objtool: UACCESS validation
Subsystems (6)
Peter Zijlstra
maintainercover objtool: UACCESS validation
Teach objtool to validate the UACCESS (SMAP, PAN) rules which are currently unenforced and (therefore obviously) violated.
UACCESS sections should be small; we want to limit the amount of code that can touch userspace. Furthermore, UACCESS state isn't scheduled, this means that anything that directly calls into the scheduler will result in random code running with UACCESS enabled and possibly getting back into the UACCESS region with UACCESS disabled and causing faults.
Forbid any CALL/RET while UACCESS is enabled; but provide an annotation to mark (a very limited) set of functions as UACCESS-safe (eg. the planned:
unsafe_copy_{to,from}_user()).
--- arch/x86/ia32/ia32_signal.c | 29 ++++--- arch/x86/include/asm/uaccess.h | 4 +- include/linux/frame.h | 49 ++++++++++- tools/objtool/Makefile | 2 +- tools/objtool/arch.h | 6 +- tools/objtool/arch/x86/decode.c | 22 ++++- tools/objtool/check.c | 180 ++++++++++++++++++++++++++++++---------- tools/objtool/check.h | 3 +- tools/objtool/elf.h | 1 + 9 files changed, 234 insertions(+), 62 deletions(-)
Peter Zijlstra
maintainerPATCH 1/6 x86/uaccess: Dont evaluate argument inside AC region
AC regions are special, make sure we don't do possibly complex evaluations there.
--- arch/x86/include/asm/uaccess.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/include/asm/uaccess.h +3 -1
@@ -431,8 +431,10 @@ do { \ ({ \ __label__ __pu_label; \ int __pu_err = -EFAULT; \ + __typeof__(*(ptr)) __pu_val; \ + __pu_val = x; \ __uaccess_begin(); \ - __put_user_size((x), (ptr), (size), __pu_label); \ + __put_user_size(__pu_val, (ptr), (size), __pu_label); \ __pu_err = 0; \ __pu_label: \ __uaccess_end(); \
Re: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
Quoted 5 lines · Peter Zijlstra
AC regions are special, make sure we don't do possibly complex evaluations there.
Probably-Signed-off-by Andy Lutomirski <[email protected]>
I put a slightly fancier version here last night so that the 0day bot could chew on it:
want to snarf it up? It's the same thing except that I added Linus' suggested change, found the offending patch that broke it (hi Linus!) and added a bit more changelog :)
Peter Zijlstra
maintainerRe: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
Quoted 7 lines · Andy Lutomirski
Quoted 5 lines · Peter Zijlstra
AC regions are special, make sure we don't do possibly complex evaluations there.
Probably-Signed-off-by Andy Lutomirski <[email protected]>
Quoted 8 lines
I put a slightly fancier version here last night so that the 0day bot could chew on it:
want to snarf it up? It's the same thing except that I added Linus' suggested change, found the offending patch that broke it (hi Linus!) and added a bit more changelog :)
Sure; I'll take that one.
Borislav Petkov
maintainerRe: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
In reply to Andy LutomirskiQuoted 2 lines · Andy Lutomirski
I put a slightly fancier version here last night so that the 0day bot could chew on it:
I'm guessing 0day bot didn't complain so far? Or should we give it one more day?
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
Quoted 8 lines · Borislav Petkov
Quoted 2 lines · Andy Lutomirski
I put a slightly fancier version here last night so that the 0day bot could chew on it:
I'm guessing 0day bot didn't complain so far? Or should we give it one more day?
No complaints yet. I'll holler if it complains. This one patch could plausibly be -stable or 5.0 material, I suppose.
Re: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
In reply to Andy LutomirskiQuoted 5 lines · Andy Lutomirski
I put a slightly fancier version here last night so that the 0day bot could chew on it:
Ack. Even if it does seem to get the whitespace wrong and not align things ;)
Linus
Borislav Petkov
maintainerRe: [PATCH 1/6] x86/uaccess: Dont evaluate argument inside AC region
Quoted 2 lines · Linus Torvalds
Ack. Even if it does seem to get the whitespace wrong and not align things ;)
It is the top commit so fixed up and added your ACK. :)
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.tip-bot for Andy Lutomirski
merge bot[tip:x86/urgent] x86/uaccess: Don't leak the AC flag into __put_user() value evaluation
In reply to Peter Zijlstrax86/uaccess: Don't leak the AC flag into __put_user() value evaluation
Commit-ID: 1ee2bd5e09195d5476daefec5c64ba597a0a9920 Gitweb: https://git.kernel.org/tip/1ee2bd5e09195d5476daefec5c64ba597a0a9920 Author: Andy Lutomirski <[email protected]> AuthorDate: Fri, 22 Feb 2019 17:17:04 -0800 Committer: Borislav Petkov <[email protected]> CommitDate: Mon, 25 Feb 2019 18:55:04 +0100
x86/uaccess: Don't leak the AC flag into __put_user() value evaluation
When calling __put_user(foo(), ptr), the __put_user() macro would call foo() in between __uaccess_begin() and __uaccess_end(). If that code were buggy, then those bugs would be run without SMAP protection.
Fortunately, there seem to be few instances of the problem in the kernel. Nevertheless, __put_user() should be fixed to avoid doing this. Therefore, evaluate __put_user()'s argument before setting AC.
This issue was noticed when an objtool hack by Peter Zijlstra complained about genregs_get() and I compared the assembly output to the C source.
[ bp: Massage commit message. ]
--- arch/x86/include/asm/uaccess.h | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/x86/include/asm/uaccess.h +4 -2
@@ -284,7 +284,7 @@ do { \__put_user_goto (x, ptr,"l" ,"k" ,"ir" , label); \break ; \case 8 : \ -__put_user_goto_u64 ((__typeof__ (*ptr))(x), ptr, label); \ +__put_user_goto_u64 (x, ptr, label); \break ; \default : \__put_user_bad (); \
@@ -431,8 +431,10 @@ do { \ ({ \ __label__ __pu_label; \ int __pu_err = -EFAULT; \ + __typeof__(*(ptr)) __pu_val; \ + __pu_val = x; \ __uaccess_begin(); \ - __put_user_size((x), (ptr), (size), __pu_label); \ + __put_user_size(__pu_val, (ptr), (size), __pu_label); \ __pu_err = 0; \ __pu_label: \ __uaccess_end(); \
tip-bot for Andy Lutomirski
merge bot[tip:x86/urgent] x86/uaccess: Don't leak the AC flag into __put_user() value evaluation
In reply to Peter Zijlstrax86/uaccess: Don't leak the AC flag into __put_user() value evaluation
Commit-ID: 2a418cf3f5f1caf911af288e978d61c9844b0695 Gitweb: https://git.kernel.org/tip/2a418cf3f5f1caf911af288e978d61c9844b0695 Author: Andy Lutomirski <[email protected]> AuthorDate: Fri, 22 Feb 2019 17:17:04 -0800 Committer: Borislav Petkov <[email protected]> CommitDate: Mon, 25 Feb 2019 20:17:05 +0100
x86/uaccess: Don't leak the AC flag into __put_user() value evaluation
When calling __put_user(foo(), ptr), the __put_user() macro would call foo() in between __uaccess_begin() and __uaccess_end(). If that code were buggy, then those bugs would be run without SMAP protection.
Fortunately, there seem to be few instances of the problem in the kernel. Nevertheless, __put_user() should be fixed to avoid doing this. Therefore, evaluate __put_user()'s argument before setting AC.
This issue was noticed when an objtool hack by Peter Zijlstra complained about genregs_get() and I compared the assembly output to the C source.
[ bp: Massage commit message and fixed up whitespace. ]
--- arch/x86/include/asm/uaccess.h | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/x86/include/asm/uaccess.h +4 -2
@@ -284,7 +284,7 @@ do { \__put_user_goto (x, ptr,"l" ,"k" ,"ir" , label); \break ; \case 8 : \ -__put_user_goto_u64 ((__typeof__ (*ptr))(x), ptr, label); \ +__put_user_goto_u64 (x, ptr, label); \break ; \default : \__put_user_bad (); \
@@ -431,8 +431,10 @@ do { \ ({ \ __label__ __pu_label; \ int __pu_err = -EFAULT; \ + __typeof__(*(ptr)) __pu_val; \ + __pu_val = x; \ __uaccess_begin(); \ - __put_user_size((x), (ptr), (size), __pu_label); \ + __put_user_size(__pu_val, (ptr), (size), __pu_label); \ __pu_err = 0; \ __pu_label: \ __uaccess_end(); \
Peter Zijlstra
maintainerPATCH 2/6 x86/ia32: Fix ia32_restore_sigcontext AC leak
In reply to Peter ZijlstraDon't call load_gs_index() with AC set; delay the segment setting until after the AC section.
