summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry/mm
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry/mm')
-rw-r--r--pkg/sentry/mm/BUILD28
-rw-r--r--pkg/sentry/mm/address_space.go8
-rw-r--r--pkg/sentry/mm/aio_context.go14
-rw-r--r--pkg/sentry/mm/io.go9
-rw-r--r--pkg/sentry/mm/lifecycle.go4
-rw-r--r--pkg/sentry/mm/metadata.go5
-rw-r--r--pkg/sentry/mm/mm.go12
-rw-r--r--pkg/sentry/mm/mm_test.go3
-rw-r--r--pkg/sentry/mm/pma.go110
-rw-r--r--pkg/sentry/mm/special_mappable.go25
-rw-r--r--pkg/sentry/mm/syscalls.go24
-rw-r--r--pkg/sentry/mm/vma.go53
12 files changed, 184 insertions, 111 deletions
diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD
index a036ce53c..b4a47ccca 100644
--- a/pkg/sentry/mm/BUILD
+++ b/pkg/sentry/mm/BUILD
@@ -7,14 +7,14 @@ go_template_instance(
name = "file_refcount_set",
out = "file_refcount_set.go",
imports = {
- "platform": "gvisor.dev/gvisor/pkg/sentry/platform",
+ "memmap": "gvisor.dev/gvisor/pkg/sentry/memmap",
},
package = "mm",
prefix = "fileRefcount",
template = "//pkg/segment:generic_set",
types = {
"Key": "uint64",
- "Range": "platform.FileRange",
+ "Range": "memmap.FileRange",
"Value": "int32",
"Functions": "fileRefcountSetFunctions",
},
@@ -73,12 +73,35 @@ go_template_instance(
},
)
+go_template_instance(
+ name = "aio_mappable_refs",
+ out = "aio_mappable_refs.go",
+ package = "mm",
+ prefix = "aioMappable",
+ template = "//pkg/refs_vfs2:refs_template",
+ types = {
+ "T": "aioMappable",
+ },
+)
+
+go_template_instance(
+ name = "special_mappable_refs",
+ out = "special_mappable_refs.go",
+ package = "mm",
+ prefix = "SpecialMappable",
+ template = "//pkg/refs_vfs2:refs_template",
+ types = {
+ "T": "SpecialMappable",
+ },
+)
+
go_library(
name = "mm",
srcs = [
"address_space.go",
"aio_context.go",
"aio_context_state.go",
+ "aio_mappable_refs.go",
"debug.go",
"file_refcount_set.go",
"io.go",
@@ -92,6 +115,7 @@ go_library(
"save_restore.go",
"shm.go",
"special_mappable.go",
+ "special_mappable_refs.go",
"syscalls.go",
"vma.go",
"vma_set.go",
diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go
index 5c667117c..a93e76c75 100644
--- a/pkg/sentry/mm/address_space.go
+++ b/pkg/sentry/mm/address_space.go
@@ -166,8 +166,12 @@ func (mm *MemoryManager) Deactivate() {
// mapASLocked maps addresses in ar into mm.as. If precommit is true, mappings
// for all addresses in ar should be precommitted.
//
-// Preconditions: mm.activeMu must be locked. mm.as != nil. ar.Length() != 0.
-// ar must be page-aligned. pseg == mm.pmas.LowerBoundSegment(ar.Start).
+// Preconditions:
+// * mm.activeMu must be locked.
+// * mm.as != nil.
+// * ar.Length() != 0.
+// * ar must be page-aligned.
+// * pseg == mm.pmas.LowerBoundSegment(ar.Start).
func (mm *MemoryManager) mapASLocked(pseg pmaIterator, ar usermem.AddrRange, precommit bool) error {
// By default, map entire pmas at a time, under the assumption that there
// is no cost to mapping more of a pma than necessary.
diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go
index 379148903..7bf48cb2c 100644
--- a/pkg/sentry/mm/aio_context.go
+++ b/pkg/sentry/mm/aio_context.go
@@ -17,10 +17,8 @@ package mm
import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/context"
- "gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
- "gvisor.dev/gvisor/pkg/sentry/platform"
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
@@ -240,10 +238,10 @@ func (ctx *AIOContext) Drain() {
//
// +stateify savable
type aioMappable struct {
- refs.AtomicRefCount
+ aioMappableRefs
mfp pgalloc.MemoryFileProvider
- fr platform.FileRange
+ fr memmap.FileRange
}
var aioRingBufferSize = uint64(usermem.Addr(linux.AIORingSize).MustRoundUp())
@@ -254,13 +252,13 @@ func newAIOMappable(mfp pgalloc.MemoryFileProvider) (*aioMappable, error) {
return nil, err
}
m := aioMappable{mfp: mfp, fr: fr}
- m.EnableLeakCheck("mm.aioMappable")
+ m.EnableLeakCheck()
return &m, nil
}
// DecRef implements refs.RefCounter.DecRef.
