* Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
@ 2020-06-20 16:22 Matt Pallissard
2020-06-20 18:11 ` Yonghong Song
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-20 16:22 UTC (permalink / raw)
To: bpf
New to bpf here.
I'm trying to read values out of of mm_struct. I have code like this;
unsigned long i[10] = {};
struct task_struct *t;
struct mm_rss_stat *rss;
t = (struct task_struct *)bpf_get_current_task();
BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
BPF_CORE_READ_INTO(i, rss, count);
However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
Any advice or documentation I could sift through would be greatly appreciated. Thanks.
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-20 16:22 Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO Matt Pallissard
@ 2020-06-20 18:11 ` Yonghong Song
2020-06-20 20:06 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Yonghong Song @ 2020-06-20 18:11 UTC (permalink / raw)
To: Matt Pallissard, bpf
On 6/20/20 9:22 AM, Matt Pallissard wrote:
> New to bpf here.
>
> I'm trying to read values out of of mm_struct. I have code like this;
>
> unsigned long i[10] = {};
> struct task_struct *t;
> struct mm_rss_stat *rss;
>
> t = (struct task_struct *)bpf_get_current_task();
> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> BPF_CORE_READ_INTO(i, rss, count);
>
> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
Maybe you want to check the return value of BPF_CORE_READ_INTO.
Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g.,
due to major fault.
>
> Any advice or documentation I could sift through would be greatly appreciated. Thanks.
>
>
> Matt Pallissard
>
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-20 18:11 ` Yonghong Song
@ 2020-06-20 20:06 ` Matt Pallissard
2020-06-21 3:29 ` Andrii Nakryiko
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-20 20:06 UTC (permalink / raw)
To: Yonghong Song; +Cc: bpf
On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
>
>
> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > New to bpf here.
> >
> > I'm trying to read values out of of mm_struct. I have code like this;
> >
> > unsigned long i[10] = {};
> > struct task_struct *t;
> > struct mm_rss_stat *rss;
> >
> > t = (struct task_struct *)bpf_get_current_task();
> > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > BPF_CORE_READ_INTO(i, rss, count);
> >
> > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
>
> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> to major fault.
Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
Thanks a bunch.
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-20 20:06 ` Matt Pallissard
@ 2020-06-21 3:29 ` Andrii Nakryiko
2020-06-21 15:44 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-21 3:29 UTC (permalink / raw)
To: Matt Pallissard; +Cc: Yonghong Song, bpf
On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
>
>
>
>
> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> >
> >
> > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > New to bpf here.
> > >
> > > I'm trying to read values out of of mm_struct. I have code like this;
> > >
> > > unsigned long i[10] = {};
> > > struct task_struct *t;
> > > struct mm_rss_stat *rss;
> > >
> > > t = (struct task_struct *)bpf_get_current_task();
> > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > BPF_CORE_READ_INTO(i, rss, count);
> > >
> > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> >
> > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > to major fault.
>
> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
>
I wrote exactly such piece of code a while ago. Here's part of it for
reference, I think it will be helpful:
struct task_struct *task = (struct task_struct *)bpf_get_current_task();
const struct mm_struct *mm = BPF_CORE_READ(task, mm);
if (mm) {
u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
u64 shmem_pages = BPF_CORE_READ(mm,
rss_stat.count[MM_SHMEMPAGES].counter);
u64 active_rss = file_pages + anon_pages + shmem_pages;
/* ... */
}
> Thanks a bunch.
>
> Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-21 3:29 ` Andrii Nakryiko
@ 2020-06-21 15:44 ` Matt Pallissard
2020-06-22 15:01 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-21 15:44 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> >
> >
> >
> >
> > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > >
> > >
> > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > New to bpf here.
> > > >
> > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > >
> > > > unsigned long i[10] = {};
> > > > struct task_struct *t;
> > > > struct mm_rss_stat *rss;
> > > >
> > > > t = (struct task_struct *)bpf_get_current_task();
> > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > BPF_CORE_READ_INTO(i, rss, count);
> > > >
> > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > >
> > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > to major fault.
> >
> > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> >
>
> I wrote exactly such piece of code a while ago. Here's part of it for
> reference, I think it will be helpful:
>
> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
>
> if (mm) {
> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> u64 shmem_pages = BPF_CORE_READ(mm,
> rss_stat.count[MM_SHMEMPAGES].counter);
> u64 active_rss = file_pages + anon_pages + shmem_pages;
> /* ... */
Thank you,
After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-21 15:44 ` Matt Pallissard
@ 2020-06-22 15:01 ` Matt Pallissard
2020-06-22 16:20 ` Andrii Nakryiko
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-22 15:01 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
>
>
> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > >
> > >
> > >
> > >
> > > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > >
> > > >
> > > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > New to bpf here.
> > > > >
> > > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > > >
> > > > > unsigned long i[10] = {};
> > > > > struct task_struct *t;
> > > > > struct mm_rss_stat *rss;
> > > > >
> > > > > t = (struct task_struct *)bpf_get_current_task();
> > > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > BPF_CORE_READ_INTO(i, rss, count);
> > > > >
> > > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > >
> > > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > to major fault.
> > >
> > > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > >
> >
> > I wrote exactly such piece of code a while ago. Here's part of it for
> > reference, I think it will be helpful:
> >
> > struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> >
> > if (mm) {
> > u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > u64 shmem_pages = BPF_CORE_READ(mm,
> > rss_stat.count[MM_SHMEMPAGES].counter);
> > u64 active_rss = file_pages + anon_pages + shmem_pages;
> > /* ... */
>
> Thank you,
>
> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
>
> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-22 15:01 ` Matt Pallissard
@ 2020-06-22 16:20 ` Andrii Nakryiko
2020-06-22 17:19 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-22 16:20 UTC (permalink / raw)
To: Matt Pallissard; +Cc: Yonghong Song, bpf
On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
>
>
>
> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> >
> >
> > On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > >
> > > >
> > > >
> > > >
> > > > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > >
> > > > >
> > > > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > > New to bpf here.
> > > > > >
> > > > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > > > >
> > > > > > unsigned long i[10] = {};
> > > > > > struct task_struct *t;
> > > > > > struct mm_rss_stat *rss;
> > > > > >
> > > > > > t = (struct task_struct *)bpf_get_current_task();
> > > > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > > BPF_CORE_READ_INTO(i, rss, count);
> > > > > >
> > > > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > >
> > > > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > to major fault.
> > > >
> > > > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > >
> > >
> > > I wrote exactly such piece of code a while ago. Here's part of it for
> > > reference, I think it will be helpful:
> > >
> > > struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > >
> > > if (mm) {
> > > u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > u64 shmem_pages = BPF_CORE_READ(mm,
> > > rss_stat.count[MM_SHMEMPAGES].counter);
> > > u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > /* ... */
> >
> > Thank you,
> >
> > After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> >
> > I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
>
>
> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
Yes, you can't take all those locks from BPF. But reading atomic
counters from BPF should be no problem. You might get a slightly out
of sync readings, but whatever you are doing shouldn't expect to have
100% correct values anyways, because they might change so fast after
you read them.
>
> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
For this, you'd need to iterate across many tasks and aggregate their
results based on tasks's tgid. Check iter/task programs in selftests
(progs/bpf_iter_task.c, I think).
>
>
> Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-22 16:20 ` Andrii Nakryiko
@ 2020-06-22 17:19 ` Matt Pallissard
2020-06-22 22:09 ` Andrii Nakryiko
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-22 17:19 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > > > New to bpf here.
> > > > > > >
> > > > > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > > > > >
> > > > > > > unsigned long i[10] = {};
> > > > > > > struct task_struct *t;
> > > > > > > struct mm_rss_stat *rss;
> > > > > > >
> > > > > > > t = (struct task_struct *)bpf_get_current_task();
> > > > > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > > > BPF_CORE_READ_INTO(i, rss, count);
> > > > > > >
> > > > > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > > >
> > > > > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > > to major fault.
> > > > >
> > > > > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > >
> > > >
> > > > I wrote exactly such piece of code a while ago. Here's part of it for
> > > > reference, I think it will be helpful:
> > > >
> > > > struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > >
> > > > if (mm) {
> > > > u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > u64 shmem_pages = BPF_CORE_READ(mm,
> > > > rss_stat.count[MM_SHMEMPAGES].counter);
> > > > u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > /* ... */
> > >
> > > Thank you,
> > >
> > > After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > >
> > > I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> >
> >
> > I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
>
> Yes, you can't take all those locks from BPF. But reading atomic
> counters from BPF should be no problem. You might get a slightly out
> of sync readings, but whatever you are doing shouldn't expect to have
> 100% correct values anyways, because they might change so fast after
> you read them.
That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
>
> For this, you'd need to iterate across many tasks and aggregate their
> results based on tasks's tgid. Check iter/task programs in selftests
> (progs/bpf_iter_task.c, I think).
Sounds like a great starting point. Thanks again.
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-22 17:19 ` Matt Pallissard
@ 2020-06-22 22:09 ` Andrii Nakryiko
2020-06-23 14:54 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-22 22:09 UTC (permalink / raw)
To: Matt Pallissard; +Cc: Yonghong Song, bpf
On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
>
> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > > On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > > On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > > > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > > > > New to bpf here.
> > > > > > > >
> > > > > > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > > > > > >
> > > > > > > > unsigned long i[10] = {};
> > > > > > > > struct task_struct *t;
> > > > > > > > struct mm_rss_stat *rss;
> > > > > > > >
> > > > > > > > t = (struct task_struct *)bpf_get_current_task();
> > > > > > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > > > > BPF_CORE_READ_INTO(i, rss, count);
> > > > > > > >
> > > > > > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > > > >
> > > > > > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > > > to major fault.
> > > > > >
> > > > > > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > > >
> > > > >
> > > > > I wrote exactly such piece of code a while ago. Here's part of it for
> > > > > reference, I think it will be helpful:
> > > > >
> > > > > struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > > const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > > >
> > > > > if (mm) {
> > > > > u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > > u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > > u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > > u64 shmem_pages = BPF_CORE_READ(mm,
> > > > > rss_stat.count[MM_SHMEMPAGES].counter);
> > > > > u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > > /* ... */
> > > >
> > > > Thank you,
> > > >
> > > > After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > >
> > > > I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > >
> > >
> > > I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> >
> > Yes, you can't take all those locks from BPF. But reading atomic
> > counters from BPF should be no problem. You might get a slightly out
> > of sync readings, but whatever you are doing shouldn't expect to have
> > 100% correct values anyways, because they might change so fast after
> > you read them.
