blob: d14663a91e7f06ea79f98e89463de35cf14c2749 [file] [log] [blame]
// Copyright 2018 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef BASE_ALLOCATOR_PARTITION_ALLOCATOR_PARTITION_PAGE_H_
#define BASE_ALLOCATOR_PARTITION_ALLOCATOR_PARTITION_PAGE_H_
#include <cstdint>
#include <cstring>
#include <limits>
#include <utility>
#include "base/allocator/partition_allocator/address_pool_manager.h"
#include "base/allocator/partition_allocator/address_pool_manager_types.h"
#include "base/allocator/partition_allocator/freeslot_bitmap_constants.h"
#include "base/allocator/partition_allocator/partition_address_space.h"
#include "base/allocator/partition_allocator/partition_alloc_base/bits.h"
#include "base/allocator/partition_allocator/partition_alloc_base/compiler_specific.h"
#include "base/allocator/partition_allocator/partition_alloc_base/component_export.h"
#include "base/allocator/partition_allocator/partition_alloc_base/debug/debugging_buildflags.h"
#include "base/allocator/partition_allocator/partition_alloc_base/thread_annotations.h"
#include "base/allocator/partition_allocator/partition_alloc_buildflags.h"
#include "base/allocator/partition_allocator/partition_alloc_check.h"
#include "base/allocator/partition_allocator/partition_alloc_config.h"
#include "base/allocator/partition_allocator/partition_alloc_constants.h"
#include "base/allocator/partition_allocator/partition_alloc_forward.h"
#include "base/allocator/partition_allocator/partition_bucket.h"
#include "base/allocator/partition_allocator/partition_freelist_entry.h"
#include "base/allocator/partition_allocator/partition_tag_bitmap.h"
#include "base/allocator/partition_allocator/partition_tag_types.h"
#include "base/allocator/partition_allocator/reservation_offset_table.h"
#include "base/allocator/partition_allocator/tagging.h"
#include "build/build_config.h"
#if BUILDFLAG(STARSCAN)
#include "base/allocator/partition_allocator/starscan/state_bitmap.h"
#endif // BUILDFLAG(STARSCAN)
#if BUILDFLAG(PUT_REF_COUNT_IN_PREVIOUS_SLOT)
#include "base/allocator/partition_allocator/partition_ref_count.h"
#endif
namespace partition_alloc::internal {
// An "extent" is a span of consecutive superpages. We link the partition's next
// extent (if there is one) to the very start of a superpage's metadata area.
template <bool thread_safe>
struct PartitionSuperPageExtentEntry {
PartitionRoot<thread_safe>* root;
PartitionSuperPageExtentEntry<thread_safe>* next;
uint16_t number_of_consecutive_super_pages;
uint16_t number_of_nonempty_slot_spans;
PA_ALWAYS_INLINE void IncrementNumberOfNonemptySlotSpans();
PA_ALWAYS_INLINE void DecrementNumberOfNonemptySlotSpans();
};
static_assert(
sizeof(PartitionSuperPageExtentEntry<ThreadSafe>) <= kPageMetadataSize,
"PartitionSuperPageExtentEntry must be able to fit in a metadata slot");
static_assert(
kMaxSuperPagesInPool / kSuperPageSize <=
std::numeric_limits<
decltype(PartitionSuperPageExtentEntry<
ThreadSafe>::number_of_consecutive_super_pages)>::max(),
"number_of_consecutive_super_pages must be big enough");
// Returns the base of the first super page in the range of consecutive super
// pages.
//
// CAUTION! |extent| must point to the extent of the first super page in the
// range of consecutive super pages.
template <bool thread_safe>
PA_ALWAYS_INLINE uintptr_t SuperPagesBeginFromExtent(
const PartitionSuperPageExtentEntry<thread_safe>* extent) {
PA_DCHECK(0 < extent->number_of_consecutive_super_pages);
uintptr_t extent_as_uintptr = reinterpret_cast<uintptr_t>(extent);
PA_DCHECK(IsManagedByNormalBuckets(extent_as_uintptr));
return base::bits::AlignDown(extent_as_uintptr, kSuperPageAlignment);
}
// Returns the end of the last super page in the range of consecutive super
// pages.
//
// CAUTION! |extent| must point to the extent of the first super page in the
// range of consecutive super pages.
template <bool thread_safe>
PA_ALWAYS_INLINE uintptr_t SuperPagesEndFromExtent(
const PartitionSuperPageExtentEntry<thread_safe>* extent) {
return SuperPagesBeginFromExtent(extent) +
(extent->number_of_consecutive_super_pages * kSuperPageSize);
}
#if BUILDFLAG(STARSCAN)
using AllocationStateMap =
StateBitmap<kSuperPageSize, kSuperPageAlignment, kAlignment>;
#endif // BUILDFLAG(STARSCAN)
// Metadata of the slot span.
//
// Some notes on slot span states. It can be in one of four major states:
// 1) Active.
// 2) Full.
// 3) Empty.
// 4) Decommitted.
// An active slot span has available free slots, as well as allocated ones.
// A full slot span has no free slots. An empty slot span has no allocated
// slots, and a decommitted slot span is an empty one that had its backing
// memory released back to the system.
//
// There are three linked lists tracking slot spans. The "active" list is an
// approximation of a list of active slot spans. It is an approximation because
// full, empty and decommitted slot spans may briefly be present in the list
// until we next do a scan over it. The "empty" list holds mostly empty slot
// spans, but may briefly hold decommitted ones too. The "decommitted" list
// holds only decommitted slot spans.