-func (m *aioMappable) DecRef() {
- m.AtomicRefCount.DecRefWithDestructor(func() {
+func (m *aioMappable) DecRef(ctx context.Context) {
+ m.aioMappableRefs.DecRef(func() {
m.mfp.MemoryFile().DecRef(m.fr)
})
}
@@ -368,7 +366,7 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint
if err != nil {
return 0, err
}
- defer m.DecRef()
+ defer m.DecRef(ctx)
addr, err := mm.MMap(ctx, memmap.MMapOpts{
Length: aioRingBufferSize,
MappingIdentity: m,
diff --git a/pkg/sentry/mm/io.go b/pkg/sentry/mm/io.go
index fa776f9c6..a8ac48080 100644
--- a/pkg/sentry/mm/io.go
+++ b/pkg/sentry/mm/io.go
@@ -441,7 +441,10 @@ func (mm *MemoryManager) LoadUint32(ctx context.Context, addr usermem.Addr, opts
// handleASIOFault handles a page fault at address addr for an AddressSpaceIO
// operation spanning ioar.
//
-// Preconditions: mm.as != nil. ioar.Length() != 0. ioar.Contains(addr).
+// Preconditions:
+// * mm.as != nil.
+// * ioar.Length() != 0.
+// * ioar.Contains(addr).
func (mm *MemoryManager) handleASIOFault(ctx context.Context, addr usermem.Addr, ioar usermem.AddrRange, at usermem.AccessType) error {
// Try to map all remaining pages in the I/O operation. This RoundUp can't
// overflow because otherwise it would have been caught by CheckIORange.
@@ -629,7 +632,9 @@ func (mm *MemoryManager) withVecInternalMappings(ctx context.Context, ars userme
// at most address end on AddrRange arsit.Head(). It is used in vector I/O paths to
// truncate usermem.AddrRangeSeq when errors occur.
//
-// Preconditions: !arsit.IsEmpty(). end <= arsit.Head().End.
+// Preconditions:
+// * !arsit.IsEmpty().
+// * end <= arsit.Head().End.
func truncatedAddrRangeSeq(ars, arsit usermem.AddrRangeSeq, end usermem.Addr) usermem.AddrRangeSeq {
ar := arsit.Head()
if end <= ar.Start {
diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go
index aac56679b..09dbc06a4 100644
--- a/pkg/sentry/mm/lifecycle.go
+++ b/pkg/sentry/mm/lifecycle.go
@@ -57,6 +57,8 @@ func (mm *MemoryManager) SetMmapLayout(ac arch.Context, r *limits.LimitSet) (arc
// Fork creates a copy of mm with 1 user, as for Linux syscalls fork() or
// clone() (without CLONE_VM).
func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) {
+ mm.AddressSpace().PreFork()
+ defer mm.AddressSpace().PostFork()
mm.metadataMu.Lock()
defer mm.metadataMu.Unlock()
mm.mappingMu.RLock()
@@ -258,7 +260,7 @@ func (mm *MemoryManager) DecUsers(ctx context.Context) {
mm.executable = nil
mm.metadataMu.Unlock()
if exe != nil {
- exe.DecRef()
+ exe.DecRef(ctx)
}
mm.activeMu.Lock()
diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go
index 28e5057f7..0cfd60f6c 100644
--- a/pkg/sentry/mm/metadata.go
+++ b/pkg/sentry/mm/metadata.go
@@ -15,6 +15,7 @@
package mm
import (
+ "gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/fsbridge"
"gvisor.dev/gvisor/pkg/usermem"
@@ -147,7 +148,7 @@ func (mm *MemoryManager) Executable() fsbridge.File {
// SetExecutable sets the executable.
//
// This takes a reference on d.
-func (mm *MemoryManager) SetExecutable(file fsbridge.File) {
+func (mm *MemoryManager) SetExecutable(ctx context.Context, file fsbridge.File) {
mm.metadataMu.Lock()
// Grab a new reference.
@@ -164,7 +165,7 @@ func (mm *MemoryManager) SetExecutable(file fsbridge.File) {
// Do this without holding the lock, since it may wind up doing some
// I/O to sync the dirent, etc.
if orig != nil {
- orig.DecRef()
+ orig.DecRef(ctx)
}
}
diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go
index 6db7c3d40..8c9f11cce 100644
--- a/pkg/sentry/mm/mm.go
+++ b/pkg/sentry/mm/mm.go
@@ -25,7 +25,7 @@
// Locks taken by memmap.Mappable.Translate
// mm.privateRefs.mu
// platform.AddressSpace locks
-// platform.File locks
+// memmap.File locks
// mm.aioManager.mu
// mm.AIOContext.mu
//
@@ -242,7 +242,7 @@ type MemoryManager struct {
// +stateify savable
type vma struct {
// mappable is the virtual memory object mapped by this vma. If mappable is
- // nil, the vma represents a private anonymous mapping.
