drivers/input: Fix null pointer dereferences in input_ff_create() and input_ff_create_memless()
Subsystems (2)
PATCH RESEND 1/1 drivers/input: Fix null pointer dereferences in input_ff_create() and input_ff_create_memless()
Resending this patch as I have not received feedback since my initial submission on December 25, 2024. Please let me know if additional changes or information are required.
This patch addresses two null pointer dereference bugs detected by KASAN in input_ff_create() (drivers/input/ff-core.c) and input_ff_create_memless() (drivers/input/ff-memless.c).
- input_ff_create() log: https://syzkaller.appspot.com/bug?extid=dd5f8d6456680e55eb0a
- input_ff_create_memless() log: https://syzkaller.appspot.com/x/report.txt?x=10a51adf980000
The root cause of the null pointer dereference in input_ff_create() is that it utilises the "no_free_ptr" macro in the line:
"dev->ff = no_free_ptr(ff);"
which effectively sets "ff" to NULL and makes "dev->ff" point to what "ff" was initially pointing to. The code then attempts to incorrectly utilise "ff" after it has been set to NULL causing a null pointer dereference to occur in the following:
"
for_each_set_bit(i, dev->ffbit, FF_CNT)
__set_bit(i, ff->ffbit);
/* we can emulate RUMBLE with periodic effects */ if (test_bit(FF_PERIODIC, ff->ffbit))
__set_bit(FF_RUMBLE, dev->ffbit);
"
To fix this I changed all accesses to "ff" to be "dev->ff".
The root cause of the null pointer dereference in input_ff_create_memless() is that it also utilises the "no_free_ptr" macro in the line: "ff->private = no_free_ptr(ml);"
which sets "ml" to NULL and makes "ff->private" point to what "ml" was initially pointing to. The code then attempts to utilise "ml" after it has been set to NULL:
"
for (i = 0; i < FF_MEMLESS_EFFECTS; i++)
ml->states[i].effect = &ff->effects[i];
"
To fix this bug I moved the for loop before "ff->private = no_free_ptr(ml);".
--- drivers/input/ff-core.c | 4 ++-- drivers/input/ff-memless.c | 7 ++++--- 2 files changed, 6 insertions(+), 5 deletions(-)
drivers/input/ff-core.c +2 -2
@@ -322,10 +322,10 @@ int input_ff_create(struct input_dev *dev, unsigned int max_effects)/* Copy "true" bits into ff device bitmap */ for_each_set_bit (i, dev->ffbit,FF_CNT ) -__set_bit (i, ff->ffbit); +__set_bit (i, dev->ff->ffbit);/* we can emulate RUMBLE with periodic effects */ -if (test_bit (FF_PERIODIC , ff->ffbit)) +if (test_bit (FF_PERIODIC , dev->ff->ffbit))__set_bit (FF_RUMBLE , dev->ffbit);return 0 ;
drivers/input/ff-memless.c +4 -3
@@ -524,6 +524,10 @@ int input_ff_create_memless(struct input_dev *dev, void *data,return error; ff = dev->ff; + +for (i =0 ; i <FF_MEMLESS_EFFECTS ; i++) + ml->states[i].effect = &ff->effects[i]; + ff->private =no_free_ptr (ml); ff->upload = ml_ff_upload; ff->playback = ml_ff_playback;
@@ -538,9 +542,6 @@ int input_ff_create_memless(struct input_dev *dev, void *data,set_bit (FF_SQUARE , dev->ffbit); } -for (i =0 ; i <FF_MEMLESS_EFFECTS ; i++) - ml->states[i].effect = &ff->effects[i]; -return 0 ; }EXPORT_SYMBOL_GPL (input_ff_create_memless);
--
2.39.5
Quoted 2 lines · qasdev
Resending this patch as I have not received feedback since my initial submission on December 25, 2024. Please let me know if additional changes or information are required.
The bug isn't present in linux-next.
The original syzbot bug was already fixed in November last year. It was fixed in a rebase so it doesn't exist in git history and I really had to dig to understand what the issue was.
regards,
dan carpenter