ati_remote2.c: possible mutex_lock without mutex_unlock
Subsystems (2)
KERNEL_VERSION: 2.6.31 DESCRIBE: In driver ./drivers/input/input.c possible call to mutex_lock
from function input_devices_seq_start without mutex_unlock.
After calling input_devices_seq_start we can't know whether
mutex was locked or not.
Case 1. If mutex_lock_interruptible was not
locked due to interrupt then input_devices_seq_start returns NULL. Case 2. If mutex was successfuly locked but seq_list_start returned NULL then input_devices_seq_start returns NULL too.
The last case occurs if seq_list_start is called with pos>size of input_dev_list or pos<0.
Hence, after calling input_devices_seq_start we can not simply check that result is not NULL and call input_devices_seq_stop function which unlocks the mutex. Because in case 2 the mutex will stay locked. void * ret = input_devices_seq_start(...);
if(ret!=NULL) {
//mutex is acquired for sure
input_devices_seq_stop(...);//unlocks the mutex
} else {
//mutex may be acquired or not
} 783 static void *input_devices_seq_start(struct seq_file *seq, loff_t
*pos)
784{
785 if (mutex_lock_interruptible(&input_mutex))
786 return NULL;
787
788 return seq_list_start(&input_dev_list, *pos);
789}
663struct list_head *seq_list_start(struct list_head *head, loff_t
pos)
664{
665 struct list_head *lh;
666
667 list_for_each(lh, head)
668 if (pos-- == 0)
669 return lh;
670
671 return NULL;
672}
673
674EXPORT_SYMBOL(seq_list_start);
675
Found by: Linux Driver Verification project
Quoted 15 lines · iceberg
In driver
./drivers/input/input.cpossible call tomutex_lockfrom
functioninput_devices_seq_startwithoutmutex_unlock.After calling
input_devices_seq_startwe can't know whether
mutex was locked or not.Case 1. If
mutex_lock_interruptiblewas not locked due to interrupt theninput_devices_seq_startreturns NULL.
Case 2. If mutex was successfuly locked butseq_list_startreturned NULL theninput_devices_seq_startreturns NULL too. The last case occurs ifseq_list_startis called with pos>size ofinput_dev_listor pos<0. Hence, after callinginput_devices_seq_startwe can not simply check that result is not NULL and callinput_devices_seq_stopfunction which unlocks the mutex. Because in case 2 the mutex will stay locked. void * ret =input_devices_seq_start(...);
Quoted 6 lines
if(ret!=NULL) {
//mutex is acquired for sure
input_devices_seq_stop(...);//unlocks the mutex
} else {
//mutex may be acquired or not
}Plus, we should return EAGAIN rather than failing silently when input_handlers_seq_start() has been interrupted by signal, right?
Dmitry, how about the fix below?
From: Jiri Kosina <[email protected]>
Subject: [PATCH] Input: make input_handlers_seq_start() signal safe
input_devices_seq_start() uses mutex_lock_interruptible() to acquire the input_mutex, but doesn't properly handle the situation if mutex_lock_interruptible() really gets interrupted. In such scenario, input_handlers_seq_start() returns NULL, which ambiguous, as seq_list_start() could return NULL as well. This could lead to the situation in which input_handlers_seq_stop() will try to unlock mutex that hasn't been locked.
Plus, in such situations, the code fails silently, rather than returning EAGAIN.
---
drivers/input/input.c | 6 +++++- 1 files changed, 5 insertions(+), 1 deletions(-)
drivers/input/input.c +5 -1
@@ -882,7 +882,7 @@ static const struct file_operations input_devices_fileops = {static void *input_handlers_seq_start (struct seq_file *seq,loff_t *pos) {if (mutex_lock_interruptible (&input_mutex)) -return NULL ; +return ERR_PTR (-EAGAIN ); seq->private = (void *)(unsigned long )*pos;return seq_list_start (&input_handler_list, *pos);
@@ -896,6 +896,10 @@ static void *input_handlers_seq_next(struct seq_file *seq, void *v, loff_t *pos)static void input_handlers_seq_stop (struct seq_file *seq,void *v) { +/* seq_start could get interrupted by signal before acquiring mutex */ +if (IS_ERR (v) &&ERR_PTR (v) == -EAGAIN ) +return ; +mutex_unlock (&input_mutex); }
Signature · 2 lines
Jiri Kosina SUSE Labs, Novell Inc.
Dmitry Torokhov
maintainerHi Jiri,
Quoted 24 lines · Jiri Kosina
Quoted 15 lines · iceberg
In driver
./drivers/input/input.cpossible call tomutex_lockfrom
functioninput_devices_seq_startwithoutmutex_unlock.After calling
input_devices_seq_startwe can't know whether
mutex was locked or not.Case 1. If
mutex_lock_interruptiblewas not locked due to interrupt theninput_devices_seq_startreturns NULL.
