robust futex thread exit race
PATCH 1/1 robust futex thread exit race
Hi Ingo,
I finally found the bug that causes tst-robust8 from the glibc to fail on s390x. Turned out to be a common code problem with the processing of the robust futex list. The patch below fixes the bug for me.
Signature · 83 lines
blue skies, Martin. "Reality continues to ruin my life." - Calvin. -- Subject: [PATCH] robust futex thread exit race From: Martin Schwidefsky <[email protected]> Calling handle_futex_death in exit_robust_list for the different robust mutexes of a thread basically frees the mutex. Another thread might grab the lock immediately which updates the next pointer of the mutex. fetch_robust_entry over the next pointer might therefore branch into the robust mutex list of a different thread. This can cause two problems: 1) some mutexes held by the dead thread are not getting freed and 2) some mutexs held by a different thread are freed. The next point need to be read before calling handle_futex_death. Signed-off-by: Martin Schwidefsky <[email protected]> --- diff -urpN linux-2.6/kernel/futex.c linux-2.6-patched/kernel/futex.c --- linux-2.6/kernel/futex.c 2007-08-23 11:14:33.000000000 +0200 +++ linux-2.6-patched/kernel/futex.c 2007-09-30 16:31:57.000000000 +0200 @@ -1943,9 +1943,10 @@ static inline int fetch_robust_entry(str void exit_robust_list(struct task_struct *curr) { struct robust_list_head __user *head = curr->robust_list; - struct robust_list __user *entry, *pending; - unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; + struct robust_list __user *entry, *next_entry, *pending; + unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; unsigned long futex_offset; + int rc; /* * Fetch the list head (which was registered earlier, via @@ -1965,12 +1966,13 @@ void exit_robust_list(struct task_struct if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) return; - if (pending) - handle_futex_death((void __user *)pending + futex_offset, - curr, pip); - while (entry != &head->list) { /* + * Fetch the next entry in the list before calling + * handle_futex_death: + */ + rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); + /* * A pending lock might already be on the list, so * don't process it twice: */ @@ -1978,11 +1980,10 @@ void exit_robust_list(struct task_struct if (handle_futex_death((void __user *)entry + futex_offset, curr, pi)) return; - /* - * Fetch the next entry in the list: - */ - if (fetch_robust_entry(&entry, &entry->next, &pi)) + if (rc) return; + entry = next_entry; + pi = next_pi; /* * Avoid excessively long or circular lists: */ @@ -1991,6 +1992,10 @@ void exit_robust_list(struct task_struct cond_resched(); } + + if (pending) + handle_futex_death((void __user *)pending + futex_offset, + curr, pip); } long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
Quoted 4 lines · Martin Schwidefsky
Hi Ingo,
I finally found the bug that causes tst-robust8 from the glibc to fail on s390x. Turned out to be a common code problem with the processing of the robust futex list. The patch below fixes the bug for me.
good catch! A quick preliminary review of your patch indicates it's fine
- and it might be v2.6.23 material.
Quoted 8 lines
Calling
handle_futex_deathinexit_robust_listfor the different robust mutexes of a thread basically frees the mutex. Another thread might grab the lock immediately which updates the next pointer of the mutex.fetch_robust_entryover the next pointer might therefore branch into the robust mutex list of a different thread. This can cause two
problems: 1) some mutexes held by the dead thread are not getting freed and 2) some mutexs held by a different thread are freed. The next point need to be read before callinghandle_futex_death.
nasty race... Ulrich, Thomas, do you concur?
Ingo
Thomas Gleixner
Acked-byQuoted 11 lines · Ingo Molnar
Quoted 4 lines · Martin Schwidefsky
Hi Ingo,
I finally found the bug that causes tst-robust8 from the glibc to fail on s390x. Turned out to be a common code problem with the processing of the robust futex list. The patch below fixes the bug for me.good catch! A quick preliminary review of your patch indicates it's fine
- and it might be v2.6.23 material.Acked-by Ingo Molnar <[email protected]>
Quoted 10 lines
Quoted 8 lines
Calling
handle_futex_deathinexit_robust_listfor the different robust mutexes of a thread basically frees the mutex. Another thread might grab the lock immediately which updates the next pointer of the mutex.fetch_robust_entryover the next pointer might therefore branch into the robust mutex list of a different thread. This can cause two
problems: 1) some mutexes held by the dead thread are not getting freed and 2) some mutexs held by a different thread are freed. The next point need to be read before callinghandle_futex_death.nasty race... Ulrich, Thomas, do you concur?
