diff options
48 files changed, 6253 insertions, 892 deletions
diff --git a/pkg/abi/linux/fuse.go b/pkg/abi/linux/fuse.go index 7e30483ee..d91c97a64 100644 --- a/pkg/abi/linux/fuse.go +++ b/pkg/abi/linux/fuse.go @@ -14,12 +14,20 @@ package linux +import ( + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/marshal/primitive" +) + // +marshal type FUSEOpcode uint32 // +marshal type FUSEOpID uint64 +// FUSE_ROOT_ID is the id of root inode. +const FUSE_ROOT_ID = 1 + // Opcodes for FUSE operations. Analogous to the opcodes in include/linux/fuse.h. const ( FUSE_LOOKUP FUSEOpcode = 1 @@ -116,61 +124,28 @@ type FUSEHeaderOut struct { Unique FUSEOpID } -// FUSEWriteIn is the header written by a daemon when it makes a -// write request to the FUSE filesystem. -// -// +marshal -type FUSEWriteIn struct { - // Fh specifies the file handle that is being written to. - Fh uint64 - - // Offset is the offset of the write. - Offset uint64 - - // Size is the size of data being written. - Size uint32 - - // WriteFlags is the flags used during the write. - WriteFlags uint32 - - // LockOwner is the ID of the lock owner. - LockOwner uint64 - - // Flags is the flags for the request. - Flags uint32 - - _ uint32 -} - // FUSE_INIT flags, consistent with the ones in include/uapi/linux/fuse.h. +// Our taget version is 7.23 but we have few implemented in advance. const ( - FUSE_ASYNC_READ = 1 << 0 - FUSE_POSIX_LOCKS = 1 << 1 - FUSE_FILE_OPS = 1 << 2 - FUSE_ATOMIC_O_TRUNC = 1 << 3 - FUSE_EXPORT_SUPPORT = 1 << 4 - FUSE_BIG_WRITES = 1 << 5 - FUSE_DONT_MASK = 1 << 6 - FUSE_SPLICE_WRITE = 1 << 7 - FUSE_SPLICE_MOVE = 1 << 8 - FUSE_SPLICE_READ = 1 << 9 - FUSE_FLOCK_LOCKS = 1 << 10 - FUSE_HAS_IOCTL_DIR = 1 << 11 - FUSE_AUTO_INVAL_DATA = 1 << 12 - FUSE_DO_READDIRPLUS = 1 << 13 - FUSE_READDIRPLUS_AUTO = 1 << 14 - FUSE_ASYNC_DIO = 1 << 15 - FUSE_WRITEBACK_CACHE = 1 << 16 - FUSE_NO_OPEN_SUPPORT = 1 << 17 - FUSE_PARALLEL_DIROPS = 1 << 18 - FUSE_HANDLE_KILLPRIV = 1 << 19 - FUSE_POSIX_ACL = 1 << 20 - FUSE_ABORT_ERROR = 1 << 21 - FUSE_MAX_PAGES = 1 << 22 - FUSE_CACHE_SYMLINKS = 1 << 23 - FUSE_NO_OPENDIR_SUPPORT = 1 << 24 - FUSE_EXPLICIT_INVAL_DATA = 1 << 25 - FUSE_MAP_ALIGNMENT = 1 << 26 + FUSE_ASYNC_READ = 1 << 0 + FUSE_POSIX_LOCKS = 1 << 1 + FUSE_FILE_OPS = 1 << 2 + FUSE_ATOMIC_O_TRUNC = 1 << 3 + FUSE_EXPORT_SUPPORT = 1 << 4 + FUSE_BIG_WRITES = 1 << 5 + FUSE_DONT_MASK = 1 << 6 + FUSE_SPLICE_WRITE = 1 << 7 + FUSE_SPLICE_MOVE = 1 << 8 + FUSE_SPLICE_READ = 1 << 9 + FUSE_FLOCK_LOCKS = 1 << 10 + FUSE_HAS_IOCTL_DIR = 1 << 11 + FUSE_AUTO_INVAL_DATA = 1 << 12 + FUSE_DO_READDIRPLUS = 1 << 13 + FUSE_READDIRPLUS_AUTO = 1 << 14 + FUSE_ASYNC_DIO = 1 << 15 + FUSE_WRITEBACK_CACHE = 1 << 16 + FUSE_NO_OPEN_SUPPORT = 1 << 17 + FUSE_MAX_PAGES = 1 << 22 // From FUSE 7.28 ) // currently supported FUSE protocol version numbers. @@ -179,6 +154,13 @@ const ( FUSE_KERNEL_MINOR_VERSION = 31 ) +// Constants relevant to FUSE operations. +const ( + FUSE_NAME_MAX = 1024 + FUSE_PAGE_SIZE = 4096 + FUSE_DIRENT_ALIGN = 8 +) + // FUSEInitIn is the request sent by the kernel to the daemon, // to negotiate the version and flags. // @@ -199,7 +181,7 @@ type FUSEInitIn struct { } // FUSEInitOut is the reply sent by the daemon to the kernel -// for FUSEInitIn. +// for FUSEInitIn. We target FUSE 7.23; this struct supports 7.28. // // +marshal type FUSEInitOut struct { @@ -240,13 +222,16 @@ type FUSEInitOut struct { // if the value from daemon is too large. MaxPages uint16 - // MapAlignment is an unknown field and not used by this package at this moment. - // Use as a placeholder to be consistent with the FUSE protocol. - MapAlignment uint16 + _ uint16 _ [8]uint32 } +// FUSE_GETATTR_FH is currently the only flag of FUSEGetAttrIn.GetAttrFlags. +// If it is set, the file handle (FUSEGetAttrIn.Fh) is used to indicate the +// object instead of the node id attribute in the request header. +const FUSE_GETATTR_FH = (1 << 0) + // FUSEGetAttrIn is the request sent by the kernel to the daemon, // to get the attribute of a inode. // @@ -267,22 +252,52 @@ type FUSEGetAttrIn struct { // // +marshal type FUSEAttr struct { - Ino uint64 - Size uint64 - Blocks uint64 - Atime uint64 - Mtime uint64 - Ctime uint64 + // Ino is the inode number of this file. + Ino uint64 + + // Size is the size of this file. + Size uint64 + + // Blocks is the number of the 512B blocks allocated by this file. + Blocks uint64 + + // Atime is the time of last access. + Atime uint64 + + // Mtime is the time of last modification. + Mtime uint64 + + // Ctime is the time of last status change. + Ctime uint64 + + // AtimeNsec is the nano second part of Atime. AtimeNsec uint32 + + // MtimeNsec is the nano second part of Mtime. MtimeNsec uint32 + + // CtimeNsec is the nano second part of Ctime. CtimeNsec uint32 - Mode uint32 - Nlink uint32 - UID uint32 - GID uint32 - Rdev uint32 - BlkSize uint32 - _ uint32 + + // Mode contains the file type and mode. + Mode uint32 + + // Nlink is the number of the hard links. + Nlink uint32 + + // UID is user ID of the owner. + UID uint32 + + // GID is group ID of the owner. + GID uint32 + + // Rdev is the device ID if this is a special file. + Rdev uint32 + + // BlkSize is the block size for filesystem I/O. + BlkSize uint32 + + _ uint32 } // FUSEGetAttrOut is the reply sent by the daemon to the kernel @@ -301,3 +316,558 @@ type FUSEGetAttrOut struct { // Attr contains the metadata returned from the FUSE server Attr FUSEAttr } + +// FUSEEntryOut is the reply sent by the daemon to the kernel +// for FUSE_MKNOD, FUSE_MKDIR, FUSE_SYMLINK, FUSE_LINK and +// FUSE_LOOKUP. +// +// +marshal +type FUSEEntryOut struct { + // NodeID is the ID for current inode. + NodeID uint64 + + // Generation is the generation number of inode. + // Used to identify an inode that have different ID at different time. + Generation uint64 + + // EntryValid indicates timeout for an entry. + EntryValid uint64 + + // AttrValid indicates timeout for an entry's attributes. + AttrValid uint64 + + // EntryValidNsec indicates timeout for an entry in nanosecond. + EntryValidNSec uint32 + + // AttrValidNsec indicates timeout for an entry's attributes in nanosecond. + AttrValidNSec uint32 + + // Attr contains the attributes of an entry. + Attr FUSEAttr +} + +// FUSELookupIn is the request sent by the kernel to the daemon +// to look up a file name. +// +// Dynamically-sized objects cannot be marshalled. +type FUSELookupIn struct { + marshal.StubMarshallable + + // Name is a file name to be looked up. + Name string +} + +// MarshalBytes serializes r.name to the dst buffer. +func (r *FUSELookupIn) MarshalBytes(buf []byte) { + copy(buf, r.Name) +} + +// SizeBytes is the size of the memory representation of FUSELookupIn. +// 1 extra byte for null-terminated string. +func (r *FUSELookupIn) SizeBytes() int { + return len(r.Name) + 1 +} + +// MAX_NON_LFS indicates the maximum offset without large file support. +const MAX_NON_LFS = ((1 << 31) - 1) + +// flags returned by OPEN request. +const ( + // FOPEN_DIRECT_IO indicates bypassing page cache for this opened file. + FOPEN_DIRECT_IO = 1 << 0 + // FOPEN_KEEP_CACHE avoids invalidate of data cache on open. + FOPEN_KEEP_CACHE = 1 << 1 + // FOPEN_NONSEEKABLE indicates the file cannot be seeked. + FOPEN_NONSEEKABLE = 1 << 2 +) + +// FUSEOpenIn is the request sent by the kernel to the daemon, +// to negotiate flags and get file handle. +// +// +marshal +type FUSEOpenIn struct { + // Flags of this open request. + Flags uint32 + + _ uint32 +} + +// FUSEOpenOut is the reply sent by the daemon to the kernel +// for FUSEOpenIn. +// +// +marshal +type FUSEOpenOut struct { + // Fh is the file handler for opened file. + Fh uint64 + + // OpenFlag for the opened file. + OpenFlag uint32 + + _ uint32 +} + +// FUSE_READ flags, consistent with the ones in include/uapi/linux/fuse.h. +const ( + FUSE_READ_LOCKOWNER = 1 << 1 +) + +// FUSEReadIn is the request sent by the kernel to the daemon +// for FUSE_READ. +// +// +marshal +type FUSEReadIn struct { + // Fh is the file handle in userspace. + Fh uint64 + + // Offset is the read offset. + Offset uint64 + + // Size is the number of bytes to read. + Size uint32 + + // ReadFlags for this FUSE_READ request. + // Currently only contains FUSE_READ_LOCKOWNER. + ReadFlags uint32 + + // LockOwner is the id of the lock owner if there is one. + LockOwner uint64 + + // Flags for the underlying file. + Flags uint32 + + _ uint32 +} + +// FUSEWriteIn is the first part of the payload of the +// request sent by the kernel to the daemon +// for FUSE_WRITE (struct for FUSE version >= 7.9). +// +// The second part of the payload is the +// binary bytes of the data to be written. +// +// +marshal +type FUSEWriteIn struct { + // Fh is the file handle in userspace. + Fh uint64 + + // Offset is the write offset. + Offset uint64 + + // Size is the number of bytes to write. + Size uint32 + + // ReadFlags for this FUSE_WRITE request. + WriteFlags uint32 + + // LockOwner is the id of the lock owner if there is one. + LockOwner uint64 + + // Flags for the underlying file. + Flags uint32 + + _ uint32 +} + +// FUSEWriteOut is the payload of the reply sent by the daemon to the kernel +// for a FUSE_WRITE request. +// +// +marshal +type FUSEWriteOut struct { + // Size is the number of bytes written. + Size uint32 + + _ uint32 +} + +// FUSEReleaseIn is the request sent by the kernel to the daemon +// when there is no more reference to a file. +// +// +marshal +type FUSEReleaseIn struct { + // Fh is the file handler for the file to be released. + Fh uint64 + + // Flags of the file. + Flags uint32 + + // ReleaseFlags of this release request. + ReleaseFlags uint32 + + // LockOwner is the id of the lock owner if there is one. + LockOwner uint64 +} + +// FUSECreateMeta contains all the static fields of FUSECreateIn, +// which is used for FUSE_CREATE. +// +// +marshal +type FUSECreateMeta struct { + // Flags of the creating file. + Flags uint32 + + // Mode is the mode of the creating file. + Mode uint32 + + // Umask is the current file mode creation mask. + Umask uint32 + _ uint32 +} + +// FUSECreateIn contains all the arguments sent by the kernel to the daemon, to +// atomically create and open a new regular file. +// +// Dynamically-sized objects cannot be marshalled. +type FUSECreateIn struct { + marshal.StubMarshallable + + // CreateMeta contains mode, rdev and umash field for FUSE_MKNODS. + CreateMeta FUSECreateMeta + + // Name is the name of the node to create. + Name string +} + +// MarshalBytes serializes r.CreateMeta and r.Name to the dst buffer. +func (r *FUSECreateIn) MarshalBytes(buf []byte) { + r.CreateMeta.MarshalBytes(buf[:r.CreateMeta.SizeBytes()]) + copy(buf[r.CreateMeta.SizeBytes():], r.Name) +} + +// SizeBytes is the size of the memory representation of FUSECreateIn. +// 1 extra byte for null-terminated string. +func (r *FUSECreateIn) SizeBytes() int { + return r.CreateMeta.SizeBytes() + len(r.Name) + 1 +} + +// FUSEMknodMeta contains all the static fields of FUSEMknodIn, +// which is used for FUSE_MKNOD. +// +// +marshal +type FUSEMknodMeta struct { + // Mode of the inode to create. + Mode uint32 + + // Rdev encodes device major and minor information. + Rdev uint32 + + // Umask is the current file mode creation mask. + Umask uint32 + + _ uint32 +} + +// FUSEMknodIn contains all the arguments sent by the kernel +// to the daemon, to create a new file node. +// +// Dynamically-sized objects cannot be marshalled. +type FUSEMknodIn struct { + marshal.StubMarshallable + + // MknodMeta contains mode, rdev and umash field for FUSE_MKNODS. + MknodMeta FUSEMknodMeta + + // Name is the name of the node to create. + Name string +} + +// MarshalBytes serializes r.MknodMeta and r.Name to the dst buffer. +func (r *FUSEMknodIn) MarshalBytes(buf []byte) { + r.MknodMeta.MarshalBytes(buf[:r.MknodMeta.SizeBytes()]) + copy(buf[r.MknodMeta.SizeBytes():], r.Name) +} + +// SizeBytes is the size of the memory representation of FUSEMknodIn. +// 1 extra byte for null-terminated string. +func (r *FUSEMknodIn) SizeBytes() int { + return r.MknodMeta.SizeBytes() + len(r.Name) + 1 +} + +// FUSESymLinkIn is the request sent by the kernel to the daemon, +// to create a symbolic link. +// +// Dynamically-sized objects cannot be marshalled. +type FUSESymLinkIn struct { + marshal.StubMarshallable + + // Name of symlink to create. + Name string + + // Target of the symlink. + Target string +} + +// MarshalBytes serializes r.Name and r.Target to the dst buffer. +// Left null-termination at end of r.Name and r.Target. +func (r *FUSESymLinkIn) MarshalBytes(buf []byte) { + copy(buf, r.Name) + copy(buf[len(r.Name)+1:], r.Target) +} + +// SizeBytes is the size of the memory representation of FUSESymLinkIn. +// 2 extra bytes for null-terminated string. +func (r *FUSESymLinkIn) SizeBytes() int { + return len(r.Name) + len(r.Target) + 2 +} + +// FUSEEmptyIn is used by operations without request body. +type FUSEEmptyIn struct{ marshal.StubMarshallable } + +// MarshalBytes do nothing for marshal. +func (r *FUSEEmptyIn) MarshalBytes(buf []byte) {} + +// SizeBytes is 0 for empty request. +func (r *FUSEEmptyIn) SizeBytes() int { + return 0 +} + +// FUSEMkdirMeta contains all the static fields of FUSEMkdirIn, +// which is used for FUSE_MKDIR. +// +// +marshal +type FUSEMkdirMeta struct { + // Mode of the directory of create. + Mode uint32 + + // Umask is the user file creation mask. + Umask uint32 +} + +// FUSEMkdirIn contains all the arguments sent by the kernel +// to the daemon, to create a new directory. +// +// Dynamically-sized objects cannot be marshalled. +type FUSEMkdirIn struct { + marshal.StubMarshallable + + // MkdirMeta contains Mode and Umask of the directory to create. + MkdirMeta FUSEMkdirMeta + + // Name of the directory to create. + Name string +} + +// MarshalBytes serializes r.MkdirMeta and r.Name to the dst buffer. +func (r *FUSEMkdirIn) MarshalBytes(buf []byte) { + r.MkdirMeta.MarshalBytes(buf[:r.MkdirMeta.SizeBytes()]) + copy(buf[r.MkdirMeta.SizeBytes():], r.Name) +} + +// SizeBytes is the size of the memory representation of FUSEMkdirIn. +// 1 extra byte for null-terminated Name string. +func (r *FUSEMkdirIn) SizeBytes() int { + return r.MkdirMeta.SizeBytes() + len(r.Name) + 1 +} + +// FUSERmDirIn is the request sent by the kernel to the daemon +// when trying to remove a directory. +// +// Dynamically-sized objects cannot be marshalled. +type FUSERmDirIn struct { + marshal.StubMarshallable + + // Name is a directory name to be removed. + Name string +} + +// MarshalBytes serializes r.name to the dst buffer. +func (r *FUSERmDirIn) MarshalBytes(buf []byte) { + copy(buf, r.Name) +} + +// SizeBytes is the size of the memory representation of FUSERmDirIn. +func (r *FUSERmDirIn) SizeBytes() int { + return len(r.Name) + 1 +} + +// FUSEDirents is a list of Dirents received from the FUSE daemon server. +// It is used for FUSE_READDIR. +// +// Dynamically-sized objects cannot be marshalled. +type FUSEDirents struct { + marshal.StubMarshallable + + Dirents []*FUSEDirent +} + +// FUSEDirent is a Dirent received from the FUSE daemon server. +// It is used for FUSE_READDIR. +// +// Dynamically-sized objects cannot be marshalled. +type FUSEDirent struct { + marshal.StubMarshallable + + // Meta contains all the static fields of FUSEDirent. + Meta FUSEDirentMeta + + // Name is the filename of the dirent. + Name string +} + +// FUSEDirentMeta contains all the static fields of FUSEDirent. +// It is used for FUSE_READDIR. +// +// +marshal +type FUSEDirentMeta struct { + // Inode of the dirent. + Ino uint64 + + // Offset of the dirent. + Off uint64 + + // NameLen is the length of the dirent name. + NameLen uint32 + + // Type of the dirent. + Type uint32 +} + +// SizeBytes is the size of the memory representation of FUSEDirents. +func (r *FUSEDirents) SizeBytes() int { + var sizeBytes int + for _, dirent := range r.Dirents { + sizeBytes += dirent.SizeBytes() + } + + return sizeBytes +} + +// UnmarshalBytes deserializes FUSEDirents from the src buffer. +func (r *FUSEDirents) UnmarshalBytes(src []byte) { + for { + if len(src) <= (*FUSEDirentMeta)(nil).SizeBytes() { + break + } + + // Its unclear how many dirents there are in src. Each dirent is dynamically + // sized and so we can't make assumptions about how many dirents we can allocate. + if r.Dirents == nil { + r.Dirents = make([]*FUSEDirent, 0) + } + + // We have to allocate a struct for each dirent - there must be a better way + // to do this. Linux allocates 1 page to store all the dirents and then + // simply reads them from the page. + var dirent FUSEDirent + dirent.UnmarshalBytes(src) + r.Dirents = append(r.Dirents, &dirent) + + src = src[dirent.SizeBytes():] + } +} + +// SizeBytes is the size of the memory representation of FUSEDirent. +func (r *FUSEDirent) SizeBytes() int { + dataSize := r.Meta.SizeBytes() + len(r.Name) + + // Each Dirent must be padded such that its size is a multiple + // of FUSE_DIRENT_ALIGN. Similar to the fuse dirent alignment + // in linux/fuse.h. + return (dataSize + (FUSE_DIRENT_ALIGN - 1)) & ^(FUSE_DIRENT_ALIGN - 1) +} + +// UnmarshalBytes deserializes FUSEDirent from the src buffer. +func (r *FUSEDirent) UnmarshalBytes(src []byte) { + r.Meta.UnmarshalBytes(src) + src = src[r.Meta.SizeBytes():] + + if r.Meta.NameLen > FUSE_NAME_MAX { + // The name is too long and therefore invalid. We don't + // need to unmarshal the name since it'll be thrown away. + return + } + + buf := make([]byte, r.Meta.NameLen) + name := primitive.ByteSlice(buf) + name.UnmarshalBytes(src[:r.Meta.NameLen]) + r.Name = string(name) +} + +// FATTR_* consts are the attribute flags defined in include/uapi/linux/fuse.h. +// These should be or-ed together for setattr to know what has been changed. +const ( + FATTR_MODE = (1 << 0) + FATTR_UID = (1 << 1) + FATTR_GID = (1 << 2) + FATTR_SIZE = (1 << 3) + FATTR_ATIME = (1 << 4) + FATTR_MTIME = (1 << 5) + FATTR_FH = (1 << 6) + FATTR_ATIME_NOW = (1 << 7) + FATTR_MTIME_NOW = (1 << 8) + FATTR_LOCKOWNER = (1 << 9) + FATTR_CTIME = (1 << 10) +) + +// FUSESetAttrIn is the request sent by the kernel to the daemon, +// to set the attribute(s) of a file. +// +// +marshal +type FUSESetAttrIn struct { + // Valid indicates which attributes are modified by this request. + Valid uint32 + + _ uint32 + + // Fh is used to identify the file if FATTR_FH is set in Valid. + Fh uint64 + + // Size is the size that the request wants to change to. + Size uint64 + + // LockOwner is the owner of the lock that the request wants to change to. + LockOwner uint64 + + // Atime is the access time that the request wants to change to. + Atime uint64 + + // Mtime is the modification time that the request wants to change to. + Mtime uint64 + + // Ctime is the status change time that the request wants to change to. + Ctime uint64 + + // AtimeNsec is the nano second part of Atime. + AtimeNsec uint32 + + // MtimeNsec is the nano second part of Mtime. + MtimeNsec uint32 + + // CtimeNsec is the nano second part of Ctime. + CtimeNsec uint32 + + // Mode is the file mode that the request wants to change to. + Mode uint32 + + _ uint32 + + // UID is the user ID of the owner that the request wants to change to. + UID uint32 + + // GID is the group ID of the owner that the request wants to change to. + GID uint32 + + _ uint32 +} + +// FUSEUnlinkIn is the request sent by the kernel to the daemon +// when trying to unlink a node. +// +// Dynamically-sized objects cannot be marshalled. +type FUSEUnlinkIn struct { + marshal.StubMarshallable + + // Name of the node to unlink. + Name string +} + +// MarshalBytes serializes r.name to the dst buffer, which should +// have size len(r.Name) + 1 and last byte set to 0. +func (r *FUSEUnlinkIn) MarshalBytes(buf []byte) { + copy(buf, r.Name) +} + +// SizeBytes is the size of the memory representation of FUSEUnlinkIn. +// 1 extra byte for null-terminated Name string. +func (r *FUSEUnlinkIn) SizeBytes() int { + return len(r.Name) + 1 +} diff --git a/pkg/marshal/marshal_impl_util.go b/pkg/marshal/marshal_impl_util.go index 69f33321e..ea75e09f2 100644 --- a/pkg/marshal/marshal_impl_util.go +++ b/pkg/marshal/marshal_impl_util.go @@ -44,7 +44,7 @@ func (StubMarshallable) MarshalBytes(dst []byte) { // UnmarshalBytes implements Marshallable.UnmarshalBytes. func (StubMarshallable) UnmarshalBytes(src []byte) { - panic("Please implement your own UnMarshalBytes function") + panic("Please implement your own UnmarshalBytes function") } // Packed implements Marshallable.Packed. diff --git a/pkg/sentry/fs/g3doc/fuse.md b/pkg/sentry/fs/g3doc/fuse.md index 2ca84dd74..eccb1fb2f 100644 --- a/pkg/sentry/fs/g3doc/fuse.md +++ b/pkg/sentry/fs/g3doc/fuse.md @@ -95,6 +95,52 @@ ops can be implemented in parallel. - Implement the remaining FUSE ops and decide if we can omit rarely used operations like ioctl. +### Design Details + +#### Lifecycle for a FUSE Request + +- User invokes a syscall +- Sentry prepares corresponding request + - If FUSE device is available + - Write the request in binary + - If FUSE device is full + - Kernel task blocked until available +- Sentry notifies the readers of fuse device that it's ready for read +- FUSE daemon reads the request and processes it +- Sentry waits until a reply is written to the FUSE device + - but returns directly for async requests +- FUSE daemon writes to the fuse device +- Sentry processes the reply + - For sync requests, unblock blocked kernel task + - For async requests, execute pre-specified callback if any +- Sentry returns the syscall to the user + +#### Channels and Queues for Requests in Different Stages + +`connection.initializedChan` + +- a channel that the requests issued before connection initialization blocks + on. + +`fd.queue` + +- a queue of requests that haven’t been read by the FUSE daemon yet. + +`fd.completions` + +- a map of the requests that have been prepared but not yet received a + response, including the ones on the `fd.queue`. + +`fd.waitQueue` + +- a queue of waiters that is waiting for the fuse device fd to be available, + such as the FUSE daemon. + +`fd.fullQueueCh` + +- a channel that the kernel task will be blocked on when the fd is not + available. + # Appendix ## FUSE Protocol diff --git a/pkg/sentry/fsimpl/fuse/BUILD b/pkg/sentry/fsimpl/fuse/BUILD index 999c16bfd..045d7ab08 100644 --- a/pkg/sentry/fsimpl/fuse/BUILD +++ b/pkg/sentry/fsimpl/fuse/BUILD @@ -30,12 +30,17 @@ go_library( name = "fuse", srcs = [ "connection.go", + "connection_control.go", "dev.go", + "directory.go", + "file.go", "fusefs.go", - "init.go", "inode_refs.go", + "read_write.go", "register.go", + "regular_file.go", "request_list.go", + "request_response.go", ], visibility = ["//pkg/sentry:internal"], deps = [ @@ -44,6 +49,7 @@ go_library( "//pkg/log", "//pkg/marshal", "//pkg/refs", + "//pkg/safemem", "//pkg/sentry/fsimpl/devtmpfs", "//pkg/sentry/fsimpl/kernfs", "//pkg/sentry/kernel", @@ -60,7 +66,11 @@ go_library( go_test( name = "fuse_test", size = "small", - srcs = ["dev_test.go"], + srcs = [ + "connection_test.go", + "dev_test.go", + "utils_test.go", + ], library = ":fuse", deps = [ "//pkg/abi/linux", diff --git a/pkg/sentry/fsimpl/fuse/connection.go b/pkg/sentry/fsimpl/fuse/connection.go index eb1d1a2b7..dbc5e1954 100644 --- a/pkg/sentry/fsimpl/fuse/connection.go +++ b/pkg/sentry/fsimpl/fuse/connection.go @@ -15,31 +15,17 @@ package fuse import ( - "errors" - "fmt" "sync" - "sync/atomic" - "syscall" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" - "gvisor.dev/gvisor/pkg/marshal" "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" ) -// maxActiveRequestsDefault is the default setting controlling the upper bound -// on the number of active requests at any given time. -const maxActiveRequestsDefault = 10000 - -// 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 @@ -52,43 +38,36 @@ const ( fuseDefaultMaxPagesPerReq = 32 ) -// Request represents a FUSE operation request that hasn't been sent to the -// server yet. -// -// +stateify savable -type Request struct { - requestEntry - - id linux.FUSEOpID - hdr *linux.FUSEHeaderIn - data []byte -} - -// Response represents an actual response from the server, including the -// response payload. -// -// +stateify savable -type Response struct { - opcode linux.FUSEOpcode - hdr linux.FUSEHeaderOut - data []byte -} - // connection is the struct by which the sentry communicates with the FUSE server daemon. +// Lock order: +// - conn.fd.mu +// - conn.mu +// - conn.asyncMu type connection struct { fd *DeviceFD + // mu protects access to struct memebers. + mu sync.Mutex + + // attributeVersion is the version of connection's attributes. + attributeVersion uint64 + + // We target FUSE 7.23. // 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 + // - FUSE_PARALLEL_DIROPS (7.25) + // - FUSE_HANDLE_KILLPRIV (7.26) + // - FUSE_POSIX_ACL: affects defaultPermissions, posixACL, xattr handler (7.26) + // - FUSE_ABORT_ERROR (7.27) + // - FUSE_CACHE_SYMLINKS (7.28) + // - FUSE_NO_OPENDIR_SUPPORT (7.29) + // - FUSE_EXPLICIT_INVAL_DATA: requires page caching eviction (7.30) + // - FUSE_MAP_ALIGNMENT (7.31) // initialized after receiving FUSE_INIT reply. // Until it's set, suspend sending FUSE requests. @@ -98,10 +77,6 @@ type connection struct { // 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, @@ -109,48 +84,55 @@ type connection struct { // 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. + // Used for destory (not yet implemented). 