diff mbox series

[v3,4/5] kfence: limit currently covered allocations when pool nearly full

Message ID 20210923104803.2620285-4-elver@google.com (mailing list archive)
State New
Headers show
Series [v3,1/5] stacktrace: move filter_irq_stacks() to kernel/stacktrace.c | expand

Commit Message

Marco Elver Sept. 23, 2021, 10:48 a.m. UTC
One of KFENCE's main design principles is that with increasing uptime,
allocation coverage increases sufficiently to detect previously
undetected bugs.

We have observed that frequent long-lived allocations of the same
source (e.g. pagecache) tend to permanently fill up the KFENCE pool
with increasing system uptime, thus breaking the above requirement.
The workaround thus far had been increasing the sample interval and/or
increasing the KFENCE pool size, but is no reliable solution.

To ensure diverse coverage of allocations, limit currently covered
allocations of the same source once pool utilization reaches 75%
(configurable via `kfence.skip_covered_thresh`) or above. The effect is
retaining reasonable allocation coverage when the pool is close to full.

A side-effect is that this also limits frequent long-lived allocations
of the same source filling up the pool permanently.

Uniqueness of an allocation for coverage purposes is based on its
(partial) allocation stack trace (the source). A Counting Bloom filter
is used to check if an allocation is covered; if the allocation is
currently covered, the allocation is skipped by KFENCE.

Testing was done using:

	(a) a synthetic workload that performs frequent long-lived
	    allocations (default config values; sample_interval=1;
	    num_objects=63), and

	(b) normal desktop workloads on an otherwise idle machine where
	    the problem was first reported after a few days of uptime
	    (default config values).

In both test cases the sampled allocation rate no longer drops to zero
at any point. In the case of (b) we observe (after 2 days uptime) 15%
unique allocations in the pool, 77% pool utilization, with 20% "skipped
allocations (covered)".

Signed-off-by: Marco Elver <elver@google.com>
---
v3:
* Remove unneeded !alloc_stack_hash checks.
* Remove unneeded meta->alloc_stack_hash=0 in kfence_guarded_free().

v2:
* Switch to counting bloom filter to guarantee currently covered
  allocations being skipped.
* Use a module param for skip_covered threshold.
* Use kfence pool address as hash entropy.
* Use filter_irq_stacks().
---
 mm/kfence/core.c   | 103 ++++++++++++++++++++++++++++++++++++++++++++-
 mm/kfence/kfence.h |   2 +
 2 files changed, 103 insertions(+), 2 deletions(-)

Comments

Dmitry Vyukov Sept. 23, 2021, 11:18 a.m. UTC | #1
On Thu, 23 Sept 2021 at 12:48, Marco Elver <elver@google.com> wrote:
>
> One of KFENCE's main design principles is that with increasing uptime,
> allocation coverage increases sufficiently to detect previously
> undetected bugs.
>
> We have observed that frequent long-lived allocations of the same
> source (e.g. pagecache) tend to permanently fill up the KFENCE pool
> with increasing system uptime, thus breaking the above requirement.
> The workaround thus far had been increasing the sample interval and/or
> increasing the KFENCE pool size, but is no reliable solution.
>
> To ensure diverse coverage of allocations, limit currently covered
> allocations of the same source once pool utilization reaches 75%
> (configurable via `kfence.skip_covered_thresh`) or above. The effect is
> retaining reasonable allocation coverage when the pool is close to full.
>
> A side-effect is that this also limits frequent long-lived allocations
> of the same source filling up the pool permanently.
>
> Uniqueness of an allocation for coverage purposes is based on its
> (partial) allocation stack trace (the source). A Counting Bloom filter
> is used to check if an allocation is covered; if the allocation is
> currently covered, the allocation is skipped by KFENCE.
>
> Testing was done using:
>
>         (a) a synthetic workload that performs frequent long-lived
>             allocations (default config values; sample_interval=1;
>             num_objects=63), and
>
>         (b) normal desktop workloads on an otherwise idle machine where
>             the problem was first reported after a few days of uptime
>             (default config values).
>
> In both test cases the sampled allocation rate no longer drops to zero
> at any point. In the case of (b) we observe (after 2 days uptime) 15%
> unique allocations in the pool, 77% pool utilization, with 20% "skipped
> allocations (covered)".
>
> Signed-off-by: Marco Elver <elver@google.com>

Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

> ---
> v3:
> * Remove unneeded !alloc_stack_hash checks.
> * Remove unneeded meta->alloc_stack_hash=0 in kfence_guarded_free().
>
> v2:
> * Switch to counting bloom filter to guarantee currently covered
>   allocations being skipped.
> * Use a module param for skip_covered threshold.
> * Use kfence pool address as hash entropy.
> * Use filter_irq_stacks().
> ---
>  mm/kfence/core.c   | 103 ++++++++++++++++++++++++++++++++++++++++++++-
>  mm/kfence/kfence.h |   2 +
>  2 files changed, 103 insertions(+), 2 deletions(-)
>
> diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> index db01814f8ff0..58a0f6f1acc5 100644
> --- a/mm/kfence/core.c
> +++ b/mm/kfence/core.c
> @@ -11,11 +11,13 @@
>  #include <linux/bug.h>
>  #include <linux/debugfs.h>
>  #include <linux/irq_work.h>
> +#include <linux/jhash.h>
>  #include <linux/kcsan-checks.h>
>  #include <linux/kfence.h>
>  #include <linux/kmemleak.h>
>  #include <linux/list.h>
>  #include <linux/lockdep.h>
> +#include <linux/log2.h>
>  #include <linux/memblock.h>
>  #include <linux/moduleparam.h>
>  #include <linux/random.h>
> @@ -82,6 +84,10 @@ static const struct kernel_param_ops sample_interval_param_ops = {
>  };
>  module_param_cb(sample_interval, &sample_interval_param_ops, &kfence_sample_interval, 0600);
>
> +/* Pool usage% threshold when currently covered allocations are skipped. */
> +static unsigned long kfence_skip_covered_thresh __read_mostly = 75;
> +module_param_named(skip_covered_thresh, kfence_skip_covered_thresh, ulong, 0644);
> +
>  /* The pool of pages used for guard pages and objects. */
>  char *__kfence_pool __ro_after_init;
>  EXPORT_SYMBOL(__kfence_pool); /* Export for test modules. */
> @@ -105,6 +111,25 @@ DEFINE_STATIC_KEY_FALSE(kfence_allocation_key);
>  /* Gates the allocation, ensuring only one succeeds in a given period. */
>  atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
>
> +/*
> + * A Counting Bloom filter of allocation coverage: limits currently covered
> + * allocations of the same source filling up the pool.
> + *
> + * Assuming a range of 15%-85% unique allocations in the pool at any point in
> + * time, the below parameters provide a probablity of 0.02-0.33 for false
> + * positive hits respectively:
> + *
> + *     P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
> + */
> +#define ALLOC_COVERED_HNUM     2
> +#define ALLOC_COVERED_SIZE     (1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
> +#define ALLOC_COVERED_HNEXT(h) (1664525 * (h) + 1013904223)
> +#define ALLOC_COVERED_MASK     (ALLOC_COVERED_SIZE - 1)
> +static atomic_t alloc_covered[ALLOC_COVERED_SIZE];
> +
> +/* Stack depth used to determine uniqueness of an allocation. */
> +#define UNIQUE_ALLOC_STACK_DEPTH 8UL
> +
>  /* Statistics counters for debugfs. */
>  enum kfence_counter_id {
>         KFENCE_COUNTER_ALLOCATED,
> @@ -114,6 +139,7 @@ enum kfence_counter_id {
>         KFENCE_COUNTER_BUGS,
>         KFENCE_COUNTER_SKIP_INCOMPAT,
>         KFENCE_COUNTER_SKIP_CAPACITY,
> +       KFENCE_COUNTER_SKIP_COVERED,
>         KFENCE_COUNTER_COUNT,
>  };
>  static atomic_long_t counters[KFENCE_COUNTER_COUNT];
> @@ -125,11 +151,60 @@ static const char *const counter_names[] = {
>         [KFENCE_COUNTER_BUGS]           = "total bugs",
>         [KFENCE_COUNTER_SKIP_INCOMPAT]  = "skipped allocations (incompatible)",
>         [KFENCE_COUNTER_SKIP_CAPACITY]  = "skipped allocations (capacity)",
> +       [KFENCE_COUNTER_SKIP_COVERED]   = "skipped allocations (covered)",
>  };
>  static_assert(ARRAY_SIZE(counter_names) == KFENCE_COUNTER_COUNT);
>
>  /* === Internals ============================================================ */
>
> +static inline bool should_skip_covered(void)
> +{
> +       unsigned long thresh = (CONFIG_KFENCE_NUM_OBJECTS * kfence_skip_covered_thresh) / 100;
> +
> +       return atomic_long_read(&counters[KFENCE_COUNTER_ALLOCATED]) > thresh;
> +}
> +
> +static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
> +{
> +       /* Some randomness across reboots / different machines. */
> +       u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));
> +
> +       num_entries = min(num_entries, UNIQUE_ALLOC_STACK_DEPTH);
> +       num_entries = filter_irq_stacks(stack_entries, num_entries);
> +       return jhash(stack_entries, num_entries * sizeof(stack_entries[0]), seed);
> +}
> +
> +/*
> + * Adds (or subtracts) count @val for allocation stack trace hash
> + * @alloc_stack_hash from Counting Bloom filter.
> + */
> +static void alloc_covered_add(u32 alloc_stack_hash, int val)
> +{
> +       int i;
> +
> +       for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
> +               atomic_add(val, &alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]);
> +               alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
> +       }
> +}
> +
> +/*
> + * Returns true if the allocation stack trace hash @alloc_stack_hash is
> + * currently contained (non-zero count) in Counting Bloom filter.
> + */
> +static bool alloc_covered_contains(u32 alloc_stack_hash)
> +{
> +       int i;
> +
> +       for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
> +               if (!atomic_read(&alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]))
> +                       return false;
> +               alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
> +       }
> +
> +       return true;
> +}
> +
>  static bool kfence_protect(unsigned long addr)
>  {
>         return !KFENCE_WARN_ON(!kfence_protect_page(ALIGN_DOWN(addr, PAGE_SIZE), true));
> @@ -269,7 +344,8 @@ static __always_inline void for_each_canary(const struct kfence_metadata *meta,
>  }
>
>  static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t gfp,
> -                                 unsigned long *stack_entries, size_t num_stack_entries)
> +                                 unsigned long *stack_entries, size_t num_stack_entries,
> +                                 u32 alloc_stack_hash)
>  {
>         struct kfence_metadata *meta = NULL;
>         unsigned long flags;
> @@ -332,6 +408,8 @@ static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
>         /* Pairs with READ_ONCE() in kfence_shutdown_cache(). */
>         WRITE_ONCE(meta->cache, cache);
>         meta->size = size;
> +       meta->alloc_stack_hash = alloc_stack_hash;
> +
>         for_each_canary(meta, set_canary_byte);
>
>         /* Set required struct page fields. */
> @@ -344,6 +422,8 @@ static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
>
>         raw_spin_unlock_irqrestore(&meta->lock, flags);
>
> +       alloc_covered_add(alloc_stack_hash, 1);
> +
>         /* Memory initialization. */
>
>         /*
> @@ -412,6 +492,8 @@ static void kfence_guarded_free(void *addr, struct kfence_metadata *meta, bool z
>
>         raw_spin_unlock_irqrestore(&meta->lock, flags);
>
> +       alloc_covered_add(meta->alloc_stack_hash, -1);
> +
>         /* Protect to detect use-after-frees. */
>         kfence_protect((unsigned long)addr);
>
> @@ -752,6 +834,7 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
>  {
>         unsigned long stack_entries[KFENCE_STACK_DEPTH];
>         size_t num_stack_entries;
> +       u32 alloc_stack_hash;
>
>         /*
>          * Perform size check before switching kfence_allocation_gate, so that
> @@ -799,7 +882,23 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
>
>         num_stack_entries = stack_trace_save(stack_entries, KFENCE_STACK_DEPTH, 0);
>
> -       return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries);
> +       /*
> +        * Do expensive check for coverage of allocation in slow-path after
> +        * allocation_gate has already become non-zero, even though it might
> +        * mean not making any allocation within a given sample interval.
> +        *
> +        * This ensures reasonable allocation coverage when the pool is almost
> +        * full, including avoiding long-lived allocations of the same source
> +        * filling up the pool (e.g. pagecache allocations).
> +        */
> +       alloc_stack_hash = get_alloc_stack_hash(stack_entries, num_stack_entries);
> +       if (should_skip_covered() && alloc_covered_contains(alloc_stack_hash)) {
> +               atomic_long_inc(&counters[KFENCE_COUNTER_SKIP_COVERED]);
> +               return NULL;
> +       }
> +
> +       return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries,
> +                                   alloc_stack_hash);
>  }
>
>  size_t kfence_ksize(const void *addr)
> diff --git a/mm/kfence/kfence.h b/mm/kfence/kfence.h
> index c1f23c61e5f9..2a2d5de9d379 100644
> --- a/mm/kfence/kfence.h
> +++ b/mm/kfence/kfence.h
> @@ -87,6 +87,8 @@ struct kfence_metadata {
>         /* Allocation and free stack information. */
>         struct kfence_track alloc_track;
>         struct kfence_track free_track;
> +       /* For updating alloc_covered on frees. */
> +       u32 alloc_stack_hash;
>  };
>
>  extern struct kfence_metadata kfence_metadata[CONFIG_KFENCE_NUM_OBJECTS];
> --
> 2.33.0.464.g1972c5931b-goog
>
Alexander Potapenko Sept. 23, 2021, 1:23 p.m. UTC | #2
On Thu, Sep 23, 2021 at 1:19 PM Dmitry Vyukov <dvyukov@google.com> wrote:
>
> On Thu, 23 Sept 2021 at 12:48, Marco Elver <elver@google.com> wrote:
> >
> > One of KFENCE's main design principles is that with increasing uptime,
> > allocation coverage increases sufficiently to detect previously
> > undetected bugs.
> >
> > We have observed that frequent long-lived allocations of the same
> > source (e.g. pagecache) tend to permanently fill up the KFENCE pool
> > with increasing system uptime, thus breaking the above requirement.
> > The workaround thus far had been increasing the sample interval and/or
> > increasing the KFENCE pool size, but is no reliable solution.
> >
> > To ensure diverse coverage of allocations, limit currently covered
> > allocations of the same source once pool utilization reaches 75%
> > (configurable via `kfence.skip_covered_thresh`) or above. The effect is
> > retaining reasonable allocation coverage when the pool is close to full.
> >
> > A side-effect is that this also limits frequent long-lived allocations
> > of the same source filling up the pool permanently.
> >
> > Uniqueness of an allocation for coverage purposes is based on its
> > (partial) allocation stack trace (the source). A Counting Bloom filter
> > is used to check if an allocation is covered; if the allocation is
> > currently covered, the allocation is skipped by KFENCE.
> >
> > Testing was done using:
> >
> >         (a) a synthetic workload that performs frequent long-lived
> >             allocations (default config values; sample_interval=1;
> >             num_objects=63), and
> >
> >         (b) normal desktop workloads on an otherwise idle machine where
> >             the problem was first reported after a few days of uptime
> >             (default config values).
> >
> > In both test cases the sampled allocation rate no longer drops to zero
> > at any point. In the case of (b) we observe (after 2 days uptime) 15%
> > unique allocations in the pool, 77% pool utilization, with 20% "skipped
> > allocations (covered)".
> >
> > Signed-off-by: Marco Elver <elver@google.com>
>
> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Alexander Potapenko <glider@google.com>