>
> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
>
Not sure about vm_area_struct (where is it in the example above?), but
mm_struct won't go away, because current task won't go away, because
BPF program is running in the context of current. Similarly for
bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
think you don't have to worry about that.
>
> > > If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> >
> > For this, you'd need to iterate across many tasks and aggregate their
> > results based on tasks's tgid. Check iter/task programs in selftests
> > (progs/bpf_iter_task.c, I think).
>
> Sounds like a great starting point. Thanks again.
>
> Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-22 22:09 ` Andrii Nakryiko
@ 2020-06-23 14:54 ` Matt Pallissard
2020-06-23 16:35 ` Yonghong Song
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-23 14:54 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> >
> > On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > > On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > > > On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > > > On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > > > On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > > > > > On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > > > > > New to bpf here.
> > > > > > > > >
> > > > > > > > > I'm trying to read values out of of mm_struct. I have code like this;
> > > > > > > > >
> > > > > > > > > unsigned long i[10] = {};
> > > > > > > > > struct task_struct *t;
> > > > > > > > > struct mm_rss_stat *rss;
> > > > > > > > >
> > > > > > > > > t = (struct task_struct *)bpf_get_current_task();
> > > > > > > > > BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > > > > > BPF_CORE_READ_INTO(i, rss, count);
> > > > > > > > >
> > > > > > > > > However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > > > > >
> > > > > > > > Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > > > > Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > > > > to major fault.
> > > > > > >
> > > > > > > Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > > > >
> > > > > >
> > > > > > I wrote exactly such piece of code a while ago. Here's part of it for
> > > > > > reference, I think it will be helpful:
> > > > > >
> > > > > > struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > > > const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > > > >
> > > > > > if (mm) {
> > > > > > u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > > > u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > > > u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > > > u64 shmem_pages = BPF_CORE_READ(mm,
> > > > > > rss_stat.count[MM_SHMEMPAGES].counter);
> > > > > > u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > > > /* ... */
> > > > >
> > > > > Thank you,
> > > > >
> > > > > After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > > >
> > > > > I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > > >
> > > >
> > > > I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > >
> > > Yes, you can't take all those locks from BPF. But reading atomic
> > > counters from BPF should be no problem. You might get a slightly out
> > > of sync readings, but whatever you are doing shouldn't expect to have
> > > 100% correct values anyways, because they might change so fast after
> > > you read them.
> >
> > That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> >
>
> Not sure about vm_area_struct (where is it in the example above?), but
> mm_struct won't go away, because current task won't go away, because
> BPF program is running in the context of current. Similarly for
> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> think you don't have to worry about that.
I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > > > If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > >
> > > For this, you'd need to iterate across many tasks and aggregate their
> > > results based on tasks's tgid. Check iter/task programs in selftests
> > > (progs/bpf_iter_task.c, I think).
When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
As an aside; is there any documentation for bpf_iter outside of the selftests?
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 14:54 ` Matt Pallissard
@ 2020-06-23 16:35 ` Yonghong Song
2020-06-23 17:58 ` Andrii Nakryiko
0 siblings, 1 reply; 18+ messages in thread
From: Yonghong Song @ 2020-06-23 16:35 UTC (permalink / raw)
To: Matt Pallissard, Andrii Nakryiko; +Cc: bpf
On 6/23/20 7:54 AM, Matt Pallissard wrote:
>
> On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
>> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
>>>
>>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
>>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
>>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
>>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
>>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
>>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
>>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
>>>>>>>>>> New to bpf here.
>>>>>>>>>>
>>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
>>>>>>>>>>
>>>>>>>>>> unsigned long i[10] = {};
>>>>>>>>>> struct task_struct *t;
>>>>>>>>>> struct mm_rss_stat *rss;
>>>>>>>>>>
>>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
>>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
>>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
>>>>>>>>>>
>>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
>>>>>>>>>
>>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
>>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
>>>>>>>>> to major fault.
>>>>>>>>
>>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
>>>>>>>>
>>>>>>>
>>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
>>>>>>> reference, I think it will be helpful:
>>>>>>>
>>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
>>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
>>>>>>>
>>>>>>> if (mm) {
>>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
>>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
>>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
>>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
>>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
>>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
>>>>>>> /* ... */
>>>>>>
>>>>>> Thank you,
>>>>>>
>>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
>>>>>>
>>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
>>>>>
>>>>>
>>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
>>>>
>>>> Yes, you can't take all those locks from BPF. But reading atomic
>>>> counters from BPF should be no problem. You might get a slightly out
>>>> of sync readings, but whatever you are doing shouldn't expect to have
>>>> 100% correct values anyways, because they might change so fast after
>>>> you read them.
>>>
>>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
>>>
>>
>> Not sure about vm_area_struct (where is it in the example above?), but
>> mm_struct won't go away, because current task won't go away, because
>> BPF program is running in the context of current. Similarly for
>> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
>> think you don't have to worry about that.
>
> I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
>
>
>>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
>>>>
>>>> For this, you'd need to iterate across many tasks and aggregate their
>>>> results based on tasks's tgid. Check iter/task programs in selftests
>>>> (progs/bpf_iter_task.c, I think).
>
>
> When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
what is the exact issue. Yes, please mimic what selftests did.
>
> As an aside; is there any documentation for bpf_iter outside of the selftests?
Unfortunately, no. The commit messages of the original patch set might help.
https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
>
> Matt Pallissard
>
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 16:35 ` Yonghong Song
@ 2020-06-23 17:58 ` Andrii Nakryiko
2020-06-23 18:11 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-23 17:58 UTC (permalink / raw)
To: Yonghong Song; +Cc: Matt Pallissard, bpf
On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
>
>
>
> On 6/23/20 7:54 AM, Matt Pallissard wrote:
> >
> > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> >>>
> >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> >>>>>>>>>> New to bpf here.
> >>>>>>>>>>
> >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> >>>>>>>>>>
> >>>>>>>>>> unsigned long i[10] = {};
> >>>>>>>>>> struct task_struct *t;
> >>>>>>>>>> struct mm_rss_stat *rss;
> >>>>>>>>>>
> >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> >>>>>>>>>>
> >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> >>>>>>>>>
> >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> >>>>>>>>> to major fault.
> >>>>>>>>
> >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> >>>>>>>>
> >>>>>>>
> >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> >>>>>>> reference, I think it will be helpful:
> >>>>>>>
> >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> >>>>>>>
> >>>>>>> if (mm) {
> >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> >>>>>>> /* ... */
> >>>>>>
> >>>>>> Thank you,
> >>>>>>
> >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> >>>>>>
> >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> >>>>>
> >>>>>
> >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> >>>>
> >>>> Yes, you can't take all those locks from BPF. But reading atomic
> >>>> counters from BPF should be no problem. You might get a slightly out
> >>>> of sync readings, but whatever you are doing shouldn't expect to have
> >>>> 100% correct values anyways, because they might change so fast after
> >>>> you read them.
> >>>
> >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> >>>
> >>
> >> Not sure about vm_area_struct (where is it in the example above?), but
> >> mm_struct won't go away, because current task won't go away, because
> >> BPF program is running in the context of current. Similarly for
> >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> >> think you don't have to worry about that.
> >
> > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> >
> >
> >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> >>>>
> >>>> For this, you'd need to iterate across many tasks and aggregate their
> >>>> results based on tasks's tgid. Check iter/task programs in selftests
> >>>> (progs/bpf_iter_task.c, I think).
> >
> >
> > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
>
> Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> what is the exact issue. Yes, please mimic what selftests did.
>
It's just an artifact of how libbpf logs error in such case. It did
search for "bpf_iter_task" type, though. But Matt probably doesn't
have a recent enough kernel or didn't build it with
CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
> >
> > As an aside; is there any documentation for bpf_iter outside of the selftests?
>
> Unfortunately, no. The commit messages of the original patch set might help.
> https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
>
> >
> > Matt Pallissard
> >
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 17:58 ` Andrii Nakryiko
@ 2020-06-23 18:11 ` Matt Pallissard
2020-06-23 18:36 ` Andrii Nakryiko
0 siblings, 1 reply; 18+ messages in thread
From: Matt Pallissard @ 2020-06-23 18:11 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
> On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
> >
> >
> >
> > On 6/23/20 7:54 AM, Matt Pallissard wrote:
> > >
> > > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> > >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> > >>>
> > >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > >>>>>>>>>> New to bpf here.
> > >>>>>>>>>>
> > >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> > >>>>>>>>>>
> > >>>>>>>>>> unsigned long i[10] = {};
> > >>>>>>>>>> struct task_struct *t;
> > >>>>>>>>>> struct mm_rss_stat *rss;
> > >>>>>>>>>>
> > >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> > >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> > >>>>>>>>>>
> > >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > >>>>>>>>>
> > >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > >>>>>>>>> to major fault.
> > >>>>>>>>
> > >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > >>>>>>>>
> > >>>>>>>
> > >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> > >>>>>>> reference, I think it will be helpful:
> > >>>>>>>
> > >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > >>>>>>>
> > >>>>>>> if (mm) {
> > >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> > >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> > >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> > >>>>>>> /* ... */
> > >>>>>>
> > >>>>>> Thank you,
> > >>>>>>
> > >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > >>>>>>
> > >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > >>>>>
> > >>>>>
> > >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > >>>>
> > >>>> Yes, you can't take all those locks from BPF. But reading atomic
> > >>>> counters from BPF should be no problem. You might get a slightly out
> > >>>> of sync readings, but whatever you are doing shouldn't expect to have
> > >>>> 100% correct values anyways, because they might change so fast after
> > >>>> you read them.
> > >>>
> > >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > >>>
> > >>
> > >> Not sure about vm_area_struct (where is it in the example above?), but
> > >> mm_struct won't go away, because current task won't go away, because
> > >> BPF program is running in the context of current. Similarly for
> > >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> > >> think you don't have to worry about that.
> > >
> > > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > >
> > >
> > >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > >>>>
> > >>>> For this, you'd need to iterate across many tasks and aggregate their
> > >>>> results based on tasks's tgid. Check iter/task programs in selftests
> > >>>> (progs/bpf_iter_task.c, I think).
> > >
> > >
> > > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
> >
> > Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> > what is the exact issue. Yes, please mimic what selftests did.