//
// The significant slot span transitions are:
// - Free() will detect when a full slot span has a slot freed and immediately
// return the slot span to the head of the active list.
// - Free() will detect when a slot span is fully emptied. It _may_ add it to
// the empty list or it _may_ leave it on the active list until a future
// list scan.
// - Alloc() _may_ scan the active page list in order to fulfil the request.
// If it does this, full, empty and decommitted slot spans encountered will be
// booted out of the active list. If there are no suitable active slot spans
// found, an empty or decommitted slot spans (if one exists) will be pulled
// from the empty/decommitted list on to the active list.
#pragma pack(push, 1)
template <bool thread_safe>
struct SlotSpanMetadata {
private:
PartitionFreelistEntry* freelist_head = nullptr;
public:
// TODO(lizeb): Make as many fields as possible private or const, to
// encapsulate things more clearly.
SlotSpanMetadata<thread_safe>* next_slot_span = nullptr;
PartitionBucket<thread_safe>* const bucket = nullptr;
// CHECK()ed in AllocNewSlotSpan().
#if PA_CONFIG(HAS_64_BITS_POINTERS) && BUILDFLAG(IS_APPLE)
// System page size is not a constant on Apple OSes, but is either 4 or 16kiB
// (1 << 12 or 1 << 14), as checked in PartitionRoot::Init(). And
// PartitionPageSize() is 4 times the OS page size.
static constexpr size_t kMaxSlotsPerSlotSpan =
4 * (1 << 14) / kSmallestBucket;
#elif BUILDFLAG(IS_LINUX) && defined(ARCH_CPU_ARM64)
// System page size can be 4, 16, or 64 kiB on Linux on arm64. 64 kiB is
// currently (kMaxSlotsPerSlotSpanBits == 13) not supported by the code,
// so we use the 16 kiB maximum (64 kiB will crash).
static constexpr size_t kMaxSlotsPerSlotSpan =
4 * (1 << 14) / kSmallestBucket;
#else
// A slot span can "span" multiple PartitionPages, but then its slot size is
// larger, so it doesn't have as many slots.
static constexpr size_t kMaxSlotsPerSlotSpan =
PartitionPageSize() / kSmallestBucket;
#endif // PA_CONFIG(HAS_64_BITS_POINTERS) && BUILDFLAG(IS_APPLE)
// The maximum number of bits needed to cover all currently supported OSes.
static constexpr size_t kMaxSlotsPerSlotSpanBits = 13;
static_assert(kMaxSlotsPerSlotSpan < (1 << kMaxSlotsPerSlotSpanBits), "");
// |marked_full| isn't equivalent to being full. Slot span is marked as full
// iff it isn't on the active slot span list (or any other list).
uint32_t marked_full : 1;
// |num_allocated_slots| is 0 for empty or decommitted slot spans, which can
// be further differentiated by checking existence of the freelist.
uint32_t num_allocated_slots : kMaxSlotsPerSlotSpanBits;
uint32_t num_unprovisioned_slots : kMaxSlotsPerSlotSpanBits;
private:
const uint32_t can_store_raw_size_ : 1;
uint32_t freelist_is_sorted_ : 1;
uint32_t unused1_ : (32 - 1 - 2 * kMaxSlotsPerSlotSpanBits - 1 - 1);
// If |in_empty_cache_|==1, |empty_cache_index| is undefined and mustn't be
// used.
uint16_t in_empty_cache_ : 1;
uint16_t empty_cache_index_ : kEmptyCacheIndexBits; // < kMaxFreeableSpans.
uint16_t unused2_ : (16 - 1 - kEmptyCacheIndexBits);
// Can use only 48 bits (6B) in this bitfield, as this structure is embedded
// in PartitionPage which has 2B worth of fields and must fit in 32B.
public:
PA_COMPONENT_EXPORT(PARTITION_ALLOC)
explicit SlotSpanMetadata(PartitionBucket<thread_safe>* bucket);
// Public API
// Note the matching Alloc() functions are in PartitionPage.
PA_COMPONENT_EXPORT(PARTITION_ALLOC)
PA_NOINLINE void FreeSlowPath(size_t number_of_freed);
PA_ALWAYS_INLINE PartitionFreelistEntry* PopForAlloc(size_t size);
PA_ALWAYS_INLINE void Free(uintptr_t ptr);
// Appends the passed freelist to the slot-span's freelist. Please note that
// the function doesn't increment the tags of the passed freelist entries,
// since FreeNoHooks() did it already.
PA_ALWAYS_INLINE void AppendFreeList(PartitionFreelistEntry* head,
PartitionFreelistEntry* tail,
size_t number_of_freed);
void Decommit(PartitionRoot<thread_safe>* root);
void DecommitIfPossible(PartitionRoot<thread_safe>* root);
// Sorts the freelist in ascending addresses order.
void SortFreelist();
// Inserts the slot span into the empty ring, making space for the new slot
// span, and potentially shrinking the ring.
void RegisterEmpty();
// Pointer/address manipulation functions. These must be static as the input
// |slot_span| pointer may be the result of an offset calculation and
// therefore cannot be trusted. The objective of these functions is to
// sanitize this input.
PA_ALWAYS_INLINE static uintptr_t ToSlotSpanStart(
const SlotSpanMetadata* slot_span);
PA_ALWAYS_INLINE static SlotSpanMetadata* FromAddr(uintptr_t address);
PA_ALWAYS_INLINE static SlotSpanMetadata* FromSlotStart(uintptr_t slot_start);
PA_ALWAYS_INLINE static SlotSpanMetadata* FromObject(void* object);
PA_ALWAYS_INLINE static SlotSpanMetadata* FromObjectInnerAddr(
uintptr_t address);
PA_ALWAYS_INLINE static SlotSpanMetadata* FromObjectInnerPtr(void* ptr);
PA_ALWAYS_INLINE PartitionSuperPageExtentEntry<thread_safe>*
ToSuperPageExtent() const;
// Checks if it is feasible to store raw_size.