--- arch/x86/ia32/ia32_signal.c | 29 +++++++++++++++++++---------- 1 file changed, 19 insertions(+), 10 deletions(-)
arch/x86/ia32/ia32_signal.c +19 -10
@@ -60,17 +60,21 @@ regs->seg =GET_SEG (seg) |3 ; \ } while (0) -#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ - unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ - if (pre != cur) \ - set_user_seg(seg, pre); \ +#define LOAD_SEG (seg) { \ + pre_##seg = 3 | GET_SEG(seg); \ + cur_##seg = get_user_seg(seg); \ +} + +#define RELOAD_SEG (seg) { \ + if (pre_##seg != cur_##seg) \ + set_user_seg(seg, pre_##seg); \ }static int ia32_restore_sigcontext (struct pt_regs *regs,struct sigcontext_32 __user *sc) { +u16 pre_gs, pre_fs, pre_ds, pre_es; +u16 cur_gs, cur_fs, cur_ds, cur_es;unsigned int tmpflags, err =0 ;void __user *buf;u32 tmp;
@@ -85,10 +89,10 @@ static int ia32_restore_sigcontext(struc* the handler, but does not clobber them at least in the * normal case. */ -RELOAD_SEG (gs); -RELOAD_SEG (fs); -RELOAD_SEG (ds); -RELOAD_SEG (es); +LOAD_SEG (gs); +LOAD_SEG (fs); +LOAD_SEG (ds); +LOAD_SEG (es);COPY (di);COPY (si);COPY (bp);COPY (sp);COPY (bx);COPY (dx);COPY (cx);COPY (ip);COPY (ax);
@@ -106,6 +110,11 @@ static int ia32_restore_sigcontext(struc buf =compat_ptr (tmp); }get_user_catch (err); +RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es); + err |=fpu__restore_sig (buf,1 );force_iret ();
Re: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
Quoted 20 lines · Peter Zijlstra
Don't call
load_gs_index()with AC set; delay the segment setting until after the AC section.Signed-off-by Peter Zijlstra (Intel) <[email protected]>--- arch/x86/ia32/ia32_signal.c | 29 +++++++++++++++++++---------- 1 file changed, 19 insertions(+), 10 deletions(-)arch/x86/ia32/ia32_signal.c +0
-5@@ -60,17 +60,21 @@ regs->seg =GET_SEG (seg) |3 ; \ }while (0 ) -#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ - unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ - if (pre != cur) \
arch/x86/ia32/ia32_signal.c line 71
Quoted 10 lines
arch/x86/ia32/ia32_signal.c +8
-1-set_user_seg (seg, pre); \ +#define LOAD_SEG (seg) { \ + pre_##seg = 3 | GET_SEG(seg); \ + cur_##seg = get_user_seg(seg); \ +} + +#define RELOAD_SEG (seg) { \ + if (pre_##seg != cur_##seg) \ + set_user_seg(seg, pre_##seg); \ }
This is so tangled.
How about changing RELOAD_SEG to replace unsigned int pre = GET_SEG(seg); with unsigned int pre = (seg); to make it less magic. Then do:
unsigned int gs = GET_SEG(gs);
...
RELOAD_SEG(gs);
And now the code actually does what it looks like it does.
Peter Zijlstra
maintainerRe: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
Quoted 6 lines · Andy Lutomirski
This is so tangled.
How about changing
RELOAD_SEGto replace unsigned int pre =GET_SEG(seg); with unsigned int pre = (seg); to make it less magic. Then do:
Quoted 7 lines
unsigned int gs =
GET_SEG(gs);...
RELOAD_SEG(gs);And now the code actually does what it looks like it does.
Is this what you mean?
---
arch/x86/ia32/ia32_signal.c +19 -14
@@ -53,17 +53,16 @@#define GET_SEG (seg) ({ \ unsigned short tmp; \ get_user_ex(tmp, &sc->seg); \ - tmp; \ + tmp | 3; \ })#define COPY_SEG_CPL3 (seg) do { \ - regs->seg = GET_SEG(seg) | 3; \ + regs->seg = GET_SEG(seg); \ } while (0)#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg); \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ if (pre != cur) \ set_user_seg(seg, pre); \ }
@@ -72,6 +71,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, struct sigcontext_32 __user *sc) {unsigned int tmpflags, err =0 ; +u16 gs, fs, es, ds;void __user *buf;u32 tmp;
@@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, current->restart_block.fn = do_no_restart_syscall; get_user_try { -/* -* Reload fs and gs if they have changed in the signal -* handler. This does not handle long fs/gs base changes in -* the handler, but does not clobber them at least in the -* normal case. -*/ -RELOAD_SEG (gs); -RELOAD_SEG (fs); -RELOAD_SEG (ds); -RELOAD_SEG (es); + gs =GET_SEG (gs); + fs =GET_SEG (fs); + ds =GET_SEG (ds); + es =GET_SEG (es);COPY (di);COPY (si);COPY (bp);COPY (sp);COPY (bx);COPY (dx);COPY (cx);COPY (ip);COPY (ax);
@@ -106,6 +100,17 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, buf =compat_ptr (tmp); }get_user_catch (err); +/* +* Reload fs and gs if they have changed in the signal +* handler. This does not handle long fs/gs base changes in +* the handler, but does not clobber them at least in the +* normal case. +*/ +RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es); + err |=fpu__restore_sig (buf,1 );force_iret ();
Re: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
Quoted 14 lines · Peter Zijlstra
Quoted 10 lines · Andy Lutomirski
This is so tangled.
How about changing
RELOAD_SEGto replace unsigned int pre =GET_SEG(seg); with unsigned int pre = (seg); to make it less magic. Then do:unsigned int gs =
GET_SEG(gs);
Quoted 7 lines
Quoted 5 lines
...
RELOAD_SEG(gs);And now the code actually does what it looks like it does.
Is this what you mean?
Yes, except:
Quoted 7 lines
---
arch/x86/ia32/ia32_signal.c +0
-0
arch/x86/ia32/ia32_signal.c line 58
Quoted 7 lines
arch/x86/ia32/ia32_signal.c +1
-1@@ -53,17 +53,16 @@#define GET_SEG (seg) ({ \ unsigned short tmp; \ get_user_ex(tmp, &sc->seg); \ - tmp; \ + tmp | 3; \ })
Drop this part.
arch/x86/ia32/ia32_signal.c line 61
Quoted 4 lines
arch/x86/ia32/ia32_signal.c +1
-1#define COPY_SEG_CPL3 (seg) do { \ - regs->seg = GET_SEG(seg) | 3; \ + regs->seg = GET_SEG(seg); \ } while (0)
And this.
Unfortunately, whether we want the | 3 varies by segment. For FS and GS, we definitely don’t want it, since 0 is a common and important value, and 3 is a deeply screwed up value. (3 is legal to *write* to GS, and it sticks, but IRET silently changes it to 0, because the original 386 designers (I assume) confused themselves.
Quoted 46 lines
arch/x86/ia32/ia32_signal.c +10
-12#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg); \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ if (pre != cur) \ set_user_seg(seg, pre); \ }@@ -72,6 +71,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, struct sigcontext_32 __user *sc) {unsigned int tmpflags, err =0 ; +u16 gs, fs, es, ds;void __user *buf;u32 tmp;@@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, current->restart_block.fn = do_no_restart_syscall; get_user_try { -/* -* Reload fs and gs if they have changed in the signal -* handler. This does not handle long fs/gs base changes in -* the handler, but does not clobber them at least in the -* normal case. -*/ -RELOAD_SEG (gs); -RELOAD_SEG (fs); -RELOAD_SEG (ds); -RELOAD_SEG (es); + gs =GET_SEG (gs); + fs =GET_SEG (fs); + ds =GET_SEG (ds); + es =GET_SEG (es);COPY (di);COPY (si);COPY (bp);COPY (sp);COPY (bx);COPY (dx);COPY (cx);COPY (ip);COPY (ax);@@ -106,6 +100,17 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, buf =compat_ptr (tmp); }get_user_catch (err); +/* +* Reload fs and gs if they have changed in the signal +* handler. This does not handle long fs/gs base changes in +* the handler, but does not clobber them at least in the
arch/x86/ia32/ia32_signal.c static int ia32_restore_sigcontext(struct pt_regs *regs, · line 116
Quoted 10 lines
arch/x86/ia32/ia32_signal.c +7
-0+* normal case. +*/ +RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es); + err |=fpu__restore_sig (buf,1 );force_iret ();
I
Peter Zijlstra
maintainerRe: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
Quoted 20 lines · Andy Lutomirski
arch/x86/ia32/ia32_signal.c line 58
Quoted 12 lines
arch/x86/ia32/ia32_signal.c +1
-1@@ -53,17 +53,16 @@#define GET_SEG (seg) ({ \ unsigned short tmp; \ get_user_ex(tmp, &sc->seg); \ - tmp; \ + tmp | 3; \ })Drop this part.
arch/x86/ia32/ia32_signal.c line 61
Quoted 4 lines
arch/x86/ia32/ia32_signal.c +1
-1#define COPY_SEG_CPL3 (seg) do { \ - regs->seg = GET_SEG(seg) | 3; \ + regs->seg = GET_SEG(seg); \ } while (0)
Quoted 14 lines
And this.
