diff options
| -rw-r--r-- | pkg/sentry/fsimpl/host/BUILD | 25 | ||||
| -rw-r--r-- | pkg/sentry/fsimpl/host/host.go | 94 | ||||
| -rw-r--r-- | pkg/sentry/fsimpl/host/socket.go | 20 |
3 files changed, 88 insertions, 51 deletions
diff --git a/pkg/sentry/fsimpl/host/BUILD b/pkg/sentry/fsimpl/host/BUILD index bd701bbc7..090ae0804 100644 --- a/pkg/sentry/fsimpl/host/BUILD +++ b/pkg/sentry/fsimpl/host/BUILD @@ -1,12 +1,37 @@ load("//tools:defs.bzl", "go_library") +load("//tools/go_generics:defs.bzl", "go_template_instance") licenses(["notice"]) +go_template_instance( + name = "inode_refs", + out = "inode_refs.go", + package = "host", + prefix = "inode", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "inode", + }, +) + +go_template_instance( + name = "connected_endpoint_refs", + out = "connected_endpoint_refs.go", + package = "host", + prefix = "ConnectedEndpoint", + template = "//pkg/refs_vfs2:refs_template", + types = { + "T": "ConnectedEndpoint", + }, +) + go_library( name = "host", srcs = [ + "connected_endpoint_refs.go", "control.go", "host.go", + "inode_refs.go", "ioctl_unsafe.go", "mmap.go", "socket.go", diff --git a/pkg/sentry/fsimpl/host/host.go b/pkg/sentry/fsimpl/host/host.go index 56869f59a..2d3821f33 100644 --- a/pkg/sentry/fsimpl/host/host.go +++ b/pkg/sentry/fsimpl/host/host.go @@ -27,7 +27,6 @@ import ( "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/fspath" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/refs" fslock "gvisor.dev/gvisor/pkg/sentry/fs/lock" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "gvisor.dev/gvisor/pkg/sentry/hostfd" @@ -41,6 +40,44 @@ import ( "gvisor.dev/gvisor/pkg/waiter" ) +func newInode(fs *filesystem, hostFD int, fileType linux.FileMode, isTTY bool) (*inode, error) { + // Determine if hostFD is seekable. If not, this syscall will return ESPIPE + // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character + // devices. + _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) + seekable := err != syserror.ESPIPE + + i := &inode{ + hostFD: hostFD, + ino: fs.NextIno(), + isTTY: isTTY, + wouldBlock: wouldBlock(uint32(fileType)), + seekable: seekable, + // NOTE(b/38213152): Technically, some obscure char devices can be memory + // mapped, but we only allow regular files. + canMap: fileType == linux.S_IFREG, + } + i.pf.inode = i + i.refs.EnableLeakCheck() + + // Non-seekable files can't be memory mapped, assert this. + if !i.seekable && i.canMap { + panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") + } + + // If the hostFD would block, we must set it to non-blocking and handle + // blocking behavior in the sentry. + if i.wouldBlock { + if err := syscall.SetNonblock(i.hostFD, true); err != nil { + return nil, err + } + if err := fdnotifier.AddFD(int32(i.hostFD), &i.queue); err != nil { + return nil, err + } + } + return i, nil +} + // NewFDOptions contains options to NewFD. type NewFDOptions struct { // If IsTTY is true, the file descriptor is a TTY. @@ -76,44 +113,11 @@ func NewFD(ctx context.Context, mnt *vfs.Mount, hostFD int, opts *NewFDOptions) flags = uint32(flagsInt) } - fileMode := linux.FileMode(s.Mode) - fileType := fileMode.FileType() - - // Determine if hostFD is seekable. If not, this syscall will return ESPIPE - // (see fs/read_write.c:llseek), e.g. for pipes, sockets, and some character - // devices. - _, err := unix.Seek(hostFD, 0, linux.SEEK_CUR) - seekable := err != syserror.ESPIPE - - i := &inode{ - hostFD: hostFD, - ino: fs.NextIno(), - isTTY: opts.IsTTY, - wouldBlock: wouldBlock(uint32(fileType)), - seekable: seekable, - // NOTE(b/38213152): Technically, some obscure char devices can be memory - // mapped, but we only allow regular files. - canMap: fileType == linux.S_IFREG, - } - i.pf.inode = i - - // Non-seekable