> >
>
> It's just an artifact of how libbpf logs error in such case. It did
> search for "bpf_iter_task" type, though. But Matt probably doesn't
> have a recent enough kernel or didn't build it with
> CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
That shouldn't be the case, I generated vmlinux.h from my currently running machine.
I'm using an upstream kernel.
> ~ uname -r
> 5.7.2-arch1-1
Which has the BTF debug info enabled.
> ~ zgrep BTF= /proc/config.gz
> CONFIG_DEBUG_INFO_BTF=y
I assume that it was built with the version of pahole that's in the upstream repos.
> ~ pacman -Ss pahole
> extra/pahole 1.17-1 [installed]
Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
> > > As an aside; is there any documentation for bpf_iter outside of the selftests?
> >
> > Unfortunately, no. The commit messages of the original patch set might help.
> > https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 18:11 ` Matt Pallissard
@ 2020-06-23 18:36 ` Andrii Nakryiko
2020-06-23 22:05 ` Matt Pallissard
0 siblings, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-23 18:36 UTC (permalink / raw)
To: Matt Pallissard; +Cc: Yonghong Song, bpf
On Tue, Jun 23, 2020 at 11:11 AM Matt Pallissard <matt@pallissard.net> wrote:
>
>
> On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
> > On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
> > >
> > >
> > >
> > > On 6/23/20 7:54 AM, Matt Pallissard wrote:
> > > >
> > > > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> > > >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > >>>
> > > >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > > >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > >>>>>>>>>> New to bpf here.
> > > >>>>>>>>>>
> > > >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> > > >>>>>>>>>>
> > > >>>>>>>>>> unsigned long i[10] = {};
> > > >>>>>>>>>> struct task_struct *t;
> > > >>>>>>>>>> struct mm_rss_stat *rss;
> > > >>>>>>>>>>
> > > >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> > > >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> > > >>>>>>>>>>
> > > >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > >>>>>>>>>
> > > >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > >>>>>>>>> to major fault.
> > > >>>>>>>>
> > > >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > >>>>>>>>
> > > >>>>>>>
> > > >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> > > >>>>>>> reference, I think it will be helpful:
> > > >>>>>>>
> > > >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > >>>>>>>
> > > >>>>>>> if (mm) {
> > > >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> > > >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> > > >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > >>>>>>> /* ... */
> > > >>>>>>
> > > >>>>>> Thank you,
> > > >>>>>>
> > > >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > >>>>>>
> > > >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > > >>>>>
> > > >>>>>
> > > >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > > >>>>
> > > >>>> Yes, you can't take all those locks from BPF. But reading atomic
> > > >>>> counters from BPF should be no problem. You might get a slightly out
> > > >>>> of sync readings, but whatever you are doing shouldn't expect to have
> > > >>>> 100% correct values anyways, because they might change so fast after
> > > >>>> you read them.
> > > >>>
> > > >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > > >>>
> > > >>
> > > >> Not sure about vm_area_struct (where is it in the example above?), but
> > > >> mm_struct won't go away, because current task won't go away, because
> > > >> BPF program is running in the context of current. Similarly for
> > > >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> > > >> think you don't have to worry about that.
> > > >
> > > > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > > >
> > > >
> > > >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > > >>>>
> > > >>>> For this, you'd need to iterate across many tasks and aggregate their
> > > >>>> results based on tasks's tgid. Check iter/task programs in selftests
> > > >>>> (progs/bpf_iter_task.c, I think).
> > > >
> > > >
> > > > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
> > >
> > > Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> > > what is the exact issue. Yes, please mimic what selftests did.
> > >
> >
> > It's just an artifact of how libbpf logs error in such case. It did
> > search for "bpf_iter_task" type, though. But Matt probably doesn't
> > have a recent enough kernel or didn't build it with
> > CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
>
> That shouldn't be the case, I generated vmlinux.h from my currently running machine.
>
>
> I'm using an upstream kernel.
> > ~ uname -r
> > 5.7.2-arch1-1
>
> Which has the BTF debug info enabled.
> > ~ zgrep BTF= /proc/config.gz
> > CONFIG_DEBUG_INFO_BTF=y
>
>
> I assume that it was built with the version of pahole that's in the upstream repos.
> > ~ pacman -Ss pahole
> > extra/pahole 1.17-1 [installed]
>
>
> Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
>
Ok, can you show your code (BPF and user-space side) and libbpf debug logs then?
>
> > > > As an aside; is there any documentation for bpf_iter outside of the selftests?
> > >
> > > Unfortunately, no. The commit messages of the original patch set might help.
> > > https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
>
> Matt Pallissard
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 18:36 ` Andrii Nakryiko
@ 2020-06-23 22:05 ` Matt Pallissard
2020-06-23 22:13 ` Andrii Nakryiko
2020-06-23 22:16 ` Yonghong Song
0 siblings, 2 replies; 18+ messages in thread
From: Matt Pallissard @ 2020-06-23 22:05 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-23T11:36:06 -0700, Andrii Nakryiko wrote:
> On Tue, Jun 23, 2020 at 11:11 AM Matt Pallissard <matt@pallissard.net> wrote:
> >
> >
> > On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
> > > On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
> > > >
> > > >
> > > >
> > > > On 6/23/20 7:54 AM, Matt Pallissard wrote:
> > > > >
> > > > > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> > > > >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > >>>
> > > > >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > > > >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > > >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > >>>>>>>>>> New to bpf here.
> > > > >>>>>>>>>>
> > > > >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> > > > >>>>>>>>>>
> > > > >>>>>>>>>> unsigned long i[10] = {};
> > > > >>>>>>>>>> struct task_struct *t;
> > > > >>>>>>>>>> struct mm_rss_stat *rss;
> > > > >>>>>>>>>>
> > > > >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> > > > >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> > > > >>>>>>>>>>
> > > > >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > >>>>>>>>>
> > > > >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > >>>>>>>>> to major fault.
> > > > >>>>>>>>
> > > > >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > >>>>>>>>
> > > > >>>>>>>
> > > > >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> > > > >>>>>>> reference, I think it will be helpful:
> > > > >>>>>>>
> > > > >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > > >>>>>>>
> > > > >>>>>>> if (mm) {
> > > > >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> > > > >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> > > > >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > >>>>>>> /* ... */
> > > > >>>>>>
> > > > >>>>>> Thank you,
> > > > >>>>>>
> > > > >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > > >>>>>>
> > > > >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > > > >>>>>
> > > > >>>>>
> > > > >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > > > >>>>
> > > > >>>> Yes, you can't take all those locks from BPF. But reading atomic
> > > > >>>> counters from BPF should be no problem. You might get a slightly out
> > > > >>>> of sync readings, but whatever you are doing shouldn't expect to have
> > > > >>>> 100% correct values anyways, because they might change so fast after
> > > > >>>> you read them.
> > > > >>>
> > > > >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > > > >>>
> > > > >>
> > > > >> Not sure about vm_area_struct (where is it in the example above?), but
> > > > >> mm_struct won't go away, because current task won't go away, because
> > > > >> BPF program is running in the context of current. Similarly for
> > > > >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> > > > >> think you don't have to worry about that.
> > > > >
> > > > > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > > > >
> > > > >
> > > > >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > > > >>>>
> > > > >>>> For this, you'd need to iterate across many tasks and aggregate their
> > > > >>>> results based on tasks's tgid. Check iter/task programs in selftests
> > > > >>>> (progs/bpf_iter_task.c, I think).
> > > > >
> > > > >
> > > > > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
> > > >
> > > > Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> > > > what is the exact issue. Yes, please mimic what selftests did.
> > > >
> > >
> > > It's just an artifact of how libbpf logs error in such case. It did
> > > search for "bpf_iter_task" type, though. But Matt probably doesn't
> > > have a recent enough kernel or didn't build it with
> > > CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
> >
> > That shouldn't be the case, I generated vmlinux.h from my currently running machine.
> >
> >
> > I'm using an upstream kernel.
> > > ~ uname -r
> > > 5.7.2-arch1-1
> >
> > Which has the BTF debug info enabled.
> > > ~ zgrep BTF= /proc/config.gz
> > > CONFIG_DEBUG_INFO_BTF=y
> >
> >
> > I assume that it was built with the version of pahole that's in the upstream repos.
> > > ~ pacman -Ss pahole
> > > extra/pahole 1.17-1 [installed]
> >
> >
> > Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
> >
>
> Ok, can you show your code (BPF and user-space side) and libbpf debug logs then?
Sure. The userspace section in question is below. I don't make it past `bpf_object__load`. Same userspace code works fine for tracepoints.
