From 84ff7a09c371bc7417eabfda19bf7f113ec917b6 Mon Sep 17 00:00:00 2001 From: Will Deacon Date: Mon, 8 Apr 2019 12:45:09 +0100 Subject: [PATCH 1/5] arm64: futex: Fix FUTEX_WAKE_OP atomic ops with non-zero result value Rather embarrassingly, our futex() FUTEX_WAKE_OP implementation doesn't explicitly set the return value on the non-faulting path and instead leaves it holding the result of the underlying atomic operation. This means that any FUTEX_WAKE_OP atomic operation which computes a non-zero value will be reported as having failed. Regrettably, I wrote the buggy code back in 2011 and it was upstreamed as part of the initial arm64 support in 2012. The reasons we appear to get away with this are: 1. FUTEX_WAKE_OP is rarely used and therefore doesn't appear to get exercised by futex() test applications 2. If the result of the atomic operation is zero, the system call behaves correctly 3. Prior to version 2.25, the only operation used by GLIBC set the futex to zero, and therefore worked as expected. From 2.25 onwards, FUTEX_WAKE_OP is not used by GLIBC at all. Fix the implementation by ensuring that the return value is either 0 to indicate that the atomic operation completed successfully, or -EFAULT if we encountered a fault when accessing the user mapping. Cc: Fixes: 6170a97460db ("arm64: Atomic operations") Signed-off-by: Will Deacon --- arch/arm64/include/asm/futex.h | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/arch/arm64/include/asm/futex.h b/arch/arm64/include/asm/futex.h index cccb83ad7fa8..e1d95f08f8e1 100644 --- a/arch/arm64/include/asm/futex.h +++ b/arch/arm64/include/asm/futex.h @@ -30,8 +30,8 @@ do { \ " prfm pstl1strm, %2\n" \ "1: ldxr %w1, %2\n" \ insn "\n" \ -"2: stlxr %w3, %w0, %2\n" \ -" cbnz %w3, 1b\n" \ +"2: stlxr %w0, %w3, %2\n" \ +" cbnz %w0, 1b\n" \ " dmb ish\n" \ "3:\n" \ " .pushsection .fixup,\"ax\"\n" \ @@ -50,30 +50,30 @@ do { \ static inline int arch_futex_atomic_op_inuser(int op, int oparg, int *oval, u32 __user *_uaddr) { - int oldval = 0, ret, tmp; + int oldval, ret, tmp; u32 __user *uaddr = __uaccess_mask_ptr(_uaddr); pagefault_disable(); switch (op) { case FUTEX_OP_SET: - __futex_atomic_op("mov %w0, %w4", + __futex_atomic_op("mov %w3, %w4", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_ADD: - __futex_atomic_op("add %w0, %w1, %w4", + __futex_atomic_op("add %w3, %w1, %w4", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_OR: - __futex_atomic_op("orr %w0, %w1, %w4", + __futex_atomic_op("orr %w3, %w1, %w4", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_ANDN: - __futex_atomic_op("and %w0, %w1, %w4", + __futex_atomic_op("and %w3, %w1, %w4", ret, oldval, uaddr, tmp, ~oparg); break; case FUTEX_OP_XOR: - __futex_atomic_op("eor %w0, %w1, %w4", + __futex_atomic_op("eor %w3, %w1, %w4", ret, oldval, uaddr, tmp, oparg); break; default: From 6b4f4bc9cb22875f97023984a625386f0c7cc1c0 Mon Sep 17 00:00:00 2001 From: Will Deacon Date: Thu, 28 Feb 2019 11:58:08 +0000 Subject: [PATCH 2/5] locking/futex: Allow low-level atomic operations to return -EAGAIN Some futex() operations, including FUTEX_WAKE_OP, require the kernel to perform an atomic read-modify-write of the futex word via the userspace mapping. These operations are implemented by each architecture in arch_futex_atomic_op_inuser() and futex_atomic_cmpxchg_inatomic(), which are called in atomic context with the relevant hash bucket locks held. Although these routines may return -EFAULT in response to a page fault generated when accessing userspace, they are expected to succeed (i.e. return 0) in all other cases. This poses a problem for architectures that do not provide bounded forward progress guarantees or fairness of contended atomic operations and can lead to starvation in some cases. In these problematic scenarios, we must return back to the core futex code so that we can drop the hash bucket locks and reschedule if necessary, much like we do in the case