Unfortunately, whether we want the | 3 varies by segment. For FS and GS, we definitely don’t want it, since 0 is a common and important value, and 3 is a deeply screwed up value. (3 is legal to *write* to GS, and it sticks, but IRET silently changes it to 0, because the original 386 designers (I assume) confused themselves.
Quoted 6 lines
arch/x86/ia32/ia32_signal.c +1
-2#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg); \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \
And here ?
Quoted 3 lines
arch/x86/ia32/ia32_signal.c line 67
Quoted 3 lines
arch/x86/ia32/ia32_signal.c +0
-0if (pre != cur) \set_user_seg (seg, pre); \ }
Thing is; afaict the current code _always_ does |3 on any GET_SET() result.
If you want to change that; I'm fine with that, but lets not do that in this same patch.
So then?
arch/x86/ia32/ia32_signal.c +17 -12
@@ -61,9 +61,8 @@ } while (0)#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg) | 3; \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ if (pre != cur) \ set_user_seg(seg, pre); \ }
@@ -72,6 +71,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, struct sigcontext_32 __user *sc) {unsigned int tmpflags, err =0 ; +u16 gs, fs, es, ds;void __user *buf;u32 tmp;
@@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, current->restart_block.fn = do_no_restart_syscall; get_user_try { -/* -* Reload fs and gs if they have changed in the signal -* handler. This does not handle long fs/gs base changes in -* the handler, but does not clobber them at least in the -* normal case. -*/ -RELOAD_SEG (gs); -RELOAD_SEG (fs); -RELOAD_SEG (ds); -RELOAD_SEG (es); + gs =GET_SEG (gs); + fs =GET_SEG (fs); + ds =GET_SEG (ds); + es =GET_SEG (es);COPY (di);COPY (si);COPY (bp);COPY (sp);COPY (bx);COPY (dx);COPY (cx);COPY (ip);COPY (ax);
@@ -106,6 +100,17 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, buf =compat_ptr (tmp); }get_user_catch (err); +/* +* Reload fs and gs if they have changed in the signal +* handler. This does not handle long fs/gs base changes in +* the handler, but does not clobber them at least in the +* normal case. +*/ +RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es); + err |=fpu__restore_sig (buf,1 );force_iret ();
Peter Zijlstra
maintainerRe: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
arch/x86/ia32/ia32_signal.c static int ia32_restore_sigcontext(struct pt_regs *regs, · line 112
Quoted 4 lines · Peter Zijlstra
arch/x86/ia32/ia32_signal.c +4
-0+RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es);
Also; is that the canonical order ? It bugs the hell out of me to not have that alphabetically correct. Shall I correct that?
Re: [PATCH 2/6] x86/ia32: Fix ia32_restore_sigcontext AC leak
In reply to Peter ZijlstraQuoted 44 lines · Peter Zijlstra
Quoted 20 lines · Andy Lutomirski
arch/x86/ia32/ia32_signal.c line 58
Quoted 12 lines
arch/x86/ia32/ia32_signal.c +1
-1@@ -53,17 +53,16 @@#define GET_SEG (seg) ({ \ unsigned short tmp; \ get_user_ex(tmp, &sc->seg); \ - tmp; \ + tmp | 3; \ })Drop this part.
arch/x86/ia32/ia32_signal.c line 61
Quoted 4 lines
arch/x86/ia32/ia32_signal.c +1
-1#define COPY_SEG_CPL3 (seg) do { \ - regs->seg = GET_SEG(seg) | 3; \ + regs->seg = GET_SEG(seg); \ } while (0)Quoted 14 lines
And this.
Unfortunately, whether we want the | 3 varies by segment. For FS and GS, we definitely don’t want it, since 0 is a common and important value, and 3 is a deeply screwed up value. (3 is legal to *write* to GS, and it sticks, but IRET silently changes it to 0, because the original 386 designers (I assume) confused themselves.
Quoted 6 lines
arch/x86/ia32/ia32_signal.c +1
-2#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg); \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \And here ?
Quoted 3 lines
arch/x86/ia32/ia32_signal.c line 68
Quoted 3 lines
arch/x86/ia32/ia32_signal.c +0
-0if (pre != cur) \set_user_seg (seg, pre); \ }
Quoted 5 lines
Thing is; afaict the current code _always_ does |3 on any
GET_SET()result.If you want to change that; I'm fine with that, but lets not do that in this same patch.
Ugh, you're right. I bet I can come up with some awful case involving ptrace and sigreturn where this causes problems.
I have a patch series I need to dust off that deletes this entire file. It's this and its parents:
Quoted 55 lines
So then?
arch/x86/ia32/ia32_signal.c +10
-12@@ -61,9 +61,8 @@ } while (0)#define RELOAD_SEG (seg) { \ - unsigned int pre = GET_SEG(seg); \ + unsigned int pre = (seg) | 3; \ unsigned int cur = get_user_seg(seg); \ - pre |= 3; \ if (pre != cur) \ set_user_seg(seg, pre); \ }@@ -72,6 +71,7 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, struct sigcontext_32 __user *sc) {unsigned int tmpflags, err =0 ; +u16 gs, fs, es, ds;void __user *buf;u32 tmp;@@ -79,16 +79,10 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, current->restart_block.fn = do_no_restart_syscall; get_user_try { -/* -* Reload fs and gs if they have changed in the signal -* handler. This does not handle long fs/gs base changes in -* the handler, but does not clobber them at least in the -* normal case. -*/ -RELOAD_SEG (gs); -RELOAD_SEG (fs); -RELOAD_SEG (ds); -RELOAD_SEG (es); + gs =GET_SEG (gs); + fs =GET_SEG (fs); + ds =GET_SEG (ds); + es =GET_SEG (es);COPY (di);COPY (si);COPY (bp);COPY (sp);COPY (bx);COPY (dx);COPY (cx);COPY (ip);COPY (ax);@@ -106,6 +100,17 @@ static int ia32_restore_sigcontext(struct pt_regs *regs, buf =compat_ptr (tmp); }get_user_catch (err); +/* +* Reload fs and gs if they have changed in the signal +* handler. This does not handle long fs/gs base changes in +* the handler, but does not clobber them at least in the
arch/x86/ia32/ia32_signal.c static int ia32_restore_sigcontext(struct pt_regs *regs, · line 116
Quoted 10 lines
arch/x86/ia32/ia32_signal.c +7
-0+* normal case. +*/ +RELOAD_SEG (gs); +RELOAD_SEG (fs); +RELOAD_SEG (ds); +RELOAD_SEG (es); + err |=fpu__restore_sig (buf,1 );force_iret ();
Looks good to me. The order of the restores shouldn't matter at all.
Peter Zijlstra
maintainerPATCH 3/6 objtool: Set insn->func for alternatives
In reply to Peter ZijlstraMake sure we set the function association for alternative function sequences; they are after all still part of the function.
--- tools/objtool/check.c | 1 + 1 file changed, 1 insertion(+)
tools/objtool/check.c +1 -0
@@ -695,6 +695,7 @@ static int handle_group_alt(struct objto last_new_insn = insn; insn->ignore = orig_insn->ignore_alts; + insn->func = orig_insn->func;if (insn->type !=INSN_JUMP_CONDITIONAL && insn->type !=INSN_JUMP_UNCONDITIONAL )
Peter Zijlstra
maintainerPATCH 4/6 objtool: Replace STACK_FRAME_NON_STANDARD annotation
In reply to Peter ZijlstraReplace the existing STACK_FRAME_NON_STANDARD annotation with a 'better' scheme.
