diff mbox series

[1/2] bpf: Adjust BPF stack helper functions to accommodate skip > 0

Message ID 20220310082202.1229345-1-namhyung@kernel.org (mailing list archive)
State Superseded
Delegated to: BPF
Headers show
Series [1/2] bpf: Adjust BPF stack helper functions to accommodate skip > 0 | expand

Checks

Context Check Description
bpf/vmtest-bpf-next-PR success PR summary
bpf/vmtest-bpf-next success VM_Test
netdev/tree_selection success Guessing tree name failed - patch did not apply, async

Commit Message

Namhyung Kim March 10, 2022, 8:22 a.m. UTC 
Let's say that the caller has storage for num_elem stack frames.  Then,
the BPF stack helper functions walk the stack for only num_elem frames.
This means that if skip > 0, one keeps only 'num_elem - skip' frames.

This is because it sets init_nr in the perf_callchain_entry to the end
of the buffer to save num_elem entries only.  I believe it was because
the perf callchain code unwound the stack frames until it reached the
global max size (sysctl_perf_event_max_stack).

However it now has perf_callchain_entry_ctx.max_stack to limit the
iteration locally.  This simplifies the code to handle init_nr in the
BPF callstack entries and removes the confusion with the perf_event's
__PERF_SAMPLE_CALLCHAIN_EARLY which sets init_nr to 0.

Also change the comment on bpf_get_stack() in the header file to be
more explicit what the return value means.

Based-on-patch-by: Eugene Loh <eugene.loh@oracle.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
---
 include/uapi/linux/bpf.h |  4 +--
 kernel/bpf/stackmap.c    | 56 +++++++++++++++++-----------------------
 2 files changed, 26 insertions(+), 34 deletions(-)

Comments

Yonghong Song March 10, 2022, 10:54 p.m. UTC | #1 
On 3/10/22 12:22 AM, Namhyung Kim wrote:
> Let's say that the caller has storage for num_elem stack frames.  Then,
> the BPF stack helper functions walk the stack for only num_elem frames.
> This means that if skip > 0, one keeps only 'num_elem - skip' frames.
> 
> This is because it sets init_nr in the perf_callchain_entry to the end
> of the buffer to save num_elem entries only.  I believe it was because
> the perf callchain code unwound the stack frames until it reached the
> global max size (sysctl_perf_event_max_stack).
> 
> However it now has perf_callchain_entry_ctx.max_stack to limit the
> iteration locally.  This simplifies the code to handle init_nr in the
> BPF callstack entries and removes the confusion with the perf_event's
> __PERF_SAMPLE_CALLCHAIN_EARLY which sets init_nr to 0.
> 
> Also change the comment on bpf_get_stack() in the header file to be
> more explicit what the return value means.
> 
> Based-on-patch-by: Eugene Loh <eugene.loh@oracle.com>
> Signed-off-by: Namhyung Kim <namhyung@kernel.org>

The change looks good to me. This patch actually fixed a bug
discussed below:

 
https://lore.kernel.org/bpf/30a7b5d5-6726-1cc2-eaee-8da2828a9a9c@oracle.com/

A reference to the above link in the commit message
will be useful for people to understand better with an
example.

Also, the following fixes tag should be added:

Fixes: c195651e565a ("bpf: add bpf_get_stack helper")

Since the bug needs skip > 0 which is seldomly used,
and the current returned stack is still correct although
with less entries, I guess that is why less people
complains.