+ // nil, the vma represents an anonymous mapping.
mappable memmap.Mappable
// off is the offset into mappable at which this vma begins. If mappable is
@@ -396,7 +396,7 @@ type pma struct {
// file is the file mapped by this pma. Only pmas for which file ==
// MemoryManager.mfp.MemoryFile() may be saved. pmas hold a reference to
// the corresponding file range while they exist.
- file platform.File `state:"nosave"`
+ file memmap.File `state:"nosave"`
// off is the offset into file at which this pma begins.
//
@@ -436,7 +436,7 @@ type pma struct {
private bool
// If internalMappings is not empty, it is the cached return value of
- // file.MapInternal for the platform.FileRange mapped by this pma.
+ // file.MapInternal for the memmap.FileRange mapped by this pma.
internalMappings safemem.BlockSeq `state:"nosave"`
}
@@ -469,10 +469,10 @@ func (fileRefcountSetFunctions) MaxKey() uint64 {
func (fileRefcountSetFunctions) ClearValue(_ *int32) {
}
-func (fileRefcountSetFunctions) Merge(_ platform.FileRange, rc1 int32, _ platform.FileRange, rc2 int32) (int32, bool) {
+func (fileRefcountSetFunctions) Merge(_ memmap.FileRange, rc1 int32, _ memmap.FileRange, rc2 int32) (int32, bool) {
return rc1, rc1 == rc2
}
-func (fileRefcountSetFunctions) Split(_ platform.FileRange, rc int32, _ uint64) (int32, int32) {
+func (fileRefcountSetFunctions) Split(_ memmap.FileRange, rc int32, _ uint64) (int32, int32) {
return rc, rc
}
diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go
index fdc308542..acac3d357 100644
--- a/pkg/sentry/mm/mm_test.go
+++ b/pkg/sentry/mm/mm_test.go
@@ -51,7 +51,8 @@ func TestUsageASUpdates(t *testing.T) {
defer mm.DecUsers(ctx)
addr, err := mm.MMap(ctx, memmap.MMapOpts{
- Length: 2 * usermem.PageSize,
+ Length: 2 * usermem.PageSize,
+ Private: true,
})
if err != nil {
t.Fatalf("MMap got err %v want nil", err)
diff --git a/pkg/sentry/mm/pma.go b/pkg/sentry/mm/pma.go
index 62e4c20af..30facebf7 100644
--- a/pkg/sentry/mm/pma.go
+++ b/pkg/sentry/mm/pma.go
@@ -21,7 +21,6 @@ import (
"gvisor.dev/gvisor/pkg/safecopy"
"gvisor.dev/gvisor/pkg/safemem"
"gvisor.dev/gvisor/pkg/sentry/memmap"
- "gvisor.dev/gvisor/pkg/sentry/platform"
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -32,7 +31,9 @@ import (
// iterator to the pma containing ar.Start. Otherwise it returns a terminal
// iterator.
//
-// Preconditions: mm.activeMu must be locked. ar.Length() != 0.
+// Preconditions:
+// * mm.activeMu must be locked.
+// * ar.Length() != 0.
func (mm *MemoryManager) existingPMAsLocked(ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool, needInternalMappings bool) pmaIterator {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 {
@@ -90,10 +91,13 @@ func (mm *MemoryManager) existingVecPMAsLocked(ars usermem.AddrRangeSeq, at user
//
// - An error that is non-nil if pmas exist for only a subset of ar.
//
-// Preconditions: mm.mappingMu must be locked. mm.activeMu must be locked for
-// writing. ar.Length() != 0. vseg.Range().Contains(ar.Start). vmas must exist
-// for all addresses in ar, and support accesses of type at (i.e. permission
-// checks must have been performed against vmas).
+// Preconditions:
+// * mm.mappingMu must be locked.
+// * mm.activeMu must be locked for writing.
+// * ar.Length() != 0.
+// * vseg.Range().Contains(ar.Start).
+// * vmas must exist for all addresses in ar, and support accesses of type at
+// (i.e. permission checks must have been performed against vmas).
func (mm *MemoryManager) getPMAsLocked(ctx context.Context, vseg vmaIterator, ar usermem.AddrRange, at usermem.AccessType) (pmaIterator, pmaGapIterator, error) {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 {
@@ -136,9 +140,11 @@ func (mm *MemoryManager) getPMAsLocked(ctx context.Context, vseg vmaIterator, ar
// exist. If this is not equal to ars, it returns a non-nil error explaining
// why.
