summaryrefslogtreecommitdiffhomepage
path: root/pkg/sentry
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/sentry')
-rw-r--r--pkg/sentry/fsimpl/fuse/BUILD4
-rw-r--r--pkg/sentry/fsimpl/fuse/connection.go234
-rw-r--r--pkg/sentry/fsimpl/fuse/dev.go13
-rw-r--r--pkg/sentry/fsimpl/fuse/dev_test.go17
-rw-r--r--pkg/sentry/fsimpl/fuse/fusefs.go22
-rw-r--r--pkg/sentry/fsimpl/fuse/init.go166
-rw-r--r--pkg/sentry/platform/kvm/kvm_const_arm64.go11
-rw-r--r--pkg/sentry/platform/kvm/machine_arm64.go72
-rw-r--r--pkg/sentry/platform/kvm/testutil/testutil_arm64.go4
-rw-r--r--pkg/sentry/platform/kvm/testutil/testutil_arm64.s14
-rw-r--r--pkg/sentry/syscalls/linux/sys_file.go8
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/fd.go4
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/filesystem.go50
-rw-r--r--pkg/sentry/syscalls/linux/vfs2/setstat.go50
-rw-r--r--pkg/sentry/vfs/epoll.go4
-rw-r--r--pkg/sentry/vfs/file_description.go7
16 files changed, 524 insertions, 156 deletions
diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD
index 67649e811..999111deb 100644
--- a/pkg/sentry/fsimpl/fuse/BUILD
+++ b/pkg/sentry/fsimpl/fuse/BUILD
@@ -21,6 +21,7 @@ go_library(
"connection.go",
"dev.go",
"fusefs.go",
+ "init.go",
"register.go",
"request_list.go",
],
@@ -44,14 +45,13 @@ go_library(
)
go_test(
- name = "dev_test",
+ name = "fuse_test",
size = "small",
srcs = ["dev_test.go"],
library = ":fuse",
deps = [
"//pkg/abi/linux",
"//pkg/sentry/fsimpl/testutil",
- "//pkg/sentry/fsimpl/tmpfs",
"//pkg/sentry/kernel",
"//pkg/sentry/kernel/auth",
"//pkg/sentry/vfs",
diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go
index f330da0bd..6df2728ab 100644
--- a/pkg/sentry/fsimpl/fuse/connection.go
+++ b/pkg/sentry/fsimpl/fuse/connection.go
@@ -17,6 +17,8 @@ package fuse
import (
"errors"
"fmt"
+ "sync"
+ "sync/atomic"
"syscall"
"gvisor.dev/gvisor/pkg/abi/linux"
@@ -25,18 +27,29 @@ import (
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/vfs"
+ "gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/waiter"
"gvisor.dev/gvisor/tools/go_marshal/marshal"
)
-// MaxActiveRequestsDefault is the default setting controlling the upper bound
+// maxActiveRequestsDefault is the default setting controlling the upper bound
// on the number of active requests at any given time.
-const MaxActiveRequestsDefault = 10000
+const maxActiveRequestsDefault = 10000
-var (
- // Ordinary requests have even IDs, while interrupts IDs are odd.
- InitReqBit uint64 = 1
- ReqIDStep uint64 = 2
+// Ordinary requests have even IDs, while interrupts IDs are odd.
+// Used to increment the unique ID for each FUSE request.
+var reqIDStep uint64 = 2
+
+const (
+ // fuseDefaultMaxBackground is the default value for MaxBackground.
+ fuseDefaultMaxBackground = 12
+
+ // fuseDefaultCongestionThreshold is the default value for CongestionThreshold,
+ // and is 75% of the default maximum of MaxGround.
+ fuseDefaultCongestionThreshold = (fuseDefaultMaxBackground * 3 / 4)
+
+ // fuseDefaultMaxPagesPerReq is the default value for MaxPagesPerReq.
+ fuseDefaultMaxPagesPerReq = 32
)
// Request represents a FUSE operation request that hasn't been sent to the
@@ -61,17 +74,125 @@ type Response struct {
data []byte
}
-// Connection is the struct by which the sentry communicates with the FUSE server daemon.
-type Connection struct {
+// connection is the struct by which the sentry communicates with the FUSE server daemon.
+type connection struct {
fd *DeviceFD
- // MaxWrite is the daemon's maximum size of a write buffer.
- // This is negotiated during FUSE_INIT.
- MaxWrite uint32
+ // The following FUSE_INIT flags are currently unsupported by this implementation:
+ // - FUSE_ATOMIC_O_TRUNC: requires open(..., O_TRUNC)
+ // - FUSE_EXPORT_SUPPORT
+ // - FUSE_HANDLE_KILLPRIV
+ // - FUSE_POSIX_LOCKS: requires POSIX locks
+ // - FUSE_FLOCK_LOCKS: requires POSIX locks
+ // - FUSE_AUTO_INVAL_DATA: requires page caching eviction
+ // - FUSE_EXPLICIT_INVAL_DATA: requires page caching eviction
+ // - FUSE_DO_READDIRPLUS/FUSE_READDIRPLUS_AUTO: requires FUSE_READDIRPLUS implementation
+ // - FUSE_ASYNC_DIO
+ // - FUSE_POSIX_ACL: affects defaultPermissions, posixACL, xattr handler
+
+ // initialized after receiving FUSE_INIT reply.
+ // Until it's set, suspend sending FUSE requests.
+ // Use SetInitialized() and IsInitialized() for atomic access.
+ initialized int32
+
+ // initializedChan is used to block requests before initialization.