Anyway, ack the patch:
Acked-by: Yonghong Song <yhs@fb.com>


> ---
>   include/uapi/linux/bpf.h |  4 +--
>   kernel/bpf/stackmap.c    | 56 +++++++++++++++++-----------------------
>   2 files changed, 26 insertions(+), 34 deletions(-)
> 
> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> index b0383d371b9a..77f4a022c60c 100644
> --- a/include/uapi/linux/bpf.h
> +++ b/include/uapi/linux/bpf.h
> @@ -2975,8 +2975,8 @@ union bpf_attr {
>    *
>    * 			# sysctl kernel.perf_event_max_stack=<new value>
>    * 	Return
> - * 		A non-negative value equal to or less than *size* on success,
> - * 		or a negative error in case of failure.
> + * 		The non-negative copied *buf* length equal to or less than
> + * 		*size* on success, or a negative error in case of failure.
>    *
>    * long bpf_skb_load_bytes_relative(const void *skb, u32 offset, void *to, u32 len, u32 start_header)
[...]
Hao Luo March 11, 2022, 12:23 a.m. UTC | #2 
On Thu, Mar 10, 2022 at 2:54 PM Yonghong Song <yhs@fb.com> wrote:
>
>
>
> On 3/10/22 12:22 AM, Namhyung Kim wrote:
> > Let's say that the caller has storage for num_elem stack frames.  Then,
> > the BPF stack helper functions walk the stack for only num_elem frames.
> > This means that if skip > 0, one keeps only 'num_elem - skip' frames.
> >
> > This is because it sets init_nr in the perf_callchain_entry to the end
> > of the buffer to save num_elem entries only.  I believe it was because
> > the perf callchain code unwound the stack frames until it reached the
> > global max size (sysctl_perf_event_max_stack).
> >
> > However it now has perf_callchain_entry_ctx.max_stack to limit the
> > iteration locally.  This simplifies the code to handle init_nr in the
> > BPF callstack entries and removes the confusion with the perf_event's
> > __PERF_SAMPLE_CALLCHAIN_EARLY which sets init_nr to 0.
> >
> > Also change the comment on bpf_get_stack() in the header file to be
> > more explicit what the return value means.
> >
> > Based-on-patch-by: Eugene Loh <eugene.loh@oracle.com>
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
>
> The change looks good to me. This patch actually fixed a bug
> discussed below:
>
>
> https://lore.kernel.org/bpf/30a7b5d5-6726-1cc2-eaee-8da2828a9a9c@oracle.com/
>
> A reference to the above link in the commit message
> will be useful for people to understand better with an
> example.
>
> Also, the following fixes tag should be added:
>
> Fixes: c195651e565a ("bpf: add bpf_get_stack helper")
>
> Since the bug needs skip > 0 which is seldomly used,
> and the current returned stack is still correct although
> with less entries, I guess that is why less people
> complains.
>
> Anyway, ack the patch:
> Acked-by: Yonghong Song <yhs@fb.com>
>
>
> > ---
> >   include/uapi/linux/bpf.h |  4 +--
> >   kernel/bpf/stackmap.c    | 56 +++++++++++++++++-----------------------
> >   2 files changed, 26 insertions(+), 34 deletions(-)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index b0383d371b9a..77f4a022c60c 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -2975,8 +2975,8 @@ union bpf_attr {
> >    *
> >    *                  # sysctl kernel.perf_event_max_stack=<new value>
> >    *  Return
> > - *           A non-negative value equal to or less than *size* on success,
> > - *           or a negative error in case of failure.
> > + *           The non-negative copied *buf* length equal to or less than
> > + *           *size* on success, or a negative error in case of failure.
> >    *
> >    * long bpf_skb_load_bytes_relative(const void *skb, u32 offset, void *to, u32 len, u32 start_header)

Namhyung, I think you also need to mirror the change in
tools/include/uapi/linux/bpf.h
Namhyung Kim March 11, 2022, 12:32 a.m. UTC | #3 
Hello,

On Thu, Mar 10, 2022 at 2:55 PM Yonghong Song <yhs@fb.com> wrote:
>
>
>
> On 3/10/22 12:22 AM, Namhyung Kim wrote:
> > Let's say that the caller has storage for num_elem stack frames.  Then,
> > the BPF stack helper functions walk the stack for only num_elem frames.
> > This means that if skip > 0, one keeps only 'num_elem - skip' frames.
> >
> > This is because it sets init_nr in the perf_callchain_entry to the end
> > of the buffer to save num_elem entries only.  I believe it was because
> > the perf callchain code unwound the stack frames until it reached the
> > global max size (sysctl_perf_event_max_stack).
> >
> > However it now has perf_callchain_entry_ctx.max_stack to limit the
> > iteration locally.  This simplifies the code to handle init_nr in the
> > BPF callstack entries and removes the confusion with the perf_event's
> > __PERF_SAMPLE_CALLCHAIN_EARLY which sets init_nr to 0.
> >
> > Also change the comment on bpf_get_stack() in the header file to be
> > more explicit what the return value means.
> >
> > Based-on-patch-by: Eugene Loh <eugene.loh@oracle.com>
> > Signed-off-by: Namhyung Kim <namhyung@kernel.org>
>
> The change looks good to me. This patch actually fixed a bug
> discussed below:
>
>
> https://lore.kernel.org/bpf/30a7b5d5-6726-1cc2-eaee-8da2828a9a9c@oracle.com/
>
> A reference to the above link in the commit message
> will be useful for people to understand better with an
> example.

Ok, will do.

>
> Also, the following fixes tag should be added:
>
> Fixes: c195651e565a ("bpf: add bpf_get_stack helper")

Sure.

