--- /dev/null
+/*
+ * Generic stack depot for storing stack traces.
+ *
+ * Some debugging tools need to save stack traces of certain events which can
+ * be later presented to the user. For example, KASAN needs to safe alloc and
+ * free stacks for each object, but storing two stack traces per object
+ * requires too much memory (e.g. SLUB_DEBUG needs 256 bytes per object for
+ * that).
+ *
+ * Instead, stack depot maintains a hashtable of unique stacktraces. Since alloc
+ * and free stacks repeat a lot, we save about 100x space.
+ * Stacks are never removed from depot, so we store them contiguously one after
+ * another in a contiguos memory allocation.
+ *
+ * Author: Alexander Potapenko <glider@google.com>
+ * Copyright (C) 2016 Google, Inc.
+ *
+ * Based on code by Dmitry Chernenkov.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ */
+
+#include <linux/gfp.h>
+#include <linux/jhash.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/percpu.h>
+#include <linux/printk.h>
+#include <linux/slab.h>
+#include <linux/stacktrace.h>
+#include <linux/stackdepot.h>
+#include <linux/string.h>
+#include <linux/types.h>
+
+#define DEPOT_STACK_BITS (sizeof(depot_stack_handle_t) * 8)
+
+#define STACK_ALLOC_ORDER 2 /* 'Slab' size order for stack depot, 4 pages */
+#define STACK_ALLOC_SIZE (1LL << (PAGE_SHIFT + STACK_ALLOC_ORDER))
+#define STACK_ALLOC_ALIGN 4
+#define STACK_ALLOC_OFFSET_BITS (STACK_ALLOC_ORDER + PAGE_SHIFT - \
+                                       STACK_ALLOC_ALIGN)
+#define STACK_ALLOC_INDEX_BITS (DEPOT_STACK_BITS - STACK_ALLOC_OFFSET_BITS)
+#define STACK_ALLOC_SLABS_CAP 1024
+#define STACK_ALLOC_MAX_SLABS \
+       (((1LL << (STACK_ALLOC_INDEX_BITS)) < STACK_ALLOC_SLABS_CAP) ? \
+        (1LL << (STACK_ALLOC_INDEX_BITS)) : STACK_ALLOC_SLABS_CAP)
+
+/* The compact structure to store the reference to stacks. */
+union handle_parts {
+       depot_stack_handle_t handle;
+       struct {
+               u32 slabindex : STACK_ALLOC_INDEX_BITS;
+               u32 offset : STACK_ALLOC_OFFSET_BITS;
+       };
+};
+
+struct stack_record {
+       struct stack_record *next;      /* Link in the hashtable */
+       u32 hash;                       /* Hash in the hastable */
+       u32 size;                       /* Number of frames in the stack */
+       union handle_parts handle;
+       unsigned long entries[1];       /* Variable-sized array of entries. */
+};
+
+static void *stack_slabs[STACK_ALLOC_MAX_SLABS];
+
+static int depot_index;
+static int next_slab_inited;
+static size_t depot_offset;
+static DEFINE_SPINLOCK(depot_lock);
+
+static bool init_stack_slab(void **prealloc)
+{
+       if (!*prealloc)
+               return false;
+       /*
+        * This smp_load_acquire() pairs with smp_store_release() to
+        * |next_slab_inited| below and in depot_alloc_stack().
+        */
+       if (smp_load_acquire(&next_slab_inited))
+               return true;
+       if (stack_slabs[depot_index] == NULL) {
+               stack_slabs[depot_index] = *prealloc;
+       } else {
+               stack_slabs[depot_index + 1] = *prealloc;
+               /*
+                * This smp_store_release pairs with smp_load_acquire() from
+                * |next_slab_inited| above and in depot_save_stack().
+                */
+               smp_store_release(&next_slab_inited, 1);
+       }
+       *prealloc = NULL;
+       return true;
+}
+
+/* Allocation of a new stack in raw storage */
+static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
+               u32 hash, void **prealloc, gfp_t alloc_flags)
+{
+       int required_size = offsetof(struct stack_record, entries) +
+               sizeof(unsigned long) * size;
+       struct stack_record *stack;
+
+       required_size = ALIGN(required_size, 1 << STACK_ALLOC_ALIGN);
+
+       if (unlikely(depot_offset + required_size > STACK_ALLOC_SIZE)) {
+               if (unlikely(depot_index + 1 >= STACK_ALLOC_MAX_SLABS)) {
+                       WARN_ONCE(1, "Stack depot reached limit capacity");
+                       return NULL;
+               }
+               depot_index++;
+               depot_offset = 0;
+               /*
+                * smp_store_release() here pairs with smp_load_acquire() from
+                * |next_slab_inited| in depot_save_stack() and
+                * init_stack_slab().
+                */
+               if (depot_index + 1 < STACK_ALLOC_MAX_SLABS)
+                       smp_store_release(&next_slab_inited, 0);
+       }
+       init_stack_slab(prealloc);
+       if (stack_slabs[depot_index] == NULL)
+               return NULL;
+
+       stack = stack_slabs[depot_index] + depot_offset;
+
+       stack->hash = hash;
+       stack->size = size;
+       stack->handle.slabindex = depot_index;
+       stack->handle.offset = depot_offset >> STACK_ALLOC_ALIGN;
+       memcpy(stack->entries, entries, size * sizeof(unsigned long));
+       depot_offset += required_size;
+
+       return stack;
+}
+
+#define STACK_HASH_ORDER 20
+#define STACK_HASH_SIZE (1L << STACK_HASH_ORDER)