>
> > ---
> > v3:
> > * Remove unneeded !alloc_stack_hash checks.
> > * Remove unneeded meta->alloc_stack_hash=0 in kfence_guarded_free().
> >
> > v2:
> > * Switch to counting bloom filter to guarantee currently covered
> >   allocations being skipped.
> > * Use a module param for skip_covered threshold.
> > * Use kfence pool address as hash entropy.
> > * Use filter_irq_stacks().
> > ---
> >  mm/kfence/core.c   | 103 ++++++++++++++++++++++++++++++++++++++++++++-
> >  mm/kfence/kfence.h |   2 +
> >  2 files changed, 103 insertions(+), 2 deletions(-)
> >
> > diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> > index db01814f8ff0..58a0f6f1acc5 100644
> > --- a/mm/kfence/core.c
> > +++ b/mm/kfence/core.c
> > @@ -11,11 +11,13 @@
> >  #include <linux/bug.h>
> >  #include <linux/debugfs.h>
> >  #include <linux/irq_work.h>
> > +#include <linux/jhash.h>
> >  #include <linux/kcsan-checks.h>
> >  #include <linux/kfence.h>
> >  #include <linux/kmemleak.h>
> >  #include <linux/list.h>
> >  #include <linux/lockdep.h>
> > +#include <linux/log2.h>
> >  #include <linux/memblock.h>
> >  #include <linux/moduleparam.h>
> >  #include <linux/random.h>
> > @@ -82,6 +84,10 @@ static const struct kernel_param_ops sample_interval_param_ops = {
> >  };
> >  module_param_cb(sample_interval, &sample_interval_param_ops, &kfence_sample_interval, 0600);
> >
> > +/* Pool usage% threshold when currently covered allocations are skipped. */
> > +static unsigned long kfence_skip_covered_thresh __read_mostly = 75;
> > +module_param_named(skip_covered_thresh, kfence_skip_covered_thresh, ulong, 0644);
> > +
> >  /* The pool of pages used for guard pages and objects. */
> >  char *__kfence_pool __ro_after_init;
> >  EXPORT_SYMBOL(__kfence_pool); /* Export for test modules. */
> > @@ -105,6 +111,25 @@ DEFINE_STATIC_KEY_FALSE(kfence_allocation_key);
> >  /* Gates the allocation, ensuring only one succeeds in a given period. */
> >  atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
> >
> > +/*
> > + * A Counting Bloom filter of allocation coverage: limits currently covered
> > + * allocations of the same source filling up the pool.
> > + *
> > + * Assuming a range of 15%-85% unique allocations in the pool at any point in

Where do these 85% come from?

> > + * time, the below parameters provide a probablity of 0.02-0.33 for false
> > + * positive hits respectively:
> > + *
> > + *     P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
> > + */
> > +#define ALLOC_COVERED_HNUM     2
> > +#define ALLOC_COVERED_SIZE     (1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
> > +#define ALLOC_COVERED_HNEXT(h) (1664525 * (h) + 1013904223)

Unless we are planning to change these primes, can you use
next_pseudo_random32() instead?


