| Message ID | 20160803081140.GA7833@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Aug 03, 2016 at 10:11:40AM +0200, Ingo Molnar wrote: > An added advantage would be that during review it would stick out like a sore > thumb if anyone used a 'weird' permission variant. > > For example, if you saw these lines in a driver patch: > > + __ATTR(l1, 0444, driver_show_l4, NULL); > + __ATTR(l3, 0446, driver_show_l4, NULL); > + __ATTR(l2, 04444, driver_show_l4, NULL); > + __ATTR(l4, 0444, driver_show_l4, NULL); > > ... would you notice it at a glance that it contains two security holes? I've tried to deal with that in the past with the __ATTR_RW() and __ATTR_RO() and __ATTR_WO() macros that more should be using. I swept the tree a few years ago to try to fix up most of them, but I know I didn't catch them all, and more files have been added since then. > While the weird permissions in this: > > + __ATTR(l1, PERM_r__r__r__, driver_show_l4, NULL); > + __ATTR(l3, PERM_r__r__rw_, driver_show_l4, NULL); > + __ATTR(l2, PERM_sr__r__r__, driver_show_l4, NULL); > + __ATTR(l4, PERM_r__r__r__, driver_show_l4, NULL); > > Wouln't even build, because the dangerous patterns of PERM_r__r__rw_ or > PERM_sr__r__r__ are not defined to begin with. Because of that, odds are people will just stick to the octal numbers, because they think they want something other than the ones you defined for foolish reasons :) That being said, I do like them much better than the macros we have today, which I always have to go and look up every time I see them... thanks, greg k-h
* Greg Kroah-Hartman <gregkh@linuxfoundation.org> wrote: > On Wed, Aug 03, 2016 at 10:11:40AM +0200, Ingo Molnar wrote: > > An added advantage would be that during review it would stick out like a sore > > thumb if anyone used a 'weird' permission variant. > > > > For example, if you saw these lines in a driver patch: > > > > + __ATTR(l1, 0444, driver_show_l4, NULL); > > + __ATTR(l3, 0446, driver_show_l4, NULL); > > + __ATTR(l2, 04444, driver_show_l4, NULL); > > + __ATTR(l4, 0444, driver_show_l4, NULL); > > > > ... would you notice it at a glance that it contains two security holes? > > I've tried to deal with that in the past with the __ATTR_RW() and > __ATTR_RO() and __ATTR_WO() macros that more should be using. I swept > the tree a few years ago to try to fix up most of them, but I know I > didn't catch them all, and more files have been added since then. > > > While the weird permissions in this: > > > > + __ATTR(l1, PERM_r__r__r__, driver_show_l4, NULL); > > + __ATTR(l3, PERM_r__r__rw_, driver_show_l4, NULL); > > + __ATTR(l2, PERM_sr__r__r__, driver_show_l4, NULL); > > + __ATTR(l4, PERM_r__r__r__, driver_show_l4, NULL); > > > > Wouln't even build, because the dangerous patterns of PERM_r__r__rw_ or > > PERM_sr__r__r__ are not defined to begin with. > > Because of that, odds are people will just stick to the octal numbers, > because they think they want something other than the ones you defined > for foolish reasons :) For code I maintain I'd insist on contributors using the human readable versions, because in the past I've mixed up octals (and the symbolic helpers we have today) myself and I find the 'ls -l' format much easier to read because that's the primary file permission format I see every day working on code. > That being said, I do like them much better than the macros we have today, which > I always have to go and look up every time I see them... Same here! I'm sure core VFS developers know all of the octals and the helpers by heart, but the set of maintainers accepting debugfs and sysfs file permission patches is much wider than that, so every little bit of clarity helps. Thanks, Ingo
