Skip to contentSkip to conversation

kprobes/arm: Improve kprobes implementation on arm

PATCH Superseded Masami Hiramatsu 2 participants 13 messages Feb 13, 2017 – Feb 14, 2017 Arm (32-bit)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover kprobes/arm: Improve kprobes implementation on arm

Hi,

Here are patches which improve kprobe on arm implementation. This includes some improves ported from x86 for multiple kretprobes on same function and recursing kprobes on FIQ (NMI) path. Also, I've fixed a bug(?) on recursing path.

- [1/3]: Port an improvement (and fix) for recursing kprobe on single-stepping by probing FIQ/NMI context.
- [2/3]: Skip single-stepping (and counting nmissed) if the recursing kprobe was hit on a conditional instruction which should not be executed.
- [3/3]: Fix to show correct return address with multiple kretprobe events on same function.

David, I think arm64 also has some conditinal instructions which should be skipped to handle it and single stepping if the condition is false. Or, user will see the probe events even when the instruction is not executed.

Thank you,

---

Masami Hiramatsu (3):
kprobes/arm: Allow to handle reentered kprobe on single-stepping kprobes/arm: Skip single-stepping in recursing path if possible kprobes/arm: Fix the return address of multiple kretprobes

 arch/arm/probes/kprobes/core.c |   49 +++++++++++++++++++++++++++++++---------
 1 file changed, 38 insertions(+), 11 deletions(-)

--
Masami Hiramatsu

#1 lore raw

PATCH 1/3 kprobes/arm: Allow to handle reentered kprobe on single-stepping

This is arm port of commit 6a5022a56ac3 ("kprobes/x86: Allow to handle reentered kprobe on single-stepping")

Since the FIQ handlers can interrupt in the single stepping (or preparing the single stepping, do_debug etc.), we should consider a kprobe is hit in the NMI handler. Even in that case, the kprobe is allowed to be reentered as same as the kprobes hit in kprobe handlers
(KPROBE_HIT_ACTIVE or KPROBE_HIT_SSDONE).

The real issue will happen when a kprobe hit while another reentered kprobe is processing (KPROBE_REENTER), because we already consumed a saved-area for the previous kprobe.

---
 arch/arm/probes/kprobes/core.c |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/probes/kprobes/core.c +6 -0
index a4ec240..264fedb 100644
@@ -270,6 +270,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
 			case KPROBE_HIT_SSDONE:
+			case KPROBE_HIT_SS:
 				/* A pre- or post-handler probe got us here. */
 				kprobes_inc_nmissed_count(p);
 				save_previous_kprobe(kcb);
@@ -278,6 +279,11 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				singlestep(p, regs, kcb);
 				restore_previous_kprobe(kcb);
 				break;
+			case KPROBE_REENTER:
+				/* A nested probe was hit in FIQ, it is a BUG */
+				pr_warn("Unrecoverable kprobe detected at %p.\n",
+					p->addr);
+				/* fall through */
 			default:
 				/* impossible cases */
 				BUG();
Acked-by on v1 1/3
#2 lore raw

Re: [BUGFIX PATCH 1/3] kprobes/arm: Allow to handle reentered kprobe on single-stepping

Quoted 11 lines · Masami Hiramatsu

This is arm port of commit 6a5022a56ac3 ("kprobes/x86: Allow to handle reentered kprobe on single-stepping")

Since the FIQ handlers can interrupt in the single stepping (or preparing the single stepping, do_debug etc.), we should consider a kprobe is hit in the NMI handler. Even in that case, the kprobe is allowed to be reentered as same as the kprobes hit in kprobe handlers
(KPROBE_HIT_ACTIVE or KPROBE_HIT_SSDONE).

The real issue will happen when a kprobe hit while another

Could to with 'is' being inserted above ^^^
(I know this is a copy of the x86 commit message)

Quoted 5 lines

reentered kprobe is processing (KPROBE_REENTER), because we already consumed a saved-area for the previous kprobe.