Yes. Where do they sell those brown paperbags again ?
tglx
Quoted 25 lines · Thomas Gleixner
Quoted 11 lines · Ingo Molnar
Quoted 4 lines · Martin Schwidefsky
Hi Ingo,
I finally found the bug that causes tst-robust8 from the glibc to fail on s390x. Turned out to be a common code problem with the processing of the robust futex list. The patch below fixes the bug for me.good catch! A quick preliminary review of your patch indicates it's fine
- and it might be v2.6.23 material.Acked-by Ingo Molnar <[email protected]>Acked-by Thomas Gleixner <[email protected]>Quoted 9 lines
Quoted 8 lines
Calling
handle_futex_deathinexit_robust_listfor the different robust mutexes of a thread basically frees the mutex. Another thread might grab the lock immediately which updates the next pointer of the mutex.fetch_robust_entryover the next pointer might therefore branch into the robust mutex list of a different thread. This can cause two
problems: 1) some mutexes held by the dead thread are not getting freed and 2) some mutexs held by a different thread are freed. The next point need to be read before callinghandle_futex_death.
Quoted 3 lines
Quoted 1 line
nasty race... Ulrich, Thomas, do you concur?
Yes. Where do they sell those brown paperbags again ?
i've still got plenty of them stacked up, can pass over a few. (i'm getting them cheap from the manufacturer - large quantity discount)
Ingo
Quoted 9 lines · Ingo Molnar
Quoted 4 lines · Martin Schwidefsky
Hi Ingo,
I finally found the bug that causes tst-robust8 from the glibc to fail on s390x. Turned out to be a common code problem with the processing of the robust futex list. The patch below fixes the bug for me.good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in kernel/compat_futex.c too.
Ingo
Quoted 4 lines · Ingo Molnar
Quoted 2 lines
good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in
kernel/compat_futex.ctoo.
True. I'll update the patch and send it again with the Acked-bys.
Signature · 4 lines
blue skies, Martin. "Reality continues to ruin my life." - Calvin.
Quoted 8 lines · Martin Schwidefsky
Quoted 4 lines · Ingo Molnar
Quoted 2 lines
good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in
kernel/compat_futex.ctoo.True. I'll update the patch and send it again with the Acked-bys.
thanks! Logistics-wise: given that v2.6.22 has this same bug too, it's not a .23 regression per se and could in theory be merged after the 2.6.23 release too, into 2.6.23.1.
Ingo
Quoted 10 lines · Ingo Molnar
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in
kernel/compat_futex.ctoo.True. I'll update the patch and send it again with the Acked-bys.
thanks! Logistics-wise: given that v2.6.22 has this same bug too, it's not a .23 regression per se and could in theory be merged after the 2.6.23 release too, into 2.6.23.1.
Here we go. Tested with current git and latest glibc tst-robust8 on s390-64, native and compat.
Signature · 154 lines
blue skies, Martin. "Reality continues to ruin my life." - Calvin. -- Subject: [PATCH] robust futex thread exit race From: Martin Schwidefsky <[email protected]> Calling handle_futex_death in exit_robust_list for the different robust mutexes of a thread basically frees the mutex. Another thread might grab the lock immediately which updates the next pointer of the mutex. fetch_robust_entry over the next pointer might therefore branch into the robust mutex list of a different thread. This can cause two problems: 1) some mutexes held by the dead thread are not getting freed and 2) some mutexs held by a different thread are freed. The next pointer need to be read before calling fetch_robust_entry. Signed-off-by: Martin Schwidefsky <[email protected]> Acked-by: Ingo Molnar <[email protected]> Acked-by: Thomas Gleixner <[email protected]> --- kernel/futex.c | 26 ++++++++++++++++---------- kernel/futex_compat.c | 28 ++++++++++++++++++---------- 2 files changed, 34 insertions(+), 20 deletions(-) diff -urpN linux-2.6/kernel/futex.c linux-2.6-patched/kernel/futex.c --- linux-2.6/kernel/futex.c 2007-08-23 11:14:33.000000000 +0200 +++ linux-2.6-patched/kernel/futex.c 2007-09-30 19:23:12.000000000 +0200 @@ -1943,9 +1943,10 @@ static inline int fetch_robust_entry(str void exit_robust_list(struct task_struct *curr) { struct robust_list_head __user *head = curr->robust_list; - struct robust_list __user *entry, *pending; - unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; + struct robust_list __user *entry, *next_entry, *pending; + unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; unsigned long futex_offset; + int rc; /* * Fetch the list head (which was registered earlier, via @@ -1965,12 +1966,14 @@ void exit_robust_list(struct task_struct if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) return; - if (pending) - handle_futex_death((void __user *)pending + futex_offset, - curr, pip); - + next_entry = NULL; /* avoid warning with gcc */ while (entry != &head->list) { /* + * Fetch the next entry in the list before calling + * handle_futex_death: + */ + rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); + /* * A pending lock might already be on the list, so * don't process it twice: */ @@ -1978,11 +1981,10 @@ void exit_robust_list(struct task_struct if (handle_futex_death((void __user *)entry + futex_offset, curr, pi)) return; - /* - * Fetch the next entry in the list: - */ - if (fetch_robust_entry(&entry, &entry->next, &pi)) + if (rc) return; + entry = next_entry; + pi = next_pi; /* * Avoid excessively long or circular lists: */ @@ -1991,6 +1993,10 @@ void exit_robust_list(struct task_struct cond_resched(); } + + if (pending) + handle_futex_death((void __user *)pending + futex_offset, + curr, pip); } long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, diff -urpN linux-2.6/kernel/futex_compat.c linux-2.6-patched/kernel/futex_compat.c --- linux-2.6/kernel/futex_compat.c 2007-09-12 10:29:09.000000000 +0200 +++ linux-2.6-patched/kernel/futex_compat.c 2007-09-30 18:29:16.000000000 +0200 @@ -38,10 +38,11 @@ fetch_robust_entry(compat_uptr_t *uentry void compat_exit_robust_list(struct task_struct *curr) { struct compat_robust_list_head __user *head = curr->compat_robust_list; - struct robust_list __user *entry, *pending; - unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; - compat_uptr_t uentry, upending; + struct robust_list __user *entry, *next_entry, *pending; + unsigned int limit = ROBUST_LIST_LIMIT, pi, next_pi, pip; + compat_uptr_t uentry, next_uentry, upending; compat_long_t futex_offset; + int rc; /* * Fetch the list head (which was registered earlier, via @@ -61,11 +62,16 @@ void compat_exit_robust_list(struct task if (fetch_robust_entry(&upending, &pending, &head->list_op_pending, &pip)) return; - if (pending) - handle_futex_death((void __user *)pending + futex_offset, curr, pip); + next_entry = NULL; /* avoid warning with gcc */ while (entry != (struct robust_list __user *) &head->list) { /* + * Fetch the next entry in the list before calling + * handle_futex_death: + */ + rc = fetch_robust_entry(&next_uentry, &next_entry, + (compat_uptr_t __user *)&entry->next, &next_pi); + /* * A pending lock might already be on the list, so * dont process it twice: */ @@ -74,12 +80,11 @@ void compat_exit_robust_list(struct task curr, pi)) return; - /* - * Fetch the next entry in the list: - */ - if (fetch_robust_entry(&uentry, &entry, - (compat_uptr_t __user *)&entry->next, &pi)) + if (rc) return; + uentry = next_uentry; + entry = next_entry; + pi = next_pi; /* * Avoid excessively long or circular lists: */ @@ -88,6 +93,9 @@ void compat_exit_robust_list(struct task cond_resched(); } + if (pending) + handle_futex_death((void __user *)pending + futex_offset, + curr, pip); } asmlinkage long
Quoted 6 lines · Martin Schwidefsky
Quoted 5 lines · Ingo Molnar
Quoted 5 lines
Quoted 4 lines
Quoted 2 lines
good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in
kernel/compat_futex.ctoo.
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
True. I'll update the patch and send it again with the Acked-bys.
thanks! Logistics-wise: given that v2.6.22 has this same bug too, it's not a .23 regression per se and could in theory be merged after the 2.6.23 release too, into 2.6.23.1.
Here we go. Tested with current git and latest glibc tst-robust8 on s390-64, native and compat.
excellent! This patch too looks good to me. I booted and tested it on x86 too (32-bit-native, 64-bit-native and 32-bit-compat-on-64-bit.)
Ingo
From: Ingo Molnar <[email protected]>
Date: Sun, 30 Sep 2007 21:55:38 +0200
Quoted 15 lines · Martin Schwidefsky
Quoted 13 lines · Ingo Molnar
Quoted 10 lines
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
good catch! A quick preliminary review of your patch indicates it's fine - and it might be v2.6.23 material.
there's a symmetric bug in
kernel/compat_futex.ctoo.True. I'll update the patch and send it again with the Acked-bys.
thanks! Logistics-wise: given that v2.6.22 has this same bug too, it's not a .23 regression per se and could in theory be merged after the 2.6.23 release too, into 2.6.23.1.
Quoted 5 lines
Quoted 2 lines
Here we go. Tested with current git and latest glibc tst-robust8 on s390-64, native and compat.
excellent! This patch too looks good to me. I booted and tested it on x86 too (32-bit-native, 64-bit-native and 32-bit-compat-on-64-bit.)
I would personally prefer it if this patch got scheduled into 2.6.23, it is a bug fix and it might even account for some of the futex weirdness we're still seeing on sparc64.
I'll toss this patch to some folks seeing those issues for testing.
But even if it doesn't fix their bug, it still belongs in 2.6.23 and there is no reason to delay this fix.