diff mbox series

[5/5] slab: Allocate and use per-call-site caches

Message ID 20240809073309.2134488-5-kees@kernel.org (mailing list archive)
State New
Headers show
Series slab: Allocate and use per-call-site caches | expand

Commit Message

Kees Cook Aug. 9, 2024, 7:33 a.m. UTC
Use separate per-call-site kmem_cache or kmem_buckets. These are
allocated on demand to avoid wasting memory for unused caches.

A few caches need to be allocated very early to support allocating the
caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
GFP_ATOMIC allocations are currently left to be allocated from
KMALLOC_NORMAL.

With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.

Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
text that compares the features.

Improvements needed:
- Retain call site gfp flags in alloc_tag meta field to:
  - pre-allocate all GFP_ATOMIC caches (since their caches cannot
    be allocated on demand unless we want them to be GFP_ATOMIC
    themselves...)
  - Separate MEMCG allocations as well
- Allocate individual caches within kmem_buckets on demand to
  further reduce memory usage overhead.

Signed-off-by: Kees Cook <kees@kernel.org>
---
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com>
Cc: linux-mm@kvack.org
---
 include/linux/alloc_tag.h |   8 +++
 lib/alloc_tag.c           | 121 +++++++++++++++++++++++++++++++++++---
 mm/Kconfig                |  19 +++++-
 mm/slab_common.c          |   1 +
 mm/slub.c                 |  31 +++++++++-
 5 files changed, 170 insertions(+), 10 deletions(-)

Comments

Xiu Jianfeng Aug. 17, 2024, 1:30 a.m. UTC | #1
Hi Kees,

On 2024/8/9 15:33, Kees Cook wrote:
> Use separate per-call-site kmem_cache or kmem_buckets. These are
> allocated on demand to avoid wasting memory for unused caches.
> 
> A few caches need to be allocated very early to support allocating the
> caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
> GFP_ATOMIC allocations are currently left to be allocated from
> KMALLOC_NORMAL.
> 
> With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.
> 
> Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
> CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
> text that compares the features.
> 
> Improvements needed:
> - Retain call site gfp flags in alloc_tag meta field to:
>   - pre-allocate all GFP_ATOMIC caches (since their caches cannot
>     be allocated on demand unless we want them to be GFP_ATOMIC
>     themselves...)
>   - Separate MEMCG allocations as well
> - Allocate individual caches within kmem_buckets on demand to
>   further reduce memory usage overhead.
> 
> Signed-off-by: Kees Cook <kees@kernel.org>
> ---
> Cc: Suren Baghdasaryan <surenb@google.com>
> Cc: Kent Overstreet <kent.overstreet@linux.dev>
> Cc: Vlastimil Babka <vbabka@suse.cz>
> Cc: Christoph Lameter <cl@linux.com>
> Cc: Pekka Enberg <penberg@kernel.org>
> Cc: David Rientjes <rientjes@google.com>
> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Roman Gushchin <roman.gushchin@linux.dev>
> Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> Cc: linux-mm@kvack.org
> ---
>  include/linux/alloc_tag.h |   8 +++
>  lib/alloc_tag.c           | 121 +++++++++++++++++++++++++++++++++++---
>  mm/Kconfig                |  19 +++++-
>  mm/slab_common.c          |   1 +
>  mm/slub.c                 |  31 +++++++++-
>  5 files changed, 170 insertions(+), 10 deletions(-)
> 

[...]

> diff --git a/mm/slub.c b/mm/slub.c
> index 3520acaf9afa..d14102c4b4d7 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -4135,6 +4135,35 @@ void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
>  }
>  EXPORT_SYMBOL(__kmalloc_large_node_noprof);
>  
> +static __always_inline
> +struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
> +			       unsigned long caller)
> +{
> +#ifdef CONFIG_SLAB_PER_SITE
> +	struct alloc_tag *tag = current->alloc_tag;

There is a compile error here if CONFIG_MEM_ALLOC_PROFILING is disabled
when I test this patchset.

mm/slub.c: In function ‘choose_slab’:
mm/slub.c:4187:40: error: ‘struct task_struct’ has no member named
‘alloc_tag’
 4187 |         struct alloc_tag *tag = current->alloc_tag;
      |                                        ^~
  CC      mm/page_reporting.o