+#define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
+#define STACK_HASH_SEED 0x9747b28c
+
+static struct stack_record *stack_table[STACK_HASH_SIZE] = {
+       [0 ...  STACK_HASH_SIZE - 1] = NULL
+};
+
+/* Calculate hash for a stack */
+static inline u32 hash_stack(unsigned long *entries, unsigned int size)
+{
+       return jhash2((u32 *)entries,
+                              size * sizeof(unsigned long) / sizeof(u32),
+                              STACK_HASH_SEED);
+}
+
+/* Find a stack that is equal to the one stored in entries in the hash */
+static inline struct stack_record *find_stack(struct stack_record *bucket,
+                                            unsigned long *entries, int size,
+                                            u32 hash)
+{
+       struct stack_record *found;
+
+       for (found = bucket; found; found = found->next) {
+               if (found->hash == hash &&
+                   found->size == size &&
+                   !memcmp(entries, found->entries,
+                           size * sizeof(unsigned long))) {
+                       return found;
+               }
+       }
+       return NULL;
+}
+
+void depot_fetch_stack(depot_stack_handle_t handle, struct stack_trace *trace)
+{
+       union handle_parts parts = { .handle = handle };
+       void *slab = stack_slabs[parts.slabindex];
+       size_t offset = parts.offset << STACK_ALLOC_ALIGN;
+       struct stack_record *stack = slab + offset;
+
+       trace->nr_entries = trace->max_entries = stack->size;
+       trace->entries = stack->entries;
+       trace->skip = 0;
+}
+
+/**
+ * depot_save_stack - save stack in a stack depot.
+ * @trace - the stacktrace to save.
+ * @alloc_flags - flags for allocating additional memory if required.
+ *
+ * Returns the handle of the stack struct stored in depot.
+ */
+depot_stack_handle_t depot_save_stack(struct stack_trace *trace,
+                                   gfp_t alloc_flags)
+{
+       u32 hash;
+       depot_stack_handle_t retval = 0;
+       struct stack_record *found = NULL, **bucket;
+       unsigned long flags;
+       struct page *page = NULL;
+       void *prealloc = NULL;
+
+       if (unlikely(trace->nr_entries == 0))
+               goto fast_exit;
+
+       hash = hash_stack(trace->entries, trace->nr_entries);
+       /* Bad luck, we won't store this stack. */
+       if (hash == 0)
+               goto exit;
+
+       bucket = &stack_table[hash & STACK_HASH_MASK];
+
+       /*
+        * Fast path: look the stack trace up without locking.
+        * The smp_load_acquire() here pairs with smp_store_release() to
+        * |bucket| below.
+        */
+       found = find_stack(smp_load_acquire(bucket), trace->entries,
+                          trace->nr_entries, hash);
+       if (found)
+               goto exit;
+
+       /*
+        * Check if the current or the next stack slab need to be initialized.
+        * If so, allocate the memory - we won't be able to do that under the
+        * lock.
+        *
+        * The smp_load_acquire() here pairs with smp_store_release() to
+        * |next_slab_inited| in depot_alloc_stack() and init_stack_slab().
+        */
+       if (unlikely(!smp_load_acquire(&next_slab_inited))) {
+               /*
+                * Zero out zone modifiers, as we don't have specific zone
+                * requirements. Keep the flags related to allocation in atomic
+                * contexts and I/O.
+                */
+               alloc_flags &= ~GFP_ZONEMASK;
+               alloc_flags &= (GFP_ATOMIC | GFP_KERNEL);
+               page = alloc_pages(alloc_flags, STACK_ALLOC_ORDER);
+               if (page)
+                       prealloc = page_address(page);
+       }
+
+       spin_lock_irqsave(&depot_lock, flags);
+
+       found = find_stack(*bucket, trace->entries, trace->nr_entries, hash);
+       if (!found) {
+               struct stack_record *new =
+                       depot_alloc_stack(trace->entries, trace->nr_entries,
+                                         hash, &prealloc, alloc_flags);
+               if (new) {
+                       new->next = *bucket;
+                       /*
+                        * This smp_store_release() pairs with
+                        * smp_load_acquire() from |bucket| above.
+                        */
+                       smp_store_release(bucket, new);
+                       found = new;
+               }
+       } else if (prealloc) {
+               /*
+                * We didn't need to store this stack trace, but let's keep
+                * the preallocated memory for the future.
+                */
+               WARN_ON(!init_stack_slab(&prealloc));
+       }
+
+       spin_unlock_irqrestore(&depot_lock, flags);
+exit:
+       if (prealloc) {
+               /* Nobody used this memory, ok to free it. */
+               free_pages((unsigned long)prealloc, STACK_ALLOC_ORDER);
+       }
+       if (found)
+               retval = found->handle.handle;
+fast_exit:
+       return retval;
+}
 