PA_ALWAYS_INLINE bool CanStoreRawSize() const { return can_store_raw_size_; }
// The caller is responsible for ensuring that raw_size can be stored before
// calling Set/GetRawSize.
PA_ALWAYS_INLINE void SetRawSize(size_t raw_size);
PA_ALWAYS_INLINE size_t GetRawSize() const;
// Only meaningful when `this` refers to a slot span in a direct map
// bucket.
PA_ALWAYS_INLINE PartitionTag* DirectMapMTETag();
PA_ALWAYS_INLINE PartitionFreelistEntry* get_freelist_head() const {
return freelist_head;
}
PA_ALWAYS_INLINE void SetFreelistHead(PartitionFreelistEntry* new_head);
// Returns size of the region used within a slot. The used region comprises
// of actual allocated data, extras and possibly empty space in the middle.
PA_ALWAYS_INLINE size_t GetUtilizedSlotSize() const {
// The returned size can be:
// - The slot size for small buckets.
// - Exact size needed to satisfy allocation (incl. extras), for large
// buckets and direct-mapped allocations (see also the comment in
// CanStoreRawSize() for more info).
if (PA_LIKELY(!CanStoreRawSize())) {
return bucket->slot_size;
}
return GetRawSize();
}
// This includes padding due to rounding done at allocation; we don't know the
// requested size at deallocation, so we use this in both places.
PA_ALWAYS_INLINE size_t GetSlotSizeForBookkeeping() const {
// This could be more precise for allocations where CanStoreRawSize()
// returns true (large allocations). However this is called for *every*
// allocation, so we don't want an extra branch there.
return bucket->slot_size;
}
// Returns the size available to the app. It can be equal or higher than the
// requested size. If higher, the overage won't exceed what's actually usable
// by the app without a risk of running out of an allocated region or into
// PartitionAlloc's internal data (like extras).
PA_ALWAYS_INLINE size_t
GetUsableSize(PartitionRoot<thread_safe>* root) const {
// The returned size can be:
// - The slot size minus extras, for small buckets. This could be more than
// requested size.
// - Raw size minus extras, for large buckets and direct-mapped allocations
// (see also the comment in CanStoreRawSize() for more info). This is
// equal to requested size.
return root->AdjustSizeForExtrasSubtract(GetUtilizedSlotSize());
}
// Returns the total size of the slots that are currently provisioned.
PA_ALWAYS_INLINE size_t GetProvisionedSize() const {
size_t num_provisioned_slots =
bucket->get_slots_per_span() - num_unprovisioned_slots;
size_t provisioned_size = num_provisioned_slots * bucket->slot_size;
PA_DCHECK(provisioned_size <= bucket->get_bytes_per_span());
return provisioned_size;
}
// Return the number of entries in the freelist.
size_t GetFreelistLength() const {
size_t num_provisioned_slots =
bucket->get_slots_per_span() - num_unprovisioned_slots;
return num_provisioned_slots - num_allocated_slots;
}
PA_ALWAYS_INLINE void Reset();
// TODO(ajwong): Can this be made private? https://crbug.com/787153
PA_COMPONENT_EXPORT(PARTITION_ALLOC)
static const SlotSpanMetadata* get_sentinel_slot_span();
// The sentinel is not supposed to be modified and hence we mark it as const
// under the hood. However, we often store it together with mutable metadata
// objects and need a non-const pointer.
// You can use this function for this case, but you need to ensure that the
// returned object will not be written to.
static SlotSpanMetadata* get_sentinel_slot_span_non_const();
// Slot span state getters.
PA_ALWAYS_INLINE bool is_active() const;
PA_ALWAYS_INLINE bool is_full() const;
PA_ALWAYS_INLINE bool is_empty() const;
PA_ALWAYS_INLINE bool is_decommitted() const;
PA_ALWAYS_INLINE bool in_empty_cache() const { return in_empty_cache_; }
PA_ALWAYS_INLINE bool freelist_is_sorted() const {
return freelist_is_sorted_;
}
PA_ALWAYS_INLINE void set_freelist_sorted() { freelist_is_sorted_ = true; }
private:
// sentinel_slot_span_ is used as a sentinel to indicate that there is no slot
// span in the active list. We could use nullptr, but in that case we need to
// add a null-check branch to the hot allocation path. We want to avoid that.
//
// Note, this declaration is kept in the header as opposed to an anonymous
// namespace so the getter can be fully inlined.
static const SlotSpanMetadata sentinel_slot_span_;
// For the sentinel.
constexpr SlotSpanMetadata() noexcept
: marked_full(0),
num_allocated_slots(0),
num_unprovisioned_slots(0),
can_store_raw_size_(false),
freelist_is_sorted_(true),
unused1_(0),
in_empty_cache_(0),
empty_cache_index_(0),
unused2_(0) {}
};
#pragma pack(pop)
static_assert(sizeof(SlotSpanMetadata<ThreadSafe>) <= kPageMetadataSize,
"SlotSpanMetadata must fit into a Page Metadata slot.");
// Metadata of a non-first partition page in a slot span.
struct SubsequentPageMetadata {
// Raw size is the size needed to satisfy the allocation (requested size +
// extras). If available, it can be used to report better statistics or to
// bring protective cookie closer to the allocated memory.