maybe CONFIG_SLAB_PER_SITE should depend on CONFIG_MEM_ALLOC_PROFILING


> +
> +	if (!b && tag && tag->meta.sized &&
> +	    kmalloc_type(flags, caller) == KMALLOC_NORMAL &&
> +	    (flags & GFP_ATOMIC) != GFP_ATOMIC) {
> +		void *p = READ_ONCE(tag->meta.cache);
> +
> +		if (!p && slab_state >= UP) {
> +			alloc_tag_site_init(&tag->ct, true);
> +			p = READ_ONCE(tag->meta.cache);
> +		}
> +
> +		if (tag->meta.sized < SIZE_MAX) {
> +			if (p)
> +				return p;
> +			/* Otherwise continue with default buckets. */
> +		} else {
> +			b = p;
> +		}
> +	}
> +#endif
> +	return kmalloc_slab(size, b, flags, caller);
> +}
> +
>  static __always_inline
>  void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
>  			unsigned long caller)
> @@ -4152,7 +4181,7 @@ void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
>  	if (unlikely(!size))
>  		return ZERO_SIZE_PTR;
>  
> -	s = kmalloc_slab(size, b, flags, caller);
> +	s = choose_slab(size, b, flags, caller);
>  
>  	ret = slab_alloc_node(s, NULL, flags, node, caller, size);
>  	ret = kasan_kmalloc(s, ret, size, flags);
Kees Cook Aug. 22, 2024, 5:47 p.m. UTC | #2
On Sat, Aug 17, 2024 at 09:30:58AM +0800, Xiu Jianfeng wrote:
> Hi Kees,
> 
> On 2024/8/9 15:33, Kees Cook wrote:
> > Use separate per-call-site kmem_cache or kmem_buckets. These are
> > allocated on demand to avoid wasting memory for unused caches.
> > 
> > A few caches need to be allocated very early to support allocating the
> > caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
> > GFP_ATOMIC allocations are currently left to be allocated from
> > KMALLOC_NORMAL.
> > 
> > With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.
> > 
> > Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
> > CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
> > text that compares the features.
> > 
> > Improvements needed:
> > - Retain call site gfp flags in alloc_tag meta field to:
> >   - pre-allocate all GFP_ATOMIC caches (since their caches cannot
> >     be allocated on demand unless we want them to be GFP_ATOMIC
> >     themselves...)
> >   - Separate MEMCG allocations as well
> > - Allocate individual caches within kmem_buckets on demand to
> >   further reduce memory usage overhead.
> > 
> > Signed-off-by: Kees Cook <kees@kernel.org>
> > ---
> > Cc: Suren Baghdasaryan <surenb@google.com>
> > Cc: Kent Overstreet <kent.overstreet@linux.dev>
> > Cc: Vlastimil Babka <vbabka@suse.cz>
> > Cc: Christoph Lameter <cl@linux.com>
> > Cc: Pekka Enberg <penberg@kernel.org>
> > Cc: David Rientjes <rientjes@google.com>
> > Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: Roman Gushchin <roman.gushchin@linux.dev>
> > Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> > Cc: linux-mm@kvack.org
> > ---
> >  include/linux/alloc_tag.h |   8 +++
> >  lib/alloc_tag.c           | 121 +++++++++++++++++++++++++++++++++++---
> >  mm/Kconfig                |  19 +++++-
> >  mm/slab_common.c          |   1 +
> >  mm/slub.c                 |  31 +++++++++-
> >  5 files changed, 170 insertions(+), 10 deletions(-)
> > 
> 
> [...]
> 
> > diff --git a/mm/slub.c b/mm/slub.c
> > index 3520acaf9afa..d14102c4b4d7 100644
> > --- a/mm/slub.c
> > +++ b/mm/slub.c
> > @@ -4135,6 +4135,35 @@ void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
> >  }
> >  EXPORT_SYMBOL(__kmalloc_large_node_noprof);
> >  
> > +static __always_inline
> > +struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
> > +			       unsigned long caller)
> > +{
> > +#ifdef CONFIG_SLAB_PER_SITE
> > +	struct alloc_tag *tag = current->alloc_tag;
> 
> There is a compile error here if CONFIG_MEM_ALLOC_PROFILING is disabled
> when I test this patchset.
> 
> mm/slub.c: In function ‘choose_slab’:
> mm/slub.c:4187:40: error: ‘struct task_struct’ has no member named
> ‘alloc_tag’
>  4187 |         struct alloc_tag *tag = current->alloc_tag;
>       |                                        ^~
>   CC      mm/page_reporting.o
> 
> maybe CONFIG_SLAB_PER_SITE should depend on CONFIG_MEM_ALLOC_PROFILING

Thanks! I tried to make the Kconfig use the right dependencies, but I
clearly missed something. There is also some weird behavior between
"depends" and "select". I will get this fixed for the next version.

-Kees
Suren Baghdasaryan Aug. 29, 2024, 5:03 p.m. UTC | #3
On Fri, Aug 9, 2024 at 12:33 AM Kees Cook <kees@kernel.org> wrote:
>
> Use separate per-call-site kmem_cache or kmem_buckets. These are
> allocated on demand to avoid wasting memory for unused caches.
>
> A few caches need to be allocated very early to support allocating the
> caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
> GFP_ATOMIC allocations are currently left to be allocated from
> KMALLOC_NORMAL.
>
> With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.
>
> Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
> CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
> text that compares the features.
>
> Improvements needed:
> - Retain call site gfp flags in alloc_tag meta field to:
>   - pre-allocate all GFP_ATOMIC caches (since their caches cannot
>     be allocated on demand unless we want them to be GFP_ATOMIC
>     themselves...)

I'm currently working on a feature to identify allocations with
__GFP_ACCOUNT known at compile time (similar to how you handle the
size in the previous patch). Might be something you can reuse/extend.

>   - Separate MEMCG allocations as well

Do you mean allocations with __GFP_ACCOUNT or something else?

> - Allocate individual caches within kmem_buckets on demand to
>   further reduce memory usage overhead.
>
> Signed-off-by: Kees Cook <kees@kernel.org>
> ---
> Cc: Suren Baghdasaryan <surenb@google.com>
> Cc: Kent Overstreet <kent.overstreet@linux.dev>
> Cc: Vlastimil Babka <vbabka@suse.cz>
> Cc: Christoph Lameter <cl@linux.com>
> Cc: Pekka Enberg <penberg@kernel.org>
> Cc: David Rientjes <rientjes@google.com>
> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Cc: Roman Gushchin <roman.gushchin@linux.dev>
> Cc: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> Cc: linux-mm@kvack.org
> ---
>  include/linux/alloc_tag.h |   8 +++
>  lib/alloc_tag.c           | 121 +++++++++++++++++++++++++++++++++++---
>  mm/Kconfig                |  19 +++++-
>  mm/slab_common.c          |   1 +
>  mm/slub.c                 |  31 +++++++++-
>  5 files changed, 170 insertions(+), 10 deletions(-)
>
> diff --git a/include/linux/alloc_tag.h b/include/linux/alloc_tag.h
> index f5d8c5849b82..c95628f9b049 100644
> --- a/include/linux/alloc_tag.h
> +++ b/include/linux/alloc_tag.h
> @@ -24,6 +24,7 @@ struct alloc_tag_counters {
>  struct alloc_meta {
>         /* 0 means non-slab, SIZE_MAX means dynamic, and everything else is fixed-size. */
>         size_t sized;
> +       void *cache;

I see now where that meta.cache in the previous patch came from...
That part should be moved here.

