media: v4l2-ctrls: split up into four source files
[ This code is older, but it showed up as a new warning because of
moving the files around - dan ]
Hello Hans Verkuil,
The patch 71c689dc2e73: "media: v4l2-ctrls: split up into four source files" from Apr 27, 2021, leads to the following Smatch static checker warning:
drivers/media/v4l2-core/v4l2-ctrls-api.c:374 v4l2_g_ext_ctrls_common() warn: uncapped user size for kvmalloc() will WARN drivers/media/v4l2-core/v4l2-ctrls-api.c:545 try_set_ext_ctrls_common() warn: uncapped user size for kvmalloc() will WARN
drivers/media/v4l2-core/v4l2-ctrls-api.c
351 int v4l2_g_ext_ctrls_common(struct v4l2_ctrl_handler *hdl,
352 struct v4l2_ext_controls *cs,
353 struct video_device *vdev)
354 {
355 struct v4l2_ctrl_helper helper[4];
356 struct v4l2_ctrl_helper *helpers = helper;
357 int ret;
358 int i, j;
359 bool is_default, is_request;
360
361 is_default = (cs->which == V4L2_CTRL_WHICH_DEF_VAL);
362 is_request = (cs->which == V4L2_CTRL_WHICH_REQUEST_VAL);
363
364 cs->error_idx = cs->count;
365 cs->which = V4L2_CTRL_ID2WHICH(cs->which);
366
367 if (!hdl)
368 return -EINVAL;
369
370 if (cs->count == 0)
371 return class_check(hdl, cs->which);
372
373 if (cs->count > ARRAY_SIZE(helper)) {--> 374 helpers = kvmalloc_array(cs->count, sizeof(helper[0]),
These days if "cs->count" is larger than INT_MAX it will trigger a WARN() because basically "people shouldn't be so trusting of user space". kvmalloc() used to be able to allocate more than INT_MAX but that led to integer overflow problmes and security bugs.
This "cs->count" value comes from the user via the ioctl. I don't know what a sensible upper bound is though.
375 GFP_KERNEL);
376 if (!helpers)
377 return -ENOMEM;
378 }
379 regards,
dan carpenter
Hi Dan,
Quoted 38 lines · Dan Carpenter
[ This code is older, but it showed up as a new warning because of
moving the files around - dan ]Hello Hans Verkuil,
The patch
71c689dc2e73: "media: v4l2-ctrls: split up into four source files" from Apr 27, 2021, leads to the following Smatch static checker warning:
drivers/media/v4l2-core/v4l2-ctrls-api.c:374v4l2_g_ext_ctrls_common()warn: uncapped user size forkvmalloc()will WARNdrivers/media/v4l2-core/v4l2-ctrls-api.c:545try_set_ext_ctrls_common()warn: uncapped user size forkvmalloc()will WARN
drivers/media/v4l2-core/v4l2-ctrls-api.c351 int v4l2_g_ext_ctrls_common(struct v4l2_ctrl_handler *hdl, 352 struct v4l2_ext_controls *cs, 353 struct video_device *vdev) 354 { 355 struct v4l2_ctrl_helper helper[4]; 356 struct v4l2_ctrl_helper *helpers = helper; 357 int ret; 358 int i, j; 359 bool is_default, is_request; 360 361 is_default = (cs->which == V4L2_CTRL_WHICH_DEF_VAL); 362 is_request = (cs->which == V4L2_CTRL_WHICH_REQUEST_VAL); 363 364 cs->error_idx = cs->count; 365 cs->which = V4L2_CTRL_ID2WHICH(cs->which); 366 367 if (!hdl) 368 return -EINVAL; 369 370 if (cs->count == 0) 371 return class_check(hdl, cs->which); 372 373 if (cs->count > ARRAY_SIZE(helper)) {--> 374 helpers = kvmalloc_array(cs->count,
sizeof(helper[0]),
Quoted 7 lines
These days if "
cs->count" is larger thanINT_MAXit will trigger aWARN()because basically "people shouldn't be so trusting of user space".kvmalloc()used to be able to allocate more thanINT_MAXbut that led to integer overflow problmes and security bugs.This "
cs->count" value comes from the user via the ioctl. I don't know what a sensible upper bound is though.
cs->count is capped to 1024 in drivers/media/v4l2-core/v4l2-ioctl.c. Search for V4L2_CID_MAX_CTRLS.
There is no way for smatch to know this of course.
Regards,
Hans
Quoted 9 lines
375 GFP_KERNEL); 376 if (!helpers) 377 return -ENOMEM; 378 } 379regards,
dan carpenter
Quoted 46 lines · Hans Verkuil
Hi Dan,
Quoted 43 lines · Dan Carpenter
[ This code is older, but it showed up as a new warning because of
moving the files around - dan ]Hello Hans Verkuil,
The patch
71c689dc2e73: "media: v4l2-ctrls: split up into four source files" from Apr 27, 2021, leads to the following Smatch static checker warning:
drivers/media/v4l2-core/v4l2-ctrls-api.c:374v4l2_g_ext_ctrls_common()warn: uncapped user size forkvmalloc()will WARNdrivers/media/v4l2-core/v4l2-ctrls-api.c:545try_set_ext_ctrls_common()warn: uncapped user size forkvmalloc()will WARN
drivers/media/v4l2-core/v4l2-ctrls-api.c351 int v4l2_g_ext_ctrls_common(struct v4l2_ctrl_handler *hdl, 352 struct v4l2_ext_controls *cs, 353 struct video_device *vdev) 354 { 355 struct v4l2_ctrl_helper helper[4]; 356 struct v4l2_ctrl_helper *helpers = helper; 357 int ret; 358 int i, j; 359 bool is_default, is_request; 360 361 is_default = (cs->which == V4L2_CTRL_WHICH_DEF_VAL); 362 is_request = (cs->which == V4L2_CTRL_WHICH_REQUEST_VAL); 363 364 cs->error_idx = cs->count; 365 cs->which = V4L2_CTRL_ID2WHICH(cs->which); 366 367 if (!hdl) 368 return -EINVAL; 369 370 if (cs->count == 0) 371 return class_check(hdl, cs->which); 372 373 if (cs->count > ARRAY_SIZE(helper)) {--> 374 helpers = kvmalloc_array(cs->count,
sizeof(helper[0]),These days if "
cs->count" is larger thanINT_MAXit will trigger aWARN()because basically "people shouldn't be so trusting of user space".kvmalloc()used to be able to allocate more thanINT_MAXbut that led to integer overflow problmes and security bugs.
Quoted 7 lines
Quoted 2 lines
This "
cs->count" value comes from the user via the ioctl. I don't know what a sensible upper bound is though.
cs->countis capped to 1024 indrivers/media/v4l2-core/v4l2-ioctl.c. Search forV4L2_CID_MAX_CTRLS.There is no way for smatch to know this of course.
Yeah. It's tricky. The switch statement in check_array_args() is too much data for Smatch and the void *arg confuses Smatch as well. :/
Thanks for looking at this.
regards,
dan carpenter