---

Quoted 28 lines
 arch/arm/probes/kprobes/core.c |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/probes/kprobes/core.c +6 -0
index a4ec240..264fedb 100644
@@ -270,6 +270,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
 			case KPROBE_HIT_SSDONE:
+			case KPROBE_HIT_SS:
 				/* A pre- or post-handler probe got us here. */
 				kprobes_inc_nmissed_count(p);
 				save_previous_kprobe(kcb);
@@ -278,6 +279,11 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				singlestep(p, regs, kcb);
 				restore_previous_kprobe(kcb);
 				break;
+			case KPROBE_REENTER:
+				/* A nested probe was hit in FIQ, it is a BUG */
+				pr_warn("Unrecoverable kprobe detected at %p.\n",
+					p->addr);
+				/* fall through */
 			default:
 				/* impossible cases */
 				BUG();
#3 lore raw

Re: [BUGFIX PATCH 1/3] kprobes/arm: Allow to handle reentered kprobe on single-stepping

Quoted 11 lines · On Tue, 14 Feb 2017 10:01:43 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 10 lines · Masami Hiramatsu

This is arm port of commit 6a5022a56ac3 ("kprobes/x86: Allow to handle reentered kprobe on single-stepping")

Since the FIQ handlers can interrupt in the single stepping (or preparing the single stepping, do_debug etc.), we should consider a kprobe is hit in the NMI handler. Even in that case, the kprobe is allowed to be reentered as same as the kprobes hit in kprobe handlers
(KPROBE_HIT_ACTIVE or KPROBE_HIT_SSDONE).

Quoted 4 lines
Quoted 1 line

The real issue will happen when a kprobe hit while another

Could to with 'is' being inserted above ^^^
(I know this is a copy of the x86 commit message)

Ah! yes, it should be corrected.

Quoted 8 lines
Quoted 6 lines

reentered kprobe is processing (KPROBE_REENTER), because we already consumed a saved-area for the previous kprobe.

---

Thank you!

Quoted 29 lines
Quoted 29 lines
 arch/arm/probes/kprobes/core.c |    6 ++++++
 1 file changed, 6 insertions(+)
arch/arm/probes/kprobes/core.c +6 -0
index a4ec240..264fedb 100644
@@ -270,6 +270,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
 			case KPROBE_HIT_SSDONE:
+			case KPROBE_HIT_SS:
 				/* A pre- or post-handler probe got us here. */
 				kprobes_inc_nmissed_count(p);
 				save_previous_kprobe(kcb);
@@ -278,6 +279,11 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				singlestep(p, regs, kcb);
 				restore_previous_kprobe(kcb);
 				break;
+			case KPROBE_REENTER:
+				/* A nested probe was hit in FIQ, it is a BUG */
+				pr_warn("Unrecoverable kprobe detected at %p.\n",
+					p->addr);
+				/* fall through */
 			default:
 				/* impossible cases */
 				BUG();
Signature · 1 line
Masami Hiramatsu <[email protected]>
#4 lore raw

PATCH 2/3 kprobes/arm: Skip single-stepping in recursing path if possible

In reply to Masami Hiramatsu

Kprobes/arm skips single-stepping (moreover handling the event) if the conditional instruction must not be executed. This also apply the rule when we hit the recursing kprobe, so that kprobe does not count nmissed up in that case.

---
 arch/arm/probes/kprobes/core.c |   19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)
arch/arm/probes/kprobes/core.c +10 -9
index 264fedb..84989ae 100644
@@ -265,7 +265,15 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 #endif
 