On Wed, Aug 03, 2016 at 10:39:03AM +0200, Ingo Molnar wrote: > > > While the weird permissions in this: > > > > > > + __ATTR(l1, PERM_r__r__r__, driver_show_l4, NULL); > > > + __ATTR(l3, PERM_r__r__rw_, driver_show_l4, NULL); > > > + __ATTR(l2, PERM_sr__r__r__, driver_show_l4, NULL); > > > + __ATTR(l4, PERM_r__r__r__, driver_show_l4, NULL); > > > > > > Wouln't even build, because the dangerous patterns of PERM_r__r__rw_ or > > > PERM_sr__r__r__ are not defined to begin with. > > > > Because of that, odds are people will just stick to the octal numbers, > > because they think they want something other than the ones you defined > > for foolish reasons :) > > For code I maintain I'd insist on contributors using the human readable versions, > because in the past I've mixed up octals (and the symbolic helpers we have today) > myself and I find the 'ls -l' format much easier to read because that's the > primary file permission format I see every day working on code. FWIW, the only "human readable" ones for me are the octal ones, which are also the same as those I'm using every day with "chmod" or "find" and that I find hard to get wrong. But I agree that the PERM_* idea above are a nice alternative since they match the "ls -l" output, and you can even add the directory flag there with "d" like "ls" does. You could also have PERM_0444 and similar for those who are more at ease with the octal numers without defining the few ones that are definitely wrong, as a safety belt. > > That being said, I do like them much better than the macros we have today, which > > I always have to go and look up every time I see them... > > Same here! Same for me. I never use S_I* and never know where to look for their definitions when I see them. Willy
Hi Greg, >> An added advantage would be that during review it would stick out like a sore >> thumb if anyone used a 'weird' permission variant. >> >> For example, if you saw these lines in a driver patch: >> >> + __ATTR(l1, 0444, driver_show_l4, NULL); >> + __ATTR(l3, 0446, driver_show_l4, NULL); >> + __ATTR(l2, 04444, driver_show_l4, NULL); >> + __ATTR(l4, 0444, driver_show_l4, NULL); >> >> ... would you notice it at a glance that it contains two security holes? > > I've tried to deal with that in the past with the __ATTR_RW() and > __ATTR_RO() and __ATTR_WO() macros that more should be using. I swept > the tree a few years ago to try to fix up most of them, but I know I > didn't catch them all, and more files have been added since then. I said in another response that maybe module_param_rw and module_param_ro will make some sense. Not sure if they are easier to read or not. I mean for each usage, we could look at the tree and see what values are actually used. My bet is that for module_param only a few ones are used. I have the feeling it is 0444 or 0644 and nothing else. Maybe some outlaws with 0400 and 0600 that don't even need to be that secretive. Regards Marcel
On Wed, 2016-08-03 at 10:11 +0200, Ingo Molnar wrote: > * Linus Torvalds <torvalds@linux-foundation.org> wrote: > > [ So I answered similarly to another patch, but I'll just re-iterate > > and change the subject line so that it stands out a bit from the > > millions of actual patches ] > > > > On Tue, Aug 2, 2016 at 1:42 PM, Pavel Machek <pavel@ucw.cz> wrote: > > > Everyone knows what 0644 is, but noone can read S_IRUSR | S_IWUSR | > > > S_IRCRP | S_IROTH (*). Please don't do this. > > Absolutely. It's *much* easier to parse and understand the octal > > numbers, while the symbolic macro names are just random line noise and > > hard as hell to understand. You really have to think about it. > > > > So we should rather go the other way: convert existing bad symbolic > > permission bit macro use to just use the octal numbers. > In addition to that I'd love to have something even easier to read, a few common > variants of the permissions field of 'ls -l' pre-defined. I did some quick > grepping, and collected the main variants that are in use: > > PERM_r________ 0400 > PERM_r__r_____ 0440 > PERM_r__r__r__ 0444 [etc] While the proposed PERM_ variants are easily read, using a single style instead of 2+ incompatible symbolic styles makes treewide misuse identification via grep style tools easier.