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 + // aborted via sysfs, and will send ECONNABORTED to read after disconnection (instead of ENODEV). + // Set only if abortErr is true and via fuse control fs (not yet implemented). + // TODO(gvisor.dev/issue/3525): set this to true when user aborts. + aborted bool - // congestionThreshold for NumBackground. - // Negotiated in FUSE_INIT. - congestionThreshold uint16 + // numWating is the number of requests waiting to be + // sent to FUSE device or being processed by FUSE daemon. + numWaiting uint32 - // maxBackground is the maximum number of NumBackground. - // Block connection when it is reached. - // Negotiated in FUSE_INIT. - maxBackground uint16 + // Terminology note: + // + // - `asyncNumMax` is the `MaxBackground` in the FUSE_INIT_IN struct. + // + // - `asyncCongestionThreshold` is the `CongestionThreshold` in the FUSE_INIT_IN struct. + // + // We call the "background" requests in unix term as async requests. + // The "async requests" in unix term is our async requests that expect a reply, + // i.e. `!request.noReply` - // numActiveBackground is the number of requests in background and has being marked as active. - numActiveBackground uint16 + // asyncMu protects the async request fields. + asyncMu sync.Mutex - // numWating is the number of requests waiting for completion. - numWaiting uint32 + // asyncNum is the number of async requests. + // Protected by asyncMu. + asyncNum uint16 - // TODO(gvisor.dev/issue/3185): BgQueue - // some queue for background queued requests. + // asyncCongestionThreshold the number of async requests. + // Negotiated in FUSE_INIT as "CongestionThreshold". + // TODO(gvisor.dev/issue/3529): add congestion control. + // Protected by asyncMu. + asyncCongestionThreshold uint16 - // bgLock protects: - // MaxBackground, CongestionThreshold, NumBackground, - // NumActiveBackground, BgQueue, Blocked. - bgLock sync.Mutex + // asyncNumMax is the maximum number of asyncNum. + // Connection blocks the async requests when it is reached. + // Negotiated in FUSE_INIT as "MaxBackground". + // Protected by asyncMu. + asyncNumMax uint16 // maxRead is the maximum size of a read buffer in in bytes. + // Initialized from a fuse fs parameter. maxRead uint32 // maxWrite is the maximum size of a write buffer in bytes. @@ -165,23 +147,20 @@ type connection struct { // Negotiated and only set in INIT. minor uint32 - // asyncRead if read pages asynchronously. + // atomicOTrunc is true when FUSE does not send a separate SETATTR request + // before open with O_TRUNC flag. // Negotiated and only set in INIT. - asyncRead bool + atomicOTrunc bool - // abortErr is true if kernel need to return an unique read error after abort. + // asyncRead if read pages asynchronously. // Negotiated and only set in INIT. - abortErr bool + asyncRead 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 @@ -189,116 +168,70 @@ type connection struct { // dontMask if filestestem does not apply umask to creation modes. // Negotiated in INIT. dontMask bool + + // noOpen if FUSE server doesn't support open operation. + // This flag only influence performance, not correctness of the program. + noOpen bool } // newFUSEConnection creates a FUSE connection to fd. -func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, maxInFlightRequests uint64) (*connection, error) { +func newFUSEConnection(_ context.Context, fd *vfs.FileDescription, opts *filesystemOptions) (*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) - fuseFD.mounted = true // Create the writeBuf for the header to be stored in. hdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) fuseFD.writeBuf = make([]byte, hdrLen) fuseFD.completions = make(map[linux.FUSEOpID]*futureResponse) - fuseFD.fullQueueCh = make(chan struct{}, maxInFlightRequests) + fuseFD.fullQueueCh = make(chan struct{}, opts.maxActiveRequests) fuseFD.writeCursor = 0 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) { - conn.fd.mu.Lock() - defer conn.fd.mu.Unlock() - conn.fd.nextOpID += linux.FUSEOpID(reqIDStep) - - hdrLen := (*linux.FUSEHeaderIn)(nil).SizeBytes() - hdr := linux.FUSEHeaderIn{ - Len: uint32(hdrLen + payload.SizeBytes()), - Opcode: opcode, - Unique: conn.fd.nextOpID, - NodeID: ino, - UID: uint32(creds.EffectiveKUID), - GID: uint32(creds.EffectiveKGID), - PID: pid, - } - - buf := make([]byte, hdr.Len) - hdr.MarshalUnsafe(buf[:hdrLen]) - payload.MarshalUnsafe(buf[hdrLen:]) - - return &Request{ - id: hdr.Unique, - hdr: &hdr, - data: buf, + fd: fuseFD, + asyncNumMax: fuseDefaultMaxBackground, + asyncCongestionThreshold: fuseDefaultCongestionThreshold, + maxRead: opts.maxRead, + maxPages: fuseDefaultMaxPagesPerReq, + initializedChan: make(chan struct{}), + connected: true, }, nil } -// Call makes a request to the server and blocks the invoking task until a -// 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. +// It's a simple wrapper around Call(). +func (conn *connection) CallAsync(t *kernel.Task, r *Request) error { + r.async = true + _, err := conn.Call(t, r) + return err } -// 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. +// Call makes a request to the server. +// Block before the connection is initialized. // When the Request is FUSE_INIT, it will not be blocked before initialization. -func (conn *connection) CallAsync(t *kernel.Task, r *Request) error { +// Task should never be nil. +// +// For a sync request, it blocks the invoking task until +// a server responds with a response. +// +// For an async request (that do not expect a response immediately), +// it returns directly unless being blocked either before initialization +// or when there are too many async requests ongoing. +// +// Example for async request: +// init, readahead, write, async read/write, fuse_notify_reply, +// non-sync release, interrupt, forget. +// +// The forget request does not have a reply, +// as documented in include/uapi/linux/fuse.h:FUSE_FORGET. +func (conn *connection) Call(t *kernel.Task, r *Request) (*Response, 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 + return nil, 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 } @@ -315,31 +248,6 @@ func (conn *connection) call(t *kernel.Task, r *Request) (*Response, error) { return fut.resolve(t) } -// Error returns the error of the FUSE call. -func (r *Response) Error() error { - errno := r.hdr.Error - if errno >= 0 { - return nil - } - - sysErrNo := syscall.Errno(-errno) - return error(sysErrNo) -} - -// UnmarshalPayload unmarshals the response data into m. -func (r *Response) UnmarshalPayload(m marshal.Marshallable) error { - hdrLen := r.hdr.SizeBytes() - haveDataLen := r.hdr.Len - uint32(hdrLen) - wantDataLen := uint32(m.SizeBytes()) - - if haveDataLen < wantDataLen { - return fmt.Errorf("payload too small. Minimum data lenth required: %d, but got data length %d", wantDataLen, haveDataLen) - } - - m.UnmarshalUnsafe(r.data[hdrLen:]) - return nil -} - // 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) { @@ -358,11 +266,6 @@ func (conn *connection) callFuture(t *kernel.Task, r *Request) (*futureResponse, // if there are always too many ongoing requests all the time. The // supported maxActiveRequests setting should be really high to avoid this. for conn.fd.numActiveRequests == conn.fd.fs.opts.maxActiveRequests { - if t == nil { - // Since there is no task that is waiting. We must error out. - return nil, errors.New("FUSE request queue full") - } - log.Infof("Blocking request %v from being queued. Too many active requests: %v", r.id, conn.fd.numActiveRequests) conn.fd.mu.Unlock() @@ -378,9 +281,19 @@ 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) { + // Check connected again holding conn.mu. + conn.mu.Lock() + if !conn.connected { + conn.mu.Unlock() + // we checked connected before, + // this must be due to aborted connection. + return nil, syserror.ECONNABORTED + } + conn.mu.Unlock() + conn.fd.queue.PushBack(r) - conn.fd.numActiveRequests += 1 - fut := newFutureResponse(r.hdr.Opcode) + conn.fd.numActiveRequests++ + fut := newFutureResponse(r) conn.fd.completions[r.id] = fut // Signal the readers that there is something to read. @@ -388,50 +301,3 @@ func (conn *connection) callFutureLocked(t *kernel.Task, r *Request) (*futureRes return fut, nil } - -// futureResponse represents an in-flight request, that may or may not have -// completed yet. Convert it to a resolved Response by calling Resolve, but note -// that this may block. -// -// +stateify savable -type futureResponse struct { - opcode linux.FUSEOpcode - ch chan struct{} - hdr *linux.FUSEHeaderOut - data []byte -} - -// newFutureResponse creates a future response to a FUSE request. -func newFutureResponse(opcode linux.FUSEOpcode) *futureResponse { - return &futureResponse{ - opcode: opcode, - ch: make(chan struct{}), - } -} - -// resolve blocks the task until the server responds to its corresponding request, -// then returns a resolved response. -func (f *futureResponse) resolve(t *kernel.Task) (*Response, error) { - // If there is no Task associated with this request - then we don't try to resolve - // the response. Instead, the task writing the response (proxy to the server) will - // process the response on our behalf. - if t == nil { - log.Infof("fuse.Response.resolve: Not waiting on a response from server.") - return nil, nil - } - - if err := t.Block(f.ch); err != nil { - return nil, err - } - - return f.getResponse(), nil -} - -// getResponse creates a Response from the data the futureResponse has. -func (f *futureResponse) getResponse() *Response { - return &Response{ - opcode: f.opcode, - hdr: *f.hdr, - data: f.data, - } -} diff --git a/pkg/sentry/fsimpl/fuse/init.go b/pkg/sentry/fsimpl/fuse/connection_control.go index 779c2bd3f..bfde78559 100644 --- a/pkg/sentry/fsimpl/fuse/init.go +++ b/pkg/sentry/fsimpl/fuse/connection_control.go @@ -15,7 +15,11 @@ package fuse import ( + "sync/atomic" + "syscall" + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" ) @@ -29,9 +33,10 @@ const ( // Follow the same behavior as unix fuse implementation. fuseMaxTimeGranNs = 1000000000 - // Minimum value for MaxWrite. + // Minimum value for MaxWrite and MaxRead. // Follow the same behavior as unix fuse implementation. fuseMinMaxWrite = 4096 + fuseMinMaxRead = 4096 // Temporary default value for max readahead, 128kb. fuseDefaultMaxReadahead = 131072 @@ -49,6 +54,26 @@ var ( MaxUserCongestionThreshold uint16 = fuseDefaultCongestionThreshold ) +// 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 +} + // InitSend sends a FUSE_INIT request. func (conn *connection) InitSend(creds *auth.Credentials, pid uint32) error { in := linux.FUSEInitIn{ @@ -70,29 +95,31 @@ func (conn *connection) InitSend(creds *auth.Credentials, pid uint32) error { } // InitRecv receives a FUSE_INIT reply and process it. +// +// Preconditions: conn.asyncMu must not be held if minor verion is newer than 13. 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 { + initRes := fuseInitRes{initLen: res.DataLen()} + if err := res.UnmarshalPayload(&initRes); err != nil { return err } - return conn.initProcessReply(&out, hasSysAdminCap) + return conn.initProcessReply(&initRes.initOut, hasSysAdminCap) } // Process the FUSE_INIT reply from the FUSE server. +// It tries to acquire the conn.asyncMu lock if minor version is newer than 13. func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap bool) error { + // No matter error or not, always set initialzied. + // to unblock the blocked requests. + defer conn.SetInitialized() + // 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 } @@ -100,29 +127,14 @@ func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap 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 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 } // No support for the following flags before minor version 6. @@ -131,8 +143,6 @@ func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap 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). @@ -148,19 +158,90 @@ func (conn *connection) initProcessReply(out *linux.FUSEInitOut, hasSysAdminCap } } - // No support for negotiating MaxWrite before minor version 5. - if out.Minor >= 5 { - conn.maxWrite = out.MaxWrite - } else { - conn.maxWrite = fuseMinMaxWrite + // No support for limits before minor version 13. + if out.Minor >= 13 { + conn.asyncMu.Lock() + + if out.MaxBackground > 0 { + conn.asyncNumMax = out.MaxBackground + + if !hasSysAdminCap && + conn.asyncNumMax > MaxUserBackgroundRequest { + conn.asyncNumMax = MaxUserBackgroundRequest + } + } + + if out.CongestionThreshold > 0 { + conn.asyncCongestionThreshold = out.CongestionThreshold + + if !hasSysAdminCap && + conn.asyncCongestionThreshold > MaxUserCongestionThreshold { + conn.asyncCongestionThreshold = MaxUserCongestionThreshold + } + } + + conn.asyncMu.Unlock() } - if conn.maxWrite < fuseMinMaxWrite { - conn.maxWrite = fuseMinMaxWrite + + return nil +} + +// Abort this FUSE connection. +// It tries to acquire conn.fd.mu, conn.lock, conn.bgLock in order. +// All possible requests waiting or blocking will be aborted. +// +// Preconditions: conn.fd.mu is locked. +func (conn *connection) Abort(ctx context.Context) { + conn.mu.Lock() + conn.asyncMu.Lock() + + if !conn.connected { + conn.asyncMu.Unlock() + conn.mu.Unlock() + conn.fd.mu.Unlock() + return } - // Set connection as initialized and unblock the requests - // issued before init. - conn.SetInitialized() + conn.connected = false - return nil + // Empty the `fd.queue` that holds the requests + // not yet read by the FUSE daemon yet. + // These are a subset of the requests in `fuse.completion` map. + for !conn.fd.queue.Empty() { + req := conn.fd.queue.Front() + conn.fd.queue.Remove(req) + } + + var terminate []linux.FUSEOpID + + // 2. Collect the requests have not been sent to FUSE daemon, + // or have not received a reply. + for unique := range conn.fd.completions { + terminate = append(terminate, unique) + } + + // Release locks to avoid deadlock. + conn.asyncMu.Unlock() + conn.mu.Unlock() + + // 1. The requets blocked before initialization. + // Will reach call() `connected` check and return. + if !conn.Initialized() { + conn.SetInitialized() + } + + // 2. Terminate the requests collected above. + // Set ECONNABORTED error. + // sendError() will remove them from `fd.completion` map. + // Will enter the path of a normally received error. + for _, toTerminate := range terminate { + conn.fd.sendError(ctx, -int32(syscall.ECONNABORTED), toTerminate) + } + + // 3. The requests not yet written to FUSE device. + // Early terminate. + // Will reach callFutureLocked() `connected` check and return. + close(conn.fd.fullQueueCh) + + // TODO(gvisor.dev/issue/3528): Forget all pending forget reqs. } diff --git a/pkg/sentry/fsimpl/fuse/connection_test.go b/pkg/sentry/fsimpl/fuse/connection_test.go new file mode 100644 index 000000000..91d16c1cf --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/connection_test.go @@ -0,0 +1,117 @@ +// 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 ( + "math/rand" + "syscall" + "testing" + + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/syserror" +) + +// TestConnectionInitBlock tests if initialization +// correctly blocks and unblocks the connection. +// Since it's unfeasible to test kernelTask.Block() in unit test, +// the code in Call() are not tested here. +func TestConnectionInitBlock(t *testing.T) { + s := setup(t) + defer s.Destroy() + + k := kernel.KernelFromContext(s.Ctx) + + conn, _, err := newTestConnection(s, k, maxActiveRequestsDefault) + if err != nil { + t.Fatalf("newTestConnection: %v", err) + } + + select { + case <-conn.initializedChan: + t.Fatalf("initializedChan should be blocking before SetInitialized") + default: + } + + conn.SetInitialized() + + select { + case <-conn.initializedChan: + default: + t.Fatalf("initializedChan should not be blocking after SetInitialized") + } +} + +func TestConnectionAbort(t *testing.T) { + s := setup(t) + defer s.Destroy() + + k := kernel.KernelFromContext(s.Ctx) + creds := auth.CredentialsFromContext(s.Ctx) + task := kernel.TaskFromContext(s.Ctx) + + const numRequests uint64 = 256 + + conn, _, err := newTestConnection(s, k, numRequests) + if err != nil { + t.Fatalf("newTestConnection: %v", err) + } + + testObj := &testPayload{ + data: rand.Uint32(), + } + + var futNormal []*futureResponse + + for i := 0; i < int(numRequests); i++ { + req, err := conn.NewRequest(creds, uint32(i), uint64(i), 0, testObj) + if err != nil { + t.Fatalf("NewRequest creation failed: %v", err) + } + fut, err := conn.callFutureLocked(task, req) + if err != nil { + t.Fatalf("callFutureLocked failed: %v", err) + } + futNormal = append(futNormal, fut) + } + + conn.Abort(s.Ctx) + + // Abort should unblock the initialization channel. + // Note: no test requests are actually blocked on `conn.initializedChan`. + select { + case <-conn.initializedChan: + default: + t.Fatalf("initializedChan should not be blocking after SetInitialized") + } + + // Abort will return ECONNABORTED error to unblocked requests. + for _, fut := range futNormal { + if fut.getResponse().hdr.Error != -int32(syscall.ECONNABORTED) { + t.Fatalf("Incorrect error code received for aborted connection: %v", fut.getResponse().hdr.Error) + } + } + + // After abort, Call() should return directly with ENOTCONN. + req, err := conn.NewRequest(creds, 0, 0, 0, testObj) + if err != nil { + t.Fatalf("NewRequest creation failed: %v", err) + } + _, err = conn.Call(task, req) + if err != syserror.ENOTCONN { + t.Fatalf("Incorrect error code received for Call() after connection aborted") + } + +} diff --git a/pkg/sentry/fsimpl/fuse/dev.go b/pkg/sentry/fsimpl/fuse/dev.go index e522ff9a0..5539466ff 100644 --- a/pkg/sentry/fsimpl/fuse/dev.go +++ b/pkg/sentry/fsimpl/fuse/dev.go @@ -19,7 +19,6 @@ import ( "gvisor.dev/gvisor/pkg/abi/linux" "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" @@ -56,9 +55,6 @@ type DeviceFD struct { vfs.DentryMetadataFileDescriptionImpl vfs.NoLockFD - // mounted specifies whether a FUSE filesystem was mounted using the DeviceFD. - mounted bool - // nextOpID is used to create new requests. nextOpID linux.FUSEOpID @@ -100,13 +96,15 @@ type DeviceFD struct { // Release implements vfs.FileDescriptionImpl.Release. func (fd *DeviceFD) Release(context.Context) { - fd.fs.conn.connected = false + if fd.fs != nil { + 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) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. - if !fd.mounted { + if fd.fs == nil { return 0, syserror.EPERM } @@ -116,10 +114,16 @@ func (fd *DeviceFD) PRead(ctx context.Context, dst usermem.IOSequence, offset in // Read implements vfs.FileDescriptionImpl.Read. func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. - if !fd.mounted { + if fd.fs == nil { return 0, syserror.EPERM } + // Return ENODEV if the filesystem is umounted. + if fd.fs.umounted { + // TODO(gvisor.dev/issue/3525): return ECONNABORTED if aborted via fuse control fs. + return 0, syserror.ENODEV + } + // We require that any Read done on this filesystem have a sane minimum // read buffer. It must have the capacity for the fixed parts of any request // header (Linux uses the request header and the FUSEWriteIn header for this @@ -143,58 +147,82 @@ func (fd *DeviceFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.R } // readLocked implements the reading of the fuse device while locked with DeviceFD.mu. +// +// Preconditions: dst is large enough for any reasonable request. func (fd *DeviceFD) readLocked(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { - if fd.queue.Empty() { - return 0, syserror.ErrWouldBlock - } + var req *Request - var readCursor uint32 - var bytesRead int64 - for { - req := fd.queue.Front() - if dst.NumBytes() < int64(req.hdr.Len) { - // The request is too large. Cannot process it. All requests must be smaller than the - // negotiated size as specified by Connection.MaxWrite set as part of the FUSE_INIT - // handshake. - errno := -int32(syscall.EIO) - if req.hdr.Opcode == linux.FUSE_SETXATTR { - errno = -int32(syscall.E2BIG) - } + // Find the first valid request. + // For the normal case this loop only execute once. + for !fd.queue.Empty() { + req = fd.queue.Front() - // Return the error to the calling task. - if err := fd.sendError(ctx, errno, req); err != nil { - return 0, err - } + if int64(req.hdr.Len)+int64(len(req.payload)) <= dst.NumBytes() { + break + } - // We're done with this request. - fd.queue.Remove(req) + // The request is too large. Cannot process it. All requests must be smaller than the + // negotiated size as specified by Connection.MaxWrite set as part of the FUSE_INIT + // handshake. + errno := -int32(syscall.EIO) + if req.hdr.Opcode == linux.FUSE_SETXATTR { + errno = -int32(syscall.E2BIG) + } - // Restart the read as this request was invalid. - log.Warningf("fuse.DeviceFD.Read: request found was too large. Restarting read.") - return fd.readLocked(ctx, dst, opts) + // Return the error to the calling task. + if err := fd.sendError(ctx, errno, req.hdr.Unique); err != nil { + return 0, err } - n, err := dst.CopyOut(ctx, req.data[readCursor:]) + // We're done with this request. + fd.queue.Remove(req) + req = nil + } + + if req == nil { + return 0, syserror.ErrWouldBlock + } + + // We already checked the size: dst must be able to fit the whole request. + // Now we write the marshalled header, the payload, + // and the potential additional payload + // to the user memory IOSequence. + + n, err := dst.CopyOut(ctx, req.data) + if err != nil { + return 0, err + } + if n != len(req.data) { + return 0, syserror.EIO + } + + if req.hdr.Opcode == linux.FUSE_WRITE { + written, err := dst.DropFirst(n).CopyOut(ctx, req.payload) if err != nil { return 0, err } - readCursor += uint32(n) - bytesRead += int64(n) - - if readCursor >= req.hdr.Len { - // Fully done with this req, remove it from the queue. - fd.queue.Remove(req) - break + if written != len(req.payload) { + return 0, syserror.EIO } + n += int(written) + } + + // Fully done with this req, remove it from the queue. + fd.queue.Remove(req) + + // Remove noReply ones from map of requests expecting a reply. + if req.noReply { + fd.numActiveRequests -= 1 + delete(fd.completions, req.hdr.Unique) } - return bytesRead, nil + return int64(n), nil } // PWrite implements vfs.FileDescriptionImpl.PWrite. func (fd *DeviceFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. - if !fd.mounted { + if fd.fs == nil { return 0, syserror.EPERM } @@ -211,10 +239,15 @@ func (fd *DeviceFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs. // writeLocked implements writing to the fuse device while locked with DeviceFD.mu. func (fd *DeviceFD) writeLocked(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. - if !fd.mounted { + if fd.fs == nil { return 0, syserror.EPERM } + // Return ENODEV if the filesystem is umounted. + if fd.fs.umounted { + return 0, syserror.ENODEV + } + var cn, n int64 hdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) @@ -276,7 +309,8 @@ func (fd *DeviceFD) writeLocked(ctx context.Context, src usermem.IOSequence, opt fut, ok := fd.completions[hdr.Unique] if !ok { - // Server sent us a response for a request we never sent? + // Server sent us a response for a request we never sent, + // or for which we already received a reply (e.g. aborted), an unlikely event. return 0, syserror.EINVAL } @@ -307,8 +341,23 @@ func (fd *DeviceFD) writeLocked(ctx context.Context, src usermem.IOSequence, opt // Readiness implements vfs.FileDescriptionImpl.Readiness. func (fd *DeviceFD) Readiness(mask waiter.EventMask) waiter.EventMask { + fd.mu.Lock() + defer fd.mu.Unlock() + return fd.readinessLocked(mask) +} + +// readinessLocked implements checking the readiness of the fuse device while +// locked with DeviceFD.mu. +func (fd *DeviceFD) readinessLocked(mask waiter.EventMask) waiter.EventMask { var ready waiter.EventMask - ready |= waiter.EventOut // FD is always writable + + if fd.fs.umounted { + ready |= waiter.EventErr + return ready & mask + } + + // FD is always writable. + ready |= waiter.EventOut if !fd.queue.Empty() { // Have reqs available, FD is readable. ready |= waiter.EventIn @@ -330,7 +379,7 @@ func (fd *DeviceFD) EventUnregister(e *waiter.Entry) { // Seek implements vfs.FileDescriptionImpl.Seek. func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { // Operations on /dev/fuse don't make sense until a FUSE filesystem is mounted. - if !fd.mounted { + if fd.fs == nil { return 0, syserror.EPERM } @@ -338,59 +387,59 @@ func (fd *DeviceFD) Seek(ctx context.Context, offset int64, whence int32) (int64 } // sendResponse sends a response to the waiting task (if any). +// +// Preconditions: fd.mu must be held. func (fd *DeviceFD) sendResponse(ctx context.Context, fut *futureResponse) error { - // See if the running task need to perform some action before returning. - // Since we just finished writing the future, we can be sure that - // getResponse generates a populated response. - if err := fd.noReceiverAction(ctx, fut.getResponse()); err != nil { - return err - } + // Signal the task waiting on a response if any. + defer close(fut.ch) // Signal that the queue is no longer full. select { case fd.fullQueueCh <- struct{}{}: default: } - fd.numActiveRequests -= 1 + fd.numActiveRequests-- + + if fut.async { + return fd.asyncCallBack(ctx, fut.getResponse()) + } - // Signal the task waiting on a response. - close(fut.ch) return nil } -// sendError sends an error response to the waiting task (if any). -func (fd *DeviceFD) sendError(ctx context.Context, errno int32, req *Request) error { +// sendError sends an error response to the waiting task (if any) by calling sendResponse(). +// +// Preconditions: fd.mu must be held. +func (fd *DeviceFD) sendError(ctx context.Context, errno int32, unique linux.FUSEOpID) error { // Return the error to the calling task. outHdrLen := uint32((*linux.FUSEHeaderOut)(nil).SizeBytes()) respHdr := linux.FUSEHeaderOut{ Len: outHdrLen, Error: errno, - Unique: req.hdr.Unique, + Unique: unique, } fut, ok := fd.completions[respHdr.Unique] if !ok { - // Server sent us a response for a request we never sent? + // A response for a request we never sent, + // or for which we already received a reply (e.g. aborted). return syserror.EINVAL } delete(fd.completions, respHdr.Unique) fut.hdr = &respHdr - if err := fd.sendResponse(ctx, fut); err != nil { - return err - } - - return nil + return fd.sendResponse(ctx, fut) } -// noReceiverAction has the calling kernel.Task do some action if its known that no -// receiver is going to be waiting on the future channel. This is to be used by: -// FUSE_INIT. -func (fd *DeviceFD) noReceiverAction(ctx context.Context, r *Response) error { - if r.opcode == linux.FUSE_INIT { +// asyncCallBack executes pre-defined callback function for async requests. +// Currently used by: FUSE_INIT. +func (fd *DeviceFD) asyncCallBack(ctx context.Context, r *Response) error { + switch r.opcode { + case linux.FUSE_INIT: creds := auth.CredentialsFromContext(ctx) rootUserNs := kernel.KernelFromContext(ctx).RootUserNamespace() return fd.fs.conn.InitRecv(r, creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, rootUserNs)) + // TODO(gvisor.dev/issue/3247): support async read: correctly process the response. } return nil diff --git a/pkg/sentry/fsimpl/fuse/dev_test.go b/pkg/sentry/fsimpl/fuse/dev_test.go index c18921dd8..5986133e9 100644 --- a/pkg/sentry/fsimpl/fuse/dev_test.go +++ b/pkg/sentry/fsimpl/fuse/dev_test.go @@ -16,12 +16,10 @@ package fuse import ( "fmt" - "io" "math/rand" "testing" "gvisor.dev/gvisor/pkg/abi/linux" - "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" "gvisor.dev/gvisor/pkg/sentry/kernel" "gvisor.dev/gvisor/pkg/sentry/kernel/auth" @@ -35,10 +33,6 @@ import ( // will simply echo the payload back with the appropriate headers. const echoTestOpcode linux.FUSEOpcode = 1000 -type testPayload struct { - data uint32 -} - // TestFUSECommunication tests that the communication layer between the Sentry and the // FUSE server daemon works as expected. func TestFUSECommunication(t *testing.T) { @@ -327,102 +321,3 @@ func fuseServerRun(t *testing.T, s *testutil.System, k *kernel.Kernel, fd *vfs.F } } } - -func setup(t *testing.T) *testutil.System { - k, err := testutil.Boot() - if err != nil { - t.Fatalf("Error creating kernel: %v", err) - } - - ctx := k.SupervisorContext() - creds := auth.CredentialsFromContext(ctx) - - k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ - AllowUserList: true, - AllowUserMount: true, - }) - - mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{}) - if err != nil { - t.Fatalf("NewMountNamespace(): %v", err) - } - - return testutil.NewSystem(ctx, t, k.VFS(), mntns) -} - -// 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) { - vfsObj := &vfs.VirtualFilesystem{} - fuseDev := &DeviceFD{} - - if err := vfsObj.Init(system.Ctx); err != nil { - return nil, nil, err - } - - vd := vfsObj.NewAnonVirtualDentry("genCountFD") - defer vd.DecRef(system.Ctx) - if err := fuseDev.vfsfd.Init(fuseDev, linux.O_RDWR|linux.O_CREAT, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{}); err != nil { - return nil, nil, err - } - - fsopts := filesystemOptions{ - maxActiveRequests: maxActiveRequests, - } - fs, err := NewFUSEFilesystem(system.Ctx, 0, &fsopts, &fuseDev.vfsfd) - if err != nil { - return nil, nil, err - } - - return fs.conn, &fuseDev.vfsfd, nil -} - -// SizeBytes implements marshal.Marshallable.SizeBytes. -func (t *testPayload) SizeBytes() int { - return 4 -} - -// MarshalBytes implements marshal.Marshallable.MarshalBytes. -func (t *testPayload) MarshalBytes(dst []byte) { - usermem.ByteOrder.PutUint32(dst[:4], t.data) -} - -// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. -func (t *testPayload) UnmarshalBytes(src []byte) { - *t = testPayload{data: usermem.ByteOrder.Uint32(src[:4])} -} - -// Packed implements