perf/x86: Annotate struct bts_buffer with __counted_by()
Subsystems (2)
PATCH 1/1 perf/x86: Annotate struct bts_buffer with __counted_by()
Add the __counted_by() compiler attribute to the flexible array member buf to improve access bounds-checking via CONFIG_UBSAN_BOUNDS and CONFIG_FORTIFY_SOURCE.
Use struct_size() to calculate the number of bytes to allocate for a new bts_buffer. Compared to offsetof(), struct_size() has additional compile-time checks (e.g., __must_be_array()).
No functional changes intended.
--- arch/x86/events/intel/bts.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/events/intel/bts.c +2 -2
@@ -58,7 +58,7 @@ struct bts_buffer {local_t head;unsigned long end;void **data_pages; -struct bts_phys buf[]; + struct bts_phys buf[] __counted_by(nr_bufs); };static struct pmu bts_pmu;
@@ -101,7 +101,7 @@ bts_buffer_setup_aux(struct perf_event *event, void **pages,if (overwrite && nbuf >1 )return NULL ; - buf = kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node); + buf =kzalloc_node (struct_size (buf, buf, nbuf),GFP_KERNEL , node);if (!buf)return NULL ;
Signature · 1 line
2.48.1
The following commit has been merged into the perf/core branch of tip:
The following commit has been merged into the perf/core branch of tip:
Commit-ID: 077dcef270361089c322a969b792438b33cfb479 Gitweb: https://git.kernel.org/tip/077dcef270361089c322a969b792438b33cfb479 Author: Thorsten Blum <[email protected]> AuthorDate: Tue, 04 Mar 2025 19:30:57 +01:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Tue, 04 Mar 2025 19:58:01 +01:00
perf/x86: Annotate struct bts_buffer with __counted_by()
Add the __counted_by() compiler attribute to the flexible array member buf to improve access bounds-checking via CONFIG_UBSAN_BOUNDS and CONFIG_FORTIFY_SOURCE.
Use struct_size() to calculate the number of bytes to allocate for a new bts_buffer. Compared to offsetof(), struct_size() has additional compile-time checks (e.g., __must_be_array()).
No functional changes intended.
--- arch/x86/events/intel/bts.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/events/intel/bts.c +2 -2
@@ -58,7 +58,7 @@ struct bts_buffer {local_t head;unsigned long end;void **data_pages; -struct bts_phys buf[]; + struct bts_phys buf[] __counted_by(nr_bufs); };static struct pmu bts_pmu;
@@ -101,7 +101,7 @@ bts_buffer_setup_aux(struct perf_event *event, void **pages,if (overwrite && nbuf >1 )return NULL ; - buf = kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node); + buf =kzalloc_node (struct_size (buf, buf, nbuf),GFP_KERNEL , node);if (!buf)return NULL ;
Quoted 43 lines · * tip-bot2 for Thorsten Blum <[email protected]> wrote:
The following commit has been merged into the perf/core branch of tip:
Commit-ID: 077dcef270361089c322a969b792438b33cfb479 Gitweb: https://git.kernel.org/tip/077dcef270361089c322a969b792438b33cfb479 Author: Thorsten Blum <[email protected]> AuthorDate: Tue, 04 Mar 2025 19:30:57 +01:00 Committer: Ingo Molnar <[email protected]> CommitterDate: Tue, 04 Mar 2025 19:58:01 +01:00perf/x86: Annotate struct
bts_bufferwith__counted_by()Add the
__counted_by()compiler attribute to the flexible array member buf to improve access bounds-checking viaCONFIG_UBSAN_BOUNDSandCONFIG_FORTIFY_SOURCE.Use
struct_size()to calculate the number of bytes to allocate for a newbts_buffer. Compared tooffsetof(),struct_size()has additional compile-time checks (e.g.,__must_be_array()).No functional changes intended.
Signed-off-by Thorsten Blum <[email protected]>Signed-off-by Ingo Molnar <[email protected]>Cc Peter Zijlstra <[email protected]>--- arch/x86/events/intel/bts.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)arch/x86/events/intel/bts.c +1
-1@@ -58,7 +58,7 @@ struct bts_buffer {local_t head;unsigned long end;void **data_pages; -struct bts_phys buf[]; + struct bts_phys buf[] __counted_by(nr_bufs); };static struct pmu bts_pmu;
arch/x86/events/intel/bts.c bts_buffer_setup_aux(struct perf_event *event, void **pages, · line 102
Quoted 2 lines
arch/x86/events/intel/bts.c +0
-0@@ -101,7 +101,7 @@ bts_buffer_setup_aux(struct perf_event *event, void **pages,if (overwrite && nbuf >1 )return NULL ;
Actually, on a second thought:
arch/x86/events/intel/bts.c bts_buffer_setup_aux(struct perf_event *event, void **pages, · line 104
Quoted 2 lines
arch/x86/events/intel/bts.c +1
-1- buf = kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node); + buf =kzalloc_node (struct_size (buf, buf, nbuf),GFP_KERNEL , node);
Firstly, in what world is 'buf, buf' more readable? One is a member of a structure, the other is the name of the structure - and they match, which shows that this function's naming conventions are a mess.