//
-// Preconditions: mm.mappingMu must be locked. mm.activeMu must be locked for
-// writing. vmas must exist for all addresses in ars, and support accesses of
-// type at (i.e. permission checks must have been performed against vmas).
+// Preconditions:
+// * mm.mappingMu must be locked.
+// * mm.activeMu must be locked for writing.
+// * vmas must exist for all addresses in ars, and support accesses of type at
+// (i.e. permission checks must have been performed against vmas).
func (mm *MemoryManager) getVecPMAsLocked(ctx context.Context, ars usermem.AddrRangeSeq, at usermem.AccessType) (usermem.AddrRangeSeq, error) {
for arsit := ars; !arsit.IsEmpty(); arsit = arsit.Tail() {
ar := arsit.Head()
@@ -519,8 +525,10 @@ func privateAligned(ar usermem.AddrRange) usermem.AddrRange {
// the memory it maps, isPMACopyOnWriteLocked will take ownership of the memory
// and update the pma to indicate that it does not require copy-on-write.
//
-// Preconditions: vseg.Range().IsSupersetOf(pseg.Range()). mm.mappingMu must be
-// locked. mm.activeMu must be locked for writing.
+// Preconditions:
+// * vseg.Range().IsSupersetOf(pseg.Range()).
+// * mm.mappingMu must be locked.
+// * mm.activeMu must be locked for writing.
func (mm *MemoryManager) isPMACopyOnWriteLocked(vseg vmaIterator, pseg pmaIterator) bool {
pma := pseg.ValuePtr()
if !pma.needCOW {
@@ -569,8 +577,10 @@ func (mm *MemoryManager) Invalidate(ar usermem.AddrRange, opts memmap.Invalidate
// invalidateLocked removes pmas and AddressSpace mappings of those pmas for
// addresses in ar.
//
-// Preconditions: mm.activeMu must be locked for writing. ar.Length() != 0. ar
-// must be page-aligned.
+// Preconditions:
+// * mm.activeMu must be locked for writing.
+// * ar.Length() != 0.
+// * ar must be page-aligned.
func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivate, invalidateShared bool) {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() {
@@ -604,7 +614,7 @@ func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivat
}
}
-// Pin returns the platform.File ranges currently mapped by addresses in ar in
+// Pin returns the memmap.File ranges currently mapped by addresses in ar in
// mm, acquiring a reference on the returned ranges which the caller must
// release by calling Unpin. If not all addresses are mapped, Pin returns a
// non-nil error. Note that Pin may return both a non-empty slice of
@@ -614,7 +624,9 @@ func (mm *MemoryManager) invalidateLocked(ar usermem.AddrRange, invalidatePrivat
// most I/O. It should only be used in contexts that would use get_user_pages()
// in the Linux kernel.
//
-// Preconditions: ar.Length() != 0. ar must be page-aligned.
+// Preconditions:
+// * ar.Length() != 0.
+// * ar must be page-aligned.
func (mm *MemoryManager) Pin(ctx context.Context, ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool) ([]PinnedRange, error) {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() {
@@ -674,15 +686,15 @@ type PinnedRange struct {
Source usermem.AddrRange
// File is the mapped file.
- File platform.File
+ File memmap.File
// Offset is the offset into File at which this PinnedRange begins.
Offset uint64
}
-// FileRange returns the platform.File offsets mapped by pr.
-func (pr PinnedRange) FileRange() platform.FileRange {
- return platform.FileRange{pr.Offset, pr.Offset + uint64(pr.Source.Length())}
+// FileRange returns the memmap.File offsets mapped by pr.
+func (pr PinnedRange) FileRange() memmap.FileRange {
+ return memmap.FileRange{pr.Offset, pr.Offset + uint64(pr.Source.Length())}
}
// Unpin releases the reference held by prs.
@@ -694,9 +706,13 @@ func Unpin(prs []PinnedRange) {
// movePMAsLocked moves all pmas in oldAR to newAR.
//
-// Preconditions: mm.activeMu must be locked for writing. oldAR.Length() != 0.
-// oldAR.Length() <= newAR.Length(). !oldAR.Overlaps(newAR).
-// mm.pmas.IsEmptyRange(newAR). oldAR and newAR must be page-aligned.
+// Preconditions:
+// * mm.activeMu must be locked for writing.
+// * oldAR.Length() != 0.
+// * oldAR.Length() <= newAR.Length().
+// * !oldAR.Overlaps(newAR).
+// * mm.pmas.IsEmptyRange(newAR).
+// * oldAR and newAR must be page-aligned.
func (mm *MemoryManager) movePMAsLocked(oldAR, newAR usermem.AddrRange) {
if checkInvariants {
if !oldAR.WellFormed() || oldAR.Length() <= 0 || !oldAR.IsPageAligned() {
@@ -752,9 +768,11 @@ func (mm *MemoryManager) movePMAsLocked(oldAR, newAR usermem.AddrRange) {
// - An error that is non-nil if internal mappings exist for only a subset of
// ar.