//
// It can be used only if:
// - there is no more than one slot in the slot span (otherwise we wouldn't
// know which slot the raw size applies to)
// - there is more than one partition page in the slot span (the metadata of
// the first one is used to store slot information, but the second one is
// available for extra information)
size_t raw_size;
// Specific to when `this` is used in a direct map bucket. Since direct
// maps don't have as many tags as the typical normal bucket slot span,
// we can get away with just hiding the sole tag in here.
//
// See `//base/memory/mtecheckedptr.md` for details.
PartitionTag direct_map_tag;
};
// Each partition page has metadata associated with it. The metadata of the
// first page of a slot span, describes that slot span. If a slot span spans
// more than 1 page, the page metadata may contain rudimentary additional
// information.
// "Pack" the union so that common page metadata still fits within
// kPageMetadataSize. (SlotSpanMetadata is also "packed".)
#pragma pack(push, 1)
template <bool thread_safe>
struct PartitionPage {
union {
SlotSpanMetadata<thread_safe> slot_span_metadata;
SubsequentPageMetadata subsequent_page_metadata;
// sizeof(PartitionPageMetadata) must always be:
// - a power of 2 (for fast modulo operations)
// - below kPageMetadataSize
//
// This makes sure that this is respected no matter the architecture.
char optional_padding[kPageMetadataSize - sizeof(uint8_t) - sizeof(bool)];
};
// The first PartitionPage of the slot span holds its metadata. This offset
// tells how many pages in from that first page we are.
// For direct maps, the first page metadata (that isn't super page extent
// entry) uses this field to tell how many pages to the right the direct map
// metadata starts.
//
// 6 bits is enough to represent all possible offsets, given that the smallest
// partition page is 16kiB and the offset won't exceed 1MiB.
static constexpr uint16_t kMaxSlotSpanMetadataBits = 6;
static constexpr uint16_t kMaxSlotSpanMetadataOffset =
(1 << kMaxSlotSpanMetadataBits) - 1;
uint8_t slot_span_metadata_offset : kMaxSlotSpanMetadataBits;
// |is_valid| tells whether the page is part of a slot span. If |false|,
// |has_valid_span_after_this| tells whether it's an unused region in between
// slot spans within the super page.
// Note, |is_valid| has been added for clarity, but if we ever need to save
// this bit, it can be inferred from:
// |!slot_span_metadata_offset && slot_span_metadata->bucket|.
bool is_valid : 1;
bool has_valid_span_after_this : 1;
uint8_t unused;
PA_ALWAYS_INLINE static PartitionPage* FromAddr(uintptr_t address);
};
#pragma pack(pop)
static_assert(sizeof(PartitionPage<ThreadSafe>) == kPageMetadataSize,
"PartitionPage must be able to fit in a metadata slot");
// Certain functions rely on PartitionPage being either SlotSpanMetadata or
// SubsequentPageMetadata, and therefore freely casting between each other.
static_assert(offsetof(PartitionPage<ThreadSafe>, slot_span_metadata) == 0, "");
static_assert(offsetof(PartitionPage<ThreadSafe>, subsequent_page_metadata) ==
0,
"");
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionPage<thread_safe>* PartitionSuperPageToMetadataArea(
uintptr_t super_page) {
// This can't be just any super page, but it has to be the first super page of
// the reservation, as we assume here that the metadata is near its beginning.
PA_DCHECK(IsReservationStart(super_page));
PA_DCHECK(!(super_page & kSuperPageOffsetMask));
// The metadata area is exactly one system page (the guard page) into the
// super page.
return reinterpret_cast<PartitionPage<thread_safe>*>(super_page +
SystemPageSize());
}
PA_ALWAYS_INLINE const SubsequentPageMetadata* GetSubsequentPageMetadata(
const PartitionPage<ThreadSafe>* page) {
return &(page + 1)->subsequent_page_metadata;
}
PA_ALWAYS_INLINE SubsequentPageMetadata* GetSubsequentPageMetadata(
PartitionPage<ThreadSafe>* page) {
return &(page + 1)->subsequent_page_metadata;
}
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionSuperPageExtentEntry<thread_safe>*
PartitionSuperPageToExtent(uintptr_t super_page) {
// The very first entry of the metadata is the super page extent entry.
return reinterpret_cast<PartitionSuperPageExtentEntry<thread_safe>*>(
PartitionSuperPageToMetadataArea<thread_safe>(super_page));
}
#if BUILDFLAG(STARSCAN)
// Size that should be reserved for state bitmap (if present) inside a super
// page. Elements of a super page are partition-page-aligned, hence the returned
// size is a multiple of partition page size.
PAGE_ALLOCATOR_CONSTANTS_DECLARE_CONSTEXPR PA_ALWAYS_INLINE size_t
ReservedStateBitmapSize() {
return base::bits::AlignUp(sizeof(AllocationStateMap), PartitionPageSize());
}
// Size that should be committed for state bitmap (if present) inside a super
// page. It is a multiple of system page size.
PAGE_ALLOCATOR_CONSTANTS_DECLARE_CONSTEXPR PA_ALWAYS_INLINE size_t
CommittedStateBitmapSize() {
return base::bits::AlignUp(sizeof(AllocationStateMap), SystemPageSize());
}
// Returns the address/pointer to the state bitmap in the super page. It's the
// caller's responsibility to ensure that the bitmaps even exist.