+ initializedChan chan struct{}
+
+ // blocked when there are too many outstading backgrounds requests (NumBackground == MaxBackground).
+ // TODO(gvisor.dev/issue/3185): update the numBackground accordingly; use a channel to block.
+ blocked bool
+
+ // connected (connection established) when a new FUSE file system is created.
+ // Set to false when:
+ // umount,
+ // connection abort,
+ // device release.
+ connected bool
+
+ // aborted via sysfs.
+ // TODO(gvisor.dev/issue/3185): abort all queued requests.
+ aborted bool
+
+ // connInitError if FUSE_INIT encountered error (major version mismatch).
+ // Only set in INIT.
+ connInitError bool
+
+ // connInitSuccess if FUSE_INIT is successful.
+ // Only set in INIT.
+ // Used for destory.
+ connInitSuccess bool
+
+ // TODO(gvisor.dev/issue/3185): All the queue logic are working in progress.
+
+ // NumberBackground is the number of requests in the background.
+ numBackground uint16
+
+ // congestionThreshold for NumBackground.
+ // Negotiated in FUSE_INIT.
+ congestionThreshold uint16
+
+ // maxBackground is the maximum number of NumBackground.
+ // Block connection when it is reached.
+ // Negotiated in FUSE_INIT.
+ maxBackground uint16
+
+ // numActiveBackground is the number of requests in background and has being marked as active.
+ numActiveBackground uint16
+
+ // numWating is the number of requests waiting for completion.
+ numWaiting uint32
+
+ // TODO(gvisor.dev/issue/3185): BgQueue
+ // some queue for background queued requests.
+
+ // bgLock protects:
+ // MaxBackground, CongestionThreshold, NumBackground,
+ // NumActiveBackground, BgQueue, Blocked.
+ bgLock sync.Mutex
+
+ // maxRead is the maximum size of a read buffer in in bytes.
+ maxRead uint32
+
+ // maxWrite is the maximum size of a write buffer in bytes.
+ // Negotiated in FUSE_INIT.
+ maxWrite uint32
+
+ // maxPages is the maximum number of pages for a single request to use.
+ // Negotiated in FUSE_INIT.
+ maxPages uint16
+
+ // minor version of the FUSE protocol.
+ // Negotiated and only set in INIT.
+ minor uint32
+
+ // asyncRead if read pages asynchronously.
+ // Negotiated and only set in INIT.
+ asyncRead bool
+
+ // abortErr is true if kernel need to return an unique read error after abort.
+ // Negotiated and only set in INIT.
+ abortErr bool
+
+ // writebackCache is true for write-back cache policy,
+ // false for write-through policy.
+ // Negotiated and only set in INIT.
+ writebackCache bool
+
+ // cacheSymlinks if filesystem needs to cache READLINK responses in page cache.
+ // Negotiated and only set in INIT.
+ cacheSymlinks bool
+
+ // bigWrites if doing multi-page cached writes.
+ // Negotiated and only set in INIT.
+ bigWrites bool
+
+ // dontMask if filestestem does not apply umask to creation modes.
+ // Negotiated in INIT.
+ dontMask bool
}
-// NewFUSEConnection creates a FUSE connection to fd
-func NewFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*Connection, error) {
+// newFUSEConnection creates a FUSE connection to fd.
+func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*connection, error) {
// Mark the device as ready so it can be used. /dev/fuse can only be used if the FD was used to
// mount a FUSE filesystem.
fuseFD := fd.Impl().(*DeviceFD)
@@ -84,16 +205,41 @@ func NewFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRe
fuseFD.fullQueueCh = make(chan struct{}, maxInFlightRequests)
fuseFD.writeCursor = 0
- return &Connection{
- fd: fuseFD,
+ return &connection{
+ fd: fuseFD,
+ maxBackground: fuseDefaultMaxBackground,
+ congestionThreshold: fuseDefaultCongestionThreshold,
+ maxPages: fuseDefaultMaxPagesPerReq,
+ initializedChan: make(chan struct{}),
+ connected: true,
}, nil
}
+// SetInitialized atomically sets the connection as initialized.
+func (conn *connection) SetInitialized() {
+ // Unblock the requests sent before INIT.
+ close(conn.initializedChan)
+
+ // Close the channel first to avoid the non-atomic situation
+ // where conn.initialized is true but there are
+ // tasks being blocked on the channel.
+ // And it prevents the newer tasks from gaining
+ // unnecessary higher chance to be issued before the blocked one.
+
+ atomic.StoreInt32(&(conn.initialized), int32(1))
+}
+
+// IsInitialized atomically check if the connection is initialized.
+// pairs with SetInitialized().
+func (conn *connection) Initialized() bool {
+ return atomic.LoadInt32(&(conn.initialized)) != 0
+}
+
// NewRequest creates a new request that can be sent to the FUSE server.
-func (conn *Connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint64, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*Request, error) {
+func (conn *connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint64, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*Request, error) {
conn.fd.mu.Lock()
defer conn.fd.mu.Unlock()
- conn.fd.nextOpID += linux.FUSEOpID(ReqIDStep)
+ conn.fd.nextOpID += linux.FUSEOpID(reqIDStep)
hdrLen := (*linux.FUSEHeaderIn)(nil).SizeBytes()
hdr := linux.FUSEHeaderIn{
@@ -118,13 +264,49 @@ func (conn *Connection) NewRequest(creds *auth.Credentials, pid uint32, ino uint
}
// Call makes a request to the server and blocks the invoking task until a
-// server responds with a response.