marshal.Marshallable.Packed. -func (t *testPayload) Packed() bool { - return true -} - -// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. -func (t *testPayload) MarshalUnsafe(dst []byte) { - t.MarshalBytes(dst) -} - -// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. -func (t *testPayload) UnmarshalUnsafe(src []byte) { - t.UnmarshalBytes(src) -} - -// CopyOutN implements marshal.Marshallable.CopyOutN. -func (t *testPayload) CopyOutN(cc marshal.CopyContext, addr usermem.Addr, limit int) (int, error) { - panic("not implemented") -} - -// CopyOut implements marshal.Marshallable.CopyOut. -func (t *testPayload) CopyOut(cc marshal.CopyContext, addr usermem.Addr) (int, error) { - panic("not implemented") -} - -// CopyIn implements marshal.Marshallable.CopyIn. -func (t *testPayload) CopyIn(cc marshal.CopyContext, addr usermem.Addr) (int, error) { - panic("not implemented") -} - -// WriteTo implements io.WriterTo.WriteTo. -func (t *testPayload) WriteTo(w io.Writer) (int64, error) { - panic("not implemented") -} diff --git a/pkg/sentry/fsimpl/fuse/directory.go b/pkg/sentry/fsimpl/fuse/directory.go new file mode 100644 index 000000000..ee77dbaa3 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/directory.go @@ -0,0 +1,105 @@ +// 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 ( + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "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/usermem" +) + +type directoryFD struct { + fileDescription +} + +// Allocate implements directoryFD.Allocate. +func (*directoryFD) Allocate(ctx context.Context, mode, offset, length uint64) error { + return syserror.EISDIR +} + +// PRead implements FileDescriptionImpl.PRead. +func (*directoryFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.EISDIR +} + +// Read implements FileDescriptionImpl.Read. +func (*directoryFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, syserror.EISDIR +} + +// PWrite implements FileDescriptionImpl.PWrite. +func (*directoryFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.EISDIR +} + +// Write implements FileDescriptionImpl.Write. +func (*directoryFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return 0, syserror.EISDIR +} + +// IterDirents implements FileDescriptionImpl.IterDirents. +func (dir *directoryFD) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback) error { + fusefs := dir.inode().fs + task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) + + in := linux.FUSEReadIn{ + Fh: dir.Fh, + Offset: uint64(atomic.LoadInt64(&dir.off)), + Size: linux.FUSE_PAGE_SIZE, + Flags: dir.statusFlags(), + } + + // TODO(gVisor.dev/issue/3404): Support FUSE_READDIRPLUS. + req, err := fusefs.conn.NewRequest(creds, uint32(task.ThreadID()), dir.inode().nodeID, linux.FUSE_READDIR, &in) + if err != nil { + return err + } + + res, err := fusefs.conn.Call(task, req) + if err != nil { + return err + } + if err := res.Error(); err != nil { + return err + } + + var out linux.FUSEDirents + if err := res.UnmarshalPayload(&out); err != nil { + return err + } + + for _, fuseDirent := range out.Dirents { + nextOff := int64(fuseDirent.Meta.Off) + dirent := vfs.Dirent{ + Name: fuseDirent.Name, + Type: uint8(fuseDirent.Meta.Type), + Ino: fuseDirent.Meta.Ino, + NextOff: nextOff, + } + + if err := callback.Handle(dirent); err != nil { + return err + } + atomic.StoreInt64(&dir.off, nextOff) + } + + return nil +} diff --git a/pkg/sentry/fsimpl/fuse/file.go b/pkg/sentry/fsimpl/fuse/file.go new file mode 100644 index 000000000..83f2816b7 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/file.go @@ -0,0 +1,133 @@ +// 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/context" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +// fileDescription implements vfs.FileDescriptionImpl for fuse. +type fileDescription struct { + vfsfd vfs.FileDescription + vfs.FileDescriptionDefaultImpl + vfs.DentryMetadataFileDescriptionImpl + vfs.NoLockFD + + // the file handle used in userspace. + Fh uint64 + + // Nonseekable is indicate cannot perform seek on a file. + Nonseekable bool + + // DirectIO suggest fuse to use direct io operation. + DirectIO bool + + // OpenFlag is the flag returned by open. + OpenFlag uint32 + + // off is the file offset. + off int64 +} + +func (fd *fileDescription) dentry() *kernfs.Dentry { + return fd.vfsfd.Dentry().Impl().(*kernfs.Dentry) +} + +func (fd *fileDescription) inode() *inode { + return fd.dentry().Inode().(*inode) +} + +func (fd *fileDescription) filesystem() *vfs.Filesystem { + return fd.vfsfd.VirtualDentry().Mount().Filesystem() +} + +func (fd *fileDescription) statusFlags() uint32 { + return fd.vfsfd.StatusFlags() +} + +// Release implements vfs.FileDescriptionImpl.Release. +func (fd *fileDescription) Release(ctx context.Context) { + // no need to release if FUSE server doesn't implement Open. + conn := fd.inode().fs.conn + if conn.noOpen { + return + } + + in := linux.FUSEReleaseIn{ + Fh: fd.Fh, + Flags: fd.statusFlags(), + } + // TODO(gvisor.dev/issue/3245): add logic when we support file lock owner. + var opcode linux.FUSEOpcode + if fd.inode().Mode().IsDir() { + opcode = linux.FUSE_RELEASEDIR + } else { + opcode = linux.FUSE_RELEASE + } + kernelTask := kernel.TaskFromContext(ctx) + // ignoring errors and FUSE server reply is analogous to Linux's behavior. + req, err := conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(kernelTask.ThreadID()), fd.inode().nodeID, opcode, &in) + if err != nil { + // No way to invoke Call() with an errored request. + return + } + // The reply will be ignored since no callback is defined in asyncCallBack(). + conn.CallAsync(kernelTask, req) +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *fileDescription) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + return 0, nil +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *fileDescription) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + return 0, nil +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *fileDescription) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + return 0, nil +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *fileDescription) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + return 0, nil +} + +// Seek implements vfs.FileDescriptionImpl.Seek. +func (fd *fileDescription) Seek(ctx context.Context, offset int64, whence int32) (int64, error) { + return 0, nil +} + +// Stat implements vfs.FileDescriptionImpl.Stat. +func (fd *fileDescription) Stat(ctx context.Context, opts vfs.StatOptions) (linux.Statx, error) { + fs := fd.filesystem() + inode := fd.inode() + return inode.Stat(ctx, fs, opts) +} + +// SetStat implements vfs.FileDescriptionImpl.SetStat. +func (fd *fileDescription) SetStat(ctx context.Context, opts vfs.SetStatOptions) error { + fs := fd.filesystem() + creds := auth.CredentialsFromContext(ctx) + return fd.inode().setAttr(ctx, fs, creds, opts, true, fd.Fh) +} diff --git a/pkg/sentry/fsimpl/fuse/fusefs.go b/pkg/sentry/fsimpl/fuse/fusefs.go index 810819ae4..f1ffd2343 100644 --- a/pkg/sentry/fsimpl/fuse/fusefs.go +++ b/pkg/sentry/fsimpl/fuse/fusefs.go @@ -16,21 +16,30 @@ package fuse import ( + "math" "strconv" + "sync" + "sync/atomic" "gvisor.dev/gvisor/pkg/abi/linux" "gvisor.dev/gvisor/pkg/context" "gvisor.dev/gvisor/pkg/log" + "gvisor.dev/gvisor/pkg/marshal" "gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs" "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" ) // Name is the default filesystem name. const Name = "fuse" +// maxActiveRequestsDefault is the default setting controlling the upper bound +// on the number of active requests at any given time. +const maxActiveRequestsDefault = 10000 + // FilesystemType implements vfs.FilesystemType. type FilesystemType struct{} @@ -56,6 +65,11 @@ type filesystemOptions struct { // exist at any time. Any further requests will block when trying to // Call the server. maxActiveRequests uint64 + + // maxRead is the max number of bytes to read, + // specified as "max_read" in fs parameters. + // If not specified by user, use math.MaxUint32 as default value. + maxRead uint32 } // filesystem implements vfs.FilesystemImpl. @@ -69,6 +83,9 @@ type filesystem struct { // opts is the options the fusefs is initialized with. opts *filesystemOptions + + // umounted is true if filesystem.Release() has been called. + umounted bool } // Name implements vfs.FilesystemType.Name. @@ -142,14 +159,29 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt // Set the maxInFlightRequests option. fsopts.maxActiveRequests = maxActiveRequestsDefault + if maxReadStr, ok := mopts["max_read"]; ok { + delete(mopts, "max_read") + maxRead, err := strconv.ParseUint(maxReadStr, 10, 32) + if err != nil { + log.Warningf("%s.GetFilesystem: invalid max_read: max_read=%s", fsType.Name(), maxReadStr) + return nil, nil, syserror.EINVAL + } + if maxRead < fuseMinMaxRead { + maxRead = fuseMinMaxRead + } + fsopts.maxRead = uint32(maxRead) + } else { + fsopts.maxRead = math.MaxUint32 + } + // Check for unparsed options. if len(mopts) != 0 { - log.Warningf("%s.GetFilesystem: unknown options: %v", fsType.Name(), mopts) + log.Warningf("%s.GetFilesystem: unsupported or unknown options: %v", fsType.Name(), mopts) return nil, nil, syserror.EINVAL } // Create a new FUSE filesystem. - fs, err := NewFUSEFilesystem(ctx, devMinor, &fsopts, fuseFd) + fs, err := newFUSEFilesystem(ctx, devMinor, &fsopts, fuseFd) if err != nil { log.Warningf("%s.NewFUSEFilesystem: failed with error: %v", fsType.Name(), err) return nil, nil, err @@ -165,26 +197,27 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt } // root is the fusefs root directory. - root := fs.newInode(creds, fsopts.rootMode) + root := fs.newRootInode(creds, fsopts.rootMode) return fs.VFSFilesystem(), root.VFSDentry(), nil } -// NewFUSEFilesystem creates a new FUSE filesystem. -func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOptions, device *vfs.FileDescription) (*filesystem, error) { - fs := &filesystem{ - devMinor: devMinor, - opts: opts, - } - - conn, err := newFUSEConnection(ctx, device, opts.maxActiveRequests) +// newFUSEFilesystem creates a new FUSE filesystem. +func newFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOptions, device *vfs.FileDescription) (*filesystem, error) { + conn, err := newFUSEConnection(ctx, device, opts) if err != nil { log.Warningf("fuse.NewFUSEFilesystem: NewFUSEConnection failed with error: %v", err) return nil, syserror.EINVAL } - fs.conn = conn fuseFD := device.Impl().(*DeviceFD) + + fs := &filesystem{ + devMinor: devMinor, + opts: opts, + conn: conn, + } + fuseFD.fs = fs return fs, nil @@ -192,6 +225,15 @@ func NewFUSEFilesystem(ctx context.Context, devMinor uint32, opts *filesystemOpt // Release implements vfs.FilesystemImpl.Release. func (fs *filesystem) Release(ctx context.Context) { + fs.conn.fd.mu.Lock() + + fs.umounted = true + fs.conn.Abort(ctx) + // Notify all the waiters on this fd. + fs.conn.fd.waitQueue.Notify(waiter.EventIn) + + fs.conn.fd.mu.Unlock() + fs.Filesystem.VFSFilesystem().VirtualFilesystem().PutAnonBlockDevMinor(fs.devMinor) fs.Filesystem.Release(ctx) } @@ -205,14 +247,49 @@ type inode struct { kernfs.InodeNotSymlink kernfs.OrderedChildren + dentry kernfs.Dentry + + // the owning filesystem. fs is immutable. + fs *filesystem + + // metaDataMu protects the metadata of this inode. + metadataMu sync.Mutex + + nodeID uint64 + locks vfs.FileLocks - dentry kernfs.Dentry + // size of the file. + size uint64 + + // attributeVersion is the version of inode's attributes. + attributeVersion uint64 + + // attributeTime is the remaining vaild time of attributes. + attributeTime uint64 + + // version of the inode. + version uint64 + + // link is result of following a symbolic link. + link string +} + +func (fs *filesystem) newRootInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry { + i := &inode{fs: fs} + i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, 1, linux.ModeDirectory|0755) + i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) + i.dentry.Init(i) + i.nodeID = 1 + + return &i.dentry } -func (fs *filesystem) newInode(creds *auth.Credentials, mode linux.FileMode) *kernfs.Dentry { - i := &inode{} - i.InodeAttrs.Init(creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.ModeDirectory|0755) +func (fs *filesystem) newInode(nodeID uint64, attr linux.FUSEAttr) *kernfs.Dentry { + i := &inode{fs: fs, nodeID: nodeID} + creds := auth.Credentials{EffectiveKGID: auth.KGID(attr.UID), EffectiveKUID: auth.KUID(attr.UID)} + i.InodeAttrs.Init(&creds, linux.UNNAMED_MAJOR, fs.devMinor, fs.NextIno(), linux.FileMode(attr.Mode)) + atomic.StoreUint64(&i.size, attr.Size) i.OrderedChildren.Init(kernfs.OrderedChildrenOptions{}) i.EnableLeakCheck() i.dentry.Init(i) @@ -222,13 +299,301 @@ 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) { - fd, err := kernfs.NewGenericDirectoryFD(rp.Mount(), vfsd, &i.OrderedChildren, &i.locks, &opts, kernfs.GenericDirectoryFDOptions{ - SeekEnd: kernfs.SeekEndStaticEntries, - }) + isDir := i.InodeAttrs.Mode().IsDir() + // return error if specified to open directory but inode is not a directory. + if !isDir && opts.Mode.IsDir() { + return nil, syserror.ENOTDIR + } + if opts.Flags&linux.O_LARGEFILE == 0 && atomic.LoadUint64(&i.size) > linux.MAX_NON_LFS { + return nil, syserror.EOVERFLOW + } + + var fd *fileDescription + var fdImpl vfs.FileDescriptionImpl + if isDir { + directoryFD := &directoryFD{} + fd = &(directoryFD.fileDescription) + fdImpl = directoryFD + } else { + regularFD := ®ularFileFD{} + fd = &(regularFD.fileDescription) + fdImpl = regularFD + } + // FOPEN_KEEP_CACHE is the defualt flag for noOpen. + fd.OpenFlag = linux.FOPEN_KEEP_CACHE + + // Only send open request when FUSE server support open or is opening a directory. + if !i.fs.conn.noOpen || isDir { + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("fusefs.Inode.Open: couldn't get kernel task from context") + return nil, syserror.EINVAL + } + + // Build the request. + var opcode linux.FUSEOpcode + if isDir { + opcode = linux.FUSE_OPENDIR + } else { + opcode = linux.FUSE_OPEN + } + + in := linux.FUSEOpenIn{Flags: opts.Flags & ^uint32(linux.O_CREAT|linux.O_EXCL|linux.O_NOCTTY)} + if !i.fs.conn.atomicOTrunc { + in.Flags &= ^uint32(linux.O_TRUNC) + } + + req, err := i.fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(kernelTask.ThreadID()), i.nodeID, opcode, &in) + if err != nil { + return nil, err + } + + // Send the request and receive the reply. + res, err := i.fs.conn.Call(kernelTask, req) + if err != nil { + return nil, err + } + if err := res.Error(); err == syserror.ENOSYS && !isDir { + i.fs.conn.noOpen = true + } else if err != nil { + return nil, err + } else { + out := linux.FUSEOpenOut{} + if err := res.UnmarshalPayload(&out); err != nil { + return nil, err + } + + // Process the reply. + fd.OpenFlag = out.OpenFlag + if isDir { + fd.OpenFlag &= ^uint32(linux.FOPEN_DIRECT_IO) + } + + fd.Fh = out.Fh + } + } + + // TODO(gvisor.dev/issue/3234): invalidate mmap after implemented it for FUSE Inode + fd.DirectIO = fd.OpenFlag&linux.FOPEN_DIRECT_IO != 0 + fdOptions := &vfs.FileDescriptionOptions{} + if fd.OpenFlag&linux.FOPEN_NONSEEKABLE != 0 { + fdOptions.DenyPRead = true + fdOptions.DenyPWrite = true + fd.Nonseekable = true + } + + // If we don't send SETATTR before open (which is indicated by atomicOTrunc) + // and O_TRUNC is set, update the inode's version number and clean existing data + // by setting the file size to 0. + if i.fs.conn.atomicOTrunc && opts.Flags&linux.O_TRUNC != 0 { + i.fs.conn.mu.Lock() + i.fs.conn.attributeVersion++ + i.attributeVersion = i.fs.conn.attributeVersion + atomic.StoreUint64(&i.size, 0) + i.fs.conn.mu.Unlock() + i.attributeTime = 0 + } + + if err := fd.vfsfd.Init(fdImpl, opts.Flags, rp.Mount(), vfsd, fdOptions); err != nil { + return nil, err + } + return &fd.vfsfd, nil +} + +// Lookup implements kernfs.Inode.Lookup. +func (i *inode) Lookup(ctx context.Context, name string) (*vfs.Dentry, error) { + in := linux.FUSELookupIn{Name: name} + return i.newEntry(ctx, name, 0, linux.FUSE_LOOKUP, &in) +} + +// IterDirents implements kernfs.Inode.IterDirents. +func (*inode) IterDirents(ctx context.Context, callback vfs.IterDirentsCallback, offset, relOffset int64) (int64, error) { + return offset, nil +} + +// Valid implements kernfs.Inode.Valid. +func (*inode) Valid(ctx context.Context) bool { + return true +} + +// NewFile implements kernfs.Inode.NewFile. +func (i *inode) NewFile(ctx context.Context, name string, opts vfs.OpenOptions) (*vfs.Dentry, error) { + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("fusefs.Inode.NewFile: couldn't get kernel task from context", i.nodeID) + return nil, syserror.EINVAL + } + in := linux.FUSECreateIn{ + CreateMeta: linux.FUSECreateMeta{ + Flags: opts.Flags, + Mode: uint32(opts.Mode) | linux.S_IFREG, + Umask: uint32(kernelTask.FSContext().Umask()), + }, + Name: name, + } + return i.newEntry(ctx, name, linux.S_IFREG, linux.FUSE_CREATE, &in) +} + +// NewNode implements kernfs.Inode.NewNode. +func (i *inode) NewNode(ctx context.Context, name string, opts vfs.MknodOptions) (*vfs.Dentry, error) { + in := linux.FUSEMknodIn{ + MknodMeta: linux.FUSEMknodMeta{ + Mode: uint32(opts.Mode), + Rdev: linux.MakeDeviceID(uint16(opts.DevMajor), opts.DevMinor), + Umask: uint32(kernel.TaskFromContext(ctx).FSContext().Umask()), + }, + Name: name, + } + return i.newEntry(ctx, name, opts.Mode.FileType(), linux.FUSE_MKNOD, &in) +} + +// NewSymlink implements kernfs.Inode.NewSymlink. +func (i *inode) NewSymlink(ctx context.Context, name, target string) (*vfs.Dentry, error) { + in := linux.FUSESymLinkIn{ + Name: name, + Target: target, + } + return i.newEntry(ctx, name, linux.S_IFLNK, linux.FUSE_SYMLINK, &in) +} + +// Unlink implements kernfs.Inode.Unlink. +func (i *inode) Unlink(ctx context.Context, name string, child *vfs.Dentry) error { + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("fusefs.Inode.newEntry: couldn't get kernel task from context", i.nodeID) + return syserror.EINVAL + } + in := linux.FUSEUnlinkIn{Name: name} + req, err := i.fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(kernelTask.ThreadID()), i.nodeID, linux.FUSE_UNLINK, &in) + if err != nil { + return err + } + res, err := i.fs.conn.Call(kernelTask, req) + if err != nil { + return err + } + // only return error, discard res. + if err := res.Error(); err != nil { + return err + } + return i.dentry.RemoveChildLocked(name, child) +} + +// NewDir implements kernfs.Inode.NewDir. +func (i *inode) NewDir(ctx context.Context, name string, opts vfs.MkdirOptions) (*vfs.Dentry, error) { + in := linux.FUSEMkdirIn{ + MkdirMeta: linux.FUSEMkdirMeta{ + Mode: uint32(opts.Mode), + Umask: uint32(kernel.TaskFromContext(ctx).FSContext().Umask()), + }, + Name: name, + } + return i.newEntry(ctx, name, linux.S_IFDIR, linux.FUSE_MKDIR, &in) +} + +// RmDir implements kernfs.Inode.RmDir. +func (i *inode) RmDir(ctx context.Context, name string, child *vfs.Dentry) error { + fusefs := i.fs + task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) + + in := linux.FUSERmDirIn{Name: name} + req, err := fusefs.conn.NewRequest(creds, uint32(task.ThreadID()), i.nodeID, linux.FUSE_RMDIR, &in) + if err != nil { + return err + } + + res, err := i.fs.conn.Call(task, req) + if err != nil { + return err + } + if err := res.Error(); err != nil { + return err + } + + // TODO(Before merging): When creating new nodes, should we add nodes to the ordered children? + // If so we'll probably need to call this. We will also need to add them with the writable flag when + // appropriate. + // return i.OrderedChildren.RmDir(ctx, name, child) + + return nil +} + +// newEntry calls FUSE server for entry creation and allocates corresponding entry according to response. +// Shared by FUSE_MKNOD, FUSE_MKDIR, FUSE_SYMLINK, FUSE_LINK and FUSE_LOOKUP. +func (i *inode) newEntry(ctx context.Context, name string, fileType linux.FileMode, opcode linux.FUSEOpcode, payload marshal.Marshallable) (*vfs.Dentry, error) { + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("fusefs.Inode.newEntry: couldn't get kernel task from context", i.nodeID) + return nil, syserror.EINVAL + } + req, err := i.fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(kernelTask.ThreadID()), i.nodeID, opcode, payload) + if err != nil { + return nil, err + } + res, err := i.fs.conn.Call(kernelTask, req) if err != nil { return nil, err } - return fd.VFSFileDescription(), nil + if err := res.Error(); err != nil { + return nil, err + } + out := linux.FUSEEntryOut{} + if err := res.UnmarshalPayload(&out); err != nil { + return nil, err + } + if opcode != linux.FUSE_LOOKUP && ((out.Attr.Mode&linux.S_IFMT)^uint32(fileType) != 0 || out.NodeID == 0 || out.NodeID == linux.FUSE_ROOT_ID) { + return nil, syserror.EIO + } + child := i.fs.newInode(out.NodeID, out.Attr) + if opcode == linux.FUSE_LOOKUP { + i.dentry.InsertChildLocked(name, child) + } else { + i.dentry.InsertChild(name, child) + } + return child.VFSDentry(), nil +} + +// Getlink implements kernfs.Inode.Getlink. +func (i *inode) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) { + path, err := i.Readlink(ctx, mnt) + return vfs.VirtualDentry{}, path, err +} + +// Readlink implements kernfs.Inode.Readlink. +func (i *inode) Readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { + if i.Mode().FileType()&linux.S_IFLNK == 0 { + return "", syserror.EINVAL + } + if len(i.link) == 0 { + kernelTask := kernel.TaskFromContext(ctx) + if kernelTask == nil { + log.Warningf("fusefs.Inode.Readlink: couldn't get kernel task from context") + return "", syserror.EINVAL + } + req, err := i.fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(kernelTask.ThreadID()), i.nodeID, linux.FUSE_READLINK, &linux.FUSEEmptyIn{}) + if err != nil { + return "", err + } + res, err := i.fs.conn.Call(kernelTask, req) + if err != nil { + return "", err + } + i.link = string(res.data[res.hdr.SizeBytes():]) + if !mnt.Options().ReadOnly { + i.attributeTime = 0 + } + } + return i.link, nil +} + +// getFUSEAttr returns a linux.FUSEAttr of this inode stored in local cache. +// TODO(gvisor.dev/issue/3679): Add support for other fields. +func (i *inode) getFUSEAttr() linux.FUSEAttr { + return linux.FUSEAttr{ + Ino: i.Ino(), + Size: atomic.LoadUint64(&i.size), + Mode: uint32(i.Mode()), + } } // statFromFUSEAttr makes attributes from linux.FUSEAttr to linux.Statx. The @@ -284,47 +649,89 @@ func statFromFUSEAttr(attr linux.FUSEAttr, mask, devMinor uint32) linux.Statx { return stat } -// Stat implements kernfs.Inode.Stat. -func (i *inode) Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { - fusefs := fs.Impl().(*filesystem) - conn := fusefs.conn - task, creds := kernel.TaskFromContext(ctx), auth.CredentialsFromContext(ctx) +// getAttr gets the attribute of this inode by issuing a FUSE_GETATTR request +// or read from local cache. It updates the corresponding attributes if +// necessary. +func (i *inode) getAttr(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions, flags uint32, fh uint64) (linux.FUSEAttr, error) { + attributeVersion := atomic.LoadUint64(&i.fs.conn.attributeVersion) + + // TODO(gvisor.dev/issue/3679): send the request only if + // - invalid local cache for fields specified in the opts.Mask + // - forced update + // - i.attributeTime expired + // If local cache is still valid, return local cache. + // Currently we always send a request, + // and we always set the metadata with the new result, + // unless attributeVersion has changed. + + task := kernel.TaskFromContext(ctx) if task == nil { log.Warningf("couldn't get kernel task from context") - return linux.Statx{}, syserror.EINVAL + return linux.FUSEAttr{}, syserror.EINVAL } - var in linux.FUSEGetAttrIn - // We don't set any attribute in the request, because in VFS2 fstat(2) will - // finally be translated into vfs.FilesystemImpl.StatAt() (see - // pkg/sentry/syscalls/linux/vfs2/stat.go), resulting in the same flow - // as stat(2). Thus GetAttrFlags and Fh variable will never be used in VFS2. - req, err := conn.NewRequest(creds, uint32(task.ThreadID()), i.Ino(), linux.FUSE_GETATTR, &in) + creds := auth.CredentialsFromContext(ctx) + + in := linux.FUSEGetAttrIn{ + GetAttrFlags: flags, + Fh: fh, + } + req, err := i.fs.conn.NewRequest(creds, uint32(task.ThreadID()), i.nodeID, linux.FUSE_GETATTR, &in) if err != nil { - return linux.Statx{}, err + return linux.FUSEAttr{}, err } - res, err := conn.Call(task, req) + res, err := i.fs.conn.Call(task, req) if err != nil { - return linux.Statx{}, err + return linux.FUSEAttr{}, err } if err := res.Error(); err != nil { - return linux.Statx{}, err + return linux.FUSEAttr{}, err } var out linux.FUSEGetAttrOut if err := res.UnmarshalPayload(&out); err != nil { - return linux.Statx{}, err + return linux.FUSEAttr{}, err + } + + // Local version is newer, return the local one. + // Skip the update. + if attributeVersion != 0 && atomic.LoadUint64(&i.attributeVersion) > attributeVersion { + return i.getFUSEAttr(), nil } - // Set all metadata into kernfs.InodeAttrs. - if err := i.SetStat(ctx, fs, creds, vfs.SetStatOptions{ - Stat: statFromFUSEAttr(out.Attr, linux.STATX_ALL, fusefs.devMinor), + // Set the metadata of kernfs.InodeAttrs. + if err := i.SetInodeStat(ctx, fs, creds, vfs.SetStatOptions{ + Stat: statFromFUSEAttr(out.Attr, linux.STATX_ALL, i.fs.devMinor), }); err != nil { + return linux.FUSEAttr{}, err + } + + // Set the size if no error (after SetStat() check). + atomic.StoreUint64(&i.size, out.Attr.Size) + + return out.Attr, nil +} + +// reviseAttr attempts to update the attributes for internal purposes +// by calling getAttr with a pre-specified mask. +// Used by read, write, lseek. +func (i *inode) reviseAttr(ctx context.Context, flags uint32, fh uint64) error { + // Never need atime for internal purposes. + _, err := i.getAttr(ctx, i.fs.VFSFilesystem(), vfs.StatOptions{ + Mask: linux.STATX_BASIC_STATS &^ linux.STATX_ATIME, + }, flags, fh) + return err +} + +// Stat implements kernfs.Inode.Stat. +func (i *inode) Stat(ctx context.Context, fs *vfs.Filesystem, opts vfs.StatOptions) (linux.Statx, error) { + attr, err := i.getAttr(ctx, fs, opts, 0, 0) + if err != nil { return linux.Statx{}, err } - return statFromFUSEAttr(out.Attr, opts.Mask, fusefs.devMinor), nil + return statFromFUSEAttr(attr, opts.Mask, i.fs.devMinor), nil } // DecRef implements kernfs.Inode. @@ -337,3 +744,84 @@ func (i *inode) StatFS(ctx context.Context, fs *vfs.Filesystem) (linux.Statfs, e // TODO(gvisor.dev/issues/3413): Complete the implementation of statfs. return vfs.GenericStatFS(linux.FUSE_SUPER_MAGIC), nil } + +// fattrMaskFromStats converts vfs.SetStatOptions.Stat.Mask to linux stats mask +// aligned with the attribute mask defined in include/linux/fs.h. +func fattrMaskFromStats(mask uint32) uint32 { + var fuseAttrMask uint32 + maskMap := map[uint32]uint32{ + linux.STATX_MODE: linux.FATTR_MODE, + linux.STATX_UID: linux.FATTR_UID, + linux.STATX_GID: linux.FATTR_GID, + linux.STATX_SIZE: linux.FATTR_SIZE, + linux.STATX_ATIME: linux.FATTR_ATIME, + linux.STATX_MTIME: linux.FATTR_MTIME, + linux.STATX_CTIME: linux.FATTR_CTIME, + } + for statxMask, fattrMask := range maskMap { + if mask&statxMask != 0 { + fuseAttrMask |= fattrMask + } + } + return fuseAttrMask +} + +// SetStat implements kernfs.Inode.SetStat. +func (i *inode) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + return i.setAttr(ctx, fs, creds, opts, false, 0) +} + +func (i *inode) setAttr(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions, useFh bool, fh uint64) error { + conn := i.fs.conn + task := kernel.TaskFromContext(ctx) + if task == nil { + log.Warningf("couldn't get kernel task from context") + return