core: Relax gpiod_chip_open() for symbolic links
Subsystems (2)
PATCH 1/1 core: Relax gpiod_chip_open() for symbolic links
User may ask device helper tool, for example, udev, to create a specific symbolic link to a device node. GPIO chip character device node is not exceptional. However, libgpiod in the commit d9b1c1f14c6b ("core: harden gpiod_chip_open()") went way too far in the hardening device node check.
Relax that hardening for symbolic link to fix the regression.
Reproducer:
% gpioinfo /dev/gpiochip5
gpiochip5 - 16 lines:
line 0: "MUX33_DIR" "uart1-rx-oe" output active-high [used]
...
% ln -sf /dev/gpiochip5 /dev/MyGPIO_5
% gpioinfo /dev/MyGPIO_5
gpioinfo: looking up chip /dev/MyGPIO_5: Inappropriate ioctl for device
--- lib/core.c | 32 ++++++++++++++++++++++---------- 1 file changed, 22 insertions(+), 10 deletions(-)
lib/core.c +22 -10
@@ -8,6 +8,7 @@/* Low-level, core library code. */ #include <errno.h> +#include <limits.h> #include <fcntl.h> #include <gpiod.h> #include <linux/gpio.h>
@@ -75,7 +76,7 @@ struct gpiod_chip {static bool is_gpiochip_cdev (const char *path) { -char *name, *pathcpy, *sysfsp, sysfsdev[16 ], devstr[16 ]; +char *name, *realname, *sysfsp, sysfsdev[16 ], devstr[16 ];struct stat statbuf;bool ret = false;int rv, fd;
@@ -85,6 +86,21 @@ static bool is_gpiochip_cdev(const char *path)if (rv) goto out; +/* +* Is it a symbolic link? +* We have to resolve symbolic link before checking the rest. +*/ +if (S_ISLNK (statbuf.st_mode)) + realname =realpath (path,NULL ); +else + realname =strdup (path); +if (realname ==NULL ) + goto out; + + rv =stat (realname, &statbuf); +if (rv) + goto out_free_realname; +/* Is it a character device? */ if (!S_ISCHR (statbuf.st_mode)) {/*
@@ -94,20 +110,16 @@ static bool is_gpiochip_cdev(const char *path)* libgpiod from before the introduction of this routine. */ errno =ENOTTY ; - goto out; + goto out_free_realname; }/* Get the basename. */ - pathcpy =strdup (path); -if (!pathcpy) - goto out; - - name =basename (pathcpy); + name =basename (realname);/* Do we have a corresponding sysfs attribute? */ rv =asprintf (&sysfsp,"/sys/bus/gpio/devices/%s/dev" , name);if (rv <0 ) - goto out_free_pathcpy; + goto out_free_realname;if (access (sysfsp,R_OK ) !=0 ) {/*
@@ -149,8 +161,8 @@ static bool is_gpiochip_cdev(const char *path) out_free_sysfsp:free (sysfsp); -out_free_pathcpy: -free (pathcpy); +out_free_realname: +free (realname); out:return ret; }
Signature · 1 line
2.25.0
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):
Quoted 18 lines
User may ask device helper tool, for example, udev, to create a specific symbolic link to a device node. GPIO chip character device node is not exceptional. However, libgpiod in the commit
d9b1c1f14c6b("core: hardengpiod_chip_open()") went way too far in the hardening device node check.Relax that hardening for symbolic link to fix the regression.
Reproducer:
% gpioinfo
/dev/gpiochip5
gpiochip5 - 16 lines:
line 0: "MUX33_DIR" "uart1-rx-oe" output active-high [used]
...% ln -sf
/dev/gpiochip5/dev/MyGPIO_5
Quoted 7 lines
% gpioinfo
/dev/MyGPIO_5
gpioinfo: looking up chip/dev/MyGPIO_5: Inappropriate ioctl for deviceFixes d9b1c1f14c6b ("core: harden gpiod_chip_open()")Cc Bartosz Golaszewski <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>
Hi Andy,
thanks for this - it makes perfect sense. One nit though: could you keep the includes ordered alphabetically? Also: it would be great if you could add a test case for this to tests/tests-chip.c.