//
-// Preconditions: mm.activeMu must be locked for writing.
-// pseg.Range().Contains(ar.Start). pmas must exist for all addresses in ar.
-// ar.Length() != 0.
+// Preconditions:
+// * mm.activeMu must be locked for writing.
+// * pseg.Range().Contains(ar.Start).
+// * pmas must exist for all addresses in ar.
+// * ar.Length() != 0.
//
// Postconditions: getPMAInternalMappingsLocked does not invalidate iterators
// into mm.pmas.
@@ -784,8 +802,9 @@ func (mm *MemoryManager) getPMAInternalMappingsLocked(pseg pmaIterator, ar userm
// internal mappings exist. If this is not equal to ars, it returns a non-nil
// error explaining why.
//
-// Preconditions: mm.activeMu must be locked for writing. pmas must exist for
-// all addresses in ar.
+// Preconditions:
+// * mm.activeMu must be locked for writing.
+// * pmas must exist for all addresses in ar.
//
// Postconditions: getVecPMAInternalMappingsLocked does not invalidate iterators
// into mm.pmas.
@@ -804,9 +823,12 @@ func (mm *MemoryManager) getVecPMAInternalMappingsLocked(ars usermem.AddrRangeSe
// internalMappingsLocked returns internal mappings for addresses in ar.
//
-// Preconditions: mm.activeMu must be locked. Internal mappings must have been
-// previously established for all addresses in ar. ar.Length() != 0.
-// pseg.Range().Contains(ar.Start).
+// Preconditions:
+// * mm.activeMu must be locked.
+// * Internal mappings must have been previously established for all addresses
+// in ar.
+// * ar.Length() != 0.
+// * pseg.Range().Contains(ar.Start).
func (mm *MemoryManager) internalMappingsLocked(pseg pmaIterator, ar usermem.AddrRange) safemem.BlockSeq {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 {
@@ -840,8 +862,10 @@ func (mm *MemoryManager) internalMappingsLocked(pseg pmaIterator, ar usermem.Add
// vecInternalMappingsLocked returns internal mappings for addresses in ars.
//
-// Preconditions: mm.activeMu must be locked. Internal mappings must have been
-// previously established for all addresses in ars.
+// Preconditions:
+// * mm.activeMu must be locked.
+// * Internal mappings must have been previously established for all addresses
+// in ars.
func (mm *MemoryManager) vecInternalMappingsLocked(ars usermem.AddrRangeSeq) safemem.BlockSeq {
var ims []safemem.Block
for ; !ars.IsEmpty(); ars = ars.Tail() {
@@ -857,7 +881,7 @@ func (mm *MemoryManager) vecInternalMappingsLocked(ars usermem.AddrRangeSeq) saf
}
// incPrivateRef acquires a reference on private pages in fr.
-func (mm *MemoryManager) incPrivateRef(fr platform.FileRange) {
+func (mm *MemoryManager) incPrivateRef(fr memmap.FileRange) {
mm.privateRefs.mu.Lock()
defer mm.privateRefs.mu.Unlock()
refSet := &mm.privateRefs.refs
@@ -878,8 +902,8 @@ func (mm *MemoryManager) incPrivateRef(fr platform.FileRange) {
}
// decPrivateRef releases a reference on private pages in fr.
-func (mm *MemoryManager) decPrivateRef(fr platform.FileRange) {
- var freed []platform.FileRange
+func (mm *MemoryManager) decPrivateRef(fr memmap.FileRange) {
+ var freed []memmap.FileRange
mm.privateRefs.mu.Lock()
refSet := &mm.privateRefs.refs
@@ -951,7 +975,7 @@ func (pmaSetFunctions) Merge(ar1 usermem.AddrRange, pma1 pma, ar2 usermem.AddrRa
// Discard internal mappings instead of trying to merge them, since merging
// them requires an allocation and getting them again from the
- // platform.File might not.
+ // memmap.File might not.
pma1.internalMappings = safemem.BlockSeq{}
return pma1, true
}
@@ -970,7 +994,9 @@ func (pmaSetFunctions) Split(ar usermem.AddrRange, p pma, split usermem.Addr) (p
// findOrSeekPrevUpperBoundPMA returns mm.pmas.UpperBoundSegment(addr), but may do
// so by scanning linearly backward from pgap.
//
-// Preconditions: mm.activeMu must be locked. addr <= pgap.Start().
+// Preconditions:
+// * mm.activeMu must be locked.