syserror.EINVAL + } + + // We should retain the original file type when assigning new mode. + fileType := uint16(i.Mode()) & linux.S_IFMT + fattrMask := fattrMaskFromStats(opts.Stat.Mask) + if useFh { + fattrMask |= linux.FATTR_FH + } + in := linux.FUSESetAttrIn{ + Valid: fattrMask, + Fh: fh, + Size: opts.Stat.Size, + Atime: uint64(opts.Stat.Atime.Sec), + Mtime: uint64(opts.Stat.Mtime.Sec), + Ctime: uint64(opts.Stat.Ctime.Sec), + AtimeNsec: opts.Stat.Atime.Nsec, + MtimeNsec: opts.Stat.Mtime.Nsec, + CtimeNsec: opts.Stat.Ctime.Nsec, + Mode: uint32(fileType | opts.Stat.Mode), + UID: opts.Stat.UID, + GID: opts.Stat.GID, + } + req, err := conn.NewRequest(creds, uint32(task.ThreadID()), i.nodeID, linux.FUSE_SETATTR, &in) + if err != nil { + return err + } + + res, err := conn.Call(task, req) + if err != nil { + return err + } + if err := res.Error(); err != nil { + return err + } + out := linux.FUSEGetAttrOut{} + if err := res.UnmarshalPayload(&out); err != nil { + return err + } + + // Set the metadata of kernfs.InodeAttrs. + if err := i.SetInodeStat(ctx, fs, creds, vfs.SetStatOptions{ + Stat: statFromFUSEAttr(out.Attr, linux.STATX_ALL, i.fs.devMinor), + }); err != nil { + return err + } + + return nil +} diff --git a/pkg/sentry/fsimpl/fuse/read_write.go b/pkg/sentry/fsimpl/fuse/read_write.go new file mode 100644 index 000000000..625d1547f --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/read_write.go @@ -0,0 +1,242 @@ +// 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 ( + "io" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "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/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +// ReadInPages sends FUSE_READ requests for the size after round it up to +// a multiple of page size, blocks on it for reply, processes the reply +// and returns the payload (or joined payloads) as a byte slice. +// This is used for the general purpose reading. +// We do not support direct IO (which read the exact number of bytes) +// at this moment. +func (fs *filesystem) ReadInPages(ctx context.Context, fd *regularFileFD, off uint64, size uint32) ([][]byte, uint32, error) { + attributeVersion := atomic.LoadUint64(&fs.conn.attributeVersion) + + t := kernel.TaskFromContext(ctx) + if t == nil { + log.Warningf("fusefs.Read: couldn't get kernel task from context") + return nil, 0, syserror.EINVAL + } + + // Round up to a multiple of page size. + readSize, _ := usermem.PageRoundUp(uint64(size)) + + // One request cannnot exceed either maxRead or maxPages. + maxPages := fs.conn.maxRead >> usermem.PageShift + if maxPages > uint32(fs.conn.maxPages) { + maxPages = uint32(fs.conn.maxPages) + } + + var outs [][]byte + var sizeRead uint32 + + // readSize is a multiple of usermem.PageSize. + // Always request bytes as a multiple of pages. + pagesRead, pagesToRead := uint32(0), uint32(readSize>>usermem.PageShift) + + // Reuse the same struct for unmarshalling to avoid unnecessary memory allocation. + in := linux.FUSEReadIn{ + Fh: fd.Fh, + LockOwner: 0, // TODO(gvisor.dev/issue/3245): file lock + ReadFlags: 0, // TODO(gvisor.dev/issue/3245): |= linux.FUSE_READ_LOCKOWNER + Flags: fd.statusFlags(), + } + + // This loop is intended for fragmented read where the bytes to read is + // larger than either the maxPages or maxRead. + // For the majority of reads with normal size, this loop should only + // execute once. + for pagesRead < pagesToRead { + pagesCanRead := pagesToRead - pagesRead + if pagesCanRead > maxPages { + pagesCanRead = maxPages + } + + in.Offset = off + (uint64(pagesRead) << usermem.PageShift) + in.Size = pagesCanRead << usermem.PageShift + + req, err := fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(t.ThreadID()), fd.inode().nodeID, linux.FUSE_READ, &in) + if err != nil { + return nil, 0, err + } + + // TODO(gvisor.dev/issue/3247): support async read. + + res, err := fs.conn.Call(t, req) + if err != nil { + return nil, 0, err + } + if err := res.Error(); err != nil { + return nil, 0, err + } + + // Not enough bytes in response, + // either we reached EOF, + // or the FUSE server sends back a response + // that cannot even fit the hdr. + if len(res.data) <= res.hdr.SizeBytes() { + // We treat both case as EOF here for now + // since there is no reliable way to detect + // the over-short hdr case. + break + } + + // Directly using the slice to avoid extra copy. + out := res.data[res.hdr.SizeBytes():] + + outs = append(outs, out) + sizeRead += uint32(len(out)) + + pagesRead += pagesCanRead + } + + defer fs.ReadCallback(ctx, fd, off, size, sizeRead, attributeVersion) + + // No bytes returned: offset >= EOF. + if len(outs) == 0 { + return nil, 0, io.EOF + } + + return outs, sizeRead, nil +} + +// ReadCallback updates several information after receiving a read response. +// Due to readahead, sizeRead can be larger than size. +func (fs *filesystem) ReadCallback(ctx context.Context, fd *regularFileFD, off uint64, size uint32, sizeRead uint32, attributeVersion uint64) { + // TODO(gvisor.dev/issue/3247): support async read. + // If this is called by an async read, correctly process it. + // May need to update the signature. + + i := fd.inode() + // TODO(gvisor.dev/issue/1193): Invalidate or update atime. + + // Reached EOF. + if sizeRead < size { + // TODO(gvisor.dev/issue/3630): If we have writeback cache, then we need to fill this hole. + // Might need to update the buf to be returned from the Read(). + + // Update existing size. + newSize := off + uint64(sizeRead) + fs.conn.mu.Lock() + if attributeVersion == i.attributeVersion && newSize < atomic.LoadUint64(&i.size) { + fs.conn.attributeVersion++ + i.attributeVersion = i.fs.conn.attributeVersion + atomic.StoreUint64(&i.size, newSize) + } + fs.conn.mu.Unlock() + } +} + +// Write sends FUSE_WRITE requests and return the bytes +// written according to the response. +// +// Preconditions: len(data) == size. +func (fs *filesystem) Write(ctx context.Context, fd *regularFileFD, off uint64, size uint32, data []byte) (uint32, error) { + t := kernel.TaskFromContext(ctx) + if t == nil { + log.Warningf("fusefs.Read: couldn't get kernel task from context") + return 0, syserror.EINVAL + } + + // One request cannnot exceed either maxWrite or maxPages. + maxWrite := uint32(fs.conn.maxPages) << usermem.PageShift + if maxWrite > fs.conn.maxWrite { + maxWrite = fs.conn.maxWrite + } + + // Reuse the same struct for unmarshalling to avoid unnecessary memory allocation. + in := linux.FUSEWriteIn{ + Fh: fd.Fh, + // TODO(gvisor.dev/issue/3245): file lock + LockOwner: 0, + // TODO(gvisor.dev/issue/3245): |= linux.FUSE_READ_LOCKOWNER + // TODO(gvisor.dev/issue/3237): |= linux.FUSE_WRITE_CACHE (not added yet) + WriteFlags: 0, + Flags: fd.statusFlags(), + } + + var written uint32 + + // This loop is intended for fragmented write where the bytes to write is + // larger than either the maxWrite or maxPages or when bigWrites is false. + // Unless a small value for max_write is explicitly used, this loop + // is expected to execute only once for the majority of the writes. + for written < size { + toWrite := size - written + + // Limit the write size to one page. + // Note that the bigWrites flag is obsolete, + // latest libfuse always sets it on. + if !fs.conn.bigWrites && toWrite > usermem.PageSize { + toWrite = usermem.PageSize + } + + // Limit the write size to maxWrite. + if toWrite > maxWrite { + toWrite = maxWrite + } + + in.Offset = off + uint64(written) + in.Size = toWrite + + req, err := fs.conn.NewRequest(auth.CredentialsFromContext(ctx), uint32(t.ThreadID()), fd.inode().nodeID, linux.FUSE_WRITE, &in) + if err != nil { + return 0, err + } + + req.payload = data[written : written+toWrite] + + // TODO(gvisor.dev/issue/3247): support async write. + + res, err := fs.conn.Call(t, req) + if err != nil { + return 0, err + } + if err := res.Error(); err != nil { + return 0, err + } + + out := linux.FUSEWriteOut{} + if err := res.UnmarshalPayload(&out); err != nil { + return 0, err + } + + // Write more than requested? EIO. + if out.Size > toWrite { + return 0, syserror.EIO + } + + written += out.Size + + // Break if short write. Not necessarily an error. + if out.Size != toWrite { + break + } + } + + return written, nil +} diff --git a/pkg/sentry/fsimpl/fuse/regular_file.go b/pkg/sentry/fsimpl/fuse/regular_file.go new file mode 100644 index 000000000..5bdd096c3 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/regular_file.go @@ -0,0 +1,230 @@ +// 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 ( + "io" + "math" + "sync" + "sync/atomic" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/context" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/syserror" + "gvisor.dev/gvisor/pkg/usermem" +) + +type regularFileFD struct { + fileDescription + + // off is the file offset. + off int64 + // offMu protects off. + offMu sync.Mutex +} + +// PRead implements vfs.FileDescriptionImpl.PRead. +func (fd *regularFileFD) PRead(ctx context.Context, dst usermem.IOSequence, offset int64, opts vfs.ReadOptions) (int64, error) { + if offset < 0 { + return 0, syserror.EINVAL + } + + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, syserror.EOPNOTSUPP + } + + size := dst.NumBytes() + if size == 0 { + // Early return if count is 0. + return 0, nil + } else if size > math.MaxUint32 { + // FUSE only supports uint32 for size. + // Overflow. + return 0, syserror.EINVAL + } + + // TODO(gvisor.dev/issue/3678): Add direct IO support. + + inode := fd.inode() + + // Reading beyond EOF, update file size if outdated. + if uint64(offset+size) > atomic.LoadUint64(&inode.size) { + if err := inode.reviseAttr(ctx, linux.FUSE_GETATTR_FH, fd.Fh); err != nil { + return 0, err + } + // If the offset after update is still too large, return error. + if uint64(offset) >= atomic.LoadUint64(&inode.size) { + return 0, io.EOF + } + } + + // Truncate the read with updated file size. + fileSize := atomic.LoadUint64(&inode.size) + if uint64(offset+size) > fileSize { + size = int64(fileSize) - offset + } + + buffers, n, err := inode.fs.ReadInPages(ctx, fd, uint64(offset), uint32(size)) + if err != nil { + return 0, err + } + + // TODO(gvisor.dev/issue/3237): support indirect IO (e.g. caching), + // store the bytes that were read ahead. + + // Update the number of bytes to copy for short read. + if n < uint32(size) { + size = int64(n) + } + + // Copy the bytes read to the dst. + // This loop is intended for fragmented reads. + // For the majority of reads, this loop only execute once. + var copied int64 + for _, buffer := range buffers { + toCopy := int64(len(buffer)) + if copied+toCopy > size { + toCopy = size - copied + } + cp, err := dst.DropFirst64(copied).CopyOut(ctx, buffer[:toCopy]) + if err != nil { + return 0, err + } + if int64(cp) != toCopy { + return 0, syserror.EIO + } + copied += toCopy + } + + return copied, nil +} + +// Read implements vfs.FileDescriptionImpl.Read. +func (fd *regularFileFD) Read(ctx context.Context, dst usermem.IOSequence, opts vfs.ReadOptions) (int64, error) { + fd.offMu.Lock() + n, err := fd.PRead(ctx, dst, fd.off, opts) + fd.off += n + fd.offMu.Unlock() + return n, err +} + +// PWrite implements vfs.FileDescriptionImpl.PWrite. +func (fd *regularFileFD) PWrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (int64, error) { + n, _, err := fd.pwrite(ctx, src, offset, opts) + return n, err +} + +// Write implements vfs.FileDescriptionImpl.Write. +func (fd *regularFileFD) Write(ctx context.Context, src usermem.IOSequence, opts vfs.WriteOptions) (int64, error) { + fd.offMu.Lock() + n, off, err := fd.pwrite(ctx, src, fd.off, opts) + fd.off = off + fd.offMu.Unlock() + return n, err +} + +// pwrite returns the number of bytes written, final offset and error. The +// final offset should be ignored by PWrite. +func (fd *regularFileFD) pwrite(ctx context.Context, src usermem.IOSequence, offset int64, opts vfs.WriteOptions) (written, finalOff int64, err error) { + if offset < 0 { + return 0, offset, syserror.EINVAL + } + + // Check that flags are supported. + // + // TODO(gvisor.dev/issue/2601): Support select preadv2 flags. + if opts.Flags&^linux.RWF_HIPRI != 0 { + return 0, offset, syserror.EOPNOTSUPP + } + + inode := fd.inode() + inode.metadataMu.Lock() + defer inode.metadataMu.Unlock() + + // If the file is opened with O_APPEND, update offset to file size. + // Note: since our Open() implements the interface of kernfs, + // and kernfs currently does not support O_APPEND, this will never + // be true before we switch out from kernfs. + if fd.vfsfd.StatusFlags()&linux.O_APPEND != 0 { + // Locking inode.metadataMu is sufficient for reading size + offset = int64(inode.size) + } + + srclen := src.NumBytes() + + if srclen > math.MaxUint32 { + // FUSE only supports uint32 for size. + // Overflow. + return 0, offset, syserror.EINVAL + } + if end := offset + srclen; end < offset { + // Overflow. + return 0, offset, syserror.EINVAL + } + + srclen, err = vfs.CheckLimit(ctx, offset, srclen) + if err != nil { + return 0, offset, err + } + + if srclen == 0 { + // Return before causing any side effects. + return 0, offset, nil + } + + src = src.TakeFirst64(srclen) + + // TODO(gvisor.dev/issue/3237): Add cache support: + // buffer cache. Ideally we write from src to our buffer cache first. + // The slice passed to fs.Write() should be a slice from buffer cache. + data := make([]byte, srclen) + // Reason for making a copy here: connection.Call() blocks on kerneltask, + // which in turn acquires mm.activeMu lock. Functions like CopyInTo() will + // attemp to acquire the mm.activeMu lock as well -> deadlock. + // We must finish reading from the userspace memory before + // t.Block() deactivates it. + cp, err := src.CopyIn(ctx, data) + if err != nil { + return 0, offset, err + } + if int64(cp) != srclen { + return 0, offset, syserror.EIO + } + + n, err := fd.inode().fs.Write(ctx, fd, uint64(offset), uint32(srclen), data) + if err != nil { + return 0, offset, err + } + + if n == 0 { + // We have checked srclen != 0 previously. + // If err == nil, then it's a short write and we return EIO. + return 0, offset, syserror.EIO + } + + written = int64(n) + finalOff = offset + written + + if finalOff > int64(inode.size) { + atomic.StoreUint64(&inode.size, uint64(finalOff)) + atomic.AddUint64(&inode.fs.conn.attributeVersion, 1) + } + + return +} diff --git a/pkg/sentry/fsimpl/fuse/request_response.go b/pkg/sentry/fsimpl/fuse/request_response.go new file mode 100644 index 000000000..7fa00569b --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/request_response.go @@ -0,0 +1,229 @@ +// 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 ( + "fmt" + "syscall" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/usermem" +) + +// fuseInitRes is a variable-length wrapper of linux.FUSEInitOut. The FUSE +// server may implement an older version of FUSE protocol, which contains a +// linux.FUSEInitOut with less attributes. +// +// Dynamically-sized objects cannot be marshalled. +type fuseInitRes struct { + marshal.StubMarshallable + + // initOut contains the response from the FUSE server. + initOut linux.FUSEInitOut + + // initLen is the total length of bytes of the response. + initLen uint32 +} + +// UnmarshalBytes deserializes src to the initOut attribute in a fuseInitRes. +func (r *fuseInitRes) UnmarshalBytes(src []byte) { + out := &r.initOut + + // Introduced before FUSE kernel version 7.13. + out.Major = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + out.Minor = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + out.MaxReadahead = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + out.Flags = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + out.MaxBackground = uint16(usermem.ByteOrder.Uint16(src[:2])) + src = src[2:] + out.CongestionThreshold = uint16(usermem.ByteOrder.Uint16(src[:2])) + src = src[2:] + out.MaxWrite = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + + // Introduced in FUSE kernel version 7.23. + if len(src) >= 4 { + out.TimeGran = uint32(usermem.ByteOrder.Uint32(src[:4])) + src = src[4:] + } + // Introduced in FUSE kernel version 7.28. + if len(src) >= 2 { + out.MaxPages = uint16(usermem.ByteOrder.Uint16(src[:2])) + src = src[2:] + } +} + +// SizeBytes is the size of the payload of the FUSE_INIT response. +func (r *fuseInitRes) SizeBytes() int { + return int(r.initLen) +} + +// Ordinary requests have even IDs, while interrupts IDs are odd. +// Used to increment the unique ID for each FUSE request. +var reqIDStep uint64 = 2 + +// Request represents a FUSE operation request that hasn't been sent to the +// server yet. +// +// +stateify savable +type Request struct { + requestEntry + + id linux.FUSEOpID + hdr *linux.FUSEHeaderIn + data []byte + + // payload for this request: extra bytes to write after + // the data slice. Used by FUSE_WRITE. + payload []byte + + // If this request is async. + async bool + // If we don't care its response. + // Manually set by the caller. + noReply bool +} + +// 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) { + conn.fd.mu.Lock() + defer conn.fd.mu.Unlock() + conn.fd.nextOpID += linux.FUSEOpID(reqIDStep) + + hdrLen := (*linux.FUSEHeaderIn)(nil).SizeBytes() + hdr := linux.FUSEHeaderIn{ + Len: uint32(hdrLen + payload.SizeBytes()), + Opcode: opcode, + Unique: conn.fd.nextOpID, + NodeID: ino, + UID: uint32(creds.EffectiveKUID), + GID: uint32(creds.EffectiveKGID), + PID: pid, + } + + buf := make([]byte, hdr.Len) + + // TODO(gVisor.dev/issue/3698): Use the unsafe version once go_marshal is safe to use again. + hdr.MarshalBytes(buf[:hdrLen]) + payload.MarshalBytes(buf[hdrLen:]) + + return &Request{ + id: hdr.Unique, + hdr: &hdr, + data: buf, + }, nil +} + +// futureResponse represents an in-flight request, that may or may not have +// completed yet. Convert it to a resolved Response by calling Resolve, but note +// that this may block. +// +// +stateify savable +type futureResponse struct { + opcode linux.FUSEOpcode + ch chan struct{} + hdr *linux.FUSEHeaderOut + data []byte + + // If this request is async. + async bool +} + +// newFutureResponse creates a future response to a FUSE request. +func newFutureResponse(req *Request) *futureResponse { + return &futureResponse{ + opcode: req.hdr.Opcode, + ch: make(chan struct{}), + async: req.async, + } +} + +// resolve blocks the task until the server responds to its corresponding request, +// then returns a resolved response. +func (f *futureResponse) resolve(t *kernel.Task) (*Response, error) { + // Return directly for async requests. + if f.async { + return nil, nil + } + + if err := t.Block(f.ch); err != nil { + return nil, err + } + + return f.getResponse(), nil +} + +// getResponse creates a Response from the data the futureResponse has. +func (f *futureResponse) getResponse() *Response { + return &Response{ + opcode: f.opcode, + hdr: *f.hdr, + data: f.data, + } +} + +// Response represents an actual response from the server, including the +// response payload. +// +// +stateify savable +type Response struct { + opcode linux.FUSEOpcode + hdr linux.FUSEHeaderOut + data []byte +} + +// Error returns the error of the FUSE call. +func (r *Response) Error() error { + errno := r.hdr.Error + if errno >= 0 { + return nil + } + + sysErrNo := syscall.Errno(-errno) + return error(sysErrNo) +} + +// DataLen returns the size of the response without the header. +func (r *Response) DataLen() uint32 { + return r.hdr.Len - uint32(r.hdr.SizeBytes()) +} + +// UnmarshalPayload unmarshals the response data into m. +func (r *Response) UnmarshalPayload(m marshal.Marshallable) error { + hdrLen := r.hdr.SizeBytes() + haveDataLen := r.hdr.Len - uint32(hdrLen) + wantDataLen := uint32(m.SizeBytes()) + + if haveDataLen < wantDataLen { + return fmt.Errorf("payload too small. Minimum data lenth required: %d, but got data length %d", wantDataLen, haveDataLen) + } + + // The response data is empty unless there is some payload. And so, doesn't + // need to be unmarshalled. + if r.data == nil { + return nil + } + + // TODO(gVisor.dev/issue/3698): Use the unsafe version once go_marshal is safe to use again. + m.UnmarshalBytes(r.data[hdrLen:]) + return nil +} diff --git a/pkg/sentry/fsimpl/fuse/utils_test.go b/pkg/sentry/fsimpl/fuse/utils_test.go new file mode 100644 index 000000000..e1d9e3365 --- /dev/null +++ b/pkg/sentry/fsimpl/fuse/utils_test.go @@ -0,0 +1,132 @@ +// 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 ( + "io" + "testing" + + "gvisor.dev/gvisor/pkg/abi/linux" + "gvisor.dev/gvisor/pkg/marshal" + "gvisor.dev/gvisor/pkg/sentry/fsimpl/testutil" + "gvisor.dev/gvisor/pkg/sentry/kernel" + "gvisor.dev/gvisor/pkg/sentry/kernel/auth" + "gvisor.dev/gvisor/pkg/sentry/vfs" + "gvisor.dev/gvisor/pkg/usermem" +) + +func setup(t *testing.T) *testutil.System { + k, err := testutil.Boot() + if err != nil { + t.Fatalf("Error creating kernel: %v", err) + } + + ctx := k.SupervisorContext() + creds := auth.CredentialsFromContext(ctx) + + k.VFS().MustRegisterFilesystemType(Name, &FilesystemType{}, &vfs.RegisterFilesystemTypeOptions{ + AllowUserList: true, + AllowUserMount: true, + }) + + mntns, err := k.VFS().NewMountNamespace(ctx, creds, "", "tmpfs", &vfs.MountOptions{}) + if err != nil { + t.Fatalf("NewMountNamespace(): %v", err) + } + + return testutil.NewSystem(ctx, t, k.VFS(), mntns) +} + +// 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) { + vfsObj := &vfs.VirtualFilesystem{} + fuseDev := &DeviceFD{} + + if err := vfsObj.Init(system.Ctx); err != nil { + return nil, nil, err + } + + vd := vfsObj.NewAnonVirtualDentry("genCountFD") + defer vd.DecRef(system.Ctx) + if err := fuseDev.vfsfd.Init(fuseDev, linux.O_RDWR|linux.O_CREAT, vd.Mount(), vd.Dentry(), &vfs.FileDescriptionOptions{}); err != nil { + return nil, nil, err + } + + fsopts := filesystemOptions{ + maxActiveRequests: maxActiveRequests, + } + fs, err := newFUSEFilesystem(system.Ctx, 0, &fsopts, &fuseDev.vfsfd) + if err != nil { + return nil, nil, err + } + + return fs.conn, &fuseDev.vfsfd, nil +} + +type testPayload struct { + marshal.StubMarshallable + data uint32 +} + +// SizeBytes implements marshal.Marshallable.SizeBytes. +func (t *testPayload) SizeBytes() int { + return 4 +} + +// MarshalBytes implements marshal.Marshallable.MarshalBytes. +func (t *testPayload) MarshalBytes(dst []byte) { + usermem.ByteOrder.PutUint32(dst[:4], t.data) +} + +// UnmarshalBytes implements marshal.Marshallable.UnmarshalBytes. +func (t *testPayload) UnmarshalBytes(src []byte) { + *t = testPayload{data: usermem.ByteOrder.Uint32(src[:4])} +} + +// Packed implements marshal.Marshallable.Packed. +func (t *testPayload) Packed() bool { + return true +} + +// MarshalUnsafe implements marshal.Marshallable.MarshalUnsafe. +func (t *testPayload) MarshalUnsafe(dst []byte) { + t.MarshalBytes(dst) +} + +// UnmarshalUnsafe implements marshal.Marshallable.UnmarshalUnsafe. +func (t *testPayload) UnmarshalUnsafe(src []byte) { + t.UnmarshalBytes(src) +} + +// CopyOutN implements marshal.Marshallable.CopyOutN. +func (t *testPayload) CopyOutN(task marshal.CopyContext, addr usermem.Addr, limit int) (int, error) { + panic("not implemented") +} + +// CopyOut implements marshal.Marshallable.CopyOut. +func (t *testPayload) CopyOut(task marshal.CopyContext, addr usermem.Addr) (int, error) { + panic("not implemented") +} + +// CopyIn implements marshal.Marshallable.CopyIn. +func (t *testPayload) CopyIn(task marshal.CopyContext, addr usermem.Addr) (int, error) { + panic("not implemented") +} + +// WriteTo implements io.WriterTo.WriteTo. +func (t *testPayload) WriteTo(w io.Writer) (int64, error) { + panic("not implemented") +} diff --git a/pkg/sentry/fsimpl/kernfs/filesystem.go b/pkg/sentry/fsimpl/kernfs/filesystem.go index d7d3e8f48..49f6a0f1d 100644 --- a/pkg/sentry/fsimpl/kernfs/filesystem.go +++ b/pkg/sentry/fsimpl/kernfs/filesystem.go @@ -140,7 +140,7 @@ func (fs *Filesystem) revalidateChildLocked(ctx context.Context, vfsObj *vfs.Vir } // Reference on childVFSD dropped by a corresponding Valid. child = childVFSD.Impl().(*Dentry) - parent.insertChildLocked(name, child) + parent.InsertChildLocked(name, child) } return child, nil } @@ -548,7 +548,7 @@ func (fs *Filesystem) ReadlinkAt(ctx context.Context, rp *vfs.ResolvingPath) (st if !d.Impl().(*Dentry).isSymlink() { return "", syserror.EINVAL } - return inode.Readlink(ctx) + return inode.Readlink(ctx, rp.Mount()) } // RenameAt implements vfs.FilesystemImpl.RenameAt. @@ -657,6 +657,10 @@ func (fs *Filesystem) RenameAt(ctx context.Context, rp *vfs.ResolvingPath, oldPa func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() + + // Store the name before walkExistingLocked as rp will be advanced past the + // name in the following call. + name := rp.Component() vfsd, inode, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { @@ -686,7 +690,8 @@ func (fs *Filesystem) RmdirAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.RmDir(ctx, rp.Component(), vfsd); err != nil { + + if err := parentDentry.inode.RmDir(ctx, name, vfsd); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } @@ -765,6 +770,10 @@ func (fs *Filesystem) SymlinkAt(ctx context.Context, rp *vfs.ResolvingPath, targ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error { fs.mu.Lock() defer fs.mu.Unlock() + + // Store the name before walkExistingLocked as rp will be advanced past the + // name in the following call. + name := rp.Component() vfsd, _, err := fs.walkExistingLocked(ctx, rp) fs.processDeferredDecRefsLocked(ctx) if err != nil { @@ -790,7 +799,7 @@ func (fs *Filesystem) UnlinkAt(ctx context.Context, rp *vfs.ResolvingPath) error if err := virtfs.PrepareDeleteDentry(mntns, vfsd); err != nil { return err } - if err := parentDentry.inode.Unlink(ctx, rp.Component(), vfsd); err != nil { + if err := parentDentry.inode.Unlink(ctx, name, vfsd); err != nil { virtfs.AbortDeleteDentry(vfsd) return err } diff --git a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go index 74408e322..c2109cf76 100644 --- a/pkg/sentry/fsimpl/kernfs/inode_impl_util.go +++ b/pkg/sentry/fsimpl/kernfs/inode_impl_util.go @@ -172,7 +172,7 @@ func (InodeNoDynamicLookup) Valid(ctx context.Context) bool { type InodeNotSymlink struct{} // Readlink implements Inode.Readlink. -func (InodeNotSymlink) Readlink(context.Context) (string, error) { +func (InodeNotSymlink) Readlink(context.Context, *vfs.Mount) (string, error) { return "", syserror.EINVAL } @@ -256,6 +256,13 @@ func (a *InodeAttrs) Stat(context.Context, *vfs.Filesystem, vfs.StatOptions) (li // SetStat implements Inode.SetStat. func (a *InodeAttrs) SetStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { + return a.SetInodeStat(ctx, fs, creds, opts) +} + +// SetInodeStat sets the corresponding attributes from opts to InodeAttrs. +// This function can be used by other kernfs-based filesystem implementation to +// sets the unexported attributes into kernfs.InodeAttrs. +func (a *InodeAttrs) SetInodeStat(ctx context.Context, fs *vfs.Filesystem, creds *auth.Credentials, opts vfs.SetStatOptions) error { if opts.Stat.Mask == 0 { return nil } diff --git a/pkg/sentry/fsimpl/kernfs/kernfs.go b/pkg/sentry/fsimpl/kernfs/kernfs.go index 88fcd54aa..163f26ceb 100644 --- a/pkg/sentry/fsimpl/kernfs/kernfs.go +++ b/pkg/sentry/fsimpl/kernfs/kernfs.go @@ -60,6 +60,7 @@ import ( "gvisor.dev/gvisor/pkg/sentry/kernel/auth" "gvisor.dev/gvisor/pkg/sentry/vfs" "gvisor.dev/gvisor/pkg/sync" + "gvisor.dev/gvisor/pkg/syserror" ) // Filesystem mostly implements vfs.FilesystemImpl for a generic in-memory @@ -246,15 +247,15 @@ func (d *Dentry) OnZeroWatches(context.Context) {} // Precondition: d must represent a directory inode. func (d *Dentry) InsertChild(name string, child *Dentry) { d.dirMu.Lock() - d.insertChildLocked(name, child) + d.InsertChildLocked(name, child) d.dirMu.Unlock() } -// insertChildLocked is equivalent to InsertChild, with additional +// InsertChildLocked is equivalent to InsertChild, with additional // preconditions. // // Precondition: d.dirMu must be locked. -func (d *Dentry) insertChildLocked(name string, child *Dentry) { +func (d *Dentry) InsertChildLocked(name string, child *Dentry) { if !d.isDir() { panic(fmt.Sprintf("InsertChild called on non-directory Dentry: %+v.", d)) } @@ -267,6 +268,36 @@ func (d *Dentry) insertChildLocked(name string, child *Dentry) { d.children[name] = child } +// RemoveChild removes child from the vfs dentry cache. This does not update the +// directory inode or modify the inode to be unlinked. So calling this on its own +// isn't sufficient to remove a child from a directory. +// +// Precondition: d must represent a directory inode. +func (d *Dentry) RemoveChild(name string, child *vfs.Dentry) error { + d.dirMu.Lock() + defer d.dirMu.Unlock() + return d.RemoveChildLocked(name, child) +} + +// RemoveChildLocked is equivalent to RemoveChild, with additional +// preconditions. +// +// Precondition: d.dirMu must be locked. +func (d *Dentry) RemoveChildLocked(name string, child *vfs.Dentry) error { + if !d.isDir() { + panic(fmt.Sprintf("RemoveChild called on non-directory Dentry: %+v.", d)) + } + c, ok := d.children[name] + if !ok { + return syserror.ENOENT + } + if &c.vfsd != child { + panic(fmt.Sprintf("Dentry hashed into inode doesn't match what vfs thinks! Child: %+v, vfs: %+v", c, child)) + } + delete(d.children, name) + return nil +} + // Inode returns the dentry's inode. func (d *Dentry) Inode() Inode { return d.inode @@ -425,7 +456,7 @@ type inodeDynamicLookup interface { Valid(ctx context.Context) bool // IterDirents is used to iterate over dynamically created entries. It invokes - // cb on each entry in the directory represented by the FileDescription. + // cb on each entry in the directory represented by the Inode. // 'offset' is the offset for the entire IterDirents call, which may include // results from the caller (e.g. "." and ".."). 