Bartosz
Quoted 28 lines · Bartosz Golaszewski
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):Quoted 18 lines
User may ask device helper tool, for example, udev, to create a specific symbolic link to a device node. GPIO chip character device node is not exceptional. However, libgpiod in the commit
d9b1c1f14c6b("core: hardengpiod_chip_open()") went way too far in the hardening device node check.Relax that hardening for symbolic link to fix the regression.
Reproducer:
% gpioinfo
/dev/gpiochip5
gpiochip5 - 16 lines:
line 0: "MUX33_DIR" "uart1-rx-oe" output active-high [used]
...% ln -sf
/dev/gpiochip5/dev/MyGPIO_5Quoted 7 lines
% gpioinfo
/dev/MyGPIO_5
gpioinfo: looking up chip/dev/MyGPIO_5: Inappropriate ioctl for deviceFixes d9b1c1f14c6b ("core: harden gpiod_chip_open()")Cc Bartosz Golaszewski <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>
Quoted 4 lines
Hi Andy,
thanks for this - it makes perfect sense. One nit though: could you keep the includes ordered alphabetically?
Probably not. The user space relies a lot on header ordering. And limits.h sounds like one needed to be included first in many cases. That's why I moved it to the top. I can do it if you insist, but I consider it wrong approach for the record.
Quoted 2 lines
Also: it would be great if
you could add a test case for this totests/tests-chip.c.
I will look at it if I can do quickly something.
For the record I have tested it on Intel Edison with real GPIO chips (so, that's how reproducer code appears in the commit message).
Thanks for review!
Signature · 2 lines
With Best Regards, Andy Shevchenko
pt., 7 lut 2020 o 11:30 Andy Shevchenko
<[email protected]> napisał(a):
Quoted 32 lines · Bartosz Golaszewski
Quoted 30 lines
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):Quoted 18 lines
User may ask device helper tool, for example, udev, to create a specific symbolic link to a device node. GPIO chip character device node is not exceptional. However, libgpiod in the commit
d9b1c1f14c6b("core: hardengpiod_chip_open()") went way too far in the hardening device node check.Relax that hardening for symbolic link to fix the regression.
Reproducer:
% gpioinfo
/dev/gpiochip5
gpiochip5 - 16 lines:
line 0: "MUX33_DIR" "uart1-rx-oe" output active-high [used]
...% ln -sf
/dev/gpiochip5/dev/MyGPIO_5Quoted 7 lines
% gpioinfo
/dev/MyGPIO_5
gpioinfo: looking up chip/dev/MyGPIO_5: Inappropriate ioctl for deviceFixes d9b1c1f14c6b ("core: harden gpiod_chip_open()")Cc Bartosz Golaszewski <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>Hi Andy,
Quoted 7 lines
Quoted 2 lines
thanks for this - it makes perfect sense. One nit though: could you keep the includes ordered alphabetically?
Probably not. The user space relies a lot on header ordering. And
limits.hsounds like one needed to be included first in many cases. That's why I moved it to the top. I can do it if you insist, but I consider it wrong approach for the record.
Nah, if anything headers may rely on some preprocessor defines coming before them, but the ordering should be of no importance.
Quoted 6 lines
Quoted 3 lines
Also: it would be great if
you could add a test case for this totests/tests-chip.c.I will look at it if I can do quickly something.
If not, don't worry - I can add it later myself.
Bartosz
Quoted 10 lines
For the record I have tested it on Intel Edison with real GPIO chips (so, that's how reproducer code appears in the commit message).
Thanks for review!
--
With Best Regards,
Andy Shevchenko
Quoted 5 lines · Bartosz Golaszewski
pt., 7 lut 2020 o 11:30 Andy Shevchenko
<[email protected]> napisał(a):Quoted 3 lines · Bartosz Golaszewski
Quoted 2 lines
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):
...
Quoted 10 lines
Quoted 7 lines
Quoted 2 lines
thanks for this - it makes perfect sense. One nit though: could you keep the includes ordered alphabetically?