>
> Since the bug needs skip > 0 which is seldomly used,
> and the current returned stack is still correct although
> with less entries, I guess that is why less people
> complains.
>
> Anyway, ack the patch:
> Acked-by: Yonghong Song <yhs@fb.com>

Thanks for your review!

Namhyung


>
>
> > ---
> >   include/uapi/linux/bpf.h |  4 +--
> >   kernel/bpf/stackmap.c    | 56 +++++++++++++++++-----------------------
> >   2 files changed, 26 insertions(+), 34 deletions(-)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index b0383d371b9a..77f4a022c60c 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -2975,8 +2975,8 @@ union bpf_attr {
> >    *
> >    *                  # sysctl kernel.perf_event_max_stack=<new value>
> >    *  Return
> > - *           A non-negative value equal to or less than *size* on success,
> > - *           or a negative error in case of failure.
> > + *           The non-negative copied *buf* length equal to or less than
> > + *           *size* on success, or a negative error in case of failure.
> >    *
> >    * long bpf_skb_load_bytes_relative(const void *skb, u32 offset, void *to, u32 len, u32 start_header)
> [...]
Namhyung Kim March 11, 2022, 12:33 a.m. UTC | #4 
Hi Hao,

On Thu, Mar 10, 2022 at 4:24 PM Hao Luo <haoluo@google.com> wrote:
> > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > index b0383d371b9a..77f4a022c60c 100644
> > > --- a/include/uapi/linux/bpf.h
> > > +++ b/include/uapi/linux/bpf.h
> > > @@ -2975,8 +2975,8 @@ union bpf_attr {
> > >    *
> > >    *                  # sysctl kernel.perf_event_max_stack=<new value>
> > >    *  Return
> > > - *           A non-negative value equal to or less than *size* on success,
> > > - *           or a negative error in case of failure.
> > > + *           The non-negative copied *buf* length equal to or less than
> > > + *           *size* on success, or a negative error in case of failure.
> > >    *
> > >    * long bpf_skb_load_bytes_relative(const void *skb, u32 offset, void *to, u32 len, u32 start_header)
>
> Namhyung, I think you also need to mirror the change in
> tools/include/uapi/linux/bpf.h

Oh, right.  Will update.

Thanks,
Namhyung
diff mbox series

Patch

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index b0383d371b9a..77f4a022c60c 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h
@@ -2975,8 +2975,8 @@  union bpf_attr {
  *
  * 			# sysctl kernel.perf_event_max_stack=<new value>
  * 	Return
- * 		A non-negative value equal to or less than *size* on success,
- * 		or a negative error in case of failure.
+ * 		The non-negative copied *buf* length equal to or less than
+ * 		*size* on success, or a negative error in case of failure.
  *
  * long bpf_skb_load_bytes_relative(const void *skb, u32 offset, void *to, u32 len, u32 start_header)
  * 	Description
diff --git a/kernel/bpf/stackmap.c b/kernel/bpf/stackmap.c
index 22c8ae94e4c1..2823dcefae10 100644
--- a/kernel/bpf/stackmap.c
+++ b/kernel/bpf/stackmap.c
@@ -166,7 +166,7 @@  static void stack_map_get_build_id_offset(struct bpf_stack_build_id *id_offs,
 }
 
 static struct perf_callchain_entry *
-get_callchain_entry_for_task(struct task_struct *task, u32 init_nr)
+get_callchain_entry_for_task(struct task_struct *task, u32 max_depth)
 {
 #ifdef CONFIG_STACKTRACE
 	struct perf_callchain_entry *entry;
@@ -177,9 +177,8 @@  get_callchain_entry_for_task(struct task_struct *task, u32 init_nr)
 	if (!entry)
 		return NULL;
 
-	entry->nr = init_nr +
-		stack_trace_save_tsk(task, (unsigned long *)(entry->ip + init_nr),
-				     sysctl_perf_event_max_stack - init_nr, 0);
+	entry->nr = stack_trace_save_tsk(task, (unsigned long *)entry->ip,
+					 max_depth, 0);
 
 	/* stack_trace_save_tsk() works on unsigned long array, while
 	 * perf_callchain_entry uses u64 array. For 32-bit systems, it is
@@ -191,7 +190,7 @@  get_callchain_entry_for_task(struct task_struct *task, u32 init_nr)
 		int i;
 
 		/* copy data from the end to avoid using extra buffer */
-		for (i = entry->nr - 1; i >= (int)init_nr; i--)
+		for (i = entry->nr - 1; i >= 0; i--)
 			to[i] = (u64)(from[i]);
 	}
 