+// * addr <= pgap.Start().
func (mm *MemoryManager) findOrSeekPrevUpperBoundPMA(addr usermem.Addr, pgap pmaGapIterator) pmaIterator {
if checkInvariants {
if !pgap.Ok() {
@@ -1012,12 +1038,14 @@ func (pseg pmaIterator) getInternalMappingsLocked() error {
return nil
}
-func (pseg pmaIterator) fileRange() platform.FileRange {
+func (pseg pmaIterator) fileRange() memmap.FileRange {
return pseg.fileRangeOf(pseg.Range())
}
-// Preconditions: pseg.Range().IsSupersetOf(ar). ar.Length != 0.
-func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) platform.FileRange {
+// Preconditions:
+// * pseg.Range().IsSupersetOf(ar).
+// * ar.Length != 0.
+func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) memmap.FileRange {
if checkInvariants {
if !pseg.Ok() {
panic("terminal pma iterator")
@@ -1032,5 +1060,5 @@ func (pseg pmaIterator) fileRangeOf(ar usermem.AddrRange) platform.FileRange {
pma := pseg.ValuePtr()
pstart := pseg.Start()
- return platform.FileRange{pma.off + uint64(ar.Start-pstart), pma.off + uint64(ar.End-pstart)}
+ return memmap.FileRange{pma.off + uint64(ar.Start-pstart), pma.off + uint64(ar.End-pstart)}
}
diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go
index 9ad52082d..2dbe5b751 100644
--- a/pkg/sentry/mm/special_mappable.go
+++ b/pkg/sentry/mm/special_mappable.go
@@ -16,10 +16,8 @@ package mm
import (
"gvisor.dev/gvisor/pkg/context"
- "gvisor.dev/gvisor/pkg/refs"
"gvisor.dev/gvisor/pkg/sentry/memmap"
"gvisor.dev/gvisor/pkg/sentry/pgalloc"
- "gvisor.dev/gvisor/pkg/sentry/platform"
"gvisor.dev/gvisor/pkg/sentry/usage"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -32,10 +30,10 @@ import (
//
// +stateify savable
type SpecialMappable struct {
- refs.AtomicRefCount
+ SpecialMappableRefs
mfp pgalloc.MemoryFileProvider
- fr platform.FileRange
+ fr memmap.FileRange
name string
}
@@ -44,15 +42,15 @@ type SpecialMappable struct {
// SpecialMappable will use the given name in /proc/[pid]/maps.
//
// Preconditions: fr.Length() != 0.
-func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr platform.FileRange) *SpecialMappable {
+func NewSpecialMappable(name string, mfp pgalloc.MemoryFileProvider, fr memmap.FileRange) *SpecialMappable {
m := SpecialMappable{mfp: mfp, fr: fr, name: name}
- m.EnableLeakCheck("mm.SpecialMappable")
+ m.EnableLeakCheck()
return &m
}
// DecRef implements refs.RefCounter.DecRef.
-func (m *SpecialMappable) DecRef() {
- m.AtomicRefCount.DecRefWithDestructor(func() {
+func (m *SpecialMappable) DecRef(ctx context.Context) {
+ m.SpecialMappableRefs.DecRef(func() {
m.mfp.MemoryFile().DecRef(m.fr)
})
}
@@ -126,7 +124,7 @@ func (m *SpecialMappable) MemoryFileProvider() pgalloc.MemoryFileProvider {
// FileRange returns the offsets into MemoryFileProvider().MemoryFile() that
// store the SpecialMappable's contents.
-func (m *SpecialMappable) FileRange() platform.FileRange {
+func (m *SpecialMappable) FileRange() memmap.FileRange {
return m.fr
}
@@ -138,9 +136,12 @@ func (m *SpecialMappable) Length() uint64 {
// NewSharedAnonMappable returns a SpecialMappable that implements the
// semantics of mmap(MAP_SHARED|MAP_ANONYMOUS) and mappings of /dev/zero.
//
-// TODO(jamieliu): The use of SpecialMappable is a lazy code reuse hack. Linux
-// uses an ephemeral file created by mm/shmem.c:shmem_zero_setup(); we should
-// do the same to get non-zero device and inode IDs.
+// TODO(gvisor.dev/issue/1624): Linux uses an ephemeral file created by
+// mm/shmem.c:shmem_zero_setup(), and VFS2 does something analogous. VFS1 uses
+// a SpecialMappable instead, incorrectly getting device and inode IDs of zero
+// and causing memory for shared anonymous mappings to be allocated up-front
+// instead of on first touch; this is to avoid exacerbating the fs.MountSource
+// leak (b/143656263). Delete this function along with VFS1.