'relOffset' is the offset // inside the entries returned by this IterDirents invocation. In other words, @@ -437,7 +468,7 @@ type inodeDynamicLookup interface { type inodeSymlink interface { // Readlink returns the target of a symbolic link. If an inode is not a // symlink, the implementation should return EINVAL. - Readlink(ctx context.Context) (string, error) + Readlink(ctx context.Context, mnt *vfs.Mount) (string, error) // Getlink returns the target of a symbolic link, as used by path // resolution: diff --git a/pkg/sentry/fsimpl/kernfs/symlink.go b/pkg/sentry/fsimpl/kernfs/symlink.go index 64731a3e4..a9812fcef 100644 --- a/pkg/sentry/fsimpl/kernfs/symlink.go +++ b/pkg/sentry/fsimpl/kernfs/symlink.go @@ -52,7 +52,7 @@ func (s *StaticSymlink) Init(creds *auth.Credentials, devMajor uint32, devMinor } // Readlink implements Inode. -func (s *StaticSymlink) Readlink(_ context.Context) (string, error) { +func (s *StaticSymlink) Readlink(_ context.Context, _ *vfs.Mount) (string, error) { return s.target, nil } diff --git a/pkg/sentry/fsimpl/proc/task_fds.go b/pkg/sentry/fsimpl/proc/task_fds.go index 94ec2ff69..47d744a57 100644 --- a/pkg/sentry/fsimpl/proc/task_fds.go +++ b/pkg/sentry/fsimpl/proc/task_fds.go @@ -214,7 +214,7 @@ func (fs *filesystem) newFDSymlink(task *kernel.Task, fd int32, ino uint64) *ker return d } -func (s *fdSymlink) Readlink(ctx context.Context) (string, error) { +func (s *fdSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { file, _ := getTaskFD(s.task, s.fd) if file == nil { return "", syserror.ENOENT diff --git a/pkg/sentry/fsimpl/proc/task_files.go b/pkg/sentry/fsimpl/proc/task_files.go index ce87b0d47..87726d8f7 100644 --- a/pkg/sentry/fsimpl/proc/task_files.go +++ b/pkg/sentry/fsimpl/proc/task_files.go @@ -668,7 +668,7 @@ func (fs *filesystem) newExeSymlink(task *kernel.Task, ino uint64) *kernfs.Dentr } // Readlink implements kernfs.Inode. -func (s *exeSymlink) Readlink(ctx context.Context) (string, error) { +func (s *exeSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { if !kernel.ContextCanTrace(ctx, s.task, false) { return "", syserror.EACCES } @@ -808,11 +808,11 @@ func (fs *filesystem) newNamespaceSymlink(task *kernel.Task, ino uint64, ns stri } // Readlink implements Inode. -func (s *namespaceSymlink) Readlink(ctx context.Context) (string, error) { +func (s *namespaceSymlink) Readlink(ctx context.Context, mnt *vfs.Mount) (string, error) { if err := checkTaskState(s.task); err != nil { return "", err } - return s.StaticSymlink.Readlink(ctx) + return s.StaticSymlink.Readlink(ctx, mnt) } // Getlink implements Inode.Getlink. diff --git a/pkg/sentry/fsimpl/proc/tasks_files.go b/pkg/sentry/fsimpl/proc/tasks_files.go index 8c41729e4..68c541046 100644 --- a/pkg/sentry/fsimpl/proc/tasks_files.go +++ b/pkg/sentry/fsimpl/proc/tasks_files.go @@ -51,7 +51,7 @@ func (fs *filesystem) newSelfSymlink(creds *auth.Credentials, ino uint64, pidns return d } -func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { +func (s *selfSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { t := kernel.TaskFromContext(ctx) if t == nil { // Who is reading this link? @@ -64,8 +64,8 @@ func (s *selfSymlink) Readlink(ctx context.Context) (string, error) { return strconv.FormatUint(uint64(tgid), 10), nil } -func (s *selfSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { - target, err := s.Readlink(ctx) +func (s *selfSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx, mnt) return vfs.VirtualDentry{}, target, err } @@ -94,7 +94,7 @@ func (fs *filesystem) newThreadSelfSymlink(creds *auth.Credentials, ino uint64, return d } -func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { +func (s *threadSelfSymlink) Readlink(ctx context.Context, _ *vfs.Mount) (string, error) { t := kernel.TaskFromContext(ctx) if t == nil { // Who is reading this link? @@ -108,8 +108,8 @@ func (s *threadSelfSymlink) Readlink(ctx context.Context) (string, error) { return fmt.Sprintf("%d/task/%d", tgid, tid), nil } -func (s *threadSelfSymlink) Getlink(ctx context.Context, _ *vfs.Mount) (vfs.VirtualDentry, string, error) { - target, err := s.Readlink(ctx) +func (s *threadSelfSymlink) Getlink(ctx context.Context, mnt *vfs.Mount) (vfs.VirtualDentry, string, error) { + target, err := s.Readlink(ctx, mnt) return vfs.VirtualDentry{}, target, err } diff --git a/pkg/syserror/syserror.go b/pkg/syserror/syserror.go index fe9f50169..f516c8e46 100644 --- a/pkg/syserror/syserror.go +++ b/pkg/syserror/syserror.go @@ -33,6 +33,7 @@ var ( EBADFD = error(syscall.EBADFD) EBUSY = error(syscall.EBUSY) ECHILD = error(syscall.ECHILD) + ECONNABORTED = error(syscall.ECONNABORTED) ECONNREFUSED = error(syscall.ECONNREFUSED) ECONNRESET = error(syscall.ECONNRESET) EDEADLK = error(syscall.EDEADLK) diff --git a/test/fuse/BUILD b/test/fuse/BUILD index 56157c96b..8e31fdd41 100644 --- a/test/fuse/BUILD +++ b/test/fuse/BUILD @@ -5,5 +5,69 @@ package(licenses = ["notice"]) syscall_test( fuse = "True", test = "//test/fuse/linux:stat_test", - vfs2 = "True", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:open_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:release_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:mknod_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:symlink_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:readlink_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:mkdir_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:read_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:write_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:rmdir_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:readdir_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:create_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:unlink_test", +) + +syscall_test( + fuse = "True", + test = "//test/fuse/linux:setstat_test", ) diff --git a/test/fuse/README.md b/test/fuse/README.md index 734c3a4e3..65add57e2 100644 --- a/test/fuse/README.md +++ b/test/fuse/README.md @@ -1,65 +1,114 @@ # gVisor FUSE Test Suite -This is an integration test suite for fuse(4) filesystem. It runs under both -gVisor and Linux, and ensures compatibility between the two. This test suite is -based on system calls test. +This is an integration test suite for fuse(4) filesystem. It runs under gVisor +sandbox container with VFS2 and FUSE function enabled. -This document describes the framework of fuse integration test and the -guidelines that should be followed when adding new fuse tests. +This document describes the framework of FUSE integration test, how to use it, +and the guidelines that should be followed when adding new testing features. ## Integration Test Framework -Please refer to the figure below. `>` is entering the function, `<` is leaving -the function, and `=` indicates sequentially entering and leaving. +By inheriting the `FuseTest` class defined in `linux/fuse_base.h`, every test +fixture can run in an environment with `mount_point_` mounted by a fake FUSE +server. It creates a `socketpair(2)` to send and receive control commands and +data between the client and the server. Because the FUSE server runs in the +background thread, gTest cannot catch its assertion failure immediately. Thus, +`TearDown()` function sends command to the FUSE server to check if all gTest +assertion in the server are successful and all requests and preset responses are +consumed. + +## Communication Diagram + +Diagram below describes how a testing thread communicates with the FUSE server +to achieve integration test. + +For the following diagram, `>` means entering the function, `<` is leaving the +function, and `=` indicates sequentially entering and leaving. Not necessarily +follow exactly the below diagram due to the nature of a multi-threaded system, +however, it is still helpful to know when the client waits for the server to +complete a command and when the server awaits the next instruction. ``` - | Client (Test Main Process) | Server (FUSE Daemon) - | | - | >TEST_F() | - | >SetUp() | - | =MountFuse() | - | >SetUpFuseServer() | - | [create communication pipes] | - | =fork() | =fork() - | >WaitCompleted() | - | [wait for MarkDone()] | - | | =ConsumeFuseInit() - | | =MarkDone() - | <WaitCompleted() | - | <SetUpFuseServer() | - | <SetUp() | - | >SetExpected() | - | [construct expected reaction] | - | | >FuseLoop() - | | >ReceiveExpected() - | | [wait data from pipe] - | [write data to pipe] | - | [wait for MarkDone()] | - | | [save data to memory] - | | =MarkDone() - | <SetExpected() | - | | <ReceiveExpected() - | | >read() - | | [wait for fs operation] - | >[Do fs operation] | - | [wait for fs response] | - | | <read() - | | =CompareRequest() - | | =write() [write fs response] - | <[Do fs operation] | - | =[Test fs operation result] | - | =[wait for MarkDone()] | - | | =MarkDone() - | >TearDown() | - | =UnmountFuse() | - | <TearDown() | - | <TEST_F() | +| Client (Testing Thread) | Server (FUSE Server Thread) +| | +| >TEST_F() | +| >SetUp() | +| =MountFuse() | +| >SetUpFuseServer() | +| [create communication socket]| +| =fork() | =fork() +| [wait server complete] | +| | =ServerConsumeFuseInit() +| | =ServerCompleteWith() +| <SetUpFuseServer() | +| <SetUp() | +| [testing main] | +| | >ServerFuseLoop() +| | [poll on socket and fd] +| >SetServerResponse() | +| [write data to socket] | +| [wait server complete] | +| | [socket event occurs] +| | >ServerHandleCommand() +| | >ServerReceiveResponse() +| | [read data from socket] +| | [save data to memory] +| | <ServerReceiveResponse() +| | =ServerCompleteWith() +| <SetServerResponse() | +| | <ServerHandleCommand() +| >[Do fs operation] | +| [wait for fs response] | +| | [fd event occurs] +| | >ServerProcessFuseRequest() +| | =[read fs request] +| | =[save fs request to memory] +| | =[write fs response] +| <[Do fs operation] | +| | <ServerProcessFuseRequest() +| | +| =[Test fs operation result] | +| | +| >GetServerActualRequest() | +| [write data to socket] | +| [wait data from server] | +| | [socket event occurs] +| | >ServerHandleCommand() +| | >ServerSendReceivedRequest() +| | [write data to socket] +| [read data from socket] | +| [wait server complete] | +| | <ServerSendReceivedRequest() +| | =ServerCompleteWith() +| <GetServerActualRequest() | +| | <ServerHandleCommand() +| | +| =[Test actual request] | +| | +| >TearDown() | +| ... | +| >GetServerNumUnsentResponses() | +| [write data to socket] | +| [wait server complete] | +| | [socket event arrive] +| | >ServerHandleCommand() +| | >ServerSendData() +| | [write data to socket] +| | <ServerSendData() +| | =ServerCompleteWith() +| [read data from socket] | +| [test if all succeeded] | +| <GetServerNumUnsentResponses() | +| | <ServerHandleCommand() +| =UnmountFuse() | +| <TearDown() | +| <TEST_F() | ``` ## Running the tests -Based on syscall tests, fuse tests can run in different environments. To enable -fuse testing environment, the test targets should be appended with `_fuse`. +Based on syscall tests, FUSE tests generate targets only with vfs2 and fuse +enabled. The corresponding targets end in `_fuse`. For example, to run fuse test in `stat_test.cc`: @@ -77,17 +126,16 @@ $ bazel test --test_tag_filters=fuse //test/fuse/... 1. Add test targets in `BUILD` and `linux/BUILD`. 2. Inherit your test from `FuseTest` base class. It allows you to: - - Run a fake FUSE server in background during each test setup. - - Create pipes for communication and provide utility functions. - - Stop FUSE server after test completes. -3. Customize your comparison function for request assessment in FUSE server. -4. Add the mapping of the size of structs if you are working on new FUSE - opcode. - - Please update `FuseTest::GetPayloadSize()` for each new FUSE opcode. -5. Build the expected request-response pair of your FUSE operation. -6. Call `SetExpected()` function to inject the expected reaction. -7. Check the response and/or errors. -8. Finally call `WaitCompleted()` to ensure the FUSE server acts correctly. + - Fork a fake FUSE server in background during each test setup. + - Create a pair of sockets for communication and provide utility + functions. + - Stop FUSE server and check if error occurs in it after test completes. +3. Build the expected opcode-response pairs of your FUSE operation. +4. Call `SetServerResponse()` to preset the next expected opcode and response. +5. Do real filesystem operations (FUSE is mounted at `mount_point_`). +6. Check FUSE response and/or errors. +7. Retrieve FUSE request by `GetServerActualRequest()`. +8. Check if the request is as expected. A few customized matchers used in syscalls test are encouraged to test the outcome of filesystem operations. Such as: @@ -101,3 +149,40 @@ SyscallFailsWithErrno(...) Please refer to [test/syscalls/README.md](../syscalls/README.md) for further details. + +## Writing a new FuseTestCmd + +A `FuseTestCmd` is a control protocol used in the communication between the +testing thread and the FUSE server. Such commands are sent from the testing +thread to the FUSE server to set up, control, or inspect the behavior of the +FUSE server in response to a sequence of FUSE requests. + +The lifecycle of a command contains following steps: + +1. The testing thread sends a `FuseTestCmd` via socket and waits for + completion. +2. The FUSE server receives the command and does corresponding action. +3. (Optional) The testing thread reads data from socket. +4. The FUSE server sends a success indicator via socket after processing. +5. The testing thread gets the success signal and continues testing. + +The success indicator, i.e. `WaitServerComplete()`, is crucial at the end of +each `FuseTestCmd` sent from the testing thread. Because we don't want to begin +filesystem operation if the requests have not been completely set up. Also, to +test FUSE interactions in a sequential manner, concurrent requests are not +supported now. + +To add a new `FuseTestCmd`, one must comply with following format: + +1. Add a new `FuseTestCmd` enum class item defined in `linux/fuse_base.h` +2. Add a `SetServerXXX()` or `GetServerXXX()` public function in `FuseTest`. + This is how the testing thread will call to send control message. Define how + many bytes you want to send along with the command and what you will expect + to receive. Finally it should block and wait for a success indicator from + the FUSE server. +3. Add a handler logic in the switch condition of `ServerHandleCommand()`. Use + `ServerSendData()` or declare a new private function such as + `ServerReceiveXXX()` or `ServerSendXXX()`. It is mandatory to set it private + since only the FUSE server (forked from `FuseTest` base class) can call it. + This is the server part of the specific `FuseTestCmd` and the format of the + data should be consistent with what the client expects in the previous step. diff --git a/test/fuse/linux/BUILD b/test/fuse/linux/BUILD index 4871bb531..7673252ec 100644 --- a/test/fuse/linux/BUILD +++ b/test/fuse/linux/BUILD @@ -11,7 +11,134 @@ cc_binary( srcs = ["stat_test.cc"], deps = [ gtest, + ":fuse_fd_util", + "//test/util:cleanup", + "//test/util:fs_util", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "open_test", + testonly = 1, + srcs = ["open_test.cc"], + deps = [ + gtest, ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "release_test", + testonly = 1, + srcs = ["release_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "mknod_test", + testonly = 1, + srcs = ["mknod_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:temp_umask", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "symlink_test", + testonly = 1, + srcs = ["symlink_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "readlink_test", + testonly = 1, + srcs = ["readlink_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "mkdir_test", + testonly = 1, + srcs = ["mkdir_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:temp_umask", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "setstat_test", + testonly = 1, + srcs = ["setstat_test.cc"], + deps = [ + gtest, + ":fuse_fd_util", + "//test/util:cleanup", + "//test/util:fs_util", + "//test/util:fuse_util", + "//test/util:temp_umask", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "rmdir_test", + testonly = 1, + srcs = ["rmdir_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fs_util", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "readdir_test", + testonly = 1, + srcs = ["readdir_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fs_util", + "//test/util:fuse_util", "//test/util:test_main", "//test/util:test_util", ], @@ -24,9 +151,80 @@ cc_library( hdrs = ["fuse_base.h"], deps = [ gtest, + "//test/util:fuse_util", "//test/util:posix_error", "//test/util:temp_path", "//test/util:test_util", "@com_google_absl//absl/strings:str_format", ], ) + +cc_library( + name = "fuse_fd_util", + testonly = 1, + srcs = ["fuse_fd_util.cc"], + hdrs = ["fuse_fd_util.h"], + deps = [ + gtest, + ":fuse_base", + "//test/util:cleanup", + "//test/util:file_descriptor", + "//test/util:fuse_util", + "//test/util:posix_error", + ], +) + +cc_binary( + name = "read_test", + testonly = 1, + srcs = ["read_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "write_test", + testonly = 1, + srcs = ["write_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "create_test", + testonly = 1, + srcs = ["create_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fs_util", + "//test/util:fuse_util", + "//test/util:temp_umask", + "//test/util:test_main", + "//test/util:test_util", + ], +) + +cc_binary( + name = "unlink_test", + testonly = 1, + srcs = ["unlink_test.cc"], + deps = [ + gtest, + ":fuse_base", + "//test/util:fuse_util", + "//test/util:temp_umask", + "//test/util:test_main", + "//test/util:test_util", + ], +) diff --git a/test/fuse/linux/create_test.cc b/test/fuse/linux/create_test.cc new file mode 100644 index 000000000..9a0219a58 --- /dev/null +++ b/test/fuse/linux/create_test.cc @@ -0,0 +1,128 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fs_util.h" +#include "test/util/fuse_util.h" +#include "test/util/temp_umask.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class CreateTest : public FuseTest { + protected: + const std::string test_file_name_ = "test_file"; + const mode_t mode = S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(CreateTest, CreateFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_name_); + + // Ensure the file doesn't exist. + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header), + .error = -ENOENT, + }; + auto iov_out = FuseGenerateIovecs(out_header); + SetServerResponse(FUSE_LOOKUP, iov_out); + + // creat(2) is equal to open(2) with open_flags O_CREAT | O_WRONLY | O_TRUNC. + const mode_t new_mask = S_IWGRP | S_IWOTH; + const int open_flags = O_CREAT | O_WRONLY | O_TRUNC; + out_header.error = 0; + out_header.len = sizeof(struct fuse_out_header) + + sizeof(struct fuse_entry_out) + sizeof(struct fuse_open_out); + struct fuse_entry_out entry_payload = DefaultEntryOut(mode & ~new_mask, 2); + struct fuse_open_out out_payload = { + .fh = 1, + .open_flags = open_flags, + }; + iov_out = FuseGenerateIovecs(out_header, entry_payload, out_payload); + SetServerResponse(FUSE_CREATE, iov_out); + + // kernfs generates a successive FUSE_OPEN after the file is created. Linux's + // fuse kernel module will not send this FUSE_OPEN after creat(2). + out_header.len = + sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out); + iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_OPEN, iov_out); + + int fd; + TempUmask mask(new_mask); + EXPECT_THAT(fd = creat(test_file_path.c_str(), mode), SyscallSucceeds()); + EXPECT_THAT(fcntl(fd, F_GETFL), + SyscallSucceedsWithValue(open_flags & O_ACCMODE)); + + struct fuse_in_header in_header; + struct fuse_create_in in_payload; + std::vector<char> name(test_file_name_.size() + 1); + auto iov_in = FuseGenerateIovecs(in_header, in_payload, name); + + // Skip the request of FUSE_LOOKUP. + SkipServerActualRequest(); + + // Get the first FUSE_CREATE. + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload) + + test_file_name_.size() + 1); + EXPECT_EQ(in_header.opcode, FUSE_CREATE); + EXPECT_EQ(in_payload.flags, open_flags); + EXPECT_EQ(in_payload.mode, mode & ~new_mask); + EXPECT_EQ(in_payload.umask, new_mask); + EXPECT_EQ(std::string(name.data()), test_file_name_); + + // Get the successive FUSE_OPEN. + struct fuse_open_in in_payload_open; + iov_in = FuseGenerateIovecs(in_header, in_payload_open); + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload_open)); + EXPECT_EQ(in_header.opcode, FUSE_OPEN); + EXPECT_EQ(in_payload_open.flags, open_flags & O_ACCMODE); + + EXPECT_THAT(close(fd), SyscallSucceeds()); + // Skip the FUSE_RELEASE. + SkipServerActualRequest(); +} + +TEST_F(CreateTest, CreateFileAlreadyExists) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_name_); + + const int open_flags = O_CREAT | O_EXCL; + + SetServerInodeLookup(test_file_name_); + + EXPECT_THAT(open(test_file_path.c_str(), mode, open_flags), + SyscallFailsWithErrno(EEXIST)); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/fuse_base.cc b/test/fuse/linux/fuse_base.cc index 9c3124472..5b45804e1 100644 --- a/test/fuse/linux/fuse_base.cc +++ b/test/fuse/linux/fuse_base.cc @@ -16,17 +16,17 @@ #include <fcntl.h> #include <linux/fuse.h> -#include <string.h> +#include <poll.h> #include <sys/mount.h> +#include <sys/socket.h> #include <sys/stat.h> #include <sys/types.h> #include <sys/uio.h> #include <unistd.h> -#include <iostream> - #include "gtest/gtest.h" #include "absl/strings/str_format.h" +#include "test/util/fuse_util.h" #include "test/util/posix_error.h" #include "test/util/temp_path.h" #include "test/util/test_util.h" @@ -39,49 +39,123 @@ void FuseTest::SetUp() { SetUpFuseServer(); } -void FuseTest::TearDown() { UnmountFuse(); } - -// Since CompareRequest is running in background thread, gTest assertions and -// expectations won't directly reflect the test result. However, the FUSE -// background server still connects to the same standard I/O as testing main -// thread. So EXPECT_XX can still be used to show different results. To -// ensure failed testing result is observable, return false and the result -// will be sent to test main thread via pipe. -bool FuseTest::CompareRequest(void* expected_mem, size_t expected_len, - void* real_mem, size_t real_len) { - if (expected_len != real_len) return false; - return memcmp(expected_mem, real_mem, expected_len) == 0; +void FuseTest::TearDown() { + EXPECT_EQ(GetServerNumUnconsumedRequests(), 0); + EXPECT_EQ(GetServerNumUnsentResponses(), 0); + UnmountFuse(); } -// SetExpected is called by the testing main thread to set expected request- -// response pair of a single FUSE operation. -void FuseTest::SetExpected(struct iovec* iov_in, int iov_in_cnt, - struct iovec* iov_out, int iov_out_cnt) { - EXPECT_THAT(RetryEINTR(writev)(set_expected_[1], iov_in, iov_in_cnt), - SyscallSucceedsWithValue(::testing::Gt(0))); - WaitCompleted(); +// Sends 3 parts of data to the FUSE server: +// 1. The `kSetResponse` command +// 2. The expected opcode +// 3. The fake FUSE response +// Then waits for the FUSE server to notify its completion. +void FuseTest::SetServerResponse(uint32_t opcode, + std::vector<struct iovec>& iovecs) { + uint32_t cmd = static_cast<uint32_t>(FuseTestCmd::kSetResponse); + EXPECT_THAT(RetryEINTR(write)(sock_[0], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + EXPECT_THAT(RetryEINTR(write)(sock_[0], &opcode, sizeof(opcode)), + SyscallSucceedsWithValue(sizeof(opcode))); - EXPECT_THAT(RetryEINTR(writev)(set_expected_[1], iov_out, iov_out_cnt), - SyscallSucceedsWithValue(::testing::Gt(0))); - WaitCompleted(); + EXPECT_THAT(RetryEINTR(writev)(sock_[0], iovecs.data(), iovecs.size()), + SyscallSucceeds()); + + WaitServerComplete(); } -// WaitCompleted waits for the FUSE server to finish its job and check if it +// Waits for the FUSE server to finish its blocking job and check if it // completes without errors. -void FuseTest::WaitCompleted() { - char success; - EXPECT_THAT(RetryEINTR(read)(done_[0], &success, sizeof(success)), - SyscallSucceedsWithValue(1)); +void FuseTest::WaitServerComplete() { + uint32_t success; + EXPECT_THAT(RetryEINTR(read)(sock_[0], &success, sizeof(success)), + SyscallSucceedsWithValue(sizeof(success))); + ASSERT_EQ(success, 1); +} + +// Sends the `kGetRequest` command to the FUSE server, then reads the next +// request into iovec struct. The order of calling this function should be +// the same as the one of SetServerResponse(). +void FuseTest::GetServerActualRequest(std::vector<struct iovec>& iovecs) { + uint32_t cmd = static_cast<uint32_t>(FuseTestCmd::kGetRequest); + EXPECT_THAT(RetryEINTR(write)(sock_[0], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + EXPECT_THAT(RetryEINTR(readv)(sock_[0], iovecs.data(), iovecs.size()), + SyscallSucceeds()); + + WaitServerComplete(); +} + +// Sends a FuseTestCmd command to the FUSE server, reads from the socket, and +// returns the corresponding data. +uint32_t FuseTest::GetServerData(uint32_t cmd) { + uint32_t data; + EXPECT_THAT(RetryEINTR(write)(sock_[0], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + EXPECT_THAT(RetryEINTR(read)(sock_[0], &data, sizeof(data)), + SyscallSucceedsWithValue(sizeof(data))); + + WaitServerComplete(); + return data; +} + +uint32_t FuseTest::GetServerNumUnconsumedRequests() { + return GetServerData( + static_cast<uint32_t>(FuseTestCmd::kGetNumUnconsumedRequests)); +} + +uint32_t FuseTest::GetServerNumUnsentResponses() { + return GetServerData( + static_cast<uint32_t>(FuseTestCmd::kGetNumUnsentResponses)); +} + +uint32_t FuseTest::GetServerTotalReceivedBytes() { + return GetServerData( + static_cast<uint32_t>(FuseTestCmd::kGetTotalReceivedBytes)); +} + +// Sends the `kSkipRequest` command to the FUSE server, which would skip +// current stored request data. +void FuseTest::SkipServerActualRequest() { + uint32_t cmd = static_cast<uint32_t>(FuseTestCmd::kSkipRequest); + EXPECT_THAT(RetryEINTR(write)(sock_[0], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + WaitServerComplete(); } -void FuseTest::MountFuse() { +// Sends the `kSetInodeLookup` command, expected mode, and the path of the +// inode to create under the mount point. +void FuseTest::SetServerInodeLookup(const std::string& path, mode_t mode, + uint64_t size) { + uint32_t cmd = static_cast<uint32_t>(FuseTestCmd::kSetInodeLookup); + EXPECT_THAT(RetryEINTR(write)(sock_[0], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + EXPECT_THAT(RetryEINTR(write)(sock_[0], &mode, sizeof(mode)), + SyscallSucceedsWithValue(sizeof(mode))); + + EXPECT_THAT(RetryEINTR(write)(sock_[0], &size, sizeof(size)), + SyscallSucceedsWithValue(sizeof(size))); + + // Pad 1 byte for null-terminate c-string. + EXPECT_THAT(RetryEINTR(write)(sock_[0], path.c_str(), path.size() + 1), + SyscallSucceedsWithValue(path.size() + 1)); + + WaitServerComplete(); +} + +void FuseTest::MountFuse(const char* mountOpts) { EXPECT_THAT(dev_fd_ = open("/dev/fuse", O_RDWR), SyscallSucceeds()); - std::string mount_opts = absl::StrFormat("fd=%d,%s", dev_fd_, kMountOpts); + std::string mount_opts = absl::StrFormat("fd=%d,%s", dev_fd_, mountOpts); mount_point_ = ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()); EXPECT_THAT(mount("fuse", mount_point_.path().c_str(), "fuse", MS_NODEV | MS_NOSUID, mount_opts.c_str()), - SyscallSucceedsWithValue(0)); + SyscallSucceeds()); } void FuseTest::UnmountFuse() { @@ -89,13 +163,13 @@ void FuseTest::UnmountFuse() { // TODO(gvisor.dev/issue/3330): ensure the process is terminated successfully. } -// ConsumeFuseInit consumes the first FUSE request and returns the -// corresponding PosixError. -PosixError FuseTest::ConsumeFuseInit() { +// Consumes the first FUSE request and returns the corresponding PosixError. +PosixError FuseTest::ServerConsumeFuseInit( + const struct fuse_init_out* out_payload) { + std::vector<char> buf(FUSE_MIN_READ_BUFFER); RETURN_ERROR_IF_SYSCALL_FAIL( - RetryEINTR(read)(dev_fd_, buf_.data(), buf_.size())); + RetryEINTR(read)(dev_fd_, buf.data(), buf.size())); - struct iovec iov_out[2]; struct fuse_out_header out_header = { .