-// NOTE: If no task is provided then the Call will simply enqueue the request
-// and return a nil response. No blocking will happen in this case. Instead,
-// this is used to signify that the processing of this request will happen by
-// the kernel.Task that writes the response. See FUSE_INIT for such an
-// invocation.
-func (conn *Connection) Call(t *kernel.Task, r *Request) (*Response, error) {
+// server responds with a response. Task should never be nil.
+// Requests will not be sent before the connection is initialized.
+// For async tasks, use CallAsync().
+func (conn *connection) Call(t *kernel.Task, r *Request) (*Response, error) {
+ // Block requests sent before connection is initalized.
+ if !conn.Initialized() {
+ if err := t.Block(conn.initializedChan); err != nil {
+ return nil, err
+ }
+ }
+
+ return conn.call(t, r)
+}
+
+// CallAsync makes an async (aka background) request.
+// Those requests either do not expect a response (e.g. release) or
+// the response should be handled by others (e.g. init).
+// Return immediately unless the connection is blocked (before initialization).
+// Async call example: init, release, forget, aio, interrupt.
+// When the Request is FUSE_INIT, it will not be blocked before initialization.
+func (conn *connection) CallAsync(t *kernel.Task, r *Request) error {
+ // Block requests sent before connection is initalized.
+ if !conn.Initialized() && r.hdr.Opcode != linux.FUSE_INIT {
+ if err := t.Block(conn.initializedChan); err != nil {
+ return err
+ }
+ }
+
+ // This should be the only place that invokes call() with a nil task.
+ _, err := conn.call(nil, r)
+ return err
+}
+
+// call makes a call without blocking checks.
+func (conn *connection) call(t *kernel.Task, r *Request) (*Response, error) {
+ if !conn.connected {
+ return nil, syserror.ENOTCONN
+ }
+
+ if conn.connInitError {
+ return nil, syserror.ECONNREFUSED
+ }
+
fut, err := conn.callFuture(t, r)
if err != nil {
return nil, err
@@ -160,7 +342,7 @@ func (r *Response) UnmarshalPayload(m marshal.Marshallable) error {
// callFuture makes a request to the server and returns a future response.
// Call resolve() when the response needs to be fulfilled.
-func (conn *Connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, error) {
+func (conn *connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, error) {
conn.fd.mu.Lock()
defer conn.fd.mu.Unlock()
@@ -195,7 +377,7 @@ func (conn *Connection) callFuture(t *kernel.Task, r *Request) (*futureResponse,
}
// callFutureLocked makes a request to the server and returns a future response.
-func (conn *Connection) callFutureLocked(t *kernel.Task, r *Request) (*futureResponse, error) {
+func (conn *connection) callFutureLocked(t *kernel.Task, r *Request) (*futureResponse, error) {
conn.fd.queue.PushBack(r)
conn.fd.numActiveRequests += 1
fut := newFutureResponse(r.hdr.Opcode)
diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go
index f3443ac71..2225076bc 100644
--- a/pkg/sentry/fsimpl/fuse/dev.go
+++ b/pkg/sentry/fsimpl/fuse/dev.go
@@ -21,6 +21,7 @@ import (
"gvisor.dev/gvisor/pkg/context"
"gvisor.dev/gvisor/pkg/log"
"gvisor.dev/gvisor/pkg/sentry/kernel"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/sync"
"gvisor.dev/gvisor/pkg/syserror"
@@ -98,7 +99,9 @@ type DeviceFD struct {
}
// Release implements vfs.FileDescriptionImpl.Release.
-func (fd *DeviceFD) Release() {}
+func (fd *DeviceFD) Release() {
+ fd.fs.conn.connected = false
+}
// PRead implements vfs.FileDescriptionImpl.PRead.
func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) {
@@ -124,7 +127,7 @@ func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.R
minBuffSize := linux.FUSE_MIN_READ_BUFFER
inHdrLen := uint32((*linux.FUSEHeaderIn)(nil).SizeBytes())
writeHdrLen := uint32((*linux.FUSEWriteIn)(nil).SizeBytes())
- negotiatedMinBuffSize := inHdrLen + writeHdrLen + fd.fs.conn.MaxWrite
+ negotiatedMinBuffSize := inHdrLen + writeHdrLen + fd.fs.conn.maxWrite
if minBuffSize < negotiatedMinBuffSize {
minBuffSize = negotiatedMinBuffSize
}
@@ -385,9 +388,9 @@ func (fd *DeviceFD) sendError(ctx context.Context, errno int32, req *Request) er
// FUSE_INIT.
func (fd *DeviceFD) noReceiverAction(ctx context.Context, r *Response) error {
if r.opcode == linux.FUSE_INIT {
- // TODO: process init response here.
- // Maybe get the creds from the context?
- // creds := auth.CredentialsFromContext(ctx)
+ creds := auth.CredentialsFromContext(ctx)
+ rootUserNs := kernel.KernelFromContext(ctx).RootUserNamespace()
+ return fd.fs.conn.InitRecv(r, creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, rootUserNs))
}
return nil
diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go
index fcd77832a..84c222ad6 100644
--- a/pkg/sentry/fsimpl/fuse/dev_test.go
+++ b/pkg/sentry/fsimpl/fuse/dev_test.go
@@ -22,7 +22,6 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil"
- "gvisor.dev/gvisor/pkg/sentry/fsimpl/tmpfs"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
"gvisor.dev/gvisor/pkg/sentry/vfs"
@@ -60,25 +59,25 @@ func TestFUSECommunication(t *testing.T) {
Name: "SingleClientSingleServer",
NumClients: 1,
NumServers: 1,
- MaxActiveRequests: MaxActiveRequestsDefault,
+ MaxActiveRequests: maxActiveRequestsDefault,
},
{
Name: "SingleClientMultipleServers",
NumClients: 1,
NumServers: 10,
- MaxActiveRequests: MaxActiveRequestsDefault,
+ MaxActiveRequests: maxActiveRequestsDefault,
},
{
Name: "MultipleClientsSingleServer",
NumClients: 10,
NumServers: 1,
- MaxActiveRequests: MaxActiveRequestsDefault,
+ MaxActiveRequests: maxActiveRequestsDefault,
},
{
Name: "MultipleClientsMultipleServers",
NumClients: 10,
NumServers: 10,
- MaxActiveRequests: MaxActiveRequestsDefault,
+ MaxActiveRequests: maxActiveRequestsDefault,
},
{
Name: "RequestCapacityFull",
@@ -145,7 +144,7 @@ func TestFUSECommunication(t *testing.T) {
// CallTest makes a request to the server and blocks the invoking
// goroutine until a server responds with a response. Doesn't block
// a kernel.Task. Analogous to Connection.Call but used for testing.
-func CallTest(conn *Connection, t *kernel.Task, r *Request, i uint32) (*Response, error) {
+func CallTest(conn *connection, t *kernel.Task, r *Request, i uint32) (*Response, error) {
conn.fd.mu.Lock()
// Wait until we're certain that a new request can be processed.
@@ -214,7 +213,7 @@ func ReadTest(serverTask *kernel.Task, fd *vfs.FileDescription, inIOseq usermem.
// fuseClientRun emulates all the actions of a normal FUSE request. It creates
// a header, a payload, calls the server, waits for the response, and processes
// the response.
-func fuseClientRun(t *testing.T, s *testutil.System, k *kernel.Kernel, conn *Connection, creds *auth.Credentials, pid uint32, inode uint64, clientDone chan struct{}) {
+func fuseClientRun(t *testing.T, s *testutil.System, k *kernel.Kernel, conn *connection, creds *auth.Credentials, pid uint32, inode uint64, clientDone chan struct{}) {
defer func() { clientDone <- struct{}{} }()
tc := k.NewThreadGroup(nil, k.RootPIDNamespace(), kernel.NewSignalHandlers(), linux.SIGCHLD, k.GlobalInit().Limits())
@@ -343,7 +342,7 @@ func setup(t *testing.T) *testutil.System {
AllowUserMount: true,
})
- mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", tmpfs.Name, &vfs.GetFilesystemOptions{})
+ mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.GetFilesystemOptions{})
if err != nil {
t.Fatalf("NewMountNamespace(): %v", err)
}
@@ -353,7 +352,7 @@ func setup(t *testing.T) *testutil.System {
// newTestConnection creates a fuse connection that the sentry can communicate with
// and the FD for the server to communicate with.
-func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveRequests uint64) (*Connection, *vfs.FileDescription, error) {
+func newTestConnection(system *testutil.System, k *kernel.Kernel, maxActiveRequests uint64) (*connection, *vfs.FileDescription, error) {
vfsObj := &vfs.VirtualFilesystem{}
fuseDev := &DeviceFD{}
diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go
index 911b6f7cb..200a93bbf 100644
--- a/pkg/sentry/fsimpl/fuse/fusefs.go
+++ b/pkg/sentry/fsimpl/fuse/fusefs.go
@@ -65,7 +65,7 @@ type filesystem struct {
// conn is used for communication between the FUSE server
// daemon and the sentry fusefs.
- conn *Connection
+ conn *connection
// opts is the options the fusefs is initialized with.
opts *filesystemOptions
@@ -140,7 +140,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
fsopts.rootMode = rootMode
// Set the maxInFlightRequests option.
- fsopts.maxActiveRequests = MaxActiveRequestsDefault
+ fsopts.maxActiveRequests = maxActiveRequestsDefault
// Check for unparsed options.
if len(mopts) != 0 {
@@ -157,8 +157,12 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
fs.VFSFilesystem().Init(vfsObj, &fsType, fs)
- // TODO: dispatch a FUSE_INIT request to the FUSE daemon server before
- // returning. Mount will not block on this dispatched request.
+ // Send a FUSE_INIT request to the FUSE daemon server before returning.
+ // This call is not blocking.
+ if err := fs.conn.InitSend(creds, uint32(kernelTask.ThreadID())); err != nil {
+ log.Warningf("%s.InitSend: failed with error: %v", fsType.Name(), err)
+ return nil, nil, err
+ }
// root is the fusefs root directory.
root := fs.newInode(creds, fsopts.rootMode)
@@ -173,7 +177,7 @@ func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOpt
opts: opts,
}
- conn, err := NewFUSEConnection(ctx, device, opts.maxActiveRequests)
+ conn, err := newFUSEConnection(ctx, device, opts.maxActiveRequests)
if err != nil {
log.Warningf("fuse.NewFUSEFilesystem: NewFUSEConnection failed with error: %v", err)
return nil, syserror.EINVAL
@@ -192,8 +196,8 @@ func (fs *filesystem) Release() {
fs.Filesystem.Release()
}
-// Inode implements kernfs.Inode.
-type Inode struct {
+// inode implements kernfs.Inode.
+type inode struct {
kernfs.InodeAttrs
kernfs.InodeNoDynamicLookup
kernfs.InodeNotSymlink
@@ -206,7 +210,7 @@ type Inode struct {
}
func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry {
- i := &Inode{}
+ i := &inode{}
i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755)
i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{})
i.dentry.Init(i)
@@ -215,7 +219,7 @@ func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *ke
}
// Open implements kernfs.Inode.Open.
-func (i *Inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
+func (i *inode) Open(ctx context.Context, rp *vfs.ResolvingPath, vfsd *vfs.Dentry, opts vfs.OpenOptions) (*vfs.FileDescription, error) {
fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts)
if err != nil {
return nil, err
diff --git a/pkg/sentry/fsimpl/fuse/init.go b/pkg/sentry/fsimpl/fuse/init.go
new file mode 100644
index 000000000..779c2bd3f
--- /dev/null
+++ b/pkg/sentry/fsimpl/fuse/init.go
@@ -0,0 +1,166 @@
+// Copyright 2020 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package fuse
+
+import (
+ "gvisor.dev/gvisor/pkg/abi/linux"
+ "gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+)
+
+// consts used by FUSE_INIT negotiation.
+const (
+ // fuseMaxMaxPages is the maximum value for MaxPages received in InitOut.
+ // Follow the same behavior as unix fuse implementation.
+ fuseMaxMaxPages = 256
+
+ // Maximum value for the time granularity for file time stamps, 1s.
+ // Follow the same behavior as unix fuse implementation.
+ fuseMaxTimeGranNs = 1000000000
+
+ // Minimum value for MaxWrite.
+ // Follow the same behavior as unix fuse implementation.
+ fuseMinMaxWrite = 4096
+
+ // Temporary default value for max readahead, 128kb.
+ fuseDefaultMaxReadahead = 131072
+
+ // The FUSE_INIT_IN flags sent to the daemon.
+ // TODO(gvisor.dev/issue/3199): complete the flags.
+ fuseDefaultInitFlags = linux.FUSE_MAX_PAGES
+)
+
+// Adjustable maximums for Connection's cogestion control parameters.
+// Used as the upperbound of the config values.
+// Currently we do not support adjustment to them.
+var (
+ MaxUserBackgroundRequest uint16 = fuseDefaultMaxBackground
+ MaxUserCongestionThreshold uint16 = fuseDefaultCongestionThreshold
+)
+
+// InitSend sends a FUSE_INIT request.
+func (conn *connection) InitSend(creds *auth.Credentials, pid uint32) error {
+ in := linux.FUSEInitIn{
+ Major: linux.FUSE_KERNEL_VERSION,
+ Minor: linux.FUSE_KERNEL_MINOR_VERSION,
+ // TODO(gvisor.dev/issue/3196): find appropriate way to calculate this
+ MaxReadahead: fuseDefaultMaxReadahead,
+ Flags: fuseDefaultInitFlags,
+ }
+
+ req, err := conn.NewRequest(creds, pid, 0, linux.FUSE_INIT, &in)
+ if err != nil {
+ return err
+ }
+
+ // Since there is no task to block on and FUSE_INIT is the request
+ // to unblock other requests, use nil.
+ return conn.CallAsync(nil, req)
+}
+
+// InitRecv receives a FUSE_INIT reply and process it.
+func (conn *connection) InitRecv(res *Response, hasSysAdminCap bool) error {
+ if err := res.Error(); err != nil {
+ return err
+ }
+
+ var out linux.FUSEInitOut
+ if err := res.UnmarshalPayload(&out); err != nil {
+ return err
+ }
+
+ return conn.initProcessReply(&out, hasSysAdminCap)
+}
+
+// Process the FUSE_INIT reply from the FUSE server.
+func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap bool) error {
+ // No support for old major fuse versions.
+ if out.Major != linux.FUSE_KERNEL_VERSION {
+ conn.connInitError = true
+
+ // Set the connection as initialized and unblock the blocked requests
+ // (i.e. return error for them).
+ conn.SetInitialized()
+
+ return nil
+ }
+
+ // Start processing the reply.
+ conn.connInitSuccess = true
+ conn.minor = out.Minor
+
+ // No support for limits before minor version 13.
+ if out.Minor >= 13 {
+ conn.bgLock.Lock()
+
+ if out.MaxBackground > 0 {
+ conn.maxBackground = out.MaxBackground
+
+ if !hasSysAdminCap &&
+ conn.maxBackground > MaxUserBackgroundRequest {
+ conn.maxBackground = MaxUserBackgroundRequest
+ }
+ }
+
+ if out.CongestionThreshold > 0 {
+ conn.congestionThreshold = out.CongestionThreshold
+
+ if !hasSysAdminCap &&
+ conn.congestionThreshold > MaxUserCongestionThreshold {
+ conn.congestionThreshold = MaxUserCongestionThreshold
+ }
+ }
+
+ conn.bgLock.Unlock()
+ }
+
+ // No support for the following flags before minor version 6.
+ if out.Minor >= 6 {
+ conn.asyncRead = out.Flags&linux.FUSE_ASYNC_READ != 0
+ conn.bigWrites = out.Flags&linux.FUSE_BIG_WRITES != 0
+ conn.dontMask = out.Flags&linux.FUSE_DONT_MASK != 0
+ conn.writebackCache = out.Flags&linux.FUSE_WRITEBACK_CACHE != 0
+ conn.cacheSymlinks = out.Flags&linux.FUSE_CACHE_SYMLINKS != 0
+ conn.abortErr = out.Flags&linux.FUSE_ABORT_ERROR != 0
+
+ // TODO(gvisor.dev/issue/3195): figure out how to use TimeGran (0 < TimeGran <= fuseMaxTimeGranNs).