PA_ALWAYS_INLINE uintptr_t SuperPageStateBitmapAddr(uintptr_t super_page) {
PA_DCHECK(!(super_page % kSuperPageAlignment));
return super_page + PartitionPageSize() +
(IsManagedByNormalBuckets(super_page)
? ReservedTagBitmapSize() + ReservedFreeSlotBitmapSize()
: 0);
}
PA_ALWAYS_INLINE AllocationStateMap* SuperPageStateBitmap(
uintptr_t super_page) {
return reinterpret_cast<AllocationStateMap*>(
SuperPageStateBitmapAddr(super_page));
}
#else
PAGE_ALLOCATOR_CONSTANTS_DECLARE_CONSTEXPR PA_ALWAYS_INLINE size_t
ReservedStateBitmapSize() {
return 0ull;
}
#endif // BUILDFLAG(STARSCAN)
// Returns the address of the tag bitmap of the `super_page`. Caller must ensure
// that bitmap exists.
PA_ALWAYS_INLINE uintptr_t SuperPageTagBitmapAddr(uintptr_t super_page) {
PA_DCHECK(IsReservationStart(super_page));
// Skip over the guard pages / metadata.
return super_page + PartitionPageSize();
}
PA_ALWAYS_INLINE uintptr_t
SuperPagePayloadStartOffset(bool is_managed_by_normal_buckets,
bool with_quarantine) {
return PartitionPageSize() +
(is_managed_by_normal_buckets
? (ReservedTagBitmapSize() + ReservedFreeSlotBitmapSize())
: 0) +
(with_quarantine ? ReservedStateBitmapSize() : 0);
}
PA_ALWAYS_INLINE uintptr_t SuperPagePayloadBegin(uintptr_t super_page,
bool with_quarantine) {
PA_DCHECK(!(super_page % kSuperPageAlignment));
return super_page +
SuperPagePayloadStartOffset(IsManagedByNormalBuckets(super_page),
with_quarantine);
}
PA_ALWAYS_INLINE uintptr_t SuperPagePayloadEndOffset() {
return kSuperPageSize - PartitionPageSize();
}
PA_ALWAYS_INLINE uintptr_t SuperPagePayloadEnd(uintptr_t super_page) {
PA_DCHECK(!(super_page % kSuperPageAlignment));
return super_page + SuperPagePayloadEndOffset();
}
PA_ALWAYS_INLINE size_t SuperPagePayloadSize(uintptr_t super_page,
bool with_quarantine) {
return SuperPagePayloadEnd(super_page) -
SuperPagePayloadBegin(super_page, with_quarantine);
}
template <bool thread_safe>
PA_ALWAYS_INLINE void PartitionSuperPageExtentEntry<
thread_safe>::IncrementNumberOfNonemptySlotSpans() {
#if BUILDFLAG(PA_DCHECK_IS_ON)
uintptr_t super_page = base::bits::AlignDown(
reinterpret_cast<uintptr_t>(this), kSuperPageAlignment);
PA_DCHECK((SuperPagePayloadSize(super_page, root->IsQuarantineAllowed()) /
PartitionPageSize()) > number_of_nonempty_slot_spans);
#endif
++number_of_nonempty_slot_spans;
}
template <bool thread_safe>
PA_ALWAYS_INLINE void PartitionSuperPageExtentEntry<
thread_safe>::DecrementNumberOfNonemptySlotSpans() {
PA_DCHECK(number_of_nonempty_slot_spans);
--number_of_nonempty_slot_spans;
}
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionSuperPageExtentEntry<thread_safe>*
SlotSpanMetadata<thread_safe>::ToSuperPageExtent() const {
uintptr_t super_page = reinterpret_cast<uintptr_t>(this) & kSuperPageBaseMask;
return PartitionSuperPageToExtent<thread_safe>(super_page);
}
// Returns whether the pointer lies within the super page's payload area (i.e.
// area devoted to slot spans). It doesn't check whether it's within a valid
// slot span. It merely ensures it doesn't fall in a meta-data region that would
// surely never contain user data.
PA_ALWAYS_INLINE bool IsWithinSuperPagePayload(uintptr_t address,
bool with_quarantine) {
// Quarantine can only be enabled for normal buckets in the current code.
PA_DCHECK(!with_quarantine || IsManagedByNormalBuckets(address));
uintptr_t super_page = address & kSuperPageBaseMask;
uintptr_t payload_start = SuperPagePayloadBegin(super_page, with_quarantine);
uintptr_t payload_end = SuperPagePayloadEnd(super_page);
return address >= payload_start && address < payload_end;
}
// Converts from an address inside a super page into a pointer to the
// PartitionPage object (within super pages's metadata) that describes the
// partition page where |address| is located. |address| doesn't have to be
// located within a valid (i.e. allocated) slot span, but must be within the
// super page's payload area (i.e. area devoted to slot spans).
//
// While it is generally valid for |ptr| to be in the middle of an allocation,
// care has to be taken with direct maps that span multiple super pages. This
// function's behavior is undefined if |ptr| lies in a subsequent super page.
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionPage<thread_safe>*
PartitionPage<thread_safe>::FromAddr(uintptr_t address) {
uintptr_t super_page = address & kSuperPageBaseMask;
#if BUILDFLAG(PA_DCHECK_IS_ON)
PA_DCHECK(IsReservationStart(super_page));
auto* extent = PartitionSuperPageToExtent<thread_safe>(super_page);
PA_DCHECK(IsWithinSuperPagePayload(address,
IsManagedByNormalBuckets(address) &&
extent->root->IsQuarantineAllowed()));
#endif
uintptr_t partition_page_index =
(address & kSuperPageOffsetMask) >> PartitionPageShift();
// Index 0 is invalid because it is the super page extent metadata and the
// last index is invalid because the whole PartitionPage is set as guard
// pages. This repeats part of the payload PA_DCHECK above, which also checks
// for other exclusions.