@@ -208,27 +207,19 @@  static long __bpf_get_stackid(struct bpf_map *map,
 {
 	struct bpf_stack_map *smap = container_of(map, struct bpf_stack_map, map);
 	struct stack_map_bucket *bucket, *new_bucket, *old_bucket;
-	u32 max_depth = map->value_size / stack_map_data_size(map);
-	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
-	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 	u32 hash, id, trace_nr, trace_len;
 	bool user = flags & BPF_F_USER_STACK;
 	u64 *ips;
 	bool hash_matches;
 
-	/* get_perf_callchain() guarantees that trace->nr >= init_nr
-	 * and trace-nr <= sysctl_perf_event_max_stack, so trace_nr <= max_depth
-	 */
-	trace_nr = trace->nr - init_nr;
-
-	if (trace_nr <= skip)
+	if (trace->nr <= skip)
 		/* skipping more than usable stack trace */
 		return -EFAULT;
 
-	trace_nr -= skip;
+	trace_nr = trace->nr - skip;
 	trace_len = trace_nr * sizeof(u64);
-	ips = trace->ip + skip + init_nr;
+	ips = trace->ip + skip;
 	hash = jhash2((u32 *)ips, trace_len / sizeof(u32), 0);
 	id = hash & (smap->n_buckets - 1);
 	bucket = READ_ONCE(smap->buckets[id]);
@@ -285,8 +276,7 @@  BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 	   u64, flags)
 {
 	u32 max_depth = map->value_size / stack_map_data_size(map);
-	/* stack_map_alloc() checks that max_depth <= sysctl_perf_event_max_stack */
-	u32 init_nr = sysctl_perf_event_max_stack - max_depth;
+	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 	bool user = flags & BPF_F_USER_STACK;
 	struct perf_callchain_entry *trace;
 	bool kernel = !user;
@@ -295,8 +285,12 @@  BPF_CALL_3(bpf_get_stackid, struct pt_regs *, regs, struct bpf_map *, map,
 			       BPF_F_FAST_STACK_CMP | BPF_F_REUSE_STACKID)))
 		return -EINVAL;
 
-	trace = get_perf_callchain(regs, init_nr, kernel, user,
-				   sysctl_perf_event_max_stack, false, false);
+	max_depth += skip;
+	if (max_depth > sysctl_perf_event_max_stack)
+		max_depth = sysctl_perf_event_max_stack;
+
+	trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
+				   false, false);
 
 	if (unlikely(!trace))
 		/* couldn't fetch the stack trace */
@@ -387,7 +381,7 @@  static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 			    struct perf_callchain_entry *trace_in,
 			    void *buf, u32 size, u64 flags)
 {
-	u32 init_nr, trace_nr, copy_len, elem_size, num_elem;
+	u32 trace_nr, copy_len, elem_size, num_elem, max_depth;
 	bool user_build_id = flags & BPF_F_USER_BUILD_ID;
 	u32 skip = flags & BPF_F_SKIP_FIELD_MASK;
 	bool user = flags & BPF_F_USER_STACK;
@@ -412,30 +406,28 @@  static long __bpf_get_stack(struct pt_regs *regs, struct task_struct *task,
 		goto err_fault;
 
 	num_elem = size / elem_size;
-	if (sysctl_perf_event_max_stack < num_elem)
-		init_nr = 0;
-	else
-		init_nr = sysctl_perf_event_max_stack - num_elem;
+	max_depth = num_elem + skip;
+	if (sysctl_perf_event_max_stack < max_depth)
+		max_depth = sysctl_perf_event_max_stack;
 
 	if (trace_in)
 		trace = trace_in;
 	else if (kernel && task)
-		trace = get_callchain_entry_for_task(task, init_nr);
+		trace = get_callchain_entry_for_task(task, max_depth);
 	else
-		trace = get_perf_callchain(regs, init_nr, kernel, user,
-					   sysctl_perf_event_max_stack,
+		trace = get_perf_callchain(regs, 0, kernel, user, max_depth,
 					   false, false);
 	if (unlikely(!trace))
 		goto err_fault;
 
-	trace_nr = trace->nr - init_nr;
-	if (trace_nr < skip)
+	if (trace->nr < skip)
 		goto err_fault;
 
-	trace_nr -= skip;
+	trace_nr = trace->nr - skip;
 	trace_nr = (trace_nr <= num_elem) ? trace_nr : num_elem;
 	copy_len = trace_nr * elem_size;
-	ips = trace->ip + skip + init_nr;
+
+	ips = trace->ip + skip;
 	if (user && user_build_id)
 		stack_map_get_build_id_offset(buf, ips, trace_nr, user);
 	else