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_init_out), .error = 0, @@ -103,72 +177,74 @@ PosixError FuseTest::ConsumeFuseInit() { }; // Returns a fake fuse_init_out with 7.0 version to avoid ECONNREFUSED // error in the initialization of FUSE connection. - struct fuse_init_out out_payload = { - .major = 7, - }; - iov_out[0].iov_len = sizeof(out_header); - iov_out[0].iov_base = &out_header; - iov_out[1].iov_len = sizeof(out_payload); - iov_out[1].iov_base = &out_payload; + auto iov_out = FuseGenerateIovecs( + out_header, *const_cast<struct fuse_init_out*>(out_payload)); - RETURN_ERROR_IF_SYSCALL_FAIL(RetryEINTR(writev)(dev_fd_, iov_out, 2)); + RETURN_ERROR_IF_SYSCALL_FAIL( + RetryEINTR(writev)(dev_fd_, iov_out.data(), iov_out.size())); return NoError(); } -// ReceiveExpected reads 1 pair of expected fuse request-response `iovec`s -// from pipe and save them into member variables of this testing instance. -void FuseTest::ReceiveExpected() { - // Set expected fuse_in request. - EXPECT_THAT(len_in_ = RetryEINTR(read)(set_expected_[0], mem_in_.data(), - mem_in_.size()), - SyscallSucceedsWithValue(::testing::Gt(0))); - MarkDone(len_in_ > 0); +// Reads 1 expected opcode and a fake response from socket and save them into +// the serial buffer of this testing instance. +void FuseTest::ServerReceiveResponse() { + ssize_t len; + uint32_t opcode; + std::vector<char> buf(FUSE_MIN_READ_BUFFER); + EXPECT_THAT(RetryEINTR(read)(sock_[1], &opcode, sizeof(opcode)), + SyscallSucceedsWithValue(sizeof(opcode))); - // Set expected fuse_out response. - EXPECT_THAT(len_out_ = RetryEINTR(read)(set_expected_[0], mem_out_.data(), - mem_out_.size()), - SyscallSucceedsWithValue(::testing::Gt(0))); - MarkDone(len_out_ > 0); + EXPECT_THAT(len = RetryEINTR(read)(sock_[1], buf.data(), buf.size()), + SyscallSucceeds()); + + responses_.AddMemBlock(opcode, buf.data(), len); } -// MarkDone writes 1 byte of success indicator through pipe. -void FuseTest::MarkDone(bool success) { - char data = success ? 1 : 0; - EXPECT_THAT(RetryEINTR(write)(done_[1], &data, sizeof(data)), - SyscallSucceedsWithValue(1)); +// Writes 1 byte of success indicator through socket. +void FuseTest::ServerCompleteWith(bool success) { + uint32_t data = success ? 1 : 0; + ServerSendData(data); } -// FuseLoop is the implementation of the fake FUSE server. Read from /dev/fuse, -// compare the request by CompareRequest (use derived function if specified), -// and write the expected response to /dev/fuse. -void FuseTest::FuseLoop() { - bool success = true; - ssize_t len = 0; +// ServerFuseLoop is the implementation of the fake FUSE server. Monitors 2 +// file descriptors: /dev/fuse and sock_[1]. Events from /dev/fuse are FUSE +// requests and events from sock_[1] are FUSE testing commands, leading by +// a FuseTestCmd data to indicate the command. +void FuseTest::ServerFuseLoop() { + const int nfds = 2; + struct pollfd fds[nfds] = { + { + .fd = dev_fd_, + .events = POLL_IN | POLLHUP | POLLERR | POLLNVAL, + }, + { + .fd = sock_[1], + .events = POLL_IN | POLLHUP | POLLERR | POLLNVAL, + }, + }; + while (true) { - ReceiveExpected(); + ASSERT_THAT(poll(fds, nfds, -1), SyscallSucceeds()); - EXPECT_THAT(len = RetryEINTR(read)(dev_fd_, buf_.data(), buf_.size()), - SyscallSucceedsWithValue(len_in_)); - if (len != len_in_) success = false; + for (int fd_idx = 0; fd_idx < nfds; ++fd_idx) { + if (fds[fd_idx].revents == 0) continue; - if (!CompareRequest(buf_.data(), len_in_, mem_in_.data(), len_in_)) { - std::cerr << "the FUSE request is not expected" << std::endl; - success = false; + ASSERT_EQ(fds[fd_idx].revents, POLL_IN); + if (fds[fd_idx].fd == sock_[1]) { + ServerHandleCommand(); + } else if (fds[fd_idx].fd == dev_fd_) { + ServerProcessFuseRequest(); + } } - - EXPECT_THAT(len = RetryEINTR(write)(dev_fd_, mem_out_.data(), len_out_), - SyscallSucceedsWithValue(len_out_)); - if (len != len_out_) success = false; - MarkDone(success); } } -// SetUpFuseServer creates 2 pipes. First is for testing client to send the -// expected request-response pair, and the other acts as a checkpoint for the -// FUSE server to notify the client that it can proceed. -void FuseTest::SetUpFuseServer() { - ASSERT_THAT(pipe(set_expected_), SyscallSucceedsWithValue(0)); - ASSERT_THAT(pipe(done_), SyscallSucceedsWithValue(0)); +// SetUpFuseServer creates 1 socketpair and fork the process. The parent thread +// becomes testing thread and the child thread becomes the FUSE server running +// in background. These 2 threads are connected via socketpair. sock_[0] is +// opened in testing thread and sock_[1] is opened in the FUSE server. +void FuseTest::SetUpFuseServer(const struct fuse_init_out* payload) { + ASSERT_THAT(socketpair(AF_UNIX, SOCK_STREAM, 0, sock_), SyscallSucceeds()); switch (fork()) { case -1: @@ -177,31 +253,194 @@ void FuseTest::SetUpFuseServer() { case 0: break; default: - ASSERT_THAT(close(set_expected_[0]), SyscallSucceedsWithValue(0)); - ASSERT_THAT(close(done_[1]), SyscallSucceedsWithValue(0)); - WaitCompleted(); + ASSERT_THAT(close(sock_[1]), SyscallSucceeds()); + WaitServerComplete(); return; } - ASSERT_THAT(close(set_expected_[1]), SyscallSucceedsWithValue(0)); - ASSERT_THAT(close(done_[0]), SyscallSucceedsWithValue(0)); - - MarkDone(ConsumeFuseInit().ok()); - - FuseLoop(); + // Begin child thread, i.e. the FUSE server. + ASSERT_THAT(close(sock_[0]), SyscallSucceeds()); + ServerCompleteWith(ServerConsumeFuseInit(payload).ok()); + ServerFuseLoop(); _exit(0); } -// GetPayloadSize is a helper function to get the number of bytes of a -// specific FUSE operation struct. -size_t FuseTest::GetPayloadSize(uint32_t opcode, bool in) { - switch (opcode) { - case FUSE_INIT: - return in ? sizeof(struct fuse_init_in) : sizeof(struct fuse_init_out); +void FuseTest::ServerSendData(uint32_t data) { + EXPECT_THAT(RetryEINTR(write)(sock_[1], &data, sizeof(data)), + SyscallSucceedsWithValue(sizeof(data))); +} + +// Reads FuseTestCmd sent from testing thread and routes to correct handler. +// Since each command should be a blocking operation, a `ServerCompleteWith()` +// is required after the switch keyword. +void FuseTest::ServerHandleCommand() { + uint32_t cmd; + EXPECT_THAT(RetryEINTR(read)(sock_[1], &cmd, sizeof(cmd)), + SyscallSucceedsWithValue(sizeof(cmd))); + + switch (static_cast<FuseTestCmd>(cmd)) { + case FuseTestCmd::kSetResponse: + ServerReceiveResponse(); + break; + case FuseTestCmd::kSetInodeLookup: + ServerReceiveInodeLookup(); + break; + case FuseTestCmd::kGetRequest: + ServerSendReceivedRequest(); + break; + case FuseTestCmd::kGetTotalReceivedBytes: + ServerSendData(static_cast<uint32_t>(requests_.UsedBytes())); + break; + case FuseTestCmd::kGetNumUnconsumedRequests: + ServerSendData(static_cast<uint32_t>(requests_.RemainingBlocks())); + break; + case FuseTestCmd::kGetNumUnsentResponses: + ServerSendData(static_cast<uint32_t>(responses_.RemainingBlocks())); + break; + case FuseTestCmd::kSkipRequest: + ServerSkipReceivedRequest(); + break; default: + FAIL() << "Unknown FuseTestCmd " << cmd; break; } - return 0; + + ServerCompleteWith(!HasFailure()); +} + +// Reads the expected file mode and the path of one file. Crafts a basic +// `fuse_entry_out` memory block and inserts into a map for future use. +// The FUSE server will always return this response if a FUSE_LOOKUP +// request with this specific path comes in. +void FuseTest::ServerReceiveInodeLookup() { + mode_t mode; + uint64_t size; + std::vector<char> buf(FUSE_MIN_READ_BUFFER); + + EXPECT_THAT(RetryEINTR(read)(sock_[1], &mode, sizeof(mode)), + SyscallSucceedsWithValue(sizeof(mode))); + + EXPECT_THAT(RetryEINTR(read)(sock_[1], &size, sizeof(size)), + SyscallSucceedsWithValue(sizeof(size))); + + EXPECT_THAT(RetryEINTR(read)(sock_[1], buf.data(), buf.size()), + SyscallSucceeds()); + + std::string path(buf.data()); + + uint32_t out_len = + sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out); + struct fuse_out_header out_header = { + .len = out_len, + .error = 0, + }; + struct fuse_entry_out out_payload = DefaultEntryOut(mode, nodeid_); + // Since this is only used in test, nodeid_ is simply increased by 1 to + // comply with the unqiueness of different path. + ++nodeid_; + + // Set the size. + out_payload.attr.size = size; + + memcpy(buf.data(), &out_header, sizeof(out_header)); + memcpy(buf.data() + sizeof(out_header), &out_payload, sizeof(out_payload)); + lookups_.AddMemBlock(FUSE_LOOKUP, buf.data(), out_len); + lookup_map_[path] = lookups_.Next(); +} + +// Sends the received request pointed by current cursor and advances cursor. +void FuseTest::ServerSendReceivedRequest() { + if (requests_.End()) { + FAIL() << "No more received request."; + return; + } + auto mem_block = requests_.Next(); + EXPECT_THAT( + RetryEINTR(write)(sock_[1], requests_.DataAtOffset(mem_block.offset), + mem_block.len), + SyscallSucceedsWithValue(mem_block.len)); +} + +// Skip the request pointed by current cursor. +void FuseTest::ServerSkipReceivedRequest() { + if (requests_.End()) { + FAIL() << "No more received request."; + return; + } + requests_.Next(); +} + +// Handles FUSE request. Reads request from /dev/fuse, checks if it has the +// same opcode as expected, and responds with the saved fake FUSE response. +// The FUSE request is copied to the serial buffer and can be retrieved one- +// by-one by calling GetServerActualRequest from testing thread. +void FuseTest::ServerProcessFuseRequest() { + ssize_t len; + std::vector<char> buf(FUSE_MIN_READ_BUFFER); + + // Read FUSE request. + EXPECT_THAT(len = RetryEINTR(read)(dev_fd_, buf.data(), buf.size()), + SyscallSucceeds()); + fuse_in_header* in_header = reinterpret_cast<fuse_in_header*>(buf.data()); + + // Check if this is a preset FUSE_LOOKUP path. + if (in_header->opcode == FUSE_LOOKUP) { + std::string path(buf.data() + sizeof(struct fuse_in_header)); + auto it = lookup_map_.find(path); + if (it != lookup_map_.end()) { + // Matches a preset path. Reply with fake data and skip saving the + // request. + ServerRespondFuseSuccess(lookups_, it->second, in_header->unique); + return; + } + } + + requests_.AddMemBlock(in_header->opcode, buf.data(), len); + + if (in_header->opcode == FUSE_RELEASE || in_header->opcode == FUSE_RELEASEDIR) + return; + // Check if there is a corresponding response. + if (responses_.End()) { + GTEST_NONFATAL_FAILURE_("No more FUSE response is expected"); + ServerRespondFuseError(in_header->unique); + return; + } + auto mem_block = responses_.Next(); + if (in_header->opcode != mem_block.opcode) { + std::string message = absl::StrFormat("Expect opcode %d but got %d", + mem_block.opcode, in_header->opcode); + GTEST_NONFATAL_FAILURE_(message.c_str()); + // We won't get correct response if opcode is not expected. Send error + // response here to avoid wrong parsing by VFS. + ServerRespondFuseError(in_header->unique); + return; + } + + // Write FUSE response. + ServerRespondFuseSuccess(responses_, mem_block, in_header->unique); +} + +void FuseTest::ServerRespondFuseSuccess(FuseMemBuffer& mem_buf, + const FuseMemBlock& block, + uint64_t unique) { + fuse_out_header* out_header = + reinterpret_cast<fuse_out_header*>(mem_buf.DataAtOffset(block.offset)); + + // Patch `unique` in fuse_out_header to avoid EINVAL caused by responding + // with an unknown `unique`. + out_header->unique = unique; + EXPECT_THAT(RetryEINTR(write)(dev_fd_, out_header, block.len), + SyscallSucceedsWithValue(block.len)); +} + +void FuseTest::ServerRespondFuseError(uint64_t unique) { + fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header), + .error = ENOSYS, + .unique = unique, + }; + EXPECT_THAT(RetryEINTR(write)(dev_fd_, &out_header, sizeof(out_header)), + SyscallSucceedsWithValue(sizeof(out_header))); } } // namespace testing diff --git a/test/fuse/linux/fuse_base.h b/test/fuse/linux/fuse_base.h index 3a2f255a9..6ad296ca2 100644 --- a/test/fuse/linux/fuse_base.h +++ b/test/fuse/linux/fuse_base.h @@ -16,8 +16,12 @@ #define GVISOR_TEST_FUSE_FUSE_BASE_H_ #include <linux/fuse.h> +#include <string.h> +#include <sys/stat.h> #include <sys/uio.h> +#include <iostream> +#include <unordered_map> #include <vector> #include "gtest/gtest.h" @@ -29,68 +33,216 @@ namespace testing { constexpr char kMountOpts[] = "rootmode=755,user_id=0,group_id=0"; -class FuseTest : public ::testing::Test { +constexpr struct fuse_init_out kDefaultFUSEInitOutPayload = {.major = 7}; + +// Internal commands used to communicate between testing thread and the FUSE +// server. See test/fuse/README.md for further detail. +enum class FuseTestCmd { + kSetResponse = 0, + kSetInodeLookup, + kGetRequest, + kGetNumUnconsumedRequests, + kGetNumUnsentResponses, + kGetTotalReceivedBytes, + kSkipRequest, +}; + +// Holds the information of a memory block in a serial buffer. +struct FuseMemBlock { + uint32_t opcode; + size_t offset; + size_t len; +}; + +// A wrapper of a simple serial buffer that can be used with read(2) and +// write(2). Contains a cursor to indicate accessing. This class is not thread- +// safe and can only be used in single-thread version. +class FuseMemBuffer { public: - FuseTest() { - buf_.resize(FUSE_MIN_READ_BUFFER); - mem_in_.resize(FUSE_MIN_READ_BUFFER); - mem_out_.resize(FUSE_MIN_READ_BUFFER); + FuseMemBuffer() : cursor_(0) { + // To read from /dev/fuse, a buffer needs at least FUSE_MIN_READ_BUFFER + // bytes to avoid EINVAL. FuseMemBuffer holds memory that can accommodate + // a sequence of FUSE request/response, so it is initiated with double + // minimal requirement. + mem_.resize(FUSE_MIN_READ_BUFFER * 2); } - void SetUp() override; - void TearDown() override; - // CompareRequest is used by the FUSE server and should be implemented to - // compare different FUSE operations. It compares the actual FUSE input - // request with the expected one set by `SetExpected()`. - virtual bool CompareRequest(void* expected_mem, size_t expected_len, - void* real_mem, size_t real_len); + // Returns whether there is no memory block. + bool Empty() { return blocks_.empty(); } + + // Returns if there is no more remaining memory blocks. + bool End() { return cursor_ == blocks_.size(); } + + // Returns how many bytes that have been received. + size_t UsedBytes() { + return Empty() ? 0 : blocks_.back().offset + blocks_.back().len; + } + + // Returns the available bytes remains in the serial buffer. + size_t AvailBytes() { return mem_.size() - UsedBytes(); } + + // Appends a memory block information that starts at the tail of the serial + // buffer. /dev/fuse requires at least FUSE_MIN_READ_BUFFER bytes to read, or + // it will issue EINVAL. If it is not enough, just double the buffer length. + void AddMemBlock(uint32_t opcode, void* data, size_t len) { + if (AvailBytes() < FUSE_MIN_READ_BUFFER) { + mem_.resize(mem_.size() << 1); + } + size_t offset = UsedBytes(); + memcpy(mem_.data() + offset, data, len); + blocks_.push_back(FuseMemBlock{opcode, offset, len}); + } + + // Returns the memory address at a specific offset. Used with read(2) or + // write(2). + char* DataAtOffset(size_t offset) { return mem_.data() + offset; } + + // Returns current memory block pointed by the cursor and increase by 1. + FuseMemBlock Next() { + if (End()) { + std::cerr << "Buffer is already exhausted." << std::endl; + return FuseMemBlock{}; + } + return blocks_[cursor_++]; + } + + // Returns the number of the blocks that has not been requested. + size_t RemainingBlocks() { return blocks_.size() - cursor_; } + + private: + size_t cursor_; + std::vector<FuseMemBlock> blocks_; + std::vector<char> mem_; +}; - // SetExpected is called by the testing main thread. Writes a request- - // response pair into FUSE server's member variables via pipe. - void SetExpected(struct iovec* iov_in, int iov_in_cnt, struct iovec* iov_out, - int iov_out_cnt); +// FuseTest base class is useful in FUSE integration test. Inherit this class +// to automatically set up a fake FUSE server and use the member functions +// to manipulate with it. Refer to test/fuse/README.md for detailed explanation. +class FuseTest : public ::testing::Test { + public: + // nodeid_ is the ID of a fake inode. We starts from 2 since 1 is occupied by + // the mount point. + FuseTest() : nodeid_(2) {} + void SetUp() override; + void TearDown() override; - // WaitCompleted waits for FUSE server to complete its processing. It - // complains if the FUSE server responds failure during tests. - void WaitCompleted(); + // Called by the testing thread to set up a fake response for an expected + // opcode via socket. This can be used multiple times to define a sequence of + // expected FUSE reactions. + void SetServerResponse(uint32_t opcode, std::vector<struct iovec>& iovecs); + + // Called by the testing thread to install a fake path under the mount point. + // e.g. a file under /mnt/dir/file and moint point is /mnt, then it will look + // up "dir/file" in this case. + // + // It sets a fixed response to the FUSE_LOOKUP requests issued with this + // path, pretending there is an inode and avoid ENOENT when testing. If mode + // is not given, it creates a regular file with mode 0600. + void SetServerInodeLookup(const std::string& path, + mode_t mode = S_IFREG | S_IRUSR | S_IWUSR, + uint64_t size = 512); + + // Called by the testing thread to ask the FUSE server for its next received + // FUSE request. Be sure to use the corresponding struct of iovec to receive + // data from server. + void GetServerActualRequest(std::vector<struct iovec>& iovecs); + + // Called by the testing thread to query the number of unconsumed requests in + // the requests_ serial buffer of the FUSE server. TearDown() ensures all + // FUSE requests received by the FUSE server were consumed by the testing + // thread. + uint32_t GetServerNumUnconsumedRequests(); + + // Called by the testing thread to query the number of unsent responses in + // the responses_ serial buffer of the FUSE server. TearDown() ensures all + // preset FUSE responses were sent out by the FUSE server. + uint32_t GetServerNumUnsentResponses(); + + // Called by the testing thread to ask the FUSE server for its total received + // bytes from /dev/fuse. + uint32_t GetServerTotalReceivedBytes(); + + // Called by the testing thread to ask the FUSE server to skip stored + // request data. + void SkipServerActualRequest(); protected: TempPath mount_point_; - private: - void MountFuse(); + // Opens /dev/fuse and inherit the file descriptor for the FUSE server. + void MountFuse(const char* mountOpts = kMountOpts); + + // Creates a socketpair for communication and forks FUSE server. + void SetUpFuseServer( + const struct fuse_init_out* payload = &kDefaultFUSEInitOutPayload); + + // Unmounts the mountpoint of the FUSE server. void UnmountFuse(); - // ConsumeFuseInit is only used during FUSE server setup. - PosixError ConsumeFuseInit(); + private: + // Sends a FuseTestCmd and gets a uint32_t data from the FUSE server. + inline uint32_t GetServerData(uint32_t cmd); + + // Waits for FUSE server to complete its processing. Complains if the FUSE + // server responds any failure during tests. + void WaitServerComplete(); - // ReceiveExpected is the FUSE server side's corresponding code of - // `SetExpected()`. Save the request-response pair into its memory. - void ReceiveExpected(); + // The FUSE server stays here and waits next command or FUSE request until it + // is terminated. + void ServerFuseLoop(); - // MarkDone is used by the FUSE server to tell testing main if it's OK to - // proceed next command. - void MarkDone(bool success); + // Used by the FUSE server to tell testing thread if it is OK to proceed next + // command. Will be issued after processing each FuseTestCmd. + void ServerCompleteWith(bool success); - // FuseLoop is where the FUSE server stay until it is terminated. - void FuseLoop(); + // Consumes the first FUSE request when mounting FUSE. Replies with a + // response with empty payload. + PosixError ServerConsumeFuseInit(const struct fuse_init_out* payload); - // SetUpFuseServer creates 2 pipes for communication and forks FUSE server. - void SetUpFuseServer(); + // A command switch that dispatch different FuseTestCmd to its handler. + void ServerHandleCommand(); - // GetPayloadSize is a helper function to get the number of bytes of a - // specific FUSE operation struct. - size_t GetPayloadSize(uint32_t opcode, bool in); + // The FUSE server side's corresponding code of `SetServerResponse()`. + // Handles `kSetResponse` command. Saves the fake response into its output + // memory queue. + void ServerReceiveResponse(); + + // The FUSE server side's corresponding code of `SetServerInodeLookup()`. + // Handles `kSetInodeLookup` command. Receives an expected file mode and + // file path under the mount point. + void ServerReceiveInodeLookup(); + + // The FUSE server side's corresponding code of `GetServerActualRequest()`. + // Handles `kGetRequest` command. Sends the next received request pointed by + // the cursor. + void ServerSendReceivedRequest(); + + // Sends a uint32_t data via socket. + inline void ServerSendData(uint32_t data); + + // The FUSE server side's corresponding code of `SkipServerActualRequest()`. + // Handles `kSkipRequest` command. Skip the request pointed by current cursor. + void ServerSkipReceivedRequest(); + + // Handles FUSE request sent to /dev/fuse by its saved responses. + void ServerProcessFuseRequest(); + + // Responds to FUSE request with a saved data. + void ServerRespondFuseSuccess(FuseMemBuffer& mem_buf, + const FuseMemBlock& block, uint64_t unique); + + // Responds an error header to /dev/fuse when bad thing happens. + void ServerRespondFuseError(uint64_t unique); int dev_fd_; - int set_expected_[2]; - int done_[2]; - - std::vector<char> buf_; - std::vector<char> mem_in_; - std::vector<char> mem_out_; - ssize_t len_in_; - ssize_t len_out_; + int sock_[2]; + + uint64_t nodeid_; + std::unordered_map<std::string, FuseMemBlock> lookup_map_; + + FuseMemBuffer requests_; + FuseMemBuffer responses_; + FuseMemBuffer lookups_; }; } // namespace testing diff --git a/test/fuse/linux/fuse_fd_util.cc b/test/fuse/linux/fuse_fd_util.cc new file mode 100644 index 000000000..30d1157bb --- /dev/null +++ b/test/fuse/linux/fuse_fd_util.cc @@ -0,0 +1,61 @@ +// 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. + +#include "test/fuse/linux/fuse_fd_util.h" + +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/types.h> +#include <sys/uio.h> + +#include <string> +#include <vector> + +#include "test/util/cleanup.h" +#include "test/util/file_descriptor.h" +#include "test/util/fuse_util.h" +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +PosixErrorOr<FileDescriptor> FuseFdTest::OpenPath(const std::string &path, + uint32_t flags, uint64_t fh) { + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + struct fuse_open_out out_payload = { + .fh = fh, + .open_flags = flags, + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_OPEN, iov_out); + + auto res = Open(path.c_str(), flags); + if (res.ok()) { + SkipServerActualRequest(); + } + return res; +} + +Cleanup FuseFdTest::CloseFD(FileDescriptor &fd) { + return Cleanup([&] { + close(fd.release()); + SkipServerActualRequest(); + }); +} + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/fuse_fd_util.h b/test/fuse/linux/fuse_fd_util.h new file mode 100644 index 000000000..066185c94 --- /dev/null +++ b/test/fuse/linux/fuse_fd_util.h @@ -0,0 +1,48 @@ +// 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. + +#ifndef GVISOR_TEST_FUSE_FUSE_FD_UTIL_H_ +#define GVISOR_TEST_FUSE_FUSE_FD_UTIL_H_ + +#include <fcntl.h> +#include <sys/stat.h> +#include <sys/types.h> + +#include <string> + +#include "test/fuse/linux/fuse_base.h" +#include "test/util/cleanup.h" +#include "test/util/file_descriptor.h" +#include "test/util/posix_error.h" + +namespace gvisor { +namespace testing { + +class FuseFdTest : public FuseTest { + public: + // Sets the FUSE server to respond to a FUSE_OPEN with corresponding flags and + // fh. Then does a real file system open on the absolute path to get an fd. + PosixErrorOr<FileDescriptor> OpenPath(const std::string &path, + uint32_t flags = O_RDONLY, + uint64_t fh = 1); + + // Returns a cleanup object that closes the fd when it is destroyed. After + // the close is done, tells the FUSE server to skip this FUSE_RELEASE. + Cleanup CloseFD(FileDescriptor &fd); +}; + +} // namespace testing +} // namespace gvisor + +#endif // GVISOR_TEST_FUSE_FUSE_FD_UTIL_H_ diff --git a/test/fuse/linux/mkdir_test.cc b/test/fuse/linux/mkdir_test.cc new file mode 100644 index 000000000..9647cb93f --- /dev/null +++ b/test/fuse/linux/mkdir_test.cc @@ -0,0 +1,88 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/temp_umask.