 	if (p) {
-		if (cur) {
+		if (!p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+			/*
+			 * Probe hit but conditional execution check failed,
+			 * so just skip the instruction and continue as if
+			 * nothing had happened.
+			 * In this case, we can skip recursing check too.
+			 */
+			singlestep_skip(p, regs);
+		} else if (cur) {
 			/* Kprobe is pending, so we're recursing. */
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
@@ -288,7 +296,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				/* impossible cases */
 				BUG();
 			}
-		} else if (p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+		} else {
 			/* Probe hit and conditional execution check ok. */
 			set_current_kprobe(p);
 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
@@ -309,13 +317,6 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				}
 				reset_current_kprobe();
 			}
-		} else {
-			/*
-			 * Probe hit but conditional execution check failed,
-			 * so just skip the instruction and continue as if
-			 * nothing had happened.
-			 */
-			singlestep_skip(p, regs);
 		}
 	} else if (cur) {
 		/* We probably hit a jprobe.  Call its break handler. */
Acked-by on v1 2/3
#5 lore raw

Re: [BUGFIX PATCH 2/3] kprobes/arm: Skip single-stepping in recursing path if possible

Quoted 4 lines · Masami Hiramatsu

Kprobes/arm skips single-stepping (moreover handling the event) if the conditional instruction must not be executed. This also apply the rule when we hit the recursing kprobe, so that kprobe does not count nmissed up in that case.

Perhaps that last sentence would read better if written something like:

"This also applies that rule when we hit a recursing kprobe, so that the nmissed count isn't incremented in that case."

Quoted 1 line
Quoted 49 lines
---
 arch/arm/probes/kprobes/core.c |   19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)
arch/arm/probes/kprobes/core.c +10 -9
index 264fedb..84989ae 100644
@@ -265,7 +265,15 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 #endif
 
 	if (p) {
-		if (cur) {
+		if (!p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+			/*
+			 * Probe hit but conditional execution check failed,
+			 * so just skip the instruction and continue as if
+			 * nothing had happened.
+			 * In this case, we can skip recursing check too.
+			 */
+			singlestep_skip(p, regs);
+		} else if (cur) {
 			/* Kprobe is pending, so we're recursing. */
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
@@ -288,7 +296,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				/* impossible cases */
 				BUG();
 			}
-		} else if (p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+		} else {
 			/* Probe hit and conditional execution check ok. */
 			set_current_kprobe(p);
 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
@@ -309,13 +317,6 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				}
 				reset_current_kprobe();
 			}
-		} else {
-			/*
-			 * Probe hit but conditional execution check failed,
-			 * so just skip the instruction and continue as if
-			 * nothing had happened.
-			 */
-			singlestep_skip(p, regs);
 		}
 	} else if (cur) {
 		/* We probably hit a jprobe.  Call its break handler. */
#6 lore raw

Re: [BUGFIX PATCH 2/3] kprobes/arm: Skip single-stepping in recursing path if possible

Quoted 10 lines · On Tue, 14 Feb 2017 10:07:17 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 4 lines · Masami Hiramatsu

Kprobes/arm skips single-stepping (moreover handling the event) if the conditional instruction must not be executed. This also apply the rule when we hit the recursing kprobe, so that kprobe does not count nmissed up in that case.

Perhaps that last sentence would read better if written something like:

"This also applies that rule when we hit a recursing kprobe, so that the nmissed count isn't incremented in that case."

OK, Thanks!

Quoted 55 lines
Quoted 3 lines
Quoted 49 lines
---
 arch/arm/probes/kprobes/core.c |   19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)
arch/arm/probes/kprobes/core.c +10 -9
index 264fedb..84989ae 100644
@@ -265,7 +265,15 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 #endif
 