struct bpf_program *prog;
struct bpf_object *obj;
char path[] = PT_BPF_OBJECT_DIR;
strcat(&path[strlen(path)], "/test.o");
libbpf_set_print(print_libbpf_log);
obj = bpf_object__open_file(path, NULL);
if (libbpf_get_error(obj))
return 1;
if(!(prog = bpf_object__find_program_by_name(obj, "dump_task")))
goto cleanup;
if (bpf_object__load(obj))
goto cleanup;
I copied the kernel code, only slightly modifying the include statements
#define bpf_iter_meta bpf_iter_meta___not_used
#define bpf_iter__task bpf_iter__task___not_used
#include <vmlinux.h>
#undef bpf_iter_meta
#undef bpf_iter__task
#include <bpf_helpers.h>
#include <bpf_tracing.h>
struct bpf_iter_meta {
struct seq_file *seq;
__u64 session_id;
__u64 seq_num;
} __attribute__((preserve_access_index));
struct bpf_iter__task {
struct bpf_iter_meta *meta;
struct task_struct *task;
} __attribute__((preserve_access_index));
SEC("iter/task")
int dump_task(struct bpf_iter__task *ctx)
{
struct seq_file *seq = ctx->meta->seq;
struct task_struct *task = ctx->task;
if (task == (void *)0) {
BPF_SEQ_PRINTF(seq, " === END ===\n");
return 0;
}
if (ctx->meta->seq_num == 0)
BPF_SEQ_PRINTF(seq, " tgid gid\n");
BPF_SEQ_PRINTF(seq, "%8d %8d\n", task->tgid, task->pid);
return 0;
}
And here is the debug output
libbpf: loading /tmp//usr/lib/pt/bpf/test.o
libbpf: section(1) .strtab, size 277, link 0, flags 0, type=3
libbpf: skip section(1) .strtab
libbpf: section(2) .text, size 0, link 0, flags 6, type=1
libbpf: skip section(2) .text
libbpf: section(3) iter/task, size 320, link 0, flags 6, type=1
libbpf: found program iter/task
libbpf: section(4) .reliter/task, size 48, link 22, flags 0, type=9
libbpf: section(5) .rodata, size 45, link 0, flags 2, type=1
libbpf: section(6) license, size 4, link 0, flags 3, type=1
libbpf: license of /tmp//usr/lib/pt/bpf/test.o is GPL
libbpf: section(7) version, size 4, link 0, flags 3, type=1
libbpf: kernel version of /tmp//usr/lib/pt/bpf/test.o is 5060b
libbpf: section(8) .debug_str, size 135270, link 0, flags 30, type=1
libbpf: skip section(8) .debug_str
libbpf: section(9) .debug_loc, size 124, link 0, flags 0, type=1
libbpf: skip section(9) .debug_loc
libbpf: section(10) .debug_abbrev, size 857, link 0, flags 0, type=1
libbpf: skip section(10) .debug_abbrev
libbpf: section(11) .debug_info, size 224491, link 0, flags 0, type=1
libbpf: skip section(11) .debug_info
libbpf: section(12) .rel.debug_info, size 160, link 22, flags 0, type=9
libbpf: skip relo .rel.debug_info(12) for section(11)
libbpf: section(13) .BTF, size 25711, link 0, flags 0, type=1
libbpf: section(14) .rel.BTF, size 80, link 22, flags 0, type=9
libbpf: skip relo .rel.BTF(14) for section(13)
libbpf: section(15) .BTF.ext, size 348, link 0, flags 0, type=1
libbpf: section(16) .rel.BTF.ext, size 288, link 22, flags 0, type=9
libbpf: skip relo .rel.BTF.ext(16) for section(15)
libbpf: section(17) .debug_frame, size 40, link 0, flags 0, type=1
libbpf: skip section(17) .debug_frame
libbpf: section(18) .rel.debug_frame, size 16, link 22, flags 0, type=9
libbpf: skip relo .rel.debug_frame(18) for section(17)
libbpf: section(19) .debug_line, size 216, link 0, flags 0, type=1
libbpf: skip section(19) .debug_line
libbpf: section(20) .rel.debug_line, size 16, link 22, flags 0, type=9
libbpf: skip relo .rel.debug_line(20) for section(19)
libbpf: section(21) .llvm_addrsig, size 6, link 22, flags 80000000, type=1879002115
libbpf: skip section(21) .llvm_addrsig
libbpf: section(22) .symtab, size 312, link 1, flags 0, type=2
libbpf: looking for externs among 13 symbols...
libbpf: collected 0 externs total
libbpf: map 'test.rodata' (global data): at sec_idx 5, offset 0, flags 480.
libbpf: map 0 is "test.rodata"
libbpf: collecting relocating info for: 'iter/task'
libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 7
libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 7
libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 17
libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 17
libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 33
libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 33
libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
libbpf: map 'test.rodata': created successfully, fd=4
libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
libbpf: prog 'iter/task': performing 6 CO-RE offset relocs
libbpf: prog 'iter/task': relo #0: kind 0, spec is [2] bpf_iter__task + 0:0 => 0.0 @ &x[0].meta
libbpf: prog 'iter/task': relo #0: no matching targets found for [2] bpf_iter__task + 0:0
libbpf: prog 'iter/task': relo #0: substituting insn #0 w/ invalid insn
libbpf: prog 'iter/task': relo #1: kind 0, spec is [8] bpf_iter_meta + 0:0 => 0.0 @ &x[0].seq
libbpf: prog 'iter/task': relo #1: no matching targets found for [8] bpf_iter_meta + 0:0
libbpf: prog 'iter/task': relo #1: substituting insn #1 w/ invalid insn
libbpf: prog 'iter/task': relo #2: kind 0, spec is [2] bpf_iter__task + 0:1 => 8.0 @ &x[0].task
libbpf: prog 'iter/task': relo #2: no matching targets found for [2] bpf_iter__task + 0:1
libbpf: prog 'iter/task': relo #2: substituting insn #2 w/ invalid insn
libbpf: prog 'iter/task': relo #3: kind 0, spec is [8] bpf_iter_meta + 0:2 => 16.0 @ &x[0].seq_num
libbpf: prog 'iter/task': relo #3: no matching targets found for [8] bpf_iter_meta + 0:2
libbpf: prog 'iter/task': relo #3: substituting insn #12 w/ invalid insn
libbpf: prog 'iter/task': relo #4: kind 0, spec is [12] task_struct + 0:71 => 1292.0 @ &x[0].tgid
libbpf: [12] task_struct: found candidate [115] task_struct
libbpf: prog 'iter/task': relo #4: matching candidate #0 task_struct against spec [115] task_struct + 0:71 => 1292.0 @ &x[0].tgid: 1
libbpf: prog 'iter/task': relo #4: patched insn #22 (LDX/ST/STX) off 1292 -> 1292
libbpf: prog 'iter/task': relo #5: kind 0, spec is [12] task_struct + 0:70 => 1288.0 @ &x[0].pid
libbpf: prog 'iter/task': relo #5: matching candidate #0 task_struct against spec [115] task_struct + 0:70 => 1288.0 @ &x[0].pid: 1
libbpf: prog 'iter/task': relo #5: patched insn #26 (LDX/ST/STX) off 1288 -> 1288
libbpf: task is not found in vmlinux BTF
libbpf: failed to load object '/tmp//usr/lib/pt/bpf/test.o'
*** stack smashing detected ***: terminated <-- I assume this is because I'm not handling my errors and cleaning up properly
Aborted (core dumped)
> > > > > As an aside; is there any documentation for bpf_iter outside of the selftests?
> > > >
> > > > Unfortunately, no. The commit messages of the original patch set might help.
> > > > https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 22:05 ` Matt Pallissard
@ 2020-06-23 22:13 ` Andrii Nakryiko
2020-06-24 15:51 ` Matt Pallissard
2020-06-23 22:16 ` Yonghong Song
1 sibling, 1 reply; 18+ messages in thread
From: Andrii Nakryiko @ 2020-06-23 22:13 UTC (permalink / raw)
To: Matt Pallissard; +Cc: Yonghong Song, bpf
On Tue, Jun 23, 2020 at 3:05 PM Matt Pallissard <matt@pallissard.net> wrote:
>
>
>
> On 2020-06-23T11:36:06 -0700, Andrii Nakryiko wrote:
> > On Tue, Jun 23, 2020 at 11:11 AM Matt Pallissard <matt@pallissard.net> wrote:
> > >
> > >
> > > On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
> > > > On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
> > > > >
> > > > >
> > > > >
> > > > > On 6/23/20 7:54 AM, Matt Pallissard wrote:
> > > > > >
> > > > > > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> > > > > >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > >>>
> > > > > >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > > > > >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > > > >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > > >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > > >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > >>>>>>>>>> New to bpf here.
> > > > > >>>>>>>>>>
> > > > > >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> > > > > >>>>>>>>>>
> > > > > >>>>>>>>>> unsigned long i[10] = {};
> > > > > >>>>>>>>>> struct task_struct *t;
> > > > > >>>>>>>>>> struct mm_rss_stat *rss;
> > > > > >>>>>>>>>>
> > > > > >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> > > > > >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> > > > > >>>>>>>>>>
> > > > > >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > > >>>>>>>>>
> > > > > >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > >>>>>>>>> to major fault.
> > > > > >>>>>>>>
> > > > > >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > > >>>>>>>>
> > > > > >>>>>>>
> > > > > >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> > > > > >>>>>>> reference, I think it will be helpful:
> > > > > >>>>>>>
> > > > > >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > > >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > > > >>>>>>>
> > > > > >>>>>>> if (mm) {
> > > > > >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > > >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > > >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > > >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> > > > > >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> > > > > >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > > >>>>>>> /* ... */
> > > > > >>>>>>
> > > > > >>>>>> Thank you,
> > > > > >>>>>>
> > > > > >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > > > >>>>>>
> > > > > >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > > > > >>>>>
> > > > > >>>>>
> > > > > >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > > > > >>>>
> > > > > >>>> Yes, you can't take all those locks from BPF. But reading atomic
> > > > > >>>> counters from BPF should be no problem. You might get a slightly out
> > > > > >>>> of sync readings, but whatever you are doing shouldn't expect to have
> > > > > >>>> 100% correct values anyways, because they might change so fast after
> > > > > >>>> you read them.
> > > > > >>>
> > > > > >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > > > > >>>
> > > > > >>
> > > > > >> Not sure about vm_area_struct (where is it in the example above?), but
> > > > > >> mm_struct won't go away, because current task won't go away, because
> > > > > >> BPF program is running in the context of current. Similarly for
> > > > > >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> > > > > >> think you don't have to worry about that.
> > > > > >
> > > > > > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > > > > >
> > > > > >
> > > > > >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > > > > >>>>
> > > > > >>>> For this, you'd need to iterate across many tasks and aggregate their
> > > > > >>>> results based on tasks's tgid. Check iter/task programs in selftests
> > > > > >>>> (progs/bpf_iter_task.c, I think).
> > > > > >
> > > > > >
> > > > > > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
> > > > >
> > > > > Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> > > > > what is the exact issue. Yes, please mimic what selftests did.
> > > > >
> > > >
> > > > It's just an artifact of how libbpf logs error in such case. It did
> > > > search for "bpf_iter_task" type, though. But Matt probably doesn't
> > > > have a recent enough kernel or didn't build it with
> > > > CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
> > >
> > > That shouldn't be the case, I generated vmlinux.h from my currently running machine.
> > >
> > >
> > > I'm using an upstream kernel.
> > > > ~ uname -r
> > > > 5.7.2-arch1-1
> > >
> > > Which has the BTF debug info enabled.
> > > > ~ zgrep BTF= /proc/config.gz
> > > > CONFIG_DEBUG_INFO_BTF=y
> > >
> > >
> > > I assume that it was built with the version of pahole that's in the upstream repos.
> > > > ~ pacman -Ss pahole
> > > > extra/pahole 1.17-1 [installed]
> > >
> > >
> > > Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
> > >
> >
> > Ok, can you show your code (BPF and user-space side) and libbpf debug logs then?
>
>
> Sure. The userspace section in question is below. I don't make it past `bpf_object__load`. Same userspace code works fine for tracepoints.
>
> struct bpf_program *prog;
> struct bpf_object *obj;
> char path[] = PT_BPF_OBJECT_DIR;
> strcat(&path[strlen(path)], "/test.o");
>
> libbpf_set_print(print_libbpf_log);
>
> obj = bpf_object__open_file(path, NULL);
> if (libbpf_get_error(obj))
> return 1;
>
> if(!(prog = bpf_object__find_program_by_name(obj, "dump_task")))
> goto cleanup;
>
> if (bpf_object__load(obj))
> goto cleanup;
>
>
> I copied the kernel code, only slightly modifying the include statements
>
> #define bpf_iter_meta bpf_iter_meta___not_used
> #define bpf_iter__task bpf_iter__task___not_used
> #include <vmlinux.h>
> #undef bpf_iter_meta
> #undef bpf_iter__task
> #include <bpf_helpers.h>
> #include <bpf_tracing.h>
>
> struct bpf_iter_meta {
> struct seq_file *seq;
> __u64 session_id;
> __u64 seq_num;
> } __attribute__((preserve_access_index));
>
> struct bpf_iter__task {
> struct bpf_iter_meta *meta;
> struct task_struct *task;
> } __attribute__((preserve_access_index));
>
> SEC("iter/task")
> int dump_task(struct bpf_iter__task *ctx)
> {
> struct seq_file *seq = ctx->meta->seq;
> struct task_struct *task = ctx->task;
>
> if (task == (void *)0) {
> BPF_SEQ_PRINTF(seq, " === END ===\n");
> return 0;
> }
>
> if (ctx->meta->seq_num == 0)
> BPF_SEQ_PRINTF(seq, " tgid gid\n");
>
> BPF_SEQ_PRINTF(seq, "%8d %8d\n", task->tgid, task->pid);
> return 0;
> }
>
>
> And here is the debug output
>
>
> libbpf: loading /tmp//usr/lib/pt/bpf/test.o
> libbpf: section(1) .strtab, size 277, link 0, flags 0, type=3
> libbpf: skip section(1) .strtab
> libbpf: section(2) .text, size 0, link 0, flags 6, type=1
> libbpf: skip section(2) .text
> libbpf: section(3) iter/task, size 320, link 0, flags 6, type=1
> libbpf: found program iter/task
> libbpf: section(4) .reliter/task, size 48, link 22, flags 0, type=9
> libbpf: section(5) .rodata, size 45, link 0, flags 2, type=1
> libbpf: section(6) license, size 4, link 0, flags 3, type=1
> libbpf: license of /tmp//usr/lib/pt/bpf/test.o is GPL
> libbpf: section(7) version, size 4, link 0, flags 3, type=1
> libbpf: kernel version of /tmp//usr/lib/pt/bpf/test.o is 5060b
> libbpf: section(8) .debug_str, size 135270, link 0, flags 30, type=1
> libbpf: skip section(8) .debug_str
> libbpf: section(9) .debug_loc, size 124, link 0, flags 0, type=1
> libbpf: skip section(9) .debug_loc
> libbpf: section(10) .debug_abbrev, size 857, link 0, flags 0, type=1
> libbpf: skip section(10) .debug_abbrev
> libbpf: section(11) .debug_info, size 224491, link 0, flags 0, type=1
> libbpf: skip section(11) .debug_info
> libbpf: section(12) .rel.debug_info, size 160, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_info(12) for section(11)
> libbpf: section(13) .BTF, size 25711, link 0, flags 0, type=1
> libbpf: section(14) .rel.BTF, size 80, link 22, flags 0, type=9
> libbpf: skip relo .rel.BTF(14) for section(13)
> libbpf: section(15) .BTF.ext, size 348, link 0, flags 0, type=1
> libbpf: section(16) .rel.BTF.ext, size 288, link 22, flags 0, type=9
> libbpf: skip relo .rel.BTF.ext(16) for section(15)
> libbpf: section(17) .debug_frame, size 40, link 0, flags 0, type=1
> libbpf: skip section(17) .debug_frame
> libbpf: section(18) .rel.debug_frame, size 16, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_frame(18) for section(17)
> libbpf: section(19) .debug_line, size 216, link 0, flags 0, type=1
> libbpf: skip section(19) .debug_line
> libbpf: section(20) .rel.debug_line, size 16, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_line(20) for section(19)
> libbpf: section(21) .llvm_addrsig, size 6, link 22, flags 80000000, type=1879002115
> libbpf: skip section(21) .llvm_addrsig
> libbpf: section(22) .symtab, size 312, link 1, flags 0, type=2
> libbpf: looking for externs among 13 symbols...
> libbpf: collected 0 externs total
> libbpf: map 'test.rodata' (global data): at sec_idx 5, offset 0, flags 480.
> libbpf: map 0 is "test.rodata"
> libbpf: collecting relocating info for: 'iter/task'
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 7
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 7
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 17
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 17
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 33
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 33
> libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> libbpf: map 'test.rodata': created successfully, fd=4
> libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> libbpf: prog 'iter/task': performing 6 CO-RE offset relocs
> libbpf: prog 'iter/task': relo #0: kind 0, spec is [2] bpf_iter__task + 0:0 => 0.0 @ &x[0].meta
> libbpf: prog 'iter/task': relo #0: no matching targets found for [2] bpf_iter__task + 0:0
> libbpf: prog 'iter/task': relo #0: substituting insn #0 w/ invalid insn
> libbpf: prog 'iter/task': relo #1: kind 0, spec is [8] bpf_iter_meta + 0:0 => 0.0 @ &x[0].seq
> libbpf: prog 'iter/task': relo #1: no matching targets found for [8] bpf_iter_meta + 0:0
> libbpf: prog 'iter/task': relo #1: substituting insn #1 w/ invalid insn
> libbpf: prog 'iter/task': relo #2: kind 0, spec is [2] bpf_iter__task + 0:1 => 8.0 @ &x[0].task
> libbpf: prog 'iter/task': relo #2: no matching targets found for [2] bpf_iter__task + 0:1
> libbpf: prog 'iter/task': relo #2: substituting insn #2 w/ invalid insn
see all these "substituting insn w/ invalid insn" messages? Your
kernel doesn't have bpf_iter__task struct in it.
You can confirm by running:
$ bpftool btf dump file /sys/kernel/btf/vmlinux | grep bpf_iter_
You said you have 5.7 kernel. Isn't bpf_iter available starting from 5.8?
> libbpf: prog 'iter/task': relo #3: kind 0, spec is [8] bpf_iter_meta + 0:2 => 16.0 @ &x[0].seq_num
> libbpf: prog 'iter/task': relo #3: no matching targets found for [8] bpf_iter_meta + 0:2
> libbpf: prog 'iter/task': relo #3: substituting insn #12 w/ invalid insn
> libbpf: prog 'iter/task': relo #4: kind 0, spec is [12] task_struct + 0:71 => 1292.0 @ &x[0].tgid
> libbpf: [12] task_struct: found candidate [115] task_struct
> libbpf: prog 'iter/task': relo #4: matching candidate #0 task_struct against spec [115] task_struct + 0:71 => 1292.0 @ &x[0].tgid: 1
> libbpf: prog 'iter/task': relo #4: patched insn #22 (LDX/ST/STX) off 1292 -> 1292
> libbpf: prog 'iter/task': relo #5: kind 0, spec is [12] task_struct + 0:70 => 1288.0 @ &x[0].pid
> libbpf: prog 'iter/task': relo #5: matching candidate #0 task_struct against spec [115] task_struct + 0:70 => 1288.0 @ &x[0].pid: 1
> libbpf: prog 'iter/task': relo #5: patched insn #26 (LDX/ST/STX) off 1288 -> 1288
> libbpf: task is not found in vmlinux BTF
> libbpf: failed to load object '/tmp//usr/lib/pt/bpf/test.o'
> *** stack smashing detected ***: terminated <-- I assume this is because I'm not handling my errors and cleaning up properly
> Aborted (core dumped)
>
> > > > > > As an aside; is there any documentation for bpf_iter outside of the selftests?
> > > > >
> > > > > Unfortunately, no. The commit messages of the original patch set might help.
> > > > > https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 22:05 ` Matt Pallissard
2020-06-23 22:13 ` Andrii Nakryiko
@ 2020-06-23 22:16 ` Yonghong Song
1 sibling, 0 replies; 18+ messages in thread
From: Yonghong Song @ 2020-06-23 22:16 UTC (permalink / raw)
To: Matt Pallissard, Andrii Nakryiko; +Cc: bpf
On 6/23/20 3:05 PM, Matt Pallissard wrote:
>
>
> On 2020-06-23T11:36:06 -0700, Andrii Nakryiko wrote:
>> On Tue, Jun 23, 2020 at 11:11 AM Matt Pallissard <matt@pallissard.net> wrote:
>>>
>>>
>>> On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
>>>> On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
>>>>>
>>>>>
>>>>>
>>>>> On 6/23/20 7:54 AM, Matt Pallissard wrote:
>>>>>>
>>>>>> On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
>>>>>>> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
>>>>>>>>
>>>>>>>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
>>>>>>>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
>>>>>>>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
>>>>>>>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
>>>>>>>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
>>>>>>>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
>>>>>>>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
>>>>>>>>>>>>>>> New to bpf here.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> unsigned long i[10] = {};
>>>>>>>>>>>>>>> struct task_struct *t;
>>>>>>>>>>>>>>> struct mm_rss_stat *rss;
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
>>>>>>>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
>>>>>>>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
>>>>>>>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
>>>>>>>>>>>>>> to major fault.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
>>>>>>>>>>>> reference, I think it will be helpful:
>>>>>>>>>>>>
>>>>>>>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
>>>>>>>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
>>>>>>>>>>>>
>>>>>>>>>>>> if (mm) {
>>>>>>>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
>>>>>>>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
>>>>>>>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
>>>>>>>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
>>>>>>>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
>>>>>>>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
>>>>>>>>>>>> /* ... */
>>>>>>>>>>>
>>>>>>>>>>> Thank you,
>>>>>>>>>>>
>>>>>>>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
>>>>>>>>>>>
>>>>>>>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
>>>>>>>>>
>>>>>>>>> Yes, you can't take all those locks from BPF. But reading atomic
>>>>>>>>> counters from BPF should be no problem. You might get a slightly out
>>>>>>>>> of sync readings, but whatever you are doing shouldn't expect to have
>>>>>>>>> 100% correct values anyways, because they might change so fast after
>>>>>>>>> you read them.