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class MkdirTest : public FuseTest { + protected: + const std::string test_dir_ = "test_dir"; + const mode_t perms_ = S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(MkdirTest, CreateDir) { + const std::string test_dir_path_ = + JoinPath(mount_point_.path().c_str(), test_dir_); + const mode_t new_umask = 0077; + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFDIR | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_MKDIR, iov_out); + TempUmask mask(new_umask); + ASSERT_THAT(mkdir(test_dir_path_.c_str(), 0777), SyscallSucceeds()); + + struct fuse_in_header in_header; + struct fuse_mkdir_in in_payload; + std::vector<char> actual_dir(test_dir_.length() + 1); + auto iov_in = FuseGenerateIovecs(in_header, in_payload, actual_dir); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, + sizeof(in_header) + sizeof(in_payload) + test_dir_.length() + 1); + EXPECT_EQ(in_header.opcode, FUSE_MKDIR); + EXPECT_EQ(in_payload.mode & 0777, perms_ & ~new_umask); + EXPECT_EQ(in_payload.umask, new_umask); + EXPECT_EQ(std::string(actual_dir.data()), test_dir_); +} + +TEST_F(MkdirTest, FileTypeError) { + const std::string test_dir_path_ = + JoinPath(mount_point_.path().c_str(), test_dir_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFREG | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_MKDIR, iov_out); + ASSERT_THAT(mkdir(test_dir_path_.c_str(), 0777), SyscallFailsWithErrno(EIO)); + SkipServerActualRequest(); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/mknod_test.cc b/test/fuse/linux/mknod_test.cc new file mode 100644 index 000000000..74c74d76b --- /dev/null +++ b/test/fuse/linux/mknod_test.cc @@ -0,0 +1,107 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/temp_umask.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class MknodTest : public FuseTest { + protected: + const std::string test_file_ = "test_file"; + const mode_t perms_ = S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(MknodTest, RegularFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + const mode_t new_umask = 0077; + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFREG | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_MKNOD, iov_out); + TempUmask mask(new_umask); + ASSERT_THAT(mknod(test_file_path.c_str(), perms_, 0), SyscallSucceeds()); + + struct fuse_in_header in_header; + struct fuse_mknod_in in_payload; + std::vector<char> actual_file(test_file_.length() + 1); + auto iov_in = FuseGenerateIovecs(in_header, in_payload, actual_file); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, + sizeof(in_header) + sizeof(in_payload) + test_file_.length() + 1); + EXPECT_EQ(in_header.opcode, FUSE_MKNOD); + EXPECT_EQ(in_payload.mode & 0777, perms_ & ~new_umask); + EXPECT_EQ(in_payload.umask, new_umask); + EXPECT_EQ(in_payload.rdev, 0); + EXPECT_EQ(std::string(actual_file.data()), test_file_); +} + +TEST_F(MknodTest, FileTypeError) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + // server return directory instead of regular file should cause an error. + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFDIR | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_MKNOD, iov_out); + ASSERT_THAT(mknod(test_file_path.c_str(), perms_, 0), + SyscallFailsWithErrno(EIO)); + SkipServerActualRequest(); +} + +TEST_F(MknodTest, NodeIDError) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = + DefaultEntryOut(S_IFREG | perms_, FUSE_ROOT_ID); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_MKNOD, iov_out); + ASSERT_THAT(mknod(test_file_path.c_str(), perms_, 0), + SyscallFailsWithErrno(EIO)); + SkipServerActualRequest(); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/open_test.cc b/test/fuse/linux/open_test.cc new file mode 100644 index 000000000..4b0c4a805 --- /dev/null +++ b/test/fuse/linux/open_test.cc @@ -0,0 +1,128 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class OpenTest : public FuseTest { + // OpenTest doesn't care the release request when close a fd, + // so doesn't check leftover requests when tearing down. + void TearDown() { UnmountFuse(); } + + protected: + const std::string test_file_ = "test_file"; + const mode_t regular_file_ = S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO; + + struct fuse_open_out out_payload_ = { + .fh = 1, + .open_flags = O_RDWR, + }; +}; + +TEST_F(OpenTest, RegularFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, regular_file_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload_); + SetServerResponse(FUSE_OPEN, iov_out); + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_path.c_str(), O_RDWR)); + + struct fuse_in_header in_header; + struct fuse_open_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_OPEN); + EXPECT_EQ(in_payload.flags, O_RDWR); + EXPECT_THAT(fcntl(fd.get(), F_GETFL), SyscallSucceedsWithValue(O_RDWR)); +} + +TEST_F(OpenTest, SetNoOpen) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, regular_file_); + + // ENOSYS indicates open is not implemented. + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + .error = -ENOSYS, + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload_); + SetServerResponse(FUSE_OPEN, iov_out); + ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_path.c_str(), O_RDWR)); + SkipServerActualRequest(); + + // check open doesn't send new request. + uint32_t recieved_before = GetServerTotalReceivedBytes(); + ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_path.c_str(), O_RDWR)); + EXPECT_EQ(GetServerTotalReceivedBytes(), recieved_before); +} + +TEST_F(OpenTest, OpenFail) { + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + .error = -ENOENT, + }; + + auto iov_out = FuseGenerateIovecs(out_header, out_payload_); + SetServerResponse(FUSE_OPENDIR, iov_out); + ASSERT_THAT(open(mount_point_.path().c_str(), O_RDWR), + SyscallFailsWithErrno(ENOENT)); + + struct fuse_in_header in_header; + struct fuse_open_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_OPENDIR); + EXPECT_EQ(in_payload.flags, O_RDWR); +} + +TEST_F(OpenTest, DirectoryFlagOnRegularFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + + SetServerInodeLookup(test_file_, regular_file_); + ASSERT_THAT(open(test_file_path.c_str(), O_RDWR | O_DIRECTORY), + SyscallFailsWithErrno(ENOTDIR)); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/read_test.cc b/test/fuse/linux/read_test.cc new file mode 100644 index 000000000..88fc299d8 --- /dev/null +++ b/test/fuse/linux/read_test.cc @@ -0,0 +1,390 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class ReadTest : public FuseTest { + void SetUp() override { + FuseTest::SetUp(); + test_file_path_ = JoinPath(mount_point_.path().c_str(), test_file_); + } + + // TearDown overrides the parent's function + // to skip checking the unconsumed release request at the end. + void TearDown() override { UnmountFuse(); } + + protected: + const std::string test_file_ = "test_file"; + const mode_t test_file_mode_ = S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO; + const uint64_t test_fh_ = 1; + const uint32_t open_flag_ = O_RDWR; + + std::string test_file_path_; + + PosixErrorOr<FileDescriptor> OpenTestFile(const std::string &path, + uint64_t size = 512) { + SetServerInodeLookup(test_file_, test_file_mode_, size); + + struct fuse_out_header out_header_open = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + struct fuse_open_out out_payload_open = { + .fh = test_fh_, + .open_flags = open_flag_, + }; + auto iov_out_open = FuseGenerateIovecs(out_header_open, out_payload_open); + SetServerResponse(FUSE_OPEN, iov_out_open); + + auto res = Open(path.c_str(), open_flag_); + if (res.ok()) { + SkipServerActualRequest(); + } + return res; + } +}; + +class ReadTestSmallMaxRead : public ReadTest { + void SetUp() override { + MountFuse(mountOpts); + SetUpFuseServer(); + test_file_path_ = JoinPath(mount_point_.path().c_str(), test_file_); + } + + protected: + constexpr static char mountOpts[] = + "rootmode=755,user_id=0,group_id=0,max_read=4096"; + // 4096 is hard-coded as the max_read in mount options. + const int size_fragment = 4096; +}; + +TEST_F(ReadTest, ReadWhole) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the read. + const int n_read = 5; + std::vector<char> data(n_read); + RandomizeBuffer(data.data(), data.size()); + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + SetServerResponse(FUSE_READ, iov_out_read); + + // Read the whole "file". + std::vector<char> buf(n_read); + EXPECT_THAT(read(fd.get(), buf.data(), n_read), + SyscallSucceedsWithValue(n_read)); + + // Check the read request. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, 0); + EXPECT_EQ(buf, data); +} + +TEST_F(ReadTest, ReadPartial) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the read. + const int n_data = 10; + std::vector<char> data(n_data); + RandomizeBuffer(data.data(), data.size()); + // Note: due to read ahead, current read implementation will treat any + // response that is longer than requested as correct (i.e. not reach the EOF). + // Therefore, the test below should make sure the size to read does not exceed + // n_data. + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + + std::vector<char> buf(n_data); + + // Read 1 bytes. + SetServerResponse(FUSE_READ, iov_out_read); + EXPECT_THAT(read(fd.get(), buf.data(), 1), SyscallSucceedsWithValue(1)); + + // Check the 1-byte read request. + GetServerActualRequest(iov_in); + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, 0); + + // Read 3 bytes. + SetServerResponse(FUSE_READ, iov_out_read); + EXPECT_THAT(read(fd.get(), buf.data(), 3), SyscallSucceedsWithValue(3)); + + // Check the 3-byte read request. + GetServerActualRequest(iov_in); + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_payload_read.offset, 1); + + // Read 5 bytes. + SetServerResponse(FUSE_READ, iov_out_read); + EXPECT_THAT(read(fd.get(), buf.data(), 5), SyscallSucceedsWithValue(5)); + + // Check the 5-byte read request. + GetServerActualRequest(iov_in); + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_payload_read.offset, 4); +} + +TEST_F(ReadTest, PRead) { + const int file_size = 512; + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_, file_size)); + + // Prepare for the read. + const int n_read = 5; + std::vector<char> data(n_read); + RandomizeBuffer(data.data(), data.size()); + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + SetServerResponse(FUSE_READ, iov_out_read); + + // Read some bytes. + std::vector<char> buf(n_read); + const int offset_read = file_size >> 1; + EXPECT_THAT(pread(fd.get(), buf.data(), n_read, offset_read), + SyscallSucceedsWithValue(n_read)); + + // Check the read request. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, offset_read); + EXPECT_EQ(buf, data); +} + +TEST_F(ReadTest, ReadZero) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Issue the read. + std::vector<char> buf; + EXPECT_THAT(read(fd.get(), buf.data(), 0), SyscallSucceedsWithValue(0)); +} + +TEST_F(ReadTest, ReadShort) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the short read. + const int n_read = 5; + std::vector<char> data(n_read >> 1); + RandomizeBuffer(data.data(), data.size()); + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + SetServerResponse(FUSE_READ, iov_out_read); + + // Read the whole "file". + std::vector<char> buf(n_read); + EXPECT_THAT(read(fd.get(), buf.data(), n_read), + SyscallSucceedsWithValue(data.size())); + + // Check the read request. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, 0); + std::vector<char> short_buf(buf.begin(), buf.begin() + data.size()); + EXPECT_EQ(short_buf, data); +} + +TEST_F(ReadTest, ReadShortEOF) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the short read. + struct fuse_out_header out_header_read = { + .len = static_cast<uint32_t>(sizeof(struct fuse_out_header)), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read); + SetServerResponse(FUSE_READ, iov_out_read); + + // Read the whole "file". + const int n_read = 10; + std::vector<char> buf(n_read); + EXPECT_THAT(read(fd.get(), buf.data(), n_read), SyscallSucceedsWithValue(0)); + + // Check the read request. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, 0); +} + +TEST_F(ReadTestSmallMaxRead, ReadSmallMaxRead) { + const int n_fragment = 10; + const int n_read = size_fragment * n_fragment; + + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_, n_read)); + + // Prepare for the read. + std::vector<char> data(size_fragment); + RandomizeBuffer(data.data(), data.size()); + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + + for (int i = 0; i < n_fragment; ++i) { + SetServerResponse(FUSE_READ, iov_out_read); + } + + // Read the whole "file". + std::vector<char> buf(n_read); + EXPECT_THAT(read(fd.get(), buf.data(), n_read), + SyscallSucceedsWithValue(n_read)); + + ASSERT_EQ(GetServerNumUnsentResponses(), 0); + ASSERT_EQ(GetServerNumUnconsumedRequests(), n_fragment); + + // Check each read segment. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + + for (int i = 0; i < n_fragment; ++i) { + GetServerActualRequest(iov_in); + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, i * size_fragment); + EXPECT_EQ(in_payload_read.size, size_fragment); + + auto it = buf.begin() + i * size_fragment; + EXPECT_EQ(std::vector<char>(it, it + size_fragment), data); + } +} + +TEST_F(ReadTestSmallMaxRead, ReadSmallMaxReadShort) { + const int n_fragment = 10; + const int n_read = size_fragment * n_fragment; + + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_, n_read)); + + // Prepare for the read. + std::vector<char> data(size_fragment); + RandomizeBuffer(data.data(), data.size()); + struct fuse_out_header out_header_read = { + .len = + static_cast<uint32_t>(sizeof(struct fuse_out_header) + data.size()), + }; + auto iov_out_read = FuseGenerateIovecs(out_header_read, data); + + for (int i = 0; i < n_fragment - 1; ++i) { + SetServerResponse(FUSE_READ, iov_out_read); + } + + // The last fragment is a short read. + std::vector<char> half_data(data.begin(), data.begin() + (data.size() >> 1)); + struct fuse_out_header out_header_read_short = { + .len = static_cast<uint32_t>(sizeof(struct fuse_out_header) + + half_data.size()), + }; + auto iov_out_read_short = + FuseGenerateIovecs(out_header_read_short, half_data); + SetServerResponse(FUSE_READ, iov_out_read_short); + + // Read the whole "file". + std::vector<char> buf(n_read); + EXPECT_THAT(read(fd.get(), buf.data(), n_read), + SyscallSucceedsWithValue(n_read - (data.size() >> 1))); + + ASSERT_EQ(GetServerNumUnsentResponses(), 0); + ASSERT_EQ(GetServerNumUnconsumedRequests(), n_fragment); + + // Check each read segment. + struct fuse_in_header in_header_read; + struct fuse_read_in in_payload_read; + auto iov_in = FuseGenerateIovecs(in_header_read, in_payload_read); + + for (int i = 0; i < n_fragment; ++i) { + GetServerActualRequest(iov_in); + EXPECT_EQ(in_payload_read.fh, test_fh_); + EXPECT_EQ(in_header_read.len, + sizeof(in_header_read) + sizeof(in_payload_read)); + EXPECT_EQ(in_header_read.opcode, FUSE_READ); + EXPECT_EQ(in_payload_read.offset, i * size_fragment); + EXPECT_EQ(in_payload_read.size, size_fragment); + + auto it = buf.begin() + i * size_fragment; + if (i != n_fragment - 1) { + EXPECT_EQ(std::vector<char>(it, it + data.size()), data); + } else { + EXPECT_EQ(std::vector<char>(it, it + half_data.size()), half_data); + } + } +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/readdir_test.cc b/test/fuse/linux/readdir_test.cc new file mode 100644 index 000000000..2afb4b062 --- /dev/null +++ b/test/fuse/linux/readdir_test.cc @@ -0,0 +1,193 @@ +// 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. + +#include <dirent.h> +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <linux/unistd.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +#define FUSE_NAME_OFFSET offsetof(struct fuse_dirent, name) +#define FUSE_DIRENT_ALIGN(x) \ + (((x) + sizeof(uint64_t) - 1) & ~(sizeof(uint64_t) - 1)) +#define FUSE_DIRENT_SIZE(d) FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + (d)->namelen) + +namespace gvisor { +namespace testing { + +namespace { + +class ReaddirTest : public FuseTest { + public: + void fill_fuse_dirent(char *buf, const char *name, uint64_t ino) { + size_t namelen = strlen(name); + size_t entlen = FUSE_NAME_OFFSET + namelen; + size_t entlen_padded = FUSE_DIRENT_ALIGN(entlen); + struct fuse_dirent *dirent; + + dirent = reinterpret_cast<struct fuse_dirent *>(buf); + dirent->ino = ino; + dirent->namelen = namelen; + memcpy(dirent->name, name, namelen); + memset(dirent->name + namelen, 0, entlen_padded - entlen); + } + + protected: + const std::string test_dir_name_ = "test_dir"; +}; + +TEST_F(ReaddirTest, SingleEntry) { + const std::string test_dir_path = + JoinPath(mount_point_.path().c_str(), test_dir_name_); + + const uint64_t ino_dir = 1024; + // We need to make sure the test dir is a directory that can be found. + mode_t expected_mode = + S_IFDIR | S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH; + struct fuse_attr dir_attr = { + .ino = ino_dir, + .size = 512, + .blocks = 4, + .mode = expected_mode, + .blksize = 4096, + }; + + // We need to make sure the test dir is a directory that can be found. + struct fuse_out_header lookup_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out lookup_payload = { + .nodeid = 1, + .entry_valid = true, + .attr_valid = true, + .attr = dir_attr, + }; + + struct fuse_out_header open_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + struct fuse_open_out open_payload = { + .fh = 1, + }; + auto iov_out = FuseGenerateIovecs(lookup_header, lookup_payload); + SetServerResponse(FUSE_LOOKUP, iov_out); + + iov_out = FuseGenerateIovecs(open_header, open_payload); + SetServerResponse(FUSE_OPENDIR, iov_out); + + FileDescriptor fd = + ASSERT_NO_ERRNO_AND_VALUE(Open(test_dir_path.c_str(), O_RDONLY)); + + // The open command makes two syscalls. Lookup the dir file and open. + // We don't need to inspect those headers in this test. + SkipServerActualRequest(); // LOOKUP. + SkipServerActualRequest(); // OPENDIR. + + // Readdir test code. + std::string dot = "."; + std::string dot_dot = ".."; + std::string test_file = "testFile"; + + // Figure out how many dirents to send over and allocate them appropriately. + // Each dirent has a dynamic name and a static metadata part. The dirent size + // is aligned to being a multiple of 8. + size_t dot_file_dirent_size = + FUSE_DIRENT_ALIGN(dot.length() + FUSE_NAME_OFFSET); + size_t dot_dot_file_dirent_size = + FUSE_DIRENT_ALIGN(dot_dot.length() + FUSE_NAME_OFFSET); + size_t test_file_dirent_size = + FUSE_DIRENT_ALIGN(test_file.length() + FUSE_NAME_OFFSET); + + // Create an appropriately sized payload. + size_t readdir_payload_size = + test_file_dirent_size + dot_file_dirent_size + dot_dot_file_dirent_size; + std::vector<char> readdir_payload_vec(readdir_payload_size); + char *readdir_payload = readdir_payload_vec.data(); + + // Use fake ino for other directories. + fill_fuse_dirent(readdir_payload, dot.c_str(), ino_dir - 2); + fill_fuse_dirent(readdir_payload + dot_file_dirent_size, dot_dot.c_str(), + ino_dir - 1); + fill_fuse_dirent( + readdir_payload + dot_file_dirent_size + dot_dot_file_dirent_size, + test_file.c_str(), ino_dir); + + struct fuse_out_header readdir_header = { + .len = uint32_t(sizeof(struct fuse_out_header) + readdir_payload_size), + }; + struct fuse_out_header readdir_header_break = { + .len = uint32_t(sizeof(struct fuse_out_header)), + }; + + iov_out = FuseGenerateIovecs(readdir_header, readdir_payload_vec); + SetServerResponse(FUSE_READDIR, iov_out); + + iov_out = FuseGenerateIovecs(readdir_header_break); + SetServerResponse(FUSE_READDIR, iov_out); + + std::vector<char> buf(4090, 0); + int nread, off = 0, i = 0; + EXPECT_THAT( + nread = syscall(__NR_getdents64, fd.get(), buf.data(), buf.size()), + SyscallSucceeds()); + for (; off < nread;) { + struct dirent64 *ent = (struct dirent64 *)(buf.data() + off); + off += ent->d_reclen; + switch (i++) { + case 0: + EXPECT_EQ(std::string(ent->d_name), dot); + break; + case 1: + EXPECT_EQ(std::string(ent->d_name), dot_dot); + break; + case 2: + EXPECT_EQ(std::string(ent->d_name), test_file); + break; + } + } + + EXPECT_THAT( + nread = syscall(__NR_getdents64, fd.get(), buf.data(), buf.size()), + SyscallSucceedsWithValue(0)); + + SkipServerActualRequest(); // READDIR. + SkipServerActualRequest(); // READDIR with no data. + + // Clean up. + fd.reset(-1); + + struct fuse_in_header in_header; + struct fuse_release_in in_payload; + + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_RELEASEDIR); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/readlink_test.cc b/test/fuse/linux/readlink_test.cc new file mode 100644 index 000000000..2cba8fc23 --- /dev/null +++ b/test/fuse/linux/readlink_test.cc @@ -0,0 +1,85 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class ReadlinkTest : public FuseTest { + protected: + const std::string test_file_ = "test_file_"; + const mode_t perms_ = S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(ReadlinkTest, ReadSymLink) { + const std::string symlink_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, S_IFLNK | perms_); + + struct fuse_out_header out_header = { + .len = static_cast<uint32_t>(sizeof(struct fuse_out_header)) + + static_cast<uint32_t>(test_file_.length()) + 1, + }; + std::string link = test_file_; + auto iov_out = FuseGenerateIovecs(out_header, link); + SetServerResponse(FUSE_READLINK, iov_out); + const std::string actual_link = + ASSERT_NO_ERRNO_AND_VALUE(ReadLink(symlink_path)); + + struct fuse_in_header in_header; + auto iov_in = FuseGenerateIovecs(in_header); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header)); + EXPECT_EQ(in_header.opcode, FUSE_READLINK); + EXPECT_EQ(0, memcmp(actual_link.c_str(), link.data(), link.size())); + + // next readlink should have link cached, so shouldn't have new request to + // server. + uint32_t recieved_before = GetServerTotalReceivedBytes(); + ASSERT_NO_ERRNO(ReadLink(symlink_path)); + EXPECT_EQ(GetServerTotalReceivedBytes(), recieved_before); +} + +TEST_F(ReadlinkTest, NotSymlink) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, S_IFREG | perms_); + + std::vector<char> buf(PATH_MAX + 1); + ASSERT_THAT(readlink(test_file_path.c_str(), buf.data(), PATH_MAX), + SyscallFailsWithErrno(EINVAL)); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/release_test.cc b/test/fuse/linux/release_test.cc new file mode 100644 index 000000000..b5adb0870 --- /dev/null +++ b/test/fuse/linux/release_test.cc @@ -0,0 +1,74 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/mount.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class ReleaseTest : public FuseTest { + protected: + const std::string test_file_ = "test_file"; +}; + +TEST_F(ReleaseTest, RegularFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + struct fuse_open_out out_payload = { + .fh = 1, + .open_flags = O_RDWR, + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_OPEN, iov_out); + FileDescriptor fd = ASSERT_NO_ERRNO_AND_VALUE(Open(test_file_path, O_RDWR)); + SkipServerActualRequest(); + ASSERT_THAT(close(fd.release()), SyscallSucceeds()); + + struct fuse_in_header in_header; + struct fuse_release_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_RELEASE); + EXPECT_EQ(in_payload.flags, O_RDWR); + EXPECT_EQ(in_payload.fh, 1); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/rmdir_test.cc b/test/fuse/linux/rmdir_test.cc new file mode 100644 index 000000000..913d3f910 --- /dev/null +++ b/test/fuse/linux/rmdir_test.cc @@ -0,0 +1,73 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <sys/uio.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fs_util.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class RmDirTest : public FuseTest { + protected: + const std::string test_dir_name_ = "test_dir"; + const mode_t test_dir_mode_ = S_IFDIR | S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(RmDirTest, NormalRmDir) { + const std::string test_dir_path_ = + JoinPath(mount_point_.path().c_str(), test_dir_name_); + + SetServerInodeLookup(test_dir_name_, test_dir_mode_); + + // RmDir code. + struct fuse_out_header rmdir_header = { + .len = sizeof(struct fuse_out_header), + }; + + auto iov_out = FuseGenerateIovecs(rmdir_header); + SetServerResponse(FUSE_RMDIR, iov_out); + + ASSERT_THAT(rmdir(test_dir_path_.c_str()), SyscallSucceeds()); + + struct fuse_in_header in_header; + std::vector<char> actual_dirname(test_dir_name_.length() + 1); + auto iov_in = FuseGenerateIovecs(in_header, actual_dirname); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header) + test_dir_name_.length() + 1); + EXPECT_EQ(in_header.opcode, FUSE_RMDIR); + EXPECT_EQ(std::string(actual_dirname.data()), test_dir_name_); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/setstat_test.cc b/test/fuse/linux/setstat_test.cc new file mode 100644 index 000000000..68301c775 --- /dev/null +++ b/test/fuse/linux/setstat_test.cc @@ -0,0 +1,338 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <sys/uio.h> +#include <unistd.h> +#include <utime.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_fd_util.h" +#include "test/util/cleanup.h" +#include "test/util/fs_util.