clean-up path. How many buffers need to be freed depends on the state. If we were parsing data, we should free up to nmsgs. However, if we were parsing descriptors, we should free up to nmsgs - 1 only. The code was unconditionally freeing up to nmsgs, potentially freeing a non-allocated buffer.
Quoted 7 lines
In most cases, it was not a problem, we would simply call
free()on a NULL pointer and that's a no-op. However, if msgs[] was full then we would access memory beyond its end and callfree()on a random pointer.Fixes 9fc53a7fc669 ("i2c-tools: add new tool 'i2ctransfer'")Signed-off-by Jean Delvare <[email protected]>
Applied, thanks!