of a page fault. Allow architectures to return -EAGAIN from their implementations of arch_futex_atomic_op_inuser() and futex_atomic_cmpxchg_inatomic(), which will cause the core futex code to reschedule if necessary and return back to the architecture code later on. Cc: Acked-by: Peter Zijlstra (Intel) Signed-off-by: Will Deacon --- kernel/futex.c | 188 ++++++++++++++++++++++++++++++------------------- 1 file changed, 117 insertions(+), 71 deletions(-) diff --git a/kernel/futex.c b/kernel/futex.c index 9e40cf7be606..6262f1534ac9 100644 --- a/kernel/futex.c +++ b/kernel/futex.c @@ -1311,13 +1311,15 @@ static int lookup_pi_state(u32 __user *uaddr, u32 uval, static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) { + int err; u32 uninitialized_var(curval); if (unlikely(should_fail_futex(true))) return -EFAULT; - if (unlikely(cmpxchg_futex_value_locked(&curval, uaddr, uval, newval))) - return -EFAULT; + err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); + if (unlikely(err)) + return err; /* If user space value changed, let the caller retry */ return curval != uval ? -EAGAIN : 0; @@ -1502,10 +1504,8 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_ if (unlikely(should_fail_futex(true))) ret = -EFAULT; - if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) { - ret = -EFAULT; - - } else if (curval != uval) { + ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); + if (!ret && (curval != uval)) { /* * If a unconditional UNLOCK_PI operation (user space did not * try the TID->0 transition) raced with a waiter setting the @@ -1700,32 +1700,32 @@ retry_private: double_lock_hb(hb1, hb2); op_ret = futex_atomic_op_inuser(op, uaddr2); if (unlikely(op_ret < 0)) { - double_unlock_hb(hb1, hb2); -#ifndef CONFIG_MMU - /* - * we don't get EFAULT from MMU faults if we don't have an MMU, - * but we might get them from range checking - */ - ret = op_ret; - goto out_put_keys; -#endif - - if (unlikely(op_ret != -EFAULT)) { + if (!IS_ENABLED(CONFIG_MMU) || + unlikely(op_ret != -EFAULT && op_ret != -EAGAIN)) { + /* + * we don't get EFAULT from MMU faults if we don't have + * an MMU, but we might get them from range checking + */ ret = op_ret; goto out_put_keys; } - ret = fault_in_user_writeable(uaddr2); - if (ret) - goto out_put_keys; + if (op_ret == -EFAULT) { + ret = fault_in_user_writeable(uaddr2); + if (ret) + goto out_put_keys; + } - if (!(flags & FLAGS_SHARED)) + if (!(flags & FLAGS_SHARED)) { + cond_resched(); goto retry_private; + } put_futex_key(&key2); put_futex_key(&key1); + cond_resched(); goto retry; } @@ -2350,7 +2350,7 @@ static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, u32 uval, uninitialized_var(curval), newval; struct task_struct *oldowner, *newowner; u32 newtid; - int ret; + int ret, err = 0; lockdep_assert_held(q->lock_ptr); @@ -2421,14 +2421,17 @@ retry: if (!pi_state->owner) newtid |= FUTEX_OWNER_DIED; - if (get_futex_value_locked(&uval, uaddr)) - goto handle_fault; + err = get_futex_value_locked(&uval, uaddr); + if (err) + goto handle_err; for (;;) { newval = (uval & FUTEX_OWNER_DIED) | newtid; - if (cmpxchg_futex_value_locked(&curval, uaddr, uval, newval)) - goto handle_fault; + err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); + if (err) + goto handle_err; + if (curval == uval) break; uval = curval; @@ -2456,23 +2459,37 @@ retry: return 0; /* - * To handle the page fault we need to drop the locks here. That gives - * the other task (either the highest priority waiter itself or the - * task which stole the rtmutex) the chance to try the fixup of the - * pi_state. So once we are back from handling the fault we need to - * check the pi_state after reacquiring the locks and before trying to - * do another fixup. When the fixup has been done already we simply - * return. + * In order to reschedule or handle a page fault, we need to drop the + * locks here. In the case of a fault, this gives the other task + * (either