> > +#define ALLOC_COVERED_MASK     (ALLOC_COVERED_SIZE - 1)
> > +static atomic_t alloc_covered[ALLOC_COVERED_SIZE];
> > +
> > +/* Stack depth used to determine uniqueness of an allocation. */
> > +#define UNIQUE_ALLOC_STACK_DEPTH 8UL
> > +
> >  /* Statistics counters for debugfs. */
> >  enum kfence_counter_id {
> >         KFENCE_COUNTER_ALLOCATED,
> > @@ -114,6 +139,7 @@ enum kfence_counter_id {
> >         KFENCE_COUNTER_BUGS,
> >         KFENCE_COUNTER_SKIP_INCOMPAT,
> >         KFENCE_COUNTER_SKIP_CAPACITY,
> > +       KFENCE_COUNTER_SKIP_COVERED,
> >         KFENCE_COUNTER_COUNT,
> >  };
> >  static atomic_long_t counters[KFENCE_COUNTER_COUNT];
> > @@ -125,11 +151,60 @@ static const char *const counter_names[] = {
> >         [KFENCE_COUNTER_BUGS]           = "total bugs",
> >         [KFENCE_COUNTER_SKIP_INCOMPAT]  = "skipped allocations (incompatible)",
> >         [KFENCE_COUNTER_SKIP_CAPACITY]  = "skipped allocations (capacity)",
> > +       [KFENCE_COUNTER_SKIP_COVERED]   = "skipped allocations (covered)",
> >  };
> >  static_assert(ARRAY_SIZE(counter_names) == KFENCE_COUNTER_COUNT);
> >
> >  /* === Internals ============================================================ */
> >
> > +static inline bool should_skip_covered(void)
> > +{
> > +       unsigned long thresh = (CONFIG_KFENCE_NUM_OBJECTS * kfence_skip_covered_thresh) / 100;
> > +
> > +       return atomic_long_read(&counters[KFENCE_COUNTER_ALLOCATED]) > thresh;
> > +}
> > +
> > +static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
> > +{
> > +       /* Some randomness across reboots / different machines. */
> > +       u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));
> > +
> > +       num_entries = min(num_entries, UNIQUE_ALLOC_STACK_DEPTH);
> > +       num_entries = filter_irq_stacks(stack_entries, num_entries);
> > +       return jhash(stack_entries, num_entries * sizeof(stack_entries[0]), seed);
> > +}
> > +
> > +/*
> > + * Adds (or subtracts) count @val for allocation stack trace hash
> > + * @alloc_stack_hash from Counting Bloom filter.
> > + */
> > +static void alloc_covered_add(u32 alloc_stack_hash, int val)
> > +{
> > +       int i;
> > +
> > +       for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
> > +               atomic_add(val, &alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]);
> > +               alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
> > +       }
> > +}
> > +
> > +/*
> > + * Returns true if the allocation stack trace hash @alloc_stack_hash is
> > + * currently contained (non-zero count) in Counting Bloom filter.
> > + */
> > +static bool alloc_covered_contains(u32 alloc_stack_hash)
> > +{
> > +       int i;
> > +
> > +       for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
> > +               if (!atomic_read(&alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]))
> > +                       return false;
> > +               alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
> > +       }
> > +
> > +       return true;
> > +}
> > +
> >  static bool kfence_protect(unsigned long addr)
> >  {
> >         return !KFENCE_WARN_ON(!kfence_protect_page(ALIGN_DOWN(addr, PAGE_SIZE), true));
> > @@ -269,7 +344,8 @@ static __always_inline void for_each_canary(const struct kfence_metadata *meta,
> >  }
> >
> >  static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t gfp,
> > -                                 unsigned long *stack_entries, size_t num_stack_entries)
> > +                                 unsigned long *stack_entries, size_t num_stack_entries,
> > +                                 u32 alloc_stack_hash)
> >  {
> >         struct kfence_metadata *meta = NULL;
> >         unsigned long flags;
> > @@ -332,6 +408,8 @@ static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
> >         /* Pairs with READ_ONCE() in kfence_shutdown_cache(). */
> >         WRITE_ONCE(meta->cache, cache);
> >         meta->size = size;
> > +       meta->alloc_stack_hash = alloc_stack_hash;
> > +
> >         for_each_canary(meta, set_canary_byte);
> >
> >         /* Set required struct page fields. */
> > @@ -344,6 +422,8 @@ static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
> >
> >         raw_spin_unlock_irqrestore(&meta->lock, flags);
> >
> > +       alloc_covered_add(alloc_stack_hash, 1);
> > +
> >         /* Memory initialization. */
> >
> >         /*
> > @@ -412,6 +492,8 @@ static void kfence_guarded_free(void *addr, struct kfence_metadata *meta, bool z
> >
> >         raw_spin_unlock_irqrestore(&meta->lock, flags);
> >
> > +       alloc_covered_add(meta->alloc_stack_hash, -1);
> > +
> >         /* Protect to detect use-after-frees. */
> >         kfence_protect((unsigned long)addr);
> >
> > @@ -752,6 +834,7 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
> >  {
> >         unsigned long stack_entries[KFENCE_STACK_DEPTH];
> >         size_t num_stack_entries;
> > +       u32 alloc_stack_hash;
> >
> >         /*
> >          * Perform size check before switching kfence_allocation_gate, so that
> > @@ -799,7 +882,23 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
> >
> >         num_stack_entries = stack_trace_save(stack_entries, KFENCE_STACK_DEPTH, 0);
> >
> > -       return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries);
> > +       /*
> > +        * Do expensive check for coverage of allocation in slow-path after
> > +        * allocation_gate has already become non-zero, even though it might
> > +        * mean not making any allocation within a given sample interval.
> > +        *
> > +        * This ensures reasonable allocation coverage when the pool is almost
> > +        * full, including avoiding long-lived allocations of the same source
> > +        * filling up the pool (e.g. pagecache allocations).
> > +        */
> > +       alloc_stack_hash = get_alloc_stack_hash(stack_entries, num_stack_entries);
> > +       if (should_skip_covered() && alloc_covered_contains(alloc_stack_hash)) {
> > +               atomic_long_inc(&counters[KFENCE_COUNTER_SKIP_COVERED]);
> > +               return NULL;
> > +       }
> > +
> > +       return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries,
> > +                                   alloc_stack_hash);
> >  }
> >
> >  size_t kfence_ksize(const void *addr)
> > diff --git a/mm/kfence/kfence.h b/mm/kfence/kfence.h
> > index c1f23c61e5f9..2a2d5de9d379 100644
> > --- a/mm/kfence/kfence.h
> > +++ b/mm/kfence/kfence.h
> > @@ -87,6 +87,8 @@ struct kfence_metadata {
> >         /* Allocation and free stack information. */
> >         struct kfence_track alloc_track;
> >         struct kfence_track free_track;
> > +       /* For updating alloc_covered on frees. */
> > +       u32 alloc_stack_hash;
> >  };
> >
> >  extern struct kfence_metadata kfence_metadata[CONFIG_KFENCE_NUM_OBJECTS];
> > --
> > 2.33.0.464.g1972c5931b-goog
> >



