Skip to contentSkip to conversation

perf/x86/intel: Work around BTS leaking kernel addresses

PATCH New Alexander Shishkin 2 participants 5 messages Aug 31, 2015 – Sep 13, 2015

Subsystems (2)
lore
Share & tools
lore

Alexander Shishkin

author reviewer
#0 lore raw

cover perf/x86/intel: Work around BTS leaking kernel addresses

Hi Peter and Ingo,

This is a respin to address Ingo's comments: I replaced the awkward address comparison against PAGE_OFFSET with a kernel_ip() in the first patch and amended the wording in the second. The original story follows.

Due to imprecise IP sampling, BTS may (and often does) leak kernel addresses when kernel tracing is not even enabled, which results in mostly syscall entry points and page_fault handler addresses being exposed. This might be a security concern for the address randomization, and it also makes life harder for gdb that makes use of bts samples.

This patchset works around the old (DS) driver and disables the new (intel_bts) for the unprivileged users on systems where perf paranoia level prohibits kernel tracing.

Not sure if these should be treated as bugfixes.

Alexander Shishkin (2):
perf/x86/intel/ds: Work around BTS leaking kernel addresses perf/x86/intel/bts: Disallow use by unprivileged users on paranoid
systems

 arch/x86/kernel/cpu/perf_event_intel_bts.c | 13 ++++++++++
 arch/x86/kernel/cpu/perf_event_intel_ds.c  | 40 +++++++++++++++++++++++++-----
 2 files changed, 47 insertions(+), 6 deletions(-)
Signature · 1 line
2.5.0
author reviewer
#1 lore raw

PATCH 1/2 perf/x86/intel/ds: Work around BTS leaking kernel addresses

BTS leaks kernel addresses even in userspace-only mode due to imprecise IP sampling, so sometimes syscall entry points or page fault handler addresses end up in a userspace trace.

Since this driver uses a relatively small buffer for BTS records and it has to iterate through them anyway, it can also take on the additional job of filtering out the records that contain kernel addresses when kernel space tracing is not enabled.

This patch changes the bts code to skip the offending records from perf output. In order to request the exact amount of space on the ring buffer, we need to do an extra pass through the records to know how many there are of the valid ones, but considering the small size of the buffer, this extra pass adds very little overhead to the nmi handler. This way we won't end up with awkward IP samples with zero IPs in the perf stream.

---
 arch/x86/kernel/cpu/perf_event_intel_ds.c | 40 ++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/perf_event_intel_ds.c +34 -6
index 84f236ab96..5db1c77555 100644
@@ -510,10 +510,11 @@ int intel_pmu_drain_bts_buffer(void)
 		u64	flags;
 	};
 	struct perf_event *event = cpuc->events[INTEL_PMC_IDX_FIXED_BTS];
-	struct bts_record *at, *top;
+	struct bts_record *at, *base, *top;
 	struct perf_output_handle handle;
 	struct perf_event_header header;
 	struct perf_sample_data data;
+	unsigned long skip = 0;
 	struct pt_regs regs;
 
 	if (!event)
@@ -522,10 +523,10 @@ int intel_pmu_drain_bts_buffer(void)
 	if (!x86_pmu.bts_active)
 		return 0;
 
-	at  = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
-	top = (struct bts_record *)(unsigned long)ds->bts_index;
+	base = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
+	top  = (struct bts_record *)(unsigned long)ds->bts_index;
 
-	if (top <= at)
+	if (top <= base)
 		return 0;
 
 	memset(&regs, 0, sizeof(regs));
@@ -535,16 +536,43 @@ int intel_pmu_drain_bts_buffer(void)
 	perf_sample_data_init(&data, 0, event->hw.last_period);
 
 	/*
+	 * BTS leaks kernel addresses in branches across the cpl boundary,
+	 * such as traps or system calls, so unless the user is asking for
+	 * kernel tracing (and right now it's not possible), we'd need to
+	 * filter them out. But first we need to count how many of those we
+	 * have in the current batch. This is an extra O(n) pass, however,
+	 * it's much faster than the other one especially considering that
+	 * n <= 2560 (BTS_BUFFER_SIZE / BTS_RECORD_SIZE * 15/16; see the
+	 * alloc_bts_buffer()).
+	 */
+	for (at = base; at < top; at++) {
+		/*
+		 * Note that right now *this* BTS code only works if
+		 * attr::exclude_kernel is set, but let's keep this extra
+		 * check here in case that changes.
+		 */
+		if (event->attr.exclude_kernel &&
+		    (kernel_ip(at->from) || kernel_ip(at->to)))
+			skip++;
+	}
+
+	/*
 	 * Prepare a generic sample, i.e. fill in the invariant fields.
 	 * We will overwrite the from and to address before we output
 	 * the sample.
 	 */
 	perf_prepare_sample(&header, &data, event, &regs);
 