PA_DCHECK(partition_page_index);
PA_DCHECK(partition_page_index < NumPartitionPagesPerSuperPage() - 1);
return PartitionSuperPageToMetadataArea<thread_safe>(super_page) +
partition_page_index;
}
// Converts from a pointer to the SlotSpanMetadata object (within a super
// pages's metadata) into a pointer to the beginning of the slot span. This
// works on direct maps too.
template <bool thread_safe>
PA_ALWAYS_INLINE uintptr_t SlotSpanMetadata<thread_safe>::ToSlotSpanStart(
const SlotSpanMetadata* slot_span) {
uintptr_t pointer_as_uint = reinterpret_cast<uintptr_t>(slot_span);
uintptr_t super_page_offset = (pointer_as_uint & kSuperPageOffsetMask);
// A valid |page| must be past the first guard System page and within
// the following metadata region.
PA_DCHECK(super_page_offset > SystemPageSize());
// Must be less than total metadata region.
PA_DCHECK(super_page_offset <
SystemPageSize() +
(NumPartitionPagesPerSuperPage() * kPageMetadataSize));
uintptr_t partition_page_index =
(super_page_offset - SystemPageSize()) >> kPageMetadataShift;
// Index 0 is invalid because it is the super page extent metadata and the
// last index is invalid because the whole PartitionPage is set as guard
// pages.
PA_DCHECK(partition_page_index);
PA_DCHECK(partition_page_index < NumPartitionPagesPerSuperPage() - 1);
uintptr_t super_page_base = (pointer_as_uint & kSuperPageBaseMask);
return super_page_base + (partition_page_index << PartitionPageShift());
}
// Converts an address inside a slot span into a pointer to the SlotSpanMetadata
// object (within super pages's metadata) that describes the slot span
// containing that slot.
//
// CAUTION! For direct-mapped allocation, |address| has to be within the first
// partition page.
template <bool thread_safe>
PA_ALWAYS_INLINE SlotSpanMetadata<thread_safe>*
SlotSpanMetadata<thread_safe>::FromAddr(uintptr_t address) {
auto* page = PartitionPage<thread_safe>::FromAddr(address);
PA_DCHECK(page->is_valid);
// Partition pages in the same slot span share the same SlotSpanMetadata
// object (located in the first PartitionPage object of that span). Adjust
// for that.
page -= page->slot_span_metadata_offset;
PA_DCHECK(page->is_valid);
PA_DCHECK(!page->slot_span_metadata_offset);
auto* slot_span = &page->slot_span_metadata;
// TODO(crbug.com/1257655): See if we can afford to make this a CHECK.
PA_DCHECK(PartitionRoot<thread_safe>::IsValidSlotSpan(slot_span));
// For direct map, if |address| doesn't point within the first partition page,
// |slot_span_metadata_offset| will be 0, |page| won't get shifted, leaving
// |slot_size| at 0.
PA_DCHECK(slot_span->bucket->slot_size);
return slot_span;
}
// Like |FromAddr|, but asserts that |slot_start| indeed points to the
// beginning of a slot. It doesn't check if the slot is actually allocated.
//
// This works on direct maps too.
template <bool thread_safe>
PA_ALWAYS_INLINE SlotSpanMetadata<thread_safe>*
SlotSpanMetadata<thread_safe>::FromSlotStart(uintptr_t slot_start) {
auto* slot_span = FromAddr(slot_start);
#if BUILDFLAG(PA_DCHECK_IS_ON)
// Checks that the pointer is a multiple of slot size.
uintptr_t slot_span_start = ToSlotSpanStart(slot_span);
PA_DCHECK(!((slot_start - slot_span_start) % slot_span->bucket->slot_size));
#endif // BUILDFLAG(PA_DCHECK_IS_ON)
return slot_span;
}
// Like |FromAddr|, but asserts that |object| indeed points to the beginning of
// an object. It doesn't check if the object is actually allocated.
//
// This works on direct maps too.
template <bool thread_safe>
PA_ALWAYS_INLINE SlotSpanMetadata<thread_safe>*
SlotSpanMetadata<thread_safe>::FromObject(void* object) {
uintptr_t object_addr = ObjectPtr2Addr(object);
auto* slot_span = FromAddr(object_addr);
#if BUILDFLAG(PA_DCHECK_IS_ON)
// Checks that the object is exactly |extras_offset| away from a multiple of
// slot size (i.e. from a slot start).
uintptr_t slot_span_start = ToSlotSpanStart(slot_span);
auto* root = PartitionRoot<thread_safe>::FromSlotSpan(slot_span);
PA_DCHECK((object_addr - slot_span_start) % slot_span->bucket->slot_size ==
root->flags.extras_offset);
#endif // BUILDFLAG(PA_DCHECK_IS_ON)
return slot_span;
}
// Like |FromAddr|, but asserts that |address| indeed points within an object.
// It doesn't check if the object is actually allocated.
//
// CAUTION! For direct-mapped allocation, |address| has to be within the first
// partition page.
template <bool thread_safe>
PA_ALWAYS_INLINE SlotSpanMetadata<thread_safe>*
SlotSpanMetadata<thread_safe>::FromObjectInnerAddr(uintptr_t address) {
auto* slot_span = FromAddr(address);
#if BUILDFLAG(PA_DCHECK_IS_ON)
// Checks that the address is within the expected object boundaries.
uintptr_t slot_span_start = ToSlotSpanStart(slot_span);
auto* root = PartitionRoot<thread_safe>::FromSlotSpan(slot_span);
uintptr_t shift_from_slot_start =
(address - slot_span_start) % slot_span->bucket->slot_size;
PA_DCHECK(shift_from_slot_start >= root->flags.extras_offset);
// Use <= to allow an address immediately past the object.