Case 2. If mutex was successfuly locked butseq_list_startreturned NULL theninput_devices_seq_startreturns NULL too. The last case occurs ifseq_list_startis called with pos>size ofinput_dev_listor pos<0. Hence, after callinginput_devices_seq_startwe can not simply check that result is not NULL and callinput_devices_seq_stopfunction which unlocks the mutex. Because in case 2 the mutex will stay locked. void * ret =input_devices_seq_start(...);Quoted 6 lines
if(ret!=NULL) { //mutex is acquired for sure input_devices_seq_stop(...);//unlocks the mutex } else { //mutex may be acquired or not }
Quoted 5 lines
Plus, we should return EAGAIN rather than failing silently when
input_handlers_seq_start()has been interrupted by signal, right?Dmitry, how about the fix below?
Umm, I don't like assuming that EAGAIN can only mean that mutex_lock_interruptible() failed, seq_file core may theoretically return -EAGAIN too. In fact, looking through seq_file.c traverse() does return -EAGAIN in certain cases...
Plus input_devices_seq_start() needs the same treatment (you fixed input_handlers_seq_start() only).
HOw about this one instead?
Signature · 25 lines
Dmitry Input: fix locking issue in /proc/bus/input/ handlers From: Dmitry Torokhov <[email protected]> input_devices_seq_start() uses mutex_lock_interruptible() to acquire the input_mutex, but doesn't properly handle the situation when the call fails (for example due to interrupt). Instead of returning NULL (which indicates that there is no more data) we should return ERR_PTR()-encoded error. We also need explicit flag indicating whether input_mutex was acquired since input_devices_seq_stop() is called whether input_devices_seq_start() was successful or not. The same applies to input_handlers_seq_start(). Reported-by: iceberg <[email protected]> Signed-off-by: Dmitry Torokhov <[email protected]> --- drivers/input/input.c | 65 ++++++++++++++++++++++++++++++++++++------------- 1 files changed, 48 insertions(+), 17 deletions(-)
drivers/input/input.c +48 -17
@@ -782,10 +782,29 @@ static unsigned int input_proc_devices_poll(struct file *file, poll_table *wait)return 0 ; } +union input_seq_state { +struct { +unsigned short pos; +bool mutex_acquired; + }; +void *p; +}; +static void *input_devices_seq_start (struct seq_file *seq,loff_t *pos) { -if (mutex_lock_interruptible (&input_mutex)) -return NULL ; +union input_seq_state *state = (union input_seq_state *)&seq->private; +int error; + +/* We need to fit into seq->private pointer */ +BUILD_BUG_ON (sizeof (union input_seq_state ) !=sizeof (seq->private)); + + error =mutex_lock_interruptible (&input_mutex); +if (error) { + state->mutex_acquired = false; +return ERR_PTR (error); + } + + state->mutex_acquired = true;return seq_list_start (&input_dev_list, *pos); }
@@ -795,9 +814,12 @@ static void *input_devices_seq_next(struct seq_file *seq, void *v, loff_t *pos)return seq_list_next (v, &input_dev_list, pos); } -static void input_devices_seq_stop (struct seq_file *seq,void *v) +static void input_seq_stop (struct seq_file *seq,void *v) { -mutex_unlock (&input_mutex); +union input_seq_state *state = (union input_seq_state *)&seq->private; + +if (state->mutex_acquired) +mutex_unlock (&input_mutex); } static void input_seq_print_bitmap(struct seq_file *seq, const char *name,
@@ -861,7 +883,7 @@ static int input_devices_seq_show(struct seq_file *seq, void *v)static const struct seq_operations input_devices_seq_ops = { .start = input_devices_seq_start, .next = input_devices_seq_next, - .stop = input_devices_seq_stop, + .stop = input_seq_stop, .show = input_devices_seq_show, };
@@ -881,40 +903,49 @@ static const struct file_operations input_devices_fileops = {static void *input_handlers_seq_start (struct seq_file *seq,loff_t *pos) { -if (mutex_lock_interruptible (&input_mutex)) -return NULL ; +union input_seq_state *state = (union input_seq_state *)&seq->private; +int error; + +/* We need to fit into seq->private pointer */ +BUILD_BUG_ON (sizeof (union input_seq_state ) !