+
+ if out.Flags&linux.FUSE_MAX_PAGES != 0 {
+ maxPages := out.MaxPages
+ if maxPages < 1 {
+ maxPages = 1
+ }
+ if maxPages > fuseMaxMaxPages {
+ maxPages = fuseMaxMaxPages
+ }
+ conn.maxPages = maxPages
+ }
+ }
+
+ // No support for negotiating MaxWrite before minor version 5.
+ if out.Minor >= 5 {
+ conn.maxWrite = out.MaxWrite
+ } else {
+ conn.maxWrite = fuseMinMaxWrite
+ }
+ if conn.maxWrite < fuseMinMaxWrite {
+ conn.maxWrite = fuseMinMaxWrite
+ }
+
+ // Set connection as initialized and unblock the requests
+ // issued before init.
+ conn.SetInitialized()
+
+ return nil
+}
diff --git a/pkg/sentry/platform/kvm/kvm_const_arm64.go b/pkg/sentry/platform/kvm/kvm_const_arm64.go
index 6f0539c29..fdc599477 100644
--- a/pkg/sentry/platform/kvm/kvm_const_arm64.go
+++ b/pkg/sentry/platform/kvm/kvm_const_arm64.go
@@ -116,6 +116,17 @@ const (
// Arm64: Exception Syndrome Register EL1.
const (
+ _ESR_ELx_EC_SHIFT = 26
+ _ESR_ELx_EC_MASK = 0x3F << _ESR_ELx_EC_SHIFT
+
+ _ESR_ELx_EC_IMP_DEF = 0x1f
+ _ESR_ELx_EC_IABT_LOW = 0x20
+ _ESR_ELx_EC_IABT_CUR = 0x21
+ _ESR_ELx_EC_PC_ALIGN = 0x22
+
+ _ESR_ELx_CM = 1 << 8
+ _ESR_ELx_WNR = 1 << 6
+
_ESR_ELx_FSC = 0x3F
_ESR_SEGV_MAPERR_L0 = 0x4
diff --git a/pkg/sentry/platform/kvm/machine_arm64.go b/pkg/sentry/platform/kvm/machine_arm64.go
index f3bf973de..9db171af9 100644
--- a/pkg/sentry/platform/kvm/machine_arm64.go
+++ b/pkg/sentry/platform/kvm/machine_arm64.go
@@ -125,71 +125,59 @@ func nonCanonical(addr uint64, signal int32, info *arch.SignalInfo) (usermem.Acc
return usermem.NoAccess, platform.ErrContextSignal
}
+// isInstructionAbort returns true if it is an instruction abort.
+//
+//go:nosplit
+func isInstructionAbort(code uint64) bool {
+ value := (code & _ESR_ELx_EC_MASK) >> _ESR_ELx_EC_SHIFT
+ return value == _ESR_ELx_EC_IABT_LOW
+}
+
+// isWriteFault returns whether it is a write fault.
+//
+//go:nosplit
+func isWriteFault(code uint64) bool {
+ if isInstructionAbort(code) {
+ return false
+ }
+
+ return (code & _ESR_ELx_WNR) != 0
+}
+
// fault generates an appropriate fault return.
//
//go:nosplit
func (c *vCPU) fault(signal int32, info *arch.SignalInfo) (usermem.AccessType, error) {
+ bluepill(c) // Probably no-op, but may not be.
faultAddr := c.GetFaultAddr()
code, user := c.ErrorCode()
+ if !user {
+ // The last fault serviced by this CPU was not a user
+ // fault, so we can't reliably trust the faultAddr or
+ // the code provided here. We need to re-execute.
+ return usermem.NoAccess, platform.ErrContextInterrupt
+ }
+
// Reset the pointed SignalInfo.
*info = arch.SignalInfo{Signo: signal}
info.SetAddr(uint64(faultAddr))
- read := true
- write := false
- execute := true
-
ret := code & _ESR_ELx_FSC
switch ret {
case _ESR_SEGV_MAPERR_L0, _ESR_SEGV_MAPERR_L1, _ESR_SEGV_MAPERR_L2, _ESR_SEGV_MAPERR_L3:
info.Code = 1 //SEGV_MAPERR
- read = false
- write = true
- execute = false
case _ESR_SEGV_ACCERR_L1, _ESR_SEGV_ACCERR_L2, _ESR_SEGV_ACCERR_L3, _ESR_SEGV_PEMERR_L1, _ESR_SEGV_PEMERR_L2, _ESR_SEGV_PEMERR_L3:
info.Code = 2 // SEGV_ACCERR.
- read = true
- write = false
- execute = false
default:
info.Code = 2
}
- if !user {
- read = true
- write = false
- execute = true
-
- }
accessType := usermem.AccessType{
- Read: read,
- Write: write,
- Execute: execute,
+ Read: !isWriteFault(uint64(code)),
+ Write: isWriteFault(uint64(code)),
+ Execute: isInstructionAbort(uint64(code)),
}
return accessType, platform.ErrContextSignal
}
-
-// retryInGuest runs the given function in guest mode.
-//
-// If the function does not complete in guest mode (due to execution of a
-// system call due to a GC stall, for example), then it will be retried. The
-// given function must be idempotent as a result of the retry mechanism.
-func (m *machine) retryInGuest(fn func()) {
- c := m.Get()
- defer m.Put(c)
- for {
- c.ClearErrorCode() // See below.
- bluepill(c) // Force guest mode.
- fn() // Execute the given function.
- _, user := c.ErrorCode()
- if user {
- // If user is set, then we haven't bailed back to host
- // mode via a kernel exception or system call. We
- // consider the full function to have executed in guest
- // mode and we can return.
- break
- }
- }
-}
diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go
index ca902c8c1..4dad877ba 100644
--- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.go
+++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.go
@@ -56,5 +56,9 @@ func CheckTestRegs(regs *arch.Registers, full bool) (err error) {
err = addRegisterMismatch(err, fmt.Sprintf("R%d", i), regs.Regs[i], need)
}
}
+ // Check tls.
+ if need := ^uint64(11); regs.TPIDR_EL0 != need {
+ err = addRegisterMismatch(err, "tpdir_el0", regs.TPIDR_EL0, need)
+ }
return
}
diff --git a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s
index 07658144e..6caf7282d 100644
--- a/pkg/sentry/platform/kvm/testutil/testutil_arm64.s
+++ b/pkg/sentry/platform/kvm/testutil/testutil_arm64.s
@@ -52,6 +52,8 @@ start:
TEXT ·FloatingPointWorks(SB),NOSPLIT,$0-8
NO_LOCAL_POINTERS
+ // gc will touch fpsimd, so we should test it.
+ // such as in <runtime.deductSweepCredit>.
FMOVD $(9.9), F0
MOVD $SYS_GETPID, R8 // getpid
SVC
@@ -102,11 +104,15 @@ isNaN:
TEXT ·TwiddleRegsSyscall(SB),NOSPLIT,$0
TWIDDLE_REGS()
+ MSR R10, TPIDR_EL0
+ // Trapped in el0_svc.
SVC
RET // never reached
TEXT ·TwiddleRegsFault(SB),NOSPLIT,$0
- TWIDDLE_REGS()
- // Branch to Register branches unconditionally to an address in <Rn>.
- JMP (R4) // <=> br x4, must fault
- RET // never reached
+ TWIDDLE_REGS()
+ MSR R10, TPIDR_EL0
+ // Trapped in el0_ia.
+ // Branch to Register branches unconditionally to an address in <Rn>.
+ JMP (R6) // <=> br x6, must fault
+ RET // never reached
diff --git a/pkg/sentry/syscalls/linux/sys_file.go b/pkg/sentry/syscalls/linux/sys_file.go
index 2797c6a72..8cf6401e7 100644
--- a/pkg/sentry/syscalls/linux/sys_file.go
+++ b/pkg/sentry/syscalls/linux/sys_file.go
@@ -1057,7 +1057,7 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
case linux.F_SETOWN_EX:
addr := args[2].Pointer()
var owner linux.FOwnerEx
- n, err := t.CopyIn(addr, &owner)
+ _, err := t.CopyIn(addr, &owner)
if err != nil {
return 0, nil, err
}
@@ -1069,21 +1069,21 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, syserror.ESRCH
}
a.SetOwnerTask(t, task)
- return uintptr(n), nil, nil
+ return 0, nil, nil
case linux.F_OWNER_PID:
tg := t.PIDNamespace().ThreadGroupWithID(kernel.ThreadID(owner.PID))
if tg == nil {
return 0, nil, syserror.ESRCH
}
a.SetOwnerThreadGroup(t, tg)
- return uintptr(n), nil, nil
+ return 0, nil, nil
case linux.F_OWNER_PGRP:
pg := t.PIDNamespace().ProcessGroupWithID(kernel.ProcessGroupID(owner.PID))
if pg == nil {
return 0, nil, syserror.ESRCH
}
a.SetOwnerProcessGroup(t, pg)
- return uintptr(n), nil, nil
+ return 0, nil, nil
default:
return 0, nil, syserror.EINVAL
}
diff --git a/pkg/sentry/syscalls/linux/vfs2/fd.go b/pkg/sentry/syscalls/linux/vfs2/fd.go
index 517394ba9..67f191551 100644
--- a/pkg/sentry/syscalls/linux/vfs2/fd.go
+++ b/pkg/sentry/syscalls/linux/vfs2/fd.go
@@ -185,11 +185,11 @@ func Fcntl(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Syscall
return 0, nil, err
case linux.F_SETOWN_EX:
var owner linux.FOwnerEx
- n, err := t.CopyIn(args[2].Pointer(), &owner)
+ _, err := t.CopyIn(args[2].Pointer(), &owner)
if err != nil {
return 0, nil, err
}
- return uintptr(n), nil, setAsyncOwner(t, file, owner.Type, owner.PID)
+ return 0, nil, setAsyncOwner(t, file, owner.Type, owner.PID)
case linux.F_GETPIPE_SZ:
pipefile, ok := file.Impl().(*pipe.VFSPipeFD)
if !ok {
diff --git a/pkg/sentry/syscalls/linux/vfs2/filesystem.go b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
index 6b14c2bef..b6d2ddd65 100644
--- a/pkg/sentry/syscalls/linux/vfs2/filesystem.go
+++ b/pkg/sentry/syscalls/linux/vfs2/filesystem.go
@@ -18,7 +18,6 @@ import (
"gvisor.dev/gvisor/pkg/abi/linux"
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/kernel"
- "gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -245,55 +244,6 @@ func renameat(t *kernel.Task, olddirfd int32, oldpathAddr usermem.Addr, newdirfd
})
}
-// Fallocate implements linux system call fallocate(2).