func NewSharedAnonMappable(length uint64, mfp pgalloc.MemoryFileProvider) (*SpecialMappable, error) {
if length == 0 {
return nil, syserror.EINVAL
diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go
index 3f496aa9f..a2555ba1a 100644
--- a/pkg/sentry/mm/syscalls.go
+++ b/pkg/sentry/mm/syscalls.go
@@ -24,7 +24,6 @@ import (
"gvisor.dev/gvisor/pkg/sentry/kernel/futex"
"gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/memmap"
- "gvisor.dev/gvisor/pkg/sentry/pgalloc"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
)
@@ -93,18 +92,6 @@ func (mm *MemoryManager) MMap(ctx context.Context, opts memmap.MMapOpts) (userme
}
} else {
opts.Offset = 0
- if !opts.Private {
- if opts.MappingIdentity != nil {
- return 0, syserror.EINVAL
- }
- m, err := NewSharedAnonMappable(opts.Length, pgalloc.MemoryFileProviderFromContext(ctx))
- if err != nil {
- return 0, err
- }
- defer m.DecRef()
- opts.MappingIdentity = m
- opts.Mappable = m
- }
}
if opts.Addr.RoundDown() != opts.Addr {
@@ -166,7 +153,9 @@ func (mm *MemoryManager) MMap(ctx context.Context, opts memmap.MMapOpts) (userme
// populateVMA obtains pmas for addresses in ar in the given vma, and maps them
// into mm.as if it is active.
//
-// Preconditions: mm.mappingMu must be locked. vseg.Range().IsSupersetOf(ar).
+// Preconditions:
+// * mm.mappingMu must be locked.
+// * vseg.Range().IsSupersetOf(ar).
func (mm *MemoryManager) populateVMA(ctx context.Context, vseg vmaIterator, ar usermem.AddrRange, precommit bool) {
if !vseg.ValuePtr().effectivePerms.Any() {
// Linux doesn't populate inaccessible pages. See
@@ -208,8 +197,9 @@ func (mm *MemoryManager) populateVMA(ctx context.Context, vseg vmaIterator, ar u
// preferable to populateVMA since it unlocks mm.mappingMu before performing
// expensive operations that don't require it to be locked.
//
-// Preconditions: mm.mappingMu must be locked for writing.
-// vseg.Range().IsSupersetOf(ar).
+// Preconditions:
+// * mm.mappingMu must be locked for writing.
+// * vseg.Range().IsSupersetOf(ar).
//
// Postconditions: mm.mappingMu will be unlocked.
func (mm *MemoryManager) populateVMAAndUnlock(ctx context.Context, vseg vmaIterator, ar usermem.AddrRange, precommit bool) {
@@ -1191,7 +1181,7 @@ func (mm *MemoryManager) MSync(ctx context.Context, addr usermem.Addr, length ui
mr := vseg.mappableRangeOf(vseg.Range().Intersect(ar))
mm.mappingMu.RUnlock()
err := id.Msync(ctx, mr)
- id.DecRef()
+ id.DecRef(ctx)
if err != nil {
return err
}
diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go
index 16d8207e9..f769d8294 100644
--- a/pkg/sentry/mm/vma.go
+++ b/pkg/sentry/mm/vma.go
@@ -27,8 +27,9 @@ import (
"gvisor.dev/gvisor/pkg/usermem"
)
-// Preconditions: mm.mappingMu must be locked for writing. opts must be valid
-// as defined by the checks in MMap.
+// Preconditions:
+// * mm.mappingMu must be locked for writing.
+// * opts must be valid as defined by the checks in MMap.
func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOpts) (vmaIterator, usermem.AddrRange, error) {
if opts.MaxPerms != opts.MaxPerms.Effective() {
panic(fmt.Sprintf("Non-effective MaxPerms %s cannot be enforced", opts.MaxPerms))
@@ -42,7 +43,12 @@ func (mm *MemoryManager) createVMALocked(ctx context.Context, opts memmap.MMapOp
Map32Bit: opts.Map32Bit,
})
if err != nil {
- return vmaIterator{}, usermem.AddrRange{}, err
+ // Can't force without opts.Unmap and opts.Fixed.
+ if opts.Force && opts.Unmap && opts.Fixed {
+ addr = opts.Addr
+ } else {
+ return vmaIterator{}, usermem.AddrRange{}, err
+ }
}
ar, _ := addr.ToRange(opts.Length)
@@ -255,8 +261,9 @@ func (mm *MemoryManager) mlockedBytesRangeLocked(ar usermem.AddrRange) uint64 {
//
// - An error that is non-nil if vmas exist for only a subset of ar.
//
-// Preconditions: mm.mappingMu must be locked for reading; it may be
-// temporarily unlocked. ar.Length() != 0.
+// Preconditions:
+// * mm.mappingMu must be locked for reading; it may be temporarily unlocked.
+// * ar.Length() != 0.
func (mm *MemoryManager) getVMAsLocked(ctx context.Context, ar usermem.AddrRange, at usermem.AccessType, ignorePermissions bool) (vmaIterator, vmaGapIterator, error) {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 {
@@ -337,8 +344,10 @@ const guardBytes = 256 * usermem.PageSize
// unmapLocked unmaps all addresses in ar and returns the resulting gap in
// mm.vmas.