=sizeof (seq->private)); + + error =mutex_lock_interruptible (&input_mutex); +if (error) { + state->mutex_acquired = false; +return ERR_PTR (error); + } + + state->mutex_acquired = true; + state->pos = *pos; - seq->private = (void *)(unsigned long )*pos;return seq_list_start (&input_handler_list, *pos); }static void *input_handlers_seq_next (struct seq_file *seq,void *v,loff_t *pos) { - seq->private = (void *)(unsigned long )(*pos +1 ); -return seq_list_next (v, &input_handler_list, pos); -} +union input_seq_state *state = (union input_seq_state *)&seq->private; -static void input_handlers_seq_stop (struct seq_file *seq,void *v) -{ -mutex_unlock (&input_mutex); + state->pos = *pos +1 ; +return seq_list_next (v, &input_handler_list, pos); }static int input_handlers_seq_show (struct seq_file *seq,void *v) { struct input_handler *handler = container_of(v, struct input_handler, node); +union input_seq_state *state = (union input_seq_state *)&seq->private; -seq_printf (seq,"N: Number=%ld Name=%s" , - (unsigned long )seq->private, handler->name); +seq_printf (seq,"N: Number=%u Name=%s" , state->pos, handler->name);if (handler->fops)seq_printf (seq," Minor=%d" , handler->minor);seq_putc (seq,'\n' );return 0 ; } +static const struct seq_operations input_handlers_seq_ops = { .start = input_handlers_seq_start, .next = input_handlers_seq_next, - .stop = input_handlers_seq_stop, + .stop = input_seq_stop, .show = input_handlers_seq_show, };
Quoted 4 lines · Dmitry Torokhov
Umm, I don't like assuming that EAGAIN can only mean that
mutex_lock_interruptible()failed,seq_filecore may theoretically return-EAGAINtoo. In fact, looking throughseq_file.ctraverse()does return-EAGAINin certain cases...
Damn, you are right -- I explicitly checked for this, but have completely overlooked the "Eoveflow:" branch in traverse(), which returns EAGAIN. So my previous patch is of course incorrect.
Quoted 10 lines
Input: fix locking issue in
/proc/bus/input/handlersFrom: Dmitry Torokhov <[email protected]>
input_devices_seq_start()usesmutex_lock_interruptible()to acquire theinput_mutex, but doesn't properly handle the situation when the call fails (for example due to interrupt). Instead of returning NULL (which indicates that there is no more data) we should return ERR_PTR()-encoded error.
Quoted 8 lines
We also need explicit flag indicating whether
input_mutexwas acquired sinceinput_devices_seq_stop()is called whetherinput_devices_seq_start()was successful or not.The same applies to
input_handlers_seq_start().Reported-by iceberg <[email protected]>Signed-off-by Dmitry Torokhov <[email protected]>
Yup, looks OK to me.
Signature · 2 lines
Jiri Kosina SUSE Labs, Novell Inc.
Dmitry Torokhov
maintainerQuoted 25 lines · Jiri Kosina
Quoted 4 lines · Dmitry Torokhov
Umm, I don't like assuming that EAGAIN can only mean that
mutex_lock_interruptible()failed,seq_filecore may theoretically return-EAGAINtoo. In fact, looking throughseq_file.ctraverse()does return-EAGAINin certain cases...Damn, you are right -- I explicitly checked for this, but have completely overlooked the "Eoveflow:" branch in
traverse(), which returns EAGAIN. So my previous patch is of course incorrect.Quoted 14 lines
Input: fix locking issue in
/proc/bus/input/handlersFrom: Dmitry Torokhov <[email protected]>
input_devices_seq_start()usesmutex_lock_interruptible()to acquire theinput_mutex, but doesn't properly handle the situation when the call fails (for example due to interrupt). Instead of returning NULL (which indicates that there is no more data) we should return ERR_PTR()-encoded error.We also need explicit flag indicating whether
input_mutexwas acquired sinceinput_devices_seq_stop()is called whetherinput_devices_seq_start()was successful or not.
Quoted 7 lines
Quoted 4 lines
The same applies to
input_handlers_seq_start().Reported-by iceberg <[email protected]>Signed-off-by Dmitry Torokhov <[email protected]>Yup, looks OK to me.
Putting you as "Reviewed-by.." then, OK?
Signature · 1 line
Dmitry
Quoted 12 lines · Dmitry Torokhov
Quoted 12 lines
Quoted 12 lines
From: Dmitry Torokhov <[email protected]>
input_devices_seq_start()usesmutex_lock_interruptible()to acquire theinput_mutex, but doesn't properly handle the situation when the call fails (for example due to interrupt). Instead of returning NULL (which indicates that there is no more data) we should return ERR_PTR()-encoded error.We also need explicit flag indicating whether
input_mutexwas acquired sinceinput_devices_seq_stop()is called whetherinput_devices_seq_start()was successful or not.
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
The same applies to
input_handlers_seq_start().Reported-by iceberg <[email protected]>Signed-off-by Dmitry Torokhov <[email protected]>Yup, looks OK to me.
Putting you as "Reviewed-by.." then, OK?
Sure, feel free to do that.
Thanks,
Signature · 2 lines
Jiri Kosina SUSE Labs, Novell Inc.