Probably not. The user space relies a lot on header ordering. And
limits.hsounds like one needed to be included first in many cases. That's why I moved it to the top. I can do it if you insist, but I consider it wrong approach for the record.Nah, if anything headers may rely on some preprocessor defines coming before them, but the ordering should be of no importance.
Okay, in any case, if you think it's better to be sorted, can you change it when applying? (I don't think we need another version simple for that)
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Also: it would be great if
you could add a test case for this totests/tests-chip.c.I will look at it if I can do quickly something.
If not, don't worry - I can add it later myself.
I briefly looked at this, but it seems not feasible to me in reasonable time, sorry.
The problems I encountered are, but not limited to:
- creating a symlink in a test case folder
- understanding how to handle interrupt of the test case (we have to remove link ourselves or framework does it for us?)
- where to put the symbolic link creation: I think it might be a (boolean) parameter to gpio-mockup testing API when we "make a chip" (when device node should appear) to enable symlink with a predefined name (like $testpath/gpiochipX-link)
- last time I did something (simple!) with GLib was several years ago
Signature · 2 lines
With Best Regards, Andy Shevchenko
pt., 7 lut 2020 o 12:28 Andy Shevchenko
<[email protected]> napisał(a):
Quoted 18 lines · Bartosz Golaszewski
Quoted 5 lines
pt., 7 lut 2020 o 11:30 Andy Shevchenko
<[email protected]> napisał(a):Quoted 3 lines · Bartosz Golaszewski
Quoted 2 lines
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):...
Quoted 8 lines
Quoted 7 lines
Quoted 2 lines
thanks for this - it makes perfect sense. One nit though: could you keep the includes ordered alphabetically?
Probably not. The user space relies a lot on header ordering. And
limits.hsounds like one needed to be included first in many cases. That's why I moved it to the top. I can do it if you insist, but I consider it wrong approach for the record.
Quoted 6 lines
Quoted 2 lines
Nah, if anything headers may rely on some preprocessor defines coming before them, but the ordering should be of no importance.
Okay, in any case, if you think it's better to be sorted, can you change it when applying? (I don't think we need another version simple for that)
Sure.
Quoted 10 lines
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Also: it would be great if
you could add a test case for this totests/tests-chip.c.I will look at it if I can do quickly something.
If not, don't worry - I can add it later myself.
I briefly looked at this, but it seems not feasible to me in reasonable time, sorry.
Quoted 10 lines
The problems I encountered are, but not limited to:
- creating a symlink in a test case folder
- understanding how to handle interrupt of the test case (we have to remove link ourselves or framework does it for us?)
- where to put the symbolic link creation: I think it might be a (boolean) parameter to gpio-mockup testing API when we "make a chip" (when device node should appear) to enable symlink with a predefined name (like $testpath/gpiochipX-link)
- last time I did something (simple!) with GLib was several years ago
Sure, I can do it myself - it will probably take me much less time.
Bart
czw., 6 lut 2020 o 19:14 Andy Shevchenko
<[email protected]> napisał(a):
Quoted 18 lines
User may ask device helper tool, for example, udev, to create a specific symbolic link to a device node. GPIO chip character device node is not exceptional. However, libgpiod in the commit
d9b1c1f14c6b("core: hardengpiod_chip_open()") went way too far in the hardening device node check.Relax that hardening for symbolic link to fix the regression.
Reproducer:
% gpioinfo
/dev/gpiochip5
gpiochip5 - 16 lines:
line 0: "MUX33_DIR" "uart1-rx-oe" output active-high [used]
...% ln -sf
/dev/gpiochip5/dev/MyGPIO_5
Quoted 7 lines
% gpioinfo
/dev/MyGPIO_5
gpioinfo: looking up chip/dev/MyGPIO_5: Inappropriate ioctl for deviceFixes d9b1c1f14c6b ("core: harden gpiod_chip_open()")Cc Bartosz Golaszewski <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>
Patch applied and backported to v1.4.x and v1.5.x stable branches.
Thanks!
Bartosz