The old annotation works by taking the address of a function; this is visible to the compiler and might affect code generation (the function pointer escapes). The new annotation simply stores the function name and has objtool do a symtab lookup.
--- include/linux/frame.h | 19 +++++++++++++--- tools/objtool/check.c | 58 +++++++++++++++++++++----------------------------- tools/objtool/check.h | 1 3 files changed, 41 insertions(+), 37 deletions(-)
include/linux/frame.h +16 -3
@@ -11,9 +11,22 @@* * For more information, see tools/objtool/Documentation/stack-validation.txt. */ -#define STACK_FRAME_NON_STANDARD (func) \ - static void __used __section(.discard.func_stack_frame_non_standard) \ - *__func_stack_frame_non_standard_##func = func +#define STACK_FRAME_NON_STANDARD (func) \ + asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.nonstd_frame\n\t" \ + ".long 999b - .\n\t" \ + ".popsection\n\t") + +/* +* SHT_STRTAB(@3) sections should start with a \0 byte such that the 0 offset +* encodes the NULL string. +*/ +asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" +".byte 0\n\t" +".popsection\n\t" );#else /* !CONFIG_STACK_VALIDATION */
tools/objtool/check.c +25 -33
82 lines not on this page · raw message
tools/objtool/check.h +0 -1
@@ -60,7 +60,6 @@ struct objtool_file {struct elf *elf;struct list_head insn_list;DECLARE_HASHTABLE (insn_hash,16 ); -struct section *whitelist;bool ignore_unreachables, c_file, hints, rodata; };
Re: [PATCH 4/6] objtool: Replace STACK_FRAME_NON_STANDARD annotation
Quoted 23 lines · Peter Zijlstra
Replace the existing
STACK_FRAME_NON_STANDARDannotation with a 'better' scheme.The old annotation works by taking the address of a function; this is visible to the compiler and might affect code generation (the function pointer escapes). The new annotation simply stores the function name and has objtool do a symtab lookup.
Signed-off-by Peter Zijlstra (Intel) <[email protected]>--- include/linux/frame.h | 19 +++++++++++++--- tools/objtool/check.c | 58 +++++++++++++++++++++----------------------------- tools/objtool/check.h | 1 3 files changed, 41 insertions(+), 37 deletions(-)include/linux/frame.h +0
-2@@ -11,9 +11,22 @@* * For more information, see tools/objtool/Documentation/stack-validation.txt. */ -#define STACK_FRAME_NON_STANDARD (func) \ - static void __used __section(.discard.func_stack_frame_non_standard) \
include/linux/frame.h line 22
Quoted 10 lines
include/linux/frame.h +9
-1- *__func_stack_frame_non_standard_##func = func +#define STACK_FRAME_NON_STANDARD (func) \ + asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.nonstd_frame\n\t" \ + ".long 999b - .\n\t" \ + ".popsection\n\t") +
I don't think this will work in the case where GCC does an IPA optimization and renames the function (e.g., renames func to func.isra.1234), right? That might be a deal breaker...
include/linux/frame.h line 29
Quoted 7 lines
include/linux/frame.h +7
-0+/* +* SHT_STRTAB(@3) sections should start with a \0 byte such that the 0 offset +* encodes the NULL string. +*/ +asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" +".byte 0\n\t" +".popsection\n\t" );
Including the file creates the section, which is a bit nasty. Instead just change STACK_FRAME_NON_STANDARD to prefix every string with a \0?
Signature · 1 line
Josh
Peter Zijlstra
maintainerRe: [PATCH 4/6] objtool: Replace STACK_FRAME_NON_STANDARD annotation
Quoted 30 lines · Josh Poimboeuf
Quoted 29 lines · Peter Zijlstra
Replace the existing
STACK_FRAME_NON_STANDARDannotation with a 'better' scheme.The old annotation works by taking the address of a function; this is visible to the compiler and might affect code generation (the function pointer escapes). The new annotation simply stores the function name and has objtool do a symtab lookup.
Signed-off-by Peter Zijlstra (Intel) <[email protected]>--- include/linux/frame.h | 19 +++++++++++++--- tools/objtool/check.c | 58 +++++++++++++++++++++----------------------------- tools/objtool/check.h | 1 3 files changed, 41 insertions(+), 37 deletions(-)include/linux/frame.h +5
-3@@ -11,9 +11,22 @@* * For more information, see tools/objtool/Documentation/stack-validation.txt. */ -#define STACK_FRAME_NON_STANDARD (func) \ - static void __used __section(.discard.func_stack_frame_non_standard) \ - *__func_stack_frame_non_standard_##func = func +#define STACK_FRAME_NON_STANDARD (func) \ + asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \
Quoted 8 lines
include/linux/frame.h line 22
Quoted 4 lines
include/linux/frame.h +4
-0+".pushsection .discard.nonstd_frame\n\t" \ +".long 999b - .\n\t" \ + ".popsection\n\t") +I don't think this will work in the case where GCC does an IPA optimization and renames the function (e.g., renames func to func.isra.1234), right? That might be a deal breaker...
*groan*... indeed :/ So back to the old scheme then? Andy, any other clever ideas?
Quoted 10 lines
include/linux/frame.h line 29
Quoted 7 lines
include/linux/frame.h +7
-0+/* +* SHT_STRTAB(@3) sections should start with a \0 byte such that the 0 offset +* encodes the NULL string. +*/ +asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" +".byte 0\n\t" +".popsection\n\t" );Including the file creates the section, which is a bit nasty. Instead just change
STACK_FRAME_NON_STANDARDto prefix every string with a \0?
That would get us \0\0 between every string, which I suppose is in-spec, but weird. I figured that since it is a .discard. section, getting it wasn't a problem.
Re: [PATCH 4/6] objtool: Replace STACK_FRAME_NON_STANDARD annotation
Quoted 6 lines · Peter Zijlstra
Quoted 2 lines
Including the file creates the section, which is a bit nasty. Instead just change
STACK_FRAME_NON_STANDARDto prefix every string with a \0?That would get us \0\0 between every string, which I suppose is in-spec, but weird. I figured that since it is a .discard. section, getting it wasn't a problem.
It would just bother me, as a power user of readelf, to see those unused sections everywhere :-) Maybe it's not a big deal though.
Signature · 1 line
Josh
Peter Zijlstra
maintainerRe: [PATCH 4/6] objtool: Replace STACK_FRAME_NON_STANDARD annotation
In reply to Josh PoimboeufQuoted 18 lines · Josh Poimboeuf
include/linux/frame.h line 22
Quoted 12 lines · Peter Zijlstra
include/linux/frame.h +9
-3-#define STACK_FRAME_NON_STANDARD (func) \ - static void __used __section(.discard.func_stack_frame_non_standard) \ - *__func_stack_frame_non_standard_##func = func +#define STACK_FRAME_NON_STANDARD (func) \ + asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.nonstd_frame\n\t" \ + ".long 999b - .\n\t" \ + ".popsection\n\t") +I don't think this will work in the case where GCC does an IPA optimization and renames the function (e.g., renames func to func.isra.1234), right? That might be a deal breaker...
Or; as has been found by 0day; the whole function gets inlined and the symbol no longer exists at all.
That's curable with a noinline, but all things considered, I think we should go back to the old horrible scheme. Andy?
Re: [PATCH 4/6] objtool: Replace STACK_FRAME_NON_STANDARD annotation
Quoted 21 lines · Peter Zijlstra
Quoted 18 lines · Josh Poimboeuf
include/linux/frame.h line 22
Quoted 12 lines · Peter Zijlstra
include/linux/frame.h +9
-3-#define STACK_FRAME_NON_STANDARD (func) \ - static void __used __section(.discard.func_stack_frame_non_standard) \ - *__func_stack_frame_non_standard_##func = func +#define STACK_FRAME_NON_STANDARD (func) \ + asm (".pushsection .discard.nonstd_frame_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.nonstd_frame\n\t" \ + ".long 999b - .\n\t" \ + ".popsection\n\t") +I don't think this will work in the case where GCC does an IPA optimization and renames the function (e.g., renames func to func.isra.1234), right? That might be a deal breaker...
Quoted 5 lines
Or; as has been found by 0day; the whole function gets inlined and the symbol no longer exists at all.
That's curable with a noinline, but all things considered, I think we should go back to the old horrible scheme. Andy?
Ugh, I guess. I'm wondering just how atrocious the generated code is.
Just as a thought experiment, here are some other options:
1. Make a tiny GCC plugin that parses a special attribute on function declarations and emits a record that describes that attribute into the object file for each referenced symbol that comes from that declaration. (I know nothing about GCC internals, so I don't know how hard this would be.)