>  };
>  #define ALLOC_META_INIT(_size) {               \
>                 .sized = (__builtin_constant_p(_size) ? (_size) : SIZE_MAX), \
> @@ -216,6 +217,13 @@ static inline void alloc_tag_sub(union codetag_ref *ref, size_t bytes) {}
>
>  #endif /* CONFIG_MEM_ALLOC_PROFILING */
>
> +#ifdef CONFIG_SLAB_PER_SITE
> +void alloc_tag_early_walk(void);
> +void alloc_tag_site_init(struct codetag *ct, bool ondemand);
> +#else
> +static inline void alloc_tag_early_walk(void) {}
> +#endif
> +
>  #define alloc_hooks_tag(_tag, _do_alloc)                               \
>  ({                                                                     \
>         struct alloc_tag * __maybe_unused _old = alloc_tag_save(_tag);  \
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index 6d2cb72bf269..e8a66a7c4a6b 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -157,6 +157,89 @@ static void __init procfs_init(void)
>         proc_create_seq("allocinfo", 0400, NULL, &allocinfo_seq_op);
>  }
>
> +#ifdef CONFIG_SLAB_PER_SITE
> +static bool ondemand_ready;
> +
> +void alloc_tag_site_init(struct codetag *ct, bool ondemand)
> +{
> +       struct alloc_tag *tag = ct_to_alloc_tag(ct);
> +       char *name;
> +       void *p, *old;
> +
> +       /* Only handle kmalloc allocations. */
> +       if (!tag->meta.sized)
> +               return;
> +
> +       /* Must be ready for on-demand allocations. */
> +       if (ondemand && !ondemand_ready)
> +               return;
> +
> +       old = READ_ONCE(tag->meta.cache);
> +       /* Already allocated? */
> +       if (old)
> +               return;
> +
> +       if (tag->meta.sized < SIZE_MAX) {
> +               /* Fixed-size allocations. */
> +               name = kasprintf(GFP_KERNEL, "f:%zu:%s:%d", tag->meta.sized, ct->function, ct->lineno);
> +               if (WARN_ON_ONCE(!name))
> +                       return;
> +               /*
> +                * As with KMALLOC_NORMAL, the entire allocation needs to be
> +                * open to usercopy access. :(
> +                */
> +               p = kmem_cache_create_usercopy(name, tag->meta.sized, 0,
> +                                              SLAB_NO_MERGE, 0, tag->meta.sized,
> +                                              NULL);
> +       } else {
> +               /* Dynamically-size allocations. */
> +               name = kasprintf(GFP_KERNEL, "d:%s:%d", ct->function, ct->lineno);
> +               if (WARN_ON_ONCE(!name))
> +                       return;
> +               p = kmem_buckets_create(name, SLAB_NO_MERGE, 0, UINT_MAX, NULL);
> +       }
> +       if (p) {
> +               if (unlikely(!try_cmpxchg(&tag->meta.cache, &old, p))) {
> +                       /* We lost the allocation race; clean up. */
> +                       if (tag->meta.sized < SIZE_MAX)
> +                               kmem_cache_destroy(p);
> +                       else
> +                               kmem_buckets_destroy(p);
> +               }
> +       }
> +       kfree(name);
> +}
> +
> +static void alloc_tag_site_init_early(struct codetag *ct)
> +{
> +       /* Explicitly initialize the caches needed to initialize caches. */
> +       if (strcmp(ct->function, "kstrdup") == 0 ||
> +           strcmp(ct->function, "kvasprintf") == 0 ||
> +           strcmp(ct->function, "pcpu_mem_zalloc") == 0)

I hope we can find a better way to distinguish these allocations.
Maybe have a specialized hook for them, like alloc_hooks_early() which
sets a bit inside ct->flags to distinguish them?

> +               alloc_tag_site_init(ct, false);
> +
> +       /* TODO: pre-allocate GFP_ATOMIC caches here. */

You could pre-allocate GFP_ATOMIC caches during
alloc_tag_module_load() only if gfp_flags are known at compile time I
think. I guess for the dynamic case choose_slab() will fall back to
kmalloc_slab()?

> +}
> +#endif
> +
> +static void alloc_tag_module_load(struct codetag_type *cttype,
> +                                 struct codetag_module *cmod)
> +{
> +#ifdef CONFIG_SLAB_PER_SITE
> +       struct codetag_iterator iter;
> +       struct codetag *ct;
> +
> +       iter = codetag_get_ct_iter(cttype);
> +       for (ct = codetag_next_ct(&iter); ct; ct = codetag_next_ct(&iter)) {
> +               if (iter.cmod != cmod)
> +                       continue;
> +
> +               /* TODO: pre-allocate GFP_ATOMIC caches here. */
> +               //alloc_tag_site_init(ct, false);
> +       }
> +#endif
> +}
> +
>  static bool alloc_tag_module_unload(struct codetag_type *cttype,
>                                     struct codetag_module *cmod)
>  {
> @@ -175,8 +258,21 @@ static bool alloc_tag_module_unload(struct codetag_type *cttype,
>
>                 if (WARN(counter.bytes,
>                          "%s:%u module %s func:%s has %llu allocated at module unload",
> -                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes))
> +                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes)) {
>                         module_unused = false;
> +               }
> +#ifdef CONFIG_SLAB_PER_SITE
> +               else if (tag->meta.sized) {
> +                       /* Remove the allocated caches, if possible. */
> +                       void *p = READ_ONCE(tag->meta.cache);
> +
> +                       WRITE_ONCE(tag->meta.cache, NULL);

I'm guessing you are not using try_cmpxchg() the same way you did in
alloc_tag_site_init() because a race with any other user is impossible
at the module unload time? If so, a comment mentioning that would be
good.