>>>>>>>>
>>>>>>>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
>>>>>>>>
>>>>>>>
>>>>>>> Not sure about vm_area_struct (where is it in the example above?), but
>>>>>>> mm_struct won't go away, because current task won't go away, because
>>>>>>> BPF program is running in the context of current. Similarly for
>>>>>>> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
>>>>>>> think you don't have to worry about that.
>>>>>>
>>>>>> I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
>>>>>>
>>>>>>
>>>>>>>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
>>>>>>>>>
>>>>>>>>> For this, you'd need to iterate across many tasks and aggregate their
>>>>>>>>> results based on tasks's tgid. Check iter/task programs in selftests
>>>>>>>>> (progs/bpf_iter_task.c, I think).
>>>>>>
>>>>>>
>>>>>> When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
>>>>>
>>>>> Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
>>>>> what is the exact issue. Yes, please mimic what selftests did.
>>>>>
>>>>
>>>> It's just an artifact of how libbpf logs error in such case. It did
>>>> search for "bpf_iter_task" type, though. But Matt probably doesn't
>>>> have a recent enough kernel or didn't build it with
>>>> CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
>>>
>>> That shouldn't be the case, I generated vmlinux.h from my currently running machine.
>>>
>>>
>>> I'm using an upstream kernel.
>>>> ~ uname -r
>>>> 5.7.2-arch1-1
>>>
>>> Which has the BTF debug info enabled.
>>>> ~ zgrep BTF= /proc/config.gz
>>>> CONFIG_DEBUG_INFO_BTF=y
>>>
>>>
>>> I assume that it was built with the version of pahole that's in the upstream repos.
>>>> ~ pacman -Ss pahole
>>>> extra/pahole 1.17-1 [installed]
>>>
>>>
>>> Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
>>>
>>
>> Ok, can you show your code (BPF and user-space side) and libbpf debug logs then?
>
>
> Sure. The userspace section in question is below. I don't make it past `bpf_object__load`. Same userspace code works fine for tracepoints.
>
> struct bpf_program *prog;
> struct bpf_object *obj;
> char path[] = PT_BPF_OBJECT_DIR;
> strcat(&path[strlen(path)], "/test.o");
>
> libbpf_set_print(print_libbpf_log);
>
> obj = bpf_object__open_file(path, NULL);
> if (libbpf_get_error(obj))
> return 1;
>
> if(!(prog = bpf_object__find_program_by_name(obj, "dump_task")))
> goto cleanup;
>
> if (bpf_object__load(obj))
> goto cleanup;
>
>
> I copied the kernel code, only slightly modifying the include statements
>
> #define bpf_iter_meta bpf_iter_meta___not_used
> #define bpf_iter__task bpf_iter__task___not_used
> #include <vmlinux.h>
> #undef bpf_iter_meta
> #undef bpf_iter__task
> #include <bpf_helpers.h>
> #include <bpf_tracing.h>
>
> struct bpf_iter_meta {
> struct seq_file *seq;
> __u64 session_id;
> __u64 seq_num;
> } __attribute__((preserve_access_index));
>
> struct bpf_iter__task {
> struct bpf_iter_meta *meta;
> struct task_struct *task;
> } __attribute__((preserve_access_index));
>
> SEC("iter/task")
> int dump_task(struct bpf_iter__task *ctx)
> {
> struct seq_file *seq = ctx->meta->seq;
> struct task_struct *task = ctx->task;
>
> if (task == (void *)0) {
> BPF_SEQ_PRINTF(seq, " === END ===\n");
> return 0;
> }
>
> if (ctx->meta->seq_num == 0)
> BPF_SEQ_PRINTF(seq, " tgid gid\n");
>
> BPF_SEQ_PRINTF(seq, "%8d %8d\n", task->tgid, task->pid);
> return 0;
> }
>
>
> And here is the debug output
>
>
> libbpf: loading /tmp//usr/lib/pt/bpf/test.o
> libbpf: section(1) .strtab, size 277, link 0, flags 0, type=3
> libbpf: skip section(1) .strtab
> libbpf: section(2) .text, size 0, link 0, flags 6, type=1
> libbpf: skip section(2) .text
> libbpf: section(3) iter/task, size 320, link 0, flags 6, type=1
> libbpf: found program iter/task
> libbpf: section(4) .reliter/task, size 48, link 22, flags 0, type=9
> libbpf: section(5) .rodata, size 45, link 0, flags 2, type=1
> libbpf: section(6) license, size 4, link 0, flags 3, type=1
> libbpf: license of /tmp//usr/lib/pt/bpf/test.o is GPL
> libbpf: section(7) version, size 4, link 0, flags 3, type=1
> libbpf: kernel version of /tmp//usr/lib/pt/bpf/test.o is 5060b
> libbpf: section(8) .debug_str, size 135270, link 0, flags 30, type=1
> libbpf: skip section(8) .debug_str
> libbpf: section(9) .debug_loc, size 124, link 0, flags 0, type=1
> libbpf: skip section(9) .debug_loc
> libbpf: section(10) .debug_abbrev, size 857, link 0, flags 0, type=1
> libbpf: skip section(10) .debug_abbrev
> libbpf: section(11) .debug_info, size 224491, link 0, flags 0, type=1
> libbpf: skip section(11) .debug_info
> libbpf: section(12) .rel.debug_info, size 160, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_info(12) for section(11)
> libbpf: section(13) .BTF, size 25711, link 0, flags 0, type=1
> libbpf: section(14) .rel.BTF, size 80, link 22, flags 0, type=9
> libbpf: skip relo .rel.BTF(14) for section(13)
> libbpf: section(15) .BTF.ext, size 348, link 0, flags 0, type=1
> libbpf: section(16) .rel.BTF.ext, size 288, link 22, flags 0, type=9
> libbpf: skip relo .rel.BTF.ext(16) for section(15)
> libbpf: section(17) .debug_frame, size 40, link 0, flags 0, type=1
> libbpf: skip section(17) .debug_frame
> libbpf: section(18) .rel.debug_frame, size 16, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_frame(18) for section(17)
> libbpf: section(19) .debug_line, size 216, link 0, flags 0, type=1
> libbpf: skip section(19) .debug_line
> libbpf: section(20) .rel.debug_line, size 16, link 22, flags 0, type=9
> libbpf: skip relo .rel.debug_line(20) for section(19)
> libbpf: section(21) .llvm_addrsig, size 6, link 22, flags 80000000, type=1879002115
> libbpf: skip section(21) .llvm_addrsig
> libbpf: section(22) .symtab, size 312, link 1, flags 0, type=2
> libbpf: looking for externs among 13 symbols...
> libbpf: collected 0 externs total
> libbpf: map 'test.rodata' (global data): at sec_idx 5, offset 0, flags 480.
> libbpf: map 0 is "test.rodata"
> libbpf: collecting relocating info for: 'iter/task'
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 7
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 7
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 17
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 17
> libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 33
> libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 33
> libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> libbpf: map 'test.rodata': created successfully, fd=4
> libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> libbpf: prog 'iter/task': performing 6 CO-RE offset relocs
> libbpf: prog 'iter/task': relo #0: kind 0, spec is [2] bpf_iter__task + 0:0 => 0.0 @ &x[0].meta
> libbpf: prog 'iter/task': relo #0: no matching targets found for [2] bpf_iter__task + 0:0
> libbpf: prog 'iter/task': relo #0: substituting insn #0 w/ invalid insn
libbpf didn't find bpf_iter__task type in the vmlinux BTF.
The following are possible locations libbpf tries to find vmlinux:
/* try canonical vmlinux BTF through sysfs first */
{ "/sys/kernel/btf/vmlinux", true /* raw BTF */ },
/* fall back to trying to find vmlinux ELF on disk
otherwise */
{ "/boot/vmlinux-%1$s" },
{ "/lib/modules/%1$s/vmlinux-%1$s" },
{ "/lib/modules/%1$s/build/vmlinux" },
{ "/usr/lib/modules/%1$s/kernel/vmlinux" },
{ "/usr/lib/debug/boot/vmlinux-%1$s" },
{ "/usr/lib/debug/boot/vmlinux-%1$s.debug" },
{ "/usr/lib/debug/lib/modules/%1$s/vmlinux" },
Could you check based on the order of the above matching whether you
vmlinux btf does have type bpf_iter__task?
You can use
bpftool btf dump file /.../vmlinux
to see whether your vmlinux has bpf_iter__task type or not.
> libbpf: prog 'iter/task': relo #1: kind 0, spec is [8] bpf_iter_meta + 0:0 => 0.0 @ &x[0].seq
> libbpf: prog 'iter/task': relo #1: no matching targets found for [8] bpf_iter_meta + 0:0
> libbpf: prog 'iter/task': relo #1: substituting insn #1 w/ invalid insn
> libbpf: prog 'iter/task': relo #2: kind 0, spec is [2] bpf_iter__task + 0:1 => 8.0 @ &x[0].task
> libbpf: prog 'iter/task': relo #2: no matching targets found for [2] bpf_iter__task + 0:1
> libbpf: prog 'iter/task': relo #2: substituting insn #2 w/ invalid insn
> libbpf: prog 'iter/task': relo #3: kind 0, spec is [8] bpf_iter_meta + 0:2 => 16.0 @ &x[0].seq_num
> libbpf: prog 'iter/task': relo #3: no matching targets found for [8] bpf_iter_meta + 0:2
> libbpf: prog 'iter/task': relo #3: substituting insn #12 w/ invalid insn
> libbpf: prog 'iter/task': relo #4: kind 0, spec is [12] task_struct + 0:71 => 1292.0 @ &x[0].tgid
> libbpf: [12] task_struct: found candidate [115] task_struct
> libbpf: prog 'iter/task': relo #4: matching candidate #0 task_struct against spec [115] task_struct + 0:71 => 1292.0 @ &x[0].tgid: 1
> libbpf: prog 'iter/task': relo #4: patched insn #22 (LDX/ST/STX) off 1292 -> 1292
> libbpf: prog 'iter/task': relo #5: kind 0, spec is [12] task_struct + 0:70 => 1288.0 @ &x[0].pid
> libbpf: prog 'iter/task': relo #5: matching candidate #0 task_struct against spec [115] task_struct + 0:70 => 1288.0 @ &x[0].pid: 1
> libbpf: prog 'iter/task': relo #5: patched insn #26 (LDX/ST/STX) off 1288 -> 1288
> libbpf: task is not found in vmlinux BTF
> libbpf: failed to load object '/tmp//usr/lib/pt/bpf/test.o'
> *** stack smashing detected ***: terminated <-- I assume this is because I'm not handling my errors and cleaning up properly
> Aborted (core dumped)
>
>>>>>> As an aside; is there any documentation for bpf_iter outside of the selftests?