 	if (p) {
-		if (cur) {
+		if (!p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+			/*
+			 * Probe hit but conditional execution check failed,
+			 * so just skip the instruction and continue as if
+			 * nothing had happened.
+			 * In this case, we can skip recursing check too.
+			 */
+			singlestep_skip(p, regs);
+		} else if (cur) {
 			/* Kprobe is pending, so we're recursing. */
 			switch (kcb->kprobe_status) {
 			case KPROBE_HIT_ACTIVE:
@@ -288,7 +296,7 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				/* impossible cases */
 				BUG();
 			}
-		} else if (p->ainsn.insn_check_cc(regs->ARM_cpsr)) {
+		} else {
 			/* Probe hit and conditional execution check ok. */
 			set_current_kprobe(p);
 			kcb->kprobe_status = KPROBE_HIT_ACTIVE;
@@ -309,13 +317,6 @@ void __kprobes kprobe_handler(struct pt_regs *regs)
 				}
 				reset_current_kprobe();
 			}
-		} else {
-			/*
-			 * Probe hit but conditional execution check failed,
-			 * so just skip the instruction and continue as if
-			 * nothing had happened.
-			 */
-			singlestep_skip(p, regs);
 		}
 	} else if (cur) {
 		/* We probably hit a jprobe.  Call its break handler. */
Signature · 1 line
Masami Hiramatsu <[email protected]>
#7 lore raw

PATCH 3/3 kprobes/arm: Fix the return address of multiple kretprobes

In reply to Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
---
 arch/arm/probes/kprobes/core.c |   24 ++++++++++++++++++++++--
 1 file changed, 22 insertions(+), 2 deletions(-)
arch/arm/probes/kprobes/core.c +22 -2
index 84989ae..023800a 100644
@@ -440,6 +440,7 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 	struct hlist_node *tmp;
 	unsigned long flags, orig_ret_address = 0;
 	unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
+	kprobe_opcode_t *correct_ret_addr = NULL;
 
 	INIT_HLIST_HEAD(&empty_rp);
 	kretprobe_hash_lock(current, &head, &flags);
@@ -462,14 +463,34 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 			/* another task is sharing our hash bucket */
 			continue;
 
+		orig_ret_address = (unsigned long)ri->ret_addr;
+
+		if (orig_ret_address != trampoline_address)
+			/*
+			 * This is the real return address. Any other
+			 * instances associated with this task are for
+			 * other calls deeper on the call stack
+			 */
+			break;
+	}
+
+	kretprobe_assert(ri, orig_ret_address, trampoline_address);
+
+	correct_ret_addr = ri->ret_addr;
+	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
+		if (ri->task != current)
+			/* another task is sharing our hash bucket */
+			continue;
+
+		orig_ret_address = (unsigned long)ri->ret_addr;
 		if (ri->rp && ri->rp->handler) {
 			__this_cpu_write(current_kprobe, &ri->rp->kp);
 			get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
+			ri->ret_addr = correct_ret_addr;
 			ri->rp->handler(ri, regs);
 			__this_cpu_write(current_kprobe, NULL);
 		}
 
-		orig_ret_address = (unsigned long)ri->ret_addr;
 		recycle_rp_inst(ri, &empty_rp);
 
 		if (orig_ret_address != trampoline_address)
@@ -481,7 +502,6 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 			break;
 	}
 
-	kretprobe_assert(ri, orig_ret_address, trampoline_address);
 	kretprobe_hash_unlock(current, &flags);
 
 	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
Acked-by on v1 3/3
#8 lore raw

Re: [BUGFIX PATCH 3/3] kprobes/arm: Fix the return address of multiple kretprobes

Quoted 22 lines · Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

Quoted 7 lines
  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

---

I don't fully understand this function, but I've checked that the ARM version now matches the x86 version (apart from the x86 specific register fixup and some comments). So, FWIW

I ran the before and after test case in the commit log on ARM and verified the result is correct. I also tried running the ARM kprobe tests with these 3 fixes but the tests fail. However, they also fail without any of these changes, so I'll investigate that further...