On Wed 2016-08-03 10:11:40, Ingo Molnar wrote: > > * Linus Torvalds <torvalds@linux-foundation.org> wrote: > > > [ So I answered similarly to another patch, but I'll just re-iterate > > and change the subject line so that it stands out a bit from the > > millions of actual patches ] > > > > On Tue, Aug 2, 2016 at 1:42 PM, Pavel Machek <pavel@ucw.cz> wrote: > > > > > > Everyone knows what 0644 is, but noone can read S_IRUSR | S_IWUSR | > > > S_IRCRP | S_IROTH (*). Please don't do this. > > > > Absolutely. It's *much* easier to parse and understand the octal > > numbers, while the symbolic macro names are just random line noise and > > hard as hell to understand. You really have to think about it. > > > > So we should rather go the other way: convert existing bad symbolic > > permission bit macro use to just use the octal numbers. > > In addition to that I'd love to have something even easier to read, a few common > variants of the permissions field of 'ls -l' pre-defined. I did some quick > grepping, and collected the main variants that are in use: > > PERM_r________ 0400 > PERM_r__r_____ 0440 > PERM_r__r__r__ 0444 I see 0400 and 0444 making sense, but does 0440 really make sense? I assume it will be uid/gid 0/0? Is gid 0 really estabilished well enough to give it special permissions? And yes, these macros actually help readability. > PERM__wx______ 0300 > PERM__wx_wx___ 0330 > PERM__wx_wx_wx 0333 Uh. This is for sysfs. Do we event want any __x variants? _wx would certainly be strange. (And yes, we can keep people from using strange permissions by simply not defining those macros.) > Allowing these would be nice too, because there were cases in the past where > people messed up the octal representation or our internal symbolic helpers, > but this representation is fundamentally self-describing and pretty 'fool proof'. > > An added advantage would be that during review it would stick out like a sore > thumb if anyone used a 'weird' permission variant. > > For example, if you saw these lines in a driver patch: > > + __ATTR(l1, 0444, driver_show_l4, NULL); > + __ATTR(l3, 0446, driver_show_l4, NULL); > + __ATTR(l2, 04444, driver_show_l4, NULL); > + __ATTR(l4, 0444, driver_show_l4, NULL); > > ... would you notice it at a glance that it contains two security holes? I see two bugs but only one hole. How can you exploit s-bit without corresponding x-bit? I'd delete these: I don't think we should encourage their use: > +#define PERM_r__r_____ 0440 > +#define PERM_rw_r_____ 0640 > +#define PERM_rw_rw_r__ 0664 > + > +#define PERM__w__w__w_ 0222 > + > +#define PERM_r_x______ 0500 > +#define PERM_r_xr_x___ 0550 > +#define PERM_r_xr_xr_x 0555 > + > +#define PERM_rwx______ 0700 > +#define PERM_rwxr_x___ 0750 > +#define PERM_rwxr_xr_x 0755 > +#define PERM_rwxrwxr_x 0775 > +#define PERM_rwxrwxrwx 0777 > + > +#define PERM__wx______ 0300 > +#define PERM__wx_wx___ 0330 > +#define PERM__wx_wx_wx 0333 Pavel
diff --git a/include/linux/stat.h b/include/linux/stat.h index 075cb0c7eb2a..863d5563427f 100644 --- a/include/linux/stat.h +++ b/include/linux/stat.h @@ -5,6 +5,38 @@ #include <asm/stat.h> #include <uapi/linux/stat.h> +/* + * Human readable symbolic definitions for common + * file permissions: + */ +#define PERM_r________ 0400 +#define PERM_r__r_____ 0440 +#define PERM_r__r__r__ 0444 + +#define PERM_rw_______ 0600 +#define PERM_rw_r_____ 0640 +#define PERM_rw_r__r__ 0644 +#define PERM_rw_rw_r__ 0664 +#define PERM_rw_rw_rw_ 0666 + +#define PERM__w_______ 0200 +#define PERM__w__w____ 0220 +#define PERM__w__w__w_ 0222 + +#define PERM_r_x______ 0500 +#define PERM_r_xr_x___ 0550 +#define PERM_r_xr_xr_x 0555 + +#define PERM_rwx______ 0700 +#define PERM_rwxr_x___ 0750 +#define PERM_rwxr_xr_x 0755 +#define PERM_rwxrwxr_x 0775 +#define PERM_rwxrwxrwx 0777 + +#define PERM__wx______ 0300 +#define PERM__wx_wx___ 0330 +#define PERM__wx_wx_wx 0333 + #define S_IRWXUGO (S_IRWXU|S_IRWXG|S_IRWXO) #define S_IALLUGO (S_ISUID|S_ISGID|S_ISVTX|S_IRWXUGO) #define S_IRUGO (S_IRUSR|S_IRGRP|S_IROTH)