>>>>>
>>>>> Unfortunately, no. The commit messages of the original patch set might help.
>>>>> https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
^ permalink raw reply [flat|nested] 18+ messages in thread
* Re: Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO
2020-06-23 22:13 ` Andrii Nakryiko
@ 2020-06-24 15:51 ` Matt Pallissard
0 siblings, 0 replies; 18+ messages in thread
From: Matt Pallissard @ 2020-06-24 15:51 UTC (permalink / raw)
To: Andrii Nakryiko; +Cc: Yonghong Song, bpf
On 2020-06-23T15:13:28 -0700, Andrii Nakryiko wrote:
> On Tue, Jun 23, 2020 at 3:05 PM Matt Pallissard <matt@pallissard.net> wrote:
> >
> >
> >
> > On 2020-06-23T11:36:06 -0700, Andrii Nakryiko wrote:
> > > On Tue, Jun 23, 2020 at 11:11 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > >
> > > >
> > > > On 2020-06-23T10:58:20 -0700, Andrii Nakryiko wrote:
> > > > > On Tue, Jun 23, 2020 at 9:36 AM Yonghong Song <yhs@fb.com> wrote:
> > > > > >
> > > > > >
> > > > > >
> > > > > > On 6/23/20 7:54 AM, Matt Pallissard wrote:
> > > > > > >
> > > > > > > On 2020-06-22T15:09:57 -0700, Andrii Nakryiko wrote:
> > > > > > >> On Mon, Jun 22, 2020 at 10:19 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > > >>>
> > > > > > >>> On 2020-06-22T09:20:03 -0700, Andrii Nakryiko wrote:
> > > > > > >>>> On Mon, Jun 22, 2020 at 8:01 AM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > > >>>>> On 2020-06-21T08:44:28 -0700, Matt Pallissard wrote:
> > > > > > >>>>>> On 2020-06-20T20:29:43 -0700, Andrii Nakryiko wrote:
> > > > > > >>>>>>> On Sat, Jun 20, 2020 at 1:07 PM Matt Pallissard <matt@pallissard.net> wrote:
> > > > > > >>>>>>>> On 2020-06-20T11:11:55 -0700, Yonghong Song wrote:
> > > > > > >>>>>>>>> On 6/20/20 9:22 AM, Matt Pallissard wrote:
> > > > > > >>>>>>>>>> New to bpf here.
> > > > > > >>>>>>>>>>
> > > > > > >>>>>>>>>> I'm trying to read values out of of mm_struct. I have code like this;
> > > > > > >>>>>>>>>>
> > > > > > >>>>>>>>>> unsigned long i[10] = {};
> > > > > > >>>>>>>>>> struct task_struct *t;
> > > > > > >>>>>>>>>> struct mm_rss_stat *rss;
> > > > > > >>>>>>>>>>
> > > > > > >>>>>>>>>> t = (struct task_struct *)bpf_get_current_task();
> > > > > > >>>>>>>>>> BPF_CORE_READ_INTO(&rss, t, mm, rss_stat);
> > > > > > >>>>>>>>>> BPF_CORE_READ_INTO(i, rss, count);
> > > > > > >>>>>>>>>>
> > > > > > >>>>>>>>>> However, all values in `i` appear to be 0 (i[MM_FILEPAGES], etc), as if no data gets copied. I'm about 100% confident that this is caused by a glaring oversight on my part.
> > > > > > >>>>>>>>>
> > > > > > >>>>>>>>> Maybe you want to check the return value of BPF_CORE_READ_INTO.
> > > > > > >>>>>>>>> Underlying it is using bpf_probe_read and bpf_probe_read may fail e.g., due
> > > > > > >>>>>>>>> to major fault.
> > > > > > >>>>>>>>
> > > > > > >>>>>>>> Doh, I should have known to check the return codes! Yes, it was failing. I knew I was overlooking something trivial.
> > > > > > >>>>>>>>
> > > > > > >>>>>>>
> > > > > > >>>>>>> I wrote exactly such piece of code a while ago. Here's part of it for
> > > > > > >>>>>>> reference, I think it will be helpful:
> > > > > > >>>>>>>
> > > > > > >>>>>>> struct task_struct *task = (struct task_struct *)bpf_get_current_task();
> > > > > > >>>>>>> const struct mm_struct *mm = BPF_CORE_READ(task, mm);
> > > > > > >>>>>>>
> > > > > > >>>>>>> if (mm) {
> > > > > > >>>>>>> u64 hiwater_rss = BPF_CORE_READ(mm, hiwater_rss);
> > > > > > >>>>>>> u64 file_pages = BPF_CORE_READ(mm, rss_stat.count[MM_FILEPAGES].counter);
> > > > > > >>>>>>> u64 anon_pages = BPF_CORE_READ(mm, rss_stat.count[MM_ANONPAGES].counter);
> > > > > > >>>>>>> u64 shmem_pages = BPF_CORE_READ(mm,
> > > > > > >>>>>>> rss_stat.count[MM_SHMEMPAGES].counter);
> > > > > > >>>>>>> u64 active_rss = file_pages + anon_pages + shmem_pages;
> > > > > > >>>>>>> /* ... */
> > > > > > >>>>>>
> > > > > > >>>>>> Thank you,
> > > > > > >>>>>>
> > > > > > >>>>>> After realizing that I was referencing the struct incorrectly, I wound up with a similar block of code. However, as I started testing it against /proc/pid/smaps[,_rollup] I noticed that my numbers didn't match up. Always smaller.
> > > > > > >>>>>>
> > > > > > >>>>>> I took a quick glance at fs/proc/task_mmu.c. I think I'll have to walk some sort of accounting structure.
> > > > > > >>>>>
> > > > > > >>>>>
> > > > > > >>>>> I started to take a hard look at fs/proc/task_mmu.c. With all the locking, globals, and compile-time constants, I'm not sure that it's even possible to correctly walk `vm_area_struct` in bpf.
> > > > > > >>>>
> > > > > > >>>> Yes, you can't take all those locks from BPF. But reading atomic
> > > > > > >>>> counters from BPF should be no problem. You might get a slightly out
> > > > > > >>>> of sync readings, but whatever you are doing shouldn't expect to have
> > > > > > >>>> 100% correct values anyways, because they might change so fast after
> > > > > > >>>> you read them.
> > > > > > >>>
> > > > > > >>> That was my initial thought. I didn't care to much about stale data, my only real concern was walking vm_area_struct and having memory freed. I wasn't sure if that could break the list underneath me. Although, that shouldn't be too difficult to get to the bottom of.
> > > > > > >>>
> > > > > > >>
> > > > > > >> Not sure about vm_area_struct (where is it in the example above?), but
> > > > > > >> mm_struct won't go away, because current task won't go away, because
> > > > > > >> BPF program is running in the context of current. Similarly for
> > > > > > >> bpf_iter, bpf_iter will actually take a refcnt on tast_struct. So I
> > > > > > >> think you don't have to worry about that.
> > > > > > >
> > > > > > > I didn't mention it explicitly in the example above. But when I originally mentioned walking an accounting structure, as procfs does, it winds up being `mm_struct->mmap,vm_[next,prev]`, with mmap being a `vm_area_struct`. But, it sounds like I should be abandoning that path and iterating over all the tasks.
> > > > > > >
> > > > > > >
> > > > > > >>>>> If anyone has suggestions for getting memory numbers from an entire process, not just a task/thread, I'd love to hear them. If not, I'll pursue this on my own.
> > > > > > >>>>
> > > > > > >>>> For this, you'd need to iterate across many tasks and aggregate their
> > > > > > >>>> results based on tasks's tgid. Check iter/task programs in selftests
> > > > > > >>>> (progs/bpf_iter_task.c, I think).
> > > > > > >
> > > > > > >
> > > > > > > When I try to replicate some of the selftest task logic. I run into some errors when I call bpf_object__load. `libbpf: task is not found in vmlinux BTF.` I'll try matching the selftest code more closely and digging into that further.
> > > > > >
> > > > > > Somehow libbpf did not prepend `task` with `bpf_iter_` prefix. Not sure
> > > > > > what is the exact issue. Yes, please mimic what selftests did.
> > > > > >
> > > > >
> > > > > It's just an artifact of how libbpf logs error in such case. It did
> > > > > search for "bpf_iter_task" type, though. But Matt probably doesn't
> > > > > have a recent enough kernel or didn't build it with
> > > > > CONFIG_DEBUG_INFO_BTF=y and pahole 1.16+?
> > > >
> > > > That shouldn't be the case, I generated vmlinux.h from my currently running machine.
> > > >
> > > >
> > > > I'm using an upstream kernel.
> > > > > ~ uname -r
> > > > > 5.7.2-arch1-1
> > > >
> > > > Which has the BTF debug info enabled.
> > > > > ~ zgrep BTF= /proc/config.gz
> > > > > CONFIG_DEBUG_INFO_BTF=y
> > > >
> > > >
> > > > I assume that it was built with the version of pahole that's in the upstream repos.
> > > > > ~ pacman -Ss pahole
> > > > > extra/pahole 1.17-1 [installed]
> > > >
> > > >
> > > > Unless I've came across some odd bug, I assume that I've implemented something incorrectly.
> > > >
> > >
> > > Ok, can you show your code (BPF and user-space side) and libbpf debug logs then?
> >
> >
> > Sure. The userspace section in question is below. I don't make it past `bpf_object__load`. Same userspace code works fine for tracepoints.