the highest priority waiter itself or the task which stole + * the rtmutex) the chance to try the fixup of the pi_state. So once we + * are back from handling the fault we need to check the pi_state after + * reacquiring the locks and before trying to do another fixup. When + * the fixup has been done already we simply return. * * Note: we hold both hb->lock and pi_mutex->wait_lock. We can safely * drop hb->lock since the caller owns the hb -> futex_q relation. * Dropping the pi_mutex->wait_lock requires the state revalidate. */ -handle_fault: +handle_err: raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); spin_unlock(q->lock_ptr); - ret = fault_in_user_writeable(uaddr); + switch (err) { + case -EFAULT: + ret = fault_in_user_writeable(uaddr); + break; + + case -EAGAIN: + cond_resched(); + ret = 0; + break; + + default: + WARN_ON_ONCE(1); + ret = err; + break; + } spin_lock(q->lock_ptr); raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); @@ -3041,10 +3058,8 @@ retry: * A unconditional UNLOCK_PI op raced against a waiter * setting the FUTEX_WAITERS bit. Try again. */ - if (ret == -EAGAIN) { - put_futex_key(&key); - goto retry; - } + if (ret == -EAGAIN) + goto pi_retry; /* * wake_futex_pi has detected invalid state. Tell user * space. @@ -3059,9 +3074,19 @@ retry: * preserve the WAITERS bit not the OWNER_DIED one. We are the * owner. */ - if (cmpxchg_futex_value_locked(&curval, uaddr, uval, 0)) { + if ((ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))) { spin_unlock(&hb->lock); - goto pi_faulted; + switch (ret) { + case -EFAULT: + goto pi_faulted; + + case -EAGAIN: + goto pi_retry; + + default: + WARN_ON_ONCE(1); + goto out_putkey; + } } /* @@ -3075,6 +3100,11 @@ out_putkey: put_futex_key(&key); return ret; +pi_retry: + put_futex_key(&key); + cond_resched(); + goto retry; + pi_faulted: put_futex_key(&key); @@ -3435,6 +3465,7 @@ err_unlock: static int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi) { u32 uval, uninitialized_var(nval), mval; + int err; /* Futex address must be 32bit aligned */ if ((((unsigned long)uaddr) % sizeof(*uaddr)) != 0) @@ -3444,42 +3475,57 @@ retry: if (get_user(uval, uaddr)) return -1; - if ((uval & FUTEX_TID_MASK) == task_pid_vnr(curr)) { - /* - * Ok, this dying thread is truly holding a futex - * of interest. Set the OWNER_DIED bit atomically - * via cmpxchg, and if the value had FUTEX_WAITERS - * set, wake up a waiter (if any). (We have to do a - * futex_wake() even if OWNER_DIED is already set - - * to handle the rare but possible case of recursive - * thread-death.) The rest of the cleanup is done in - * userspace. - */ - mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; - /* - * We are not holding a lock here, but we want to have - * the pagefault_disable/enable() protection because - * we want to handle the fault gracefully. If the - * access fails we try to fault in the futex with R/W - * verification via get_user_pages. get_user() above - * does not guarantee R/W access. If that fails we - * give up and leave the futex locked. - */ - if (cmpxchg_futex_value_locked(&nval, uaddr, uval, mval)) { + if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) + return 0; + + /* + * Ok, this dying thread is truly holding a futex + * of interest. Set the OWNER_DIED bit atomically + * via cmpxchg, and if the value had FUTEX_WAITERS + * set, wake up a waiter (if any). (We have to do a + * futex_wake() even if OWNER_DIED is already set - + * to handle the rare but possible case of recursive + * thread-death.) The rest of the cleanup is done in + * userspace. + */ + mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; + + /* + * We are not holding a lock here, but we want to have + * the pagefault_disable/enable() protection because + * we want to handle the fault gracefully. If the + * access fails we try to fault in the futex with R/W + * verification via get_user_pages. get_user() above + * does not guarantee R/W access. If that fails we + * give up and leave the futex locked. + */ + if ((err = cmpxchg_futex_value_locked(&nval, uaddr, uval, mval))) { + switch (err) { + case -EFAULT: if (fault_in_user_writeable(uaddr)) return -1; goto retry; - } - if (nval != uval) + + case -EAGAIN: + cond_resched(); goto retry; - /* - * Wake robust non-PI futexes here. The wakeup of - * PI futexes happens in exit_pi_state(): - */ - if (!pi && (uval & FUTEX_WAITERS)) - futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); + default: + WARN_ON_ONCE(1); + return err; + } } + + if (nval != uval) + goto retry; + + /* + * Wake robust non-PI futexes here. The wakeup of + * PI futexes happens in exit_pi_state(): + */ + if (!pi && (uval & FUTEX_WAITERS)) + futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); + return 0; } From 03110a5cb2161690ae5ac04994d47ed0cd6cef75 Mon Sep 17 00:00:00 2001 From: Will Deacon Date: Mon, 8 Apr 2019 14:23:17 +0100 Subject: [PATCH 3/5] arm64: futex: Bound number of LDXR/STXR loops in FUTEX_WAKE_OP Our futex implementation makes use of LDXR/STXR loops to perform atomic updates to user memory from atomic context. This can lead to latency problems if we end up spinning around the LL/SC sequence at the expense of doing something useful. Rework our futex atomic operations so that we return -EAGAIN if we fail to update the futex word after 128 attempts. The core futex code will reschedule if necessary and we'll try again later. Cc: Fixes: 6170a97460db ("arm64: Atomic operations") Signed-off-by: Will Deacon --- arch/arm64/include/asm/futex.h | 55 +++++++++++++++++++++------------- 1 file changed, 34 insertions(+), 21 deletions(-) diff --git a/arch/arm64/include/asm/futex.h b/arch/arm64/include/asm/futex.h index e1d95f08f8e1..2d78ea6932b7 100644 --- a/arch/arm64/include/asm/futex.h +++ b/arch/arm64/include/asm/futex.h @@ -23,26 +23,34 @@ #include +#define FUTEX_MAX_LOOPS 128 /* What's the largest number you can think of? */ + #define __futex_atomic_op(insn, ret, oldval, uaddr, tmp, oparg) \ do { \ + unsigned int loops = FUTEX_MAX_LOOPS; \ + \ uaccess_enable(); \ asm volatile( \ " prfm pstl1strm, %2\n" \ "1: ldxr %w1, %2\n" \ insn "\n" \ "2: stlxr %w0, %w3, %2\n" \ -" cbnz %w0, 1b\n" \ -" dmb ish\n" \ +" cbz %w0, 3f\n" \ +" sub %w4, %w4, %w0\n" \ +" cbnz %w4, 1b\n" \ +" mov %w0, %w7\n" \ "3:\n" \ +" dmb ish\n" \ " .pushsection .fixup,\"ax\"\n" \ " .align 2\n" \ -"4: mov %w0, %w5\n" \ +"4: mov %w0, %w6\n" \ " b 3b\n" \ " .popsection\n" \ _ASM_EXTABLE(1b, 4b) \ _ASM_EXTABLE(2b, 4b) \ - : "=&r" (ret), "=&r" (oldval), "+Q" (*uaddr), "=&r" (tmp) \ - : "r" (oparg), "Ir" (-EFAULT) \ + : "=&r" (ret), "=&r" (oldval), "+Q" (*uaddr), "=&r" (tmp), \ + "+r" (loops) \ + : "r" (oparg), "Ir" (-EFAULT), "Ir" (-EAGAIN) \ : "memory"); \ uaccess_disable(); \ } while (0) @@ -57,23 +65,23 @@ arch_futex_atomic_op_inuser(int op, int oparg, int *oval, u32 __user *_uaddr) switch (op) { case FUTEX_OP_SET: - __futex_atomic_op("mov %w3, %w4", + __futex_atomic_op("mov %w3, %w5", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_ADD: - __futex_atomic_op("add %w3, %w1, %w4", + __futex_atomic_op("add %w3, %w1, %w5", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_OR: - __futex_atomic_op("orr %w3, %w1, %w4", + __futex_atomic_op("orr %w3, %w1, %w5", ret, oldval, uaddr, tmp, oparg); break; case FUTEX_OP_ANDN: - __futex_atomic_op("and %w3, %w1, %w4", + __futex_atomic_op("and %w3, %w1, %w5", ret, oldval, uaddr, tmp, ~oparg); break; case FUTEX_OP_XOR: - __futex_atomic_op("eor %w3, %w1, %w4", + __futex_atomic_op("eor %w3, %w1, %w5", ret, oldval, uaddr, tmp, oparg); break; default: @@ -93,6 +101,7 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *_uaddr, u32 oldval, u32 newval) { int ret = 0; + unsigned int loops = FUTEX_MAX_LOOPS; u32 val, tmp; u32 __user *uaddr; @@ -104,20 +113,24 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *_uaddr, asm