--
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
Marco Elver Sept. 23, 2021, 1:44 p.m. UTC | #3
On Thu, 23 Sept 2021 at 15:24, Alexander Potapenko <glider@google.com> wrote:
>
> On Thu, Sep 23, 2021 at 1:19 PM Dmitry Vyukov <dvyukov@google.com> wrote:
> >
> > On Thu, 23 Sept 2021 at 12:48, Marco Elver <elver@google.com> wrote:
> > >
> > > One of KFENCE's main design principles is that with increasing uptime,
> > > allocation coverage increases sufficiently to detect previously
> > > undetected bugs.
> > >
> > > We have observed that frequent long-lived allocations of the same
> > > source (e.g. pagecache) tend to permanently fill up the KFENCE pool
> > > with increasing system uptime, thus breaking the above requirement.
> > > The workaround thus far had been increasing the sample interval and/or
> > > increasing the KFENCE pool size, but is no reliable solution.
> > >
> > > To ensure diverse coverage of allocations, limit currently covered
> > > allocations of the same source once pool utilization reaches 75%
> > > (configurable via `kfence.skip_covered_thresh`) or above. The effect is
> > > retaining reasonable allocation coverage when the pool is close to full.
> > >
> > > A side-effect is that this also limits frequent long-lived allocations
> > > of the same source filling up the pool permanently.
> > >
> > > Uniqueness of an allocation for coverage purposes is based on its
> > > (partial) allocation stack trace (the source). A Counting Bloom filter
> > > is used to check if an allocation is covered; if the allocation is
> > > currently covered, the allocation is skipped by KFENCE.
> > >
> > > Testing was done using:
> > >
> > >         (a) a synthetic workload that performs frequent long-lived
> > >             allocations (default config values; sample_interval=1;
> > >             num_objects=63), and
> > >
> > >         (b) normal desktop workloads on an otherwise idle machine where
> > >             the problem was first reported after a few days of uptime
> > >             (default config values).
> > >
> > > In both test cases the sampled allocation rate no longer drops to zero
> > > at any point. In the case of (b) we observe (after 2 days uptime) 15%
> > > unique allocations in the pool, 77% pool utilization, with 20% "skipped
> > > allocations (covered)".
> > >
> > > Signed-off-by: Marco Elver <elver@google.com>
> >
> > Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
> Acked-by: Alexander Potapenko <glider@google.com>

Thank you both!

> > > ---
> > > v3:
> > > * Remove unneeded !alloc_stack_hash checks.
> > > * Remove unneeded meta->alloc_stack_hash=0 in kfence_guarded_free().
> > >
> > > v2:
> > > * Switch to counting bloom filter to guarantee currently covered
> > >   allocations being skipped.
> > > * Use a module param for skip_covered threshold.
> > > * Use kfence pool address as hash entropy.
> > > * Use filter_irq_stacks().
> > > ---
> > >  mm/kfence/core.c   | 103 ++++++++++++++++++++++++++++++++++++++++++++-
> > >  mm/kfence/kfence.h |   2 +
> > >  2 files changed, 103 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> > > index db01814f8ff0..58a0f6f1acc5 100644
> > > --- a/mm/kfence/core.c
> > > +++ b/mm/kfence/core.c
> > > @@ -11,11 +11,13 @@
> > >  #include <linux/bug.h>
> > >  #include <linux/debugfs.h>
> > >  #include <linux/irq_work.h>
> > > +#include <linux/jhash.h>
> > >  #include <linux/kcsan-checks.h>
> > >  #include <linux/kfence.h>
> > >  #include <linux/kmemleak.h>
> > >  #include <linux/list.h>
> > >  #include <linux/lockdep.h>
> > > +#include <linux/log2.h>
> > >  #include <linux/memblock.h>
> > >  #include <linux/moduleparam.h>
> > >  #include <linux/random.h>
> > > @@ -82,6 +84,10 @@ static const struct kernel_param_ops sample_interval_param_ops = {
> > >  };
> > >  module_param_cb(sample_interval, &sample_interval_param_ops, &kfence_sample_interval, 0600);
> > >
> > > +/* Pool usage% threshold when currently covered allocations are skipped. */
> > > +static unsigned long kfence_skip_covered_thresh __read_mostly = 75;
> > > +module_param_named(skip_covered_thresh, kfence_skip_covered_thresh, ulong, 0644);
> > > +
> > >  /* The pool of pages used for guard pages and objects. */
> > >  char *__kfence_pool __ro_after_init;
> > >  EXPORT_SYMBOL(__kfence_pool); /* Export for test modules. */
> > > @@ -105,6 +111,25 @@ DEFINE_STATIC_KEY_FALSE(kfence_allocation_key);
> > >  /* Gates the allocation, ensuring only one succeeds in a given period. */
> > >  atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
> > >
> > > +/*
> > > + * A Counting Bloom filter of allocation coverage: limits currently covered
> > > + * allocations of the same source filling up the pool.
> > > + *
> > > + * Assuming a range of 15%-85% unique allocations in the pool at any point in
>
> Where do these 85% come from?

An imaginary worst case, just to illustrate the range of the false
positive probabilities (in the case of 85% it'd be 0.33). I expect
unique allocations to be around 10-15% on a freshly booted system (on
my real-system-experiment it stayed below 15%), but other workloads
may produce other unique allocations%.

> > > + * time, the below parameters provide a probablity of 0.02-0.33 for false
> > > + * positive hits respectively:
> > > + *
> > > + *     P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
> > > + */
> > > +#define ALLOC_COVERED_HNUM     2
> > > +#define ALLOC_COVERED_SIZE     (1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
> > > +#define ALLOC_COVERED_HNEXT(h) (1664525 * (h) + 1013904223)
>
> Unless we are planning to change these primes, can you use
> next_pseudo_random32() instead?

I'm worried about next_pseudo_random32() changing their implementation
to longer be deterministic or change in other ways that break our
usecase. In this case we want pseudorandomness, but we're not
implementing a PRNG.

Open-coding the constants (given they are from "Numerical Recipes") is
more reliable and doesn't introduce unwanted reliance on
next_pseudo_random32()'s behaviour.