files can't be memory mapped, assert this. - if !i.seekable && i.canMap { - panic("files that can return EWOULDBLOCK (sockets, pipes, etc.) cannot be memory mapped") - } - - // If the hostFD would block, we must set it to non-blocking and handle - // blocking behavior in the sentry. - if i.wouldBlock { - if err := syscall.SetNonblock(i.hostFD, true); err != nil { - return nil, err - } - if err := fdnotifier.AddFD(int32(i.hostFD), &i.queue); err != nil { - return nil, err - } - } - d := &kernfs.Dentry{} + i, err := newInode(fs, hostFD, linux.FileMode(s.Mode).FileType(), opts.IsTTY) + if err != nil { + return nil, err + } d.Init(i) // i.open will take a reference on d. @@ -188,7 +192,7 @@ type inode struct { locks vfs.FileLocks // When the reference count reaches zero, the host fd is closed. - refs.AtomicRefCount + refs inodeRefs // hostFD contains the host fd that this file was originally created from, // which must be available at time of restore. @@ -430,9 +434,19 @@ func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Cre return nil } +// IncRef implements kernfs.Inode. +func (i *inode) IncRef() { + i.refs.IncRef() +} + +// TryIncRef implements kernfs.Inode. +func (i *inode) TryIncRef() bool { + return i.refs.TryIncRef() +} + // DecRef implements kernfs.Inode. func (i *inode) DecRef(ctx context.Context) { - i.AtomicRefCount.DecRefWithDestructor(ctx, func(context.Context) { + i.refs.DecRef(func() { if i.wouldBlock { fdnotifier.RemoveFD(int32(i.hostFD)) } diff --git a/pkg/sentry/fsimpl/host/socket.go b/pkg/sentry/fsimpl/host/socket.go index 4979dd0a9..131145b85 100644 --- a/pkg/sentry/fsimpl/host/socket.go +++ b/pkg/sentry/fsimpl/host/socket.go @@ -22,7 +22,6 @@ import ( "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/fdnotifier" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/refs" "gvisor.dev/gvisor/pkg/sentry/socket/control" "gvisor.dev/gvisor/pkg/sentry/socket/unix/transport" "gvisor.dev/gvisor/pkg/sentry/uniqueid" @@ -59,8 +58,7 @@ func newEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue) (transpor // // +stateify savable type ConnectedEndpoint struct { - // ref keeps track of references to a ConnectedEndpoint. - ref refs.AtomicRefCount + ConnectedEndpointRefs // mu protects fd below. mu sync.RWMutex `state:"nosave"` @@ -132,9 +130,9 @@ func NewConnectedEndpoint(ctx context.Context, hostFD int, addr string, saveable return nil, err } - // AtomicRefCounters start off with a single reference. We need two. - e.ref.IncRef() - e.ref.EnableLeakCheck("host.ConnectedEndpoint") + // ConnectedEndpointRefs start off with a single reference. We need two. + e.IncRef() + e.EnableLeakCheck() return &e, nil } @@ -318,7 +316,7 @@ func (c *ConnectedEndpoint) destroyLocked() { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. func (c *ConnectedEndpoint) Release(ctx context.Context) { - c.ref.DecRefWithDestructor(ctx, func(context.Context) { + c.DecRef(func() { c.mu.Lock() c.destroyLocked() c.mu.Unlock() @@ -348,7 +346,7 @@ func (e *SCMConnectedEndpoint) Init() error { // Release implements transport.ConnectedEndpoint.Release and // transport.Receiver.Release. func (e *SCMConnectedEndpoint) Release(ctx context.Context) { - e.ref.DecRefWithDestructor(ctx, func(context.Context) { + e.DecRef(func() { e.mu.Lock() if err := syscall.Close(e.fd); err != nil { log.Warningf("Failed to close host fd %d: %v", err) @@ -378,8 +376,8 @@ func NewSCMEndpoint(ctx context.Context, hostFD int, queue *waiter.Queue, addr s return nil, err } - // AtomicRefCounters start off with a single reference. We need two. - e.ref.IncRef() - e.ref.EnableLeakCheck("host.SCMConnectedEndpoint") + // ConnectedEndpointRefs start off with a single reference. We need two. + e.IncRef() + e.EnableLeakCheck() return &e, nil } |