> +                       if (tag->meta.sized < SIZE_MAX)
> +                               kmem_cache_destroy(p);
> +                       else
> +                               kmem_buckets_destroy(p);
> +               }
> +#endif
>         }
>
>         return module_unused;
> @@ -260,15 +356,16 @@ static void __init sysctl_init(void)
>  static inline void sysctl_init(void) {}
>  #endif /* CONFIG_SYSCTL */
>
> +static const struct codetag_type_desc alloc_tag_desc = {
> +       .section        = "alloc_tags",
> +       .tag_size       = sizeof(struct alloc_tag),
> +       .module_load    = alloc_tag_module_load,
> +       .module_unload  = alloc_tag_module_unload,
> +};
> +
>  static int __init alloc_tag_init(void)
>  {
> -       const struct codetag_type_desc desc = {
> -               .section        = "alloc_tags",
> -               .tag_size       = sizeof(struct alloc_tag),
> -               .module_unload  = alloc_tag_module_unload,
> -       };
> -
> -       alloc_tag_cttype = codetag_register_type(&desc);
> +       alloc_tag_cttype = codetag_register_type(&alloc_tag_desc);
>         if (IS_ERR(alloc_tag_cttype))
>                 return PTR_ERR(alloc_tag_cttype);
>
> @@ -278,3 +375,11 @@ static int __init alloc_tag_init(void)
>         return 0;
>  }
>  module_init(alloc_tag_init);
> +
> +#ifdef CONFIG_SLAB_PER_SITE
> +void alloc_tag_early_walk(void)
> +{
> +       codetag_early_walk(&alloc_tag_desc, alloc_tag_site_init_early);
> +       ondemand_ready = true;
> +}
> +#endif
> diff --git a/mm/Kconfig b/mm/Kconfig
> index 855c63c3270d..4f01cb6dd32e 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -302,7 +302,20 @@ config SLAB_PER_SITE
>         default SLAB_FREELIST_HARDENED
>         select SLAB_BUCKETS
>         help
> -         Track sizes of kmalloc() call sites.
> +         As a defense against shared-cache "type confusion" use-after-free
> +         attacks, every kmalloc()-family call allocates from a separate
> +         kmem_cache (or when dynamically sized, kmem_buckets). Attackers
> +         will no longer be able to groom malicious objects via similarly
> +         sized allocations that share the same cache as the target object.
> +
> +         This increases the "at rest" kmalloc slab memory usage by
> +         roughly 5x (around 7MiB), and adds the potential for greater
> +         long-term memory fragmentation. However, some workloads
> +         actually see performance improvements when single allocation
> +         sites are hot.

I hope you provide the performance and overhead data in the cover
letter when you post v1.

> +
> +         For a similar defense, see CONFIG_RANDOM_KMALLOC_CACHES, which
> +         has less memory usage overhead, but is probabilistic.
>
>  config SLUB_STATS
>         default n
> @@ -331,6 +344,7 @@ config SLUB_CPU_PARTIAL
>  config RANDOM_KMALLOC_CACHES
>         default n
>         depends on !SLUB_TINY
> +       depends on !SLAB_PER_SITE
>         bool "Randomize slab caches for normal kmalloc"
>         help
>           A hardening feature that creates multiple copies of slab caches for
> @@ -345,6 +359,9 @@ config RANDOM_KMALLOC_CACHES
>           limited degree of memory and CPU overhead that relates to hardware and
>           system workload.
>
> +         For a similar defense, see CONFIG_SLAB_PER_SITE, which is
> +         deterministic, but has greater memory usage overhead.
> +
>  endmenu # Slab allocator options
>
>  config SHUFFLE_PAGE_ALLOCATOR
> diff --git a/mm/slab_common.c b/mm/slab_common.c
> index fc698cba0ebe..09506bfa972c 100644
> --- a/mm/slab_common.c
> +++ b/mm/slab_common.c
> @@ -1040,6 +1040,7 @@ void __init create_kmalloc_caches(void)
>                 kmem_buckets_cache = kmem_cache_create("kmalloc_buckets",
>                                                        sizeof(kmem_buckets),
>                                                        0, SLAB_NO_MERGE, NULL);
> +       alloc_tag_early_walk();
>  }
>
>  /**
> diff --git a/mm/slub.c b/mm/slub.c
> index 3520acaf9afa..d14102c4b4d7 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -4135,6 +4135,35 @@ void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
>  }
>  EXPORT_SYMBOL(__kmalloc_large_node_noprof);
>
> +static __always_inline
> +struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
> +                              unsigned long caller)
> +{
> +#ifdef CONFIG_SLAB_PER_SITE
> +       struct alloc_tag *tag = current->alloc_tag;
> +
> +       if (!b && tag && tag->meta.sized &&
> +           kmalloc_type(flags, caller) == KMALLOC_NORMAL &&
> +           (flags & GFP_ATOMIC) != GFP_ATOMIC) {

What if allocation is GFP_ATOMIC but a previous allocation from the
same location (same tag) happened without GFP_ATOMIC and
tag->meta.cache was allocated. Why not use that existing cache?
Same if the tag->meta.cache was pre-allocated.


> +               void *p = READ_ONCE(tag->meta.cache);
> +
> +               if (!p && slab_state >= UP) {
> +                       alloc_tag_site_init(&tag->ct, true);
> +                       p = READ_ONCE(tag->meta.cache);
> +               }
> +
> +               if (tag->meta.sized < SIZE_MAX) {
> +                       if (p)
> +                               return p;
> +                       /* Otherwise continue with default buckets. */
> +               } else {
> +                       b = p;
> +               }
> +       }
> +#endif
> +       return kmalloc_slab(size, b, flags, caller);
> +}
> +
>  static __always_inline
>  void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
>                         unsigned long caller)
> @@ -4152,7 +4181,7 @@ void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
>         if (unlikely(!size))
>                 return ZERO_SIZE_PTR;
>
> -       s = kmalloc_slab(size, b, flags, caller);
> +       s = choose_slab(size, b, flags, caller);
>
>         ret = slab_alloc_node(s, NULL, flags, node, caller, size);
>         ret = kasan_kmalloc(s, ret, size, flags);
> --
> 2.34.1
>
Kees Cook Sept. 11, 2024, 10:30 p.m. UTC | #4
On Thu, Aug 29, 2024 at 10:03:56AM -0700, Suren Baghdasaryan wrote:
> On Fri, Aug 9, 2024 at 12:33 AM Kees Cook <kees@kernel.org> wrote:
> >
> > Use separate per-call-site kmem_cache or kmem_buckets. These are
> > allocated on demand to avoid wasting memory for unused caches.
> >
> > A few caches need to be allocated very early to support allocating the
> > caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
> > GFP_ATOMIC allocations are currently left to be allocated from
> > KMALLOC_NORMAL.
> >
> > With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.
> >
> > Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
> > CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
> > text that compares the features.
> >
> > Improvements needed:
> > - Retain call site gfp flags in alloc_tag meta field to:
> >   - pre-allocate all GFP_ATOMIC caches (since their caches cannot
> >     be allocated on demand unless we want them to be GFP_ATOMIC
> >     themselves...)
> 
> I'm currently working on a feature to identify allocations with
> __GFP_ACCOUNT known at compile time (similar to how you handle the
> size in the previous patch). Might be something you can reuse/extend.

Great, yes! I'd love to check it out.

> >   - Separate MEMCG allocations as well
> 
> Do you mean allocations with __GFP_ACCOUNT or something else?

I do, yes.

> > +static void alloc_tag_site_init_early(struct codetag *ct)
> > +{
> > +       /* Explicitly initialize the caches needed to initialize caches. */
> > +       if (strcmp(ct->function, "kstrdup") == 0 ||
> > +           strcmp(ct->function, "kvasprintf") == 0 ||
> > +           strcmp(ct->function, "pcpu_mem_zalloc") == 0)
> 
> I hope we can find a better way to distinguish these allocations.
> Maybe have a specialized hook for them, like alloc_hooks_early() which
> sets a bit inside ct->flags to distinguish them?

That might be possible. I'll see how that ends up looking. I don't want
to even further fragment the alloc_hooks_... variants.

> 
> > +               alloc_tag_site_init(ct, false);
> > +
> > +       /* TODO: pre-allocate GFP_ATOMIC caches here. */
> 
> You could pre-allocate GFP_ATOMIC caches during
> alloc_tag_module_load() only if gfp_flags are known at compile time I
> think. I guess for the dynamic case choose_slab() will fall back to
> kmalloc_slab()?

Right, yes. I'd do it like the size checking: if we know at compile
time, we can depend on it, otherwise it's a run-time fallback.

> 
> > @@ -175,8 +258,21 @@ static bool alloc_tag_module_unload(struct codetag_type *cttype,
> >
> >                 if (WARN(counter.bytes,
> >                          "%s:%u module %s func:%s has %llu allocated at module unload",
> > -                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes))
> > +                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes)) {
> >                         module_unused = false;
> > +               }
> > +#ifdef CONFIG_SLAB_PER_SITE
> > +               else if (tag->meta.sized) {
> > +                       /* Remove the allocated caches, if possible. */
> > +                       void *p = READ_ONCE(tag->meta.cache);
> > +
> > +                       WRITE_ONCE(tag->meta.cache, NULL);
> 
> I'm guessing you are not using try_cmpxchg() the same way you did in
> alloc_tag_site_init() because a race with any other user is impossible
> at the module unload time? If so, a comment mentioning that would be
> good.

Correct. It should not be possible. But yes, I will add a comment.

> > diff --git a/mm/Kconfig b/mm/Kconfig
> > index 855c63c3270d..4f01cb6dd32e 100644
> > --- a/mm/Kconfig
> > +++ b/mm/Kconfig
> > @@ -302,7 +302,20 @@ config SLAB_PER_SITE
> >         default SLAB_FREELIST_HARDENED
> >         select SLAB_BUCKETS
> >         help
> > -         Track sizes of kmalloc() call sites.
> > +         As a defense against shared-cache "type confusion" use-after-free
> > +         attacks, every kmalloc()-family call allocates from a separate
> > +         kmem_cache (or when dynamically sized, kmem_buckets). Attackers
> > +         will no longer be able to groom malicious objects via similarly
> > +         sized allocations that share the same cache as the target object.
> > +
> > +         This increases the "at rest" kmalloc slab memory usage by
> > +         roughly 5x (around 7MiB), and adds the potential for greater
> > +         long-term memory fragmentation. However, some workloads
> > +         actually see performance improvements when single allocation
> > +         sites are hot.
> 
> I hope you provide the performance and overhead data in the cover
> letter when you post v1.

That's my plan. It's always odd choosing workloads, but we do seem to
have a few 'regular' benchmarks (hackbench, kernel builds, etc). Is
there anything in particular you'd want to see?

> > +static __always_inline
> > +struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
> > +                              unsigned long caller)
> > +{
> > +#ifdef CONFIG_SLAB_PER_SITE
> > +       struct alloc_tag *tag = current->alloc_tag;
> > +
> > +       if (!b && tag && tag->meta.sized &&
> > +           kmalloc_type(flags, caller) == KMALLOC_NORMAL &&
> > +           (flags & GFP_ATOMIC) != GFP_ATOMIC) {
> 
> What if allocation is GFP_ATOMIC but a previous allocation from the
> same location (same tag) happened without GFP_ATOMIC and
> tag->meta.cache was allocated. Why not use that existing cache?
> Same if the tag->meta.cache was pre-allocated.

Maybe I was being too conservative in my understanding -- I thought that
I couldn't use those caches on the chance that they may already be full?
Or is that always the risk, ad GFP_ATOMIC deals with that? If it would
be considered safe attempt the allocation from the existing cache, then
yeah, I can adjust this check.

Thanks for looking these over!

-Kees
Suren Baghdasaryan Sept. 12, 2024, 3:58 p.m. UTC | #5
On Wed, Sep 11, 2024 at 3:30 PM Kees Cook <kees@kernel.org> wrote:
>
> On Thu, Aug 29, 2024 at 10:03:56AM -0700, Suren Baghdasaryan wrote:
> > On Fri, Aug 9, 2024 at 12:33 AM Kees Cook <kees@kernel.org> wrote:
> > >
> > > Use separate per-call-site kmem_cache or kmem_buckets. These are
> > > allocated on demand to avoid wasting memory for unused caches.
> > >
> > > A few caches need to be allocated very early to support allocating the
> > > caches themselves: kstrdup(), kvasprintf(), and pcpu_mem_zalloc(). Any
> > > GFP_ATOMIC allocations are currently left to be allocated from
> > > KMALLOC_NORMAL.
> > >
> > > With a distro config, /proc/slabinfo grows from ~400 entries to ~2200.
> > >
> > > Since this feature (CONFIG_SLAB_PER_SITE) is redundant to
> > > CONFIG_RANDOM_KMALLOC_CACHES, mark it a incompatible. Add Kconfig help
> > > text that compares the features.
> > >
> > > Improvements needed:
> > > - Retain call site gfp flags in alloc_tag meta field to:
> > >   - pre-allocate all GFP_ATOMIC caches (since their caches cannot
> > >     be allocated on demand unless we want them to be GFP_ATOMIC
> > >     themselves...)
> >
> > I'm currently working on a feature to identify allocations with
> > __GFP_ACCOUNT known at compile time (similar to how you handle the
> > size in the previous patch). Might be something you can reuse/extend.
>
> Great, yes! I'd love to check it out.
>
> > >   - Separate MEMCG allocations as well
> >
> > Do you mean allocations with __GFP_ACCOUNT or something else?
>
> I do, yes.
>
> > > +static void alloc_tag_site_init_early(struct codetag *ct)
> > > +{
> > > +       /* Explicitly initialize the caches needed to initialize caches. */
> > > +       if (strcmp(ct->function, "kstrdup") == 0 ||
> > > +           strcmp(ct->function, "kvasprintf") == 0 ||
> > > +           strcmp(ct->function, "pcpu_mem_zalloc") == 0)
> >
> > I hope we can find a better way to distinguish these allocations.
> > Maybe have a specialized hook for them, like alloc_hooks_early() which
> > sets a bit inside ct->flags to distinguish them?
>
> That might be possible. I'll see how that ends up looking. I don't want
> to even further fragment the alloc_hooks_... variants.
>
> >
> > > +               alloc_tag_site_init(ct, false);
> > > +
> > > +       /* TODO: pre-allocate GFP_ATOMIC caches here. */
> >
> > You could pre-allocate GFP_ATOMIC caches during
> > alloc_tag_module_load() only if gfp_flags are known at compile time I
> > think. I guess for the dynamic case choose_slab() will fall back to
> > kmalloc_slab()?
>
> Right, yes. I'd do it like the size checking: if we know at compile
> time, we can depend on it, otherwise it's a run-time fallback.
>
> >
> > > @@ -175,8 +258,21 @@ static bool alloc_tag_module_unload(struct codetag_type *cttype,
> > >
> > >                 if (WARN(counter.bytes,
> > >                          "%s:%u module %s func:%s has %llu allocated at module unload",
> > > -                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes))
> > > +                        ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes)) {
> > >                         module_unused = false;
> > > +               }
> > > +#ifdef CONFIG_SLAB_PER_SITE
> > > +               else if (tag->meta.sized) {
> > > +                       /* Remove the allocated caches, if possible. */
> > > +                       void *p = READ_ONCE(tag->meta.cache);
> > > +
> > > +                       WRITE_ONCE(tag->meta.cache, NULL);
> >
> > I'm guessing you are not using try_cmpxchg() the same way you did in
> > alloc_tag_site_init() because a race with any other user is impossible
> > at the module unload time? If so, a comment mentioning that would be
> > good.
>
> Correct. It should not be possible. But yes, I will add a comment.
>
> > > diff --git a/mm/Kconfig b/mm/Kconfig
> > > index 855c63c3270d..4f01cb6dd32e 100644
> > > --- a/mm/Kconfig
> > > +++ b/mm/Kconfig
> > > @@ -302,7 +302,20 @@ config SLAB_PER_SITE
> > >         default SLAB_FREELIST_HARDENED
> > >         select SLAB_BUCKETS
> > >         help
> > > -         Track sizes of kmalloc() call sites.
> > > +         As a defense against shared-cache "type confusion" use-after-free
> > > +         attacks, every kmalloc()-family call allocates from a separate
> > > +         kmem_cache (or when dynamically sized, kmem_buckets). Attackers
> > > +         will no longer be able to groom malicious objects via similarly
> > > +         sized allocations that share the same cache as the target object.
> > > +
> > > +         This increases the "at rest" kmalloc slab memory usage by
> > > +         roughly 5x (around 7MiB), and adds the potential for greater
> > > +         long-term memory fragmentation. However, some workloads
> > > +         actually see performance improvements when single allocation
> > > +         sites are hot.
> >
> > I hope you provide the performance and overhead data in the cover
> > letter when you post v1.
>
> That's my plan. It's always odd choosing workloads, but we do seem to
> have a few 'regular' benchmarks (hackbench, kernel builds, etc). Is
> there anything in particular you'd want to see?

I have a stress test implemented as a loadable module to benchmark
slab and page allocation times (just a tight loop and timing it). I
can clean it up a bit and share with you.

>
> > > +static __always_inline
> > > +struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
> > > +                              unsigned long caller)
> > > +{
> > > +#ifdef CONFIG_SLAB_PER_SITE
> > > +       struct alloc_tag *tag = current->alloc_tag;
> > > +
> > > +       if (!b && tag && tag->meta.sized &&
> > > +           kmalloc_type(flags, caller) == KMALLOC_NORMAL &&
> > > +           (flags & GFP_ATOMIC) != GFP_ATOMIC) {
> >
> > What if allocation is GFP_ATOMIC but a previous allocation from the
> > same location (same tag) happened without GFP_ATOMIC and
> > tag->meta.cache was allocated. Why not use that existing cache?
> > Same if the tag->meta.cache was pre-allocated.
>
> Maybe I was being too conservative in my understanding -- I thought that
> I couldn't use those caches on the chance that they may already be full?
> Or is that always the risk, ad GFP_ATOMIC deals with that? If it would
> be considered safe attempt the allocation from the existing cache, then
> yeah, I can adjust this check.

Well, you fall back to kmalloc_slab() which also might be full. So,
how would using an existing cache be different?

>
> Thanks for looking these over!
>
> -Kees
>
> --
> Kees Cook
diff mbox series

Patch

diff --git a/include/linux/alloc_tag.h b/include/linux/alloc_tag.h
index f5d8c5849b82..c95628f9b049 100644
--- a/include/linux/alloc_tag.h
+++ b/include/linux/alloc_tag.h
@@ -24,6 +24,7 @@  struct alloc_tag_counters {
 struct alloc_meta {
 	/* 0 means non-slab, SIZE_MAX means dynamic, and everything else is fixed-size. */
 	size_t sized;
+	void *cache;
 };
 #define ALLOC_META_INIT(_size)	{		\
 		.sized = (__builtin_constant_p(_size) ? (_size) : SIZE_MAX), \
@@ -216,6 +217,13 @@  static inline void alloc_tag_sub(union codetag_ref *ref, size_t bytes) {}
 
 #endif /* CONFIG_MEM_ALLOC_PROFILING */
 
+#ifdef CONFIG_SLAB_PER_SITE
+void alloc_tag_early_walk(void);
+void alloc_tag_site_init(struct codetag *ct, bool ondemand);
+#else
+static inline void alloc_tag_early_walk(void) {}
+#endif
+
 #define alloc_hooks_tag(_tag, _do_alloc)				\
 ({									\
 	struct alloc_tag * __maybe_unused _old = alloc_tag_save(_tag);	\
diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
index 6d2cb72bf269..e8a66a7c4a6b 100644
--- a/lib/alloc_tag.c
+++ b/lib/alloc_tag.c
@@ -157,6 +157,89 @@  static void __init procfs_init(void)
 	proc_create_seq("allocinfo", 0400, NULL, &allocinfo_seq_op);
 }
 
+#ifdef CONFIG_SLAB_PER_SITE
+static bool ondemand_ready;
+
+void alloc_tag_site_init(struct codetag *ct, bool ondemand)
+{
+	struct alloc_tag *tag = ct_to_alloc_tag(ct);
+	char *name;
+	void *p, *old;
+
+	/* Only handle kmalloc allocations. */
+	if (!tag->meta.sized)
+		return;
+
+	/* Must be ready for on-demand allocations. */
+	if (ondemand && !ondemand_ready)
+		return;
+
+	old = READ_ONCE(tag->meta.cache);
+	/* Already allocated? */
+	if (old)
+		return;
+
+	if (tag->meta.sized < SIZE_MAX) {
+		/* Fixed-size allocations. */
+		name = kasprintf(GFP_KERNEL, "f:%zu:%s:%d", tag->meta.sized, ct->function, ct->lineno);
+		if (WARN_ON_ONCE(!name))
+			return;
+		/*
+		 * As with KMALLOC_NORMAL, the entire allocation needs to be
+		 * open to usercopy access. :(
+		 */
+		p = kmem_cache_create_usercopy(name, tag->meta.sized, 0,
+					       SLAB_NO_MERGE, 0, tag->meta.sized,
+					       NULL);
+	} else {
+		/* Dynamically-size allocations. */
+		name = kasprintf(GFP_KERNEL, "d:%s:%d", ct->function, ct->lineno);
+		if (WARN_ON_ONCE(!name))
+			return;
+		p = kmem_buckets_create(name, SLAB_NO_MERGE, 0, UINT_MAX, NULL);
+	}
+	if (p) {
+		if (unlikely(!try_cmpxchg(&tag->meta.cache, &old, p))) {
+			/* We lost the allocation race; clean up. */
+			if (tag->meta.sized < SIZE_MAX)
+				kmem_cache_destroy(p);
+			else
+				kmem_buckets_destroy(p);
+		}
+	}
+	kfree(name);
+}
+
+static void alloc_tag_site_init_early(struct codetag *ct)
+{
+	/* Explicitly initialize the caches needed to initialize caches. */
+	if (strcmp(ct->function, "kstrdup") == 0 ||
+	    strcmp(ct->function, "kvasprintf") == 0 ||
+	    strcmp(ct->function, "pcpu_mem_zalloc") == 0)
+		alloc_tag_site_init(ct, false);
+
+	/* TODO: pre-allocate GFP_ATOMIC caches here. */
+}
+#endif
+
+static void alloc_tag_module_load(struct codetag_type *cttype,
+				  struct codetag_module *cmod)
+{
+#ifdef CONFIG_SLAB_PER_SITE
+	struct codetag_iterator iter;
+	struct codetag *ct;
+
+	iter = codetag_get_ct_iter(cttype);
+	for (ct = codetag_next_ct(&iter); ct; ct = codetag_next_ct(&iter)) {
+		if (iter.cmod != cmod)
+			continue;
+
+		/* TODO: pre-allocate GFP_ATOMIC caches here. */
+		//alloc_tag_site_init(ct, false);
+	}
+#endif
+}
+
 static bool alloc_tag_module_unload(struct codetag_type *cttype,
 				    struct codetag_module *cmod)
 {
@@ -175,8 +258,21 @@  static bool alloc_tag_module_unload(struct codetag_type *cttype,
 
 		if (WARN(counter.bytes,
 			 "%s:%u module %s func:%s has %llu allocated at module unload",
-			 ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes))
+			 ct->filename, ct->lineno, ct->modname, ct->function, counter.bytes)) {
 			module_unused = false;
+		}
+#ifdef CONFIG_SLAB_PER_SITE
+		else if (tag->meta.sized) {
+			/* Remove the allocated caches, if possible. */
+			void *p = READ_ONCE(tag->meta.cache);
+
+			WRITE_ONCE(tag->meta.cache, NULL);
+			if (tag->meta.sized < SIZE_MAX)
+				kmem_cache_destroy(p);
+			else
+				kmem_buckets_destroy(p);
+		}
+#endif
 	}
 
 	return module_unused;
@@ -260,15 +356,16 @@  static void __init sysctl_init(void)
 static inline void sysctl_init(void) {}
 #endif /* CONFIG_SYSCTL */
 
+static const struct codetag_type_desc alloc_tag_desc = {
+	.section	= "alloc_tags",
+	.tag_size	= sizeof(struct alloc_tag),
+	.module_load	= alloc_tag_module_load,
+	.module_unload	= alloc_tag_module_unload,
+};
+
 static int __init alloc_tag_init(void)
 {
-	const struct codetag_type_desc desc = {
-		.section	= "alloc_tags",
-		.tag_size	= sizeof(struct alloc_tag),
-		.module_unload	= alloc_tag_module_unload,
-	};
-
-	alloc_tag_cttype = codetag_register_type(&desc);
+	alloc_tag_cttype = codetag_register_type(&alloc_tag_desc);
 	if (IS_ERR(alloc_tag_cttype))
 		return PTR_ERR(alloc_tag_cttype);
 
@@ -278,3 +375,11 @@  static int __init alloc_tag_init(void)
 	return 0;
 }
 module_init(alloc_tag_init);
+
+#ifdef CONFIG_SLAB_PER_SITE
+void alloc_tag_early_walk(void)
+{
+	codetag_early_walk(&alloc_tag_desc, alloc_tag_site_init_early);
+	ondemand_ready = true;
+}
+#endif
diff --git a/mm/Kconfig b/mm/Kconfig
index 855c63c3270d..4f01cb6dd32e 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -302,7 +302,20 @@  config SLAB_PER_SITE
 	default SLAB_FREELIST_HARDENED
 	select SLAB_BUCKETS
 	help
-	  Track sizes of kmalloc() call sites.
+	  As a defense against shared-cache "type confusion" use-after-free
+	  attacks, every kmalloc()-family call allocates from a separate
+	  kmem_cache (or when dynamically sized, kmem_buckets). Attackers
+	  will no longer be able to groom malicious objects via similarly
+	  sized allocations that share the same cache as the target object.
+
+	  This increases the "at rest" kmalloc slab memory usage by
+	  roughly 5x (around 7MiB), and adds the potential for greater
+	  long-term memory fragmentation. However, some workloads
+	  actually see performance improvements when single allocation
+	  sites are hot.
+
+	  For a similar defense, see CONFIG_RANDOM_KMALLOC_CACHES, which
+	  has less memory usage overhead, but is probabilistic.
 
 config SLUB_STATS
 	default n
@@ -331,6 +344,7 @@  config SLUB_CPU_PARTIAL
 config RANDOM_KMALLOC_CACHES
 	default n
 	depends on !SLUB_TINY
+	depends on !SLAB_PER_SITE
 	bool "Randomize slab caches for normal kmalloc"
 	help
 	  A hardening feature that creates multiple copies of slab caches for
@@ -345,6 +359,9 @@  config RANDOM_KMALLOC_CACHES
 	  limited degree of memory and CPU overhead that relates to hardware and
 	  system workload.
 
+	  For a similar defense, see CONFIG_SLAB_PER_SITE, which is
+	  deterministic, but has greater memory usage overhead.
+
 endmenu # Slab allocator options
 
 config SHUFFLE_PAGE_ALLOCATOR
diff --git a/mm/slab_common.c b/mm/slab_common.c
index fc698cba0ebe..09506bfa972c 100644
--- a/mm/slab_common.c
+++ b/mm/slab_common.c
@@ -1040,6 +1040,7 @@  void __init create_kmalloc_caches(void)
 		kmem_buckets_cache = kmem_cache_create("kmalloc_buckets",
 						       sizeof(kmem_buckets),
 						       0, SLAB_NO_MERGE, NULL);
+	alloc_tag_early_walk();
 }
 
 /**
diff --git a/mm/slub.c b/mm/slub.c
index 3520acaf9afa..d14102c4b4d7 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -4135,6 +4135,35 @@  void *__kmalloc_large_node_noprof(size_t size, gfp_t flags, int node)
 }
 EXPORT_SYMBOL(__kmalloc_large_node_noprof);
 
+static __always_inline
+struct kmem_cache *choose_slab(size_t size, kmem_buckets *b, gfp_t flags,
+			       unsigned long caller)
+{
+#ifdef CONFIG_SLAB_PER_SITE
+	struct alloc_tag *tag = current->alloc_tag;
+
+	if (!b && tag && tag->meta.sized &&
+	    kmalloc_type(flags, caller) == KMALLOC_NORMAL &&
+	    (flags & GFP_ATOMIC) != GFP_ATOMIC) {
+		void *p = READ_ONCE(tag->meta.cache);
+
+		if (!p && slab_state >= UP) {
+			alloc_tag_site_init(&tag->ct, true);
+			p = READ_ONCE(tag->meta.cache);
+		}
+
+		if (tag->meta.sized < SIZE_MAX) {
+			if (p)
+				return p;
+			/* Otherwise continue with default buckets. */
+		} else {
+			b = p;
+		}
+	}
+#endif
+	return kmalloc_slab(size, b, flags, caller);
+}
+
 static __always_inline
 void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
 			unsigned long caller)
@@ -4152,7 +4181,7 @@  void *__do_kmalloc_node(size_t size, kmem_buckets *b, gfp_t flags, int node,
 	if (unlikely(!size))
 		return ZERO_SIZE_PTR;
 
-	s = kmalloc_slab(size, b, flags, caller);
+	s = choose_slab(size, b, flags, caller);
 
 	ret = slab_alloc_node(s, NULL, flags, node, caller, size);
 	ret = kasan_kmalloc(s, ret, size, flags);