volatile("// futex_atomic_cmpxchg_inatomic\n" " prfm pstl1strm, %2\n" "1: ldxr %w1, %2\n" -" sub %w3, %w1, %w4\n" -" cbnz %w3, 3f\n" -"2: stlxr %w3, %w5, %2\n" -" cbnz %w3, 1b\n" -" dmb ish\n" +" sub %w3, %w1, %w5\n" +" cbnz %w3, 4f\n" +"2: stlxr %w3, %w6, %2\n" +" cbz %w3, 3f\n" +" sub %w4, %w4, %w3\n" +" cbnz %w4, 1b\n" +" mov %w0, %w8\n" "3:\n" +" dmb ish\n" +"4:\n" " .pushsection .fixup,\"ax\"\n" -"4: mov %w0, %w6\n" -" b 3b\n" +"5: mov %w0, %w7\n" +" b 4b\n" " .popsection\n" - _ASM_EXTABLE(1b, 4b) - _ASM_EXTABLE(2b, 4b) - : "+r" (ret), "=&r" (val), "+Q" (*uaddr), "=&r" (tmp) - : "r" (oldval), "r" (newval), "Ir" (-EFAULT) + _ASM_EXTABLE(1b, 5b) + _ASM_EXTABLE(2b, 5b) + : "+r" (ret), "=&r" (val), "+Q" (*uaddr), "=&r" (tmp), "+r" (loops) + : "r" (oldval), "r" (newval), "Ir" (-EFAULT), "Ir" (-EAGAIN) : "memory"); uaccess_disable(); From 8e4e0ac02b449297b86498ac24db5786ddd9f647 Mon Sep 17 00:00:00 2001 From: Will Deacon Date: Wed, 10 Apr 2019 11:49:11 +0100 Subject: [PATCH 4/5] arm64: futex: Avoid copying out uninitialised stack in failed cmpxchg() Returning an error code from futex_atomic_cmpxchg_inatomic() indicates that the caller should not make any use of *uval, and should instead act upon on the value of the error code. Although this is implemented correctly in our futex code, we needlessly copy uninitialised stack to *uval in the error case, which can easily be avoided. Signed-off-by: Will Deacon --- arch/arm64/include/asm/futex.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/arch/arm64/include/asm/futex.h b/arch/arm64/include/asm/futex.h index 2d78ea6932b7..bdb3c05070a2 100644 --- a/arch/arm64/include/asm/futex.h +++ b/arch/arm64/include/asm/futex.h @@ -134,7 +134,9 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *_uaddr, : "memory"); uaccess_disable(); - *uval = val; + if (!ret) + *uval = val; + return ret; } From 427503519739e779c0db8afe876c1b33f3ac60ae Mon Sep 17 00:00:00 2001 From: Will Deacon Date: Wed, 10 Apr 2019 11:51:54 +0100 Subject: [PATCH 5/5] futex: Update comments and docs about return values of arch futex code The architecture implementations of 'arch_futex_atomic_op_inuser()' and 'futex_atomic_cmpxchg_inatomic()' are permitted to return only -EFAULT, -EAGAIN or -ENOSYS in the case of failure. Update the comments in the asm-generic/ implementation and also a stray reference in the robust futex documentation. Signed-off-by: Will Deacon --- Documentation/robust-futexes.txt | 3 +-- include/asm-generic/futex.h | 8 ++++++-- 2 files changed, 7 insertions(+), 4 deletions(-) diff --git a/Documentation/robust-futexes.txt b/Documentation/robust-futexes.txt index 6c42c75103eb..6361fb01c9c1 100644 --- a/Documentation/robust-futexes.txt +++ b/Documentation/robust-futexes.txt @@ -218,5 +218,4 @@ All other architectures should build just fine too - but they won't have the new syscalls yet. Architectures need to implement the new futex_atomic_cmpxchg_inatomic() -inline function before writing up the syscalls (that function returns --ENOSYS right now). +inline function before writing up the syscalls. diff --git a/include/asm-generic/futex.h b/include/asm-generic/futex.h index fcb61b4659b3..8666fe7f35d7 100644 --- a/include/asm-generic/futex.h +++ b/include/asm-generic/futex.h @@ -23,7 +23,9 @@ * * Return: * 0 - On success - * <0 - On error + * -EFAULT - User access resulted in a page fault + * -EAGAIN - Atomic operation was unable to complete due to contention + * -ENOSYS - Operation not supported */ static inline int arch_futex_atomic_op_inuser(int op, u32 oparg, int *oval, u32 __user *uaddr) @@ -85,7 +87,9 @@ out_pagefault_enable: * * Return: * 0 - On success - * <0 - On error + * -EFAULT - User access resulted in a page fault + * -EAGAIN - Atomic operation was unable to complete due to contention + * -ENOSYS - Function not implemented (only if !HAVE_FUTEX_CMPXCHG) */ static inline int futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr,