diff options
Diffstat (limited to 'pkg/sentry/mm')
-rw-r--r-- | pkg/sentry/mm/BUILD | 28 | ||||
-rw-r--r-- | pkg/sentry/mm/README.md | 8 | ||||
-rw-r--r-- | pkg/sentry/mm/address_space.go | 54 | ||||
-rw-r--r-- | pkg/sentry/mm/aio_context.go | 108 | ||||
-rw-r--r-- | pkg/sentry/mm/aio_context_state.go | 2 | ||||
-rw-r--r-- | pkg/sentry/mm/debug.go | 2 | ||||
-rw-r--r-- | pkg/sentry/mm/io.go | 6 | ||||
-rw-r--r-- | pkg/sentry/mm/lifecycle.go | 42 | ||||
-rw-r--r-- | pkg/sentry/mm/metadata.go | 26 | ||||
-rw-r--r-- | pkg/sentry/mm/mm.go | 27 | ||||
-rw-r--r-- | pkg/sentry/mm/mm_test.go | 8 | ||||
-rw-r--r-- | pkg/sentry/mm/pma.go | 8 | ||||
-rw-r--r-- | pkg/sentry/mm/procfs.go | 20 | ||||
-rw-r--r-- | pkg/sentry/mm/save_restore.go | 2 | ||||
-rw-r--r-- | pkg/sentry/mm/shm.go | 4 | ||||
-rw-r--r-- | pkg/sentry/mm/special_mappable.go | 4 | ||||
-rw-r--r-- | pkg/sentry/mm/syscalls.go | 8 | ||||
-rw-r--r-- | pkg/sentry/mm/vma.go | 4 |
18 files changed, 219 insertions, 142 deletions
diff --git a/pkg/sentry/mm/BUILD b/pkg/sentry/mm/BUILD index a804b8b5c..73591dab7 100644 --- a/pkg/sentry/mm/BUILD +++ b/pkg/sentry/mm/BUILD @@ -1,6 +1,5 @@ -load("@io_bazel_rules_go//go:def.bzl", "go_test") +load("//tools:defs.bzl", "go_library", "go_test") load("//tools/go_generics:defs.bzl", "go_template_instance") -load("//tools/go_stateify:defs.bzl", "go_library") package(licenses = ["notice"]) @@ -28,7 +27,7 @@ go_template_instance( "minDegree": "8", }, imports = { - "usermem": "gvisor.dev/gvisor/pkg/sentry/usermem", + "usermem": "gvisor.dev/gvisor/pkg/usermem", }, package = "mm", prefix = "vma", @@ -48,7 +47,7 @@ go_template_instance( "minDegree": "8", }, imports = { - "usermem": "gvisor.dev/gvisor/pkg/sentry/usermem", + "usermem": "gvisor.dev/gvisor/pkg/usermem", }, package = "mm", prefix = "pma", @@ -96,17 +95,18 @@ go_library( "vma.go", "vma_set.go", ], - importpath = "gvisor.dev/gvisor/pkg/sentry/mm", visibility = ["//pkg/sentry:internal"], deps = [ "//pkg/abi/linux", "//pkg/atomicbitops", + "//pkg/context", "//pkg/log", "//pkg/refs", + "//pkg/safecopy", + "//pkg/safemem", "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/fs", "//pkg/sentry/fs/proc/seqfile", + "//pkg/sentry/fsbridge", "//pkg/sentry/kernel/auth", "//pkg/sentry/kernel/futex", "//pkg/sentry/kernel/shm", @@ -114,13 +114,11 @@ go_library( "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/platform/safecopy", - "//pkg/sentry/safemem", "//pkg/sentry/usage", - "//pkg/sentry/usermem", + "//pkg/sync", "//pkg/syserror", "//pkg/tcpip/buffer", - "//third_party/gvsync", + "//pkg/usermem", ], ) @@ -128,16 +126,16 @@ go_test( name = "mm_test", size = "small", srcs = ["mm_test.go"], - embed = [":mm"], + library = ":mm", deps = [ + "//pkg/context", "//pkg/sentry/arch", - "//pkg/sentry/context", - "//pkg/sentry/context/contexttest", + "//pkg/sentry/contexttest", "//pkg/sentry/limits", "//pkg/sentry/memmap", "//pkg/sentry/pgalloc", "//pkg/sentry/platform", - "//pkg/sentry/usermem", "//pkg/syserror", + "//pkg/usermem", ], ) diff --git a/pkg/sentry/mm/README.md b/pkg/sentry/mm/README.md index e1322e373..f4d43d927 100644 --- a/pkg/sentry/mm/README.md +++ b/pkg/sentry/mm/README.md @@ -274,7 +274,7 @@ In the sentry: methods [`platform.AddressSpace.MapFile` and `platform.AddressSpace.Unmap`][platform]. -[memmap]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/memmap/memmap.go -[mm]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/mm/mm.go -[pgalloc]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/pgalloc/pgalloc.go -[platform]: https://github.com/google/gvisor/blob/master/+/master/pkg/sentry/platform/platform.go +[memmap]: https://github.com/google/gvisor/blob/master/pkg/sentry/memmap/memmap.go +[mm]: https://github.com/google/gvisor/blob/master/pkg/sentry/mm/mm.go +[pgalloc]: https://github.com/google/gvisor/blob/master/pkg/sentry/pgalloc/pgalloc.go +[platform]: https://github.com/google/gvisor/blob/master/pkg/sentry/platform/platform.go diff --git a/pkg/sentry/mm/address_space.go b/pkg/sentry/mm/address_space.go index cfebcfd42..5c667117c 100644 --- a/pkg/sentry/mm/address_space.go +++ b/pkg/sentry/mm/address_space.go @@ -18,9 +18,9 @@ import ( "fmt" "sync/atomic" - "gvisor.dev/gvisor/pkg/atomicbitops" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // AddressSpace returns the platform.AddressSpace bound to mm. @@ -39,11 +39,18 @@ func (mm *MemoryManager) AddressSpace() platform.AddressSpace { // // When this MemoryManager is no longer needed by a task, it should call // Deactivate to release the reference. -func (mm *MemoryManager) Activate() error { +func (mm *MemoryManager) Activate(ctx context.Context) error { // Fast path: the MemoryManager already has an active // platform.AddressSpace, and we just need to indicate that we need it too. - if atomicbitops.IncUnlessZeroInt32(&mm.active) { - return nil + for { + active := atomic.LoadInt32(&mm.active) + if active == 0 { + // Fall back to the slow path. + break + } + if atomic.CompareAndSwapInt32(&mm.active, active, active+1) { + return nil + } } for { @@ -85,16 +92,20 @@ func (mm *MemoryManager) Activate() error { if as == nil { // AddressSpace is unavailable, we must wait. // - // activeMu must not be held while waiting, as the user - // of the address space we are waiting on may attempt - // to take activeMu. - // - // Don't call UninterruptibleSleepStart to register the - // wait to allow the watchdog stuck task to trigger in - // case a process is starved waiting for the address - // space. + // activeMu must not be held while waiting, as the user of the address + // space we are waiting on may attempt to take activeMu. mm.activeMu.Unlock() + + sleep := mm.p.CooperativelySchedulesAddressSpace() && mm.sleepForActivation + if sleep { + // Mark this task sleeping while waiting for the address space to + // prevent the watchdog from reporting it as a stuck task. + ctx.UninterruptibleSleepStart(false) + } <-c + if sleep { + ctx.UninterruptibleSleepFinish(false) + } continue } @@ -118,8 +129,15 @@ func (mm *MemoryManager) Activate() error { func (mm *MemoryManager) Deactivate() { // Fast path: this is not the last goroutine to deactivate the // MemoryManager. - if atomicbitops.DecUnlessOneInt32(&mm.active) { - return + for { + active := atomic.LoadInt32(&mm.active) + if active == 1 { + // Fall back to the slow path. + break + } + if atomic.CompareAndSwapInt32(&mm.active, active, active-1) { + return + } } mm.activeMu.Lock() @@ -183,8 +201,10 @@ func (mm *MemoryManager) mapASLocked(pseg pmaIterator, ar usermem.AddrRange, pre if pma.needCOW { perms.Write = false } - if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { - return err + if perms.Any() { // MapFile precondition + if err := mm.as.MapFile(pmaMapAR.Start, pma.file, pseg.fileRangeOf(pmaMapAR), perms, precommit); err != nil { + return err + } } pseg = pseg.NextSegment() } diff --git a/pkg/sentry/mm/aio_context.go b/pkg/sentry/mm/aio_context.go index 1b746d030..379148903 100644 --- a/pkg/sentry/mm/aio_context.go +++ b/pkg/sentry/mm/aio_context.go @@ -15,17 +15,16 @@ package mm import ( - "sync" - "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "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/sentry/usermem" + "gvisor.dev/gvisor/pkg/sync" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // aioManager creates and manages asynchronous I/O contexts. @@ -60,25 +59,27 @@ func (a *aioManager) newAIOContext(events uint32, id uint64) bool { } a.contexts[id] = &AIOContext{ - done: make(chan struct{}, 1), + requestReady: make(chan struct{}, 1), maxOutstanding: events, } return true } -// destroyAIOContext destroys an asynchronous I/O context. +// destroyAIOContext destroys an asynchronous I/O context. It doesn't wait for +// for pending requests to complete. Returns the destroyed AIOContext so it can +// be drained. // -// False is returned if the context does not exist. -func (a *aioManager) destroyAIOContext(id uint64) bool { +// Nil is returned if the context does not exist. +func (a *aioManager) destroyAIOContext(id uint64) *AIOContext { a.mu.Lock() defer a.mu.Unlock() ctx, ok := a.contexts[id] if !ok { - return false + return nil } delete(a.contexts, id) ctx.destroy() - return true + return ctx } // lookupAIOContext looks up the given context. @@ -103,8 +104,8 @@ type ioResult struct { // // +stateify savable type AIOContext struct { - // done is the notification channel used for all requests. - done chan struct{} `state:"nosave"` + // requestReady is the notification channel used for all requests. + requestReady chan struct{} `state:"nosave"` // mu protects below. mu sync.Mutex `state:"nosave"` @@ -130,8 +131,14 @@ func (ctx *AIOContext) destroy() { ctx.mu.Lock() defer ctx.mu.Unlock() ctx.dead = true - if ctx.outstanding == 0 { - close(ctx.done) + ctx.checkForDone() +} + +// Preconditions: ctx.mu must be held by caller. +func (ctx *AIOContext) checkForDone() { + if ctx.dead && ctx.outstanding == 0 { + close(ctx.requestReady) + ctx.requestReady = nil } } @@ -155,11 +162,12 @@ func (ctx *AIOContext) PopRequest() (interface{}, bool) { // Is there anything ready? if e := ctx.results.Front(); e != nil { - ctx.results.Remove(e) - ctx.outstanding-- - if ctx.outstanding == 0 && ctx.dead { - close(ctx.done) + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } + ctx.outstanding-- + ctx.results.Remove(e) + ctx.checkForDone() return e.data, true } return nil, false @@ -173,26 +181,58 @@ func (ctx *AIOContext) FinishRequest(data interface{}) { // Push to the list and notify opportunistically. The channel notify // here is guaranteed to be safe because outstanding must be non-zero. - // The done channel is only closed when outstanding reaches zero. + // The requestReady channel is only closed when outstanding reaches zero. ctx.results.PushBack(&ioResult{data: data}) select { - case ctx.done <- struct{}{}: + case ctx.requestReady <- struct{}{}: default: } } // WaitChannel returns a channel that is notified when an AIO request is -// completed. -// -// The boolean return value indicates whether or not the context is active. -func (ctx *AIOContext) WaitChannel() (chan struct{}, bool) { +// completed. Returns nil if the context is destroyed and there are no more +// outstanding requests. +func (ctx *AIOContext) WaitChannel() chan struct{} { ctx.mu.Lock() defer ctx.mu.Unlock() - if ctx.outstanding == 0 && ctx.dead { - return nil, false + return ctx.requestReady +} + +// Dead returns true if the context has been destroyed. +func (ctx *AIOContext) Dead() bool { + ctx.mu.Lock() + defer ctx.mu.Unlock() + return ctx.dead +} + +// CancelPendingRequest forgets about a request that hasn't yet completed. +func (ctx *AIOContext) CancelPendingRequest() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + panic("AIOContext outstanding is going negative") } - return ctx.done, true + ctx.outstanding-- + ctx.checkForDone() +} + +// Drain drops all completed requests. Pending requests remain untouched. +func (ctx *AIOContext) Drain() { + ctx.mu.Lock() + defer ctx.mu.Unlock() + + if ctx.outstanding == 0 { + return + } + size := uint32(ctx.results.Len()) + if ctx.outstanding < size { + panic("AIOContext outstanding is going negative") + } + ctx.outstanding -= size + ctx.results.Reset() + ctx.checkForDone() } // aioMappable implements memmap.MappingIdentity and memmap.Mappable for AIO @@ -333,9 +373,9 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint Length: aioRingBufferSize, MappingIdentity: m, Mappable: m, - // TODO(fvoznika): Linux does "do_mmap_pgoff(..., PROT_READ | - // PROT_WRITE, ...)" in fs/aio.c:aio_setup_ring(); why do we make this - // mapping read-only? + // Linux uses "do_mmap_pgoff(..., PROT_READ | PROT_WRITE, ...)" in + // fs/aio.c:aio_setup_ring(). Since we don't implement AIO_RING_MAGIC, + // user mode should not write to this page. Perms: usermem.Read, MaxPerms: usermem.Read, }) @@ -350,11 +390,11 @@ func (mm *MemoryManager) NewAIOContext(ctx context.Context, events uint32) (uint return id, nil } -// DestroyAIOContext destroys an asynchronous I/O context. It returns false if -// the context does not exist. -func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) bool { +// DestroyAIOContext destroys an asynchronous I/O context. It returns the +// destroyed context. nil if the context does not exist. +func (mm *MemoryManager) DestroyAIOContext(ctx context.Context, id uint64) *AIOContext { if _, ok := mm.LookupAIOContext(ctx, id); !ok { - return false + return nil } // Only unmaps after it assured that the address is a valid aio context to diff --git a/pkg/sentry/mm/aio_context_state.go b/pkg/sentry/mm/aio_context_state.go index c37fc9f7b..3dabac1af 100644 --- a/pkg/sentry/mm/aio_context_state.go +++ b/pkg/sentry/mm/aio_context_state.go @@ -16,5 +16,5 @@ package mm // afterLoad is invoked by stateify. func (a *AIOContext) afterLoad() { - a.done = make(chan struct{}, 1) + a.requestReady = make(chan struct{}, 1) } diff --git a/pkg/sentry/mm/debug.go b/pkg/sentry/mm/debug.go index df9adf708..c273c982e 100644 --- a/pkg/sentry/mm/debug.go +++ b/pkg/sentry/mm/debug.go @@ -18,7 +18,7 @@ import ( "bytes" "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) const ( diff --git a/pkg/sentry/mm/io.go b/pkg/sentry/mm/io.go index b03e7d020..fa776f9c6 100644 --- a/pkg/sentry/mm/io.go +++ b/pkg/sentry/mm/io.go @@ -15,11 +15,11 @@ package mm import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // There are two supported ways to copy data to/from application virtual diff --git a/pkg/sentry/mm/lifecycle.go b/pkg/sentry/mm/lifecycle.go index 4e9ca1de6..aac56679b 100644 --- a/pkg/sentry/mm/lifecycle.go +++ b/pkg/sentry/mm/lifecycle.go @@ -18,27 +18,27 @@ import ( "fmt" "sync/atomic" - "gvisor.dev/gvisor/pkg/atomicbitops" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) // NewMemoryManager returns a new MemoryManager with no mappings and 1 user. -func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider) *MemoryManager { +func NewMemoryManager(p platform.Platform, mfp pgalloc.MemoryFileProvider, sleepForActivation bool) *MemoryManager { return &MemoryManager{ - p: p, - mfp: mfp, - haveASIO: p.SupportsAddressSpaceIO(), - privateRefs: &privateRefs{}, - users: 1, - auxv: arch.Auxv{}, - dumpability: UserDumpable, - aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + p: p, + mfp: mfp, + haveASIO: p.SupportsAddressSpaceIO(), + privateRefs: &privateRefs{}, + users: 1, + auxv: arch.Auxv{}, + dumpability: UserDumpable, + aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + sleepForActivation: sleepForActivation, } } @@ -80,9 +80,11 @@ func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { envv: mm.envv, auxv: append(arch.Auxv(nil), mm.auxv...), // IncRef'd below, once we know that there isn't an error. - executable: mm.executable, - dumpability: mm.dumpability, - aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + executable: mm.executable, + dumpability: mm.dumpability, + aioManager: aioManager{contexts: make(map[uint64]*AIOContext)}, + sleepForActivation: mm.sleepForActivation, + vdsoSigReturnAddr: mm.vdsoSigReturnAddr, } // Copy vmas. @@ -229,7 +231,15 @@ func (mm *MemoryManager) Fork(ctx context.Context) (*MemoryManager, error) { // IncUsers increments mm's user count and returns true. If the user count is // already 0, IncUsers does nothing and returns false. func (mm *MemoryManager) IncUsers() bool { - return atomicbitops.IncUnlessZeroInt32(&mm.users) + for { + users := atomic.LoadInt32(&mm.users) + if users == 0 { + return false + } + if atomic.CompareAndSwapInt32(&mm.users, users, users+1) { + return true + } + } } // DecUsers decrements mm's user count. If the user count reaches 0, all diff --git a/pkg/sentry/mm/metadata.go b/pkg/sentry/mm/metadata.go index d2a01d48a..28e5057f7 100644 --- a/pkg/sentry/mm/metadata.go +++ b/pkg/sentry/mm/metadata.go @@ -16,8 +16,8 @@ package mm import ( "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" + "gvisor.dev/gvisor/pkg/usermem" ) // Dumpability describes if and how core dumps should be created. @@ -132,7 +132,7 @@ func (mm *MemoryManager) SetAuxv(auxv arch.Auxv) { // // An additional reference will be taken in the case of a non-nil executable, // which must be released by the caller. -func (mm *MemoryManager) Executable() *fs.Dirent { +func (mm *MemoryManager) Executable() fsbridge.File { mm.metadataMu.Lock() defer mm.metadataMu.Unlock() @@ -147,15 +147,15 @@ func (mm *MemoryManager) Executable() *fs.Dirent { // SetExecutable sets the executable. // // This takes a reference on d. -func (mm *MemoryManager) SetExecutable(d *fs.Dirent) { +func (mm *MemoryManager) SetExecutable(file fsbridge.File) { mm.metadataMu.Lock() // Grab a new reference. - d.IncRef() + file.IncRef() // Set the executable. orig := mm.executable - mm.executable = d + mm.executable = file mm.metadataMu.Unlock() @@ -167,3 +167,17 @@ func (mm *MemoryManager) SetExecutable(d *fs.Dirent) { orig.DecRef() } } + +// VDSOSigReturn returns the address of vdso_sigreturn. +func (mm *MemoryManager) VDSOSigReturn() uint64 { + mm.metadataMu.Lock() + defer mm.metadataMu.Unlock() + return mm.vdsoSigReturnAddr +} + +// SetVDSOSigReturn sets the address of vdso_sigreturn. +func (mm *MemoryManager) SetVDSOSigReturn(addr uint64) { + mm.metadataMu.Lock() + defer mm.metadataMu.Unlock() + mm.vdsoSigReturnAddr = addr +} diff --git a/pkg/sentry/mm/mm.go b/pkg/sentry/mm/mm.go index f350e0109..6db7c3d40 100644 --- a/pkg/sentry/mm/mm.go +++ b/pkg/sentry/mm/mm.go @@ -35,16 +35,15 @@ package mm import ( - "sync" - + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/safemem" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/fs" + "gvisor.dev/gvisor/pkg/sentry/fsbridge" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/safemem" - "gvisor.dev/gvisor/pkg/sentry/usermem" - "gvisor.dev/gvisor/third_party/gvsync" + "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/usermem" ) // MemoryManager implements a virtual address space. @@ -82,7 +81,7 @@ type MemoryManager struct { users int32 // mappingMu is analogous to Linux's struct mm_struct::mmap_sem. - mappingMu gvsync.DowngradableRWMutex `state:"nosave"` + mappingMu sync.RWMutex `state:"nosave"` // vmas stores virtual memory areas. Since vmas are stored by value, // clients should usually use vmaIterator.ValuePtr() instead of @@ -125,7 +124,7 @@ type MemoryManager struct { // activeMu is loosely analogous to Linux's struct // mm_struct::page_table_lock. - activeMu gvsync.DowngradableRWMutex `state:"nosave"` + activeMu sync.RWMutex `state:"nosave"` // pmas stores platform mapping areas used to implement vmas. Since pmas // are stored by value, clients should usually use pmaIterator.ValuePtr() @@ -217,7 +216,7 @@ type MemoryManager struct { // is not nil, it holds a reference on the Dirent. // // executable is protected by metadataMu. - executable *fs.Dirent + executable fsbridge.File // dumpability describes if and how this MemoryManager may be dumped to // userspace. @@ -228,6 +227,14 @@ type MemoryManager struct { // aioManager keeps track of AIOContexts used for async IOs. AIOManager // must be cloned when CLONE_VM is used. aioManager aioManager + + // sleepForActivation indicates whether the task should report to be sleeping + // before trying to activate the address space. When set to true, delays in + // activation are not reported as stuck tasks by the watchdog. + sleepForActivation bool + + // vdsoSigReturnAddr is the address of 'vdso_sigreturn'. + vdsoSigReturnAddr uint64 } // vma represents a virtual memory area. @@ -280,7 +287,7 @@ type vma struct { mlockMode memmap.MLockMode // numaPolicy is the NUMA policy for this vma set by mbind(). - numaPolicy int32 + numaPolicy linux.NumaPolicy // numaNodemask is the NUMA nodemask for this vma set by mbind(). numaNodemask uint64 diff --git a/pkg/sentry/mm/mm_test.go b/pkg/sentry/mm/mm_test.go index 4d2bfaaed..fdc308542 100644 --- a/pkg/sentry/mm/mm_test.go +++ b/pkg/sentry/mm/mm_test.go @@ -17,21 +17,21 @@ package mm import ( "testing" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" - "gvisor.dev/gvisor/pkg/sentry/context/contexttest" + "gvisor.dev/gvisor/pkg/sentry/contexttest" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" "gvisor.dev/gvisor/pkg/sentry/pgalloc" "gvisor.dev/gvisor/pkg/sentry/platform" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) func testMemoryManager(ctx context.Context) *MemoryManager { p := platform.FromContext(ctx) mfp := pgalloc.MemoryFileProviderFromContext(ctx) - mm := NewMemoryManager(p, mfp) + mm := NewMemoryManager(p, mfp, false) mm.layout = arch.MmapLayout{ MinAddr: p.MinUserAddress(), MaxAddr: p.MaxUserAddress(), diff --git a/pkg/sentry/mm/pma.go b/pkg/sentry/mm/pma.go index c976c6f45..62e4c20af 100644 --- a/pkg/sentry/mm/pma.go +++ b/pkg/sentry/mm/pma.go @@ -17,14 +17,14 @@ package mm import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" + "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/platform/safecopy" - "gvisor.dev/gvisor/pkg/sentry/safemem" "gvisor.dev/gvisor/pkg/sentry/usage" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // existingPMAsLocked checks that pmas exist for all addresses in ar, and diff --git a/pkg/sentry/mm/procfs.go b/pkg/sentry/mm/procfs.go index 8c2246bb4..6efe5102b 100644 --- a/pkg/sentry/mm/procfs.go +++ b/pkg/sentry/mm/procfs.go @@ -19,10 +19,10 @@ import ( "fmt" "strings" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/fs/proc/seqfile" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" + "gvisor.dev/gvisor/pkg/usermem" ) const ( @@ -66,8 +66,6 @@ func (mm *MemoryManager) ReadMapsDataInto(ctx context.Context, buf *bytes.Buffer var start usermem.Addr for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". mm.appendVMAMapsEntryLocked(ctx, vseg, buf) } @@ -81,7 +79,6 @@ func (mm *MemoryManager) ReadMapsDataInto(ctx context.Context, buf *bytes.Buffer // // Artifically adjust the seqfile handle so we only output vsyscall entry once. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. buf.WriteString(vsyscallMapsEntry) } } @@ -97,8 +94,6 @@ func (mm *MemoryManager) ReadMapsSeqFileData(ctx context.Context, handle seqfile start = *handle.(*usermem.Addr) } for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". vmaAddr := vseg.End() data = append(data, seqfile.SeqData{ Buf: mm.vmaMapsEntryLocked(ctx, vseg), @@ -116,7 +111,6 @@ func (mm *MemoryManager) ReadMapsSeqFileData(ctx context.Context, handle seqfile // // Artifically adjust the seqfile handle so we only output vsyscall entry once. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. vmaAddr := vsyscallEnd data = append(data, seqfile.SeqData{ Buf: []byte(vsyscallMapsEntry), @@ -154,7 +148,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI // Do not include the guard page: fs/proc/task_mmu.c:show_map_vma() => // stack_guard_page_start(). - fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", + lineLen, _ := fmt.Fprintf(b, "%08x-%08x %s%s %08x %02x:%02x %d ", vseg.Start(), vseg.End(), vma.realPerms, private, vma.off, devMajor, devMinor, ino) // Figure out our filename or hint. @@ -171,7 +165,7 @@ func (mm *MemoryManager) appendVMAMapsEntryLocked(ctx context.Context, vseg vmaI } if s != "" { // Per linux, we pad until the 74th character. - if pad := 73 - b.Len(); pad > 0 { + if pad := 73 - lineLen; pad > 0 { b.WriteString(strings.Repeat(" ", pad)) } b.WriteString(s) @@ -187,15 +181,12 @@ func (mm *MemoryManager) ReadSmapsDataInto(ctx context.Context, buf *bytes.Buffe var start usermem.Addr for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". mm.vmaSmapsEntryIntoLocked(ctx, vseg, buf) } // We always emulate vsyscall, so advertise it here. See // ReadMapsSeqFileData for additional commentary. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. buf.WriteString(vsyscallSmapsEntry) } } @@ -211,8 +202,6 @@ func (mm *MemoryManager) ReadSmapsSeqFileData(ctx context.Context, handle seqfil start = *handle.(*usermem.Addr) } for vseg := mm.vmas.LowerBoundSegment(start); vseg.Ok(); vseg = vseg.NextSegment() { - // FIXME(b/30793614): If we use a usermem.Addr for the handle, we get - // "panic: autosave error: type usermem.Addr is not registered". vmaAddr := vseg.End() data = append(data, seqfile.SeqData{ Buf: mm.vmaSmapsEntryLocked(ctx, vseg), @@ -223,7 +212,6 @@ func (mm *MemoryManager) ReadSmapsSeqFileData(ctx context.Context, handle seqfil // We always emulate vsyscall, so advertise it here. See // ReadMapsSeqFileData for additional commentary. if start != vsyscallEnd { - // FIXME(b/30793614): Can't get a pointer to constant vsyscallEnd. vmaAddr := vsyscallEnd data = append(data, seqfile.SeqData{ Buf: []byte(vsyscallSmapsEntry), diff --git a/pkg/sentry/mm/save_restore.go b/pkg/sentry/mm/save_restore.go index 93259c5a3..f56215d9a 100644 --- a/pkg/sentry/mm/save_restore.go +++ b/pkg/sentry/mm/save_restore.go @@ -17,7 +17,7 @@ package mm import ( "fmt" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" ) // InvalidateUnsavable invokes memmap.Mappable.InvalidateUnsavable on all diff --git a/pkg/sentry/mm/shm.go b/pkg/sentry/mm/shm.go index b9f2d23e5..6432731d4 100644 --- a/pkg/sentry/mm/shm.go +++ b/pkg/sentry/mm/shm.go @@ -15,10 +15,10 @@ package mm import ( - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/shm" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // DetachShm unmaps a sysv shared memory segment. diff --git a/pkg/sentry/mm/special_mappable.go b/pkg/sentry/mm/special_mappable.go index ea2d7af74..9ad52082d 100644 --- a/pkg/sentry/mm/special_mappable.go +++ b/pkg/sentry/mm/special_mappable.go @@ -15,14 +15,14 @@ package mm import ( + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/refs" - "gvisor.dev/gvisor/pkg/sentry/context" "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/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // SpecialMappable implements memmap.MappingIdentity and memmap.Mappable with diff --git a/pkg/sentry/mm/syscalls.go b/pkg/sentry/mm/syscalls.go index c2466c988..3f496aa9f 100644 --- a/pkg/sentry/mm/syscalls.go +++ b/pkg/sentry/mm/syscalls.go @@ -19,14 +19,14 @@ import ( mrand "math/rand" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/sentry/context" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "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/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // HandleUserFault handles an application page fault. sp is the faulting @@ -974,7 +974,7 @@ func (mm *MemoryManager) MLockAll(ctx context.Context, opts MLockAllOpts) error } // NumaPolicy implements the semantics of Linux's get_mempolicy(MPOL_F_ADDR). -func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { +func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (linux.NumaPolicy, uint64, error) { mm.mappingMu.RLock() defer mm.mappingMu.RUnlock() vseg := mm.vmas.FindSegment(addr) @@ -986,7 +986,7 @@ func (mm *MemoryManager) NumaPolicy(addr usermem.Addr) (int32, uint64, error) { } // SetNumaPolicy implements the semantics of Linux's mbind(). -func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy int32, nodemask uint64) error { +func (mm *MemoryManager) SetNumaPolicy(addr usermem.Addr, length uint64, policy linux.NumaPolicy, nodemask uint64) error { if !addr.IsPageAligned() { return syserror.EINVAL } diff --git a/pkg/sentry/mm/vma.go b/pkg/sentry/mm/vma.go index f2fd70799..9a14e69e6 100644 --- a/pkg/sentry/mm/vma.go +++ b/pkg/sentry/mm/vma.go @@ -18,13 +18,13 @@ import ( "fmt" "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/arch" - "gvisor.dev/gvisor/pkg/sentry/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/limits" "gvisor.dev/gvisor/pkg/sentry/memmap" - "gvisor.dev/gvisor/pkg/sentry/usermem" "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" ) // Preconditions: mm.mappingMu must be locked for writing. opts must be valid |