Quoted 60 lines
 arch/arm/probes/kprobes/core.c |   24 ++++++++++++++++++++++--
 1 file changed, 22 insertions(+), 2 deletions(-)
arch/arm/probes/kprobes/core.c +22 -2
index 84989ae..023800a 100644
@@ -440,6 +440,7 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 	struct hlist_node *tmp;
 	unsigned long flags, orig_ret_address = 0;
 	unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
+	kprobe_opcode_t *correct_ret_addr = NULL;
 
 	INIT_HLIST_HEAD(&empty_rp);
 	kretprobe_hash_lock(current, &head, &flags);
@@ -462,14 +463,34 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 			/* another task is sharing our hash bucket */
 			continue;
 
+		orig_ret_address = (unsigned long)ri->ret_addr;
+
+		if (orig_ret_address != trampoline_address)
+			/*
+			 * This is the real return address. Any other
+			 * instances associated with this task are for
+			 * other calls deeper on the call stack
+			 */
+			break;
+	}
+
+	kretprobe_assert(ri, orig_ret_address, trampoline_address);
+
+	correct_ret_addr = ri->ret_addr;
+	hlist_for_each_entry_safe(ri, tmp, head, hlist) {
+		if (ri->task != current)
+			/* another task is sharing our hash bucket */
+			continue;
+
+		orig_ret_address = (unsigned long)ri->ret_addr;
 		if (ri->rp && ri->rp->handler) {
 			__this_cpu_write(current_kprobe, &ri->rp->kp);
 			get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE;
+			ri->ret_addr = correct_ret_addr;
 			ri->rp->handler(ri, regs);
 			__this_cpu_write(current_kprobe, NULL);
 		}
 
-		orig_ret_address = (unsigned long)ri->ret_addr;
 		recycle_rp_inst(ri, &empty_rp);
 
 		if (orig_ret_address != trampoline_address)
@@ -481,7 +502,6 @@ static __used __kprobes void *trampoline_handler(struct pt_regs *regs)
 			break;
 	}
 
-	kretprobe_assert(ri, orig_ret_address, trampoline_address);
 	kretprobe_hash_unlock(current, &flags);
 
 	hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
#9 lore raw

Re: [BUGFIX PATCH 3/3] kprobes/arm: Fix the return address of multiple kretprobes

Quoted 35 lines · Jon Medhurst (Tixy)
Quoted 22 lines · Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

Quoted 7 lines
  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

---

I don't fully understand this function, but I've checked that the ARM version now matches the x86 version (apart from the x86 specific register fixup and some comments). So, FWIW

Quoted 6 lines

I ran the before and after test case in the commit log on ARM and verified the result is correct. I also tried running the ARM kprobe tests with these 3 fixes but the tests fail. However, they also fail without any of these changes, so I'll investigate that further...

Bisecting the issue led me back to Linux 4.5 and commit 25362dc496ed ("ARM: 8501/1: mm: flip priority of CONFIG_DEBUG_RODATA")

This sets CONFIG_DEBUG_RODATA to be enabled by default. If I disable that on 4.10-rc4, with the three patches in this series, then the ARM kprobes tests pass OK.

I'll stick the DEBUG_RODATA issue on my todo list (it's been around for a year, so can probably wait a little longer).

Signature · 1 line
Tixy
#10 lore raw

Re: [BUGFIX PATCH 3/3] kprobes/arm: Fix the return address of multiple kretprobes

Quoted 46 lines · On Tue, 14 Feb 2017 13:47:07 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 35 lines · Jon Medhurst (Tixy)
Quoted 22 lines · Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

Quoted 7 lines
  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

---

I don't fully understand this function, but I've checked that the ARM version now matches the x86 version (apart from the x86 specific register fixup and some comments). So, FWIW

Quoted 6 lines

I ran the before and after test case in the commit log on ARM and verified the result is correct. I also tried running the ARM kprobe tests with these 3 fixes but the tests fail. However, they also fail without any of these changes, so I'll investigate that further...

Bisecting the issue led me back to Linux 4.5 and commit 25362dc496ed ("ARM: 8501/1: mm: flip priority of CONFIG_DEBUG_RODATA")

Quoted 6 lines

This sets CONFIG_DEBUG_RODATA to be enabled by default. If I disable that on 4.10-rc4, with the three patches in this series, then the ARM kprobes tests pass OK.

I'll stick the DEBUG_RODATA issue on my todo list (it's been around for a year, so can probably wait a little longer).