PA_DCHECK(shift_from_slot_start <=
root->flags.extras_offset + slot_span->GetUsableSize(root));
#endif // BUILDFLAG(PA_DCHECK_IS_ON)
return slot_span;
}
template <bool thread_safe>
PA_ALWAYS_INLINE SlotSpanMetadata<thread_safe>*
SlotSpanMetadata<thread_safe>::FromObjectInnerPtr(void* ptr) {
return FromObjectInnerAddr(ObjectInnerPtr2Addr(ptr));
}
template <bool thread_safe>
PA_ALWAYS_INLINE void SlotSpanMetadata<thread_safe>::SetRawSize(
size_t raw_size) {
PA_DCHECK(CanStoreRawSize());
auto* subsequent_page_metadata = GetSubsequentPageMetadata(
reinterpret_cast<PartitionPage<thread_safe>*>(this));
subsequent_page_metadata->raw_size = raw_size;
}
template <bool thread_safe>
PA_ALWAYS_INLINE size_t SlotSpanMetadata<thread_safe>::GetRawSize() const {
PA_DCHECK(CanStoreRawSize());
const auto* subsequent_page_metadata = GetSubsequentPageMetadata(
reinterpret_cast<const PartitionPage<thread_safe>*>(this));
return subsequent_page_metadata->raw_size;
}
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionTag*
SlotSpanMetadata<thread_safe>::DirectMapMTETag() {
PA_DCHECK(bucket->is_direct_mapped());
auto* subsequent_page_metadata = GetSubsequentPageMetadata(
reinterpret_cast<PartitionPage<thread_safe>*>(this));
return &subsequent_page_metadata->direct_map_tag;
}
template <bool thread_safe>
PA_ALWAYS_INLINE void SlotSpanMetadata<thread_safe>::SetFreelistHead(
PartitionFreelistEntry* new_head) {
#if BUILDFLAG(PA_DCHECK_IS_ON)
// |this| is in the metadata region, hence isn't MTE-tagged. Untag |new_head|
// as well.
uintptr_t new_head_untagged = UntagPtr(new_head);
PA_DCHECK(!new_head ||
(reinterpret_cast<uintptr_t>(this) & kSuperPageBaseMask) ==
(new_head_untagged & kSuperPageBaseMask));
#endif
freelist_head = new_head;
// Inserted something new in the freelist, assume that it is not sorted
// anymore.
freelist_is_sorted_ = false;
}
template <bool thread_safe>
PA_ALWAYS_INLINE PartitionFreelistEntry*
SlotSpanMetadata<thread_safe>::PopForAlloc(size_t size) {
// Not using bucket->slot_size directly as the compiler doesn't know that
// |bucket->slot_size| is the same as |size|.
PA_DCHECK(size == bucket->slot_size);
PartitionFreelistEntry* result = freelist_head;
// Not setting freelist_is_sorted_ to false since this doesn't destroy
// ordering.
freelist_head = freelist_head->GetNext(size);
num_allocated_slots++;
return result;
}
template <bool thread_safe>
PA_ALWAYS_INLINE void SlotSpanMetadata<thread_safe>::Free(uintptr_t slot_start)
PA_EXCLUSIVE_LOCKS_REQUIRED(
PartitionRoot<thread_safe>::FromSlotSpan(this)->lock_) {
#if BUILDFLAG(PA_DCHECK_IS_ON)
auto* root = PartitionRoot<thread_safe>::FromSlotSpan(this);
root->lock_.AssertAcquired();
#endif
auto* entry = static_cast<internal::PartitionFreelistEntry*>(
SlotStartAddr2Ptr(slot_start));
// Catches an immediate double free.
PA_CHECK(entry != freelist_head);
// Look for double free one level deeper in debug.
PA_DCHECK(!freelist_head ||
entry != freelist_head->GetNext(bucket->slot_size));
entry->SetNext(freelist_head);
SetFreelistHead(entry);
// A best effort double-free check. Works only on empty slot spans.
PA_CHECK(num_allocated_slots);
--num_allocated_slots;
// If the span is marked full, or became empty, take the slow path to update
// internal state.
if (PA_UNLIKELY(marked_full || num_allocated_slots == 0)) {
FreeSlowPath(1);
} else {
// All single-slot allocations must go through the slow path to
// correctly update the raw size.
PA_DCHECK(!CanStoreRawSize());
}
}
template <bool thread_safe>
PA_ALWAYS_INLINE void SlotSpanMetadata<thread_safe>::AppendFreeList(
PartitionFreelistEntry* head,
PartitionFreelistEntry* tail,
size_t number_of_freed)
PA_EXCLUSIVE_LOCKS_REQUIRED(
PartitionRoot<thread_safe>::FromSlotSpan(this)->lock_) {
#if BUILDFLAG(PA_DCHECK_IS_ON)
auto* root = PartitionRoot<thread_safe>::FromSlotSpan(this);
root->lock_.AssertAcquired();
PA_DCHECK(!tail->GetNext(bucket->slot_size));
PA_DCHECK(number_of_freed);
PA_DCHECK(num_allocated_slots);
if (CanStoreRawSize()) {
PA_DCHECK(number_of_freed == 1);
}
{
size_t number_of_entries = 0;
for (auto* entry = head; entry;
entry = entry->GetNext(bucket->slot_size), ++number_of_entries) {
uintptr_t untagged_entry = UntagPtr(entry);
// Check that all entries belong to this slot span.
