2024-02-21 16:15:04

by Matthew Cassell

[permalink] [raw]
Subject: [PATCH] mm/util.c: Added page count to __vm_enough_memory failure warning

Commit 44b414c8715c5dcf53288 ("mm/util.c: add warning if __vm_enough_memory
fails") adds debug information which gives the process id and executable name
should __vm_enough_memory() fail. Adding the number of pages to the failure
message would benefit application developers and system administrators in
debugging overambitious memory requests by providing a point of reference to
the amount of memory causing __vm_enough_memory() to fail.

1. Set appropriate kernel tunable to reach code path for failure
message:

# echo 2 > /proc/sys/vm/overcommit_memory

2. Test program to generate failure - requests 1 gibibyte per iteration:

#include <stdlib.h>
#include <stdio.h>

int main(int argc, char **argv) {
for(;;) {
if(malloc(1<<30) == NULL)
break;

printf("allocated 1 GiB\n");
}

return 0;
}

3. Output:

Before:

__vm_enough_memory: pid: 1218, comm: a.out, not enough
memory for the allocation

After:

__vm_enough_memory: pid: 1141, comm: a.out, pages: 262145, not
enough memory for the allocation

Signed-off-by: Matthew Cassell <[email protected]>
---
mm/util.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/mm/util.c b/mm/util.c
index 5a6a9802583b..c0afb56f16ea 100644
--- a/mm/util.c
+++ b/mm/util.c
@@ -976,8 +976,8 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
if (percpu_counter_read_positive(&vm_committed_as) < allowed)
return 0;
error:
- pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
- __func__, current->pid, current->comm);
+ pr_warn_ratelimited("%s: pid: %d, comm: %s, pages: %ld, not enough memory for the allocation\n",
+ __func__, current->pid, current->comm, pages);
vm_unacct_memory(pages);

return -ENOMEM;
--
2.43.2



2024-02-22 12:18:25

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm/util.c: Added page count to __vm_enough_memory failure warning

On 21.02.24 17:02, Matthew Cassell wrote:
> Commit 44b414c8715c5dcf53288 ("mm/util.c: add warning if __vm_enough_memory
> fails") adds debug information which gives the process id and executable name
> should __vm_enough_memory() fail. Adding the number of pages to the failure
> message would benefit application developers and system administrators in
> debugging overambitious memory requests by providing a point of reference to
> the amount of memory causing __vm_enough_memory() to fail.
>
> 1. Set appropriate kernel tunable to reach code path for failure
> message:
>
> # echo 2 > /proc/sys/vm/overcommit_memory
>
> 2. Test program to generate failure - requests 1 gibibyte per iteration:
>
> #include <stdlib.h>
> #include <stdio.h>
>
> int main(int argc, char **argv) {
> for(;;) {
> if(malloc(1<<30) == NULL)
> break;
>
> printf("allocated 1 GiB\n");
> }
>
> return 0;
> }
>
> 3. Output:
>
> Before:
>
> __vm_enough_memory: pid: 1218, comm: a.out, not enough
> memory for the allocation
>
> After:
>
> __vm_enough_memory: pid: 1141, comm: a.out, pages: 262145, not
> enough memory for the allocation
>
> Signed-off-by: Matthew Cassell <[email protected]>
> ---
> mm/util.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/mm/util.c b/mm/util.c
> index 5a6a9802583b..c0afb56f16ea 100644
> --- a/mm/util.c
> +++ b/mm/util.c
> @@ -976,8 +976,8 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
> if (percpu_counter_read_positive(&vm_committed_as) < allowed)
> return 0;
> error:
> - pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
> - __func__, current->pid, current->comm);
> + pr_warn_ratelimited("%s: pid: %d, comm: %s, pages: %ld, not enough memory for the allocation\n",
> + __func__, current->pid, current->comm, pages);
> vm_unacct_memory(pages);
>
> return -ENOMEM;

I wonder if "bytes"/"kbytes" instead of pages would be more appropriate
here.

Often, this will fail due to mmap() [where we pass a size from user
space] and also "vm.overcommit_kbytes" is not in pages.