> >
> > struct bpf_program *prog;
> > struct bpf_object *obj;
> > char path[] = PT_BPF_OBJECT_DIR;
> > strcat(&path[strlen(path)], "/test.o");
> >
> > libbpf_set_print(print_libbpf_log);
> >
> > obj = bpf_object__open_file(path, NULL);
> > if (libbpf_get_error(obj))
> > return 1;
> >
> > if(!(prog = bpf_object__find_program_by_name(obj, "dump_task")))
> > goto cleanup;
> >
> > if (bpf_object__load(obj))
> > goto cleanup;
> >
> >
> > I copied the kernel code, only slightly modifying the include statements
> >
> > #define bpf_iter_meta bpf_iter_meta___not_used
> > #define bpf_iter__task bpf_iter__task___not_used
> > #include <vmlinux.h>
> > #undef bpf_iter_meta
> > #undef bpf_iter__task
> > #include <bpf_helpers.h>
> > #include <bpf_tracing.h>
> >
> > struct bpf_iter_meta {
> > struct seq_file *seq;
> > __u64 session_id;
> > __u64 seq_num;
> > } __attribute__((preserve_access_index));
> >
> > struct bpf_iter__task {
> > struct bpf_iter_meta *meta;
> > struct task_struct *task;
> > } __attribute__((preserve_access_index));
> >
> > SEC("iter/task")
> > int dump_task(struct bpf_iter__task *ctx)
> > {
> > struct seq_file *seq = ctx->meta->seq;
> > struct task_struct *task = ctx->task;
> >
> > if (task == (void *)0) {
> > BPF_SEQ_PRINTF(seq, " === END ===\n");
> > return 0;
> > }
> >
> > if (ctx->meta->seq_num == 0)
> > BPF_SEQ_PRINTF(seq, " tgid gid\n");
> >
> > BPF_SEQ_PRINTF(seq, "%8d %8d\n", task->tgid, task->pid);
> > return 0;
> > }
> >
> >
> > And here is the debug output
> >
> >
> > libbpf: loading /tmp//usr/lib/pt/bpf/test.o
> > libbpf: section(1) .strtab, size 277, link 0, flags 0, type=3
> > libbpf: skip section(1) .strtab
> > libbpf: section(2) .text, size 0, link 0, flags 6, type=1
> > libbpf: skip section(2) .text
> > libbpf: section(3) iter/task, size 320, link 0, flags 6, type=1
> > libbpf: found program iter/task
> > libbpf: section(4) .reliter/task, size 48, link 22, flags 0, type=9
> > libbpf: section(5) .rodata, size 45, link 0, flags 2, type=1
> > libbpf: section(6) license, size 4, link 0, flags 3, type=1
> > libbpf: license of /tmp//usr/lib/pt/bpf/test.o is GPL
> > libbpf: section(7) version, size 4, link 0, flags 3, type=1
> > libbpf: kernel version of /tmp//usr/lib/pt/bpf/test.o is 5060b
> > libbpf: section(8) .debug_str, size 135270, link 0, flags 30, type=1
> > libbpf: skip section(8) .debug_str
> > libbpf: section(9) .debug_loc, size 124, link 0, flags 0, type=1
> > libbpf: skip section(9) .debug_loc
> > libbpf: section(10) .debug_abbrev, size 857, link 0, flags 0, type=1
> > libbpf: skip section(10) .debug_abbrev
> > libbpf: section(11) .debug_info, size 224491, link 0, flags 0, type=1
> > libbpf: skip section(11) .debug_info
> > libbpf: section(12) .rel.debug_info, size 160, link 22, flags 0, type=9
> > libbpf: skip relo .rel.debug_info(12) for section(11)
> > libbpf: section(13) .BTF, size 25711, link 0, flags 0, type=1
> > libbpf: section(14) .rel.BTF, size 80, link 22, flags 0, type=9
> > libbpf: skip relo .rel.BTF(14) for section(13)
> > libbpf: section(15) .BTF.ext, size 348, link 0, flags 0, type=1
> > libbpf: section(16) .rel.BTF.ext, size 288, link 22, flags 0, type=9
> > libbpf: skip relo .rel.BTF.ext(16) for section(15)
> > libbpf: section(17) .debug_frame, size 40, link 0, flags 0, type=1
> > libbpf: skip section(17) .debug_frame
> > libbpf: section(18) .rel.debug_frame, size 16, link 22, flags 0, type=9
> > libbpf: skip relo .rel.debug_frame(18) for section(17)
> > libbpf: section(19) .debug_line, size 216, link 0, flags 0, type=1
> > libbpf: skip section(19) .debug_line
> > libbpf: section(20) .rel.debug_line, size 16, link 22, flags 0, type=9
> > libbpf: skip relo .rel.debug_line(20) for section(19)
> > libbpf: section(21) .llvm_addrsig, size 6, link 22, flags 80000000, type=1879002115
> > libbpf: skip section(21) .llvm_addrsig
> > libbpf: section(22) .symtab, size 312, link 1, flags 0, type=2
> > libbpf: looking for externs among 13 symbols...
> > libbpf: collected 0 externs total
> > libbpf: map 'test.rodata' (global data): at sec_idx 5, offset 0, flags 480.
> > libbpf: map 0 is "test.rodata"
> > libbpf: collecting relocating info for: 'iter/task'
> > libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 7
> > libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 7
> > libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 17
> > libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 17
> > libbpf: relo for shdr 5, symb 9, value 0, type 3, bind 0, name 0 (''), insn 33
> > libbpf: found data map 0 (test.rodata, sec 5, off 0) for insn 33
> > libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> > libbpf: map 'test.rodata': created successfully, fd=4
> > libbpf: loading kernel BTF '/sys/kernel/btf/vmlinux': 0
> > libbpf: prog 'iter/task': performing 6 CO-RE offset relocs
> > libbpf: prog 'iter/task': relo #0: kind 0, spec is [2] bpf_iter__task + 0:0 => 0.0 @ &x[0].meta
> > libbpf: prog 'iter/task': relo #0: no matching targets found for [2] bpf_iter__task + 0:0
> > libbpf: prog 'iter/task': relo #0: substituting insn #0 w/ invalid insn
> > libbpf: prog 'iter/task': relo #1: kind 0, spec is [8] bpf_iter_meta + 0:0 => 0.0 @ &x[0].seq
> > libbpf: prog 'iter/task': relo #1: no matching targets found for [8] bpf_iter_meta + 0:0
> > libbpf: prog 'iter/task': relo #1: substituting insn #1 w/ invalid insn
> > libbpf: prog 'iter/task': relo #2: kind 0, spec is [2] bpf_iter__task + 0:1 => 8.0 @ &x[0].task
> > libbpf: prog 'iter/task': relo #2: no matching targets found for [2] bpf_iter__task + 0:1
> > libbpf: prog 'iter/task': relo #2: substituting insn #2 w/ invalid insn
>
> see all these "substituting insn w/ invalid insn" messages? Your
> kernel doesn't have bpf_iter__task struct in it.
>
> You can confirm by running:
>
> $ bpftool btf dump file /sys/kernel/btf/vmlinux | grep bpf_iter_
>
> You said you have 5.7 kernel. Isn't bpf_iter available starting from 5.8?
Ah, this was it. I bumped to 8.5-rc2 and it loads now. I figured out bpftool gen skeleton via the selftests in the process, I'm a fan.
100: tracing name dump_task tag 70ea8cf44dc2d3e4 gpl
loaded_at 2020-06-24T07:23:37-0700 uid 0
xlated 320B jited 200B memlock 4096B map_ids 44
btf_id 60
I don't have any output in /sys/kernel/tracing/trace_pipe yet, but I'm sure I can fumble through the selftests some more and figure that out.
Thanks again. You two rock.
> > libbpf: prog 'iter/task': relo #3: kind 0, spec is [8] bpf_iter_meta + 0:2 => 16.0 @ &x[0].seq_num
> > libbpf: prog 'iter/task': relo #3: no matching targets found for [8] bpf_iter_meta + 0:2
> > libbpf: prog 'iter/task': relo #3: substituting insn #12 w/ invalid insn
> > libbpf: prog 'iter/task': relo #4: kind 0, spec is [12] task_struct + 0:71 => 1292.0 @ &x[0].tgid
> > libbpf: [12] task_struct: found candidate [115] task_struct
> > libbpf: prog 'iter/task': relo #4: matching candidate #0 task_struct against spec [115] task_struct + 0:71 => 1292.0 @ &x[0].tgid: 1
> > libbpf: prog 'iter/task': relo #4: patched insn #22 (LDX/ST/STX) off 1292 -> 1292
> > libbpf: prog 'iter/task': relo #5: kind 0, spec is [12] task_struct + 0:70 => 1288.0 @ &x[0].pid
> > libbpf: prog 'iter/task': relo #5: matching candidate #0 task_struct against spec [115] task_struct + 0:70 => 1288.0 @ &x[0].pid: 1
> > libbpf: prog 'iter/task': relo #5: patched insn #26 (LDX/ST/STX) off 1288 -> 1288
> > libbpf: task is not found in vmlinux BTF
> > libbpf: failed to load object '/tmp//usr/lib/pt/bpf/test.o'
> > *** stack smashing detected ***: terminated <-- I assume this is because I'm not handling my errors and cleaning up properly
> > Aborted (core dumped)
> >
> > > > > > > As an aside; is there any documentation for bpf_iter outside of the selftests?
> > > > > >
> > > > > > Unfortunately, no. The commit messages of the original patch set might help.
> > > > > > https://lore.kernel.org/bpf/20200507053916.1542319-1-yhs@fb.com/T/#mf973843af65fc51ac9b3e3673962cd3e87f705e8
^ permalink raw reply [flat|nested] 18+ messages in thread
end of thread, other threads:[~2020-06-24 15:51 UTC | newest]
Thread overview: 18+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-06-20 16:22 Accessing mm_rss_stat fields with btf/BPF_CORE_READ_INTO Matt Pallissard
2020-06-20 18:11 ` Yonghong Song
2020-06-20 20:06 ` Matt Pallissard
2020-06-21 3:29 ` Andrii Nakryiko
2020-06-21 15:44 ` Matt Pallissard
2020-06-22 15:01 ` Matt Pallissard
2020-06-22 16:20 ` Andrii Nakryiko
2020-06-22 17:19 ` Matt Pallissard
2020-06-22 22:09 ` Andrii Nakryiko
2020-06-23 14:54 ` Matt Pallissard
2020-06-23 16:35 ` Yonghong Song
2020-06-23 17:58 ` Andrii Nakryiko
2020-06-23 18:11 ` Matt Pallissard
2020-06-23 18:36 ` Andrii Nakryiko
2020-06-23 22:05 ` Matt Pallissard
2020-06-23 22:13 ` Andrii Nakryiko
2020-06-24 15:51 ` Matt Pallissard
2020-06-23 22:16 ` Yonghong Song
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.