Thanks,
-- Marco
Alexander Potapenko Sept. 23, 2021, 1:46 p.m. UTC | #4
On Thu, Sep 23, 2021 at 3:44 PM Marco Elver <elver@google.com> wrote:
>
> On Thu, 23 Sept 2021 at 15:24, Alexander Potapenko <glider@google.com> wrote:
> >
> > On Thu, Sep 23, 2021 at 1:19 PM Dmitry Vyukov <dvyukov@google.com> wrote:
> > >
> > > On Thu, 23 Sept 2021 at 12:48, Marco Elver <elver@google.com> wrote:
> > > >
> > > > One of KFENCE's main design principles is that with increasing uptime,
> > > > allocation coverage increases sufficiently to detect previously
> > > > undetected bugs.
> > > >
> > > > We have observed that frequent long-lived allocations of the same
> > > > source (e.g. pagecache) tend to permanently fill up the KFENCE pool
> > > > with increasing system uptime, thus breaking the above requirement.
> > > > The workaround thus far had been increasing the sample interval and/or
> > > > increasing the KFENCE pool size, but is no reliable solution.
> > > >
> > > > To ensure diverse coverage of allocations, limit currently covered
> > > > allocations of the same source once pool utilization reaches 75%
> > > > (configurable via `kfence.skip_covered_thresh`) or above. The effect is
> > > > retaining reasonable allocation coverage when the pool is close to full.
> > > >
> > > > A side-effect is that this also limits frequent long-lived allocations
> > > > of the same source filling up the pool permanently.
> > > >
> > > > Uniqueness of an allocation for coverage purposes is based on its
> > > > (partial) allocation stack trace (the source). A Counting Bloom filter
> > > > is used to check if an allocation is covered; if the allocation is
> > > > currently covered, the allocation is skipped by KFENCE.
> > > >
> > > > Testing was done using:
> > > >
> > > >         (a) a synthetic workload that performs frequent long-lived
> > > >             allocations (default config values; sample_interval=1;
> > > >             num_objects=63), and
> > > >
> > > >         (b) normal desktop workloads on an otherwise idle machine where
> > > >             the problem was first reported after a few days of uptime
> > > >             (default config values).
> > > >
> > > > In both test cases the sampled allocation rate no longer drops to zero
> > > > at any point. In the case of (b) we observe (after 2 days uptime) 15%
> > > > unique allocations in the pool, 77% pool utilization, with 20% "skipped
> > > > allocations (covered)".
> > > >
> > > > Signed-off-by: Marco Elver <elver@google.com>
> > >
> > > Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
> > Acked-by: Alexander Potapenko <glider@google.com>
>
> Thank you both!
>
> > > > ---
> > > > v3:
> > > > * Remove unneeded !alloc_stack_hash checks.
> > > > * Remove unneeded meta->alloc_stack_hash=0 in kfence_guarded_free().
> > > >
> > > > v2:
> > > > * Switch to counting bloom filter to guarantee currently covered
> > > >   allocations being skipped.
> > > > * Use a module param for skip_covered threshold.
> > > > * Use kfence pool address as hash entropy.
> > > > * Use filter_irq_stacks().
> > > > ---
> > > >  mm/kfence/core.c   | 103 ++++++++++++++++++++++++++++++++++++++++++++-
> > > >  mm/kfence/kfence.h |   2 +
> > > >  2 files changed, 103 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> > > > index db01814f8ff0..58a0f6f1acc5 100644
> > > > --- a/mm/kfence/core.c
> > > > +++ b/mm/kfence/core.c
> > > > @@ -11,11 +11,13 @@
> > > >  #include <linux/bug.h>
> > > >  #include <linux/debugfs.h>
> > > >  #include <linux/irq_work.h>
> > > > +#include <linux/jhash.h>
> > > >  #include <linux/kcsan-checks.h>
> > > >  #include <linux/kfence.h>
> > > >  #include <linux/kmemleak.h>
> > > >  #include <linux/list.h>
> > > >  #include <linux/lockdep.h>
> > > > +#include <linux/log2.h>
> > > >  #include <linux/memblock.h>
> > > >  #include <linux/moduleparam.h>
> > > >  #include <linux/random.h>
> > > > @@ -82,6 +84,10 @@ static const struct kernel_param_ops sample_interval_param_ops = {
> > > >  };
> > > >  module_param_cb(sample_interval, &sample_interval_param_ops, &kfence_sample_interval, 0600);
> > > >
> > > > +/* Pool usage% threshold when currently covered allocations are skipped. */
> > > > +static unsigned long kfence_skip_covered_thresh __read_mostly = 75;
> > > > +module_param_named(skip_covered_thresh, kfence_skip_covered_thresh, ulong, 0644);
> > > > +
> > > >  /* The pool of pages used for guard pages and objects. */
> > > >  char *__kfence_pool __ro_after_init;
> > > >  EXPORT_SYMBOL(__kfence_pool); /* Export for test modules. */
> > > > @@ -105,6 +111,25 @@ DEFINE_STATIC_KEY_FALSE(kfence_allocation_key);
> > > >  /* Gates the allocation, ensuring only one succeeds in a given period. */
> > > >  atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
> > > >
> > > > +/*
> > > > + * A Counting Bloom filter of allocation coverage: limits currently covered
> > > > + * allocations of the same source filling up the pool.
> > > > + *
> > > > + * Assuming a range of 15%-85% unique allocations in the pool at any point in
> >
> > Where do these 85% come from?
>
> An imaginary worst case, just to illustrate the range of the false
> positive probabilities (in the case of 85% it'd be 0.33). I expect
> unique allocations to be around 10-15% on a freshly booted system (on
> my real-system-experiment it stayed below 15%), but other workloads
> may produce other unique allocations%.
>
> > > > + * time, the below parameters provide a probablity of 0.02-0.33 for false
> > > > + * positive hits respectively:
> > > > + *
> > > > + *     P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
> > > > + */
> > > > +#define ALLOC_COVERED_HNUM     2
> > > > +#define ALLOC_COVERED_SIZE     (1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
> > > > +#define ALLOC_COVERED_HNEXT(h) (1664525 * (h) + 1013904223)
> >
> > Unless we are planning to change these primes, can you use
> > next_pseudo_random32() instead?
>
> I'm worried about next_pseudo_random32() changing their implementation
> to longer be deterministic or change in other ways that break our
> usecase. In this case we want pseudorandomness, but we're not
> implementing a PRNG.
>
> Open-coding the constants (given they are from "Numerical Recipes") is
> more reliable and doesn't introduce unwanted reliance on
> next_pseudo_random32()'s behaviour.

Okay, fair enough.

>
> Thanks,
> -- Marco
Andrew Morton Sept. 23, 2021, 11:28 p.m. UTC | #5
On Thu, 23 Sep 2021 15:44:10 +0200 Marco Elver <elver@google.com> wrote:

> > > > + * time, the below parameters provide a probablity of 0.02-0.33 for false
> > > > + * positive hits respectively:
> > > > + *
> > > > + *     P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
> > > > + */
> > > > +#define ALLOC_COVERED_HNUM     2
> > > > +#define ALLOC_COVERED_SIZE     (1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
> > > > +#define ALLOC_COVERED_HNEXT(h) (1664525 * (h) + 1013904223)
> >
> > Unless we are planning to change these primes, can you use
> > next_pseudo_random32() instead?
> 
> I'm worried about next_pseudo_random32() changing their implementation
> to longer be deterministic or change in other ways that break our
> usecase. In this case we want pseudorandomness, but we're not
> implementing a PRNG.
> 
> Open-coding the constants (given they are from "Numerical Recipes") is
> more reliable and doesn't introduce unwanted reliance on
> next_pseudo_random32()'s behaviour.

Perhaps we could summarize this in an additional comment?