-func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
- fd := args[0].Int()
- mode := args[1].Uint64()
- offset := args[2].Int64()
- length := args[3].Int64()
-
- file := t.GetFileVFS2(fd)
-
- if file == nil {
- return 0, nil, syserror.EBADF
- }
- defer file.DecRef()
-
- if !file.IsWritable() {
- return 0, nil, syserror.EBADF
- }
-
- if mode != 0 {
- return 0, nil, syserror.ENOTSUP
- }
-
- if offset < 0 || length <= 0 {
- return 0, nil, syserror.EINVAL
- }
-
- size := offset + length
-
- if size < 0 {
- return 0, nil, syserror.EFBIG
- }
-
- limit := limits.FromContext(t).Get(limits.FileSize).Cur
-
- if uint64(size) >= limit {
- t.SendSignal(&arch.SignalInfo{
- Signo: int32(linux.SIGXFSZ),
- Code: arch.SignalInfoUser,
- })
- return 0, nil, syserror.EFBIG
- }
-
- return 0, nil, file.Impl().Allocate(t, mode, uint64(offset), uint64(length))
-
- // File length modified, generate notification.
- // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported.
- // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0)
-}
-
// Rmdir implements Linux syscall rmdir(2).
func Rmdir(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
pathAddr := args[0].Pointer()
diff --git a/pkg/sentry/syscalls/linux/vfs2/setstat.go b/pkg/sentry/syscalls/linux/vfs2/setstat.go
index 6daedd173..37fa56c19 100644
--- a/pkg/sentry/syscalls/linux/vfs2/setstat.go
+++ b/pkg/sentry/syscalls/linux/vfs2/setstat.go
@@ -20,6 +20,7 @@ import (
"gvisor.dev/gvisor/pkg/sentry/arch"
"gvisor.dev/gvisor/pkg/sentry/kernel"
"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+ "gvisor.dev/gvisor/pkg/sentry/limits"
"gvisor.dev/gvisor/pkg/sentry/vfs"
"gvisor.dev/gvisor/pkg/syserror"
"gvisor.dev/gvisor/pkg/usermem"
@@ -211,6 +212,55 @@ func Ftruncate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.Sys
return 0, nil, handleSetSizeError(t, err)
}
+// Fallocate implements linux system call fallocate(2).
+func Fallocate(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
+ fd := args[0].Int()
+ mode := args[1].Uint64()
+ offset := args[2].Int64()
+ length := args[3].Int64()
+
+ file := t.GetFileVFS2(fd)
+
+ if file == nil {
+ return 0, nil, syserror.EBADF
+ }
+ defer file.DecRef()
+
+ if !file.IsWritable() {
+ return 0, nil, syserror.EBADF
+ }
+
+ if mode != 0 {
+ return 0, nil, syserror.ENOTSUP
+ }
+
+ if offset < 0 || length <= 0 {
+ return 0, nil, syserror.EINVAL
+ }
+
+ size := offset + length
+
+ if size < 0 {
+ return 0, nil, syserror.EFBIG
+ }
+
+ limit := limits.FromContext(t).Get(limits.FileSize).Cur
+
+ if uint64(size) >= limit {
+ t.SendSignal(&arch.SignalInfo{
+ Signo: int32(linux.SIGXFSZ),
+ Code: arch.SignalInfoUser,
+ })
+ return 0, nil, syserror.EFBIG
+ }
+
+ return 0, nil, file.Allocate(t, mode, uint64(offset), uint64(length))
+
+ // File length modified, generate notification.
+ // TODO(gvisor.dev/issue/1479): Reenable when Inotify is ported.
+ // file.Dirent.InotifyEvent(linux.IN_MODIFY, 0)
+}
+
// Utime implements Linux syscall utime(2).
func Utime(t *kernel.Task, args arch.SyscallArguments) (uintptr, *kernel.SyscallControl, error) {
pathAddr := args[0].Pointer()
diff --git a/pkg/sentry/vfs/epoll.go b/pkg/sentry/vfs/epoll.go
index 599c3131c..5b009b928 100644
--- a/pkg/sentry/vfs/epoll.go
+++ b/pkg/sentry/vfs/epoll.go
@@ -186,7 +186,7 @@ func (ep *EpollInstance) AddInterest(file *FileDescription, num int32, event lin
}
// Register interest in file.
- mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP
+ mask := event.Events | linux.EPOLLERR | linux.EPOLLHUP
epi := &epollInterest{
epoll: ep,
key: key,
@@ -257,7 +257,7 @@ func (ep *EpollInstance) ModifyInterest(file *FileDescription, num int32, event
}
// Update epi for the next call to ep.ReadEvents().
- mask := event.Events | linux.EPOLLERR | linux.EPOLLRDHUP
+ mask := event.Events | linux.EPOLLERR | linux.EPOLLHUP
ep.mu.Lock()
epi.mask = mask
epi.userData = event.Data
diff --git a/pkg/sentry/vfs/file_description.go b/pkg/sentry/vfs/file_description.go
index 0c42574db..93861fb4a 100644
--- a/pkg/sentry/vfs/file_description.go
+++ b/pkg/sentry/vfs/file_description.go
@@ -354,7 +354,7 @@ type FileDescriptionImpl interface {
// represented by the FileDescription.
StatFS(ctx context.Context) (linux.Statfs, error)
- // Allocate grows file represented by FileDescription to offset + length bytes.
+ // Allocate grows the file to offset + length bytes.
// Only mode == 0 is supported currently.
Allocate(ctx context.Context, mode, offset, length uint64) error
@@ -563,6 +563,11 @@ func (fd *FileDescription) StatFS(ctx context.Context) (linux.Statfs, error) {
return fd.impl.StatFS(ctx)
}
+// Allocate grows file represented by FileDescription to offset + length bytes.
+func (fd *FileDescription) Allocate(ctx context.Context, mode, offset, length uint64) error {
+ return fd.impl.Allocate(ctx, mode, offset, length)
+}
+
// Readiness implements waiter.Waitable.Readiness.
//
// It returns fd's I/O readiness.
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-14 18:06:23 +0000