//
-// Preconditions: mm.mappingMu must be locked for writing. ar.Length() != 0.
-// ar must be page-aligned.
+// Preconditions:
+// * mm.mappingMu must be locked for writing.
+// * ar.Length() != 0.
+// * ar must be page-aligned.
func (mm *MemoryManager) unmapLocked(ctx context.Context, ar usermem.AddrRange) vmaGapIterator {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() {
@@ -356,8 +365,10 @@ func (mm *MemoryManager) unmapLocked(ctx context.Context, ar usermem.AddrRange)
// gap in mm.vmas. It does not remove pmas or AddressSpace mappings; clients
// must do so before calling removeVMAsLocked.
//
-// Preconditions: mm.mappingMu must be locked for writing. ar.Length() != 0. ar
-// must be page-aligned.
+// Preconditions:
+// * mm.mappingMu must be locked for writing.
+// * ar.Length() != 0.
+// * ar must be page-aligned.
func (mm *MemoryManager) removeVMAsLocked(ctx context.Context, ar usermem.AddrRange) vmaGapIterator {
if checkInvariants {
if !ar.WellFormed() || ar.Length() <= 0 || !ar.IsPageAligned() {
@@ -377,7 +388,7 @@ func (mm *MemoryManager) removeVMAsLocked(ctx context.Context, ar usermem.AddrRa
vma.mappable.RemoveMapping(ctx, mm, vmaAR, vma.off, vma.canWriteMappableLocked())
}
if vma.id != nil {
- vma.id.DecRef()
+ vma.id.DecRef(ctx)
}
mm.usageAS -= uint64(vmaAR.Length())
if vma.isPrivateDataLocked() {
@@ -446,7 +457,7 @@ func (vmaSetFunctions) Merge(ar1 usermem.AddrRange, vma1 vma, ar2 usermem.AddrRa
}
if vma2.id != nil {
- vma2.id.DecRef()
+ vma2.id.DecRef(context.Background())
}
return vma1, true
}
@@ -462,7 +473,9 @@ func (vmaSetFunctions) Split(ar usermem.AddrRange, v vma, split usermem.Addr) (v
return v, v2
}
-// Preconditions: vseg.ValuePtr().mappable != nil. vseg.Range().Contains(addr).
+// Preconditions:
+// * vseg.ValuePtr().mappable != nil.
+// * vseg.Range().Contains(addr).
func (vseg vmaIterator) mappableOffsetAt(addr usermem.Addr) uint64 {
if checkInvariants {
if !vseg.Ok() {
@@ -486,8 +499,10 @@ func (vseg vmaIterator) mappableRange() memmap.MappableRange {
return vseg.mappableRangeOf(vseg.Range())
}
-// Preconditions: vseg.ValuePtr().mappable != nil.
-// vseg.Range().IsSupersetOf(ar). ar.Length() != 0.
+// Preconditions:
+// * vseg.ValuePtr().mappable != nil.
+// * vseg.Range().IsSupersetOf(ar).
+// * ar.Length() != 0.
func (vseg vmaIterator) mappableRangeOf(ar usermem.AddrRange) memmap.MappableRange {
if checkInvariants {
if !vseg.Ok() {
@@ -509,8 +524,10 @@ func (vseg vmaIterator) mappableRangeOf(ar usermem.AddrRange) memmap.MappableRan
return memmap.MappableRange{vma.off + uint64(ar.Start-vstart), vma.off + uint64(ar.End-vstart)}
}
-// Preconditions: vseg.ValuePtr().mappable != nil.
-// vseg.mappableRange().IsSupersetOf(mr). mr.Length() != 0.
+// Preconditions:
+// * vseg.ValuePtr().mappable != nil.
+// * vseg.mappableRange().IsSupersetOf(mr).
+// * mr.Length() != 0.
func (vseg vmaIterator) addrRangeOf(mr memmap.MappableRange) usermem.AddrRange {
if checkInvariants {
if !vseg.Ok() {
@@ -535,7 +552,9 @@ func (vseg vmaIterator) addrRangeOf(mr memmap.MappableRange) usermem.AddrRange {
// seekNextLowerBound returns mm.vmas.LowerBoundSegment(addr), but does so by
// scanning linearly forward from vseg.
//
-// Preconditions: mm.mappingMu must be locked. addr >= vseg.Start().
+// Preconditions:
+// * mm.mappingMu must be locked.
+// * addr >= vseg.Start().
func (vseg vmaIterator) seekNextLowerBound(addr usermem.Addr) vmaIterator {
if checkInvariants {
if !vseg.Ok() {