| Message ID | 20171011190359.34926-1-shakeelb@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed 11-10-17 12:03:59, Shakeel Butt wrote: > The allocations from filp cache can be directly triggered by user > space applications. A buggy application can consume a significant > amount of unaccounted system memory. Though we have not noticed > such buggy applications in our production but upon close inspection, > we found that a lot of machines spend very significant amount of > memory on these caches. > > One way to limit allocations from filp cache is to set system level > limit of maximum number of open files. However this limit is shared > between different users on the system and one user can hog this > resource. To cater that, we can charge filp to kmemcg and set the > maximum limit very high and let the memory limit of each user limit > the number of files they can open and indirectly limiting their > allocations from filp cache. > > One side effect of this change is that it will allow _sysctl() to > return ENOMEM and the man page of _sysctl() does not specify that. > However the man page also discourages to use _sysctl() at all. > > Signed-off-by: Shakeel Butt <shakeelb@google.com> OK, this makes more sense than the original patch. struct file is not really large (248B on my system) so I am not sure how much this helps though. Anyway, I have no objections to the patch but I do not feel qualified to ack it either. > --- > > Changelog since v1: > - removed names_cache charging to kmemcg > > fs/file_table.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/fs/file_table.c b/fs/file_table.c > index 61517f57f8ef..567888cdf7d3 100644 > --- a/fs/file_table.c > +++ b/fs/file_table.c > @@ -312,7 +312,7 @@ void put_filp(struct file *file) > void __init files_init(void) > { > filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, > - SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); > + SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL); > percpu_counter_init(&nr_files, 0, GFP_KERNEL); > } > > -- > 2.15.0.rc0.271.g36b669edcc-goog
diff --git a/fs/file_table.c b/fs/file_table.c index 61517f57f8ef..567888cdf7d3 100644 --- a/fs/file_table.c +++ b/fs/file_table.c @@ -312,7 +312,7 @@ void put_filp(struct file *file) void __init files_init(void) { filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, - SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); + SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL); percpu_counter_init(&nr_files, 0, GFP_KERNEL); }
The allocations from filp cache can be directly triggered by user space applications. A buggy application can consume a significant amount of unaccounted system memory. Though we have not noticed such buggy applications in our production but upon close inspection, we found that a lot of machines spend very significant amount of memory on these caches. One way to limit allocations from filp cache is to set system level limit of maximum number of open files. However this limit is shared between different users on the system and one user can hog this resource. To cater that, we can charge filp to kmemcg and set the maximum limit very high and let the memory limit of each user limit the number of files they can open and indirectly limiting their allocations from filp cache. One side effect of this change is that it will allow _sysctl() to return ENOMEM and the man page of _sysctl() does not specify that. However the man page also discourages to use _sysctl() at all. Signed-off-by: Shakeel Butt <shakeelb@google.com> --- Changelog since v1: - removed names_cache charging to kmemcg fs/file_table.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)