Hmm, I'm running arm kernel on qemu, which maybe the reason why the test case passed in my environment, since my kconfig also sets CONFIG_DEBUG_RODATA=y.

BTW, would you see that any kprobe_events didn't work with CONFIG_DEBUG_RODATA=y? (what the failure messages were?)

Thank you,

Quoted 4 lines
Signature · 2 lines
-- 
Tixy
Signature · 1 line
Masami Hiramatsu <[email protected]>
#11 lore raw

Re: [BUGFIX PATCH 3/3] kprobes/arm: Fix the return address of multiple kretprobes

Quoted 56 lines · Masami Hiramatsu
Quoted 46 lines · On Tue, 14 Feb 2017 13:47:07 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 35 lines · Jon Medhurst (Tixy)
Quoted 22 lines · Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

Quoted 7 lines
  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

---

I don't fully understand this function, but I've checked that the ARM version now matches the x86 version (apart from the x86 specific register fixup and some comments). So, FWIW

Quoted 6 lines

I ran the before and after test case in the commit log on ARM and verified the result is correct. I also tried running the ARM kprobe tests with these 3 fixes but the tests fail. However, they also fail without any of these changes, so I'll investigate that further...

Bisecting the issue led me back to Linux 4.5 and commit 25362dc496ed ("ARM: 8501/1: mm: flip priority of CONFIG_DEBUG_RODATA")

Quoted 6 lines

This sets CONFIG_DEBUG_RODATA to be enabled by default. If I disable that on 4.10-rc4, with the three patches in this series, then the ARM kprobes tests pass OK.

I'll stick the DEBUG_RODATA issue on my todo list (it's been around for a year, so can probably wait a little longer).

Quoted 6 lines

Hmm, I'm running arm kernel on qemu, which maybe the reason why the test case passed in my environment, since my kconfig also sets CONFIG_DEBUG_RODATA=y.

BTW, would you see that any kprobe_events didn't work with CONFIG_DEBUG_RODATA=y? (what the failure messages were?)

The tests I'm running are the ARM specific tests that are enabled by CONFIG_ARM_KPROBES_TEST=y. I'm running the tests on real multicore ARM hardware (Versatile Express with a TC2 CoreTile)

For me, sometimes the first test gave:

Beginning kprobe tests...
Probe ARM code
    kprobe
FAIL: test regs not OK

Other times, for the specific instruction emulation tests they return

FAIL: test_before_handler not run

Not sure how much of the diagnostic appear without setting the tests to be verbose, which I do with:

sed -e 's/VERBOSE 0/VERBOSE 1/' -i arch/arm/probes/kprobes/test-core.h

Whilst writing a reply, I looked at the test code in arch/arm/probes/kprobes/test-core.c (which I wrote some years ago) and there is possibly a clue staring at us in the comments at the top of the file...

 *
 * The above would expand to assembler looking something like:
 *
 * @ TESTCASE_START
 * bl __kprobes_test_case_start
 * .pushsection .rodata
 * "10:
 * .ascii "mov r0, r7" @ text title for test case
 * .byte 0
 * .popsection
 * @ start of inline data...
 * .word 10b @ pointer to title in .rodata
section

Note the ".pushsection .rodata" (though I don't see an immediate obvious reason why that would cause a problem. It certainly seems likely that the problem is with the ARM test code rather than actual kprobe implementation itself.

Like I said, this issue has been there for a year or more, so I wasn't planning on spending time on it for a few more days yet whilst I get on with other urgent matters.

Signature · 4 lines
Tixy


Basically, m
#12 lore raw

Re: [BUGFIX PATCH 3/3] kprobes/arm: Fix the return address of multiple kretprobes

Quoted 61 lines · On Tue, 14 Feb 2017 16:39:50 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 60 lines · Masami Hiramatsu
Quoted 46 lines · On Tue, 14 Feb 2017 13:47:07 +0000 "Jon Medhurst (Tixy)" <[email protected]>…
Quoted 35 lines · Jon Medhurst (Tixy)
Quoted 22 lines · Masami Hiramatsu

This is arm port of commit 737480a0d525 ("kprobes/x86:
Fix the return address of multiple kretprobes").