 #define DISABLE_BRANCH_PROFILING
 
 #include <linux/export.h>
+#include <linux/interrupt.h>
 #include <linux/init.h>
+#include <linux/kasan.h>
 #include <linux/kernel.h>
 #include <linux/kmemleak.h>
 #include <linux/linkage.h>
 #include <linux/string.h>
 #include <linux/types.h>
 #include <linux/vmalloc.h>
-#include <linux/kasan.h>
 
 #include "kasan.h"
 #include "../slab.h"
 #endif
 }
 
-static inline void set_track(struct kasan_track *track)
+#ifdef CONFIG_SLAB
+static inline int in_irqentry_text(unsigned long ptr)
+{
+       return (ptr >= (unsigned long)&__irqentry_text_start &&
+               ptr < (unsigned long)&__irqentry_text_end) ||
+               (ptr >= (unsigned long)&__softirqentry_text_start &&
+                ptr < (unsigned long)&__softirqentry_text_end);
+}
+
+static inline void filter_irq_stacks(struct stack_trace *trace)
+{
+       int i;
+
+       if (!trace->nr_entries)
+               return;
+       for (i = 0; i < trace->nr_entries; i++)
+               if (in_irqentry_text(trace->entries[i])) {
+                       /* Include the irqentry function into the stack. */
+                       trace->nr_entries = i + 1;
+                       break;
+               }
+}
+
+static inline depot_stack_handle_t save_stack(gfp_t flags)
+{
+       unsigned long entries[KASAN_STACK_DEPTH];
+       struct stack_trace trace = {
+               .nr_entries = 0,
+               .entries = entries,
+               .max_entries = KASAN_STACK_DEPTH,
+               .skip = 0
+       };
+
+       save_stack_trace(&trace);
+       filter_irq_stacks(&trace);
+       if (trace.nr_entries != 0 &&
+           trace.entries[trace.nr_entries-1] == ULONG_MAX)
+               trace.nr_entries--;
+
+       return depot_save_stack(&trace, flags);
+}
+
+static inline void set_track(struct kasan_track *track, gfp_t flags)
 {
-       track->cpu = raw_smp_processor_id();
        track->pid = current->pid;
-       track->when = jiffies;
+       track->stack = save_stack(flags);
 }
 
-#ifdef CONFIG_SLAB
 struct kasan_alloc_meta *get_alloc_info(struct kmem_cache *cache,
                                        const void *object)
 {
+       BUILD_BUG_ON(sizeof(struct kasan_alloc_meta) > 32);
        return (void *)object + cache->kasan_info.alloc_meta_offset;
 }
 
 struct kasan_free_meta *get_free_info(struct kmem_cache *cache,
                                      const void *object)
 {
+       BUILD_BUG_ON(sizeof(struct kasan_free_meta) > 32);
        return (void *)object + cache->kasan_info.free_meta_offset;
 }
 #endif
 
                alloc_info->state = KASAN_STATE_ALLOC;
                alloc_info->alloc_size = size;
-               set_track(&alloc_info->track);
+               set_track(&alloc_info->track, flags);
        }
 #endif
 }