2. Fiddle with the function names. Turn a function called foo() into __uaccess_safe_foo() and also emit a weak alias from that to foo. This is probably every bit as bad as taking the address.
3. Take advantage of the fact that only static functions are (for now) subject to this IPA stuff. So take the address of a static function or just declare that calling a static function is uaccess safe.
--Andy
Peter Zijlstra
maintainerPATCH 5/6 objtool: Add UACCESS validation
In reply to Peter ZijlstraIt is important that UACCESS regions are as small as possible; furthermore the UACCESS state is not scheduled, so doing anything that might directly call into the scheduler will cause random code to be ran with UACCESS enabled.
Teach objtool too track UACCESS state and warn about any CALL made while UACCESS is enabled. This very much includes the __fentry__() tracing calls and __preempt_schedule() calls.
Note that exceptions _do_ save/restore the UACCESS state, and therefore they can drive preemption. This also means that all exception handlers must have an otherwise dedundant UACCESS disable instruction; therefore ignore this warning for !STT_FUNC code (exception handlers are not normal functions).
It also provides a UACCESS_SAFE() annotation which allows explicit annotation. This is meant to be used for future things like:
unsafe_copy_{to,from}_user().
--- include/linux/frame.h | 30 +++++++++++- tools/objtool/arch.h | 4 + tools/objtool/arch/x86/decode.c | 14 +++++ tools/objtool/check.c | 100 ++++++++++++++++++++++++++++++++++++---- tools/objtool/check.h | 2 tools/objtool/elf.h | 1 6 files changed, 138 insertions(+), 13 deletions(-)
include/linux/frame.h +29 -1
@@ -28,10 +28,38 @@ asm (".pushsection .discard.nonstd_frame".byte 0\n\t" ".popsection\n\t"); +/* +* This macro marks functions as UACCESS-safe, that is, it is safe to call from an +* UACCESS enabled region (typically user_access_begin() / +* user_access_end()). +* +* These functions in turn will only call UACCESS-safe functions themselves (which +* precludes tracing, including __fentry__ and scheduling, including +* preempt_enable). +* +* UACCESS-safe functions will obviously also not change UACCESS themselves. +*/ +#define UACCESS_SAFE (func) \ + asm (".pushsection .discard.uaccess_safe_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.uaccess_safe\n\t" \ + ".long 999b - .\n\t" \ + ".popsection") + +/* +* SHT_STRTAB(@3) sections should start with a \0 byte such that the 0 offset +* encodes the NULL string. +*/ +asm(".pushsection .discard.uaccess_safe_strtab, \"S\", @3\n\t" +".byte 0\n\t" +".popsection\n\t" ); +#else /* !CONFIG_STACK_VALIDATION */ #define STACK_FRAME_NON_STANDARD (func) +#define AC_SAFE (func)#endif /* CONFIG_STACK_VALIDATION */ -#endif /* _LINUX_FRAME_H */
tools/objtool/arch.h +3 -1
@@ -33,7 +33,9 @@#define INSN_STACK 8#define INSN_BUG 9#define INSN_NOP 10 -#define INSN_OTHER 11 +#define INSN_STAC 11 +#define INSN_CLAC 12 +#define INSN_OTHER 13#define INSN_LAST INSN_OTHERenum op_dest_type {
tools/objtool/arch/x86/decode.c +13 -1
@@ -369,7 +369,19 @@ int arch_decode_instruction(struct elf *case 0x0f : -if (op2 >=0x80 && op2 <=0x8f ) { +if (op2 ==0x01 ) { + +if (modrm ==0xca ) { + + *type =INSN_CLAC ; + + }else if (modrm ==0xcb ) { + + *type =INSN_STAC ; + + } + + }else if (op2 >=0x80 && op2 <=0x8f ) { *type =INSN_JUMP_CONDITIONAL ;
tools/objtool/check.c +91 -9
144 lines not on this page · raw message
tools/objtool/check.h +1 -1
@@ -31,7 +31,7 @@ struct insn_state {int stack_size;unsigned char type;bool bp_scratch; -bool drap, end; +bool drap, end, uaccess, uaccess_safe;int drap_reg, drap_offset;struct cfi_reg vals[CFI_NUM_REGS ]; };
tools/objtool/elf.h +1 -0
@@ -62,6 +62,7 @@ struct symbol {unsigned long offset;unsigned int len;struct symbol *pfunc, *cfunc; +bool uaccess_safe; };struct rela {
Re: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 45 lines · Peter Zijlstra
It is important that UACCESS regions are as small as possible; furthermore the UACCESS state is not scheduled, so doing anything that might directly call into the scheduler will cause random code to be ran with UACCESS enabled.
Teach objtool too track UACCESS state and warn about any CALL made while UACCESS is enabled. This very much includes the
__fentry__()tracing calls and__preempt_schedule()calls.Note that exceptions _do_ save/restore the UACCESS state, and therefore they can drive preemption. This also means that all exception handlers must have an otherwise dedundant UACCESS disable instruction; therefore ignore this warning for !STT_FUNC code (exception handlers are not normal functions).
It also provides a
UACCESS_SAFE()annotation which allows explicit annotation. This is meant to be used for future things like:
unsafe_copy_{to,from}_user().Signed-off-by Peter Zijlstra (Intel) <[email protected]>--- include/linux/frame.h | 30 +++++++++++- tools/objtool/arch.h | 4 + tools/objtool/arch/x86/decode.c | 14 +++++ tools/objtool/check.c | 100 ++++++++++++++++++++++++++++++++++++---- tools/objtool/check.h | 2 tools/objtool/elf.h | 1 6 files changed, 138 insertions(+), 13 deletions(-)include/linux/frame.h +9
-0@@ -28,10 +28,38 @@ asm (".pushsection .discard.nonstd_frame".byte 0\n\t" ".popsection\n\t"); +/* +* This macro marks functions as UACCESS-safe, that is, it is safe to call from an +* UACCESS enabled region (typically user_access_begin() / +* user_access_end()). +* +* These functions in turn will only call UACCESS-safe functions themselves (which +* precludes tracing, including __fentry__ and scheduling, including +* preempt_enable). +*
include/linux/frame.h asm (".pushsection .discard.nonstd_frame · line 49
Quoted 10 lines
include/linux/frame.h +10
-0+* UACCESS-safe functions will obviously also not change UACCESS themselves. +*/ +#define UACCESS_SAFE (func) \ + asm (".pushsection .discard.uaccess_safe_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.uaccess_safe\n\t" \ + ".long 999b - .\n\t" \ + ".popsection")
Minor nit: using big numbers like 999: like this always bugs me. It relies on there not being a macro nested inside or outside this that uses the same number. My general preference is to do something like .Ldescription_\@ instead.
Otherwise this looks conceptually good :)
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 11 lines · Andy Lutomirski
include/linux/frame.h asm (".pushsection .discard.nonstd_frame · line 49
Quoted 8 lines · Peter Zijlstra
include/linux/frame.h +8
-0+#define UACCESS_SAFE (func) \ + asm (".pushsection .discard.uaccess_safe_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.uaccess_safe\n\t" \ + ".long 999b - .\n\t" \ + ".popsection")
Quoted 6 lines
Minor nit: using big numbers like 999: like this always bugs me. It relies on there not being a macro nested inside or outside this that uses the same number. My general preference is to do something like .Ldescription_\@ instead.
Otherwise this looks conceptually good :)
I seem to remember here being an issue with the \@ thing. Notably we're not using it in nospsec-branch.h.
I'll see if I can dig up why we decided not to use it there.
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 17 lines · Peter Zijlstra
Quoted 16 lines · Andy Lutomirski
include/linux/frame.h asm (".pushsection .discard.nonstd_frame · line 49
Quoted 8 lines · Peter Zijlstra
include/linux/frame.h +8
-0+#define UACCESS_SAFE (func) \ + asm (".pushsection .discard.uaccess_safe_strtab, \"S\", @3\n\t" \ + "999: .ascii \"" #func "\"\n\t" \ + " .byte 0\n\t" \ + ".popsection\n\t" \ + ".pushsection .discard.uaccess_safe\n\t" \ + ".long 999b - .\n\t" \ + ".popsection")Minor nit: using big numbers like 999: like this always bugs me. It relies on there not being a macro nested inside or outside this that uses the same number. My general preference is to do something like .Ldescription_\@ instead.