PA_DCHECK(ToSlotSpanStart(this) <= untagged_entry);
PA_DCHECK(untagged_entry <
ToSlotSpanStart(this) + bucket->get_bytes_per_span());
}
PA_DCHECK(number_of_entries == number_of_freed);
}
#endif
tail->SetNext(freelist_head);
SetFreelistHead(head);
PA_DCHECK(num_allocated_slots >= number_of_freed);
num_allocated_slots -= number_of_freed;
// If the span is marked full, or became empty, take the slow path to update
// internal state.
if (PA_UNLIKELY(marked_full || num_allocated_slots == 0)) {
FreeSlowPath(number_of_freed);
} else {
// All single-slot allocations must go through the slow path to
// correctly update the raw size.
PA_DCHECK(!CanStoreRawSize());
}
}
template <bool thread_safe>
PA_ALWAYS_INLINE bool SlotSpanMetadata<thread_safe>::is_active() const {
PA_DCHECK(this != get_sentinel_slot_span());
bool ret =
(num_allocated_slots > 0 && (freelist_head || num_unprovisioned_slots));
if (ret) {
PA_DCHECK(!marked_full);
PA_DCHECK(num_allocated_slots < bucket->get_slots_per_span());
}
return ret;
}
template <bool thread_safe>
PA_ALWAYS_INLINE bool SlotSpanMetadata<thread_safe>::is_full() const {
PA_DCHECK(this != get_sentinel_slot_span());
bool ret = (num_allocated_slots == bucket->get_slots_per_span());
if (ret) {
PA_DCHECK(!freelist_head);
PA_DCHECK(!num_unprovisioned_slots);
// May or may not be marked full, so don't check for that.
}
return ret;
}
template <bool thread_safe>
PA_ALWAYS_INLINE bool SlotSpanMetadata<thread_safe>::is_empty() const {
PA_DCHECK(this != get_sentinel_slot_span());
bool ret = (!num_allocated_slots && freelist_head);
if (ret) {
PA_DCHECK(!marked_full);
}
return ret;
}
template <bool thread_safe>
PA_ALWAYS_INLINE bool SlotSpanMetadata<thread_safe>::is_decommitted() const {
PA_DCHECK(this != get_sentinel_slot_span());
bool ret = (!num_allocated_slots && !freelist_head);
if (ret) {
PA_DCHECK(!marked_full);
PA_DCHECK(!num_unprovisioned_slots);
PA_DCHECK(!in_empty_cache_);
}
return ret;
}
template <bool thread_safe>
PA_ALWAYS_INLINE void SlotSpanMetadata<thread_safe>::Reset() {
PA_DCHECK(is_decommitted());
num_unprovisioned_slots = bucket->get_slots_per_span();
PA_DCHECK(num_unprovisioned_slots);
ToSuperPageExtent()->IncrementNumberOfNonemptySlotSpans();
next_slot_span = nullptr;
}
#if BUILDFLAG(STARSCAN)
// Returns the state bitmap from an address within a normal-bucket super page.
// It's the caller's responsibility to ensure that the bitmap exists.
PA_ALWAYS_INLINE AllocationStateMap* StateBitmapFromAddr(uintptr_t address) {
PA_DCHECK(IsManagedByNormalBuckets(address));
uintptr_t super_page = address & kSuperPageBaseMask;
return SuperPageStateBitmap(super_page);
}
#endif // BUILDFLAG(STARSCAN)
// Iterates over all slot spans in a super-page. |Callback| must return true if
// early return is needed.
template <bool thread_safe, typename Callback>
void IterateSlotSpans(uintptr_t super_page,
bool with_quarantine,
Callback callback) {
#if BUILDFLAG(PA_DCHECK_IS_ON)
PA_DCHECK(!(super_page % kSuperPageAlignment));
auto* extent_entry = PartitionSuperPageToExtent<thread_safe>(super_page);
extent_entry->root->lock_.AssertAcquired();
#endif
using Page = PartitionPage<thread_safe>;
using SlotSpan = SlotSpanMetadata<thread_safe>;
auto* const first_page =
Page::FromAddr(SuperPagePayloadBegin(super_page, with_quarantine));
auto* const last_page =
Page::FromAddr(SuperPagePayloadEnd(super_page) - PartitionPageSize());
Page* page;
SlotSpan* slot_span;
for (page = first_page; page <= last_page;) {
PA_DCHECK(!page->slot_span_metadata_offset); // Ensure slot span beginning.
if (!page->is_valid) {
if (page->has_valid_span_after_this) {
// The page doesn't represent a valid slot span, but there is another
// one somewhere after this. Keep iterating to find it.
++page;
continue;
}
// There are currently no valid spans from here on. No need to iterate
// the rest of the super page.
break;
}
slot_span = &page->slot_span_metadata;
if (callback(slot_span))
return;
page += slot_span->bucket->get_pages_per_slot_span();
}
// Each super page must have at least one valid slot span.
PA_DCHECK(page > first_page);
// Just a quick check that the search ended at a valid slot span and there
// was no unnecessary iteration over gaps afterwards.
PA_DCHECK(page == reinterpret_cast<Page*>(slot_span) +
slot_span->bucket->get_pages_per_slot_span());
}
} // namespace partition_alloc::internal
#endif // BASE_ALLOCATOR_PARTITION_ALLOCATOR_PARTITION_PAGE_H_