--
Cheers,

David / dhildenb


2024-02-22 16:32:06

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm/util.c: Added page count to __vm_enough_memory failure warning

On 22.02.24 17:24, Matthew Cassell wrote:
> Resending due to plain-text email issue that caused mailing list to get skipped.
>
> Thank you for the feedback. I agree with you and would prefer to use
> bytes/kbytes. Here are the 2 concerns that led to me keeping it as
> pages:
>
>
> 1. Reduce the impact of the patch. Here is the call trace to reach the
> failure warning:
>
> <… usual mmap() stuff …>
> mmap_region() -> security_enough_memory_mm() -> __vm_enough_memory()
>
> Within mmap_region(), the length variable originally passed to mmap()
> gets right-shifted to get the page count. My first thought was to add
> an additional an additional argument to security_enough_memory_mm() of
> type unsigned long to keep that variable, but saw a handful of calls
> to it that would have to conform to the change. Not that I do not
> think this debug statement does not warrant that, I felt the less
> impact, the better.
>
>
> 2. Concerned about losing bits. When converting back to bytes I was
> worried about the loss of precision and printing that number back to
> users:
>
> unsigned long bytes_failed = pages << (PAGE_SHIFT);
>

In which scenario would you imagine that we lose precision? In other
words, how would someone be able to create a VMA that is larger than
what we can fit into an unsigned long in bytes?

I'd simply print "pages << PAGE_SHIFT" here and not worry about that :)

--
Cheers,

David / dhildenb


2024-02-22 16:48:20

by Matthew Cassell

[permalink] [raw]
Subject: Re: [PATCH] mm/util.c: Added page count to __vm_enough_memory failure warning

Resending due to plain-text email issue that caused mailing list to get skipped.

Thank you for the feedback. I agree with you and would prefer to use
bytes/kbytes. Here are the 2 concerns that led to me keeping it as
pages:


1. Reduce the impact of the patch. Here is the call trace to reach the
failure warning:

<… usual mmap() stuff …>
mmap_region() -> security_enough_memory_mm() -> __vm_enough_memory()

Within mmap_region(), the length variable originally passed to mmap()
gets right-shifted to get the page count. My first thought was to add
an additional an additional argument to security_enough_memory_mm() of
type unsigned long to keep that variable, but saw a handful of calls
to it that would have to conform to the change. Not that I do not
think this debug statement does not warrant that, I felt the less
impact, the better.


2. Concerned about losing bits. When converting back to bytes I was
worried about the loss of precision and printing that number back to
users:

unsigned long bytes_failed = pages << (PAGE_SHIFT);


On Thu, Feb 22, 2024 at 6:18 AM David Hildenbrand <[email protected]> wrote:
>
> On 21.02.24 17:02, Matthew Cassell wrote:
> > Commit 44b414c8715c5dcf53288 ("mm/util.c: add warning if __vm_enough_memory
> > fails") adds debug information which gives the process id and executable name
> > should __vm_enough_memory() fail. Adding the number of pages to the failure
> > message would benefit application developers and system administrators in
> > debugging overambitious memory requests by providing a point of reference to
> > the amount of memory causing __vm_enough_memory() to fail.
> >
> > 1. Set appropriate kernel tunable to reach code path for failure
> > message:
> >
> > # echo 2 > /proc/sys/vm/overcommit_memory
> >
> > 2. Test program to generate failure - requests 1 gibibyte per iteration:
> >
> > #include <stdlib.h>
> > #include <stdio.h>
> >
> > int main(int argc, char **argv) {
> > for(;;) {
> > if(malloc(1<<30) == NULL)
> > break;
> >
> > printf("allocated 1 GiB\n");
> > }
> >
> > return 0;
> > }
> >
> > 3. Output:
> >
> > Before:
> >
> > __vm_enough_memory: pid: 1218, comm: a.out, not enough
> > memory for the allocation
> >
> > After:
> >
> > __vm_enough_memory: pid: 1141, comm: a.out, pages: 262145, not
> > enough memory for the allocation
> >
> > Signed-off-by: Matthew Cassell <[email protected]>
> > ---
> > mm/util.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/mm/util.c b/mm/util.c
> > index 5a6a9802583b..c0afb56f16ea 100644
> > --- a/mm/util.c
> > +++ b/mm/util.c
> > @@ -976,8 +976,8 @@ int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
> > if (percpu_counter_read_positive(&vm_committed_as) < allowed)
> > return 0;
> > error:
> > - pr_warn_ratelimited("%s: pid: %d, comm: %s, not enough memory for the allocation\n",
> > - __func__, current->pid, current->comm);
> > + pr_warn_ratelimited("%s: pid: %d, comm: %s, pages: %ld, not enough memory for the allocation\n",
> > + __func__, current->pid, current->comm, pages);
> > vm_unacct_memory(pages);
> >
> > return -ENOMEM;
>
> I wonder if "bytes"/"kbytes" instead of pages would be more appropriate
> here.
>
> Often, this will fail due to mmap() [where we pass a size from user
> space] and also "vm.overcommit_kbytes" is not in pages.
>
> --
> Cheers,
>
> David / dhildenb
>