Quoted 6 lines
Quoted 1 line
Otherwise this looks conceptually good :)
I seem to remember here being an issue with the \@ thing. Notably we're not using it in
nospsec-branch.h.I'll see if I can dig up why we decided not to use it there.
Ah, so the thing we need for inline-asm is %=, but Linus didn't like it:
https://lkml.kernel.org/r/[email protected]
Also, I think at the time we didn't use it because backporting efforts.
I can't really find what happened between:
https://lkml.kernel.org/r/[email protected]
and
https://lkml.kernel.org/r/[email protected]
But the %= thing went away.
Re: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 2 lines · Peter Zijlstra
Ah, so the thing we need for inline-asm is %=, but Linus didn't like it:
Well, to be fair, I also didn't like the random big numbers.
I think that inline asm that has local labels should just use simple enumeration, and always use the direction suffix to make those local labels be unambiguous.
Now, there are cases where we may not be able to do that, because we have internal macros within the asm that _also_ wants to do its own numbers. This isn't such a case.
For that case, you might actually want to use a fancy "get me a unique number and prefix", but it should be seen as the exception rather than the first solution, because the code starts looking really random.
Linus
Re: [PATCH 5/6] objtool: Add UACCESS validation
In reply to Peter ZijlstraQuoted 1 line · Peter Zijlstra
Also, I think at the time we didn't use it because backporting efforts.
This part may have been my fault. I spread a vicious rumor (based on reading docs instead of code) that old versions of GCC don't support '%='. But it's actually been around since GCC 2.8.
Signature · 1 line
Josh
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
In reply to Peter ZijlstraQuoted 15 lines · Peter Zijlstra
It is important that UACCESS regions are as small as possible; furthermore the UACCESS state is not scheduled, so doing anything that might directly call into the scheduler will cause random code to be ran with UACCESS enabled.
Teach objtool too track UACCESS state and warn about any CALL made while UACCESS is enabled. This very much includes the
__fentry__()tracing calls and__preempt_schedule()calls.Note that exceptions _do_ save/restore the UACCESS state, and therefore they can drive preemption. This also means that all exception handlers must have an otherwise dedundant UACCESS disable instruction; therefore ignore this warning for !STT_FUNC code (exception handlers are not normal functions).
Quoted 5 lines
It also provides a
UACCESS_SAFE()annotation which allows explicit annotation. This is meant to be used for future things like:
unsafe_copy_{to,from}_user().Signed-off-by Peter Zijlstra (Intel) <[email protected]>
So KASAN is wildly unhappy..
I can't actually find any definitions of those functions, so I can't very well mark the safe, even if we wanted to.
---
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x59: call to__asan_store8_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x6a: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x7b: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x8f: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0xa3: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0xb4: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0xc5: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0xdc: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0xf0: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x101: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x112: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x123: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x134: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x145: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x156: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x167: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x177: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x192: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x1bc: call to __asan_store8_noabort() with UACCESS enabled
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x1f2: call to__asan_load8_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x212: call to __asan_store8_noabort() with UACCESS enabled
Quoted 2 lines
Quoted 2 lines
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x1d8: call to force_valid_ss.isra.0() with UACCESS enabledarch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x3f: call to__asan_load8_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x50: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x61: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x75: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x89: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x9a: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0xab: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0xbc: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0xcd: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0xde: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0xef: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x100: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x111: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x122: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x133: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x143: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x157: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x173: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x18e: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x1a8: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x1c2: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x1ee: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: setup_sigcontext()+0x217: call to __asan_load8_noabort() with UACCESS enabled
Quoted 2 lines
Quoted 2 lines
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x618: call to__asan_load8_noabort()with UACCESS enabledarch/x86/kernel/signal.o: warning: objtool: do_signal()+0x630: call to__asan_loadN_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x681: call to __asan_load8_noabort() with UACCESS enabled
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x6a0: call to__asan_load4_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x6bd: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x7cb: call to __asan_load8_noabort() with UACCESS enabled
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x6dd: call to__asan_store8_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x6f0: call to __asan_store8_noabort() with UACCESS enabled
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x707: call to__asan_store4_noabort()with UACCESS enabled
arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x2c6: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x2e3: call to __asan_loadN_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x334: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x352: call to __asan_load4_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x374: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x49d: call to __asan_load8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x393: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x3aa: call to __asan_store8_noabort() with UACCESS enabled arch/x86/kernel/signal.o: warning: objtool: do_signal()+0x3c1: call to __asan_store4_noabort() with UACCESS enabled
Re: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 20 lines · Peter Zijlstra
Quoted 19 lines · Peter Zijlstra
It is important that UACCESS regions are as small as possible; furthermore the UACCESS state is not scheduled, so doing anything that might directly call into the scheduler will cause random code to be ran with UACCESS enabled.
Teach objtool too track UACCESS state and warn about any CALL made while UACCESS is enabled. This very much includes the
__fentry__()tracing calls and__preempt_schedule()calls.Note that exceptions _do_ save/restore the UACCESS state, and therefore they can drive preemption. This also means that all exception handlers must have an otherwise dedundant UACCESS disable instruction; therefore ignore this warning for !STT_FUNC code (exception handlers are not normal functions).
It also provides a
UACCESS_SAFE()annotation which allows explicit annotation. This is meant to be used for future things like:
unsafe_copy_{to,from}_user().
Quoted 7 lines
Quoted 1 line
Signed-off-by Peter Zijlstra (Intel) <[email protected]>So KASAN is wildly unhappy..
I can't actually find any definitions of those functions, so I can't very well mark the safe, even if we wanted to.
They are macro-generated. Use 'git grep DEFINE_ASAN'
Quoted 3 lines
---
Quoted 1 line
Quoted 1 line
arch/x86/kernel/signal.o: warning: objtool: restore_sigcontext()+0x59: call to__asan_store8_noabort()with UACCESS enabled
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 9 lines · Andrey Ryabinin
Quoted 3 lines · Peter Zijlstra
I can't actually find any definitions of those functions, so I can't very well mark the safe, even if we wanted to.
They are macro-generated. Use 'git grep
DEFINE_ASAN'
Ah, indeed! I'll go have a look.
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Andrey Ryabinin
On 2/27/19 5:08 PM, Peter Zijlstra wrote:
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
I can't actually find any definitions of those functions, so I can't very well mark the safe, even if we wanted to.
They are macro-generated. Use 'git grep
DEFINE_ASAN'Ah, indeed! I'll go have a look.
Urgh, kasan_report() is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.
Not sure what to do here; best I can come up with atm. is kill SMAP on KASAN builds.
Dmitry Vyukov
reviewerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 13 lines · Peter Zijlstra
Quoted 11 lines · Peter Zijlstra
Quoted 9 lines · Andrey Ryabinin
Quoted 3 lines · Peter Zijlstra
I can't actually find any definitions of those functions, so I can't very well mark the safe, even if we wanted to.
They are macro-generated. Use 'git grep
DEFINE_ASAN'
Quoted 7 lines
Quoted 1 line
Ah, indeed! I'll go have a look.
Urgh,
kasan_report()is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.Not sure what to do here; best I can come up with atm. is kill SMAP on KASAN builds.
If SMAP detects additional bugs, then it would be pity to disable it with KASAN (detect bugs in production but not during testing).
You mentioned that exception save/restore the UACCESS state. Is it possible to do the same in kasan_report? At the very least we need to survive report printing, what happens after that does not matter much (we've corrupted memory by now anyway).
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 5 lines · Dmitry Vyukov
Quoted 2 lines · Peter Zijlstra
Urgh,
kasan_report()is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.
Quoted 7 lines
If SMAP detects additional bugs, then it would be pity to disable it with KASAN (detect bugs in production but not during testing).
You mentioned that exception save/restore the UACCESS state. Is it possible to do the same in
kasan_report? At the very least we need to survive report printing, what happens after that does not matter much (we've corrupted memory by now anyway).
Ideally we'll put all of kasan_report() in an exception, much like we do for WARN. But there's a distinct lack of arch hooks there to play with. I suppose I can try and create some.
On top of that we'll have to mark these __asan functions with notrace.
Maybe a little something horrible like so... completely untested.