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class SetStatTest : public FuseFdTest { + public: + void SetUp() override { + FuseFdTest::SetUp(); + test_dir_path_ = JoinPath(mount_point_.path(), test_dir_); + test_file_path_ = JoinPath(mount_point_.path(), test_file_); + } + + protected: + const uint64_t fh = 23; + const std::string test_dir_ = "testdir"; + const std::string test_file_ = "testfile"; + const mode_t test_dir_mode_ = S_IFDIR | S_IRUSR | S_IWUSR | S_IXUSR; + const mode_t test_file_mode_ = S_IFREG | S_IRUSR | S_IWUSR | S_IXUSR; + + std::string test_dir_path_; + std::string test_file_path_; +}; + +TEST_F(SetStatTest, ChmodDir) { + // Set up fixture. + SetServerInodeLookup(test_dir_, test_dir_mode_); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + mode_t set_mode = S_IRGRP | S_IWGRP | S_IXGRP; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(set_mode, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(chmod(test_dir_path_.c_str(), set_mode), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_MODE); + EXPECT_EQ(in_payload.mode, S_IFDIR | set_mode); +} + +TEST_F(SetStatTest, ChownDir) { + // Set up fixture. + SetServerInodeLookup(test_dir_, test_dir_mode_); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(test_dir_mode_, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(chown(test_dir_path_.c_str(), 1025, 1025), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_UID | FATTR_GID); + EXPECT_EQ(in_payload.uid, 1025); + EXPECT_EQ(in_payload.gid, 1025); +} + +TEST_F(SetStatTest, TruncateFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(S_IFREG | S_IRUSR | S_IWUSR, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(truncate(test_file_path_.c_str(), 321), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_SIZE); + EXPECT_EQ(in_payload.size, 321); +} + +TEST_F(SetStatTest, UtimeFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(S_IFREG | S_IRUSR | S_IWUSR, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + time_t expected_atime = 1597159766, expected_mtime = 1597159765; + struct utimbuf times = { + .actime = expected_atime, + .modtime = expected_mtime, + }; + EXPECT_THAT(utime(test_file_path_.c_str(), ×), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_ATIME | FATTR_MTIME); + EXPECT_EQ(in_payload.atime, expected_atime); + EXPECT_EQ(in_payload.mtime, expected_mtime); +} + +TEST_F(SetStatTest, UtimesFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(test_file_mode_, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + struct timeval expected_times[2] = { + { + .tv_sec = 1597159766, + .tv_usec = 234945, + }, + { + .tv_sec = 1597159765, + .tv_usec = 232341, + }, + }; + EXPECT_THAT(utimes(test_file_path_.c_str(), expected_times), + SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_ATIME | FATTR_MTIME); + EXPECT_EQ(in_payload.atime, expected_times[0].tv_sec); + EXPECT_EQ(in_payload.atimensec, expected_times[0].tv_usec * 1000); + EXPECT_EQ(in_payload.mtime, expected_times[1].tv_sec); + EXPECT_EQ(in_payload.mtimensec, expected_times[1].tv_usec * 1000); +} + +TEST_F(SetStatTest, FtruncateFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenPath(test_file_path_, O_RDWR, fh)); + auto close_fd = CloseFD(fd); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(test_file_mode_, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(ftruncate(fd.get(), 321), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_SIZE | FATTR_FH); + EXPECT_EQ(in_payload.fh, fh); + EXPECT_EQ(in_payload.size, 321); +} + +TEST_F(SetStatTest, FchmodFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenPath(test_file_path_, O_RDWR, fh)); + auto close_fd = CloseFD(fd); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + mode_t set_mode = S_IROTH | S_IWOTH | S_IXOTH; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(set_mode, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(fchmod(fd.get(), set_mode), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_MODE | FATTR_FH); + EXPECT_EQ(in_payload.fh, fh); + EXPECT_EQ(in_payload.mode, S_IFREG | set_mode); +} + +TEST_F(SetStatTest, FchownFile) { + // Set up fixture. + SetServerInodeLookup(test_file_, test_file_mode_); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenPath(test_file_path_, O_RDWR, fh)); + auto close_fd = CloseFD(fd); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + .error = 0, + }; + struct fuse_attr_out out_payload = { + .attr = DefaultFuseAttr(S_IFREG | S_IRUSR | S_IWUSR | S_IXUSR, 2), + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SETATTR, iov_out); + + // Make syscall. + EXPECT_THAT(fchown(fd.get(), 1025, 1025), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_setattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.len, sizeof(in_header) + sizeof(in_payload)); + EXPECT_EQ(in_header.opcode, FUSE_SETATTR); + EXPECT_EQ(in_header.uid, 0); + EXPECT_EQ(in_header.gid, 0); + EXPECT_EQ(in_payload.valid, FATTR_UID | FATTR_GID | FATTR_FH); + EXPECT_EQ(in_payload.fh, fh); + EXPECT_EQ(in_payload.uid, 1025); + EXPECT_EQ(in_payload.gid, 1025); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/stat_test.cc b/test/fuse/linux/stat_test.cc index 172e09867..6f032cac1 100644 --- a/test/fuse/linux/stat_test.cc +++ b/test/fuse/linux/stat_test.cc @@ -18,12 +18,16 @@ #include <sys/stat.h> #include <sys/statfs.h> #include <sys/types.h> +#include <sys/uio.h> #include <unistd.h> #include <vector> #include "gtest/gtest.h" -#include "test/fuse/linux/fuse_base.h" +#include "test/fuse/linux/fuse_fd_util.h" +#include "test/util/cleanup.h" +#include "test/util/fs_util.h" +#include "test/util/fuse_util.h" #include "test/util/test_util.h" namespace gvisor { @@ -31,89 +35,45 @@ namespace testing { namespace { -class StatTest : public FuseTest { +class StatTest : public FuseFdTest { public: - bool CompareRequest(void* expected_mem, size_t expected_len, void* real_mem, - size_t real_len) override { - if (expected_len != real_len) return false; - struct fuse_in_header* real_header = - reinterpret_cast<fuse_in_header*>(real_mem); - - if (real_header->opcode != FUSE_GETATTR) { - std::cerr << "expect header opcode " << FUSE_GETATTR << " but got " - << real_header->opcode << std::endl; - return false; - } - return true; + void SetUp() override { + FuseFdTest::SetUp(); + test_file_path_ = JoinPath(mount_point_.path(), test_file_); } + protected: bool StatsAreEqual(struct stat expected, struct stat actual) { - // device number will be dynamically allocated by kernel, we cannot know - // in advance + // Device number will be dynamically allocated by kernel, we cannot know in + // advance. actual.st_dev = expected.st_dev; return memcmp(&expected, &actual, sizeof(struct stat)) == 0; } + + const std::string test_file_ = "testfile"; + const mode_t expected_mode = S_IFREG | S_IRUSR | S_IWUSR; + const uint64_t fh = 23; + + std::string test_file_path_; }; TEST_F(StatTest, StatNormal) { - struct iovec iov_in[2]; - struct iovec iov_out[2]; - - struct fuse_in_header in_header = { - .len = sizeof(struct fuse_in_header) + sizeof(struct fuse_getattr_in), - .opcode = FUSE_GETATTR, - .unique = 4, - .nodeid = 1, - .uid = 0, - .gid = 0, - .pid = 4, - .padding = 0, - }; - struct fuse_getattr_in in_payload = {0}; - iov_in[0].iov_len = sizeof(in_header); - iov_in[0].iov_base = &in_header; - iov_in[1].iov_len = sizeof(in_payload); - iov_in[1].iov_base = &in_payload; - - mode_t expected_mode = S_IRWXU | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH; - struct timespec atime = {.tv_sec = 1595436289, .tv_nsec = 134150844}; - struct timespec mtime = {.tv_sec = 1595436290, .tv_nsec = 134150845}; - struct timespec ctime = {.tv_sec = 1595436291, .tv_nsec = 134150846}; + // Set up fixture. + struct fuse_attr attr = DefaultFuseAttr(expected_mode, 1); struct fuse_out_header out_header = { .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), - .error = 0, - .unique = 4, - }; - struct fuse_attr attr = { - .ino = 1, - .size = 512, - .blocks = 4, - .atime = static_cast<uint64_t>(atime.tv_sec), - .mtime = static_cast<uint64_t>(mtime.tv_sec), - .ctime = static_cast<uint64_t>(ctime.tv_sec), - .atimensec = static_cast<uint32_t>(atime.tv_nsec), - .mtimensec = static_cast<uint32_t>(mtime.tv_nsec), - .ctimensec = static_cast<uint32_t>(ctime.tv_nsec), - .mode = expected_mode, - .nlink = 2, - .uid = 1234, - .gid = 4321, - .rdev = 12, - .blksize = 4096, }; struct fuse_attr_out out_payload = { .attr = attr, }; - iov_out[0].iov_len = sizeof(out_header); - iov_out[0].iov_base = &out_header; - iov_out[1].iov_len = sizeof(out_payload); - iov_out[1].iov_base = &out_payload; - - SetExpected(iov_in, 2, iov_out, 2); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_GETATTR, iov_out); + // Make syscall. struct stat stat_buf; EXPECT_THAT(stat(mount_point_.path().c_str(), &stat_buf), SyscallSucceeds()); + // Check filesystem operation result. struct stat expected_stat = { .st_ino = attr.ino, .st_nlink = attr.nlink, @@ -124,43 +84,133 @@ TEST_F(StatTest, StatNormal) { .st_size = static_cast<off_t>(attr.size), .st_blksize = attr.blksize, .st_blocks = static_cast<blkcnt_t>(attr.blocks), - .st_atim = atime, - .st_mtim = mtime, - .st_ctim = ctime, + .st_atim = (struct timespec){.tv_sec = static_cast<int>(attr.atime), + .tv_nsec = attr.atimensec}, + .st_mtim = (struct timespec){.tv_sec = static_cast<int>(attr.mtime), + .tv_nsec = attr.mtimensec}, + .st_ctim = (struct timespec){.tv_sec = static_cast<int>(attr.ctime), + .tv_nsec = attr.ctimensec}, }; EXPECT_TRUE(StatsAreEqual(stat_buf, expected_stat)); - WaitCompleted(); -} -TEST_F(StatTest, StatNotFound) { - struct iovec iov_in[2]; - struct iovec iov_out[2]; + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_getattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); - struct fuse_in_header in_header = { - .len = sizeof(struct fuse_in_header) + sizeof(struct fuse_getattr_in), - .opcode = FUSE_GETATTR, - .unique = 4, - }; - struct fuse_getattr_in in_payload = {0}; - iov_in[0].iov_len = sizeof(in_header); - iov_in[0].iov_base = &in_header; - iov_in[1].iov_len = sizeof(in_payload); - iov_in[1].iov_base = &in_payload; + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.opcode, FUSE_GETATTR); + EXPECT_EQ(in_payload.getattr_flags, 0); + EXPECT_EQ(in_payload.fh, 0); +} +TEST_F(StatTest, StatNotFound) { + // Set up fixture. struct fuse_out_header out_header = { .len = sizeof(struct fuse_out_header), .error = -ENOENT, - .unique = 4, }; - iov_out[0].iov_len = sizeof(out_header); - iov_out[0].iov_base = &out_header; - - SetExpected(iov_in, 2, iov_out, 1); + auto iov_out = FuseGenerateIovecs(out_header); + SetServerResponse(FUSE_GETATTR, iov_out); + // Make syscall. struct stat stat_buf; EXPECT_THAT(stat(mount_point_.path().c_str(), &stat_buf), SyscallFailsWithErrno(ENOENT)); - WaitCompleted(); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_getattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.opcode, FUSE_GETATTR); + EXPECT_EQ(in_payload.getattr_flags, 0); + EXPECT_EQ(in_payload.fh, 0); +} + +TEST_F(StatTest, FstatNormal) { + // Set up fixture. + SetServerInodeLookup(test_file_); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenPath(test_file_path_, O_RDONLY, fh)); + auto close_fd = CloseFD(fd); + + struct fuse_attr attr = DefaultFuseAttr(expected_mode, 2); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + }; + struct fuse_attr_out out_payload = { + .attr = attr, + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_GETATTR, iov_out); + + // Make syscall. + struct stat stat_buf; + EXPECT_THAT(fstat(fd.get(), &stat_buf), SyscallSucceeds()); + + // Check filesystem operation result. + struct stat expected_stat = { + .st_ino = attr.ino, + .st_nlink = attr.nlink, + .st_mode = expected_mode, + .st_uid = attr.uid, + .st_gid = attr.gid, + .st_rdev = attr.rdev, + .st_size = static_cast<off_t>(attr.size), + .st_blksize = attr.blksize, + .st_blocks = static_cast<blkcnt_t>(attr.blocks), + .st_atim = (struct timespec){.tv_sec = static_cast<int>(attr.atime), + .tv_nsec = attr.atimensec}, + .st_mtim = (struct timespec){.tv_sec = static_cast<int>(attr.mtime), + .tv_nsec = attr.mtimensec}, + .st_ctim = (struct timespec){.tv_sec = static_cast<int>(attr.ctime), + .tv_nsec = attr.ctimensec}, + }; + EXPECT_TRUE(StatsAreEqual(stat_buf, expected_stat)); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_getattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.opcode, FUSE_GETATTR); + EXPECT_EQ(in_payload.getattr_flags, 0); + EXPECT_EQ(in_payload.fh, 0); +} + +TEST_F(StatTest, StatByFileHandle) { + // Set up fixture. + SetServerInodeLookup(test_file_, expected_mode, 0); + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenPath(test_file_path_, O_RDONLY, fh)); + auto close_fd = CloseFD(fd); + + struct fuse_attr attr = DefaultFuseAttr(expected_mode, 2, 0); + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_attr_out), + }; + struct fuse_attr_out out_payload = { + .attr = attr, + }; + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_GETATTR, iov_out); + + // Make syscall. + std::vector<char> buf(1); + // Since this is an empty file, it won't issue FUSE_READ. But a FUSE_GETATTR + // will be issued before read completes. + EXPECT_THAT(read(fd.get(), buf.data(), buf.size()), SyscallSucceeds()); + + // Check FUSE request. + struct fuse_in_header in_header; + struct fuse_getattr_in in_payload; + auto iov_in = FuseGenerateIovecs(in_header, in_payload); + + GetServerActualRequest(iov_in); + EXPECT_EQ(in_header.opcode, FUSE_GETATTR); + EXPECT_EQ(in_payload.getattr_flags, FUSE_GETATTR_FH); + EXPECT_EQ(in_payload.fh, fh); } } // namespace diff --git a/test/fuse/linux/symlink_test.cc b/test/fuse/linux/symlink_test.cc new file mode 100644 index 000000000..2c3a52987 --- /dev/null +++ b/test/fuse/linux/symlink_test.cc @@ -0,0 +1,88 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class SymlinkTest : public FuseTest { + protected: + const std::string target_file_ = "target_file_"; + const std::string symlink_ = "symlink_"; + const mode_t perms_ = S_IRWXU | S_IRWXG | S_IRWXO; +}; + +TEST_F(SymlinkTest, CreateSymLink) { + const std::string symlink_path = + JoinPath(mount_point_.path().c_str(), symlink_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFLNK | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SYMLINK, iov_out); + ASSERT_THAT(symlink(target_file_.c_str(), symlink_path.c_str()), + SyscallSucceeds()); + + struct fuse_in_header in_header; + std::vector<char> actual_target_file(target_file_.length() + 1); + std::vector<char> actual_symlink(symlink_.length() + 1); + auto iov_in = + FuseGenerateIovecs(in_header, actual_symlink, actual_target_file); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, + sizeof(in_header) + symlink_.length() + target_file_.length() + 2); + EXPECT_EQ(in_header.opcode, FUSE_SYMLINK); + EXPECT_EQ(std::string(actual_target_file.data()), target_file_); + EXPECT_EQ(std::string(actual_symlink.data()), symlink_); +} + +TEST_F(SymlinkTest, FileTypeError) { + const std::string symlink_path = + JoinPath(mount_point_.path().c_str(), symlink_); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_entry_out), + }; + struct fuse_entry_out out_payload = DefaultEntryOut(S_IFREG | perms_, 5); + auto iov_out = FuseGenerateIovecs(out_header, out_payload); + SetServerResponse(FUSE_SYMLINK, iov_out); + ASSERT_THAT(symlink(target_file_.c_str(), symlink_path.c_str()), + SyscallFailsWithErrno(EIO)); + SkipServerActualRequest(); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/unlink_test.cc b/test/fuse/linux/unlink_test.cc new file mode 100644 index 000000000..5702e9b32 --- /dev/null +++ b/test/fuse/linux/unlink_test.cc @@ -0,0 +1,83 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/mount.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class UnlinkTest : public FuseTest { + protected: + const std::string test_file_ = "test_file"; +}; + +TEST_F(UnlinkTest, RegularFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header), + }; + auto iov_out = FuseGenerateIovecs(out_header); + SetServerResponse(FUSE_UNLINK, iov_out); + + ASSERT_THAT(unlink(test_file_path.c_str()), SyscallSucceeds()); + struct fuse_in_header in_header; + std::vector<char> unlinked_file(test_file_.length() + 1); + auto iov_in = FuseGenerateIovecs(in_header, unlinked_file); + GetServerActualRequest(iov_in); + + EXPECT_EQ(in_header.len, sizeof(in_header) + test_file_.length() + 1); + EXPECT_EQ(in_header.opcode, FUSE_UNLINK); + EXPECT_EQ(std::string(unlinked_file.data()), test_file_); +} + +TEST_F(UnlinkTest, NoFile) { + const std::string test_file_path = + JoinPath(mount_point_.path().c_str(), test_file_); + SetServerInodeLookup(test_file_, S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO); + + struct fuse_out_header out_header = { + .len = sizeof(struct fuse_out_header), + .error = -ENOENT, + }; + auto iov_out = FuseGenerateIovecs(out_header); + SetServerResponse(FUSE_UNLINK, iov_out); + + ASSERT_THAT(unlink(test_file_path.c_str()), SyscallFailsWithErrno(ENOENT)); + SkipServerActualRequest(); +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/fuse/linux/write_test.cc b/test/fuse/linux/write_test.cc new file mode 100644 index 000000000..1a62beb96 --- /dev/null +++ b/test/fuse/linux/write_test.cc @@ -0,0 +1,303 @@ +// 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. + +#include <errno.h> +#include <fcntl.h> +#include <linux/fuse.h> +#include <sys/stat.h> +#include <sys/statfs.h> +#include <sys/types.h> +#include <unistd.h> + +#include <string> +#include <vector> + +#include "gtest/gtest.h" +#include "test/fuse/linux/fuse_base.h" +#include "test/util/fuse_util.h" +#include "test/util/test_util.h" + +namespace gvisor { +namespace testing { + +namespace { + +class WriteTest : public FuseTest { + void SetUp() override { + FuseTest::SetUp(); + test_file_path_ = JoinPath(mount_point_.path().c_str(), test_file_); + } + + // TearDown overrides the parent's function + // to skip checking the unconsumed release request at the end. + void TearDown() override { UnmountFuse(); } + + protected: + const std::string test_file_ = "test_file"; + const mode_t test_file_mode_ = S_IFREG | S_IRWXU | S_IRWXG | S_IRWXO; + const uint64_t test_fh_ = 1; + const uint32_t open_flag_ = O_RDWR; + + std::string test_file_path_; + + PosixErrorOr<FileDescriptor> OpenTestFile(const std::string &path, + uint64_t size = 512) { + SetServerInodeLookup(test_file_, test_file_mode_, size); + + struct fuse_out_header out_header_open = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_open_out), + }; + struct fuse_open_out out_payload_open = { + .fh = test_fh_, + .open_flags = open_flag_, + }; + auto iov_out_open = FuseGenerateIovecs(out_header_open, out_payload_open); + SetServerResponse(FUSE_OPEN, iov_out_open); + + auto res = Open(path.c_str(), open_flag_); + if (res.ok()) { + SkipServerActualRequest(); + } + return res; + } +}; + +class WriteTestSmallMaxWrite : public WriteTest { + void SetUp() override { + MountFuse(); + SetUpFuseServer(&fuse_init_payload); + test_file_path_ = JoinPath(mount_point_.path().c_str(), test_file_); + } + + protected: + const static uint32_t max_write_ = 4096; + constexpr static struct fuse_init_out fuse_init_payload = { + .major = 7, + .max_write = max_write_, + }; + + const uint32_t size_fragment = max_write_; +}; + +TEST_F(WriteTest, WriteNormal) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the write. + const int n_write = 10; + struct fuse_out_header out_header_write = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_write_out), + }; + struct fuse_write_out out_payload_write = { + .size = n_write, + }; + auto iov_out_write = FuseGenerateIovecs(out_header_write, out_payload_write); + SetServerResponse(FUSE_WRITE, iov_out_write); + + // Issue the write. + std::vector<char> buf(n_write); + RandomizeBuffer(buf.data(), buf.size()); + EXPECT_THAT(write(fd.get(), buf.data(), n_write), + SyscallSucceedsWithValue(n_write)); + + // Check the write request. + struct fuse_in_header in_header_write; + struct fuse_write_in in_payload_write; + std::vector<char> payload_buf(n_write); + auto iov_in_write = + FuseGenerateIovecs(in_header_write, in_payload_write, payload_buf); + GetServerActualRequest(iov_in_write); + + EXPECT_EQ(in_payload_write.fh, test_fh_); + EXPECT_EQ(in_header_write.len, + sizeof(in_header_write) + sizeof(in_payload_write)); + EXPECT_EQ(in_header_write.opcode, FUSE_WRITE); + EXPECT_EQ(in_payload_write.offset, 0); + EXPECT_EQ(in_payload_write.size, n_write); + EXPECT_EQ(buf, payload_buf); +} + +TEST_F(WriteTest, WriteShort) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the write. + const int n_write = 10, n_written = 5; + struct fuse_out_header out_header_write = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_write_out), + }; + struct fuse_write_out out_payload_write = { + .size = n_written, + }; + auto iov_out_write = FuseGenerateIovecs(out_header_write, out_payload_write); + SetServerResponse(FUSE_WRITE, iov_out_write); + + // Issue the write. + std::vector<char> buf(n_write); + RandomizeBuffer(buf.data(), buf.size()); + EXPECT_THAT(write(fd.get(), buf.data(), n_write), + SyscallSucceedsWithValue(n_written)); + + // Check the write request. + struct fuse_in_header in_header_write; + struct fuse_write_in in_payload_write; + std::vector<char> payload_buf(n_write); + auto iov_in_write = + FuseGenerateIovecs(in_header_write, in_payload_write, payload_buf); + GetServerActualRequest(iov_in_write); + + EXPECT_EQ(in_payload_write.fh, test_fh_); + EXPECT_EQ(in_header_write.len, + sizeof(in_header_write) + sizeof(in_payload_write)); + EXPECT_EQ(in_header_write.opcode, FUSE_WRITE); + EXPECT_EQ(in_payload_write.offset, 0); + EXPECT_EQ(in_payload_write.size, n_write); + EXPECT_EQ(buf, payload_buf); +} + +TEST_F(WriteTest, WriteShortZero) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Prepare for the write. + const int n_write = 10; + struct fuse_out_header out_header_write = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_write_out), + }; + struct fuse_write_out out_payload_write = { + .size = 0, + }; + auto iov_out_write = FuseGenerateIovecs(out_header_write, out_payload_write); + SetServerResponse(FUSE_WRITE, iov_out_write); + + // Issue the write. + std::vector<char> buf(n_write); + RandomizeBuffer(buf.data(), buf.size()); + EXPECT_THAT(write(fd.get(), buf.data(), n_write), SyscallFailsWithErrno(EIO)); + + // Check the write request. + struct fuse_in_header in_header_write; + struct fuse_write_in in_payload_write; + std::vector<char> payload_buf(n_write); + auto iov_in_write = + FuseGenerateIovecs(in_header_write, in_payload_write, payload_buf); + GetServerActualRequest(iov_in_write); + + EXPECT_EQ(in_payload_write.fh, test_fh_); + EXPECT_EQ(in_header_write.len, + sizeof(in_header_write) + sizeof(in_payload_write)); + EXPECT_EQ(in_header_write.opcode, FUSE_WRITE); + EXPECT_EQ(in_payload_write.offset, 0); + EXPECT_EQ(in_payload_write.size, n_write); + EXPECT_EQ(buf, payload_buf); +} + +TEST_F(WriteTest, WriteZero) { + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_)); + + // Issue the write. + std::vector<char> buf(0); + EXPECT_THAT(write(fd.get(), buf.data(), 0), SyscallSucceedsWithValue(0)); +} + +TEST_F(WriteTest, PWrite) { + const int file_size = 512; + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_, file_size)); + + // Prepare for the write. + const int n_write = 10; + struct fuse_out_header out_header_write = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_write_out), + }; + struct fuse_write_out out_payload_write = { + .size = n_write, + }; + auto iov_out_write = FuseGenerateIovecs(out_header_write, out_payload_write); + SetServerResponse(FUSE_WRITE, iov_out_write); + + // Issue the write. + std::vector<char> buf(n_write); + RandomizeBuffer(buf.data(), buf.size()); + const int offset_write = file_size >> 1; + EXPECT_THAT(pwrite(fd.get(), buf.data(), n_write, offset_write), + SyscallSucceedsWithValue(n_write)); + + // Check the write request. + struct fuse_in_header in_header_write; + struct fuse_write_in in_payload_write; + std::vector<char> payload_buf(n_write); + auto iov_in_write = + FuseGenerateIovecs(in_header_write, in_payload_write, payload_buf); + GetServerActualRequest(iov_in_write); + + EXPECT_EQ(in_payload_write.fh, test_fh_); + EXPECT_EQ(in_header_write.len, + sizeof(in_header_write) + sizeof(in_payload_write)); + EXPECT_EQ(in_header_write.opcode, FUSE_WRITE); + EXPECT_EQ(in_payload_write.offset, offset_write); + EXPECT_EQ(in_payload_write.size, n_write); + EXPECT_EQ(buf, payload_buf); +} + +TEST_F(WriteTestSmallMaxWrite, WriteSmallMaxWrie) { + const int n_fragment = 10; + const int n_write = size_fragment * n_fragment; + + auto fd = ASSERT_NO_ERRNO_AND_VALUE(OpenTestFile(test_file_path_, n_write)); + + // Prepare for the write. + struct fuse_out_header out_header_write = { + .len = sizeof(struct fuse_out_header) + sizeof(struct fuse_write_out), + }; + struct fuse_write_out out_payload_write = { + .size = size_fragment, + }; + auto iov_out_write = FuseGenerateIovecs(out_header_write, out_payload_write); + + for (int i = 0; i < n_fragment; ++i) { + SetServerResponse(FUSE_WRITE, iov_out_write); + } + + // Issue the write. + std::vector<char> buf(n_write); + RandomizeBuffer(buf.data(), buf.size()); + EXPECT_THAT(write(fd.get(), buf.data(), n_write), + SyscallSucceedsWithValue(n_write)); + + ASSERT_EQ(GetServerNumUnsentResponses(), 0); + ASSERT_EQ(GetServerNumUnconsumedRequests(), n_fragment); + + // Check the write request. + struct fuse_in_header in_header_write; + struct fuse_write_in in_payload_write; + std::vector<char> payload_buf(size_fragment); + auto iov_in_write = + FuseGenerateIovecs(in_header_write, in_payload_write, payload_buf); + + for (int i = 0; i < n_fragment; ++i) { + GetServerActualRequest(iov_in_write); + + EXPECT_EQ(in_payload_write.fh, test_fh_); + EXPECT_EQ(in_header_write.len, + sizeof(in_header_write) + sizeof(in_payload_write)); + EXPECT_EQ(in_header_write.opcode, FUSE_WRITE); + EXPECT_EQ(in_payload_write.offset, i * size_fragment); + EXPECT_EQ(in_payload_write.size, size_fragment); + + auto it = buf.begin() + i * size_fragment; + EXPECT_EQ(std::vector<char>(it, it + size_fragment), payload_buf); + } +} + +} // namespace + +} // namespace testing +} // namespace gvisor diff --git a/test/util/BUILD b/test/util/BUILD index 2a17c33ee..fc5fb3a8d 100644 --- a/test/util/BUILD +++ b/test/util/BUILD @@ -46,6 +46,13 @@ cc_library( ) cc_library( + name = "fuse_util", + testonly = 1, + srcs = ["fuse_util.cc"], + hdrs = ["fuse_util.h"], +) + +cc_library( name = "proc_util", testonly = 1, srcs = ["proc_util.cc"], diff --git a/test/util/fuse_util.cc b/test/util/fuse_util.cc new file mode 100644 index 000000000..027f8386c --- /dev/null +++ b/test/util/fuse_util.cc @@ -0,0 +1,63 @@ +// 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. + +#include "test/util/fuse_util.h" + +#include <sys/stat.h> +#include <sys/types.h> + +#include <string> + +namespace gvisor { +namespace testing { + +// Create a default FuseAttr struct with specified mode, inode, and size. +fuse_attr DefaultFuseAttr(mode_t mode, uint64_t inode, uint64_t size) { + const int time_sec = 1595436289; + const int time_nsec = 134150844; + return (struct fuse_attr){ + .ino = inode, + .size = size, + .blocks = 4, + .atime = time_sec, + .mtime = time_sec, + .ctime = time_sec, + .atimensec = time_nsec, + .mtimensec = time_nsec, + .ctimensec = time_nsec, + .mode = mode, + .nlink = 2, + .uid = 1234, + .gid = 4321, + .rdev = 12, + .blksize = 4096, + }; +} + +// Create response body with specified mode, nodeID, and size. +fuse_entry_out DefaultEntryOut(mode_t mode, uint64_t node_id, uint64_t size) { + struct fuse_entry_out default_entry_out = { + .nodeid = node_id, + .generation = 0, + .entry_valid = 0, + .attr_valid = 0, + .entry_valid_nsec = 0, + .attr_valid_nsec = 0, + .attr = DefaultFuseAttr(mode, node_id, size), + }; + return default_entry_out; +} + +} // namespace testing +} // namespace gvisor diff --git a/test/util/fuse_util.h b/test/util/fuse_util.h new file mode 100644 index 000000000..544fe1b38 --- /dev/null +++ b/test/util/fuse_util.h @@ -0,0 +1,75 @@ +// 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. + +#ifndef GVISOR_TEST_UTIL_FUSE_UTIL_H_ +#define GVISOR_TEST_UTIL_FUSE_UTIL_H_ + +#include <linux/fuse.h> +#include <sys/uio.h> + +#include <string> +#include <vector> + +namespace gvisor { +namespace testing { + +// The fundamental generation function with a single argument. If passed by +// std::string or std::vector<char>, it will call specialized versions as +// implemented below. +template <typename T> +std::vector<struct iovec> FuseGenerateIovecs(T &first) { + return {(struct iovec){.iov_base = &first, .iov_len = sizeof(first)}}; +} + +// If an argument is of type std::string, it must be used in read-only scenario. +// Because we are setting up iovec, which contains the original address of a +// data structure, we have to drop const qualification. Usually used with +// variable-length payload data. +template <typename T = std::string> +std::vector<struct iovec> FuseGenerateIovecs(std::string &first) { + // Pad one byte for null-terminate c-string. + return {(struct iovec){.iov_base = const_cast<char *>(first.c_str()), + .iov_len = first.size() + 1}}; +} + +// If an argument is of type std::vector<char>, it must be used in write-only +// scenario and the size of the variable must be greater than or equal to the +// size of the expected data. Usually used with variable-length payload data. +template <typename T = std::vector<char>> +std::vector<struct iovec> FuseGenerateIovecs(std::vector<char> &first) { + return {(struct iovec){.iov_base = first.data(), .iov_len = first.size()}}; +} + +// A helper function to set up an array of iovec struct for testing purpose. +// Use variadic class template to generalize different numbers and different +// types of FUSE structs. +template <typename T, typename... Types> +std::vector<struct iovec> FuseGenerateIovecs(T &first, Types &...args) { + auto first_iovec = FuseGenerateIovecs(first); + auto iovecs = FuseGenerateIovecs(args...); + first_iovec.insert(std::end(first_iovec), std::begin(iovecs), + std::end(iovecs)); + return first_iovec; +} + +// Create a fuse_attr filled with the specified mode and inode. +fuse_attr DefaultFuseAttr(mode_t mode, uint64_t inode, uint64_t size = 512); + +// Return a fuse_entry_out FUSE server response body. +fuse_entry_out DefaultEntryOut(mode_t mode, uint64_t node_id, + uint64_t size = 512); + +} // namespace testing +} // namespace gvisor +#endif // GVISOR_TEST_UTIL_FUSE_UTIL_H_ |