-	if (perf_output_begin(&handle, event, header.size * (top - at)))
+	if (perf_output_begin(&handle, event, header.size *
+			      (top - base - skip)))
 		return 1;
 
-	for (; at < top; at++) {
+	for (at = base; at < top; at++) {
+		/* Filter out any records that contain kernel addresses. */
+		if (event->attr.exclude_kernel &&
+		    (kernel_ip(at->from) || kernel_ip(at->to)))
+			continue;
+
 		data.ip		= at->from;
 		data.addr	= at->to;
 
Signature · 1 line
2.5.0
#2 lore raw

[tip:perf/core] perf/x86/intel/ds: Work around BTS leaking kernel addresses

perf/x86/intel/ds: Work around BTS leaking kernel addresses
Commit-ID:  a09d31f45224c219a4d5c728fa40150dc9d3e3e5
Gitweb:     http://git.kernel.org/tip/a09d31f45224c219a4d5c728fa40150dc9d3e3e5
Author:     Alexander Shishkin <[email protected]>
AuthorDate: Mon, 31 Aug 2015 17:09:27 +0300
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sun, 13 Sep 2015 11:27:21 +0200

perf/x86/intel/ds: Work around BTS leaking kernel addresses

BTS leaks kernel addresses even in userspace-only mode due to imprecise IP sampling, so sometimes syscall entry points or page fault handler addresses end up in a userspace trace.

Since this driver uses a relatively small buffer for BTS records and it has to iterate through them anyway, it can also take on the additional job of filtering out the records that contain kernel addresses when kernel space tracing is not enabled.

This patch changes the bts code to skip the offending records from perf output. In order to request the exact amount of space on the ring buffer, we need to do an extra pass through the records to know how many there are of the valid ones, but considering the small size of the buffer, this extra pass adds very little overhead to the nmi handler. This way we won't end up with awkward IP samples with zero IPs in the perf stream.

---
 arch/x86/kernel/cpu/perf_event_intel_ds.c | 40 ++++++++++++++++++++++++++-----
 1 file changed, 34 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/perf_event_intel_ds.c +34 -6
index 84f236a..5db1c77 100644
@@ -510,10 +510,11 @@ int intel_pmu_drain_bts_buffer(void)
 		u64	flags;
 	};
 	struct perf_event *event = cpuc->events[INTEL_PMC_IDX_FIXED_BTS];
-	struct bts_record *at, *top;
+	struct bts_record *at, *base, *top;
 	struct perf_output_handle handle;
 	struct perf_event_header header;
 	struct perf_sample_data data;
+	unsigned long skip = 0;
 	struct pt_regs regs;
 
 	if (!event)
@@ -522,10 +523,10 @@ int intel_pmu_drain_bts_buffer(void)
 	if (!x86_pmu.bts_active)
 		return 0;
 
-	at  = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
-	top = (struct bts_record *)(unsigned long)ds->bts_index;
+	base = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
+	top  = (struct bts_record *)(unsigned long)ds->bts_index;
 
-	if (top <= at)
+	if (top <= base)
 		return 0;
 
 	memset(&regs, 0, sizeof(regs));
@@ -535,16 +536,43 @@ int intel_pmu_drain_bts_buffer(void)
 	perf_sample_data_init(&data, 0, event->hw.last_period);
 
 	/*
+	 * BTS leaks kernel addresses in branches across the cpl boundary,
+	 * such as traps or system calls, so unless the user is asking for
+	 * kernel tracing (and right now it's not possible), we'd need to
+	 * filter them out. But first we need to count how many of those we
+	 * have in the current batch. This is an extra O(n) pass, however,
+	 * it's much faster than the other one especially considering that
+	 * n <= 2560 (BTS_BUFFER_SIZE / BTS_RECORD_SIZE * 15/16; see the
+	 * alloc_bts_buffer()).
+	 */
+	for (at = base; at < top; at++) {
+		/*
+		 * Note that right now *this* BTS code only works if
+		 * attr::exclude_kernel is set, but let's keep this extra
+		 * check here in case that changes.
+		 */
+		if (event->attr.exclude_kernel &&
+		    (kernel_ip(at->from) || kernel_ip(at->to)))
+			skip++;
+	}
+
+	/*
 	 * Prepare a generic sample, i.e. fill in the invariant fields.
 	 * We will overwrite the from and to address before we output
 	 * the sample.
 	 */
 	perf_prepare_sample(&header, &data, event, &regs);
 