Which should be fixed first ...
I'm also not sure the code is correct ...
So I zapped this commit from tip:perf/core.
Thanks,
Ingo
Quoted 10 lines · Ingo Molnar
Actually, on a second thought:
Quoted 2 lines
- buf =
kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node);
+ buf =kzalloc_node(struct_size(buf, buf, nbuf), GFP_KERNEL, node);Firstly, in what world is 'buf, buf' more readable? One is a member of a structure, the other is the name of the structure - and they match, which shows that this function's naming conventions are a mess.
Which should be fixed first ...
Yes, I noticed this too, but since buf->buf[] is used all over the place (also in other functions), I didn't rename it in this patch.
We could just keep offsetof(struct bts_buffer, buf[nbuf]), or use struct_size_t(struct bts_buffer, buf, nbuf) and still benefit from additional compile-time checks, or rename the local variable to struct bts_buffer *bts and use struct_size(bts, buf, nbuf), for example. Any preferences or other ideas?
Quoted 1 line
I'm also not sure the code is correct ...
Which part of it?
Thanks,
Thorsten
Quoted 12 lines · Thorsten Blum
Quoted 10 lines · Ingo Molnar
Actually, on a second thought:
Quoted 2 lines
- buf =
kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node);
+ buf =kzalloc_node(struct_size(buf, buf, nbuf), GFP_KERNEL, node);Firstly, in what world is 'buf, buf' more readable? One is a member of a structure, the other is the name of the structure - and they match, which shows that this function's naming conventions are a mess.
Which should be fixed first ...
Quoted 8 lines
Yes, I noticed this too, but since buf->buf[] is used all over the place (also in other functions), I didn't rename it in this patch.
We could just keep
offsetof(struct bts_buffer, buf[nbuf]), or usestruct_size_t(struct bts_buffer, buf, nbuf)and still benefit from additional compile-time checks, or rename the local variable to structbts_buffer*bts and usestruct_size(bts, buf, nbuf), for example. Any preferences or other ideas?
To clean up this code before changing it, so that the changes become obvious to review.
Please also split out the annotation for instrumentation, it's separate from any struct_size() changes, right?
Quoted 3 lines
Quoted 1 line
I'm also not sure the code is correct ...
Which part of it?
The size calculation. On a second reading I *think* it's correct, but it's unnecessarily confusing due to the buf<->buf aliasing.
So in a cleaned up version of the code:
- If we name 'struct bts_buffer' objects 'bb'
- and bb:buf[] is the var-array
- and we rename 'nbuf' to 'nr_buf' (the number of bb:buf[] elements)
then the code right now does:
bb = kzalloc_node(offsetof(struct bts_buffer, bb[nr_buf]), GFP_KERNEL, node);
... which looks correct.
Thanks,
Ingo
Quoted 22 lines · Ingo Molnar
Quoted 12 lines · Thorsten Blum
Quoted 10 lines · Ingo Molnar
Actually, on a second thought:
Quoted 2 lines
- buf =
kzalloc_node(offsetof(struct bts_buffer, buf[nbuf]), GFP_KERNEL, node);
+ buf =kzalloc_node(struct_size(buf, buf, nbuf), GFP_KERNEL, node);Firstly, in what world is 'buf, buf' more readable? One is a member of a structure, the other is the name of the structure - and they match, which shows that this function's naming conventions are a mess.
Which should be fixed first ...
Quoted 8 lines
Yes, I noticed this too, but since buf->buf[] is used all over the place (also in other functions), I didn't rename it in this patch.
We could just keep
offsetof(struct bts_buffer, buf[nbuf]), or usestruct_size_t(struct bts_buffer, buf, nbuf)and still benefit from additional compile-time checks, or rename the local variable to structbts_buffer*bts and usestruct_size(bts, buf, nbuf), for example. Any preferences or other ideas?
Quoted 5 lines
To clean up this code before changing it, so that the changes become obvious to review.
Please also split out the annotation for instrumentation, it's separate from any
struct_size()changes, right?
Yes, I'll send a v2 with the __counted_by() annotation and submit a separate patch for struct_size() and other changes.
Quoted 13 lines
Quoted 3 lines
Quoted 1 line
I'm also not sure the code is correct ...
Which part of it?
The size calculation. On a second reading I *think* it's correct, but it's unnecessarily confusing due to the buf<->buf aliasing.
So in a cleaned up version of the code:
- If we name 'struct
bts_buffer' objects 'bb'
- and bb:buf[] is the var-array
- and we rename 'nbuf' to 'nr_buf' (the number of bb:buf[] elements)
Quoted 5 lines
then the code right now does:
bb = kzalloc_node(offsetof(struct bts_buffer, bb[nr_buf]), GFP_KERNEL, node);... which looks correct.
If bb:buf[] is the flexible array, it should be buf[nr_buf] like this:
bb = kzalloc_node(offsetof(struct bts_buffer, buf[nr_buf]), GFP_KERNEL, node);
Thanks,
Thorsten