--- arch/x86/include/asm/bug.h | 28 ++++++++++++++-------------- arch/x86/include/asm/kasan.h | 11 +++++++++++ include/asm-generic/bug.h | 1 + include/linux/kasan.h | 6 +++++- lib/bug.c | 8 +++++++- mm/kasan/generic.c | 4 ++-- mm/kasan/kasan.h | 2 +- mm/kasan/report.c | 2 +- 8 files changed, 42 insertions(+), 20 deletions(-)
arch/x86/include/asm/bug.h +14 -14
@@ -30,33 +30,33 @@#ifdef CONFIG_DEBUG_BUGVERBOSE -#define _BUG_FLAGS (ins, flags) \ +#define _BUG_FLAGS (ins, flags, ...) \ do { \ asm volatile("1:\t" ins "\n" \ ".pushsection __bug_table,\"aw\"\n" \ - "2:\t" __BUG_REL(1b) "\t# bug_entry::bug_addr\n" \ - "\t" __BUG_REL(%c0) "\t# bug_entry::file\n" \ - "\t.word %c1" "\t# bug_entry::line\n" \ - "\t.word %c2" "\t# bug_entry::flags\n" \ - "\t.org 2b+%c3\n" \ + "2:\t" __BUG_REL(1b) "\t# bug_entry::bug_addr\n" \ + "\t" __BUG_REL(%c[file]) "\t# bug_entry::file\n" \ + "\t.word %c[line]" "\t# bug_entry::line\n" \ + "\t.word %c[flags]" "\t# bug_entry::flags\n" \ + "\t.org 2b+%c[size]\n" \ ".popsection" \ - : : "i" (__FILE__), "i" (__LINE__), \ - "i" (flags), \ - "i" (sizeof(struct bug_entry))); \ + : : [file] "i" (__FILE__), [line] "i" (__LINE__), \ + [flags] "i" (flags), \ + [size] "i" (sizeof(struct bug_entry)), ##__VA_ARGS__); \ } while (0)#else /* !CONFIG_DEBUG_BUGVERBOSE */ -#define _BUG_FLAGS (ins, flags) \ +#define _BUG_FLAGS (ins, flags, ...) \ do { \ asm volatile("1:\t" ins "\n" \ ".pushsection __bug_table,\"aw\"\n" \ "2:\t" __BUG_REL(1b) "\t# bug_entry::bug_addr\n" \ - "\t.word %c0" "\t# bug_entry::flags\n" \ - "\t.org 2b+%c1\n" \ + "\t.word %c[flags]" "\t# bug_entry::flags\n" \ + "\t.org 2b+%c[size]\n" \ ".popsection" \ - : : "i" (flags), \ - "i" (sizeof(struct bug_entry))); \ + : : [flags] "i" (flags), \ + [size] "i" (sizeof(struct bug_entry)), ##__VA_ARGS__); \ } while (0)#endif /* CONFIG_DEBUG_BUGVERBOSE */
arch/x86/include/asm/kasan.h +11 -0
@@ -2,6 +2,8 @@#ifndef _ASM_X86_KASAN_H#define _ASM_X86_KASAN_H +#include <asm/bug.h> +#include <linux/const.h> #define KASAN_SHADOW_OFFSET _AC(CONFIG_KASAN_SHADOW_OFFSET, UL)#define KASAN_SHADOW_SCALE_SHIFT 3
@@ -26,8 +28,17 @@#ifndef __ASSEMBLY__#ifdef CONFIG_KASAN +void __init kasan_early_init (void );void __init kasan_init (void ); + +static __always_inline void kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip) +{ +unsigned long rdi = addr, rsi = size, rdx = is_write, rdx = ip; + +_BUG_FLAGS (ASM_UD2 ,BUGFLAG_KASAN ,"D" (rdi),"S" (rsi),"d" (rdx),"c" (rcx)); +} +#else static inline void kasan_early_init (void ) { }static inline void kasan_init (void ) { }
include/asm-generic/bug.h +1 -0
@@ -10,6 +10,7 @@#define BUGFLAG_WARNING (1 << 0)#define BUGFLAG_ONCE (1 << 1)#define BUGFLAG_DONE (1 << 2) +#define BUGFLAG_KASAN (1 << 3)#define BUGFLAG_TAINT (taint) ((taint) << 8)#define BUG_GET_TAINT (bug) ((bug)->flags >> 8)#endif
include/linux/kasan.h +5 -1
@@ -177,7 +177,7 @@ void kasan_init_tags(void);void *kasan_reset_tag (const void *addr); -void kasan_report (unsigned long addr,size_t size, +void __kasan_report (unsigned long addr,size_t size,bool is_write,unsigned long ip);#else /* CONFIG_KASAN_SW_TAGS */
@@ -191,4 +191,8 @@ static inline void *kasan_reset_tag(const void *addr)#endif /* CONFIG_KASAN_SW_TAGS */ +#ifndef kasan_report +#define kasan_report (addr, size, is_write, ip) __kasan_report(addr, size, is_write, ip) +#endif +#endif /* LINUX_KASAN_H */
lib/bug.c +7 -1
@@ -144,7 +144,7 @@ enum bug_trap_type report_bug(unsigned long bugaddr, struct pt_regs *regs) {struct bug_entry *bug;const char *file; -unsigned line, warning, once, done; +unsigned line, warning, once, done, kasan;if (!is_valid_bugaddr (bugaddr))return BUG_TRAP_TYPE_NONE ;
@@ -167,6 +167,7 @@ enum bug_trap_type report_bug(unsigned long bugaddr, struct pt_regs *regs) line = bug->line;#endif warning = (bug->flags &BUGFLAG_WARNING ) !=0 ; + kasan = (bug->flags &BUGFLAG_KASAN ) !=0 ; once = (bug->flags &BUGFLAG_ONCE ) !=0 ; done = (bug->flags &BUGFLAG_DONE ) !=0 ;
@@ -188,6 +189,11 @@ enum bug_trap_type report_bug(unsigned long bugaddr, struct pt_regs *regs)return BUG_TRAP_TYPE_WARN ; } +if (kasan) { +__kasan_report (regs->di, regs->si, regs->dx, regs->cx); +return BUG_TRAP_TYPE_WARN ; + } + printk(KERN_DEFAULT CUT_HERE);if (file)
mm/kasan/generic.c +2 -2
@@ -228,7 +228,7 @@ void __asan_unregister_globals(struct kasan_global *globals, size_t size)EXPORT_SYMBOL (__asan_unregister_globals);#define DEFINE_ASAN_LOAD_STORE (size) \ - void __asan_load##size(unsigned long addr) \ + notrace void __asan_load##size(unsigned long addr) \ { \ check_memory_region_inline(addr, size, false, _RET_IP_);\ } \
@@ -236,7 +236,7 @@ EXPORT_SYMBOL(__asan_unregister_globals); __alias(__asan_load##size) \ void __asan_load##size##_noabort(unsigned long); \ EXPORT_SYMBOL(__asan_load##size##_noabort); \ - void __asan_store##size(unsigned long addr) \ + notrace void __asan_store##size(unsigned long addr) \ { \ check_memory_region_inline(addr, size, true, _RET_IP_); \ } \
mm/kasan/kasan.h +1 -1
@@ -130,7 +130,7 @@ void check_memory_region(unsigned long addr, size_t size, bool write,void *find_first_bad_addr (void *addr,size_t size);const char *get_bug_type (struct kasan_access_info *info); -void kasan_report (unsigned long addr,size_t size, +void __kasan_report (unsigned long addr,size_t size,bool is_write,unsigned long ip);void kasan_report_invalid_free (void *object,unsigned long ip);
mm/kasan/report.c +1 -1
@@ -281,7 +281,7 @@ void kasan_report_invalid_free(void *object, unsigned long ip)end_report (&flags); } -void kasan_report (unsigned long addr,size_t size, +void __kasan_report (unsigned long addr,size_t size,bool is_write,unsigned long ip) {struct kasan_access_info info;
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 14 lines · Peter Zijlstra
Quoted 5 lines · Dmitry Vyukov
Quoted 2 lines · Peter Zijlstra
Urgh,
kasan_report()is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.Quoted 7 lines
If SMAP detects additional bugs, then it would be pity to disable it with KASAN (detect bugs in production but not during testing).
You mentioned that exception save/restore the UACCESS state. Is it possible to do the same in
kasan_report? At the very least we need to survive report printing, what happens after that does not matter much (we've corrupted memory by now anyway).
Quoted 7 lines
Ideally we'll put all of
kasan_report()in an exception, much like we do for WARN. But there's a distinct lack of arch hooks there to play with. I suppose I can try and create some.On top of that we'll have to mark these
__asanfunctions with notrace.Maybe a little something horrible like so... completely untested.
OK, I got that to compile; the next problem is:
../include/linux/kasan.h:90:1: error: built-in function ‘__asan_loadN_noabort’ must be directly called UACCESS_SAFE(__asan_loadN_noabort);
Which doesn't make any sense; since we actually generated that symbol, it clearly is not built-in. What gives?