Also, this:

+static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
+{
+	/* Some randomness across reboots / different machines. */
+	u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));

seems a bit weak.  Would it be better to seed this at boot time with
a randomish number?
Marco Elver Sept. 24, 2021, 1:01 p.m. UTC | #6
On Thu, Sep 23, 2021 at 04:28PM -0700, Andrew Morton wrote:
> On Thu, 23 Sep 2021 15:44:10 +0200 Marco Elver <elver@google.com> wrote:
[...]
> > I'm worried about next_pseudo_random32() changing their implementation
> > to longer be deterministic or change in other ways that break our
> > usecase. In this case we want pseudorandomness, but we're not
> > implementing a PRNG.
> > 
> > Open-coding the constants (given they are from "Numerical Recipes") is
> > more reliable and doesn't introduce unwanted reliance on
> > next_pseudo_random32()'s behaviour.
> 
> Perhaps we could summarize this in an additional comment?

Hmm, on second thought, while trying to write the comment realized it's
unnecessary altogether. I've switched to just using hash_32() which is
probably better suited.

> Also, this:
> 
> +static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
> +{
> +	/* Some randomness across reboots / different machines. */
> +	u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));
> 
> seems a bit weak.  Would it be better to seed this at boot time with
> a randomish number?

Sure, makes sense.

Both fixes are included in the below fixup. (Let me know if resending as
v4 is better, but I've seen the patches already appeared in -mm.)

Thank you!

-- Marco

------ >8 ------

From: Marco Elver <elver@google.com>
Date: Fri, 24 Sep 2021 14:17:38 +0200
Subject: [PATCH] fixup! kfence: limit currently covered allocations when pool
 nearly full

* Simplify and just use hash_32().
* Use more random stack_hash_seed.

Signed-off-by: Marco Elver <elver@google.com>
---
 mm/kfence/core.c | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index 58a0f6f1acc5..545999d04af4 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -10,6 +10,7 @@
 #include <linux/atomic.h>
 #include <linux/bug.h>
 #include <linux/debugfs.h>
+#include <linux/hash.h>
 #include <linux/irq_work.h>
 #include <linux/jhash.h>
 #include <linux/kcsan-checks.h>
@@ -122,14 +123,21 @@ atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
  *	P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
  */
 #define ALLOC_COVERED_HNUM	2
-#define ALLOC_COVERED_SIZE	(1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
-#define ALLOC_COVERED_HNEXT(h)	(1664525 * (h) + 1013904223)
+#define ALLOC_COVERED_ORDER	(const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2)
+#define ALLOC_COVERED_SIZE	(1 << ALLOC_COVERED_ORDER)
+#define ALLOC_COVERED_HNEXT(h)	hash_32(h, ALLOC_COVERED_ORDER)
 #define ALLOC_COVERED_MASK	(ALLOC_COVERED_SIZE - 1)
 static atomic_t alloc_covered[ALLOC_COVERED_SIZE];
 
 /* Stack depth used to determine uniqueness of an allocation. */
 #define UNIQUE_ALLOC_STACK_DEPTH 8UL
 
+/*
+ * Randomness for stack hashes, making the same collisions across reboots and
+ * different machines less likely.
+ */
+static u32 stack_hash_seed __ro_after_init;
+
 /* Statistics counters for debugfs. */
 enum kfence_counter_id {
 	KFENCE_COUNTER_ALLOCATED,
@@ -166,12 +174,9 @@ static inline bool should_skip_covered(void)
 
 static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
 {
-	/* Some randomness across reboots / different machines. */
-	u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));
-
 	num_entries = min(num_entries, UNIQUE_ALLOC_STACK_DEPTH);
 	num_entries = filter_irq_stacks(stack_entries, num_entries);
-	return jhash(stack_entries, num_entries * sizeof(stack_entries[0]), seed);
+	return jhash(stack_entries, num_entries * sizeof(stack_entries[0]), stack_hash_seed);
 }
 
 /*
@@ -759,6 +764,7 @@ void __init kfence_init(void)
 	if (!kfence_sample_interval)
 		return;
 
+	stack_hash_seed = (u32)random_get_entropy();
 	if (!kfence_init_pool()) {
 		pr_err("%s failed\n", __func__);
 		return;
diff mbox series

Patch

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index db01814f8ff0..58a0f6f1acc5 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -11,11 +11,13 @@ 
 #include <linux/bug.h>
 #include <linux/debugfs.h>
 #include <linux/irq_work.h>
+#include <linux/jhash.h>
 #include <linux/kcsan-checks.h>
 #include <linux/kfence.h>
 #include <linux/kmemleak.h>
 #include <linux/list.h>
 #include <linux/lockdep.h>
+#include <linux/log2.h>
 #include <linux/memblock.h>
 #include <linux/moduleparam.h>
 #include <linux/random.h>
@@ -82,6 +84,10 @@  static const struct kernel_param_ops sample_interval_param_ops = {
 };
 module_param_cb(sample_interval, &sample_interval_param_ops, &kfence_sample_interval, 0600);
 
+/* Pool usage% threshold when currently covered allocations are skipped. */
+static unsigned long kfence_skip_covered_thresh __read_mostly = 75;
+module_param_named(skip_covered_thresh, kfence_skip_covered_thresh, ulong, 0644);
+
 /* The pool of pages used for guard pages and objects. */
 char *__kfence_pool __ro_after_init;
 EXPORT_SYMBOL(__kfence_pool); /* Export for test modules. */
@@ -105,6 +111,25 @@  DEFINE_STATIC_KEY_FALSE(kfence_allocation_key);
 /* Gates the allocation, ensuring only one succeeds in a given period. */
 atomic_t kfence_allocation_gate = ATOMIC_INIT(1);
 