Fix the return address of subsequent kretprobes when multiple kretprobes are set on the same function.

For example:

  # cd /sys/kernel/debug/tracing
  # echo "r:event1 sys_symlink" > kprobe_events
  # echo "r:event2 sys_symlink" >> kprobe_events
  # echo 1 > events/kprobes/enable
  # ln -s /tmp/foo /tmp/bar

(without this patch)

  # cat trace | grep -v ^#
              ln-82    [000] dn.2    68.446525: event1: (kretprobe_trampoline+0x0/0x18 <- SyS_symlink)
              ln-82    [000] dn.2    68.447831: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

(with this patch)

Quoted 7 lines
  # cat trace | grep -v ^#
              ln-81    [000] dn.1    39.463469: event1: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)
              ln-81    [000] dn.1    39.464701: event2: (ret_fast_syscall+0x0/0x1c <- SyS_symlink)

---

I don't fully understand this function, but I've checked that the ARM version now matches the x86 version (apart from the x86 specific register fixup and some comments). So, FWIW

Quoted 6 lines

I ran the before and after test case in the commit log on ARM and verified the result is correct. I also tried running the ARM kprobe tests with these 3 fixes but the tests fail. However, they also fail without any of these changes, so I'll investigate that further...

Bisecting the issue led me back to Linux 4.5 and commit 25362dc496ed ("ARM: 8501/1: mm: flip priority of CONFIG_DEBUG_RODATA")

Quoted 6 lines

This sets CONFIG_DEBUG_RODATA to be enabled by default. If I disable that on 4.10-rc4, with the three patches in this series, then the ARM kprobes tests pass OK.

I'll stick the DEBUG_RODATA issue on my todo list (it's been around for a year, so can probably wait a little longer).

Hmm, I'm running arm kernel on qemu, which maybe the reason why the test case passed in my environment, since my kconfig also sets CONFIG_DEBUG_RODATA=y.

Quoted 6 lines
Quoted 2 lines

BTW, would you see that any kprobe_events didn't work with CONFIG_DEBUG_RODATA=y? (what the failure messages were?)

The tests I'm running are the ARM specific tests that are enabled by CONFIG_ARM_KPROBES_TEST=y. I'm running the tests on real multicore ARM hardware (Versatile Express with a TC2 CoreTile)

Ah, I didn't enabled it. I'll also try to run it.

Quoted 36 lines

For me, sometimes the first test gave:

Beginning kprobe tests...
Probe ARM code
    kprobe
FAIL: test regs not OK

Other times, for the specific instruction emulation tests they return

FAIL: test_before_handler not run

Not sure how much of the diagnostic appear without setting the tests to be verbose, which I do with:

sed -e 's/VERBOSE 0/VERBOSE 1/' -i arch/arm/probes/kprobes/test-core.h

Whilst writing a reply, I looked at the test code in arch/arm/probes/kprobes/test-core.c (which I wrote some years ago) and there is possibly a clue staring at us in the comments at the top of the file...

 *
 * The above would expand to assembler looking something like:
 *
 * @ TESTCASE_START
 * bl __kprobes_test_case_start
 * .pushsection .rodata
 * "10:
 * .ascii "mov r0, r7" @ text title for test case
 * .byte 0
 * .popsection
 * @ start of inline data...
 * .word 10b @ pointer to title in .rodata
section

Quoted 4 lines

Note the ".pushsection .rodata" (though I don't see an immediate obvious reason why that would cause a problem. It certainly seems likely that the problem is with the ARM test code rather than actual kprobe implementation itself.

OK, then I'll just update the series according your comment.

Quoted 4 lines

Like I said, this issue has been there for a year or more, so I wasn't planning on spending time on it for a few more days yet whilst I get on with other urgent matters.

OK, I'll also investigate it.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>