Dmitry Vyukov
reviewerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 24 lines · Peter Zijlstra
Quoted 14 lines · Peter Zijlstra
Quoted 5 lines · Dmitry Vyukov
Quoted 2 lines · Peter Zijlstra
Urgh,
kasan_report()is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.Quoted 7 lines
If SMAP detects additional bugs, then it would be pity to disable it with KASAN (detect bugs in production but not during testing).
You mentioned that exception save/restore the UACCESS state. Is it possible to do the same in
kasan_report? At the very least we need to survive report printing, what happens after that does not matter much (we've corrupted memory by now anyway).Quoted 7 lines
Ideally we'll put all of
kasan_report()in an exception, much like we do for WARN. But there's a distinct lack of arch hooks there to play with. I suppose I can try and create some.On top of that we'll have to mark these
__asanfunctions with notrace.Maybe a little something horrible like so... completely untested.
Quoted 7 lines
OK, I got that to compile; the next problem is:
../include/linux/kasan.h:90:1: error: built-in function ‘__asan_loadN_noabort’ must be directly calledUACCESS_SAFE(__asan_loadN_noabort);Which doesn't make any sense; since we actually generated that symbol, it clearly is not built-in. What gives?
I guess this warning originated for user-space where programmer does not define them and does not generally know about them and signature is not a public contract for user. And then for kernel it just stayed the same because not doing this warning would require somebody to proactively think about this potential difference and add an additional code to skip this check and even then it wasn't obvious why one will want to do this with these functions. So that's where we are now.
Dmitry Vyukov
reviewerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 34 lines · Dmitry Vyukov
Quoted 24 lines · Peter Zijlstra
Quoted 14 lines · Peter Zijlstra
Quoted 5 lines · Dmitry Vyukov
Quoted 2 lines · Peter Zijlstra
Urgh,
kasan_report()is definitely unsafe. Now, admitedly we should 'never' hit that, but it does leave us up a creek without a paddle.Quoted 7 lines
If SMAP detects additional bugs, then it would be pity to disable it with KASAN (detect bugs in production but not during testing).
You mentioned that exception save/restore the UACCESS state. Is it possible to do the same in
kasan_report? At the very least we need to survive report printing, what happens after that does not matter much (we've corrupted memory by now anyway).Quoted 7 lines
Ideally we'll put all of
kasan_report()in an exception, much like we do for WARN. But there's a distinct lack of arch hooks there to play with. I suppose I can try and create some.On top of that we'll have to mark these
__asanfunctions with notrace.Maybe a little something horrible like so... completely untested.
Quoted 7 lines
OK, I got that to compile; the next problem is:
../include/linux/kasan.h:90:1: error: built-in function ‘__asan_loadN_noabort’ must be directly calledUACCESS_SAFE(__asan_loadN_noabort);Which doesn't make any sense; since we actually generated that symbol, it clearly is not built-in. What gives?
Quoted 8 lines
I guess this warning originated for user-space where programmer does not define them and does not generally know about them and signature is not a public contract for user. And then for kernel it just stayed the same because not doing this warning would require somebody to proactively think about this potential difference and add an additional code to skip this check and even then it wasn't obvious why one will want to do this with these functions. So that's where we are now.
Maybe asm directive will help to trick the compiler?
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 11 lines · Dmitry Vyukov
Quoted 8 lines · Dmitry Vyukov
I guess this warning originated for user-space where programmer does not define them and does not generally know about them and signature is not a public contract for user. And then for kernel it just stayed the same because not doing this warning would require somebody to proactively think about this potential difference and add an additional code to skip this check and even then it wasn't obvious why one will want to do this with these functions. So that's where we are now.
Quoted 1 line
Maybe asm directive will help to trick the compiler?
So I went back and forth on the annotation; and we're back to the same we use for STACK_FRAME_NON_STANDARD() because that forces GCC to actually generate the symbol.
Without that GCC IPA will go and wreck things by either making the entire symbol go away or generating partial functions.
I'm currently doing a hard-coded list of names in objtool for this :/ But I'm having trouble with that alias crud.
Re: [PATCH 5/6] objtool: Add UACCESS validation
In reply to Peter ZijlstraQuoted 2 lines · Peter Zijlstra
So KASAN is wildly unhappy..
Yeah, well, with KASAN you definitely end up doing lots and lots for calls for just regular memory accesses.
Which we obviously need to do for most uaccess cases.
I think you should just say "ok, kasan reporting will possibly run with AC on".
Again, having AC on isn't fatal. It just makes the window where you can incorrectly access user space through a wild pointer bigger.
So the whole "run with AC on" thing isn't a big deal, apart from the scheduling case.
And we know how to fix the scheduling case by just doing the save/restore thing..
Linus
Peter Zijlstra
maintainerRe: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 5 lines · Linus Torvalds
So the whole "run with AC on" thing isn't a big deal, apart from the scheduling case.
And we know how to fix the scheduling case by just doing the save/restore thing..
Right, but I was hoping to avoid having to do that; and I think we can do that if we push all of kasan_report into an exception, like that 'patch' I just posted.
If I'm not mistaken the regular kasan house-keeping crud just prods at the shadow memory with simple code and should be perfectly safe.
Re: [PATCH 5/6] objtool: Add UACCESS validation
Quoted 7 lines · Peter Zijlstra
Right, but I was hoping to avoid having to do that; and I think we can do that if we push all of
kasan_reportinto an exception, like that 'patch' I just posted.If I'm not mistaken the regular kasan house-keeping crud just prods at the shadow memory with simple code and should be perfectly safe.
Fair enough. I do agree it would be nice to not have the save/restore simply because "it cannot matter" and we have the static tool to prove that.
Linus
Peter Zijlstra
maintainerPATCH 6/6 objtool: Add Direction Flag validation
In reply to Peter ZijlstraHaving DF escape is BAD(tm).
--- tools/objtool/arch.h | 4 +++- tools/objtool/arch/x86/decode.c | 8 ++++++++ tools/objtool/check.c | 21 +++++++++++++++++++++ tools/objtool/check.h | 2 +- 4 files changed, 33 insertions(+), 2 deletions(-)
tools/objtool/arch.h +3 -1
@@ -35,7 +35,9 @@#define INSN_NOP 10#define INSN_STAC 11#define INSN_CLAC 12 -#define INSN_OTHER 13 +#define INSN_STD 13 +#define INSN_CLD 14 +#define INSN_OTHER 15#define INSN_LAST INSN_OTHERenum op_dest_type {
tools/objtool/arch/x86/decode.c +8 -0
@@ -456,6 +456,14 @@ int arch_decode_instruction(struct elf * *type =INSN_CALL ;break ; +case 0xfc : + *type =INSN_CLD ; +break ; + +case 0xfd : + *type =INSN_STD ; +break ; +case 0xff :if (modrm_reg ==2 || modrm_reg ==3 )
tools/objtool/check.c +21 -0
@@ -1950,6 +1950,9 @@ static int validate_branch(struct objtooif (state.uaccess)WARN_FUNC ("return with UACCESS enabled" , sec, insn->offset); +if (state.df) +WARN_FUNC ("return with DF set" , sec, insn->offset); +if (func &&has_modified_stack_frame (&state)) {WARN_FUNC ("return with modified stack frame" , sec, insn->offset);
@@ -1971,6 +1974,10 @@ static int validate_branch(struct objtooWARN_FUNC ("call to %s() with UACCESS enabled" , sec, insn->offset,insn_dest_name (insn)); } +if (state.df) { +WARN_FUNC ("call to %s() with DF set" , + sec, insn->offset,insn_dest_name (insn)); + }if (insn->type ==INSN_CALL ) {if (is_fentry_call (insn))
@@ -2054,6 +2061,20 @@ static int validate_branch(struct objtoo state.uaccess = false;break ; +case INSN_STD : +if (state.df) +WARN_FUNC ("recursive STD" , sec, insn->offset); + + state.df = true; +break ; + +case INSN_CLD : +if (!state.df && insn->func) +WARN_FUNC ("redundant CLD" , sec, insn->offset); + + state.df = false; +break ; +default :break ; }
tools/objtool/check.h +1 -1
@@ -31,7 +31,7 @@ struct insn_state {int stack_size;unsigned char type;bool bp_scratch; -bool drap, end, uaccess, uaccess_safe; +bool drap, end, uaccess, uaccess_safe, df;int drap_reg, drap_offset;struct cfi_reg vals[CFI_NUM_REGS ]; };