+/*
+ * A Counting Bloom filter of allocation coverage: limits currently covered
+ * allocations of the same source filling up the pool.
+ *
+ * Assuming a range of 15%-85% unique allocations in the pool at any point in
+ * time, the below parameters provide a probablity of 0.02-0.33 for false
+ * positive hits respectively:
+ *
+ *	P(alloc_traces) = (1 - e^(-HNUM * (alloc_traces / SIZE)) ^ HNUM
+ */
+#define ALLOC_COVERED_HNUM	2
+#define ALLOC_COVERED_SIZE	(1 << (const_ilog2(CONFIG_KFENCE_NUM_OBJECTS) + 2))
+#define ALLOC_COVERED_HNEXT(h)	(1664525 * (h) + 1013904223)
+#define ALLOC_COVERED_MASK	(ALLOC_COVERED_SIZE - 1)
+static atomic_t alloc_covered[ALLOC_COVERED_SIZE];
+
+/* Stack depth used to determine uniqueness of an allocation. */
+#define UNIQUE_ALLOC_STACK_DEPTH 8UL
+
 /* Statistics counters for debugfs. */
 enum kfence_counter_id {
 	KFENCE_COUNTER_ALLOCATED,
@@ -114,6 +139,7 @@  enum kfence_counter_id {
 	KFENCE_COUNTER_BUGS,
 	KFENCE_COUNTER_SKIP_INCOMPAT,
 	KFENCE_COUNTER_SKIP_CAPACITY,
+	KFENCE_COUNTER_SKIP_COVERED,
 	KFENCE_COUNTER_COUNT,
 };
 static atomic_long_t counters[KFENCE_COUNTER_COUNT];
@@ -125,11 +151,60 @@  static const char *const counter_names[] = {
 	[KFENCE_COUNTER_BUGS]		= "total bugs",
 	[KFENCE_COUNTER_SKIP_INCOMPAT]	= "skipped allocations (incompatible)",
 	[KFENCE_COUNTER_SKIP_CAPACITY]	= "skipped allocations (capacity)",
+	[KFENCE_COUNTER_SKIP_COVERED]	= "skipped allocations (covered)",
 };
 static_assert(ARRAY_SIZE(counter_names) == KFENCE_COUNTER_COUNT);
 
 /* === Internals ============================================================ */
 
+static inline bool should_skip_covered(void)
+{
+	unsigned long thresh = (CONFIG_KFENCE_NUM_OBJECTS * kfence_skip_covered_thresh) / 100;
+
+	return atomic_long_read(&counters[KFENCE_COUNTER_ALLOCATED]) > thresh;
+}
+
+static u32 get_alloc_stack_hash(unsigned long *stack_entries, size_t num_entries)
+{
+	/* Some randomness across reboots / different machines. */
+	u32 seed = (u32)((unsigned long)__kfence_pool >> (BITS_PER_LONG - 32));
+
+	num_entries = min(num_entries, UNIQUE_ALLOC_STACK_DEPTH);
+	num_entries = filter_irq_stacks(stack_entries, num_entries);
+	return jhash(stack_entries, num_entries * sizeof(stack_entries[0]), seed);
+}
+
+/*
+ * Adds (or subtracts) count @val for allocation stack trace hash
+ * @alloc_stack_hash from Counting Bloom filter.
+ */
+static void alloc_covered_add(u32 alloc_stack_hash, int val)
+{
+	int i;
+
+	for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
+		atomic_add(val, &alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]);
+		alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
+	}
+}
+
+/*
+ * Returns true if the allocation stack trace hash @alloc_stack_hash is
+ * currently contained (non-zero count) in Counting Bloom filter.
+ */
+static bool alloc_covered_contains(u32 alloc_stack_hash)
+{
+	int i;
+
+	for (i = 0; i < ALLOC_COVERED_HNUM; i++) {
+		if (!atomic_read(&alloc_covered[alloc_stack_hash & ALLOC_COVERED_MASK]))
+			return false;
+		alloc_stack_hash = ALLOC_COVERED_HNEXT(alloc_stack_hash);
+	}
+
+	return true;
+}
+
 static bool kfence_protect(unsigned long addr)
 {
 	return !KFENCE_WARN_ON(!kfence_protect_page(ALIGN_DOWN(addr, PAGE_SIZE), true));
@@ -269,7 +344,8 @@  static __always_inline void for_each_canary(const struct kfence_metadata *meta,
 }
 
 static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t gfp,
-				  unsigned long *stack_entries, size_t num_stack_entries)
+				  unsigned long *stack_entries, size_t num_stack_entries,
+				  u32 alloc_stack_hash)
 {
 	struct kfence_metadata *meta = NULL;
 	unsigned long flags;
@@ -332,6 +408,8 @@  static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
 	/* Pairs with READ_ONCE() in kfence_shutdown_cache(). */
 	WRITE_ONCE(meta->cache, cache);
 	meta->size = size;
+	meta->alloc_stack_hash = alloc_stack_hash;
+
 	for_each_canary(meta, set_canary_byte);
 
 	/* Set required struct page fields. */
@@ -344,6 +422,8 @@  static void *kfence_guarded_alloc(struct kmem_cache *cache, size_t size, gfp_t g
 
 	raw_spin_unlock_irqrestore(&meta->lock, flags);
 
+	alloc_covered_add(alloc_stack_hash, 1);
+
 	/* Memory initialization. */
 
 	/*
@@ -412,6 +492,8 @@  static void kfence_guarded_free(void *addr, struct kfence_metadata *meta, bool z
 
 	raw_spin_unlock_irqrestore(&meta->lock, flags);
 
+	alloc_covered_add(meta->alloc_stack_hash, -1);
+
 	/* Protect to detect use-after-frees. */
 	kfence_protect((unsigned long)addr);
 
@@ -752,6 +834,7 @@  void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
 {
 	unsigned long stack_entries[KFENCE_STACK_DEPTH];
 	size_t num_stack_entries;
+	u32 alloc_stack_hash;
 
 	/*
 	 * Perform size check before switching kfence_allocation_gate, so that
@@ -799,7 +882,23 @@  void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
 
 	num_stack_entries = stack_trace_save(stack_entries, KFENCE_STACK_DEPTH, 0);
 
-	return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries);
+	/*
+	 * Do expensive check for coverage of allocation in slow-path after
+	 * allocation_gate has already become non-zero, even though it might
+	 * mean not making any allocation within a given sample interval.
+	 *
+	 * This ensures reasonable allocation coverage when the pool is almost
+	 * full, including avoiding long-lived allocations of the same source
+	 * filling up the pool (e.g. pagecache allocations).
+	 */
+	alloc_stack_hash = get_alloc_stack_hash(stack_entries, num_stack_entries);
+	if (should_skip_covered() && alloc_covered_contains(alloc_stack_hash)) {
+		atomic_long_inc(&counters[KFENCE_COUNTER_SKIP_COVERED]);
+		return NULL;
+	}
+
+	return kfence_guarded_alloc(s, size, flags, stack_entries, num_stack_entries,
+				    alloc_stack_hash);
 }
 
 size_t kfence_ksize(const void *addr)
diff --git a/mm/kfence/kfence.h b/mm/kfence/kfence.h
index c1f23c61e5f9..2a2d5de9d379 100644
--- a/mm/kfence/kfence.h
+++ b/mm/kfence/kfence.h
@@ -87,6 +87,8 @@  struct kfence_metadata {
 	/* Allocation and free stack information. */
 	struct kfence_track alloc_track;
 	struct kfence_track free_track;
+	/* For updating alloc_covered on frees. */
+	u32 alloc_stack_hash;
 };
 
 extern struct kfence_metadata kfence_metadata[CONFIG_KFENCE_NUM_OBJECTS];