-	if (perf_output_begin(&handle, event, header.size * (top - at)))
+	if (perf_output_begin(&handle, event, header.size *
+			      (top - base - skip)))
 		return 1;
 
-	for (; at < top; at++) {
+	for (at = base; at < top; at++) {
+		/* Filter out any records that contain kernel addresses. */
+		if (event->attr.exclude_kernel &&
+		    (kernel_ip(at->from) || kernel_ip(at->to)))
+			continue;
+
 		data.ip		= at->from;
 		data.addr	= at->to;
 
author reviewer
#3 lore raw

PATCH 2/2 perf/x86/intel/bts: Disallow use by unprivileged users on paranoid systems

In reply to Alexander Shishkin

BTS leaks kernel addresses even in userspace-only mode due to imprecise IP sampling, so sometimes syscall entry points or page fault handler addresses end up in a userspace trace.

Now, intel_bts driver exports trace data zero-copy, it does not scan through it to filter out the kernel addresses and it's would be a O(n) job.

To work around this situation, this patch forbids the use of intel_bts driver by unprivileged users on systems with the paranoid setting above the (kernel's) default "1", which still allows kernel profiling. In other words, using intel_bts driver implies kernel tracing, regardless of the "exclude_kernel" attribute setting.

---
 arch/x86/kernel/cpu/perf_event_intel_bts.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)
arch/x86/kernel/cpu/perf_event_intel_bts.c +13 -0
index 80df16e020..014b0c34cd 100644
@@ -495,6 +495,19 @@ static int bts_event_init(struct perf_event *event)
 	if (x86_add_exclusive(x86_lbr_exclusive_bts))
 		return -EBUSY;
 
+	/*
+	 * BTS leaks kernel addresses even when CPL0 tracing is
+	 * disabled, so disallow intel_bts driver for unprivileged
+	 * users on paranoid systems since it provides trace data
+	 * to the user in a zero-copy fashion.
+	 *
+	 * Note that the default paranoia setting permits unprivileged
+	 * users to profile the kernel.
+	 */
+	if (event->attr.exclude_kernel && perf_paranoid_kernel() &&
+	    !capable(CAP_SYS_ADMIN))
+		return -EACCES;
+
 	ret = x86_reserve_hardware();
 	if (ret) {
 		x86_del_exclusive(x86_lbr_exclusive_bts);
Signature · 1 line
2.5.0
#4 lore raw

[tip:perf/core] perf/x86/intel/bts: Disallow use by unprivileged users on paranoid systems

perf/x86/intel/bts: Disallow use by unprivileged users on paranoid systems
Commit-ID:  d2878d642a4edd1d57c691dc3e4d7847cbf9d442
Gitweb:     http://git.kernel.org/tip/d2878d642a4edd1d57c691dc3e4d7847cbf9d442
Author:     Alexander Shishkin <[email protected]>
AuthorDate: Mon, 31 Aug 2015 17:09:28 +0300
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sun, 13 Sep 2015 11:27:22 +0200

perf/x86/intel/bts: Disallow use by unprivileged users on paranoid systems

BTS leaks kernel addresses even in userspace-only mode due to imprecise IP sampling, so sometimes syscall entry points or page fault handler addresses end up in a userspace trace.

Now, intel_bts driver exports trace data zero-copy, it does not scan through it to filter out the kernel addresses and it's would be a O(n) job.

To work around this situation, this patch forbids the use of intel_bts driver by unprivileged users on systems with the paranoid setting above the (kernel's) default "1", which still allows kernel profiling. In other words, using intel_bts driver implies kernel tracing, regardless of the "exclude_kernel" attribute setting.

---
 arch/x86/kernel/cpu/perf_event_intel_bts.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)
arch/x86/kernel/cpu/perf_event_intel_bts.c +13 -0
index d1c0f25..2cad71d 100644
@@ -495,6 +495,19 @@ static int bts_event_init(struct perf_event *event)
 	if (x86_add_exclusive(x86_lbr_exclusive_bts))
 		return -EBUSY;
 
+	/*
+	 * BTS leaks kernel addresses even when CPL0 tracing is
+	 * disabled, so disallow intel_bts driver for unprivileged
+	 * users on paranoid systems since it provides trace data
+	 * to the user in a zero-copy fashion.
+	 *
+	 * Note that the default paranoia setting permits unprivileged
+	 * users to profile the kernel.
+	 */
+	if (event->attr.exclude_kernel && perf_paranoid_kernel() &&
+	    !capable(CAP_SYS_ADMIN))
+		return -EACCES;
+
 	ret = x86_reserve_hardware();
 	if (ret) {
 		x86_del_exclusive(x86_lbr_exclusive_bts);