minix/servers/vfs/README
David van Moolenbroek 87337273e4 Remove support for reopening character devices
Previously, VFS would reopen a character device after a driver crash
if the associated file descriptor was opened with the O_REOPEN flag.
This patch removes support for this feature. The code was complex,
full of uncovered corner cases, and hard to test. Moreover, it did not
actually hide the crash from user applications: they would get an
error code to indicate that something went wrong, and have to decide
based on the nature of the underlying device how to continue.

- remove support for O_REOPEN, and make playwave(1) reopen its device;
- remove support for the DEV_REOPEN protocol message;
- remove all code in VFS related to reopening character devices;
- no longer change VFS filp reference count and FD bitmap upon filp
  invalidation; instead, make get_filp* fail all calls on invalidated
  FDs except when obtained with the locktype VNODE_OPCL which is used
  by close_fd only;
- remove the VFS fproc file descriptor bitmap entirely, returning to
  the situation that a FD is in use if its slot points to a filp; use
  FILP_CLOSED as single means of marking a filp as invalidated.

Change-Id: I34f6bc69a036b3a8fc667c1f80435ff3af56558f
2014-03-01 09:04:52 +01:00

699 lines
45 KiB
Text

## Description of VFS Thomas Veerman 21-3-2013
## This file is organized such that it can be read both in a Wiki and on
## the MINIX terminal using e.g. vi or less. Please, keep the file in the
## source tree as the canonical version and copy changes into the Wiki.
#pragma section-numbers 2
= VFS internals =
<<TableOfContents(2)>>
## Table of contents
## 1 ..... General description of responsibilities
## 2 ..... General architecture
## 3 ..... Worker threads
## 4 ..... Locking
## 4.1 .... Locking requirements
## 4.2 .... Three-level Lock
## 4.3 .... Data structures subject to locking
## 4.4 .... Locking order
## 4.5 .... Vmnt (file system) locking
## 4.6 .... Vnode (open file) locking
## 4.7 .... Filp (file position) locking
## 4.8 .... Lock characteristics per request type
## 5 ..... Recovery from driver crashes
## 5.1 .... Recovery from block drivers crashes
## 5.2 .... Recovery from character driver crashes
## 5.3 .... Recovery from File Server crashes
== General description of responsibilities ==
## 1 General description of responsibilities
VFS implements the file system in cooperation with one or more File Servers
(FS). The File Servers take care of the actual file system on a partition. That
is, they interpret the data structure on disk, write and read data to/from
disk, etc. VFS sits on top of those File Servers and communicates with
them. Looking inside VFS, we can identify several roles. First, a role of VFS
is to handle most POSIX system calls that are supported by Minix. Additionally,
it supports a few calls necessary for libc. The following system calls are
handled by VFS:
access, chdir, chmod, chown, chroot, close, creat, fchdir, fcntl, fstat,
fstatvfs, fsync, ftruncate, getdents, getvfsstat, ioctl, link, llseek, lseek,
lstat, mkdir, mknod, mount, open, pipe, read, readlink, rename, rmdir, select,
stat, statvfs, symlink, sync, truncate, umask, umount, unlink, utime, write.
Second, it maintains part of the state belonging to a process (process state is
spread out over the kernel, VM, PM, and VFS). For example, it maintains state
for select(2) calls, file descriptors and file positions. Also, it cooperates
with the Process Manager to handle the fork, exec, and exit system calls.
Third, VFS keeps track of endpoints that are supposed to be drivers for
character or block special files. File Servers can be regarded as drivers for
block special files, although they are handled entirely different compared
to other drivers.
The following diagram depicts how a read() on a file in /home is being handled:
{{{
----------------
| user process |
----------------
^ ^
| |
read(2) \
| \
V \
---------------- |
| VFS | |
---------------- |
^ |
| |
V |
------- -------- ---------
| MFS | | MFS | | MFS |
| / | | /usr | | /home |
------- -------- ---------
}}}
Diagram 1: handling of read(2) system call
The user process executes the read system call which is delivered to VFS. VFS
verifies the read is done on a valid (open) file and forwards the request
to the FS responsible for the file system on which the file resides. The FS
reads the data, copies it directly to the user process, and replies to VFS
it has executed the request. Subsequently, VFS replies to the user process
the operation is done and the user process continues to run.
== General architecture ==
## 2 General architecture
VFS works roughly identical to every other server and driver in Minix; it
fetches a message (internally referred to as a job in some cases), executes
the request embedded in the message, returns a reply, and fetches the next
job. There are several sources for new jobs: from user processes, from PM, from
the kernel, and from suspended jobs inside VFS itself (suspended operations
on pipes, locks, or character special files). File Servers are regarded as
normal user processes in this case, but their abilities are limited. This
is to prevent deadlocks. Once a job is received, a worker thread starts
executing it. During the lifetime of a job, the worker thread might need
to talk to several File Servers. The protocol VFS speaks with File Servers
is fully documented on the Wiki at [0]. The protocol fields are defined in
<minix/vfsif.h>. If the job is an operation on a character or block special
file and the need to talk to a driver arises, VFS uses the Character and
Block Device Protocol. See [1]. This is sadly not official documentation,
but it is an accurate description of how it works. Luckily, driver writers
can use the libchardriver and libblockdriver libraries and don't have to
know the details of the protocol.
== Worker threads ==
## 3 Worker threads
Upon start up, VFS spawns a configurable amount of worker threads. The
main thread fetches requests and replies, and hands them off to idle or
reply-pending workers, respectively. If no worker threads are available,
the request is queued. All standard system calls are handled by such worker
threads. One of the threads is reserved to handle new requests from system
processes (i.e., File Servers and drivers) when there are no normal worker
threads available; all normal threads might be blocked on a single worker
thread that caused a system process to send a request on its own. To unblock
all normal threads, we need to reserve one spare thread to handle that
situation. VFS drives all File Servers and drivers asynchronously. While
waiting for a reply, a worker thread is blocked and other workers can keep
processing requests. Upon reply the worker thread is unblocked.
As mentioned above, the main thread is responsible for retrieving new jobs and
replies to current jobs and start or unblock the proper worker thread.
Driver replies are processed directly from the main thread. As a consequence,
these processing routines may not block their calling thread. In some cases,
these routines may resume a thread that is blocked waiting for the reply. This
is always the case for block driver replies, and may or may not be the case for
character driver replies. The character driver reply processing routines may
also unblock suspended processes which in turn generate new jobs to be handled
by the main loop (e.g., suspended reads and writes on pipes). So depending
on the reply a new thread may have to be started.
Worker threads are strictly tied to a process, and each process can have at
most one worker thread running for it. Generally speaking, there are two types
of work supported by worker threads: normal work, and work from PM. The main
subtype of normal work is the handling of a system call made by the process
itself. The process is blocked while VFS is handling the system call, so no new
system call can arrive from a process while VFS has not completed a previous
system call from that process. For that reason, if there are no worker threads
available to handle the work, the work is queued in the corresponding process
entry of the fproc table.
The other main type of work consists of requests from PM. The protocol PM
speaks with VFS is asynchronous. PM is allowed to send up to one request per
process to VFS, in addition to a request to initiate a reboot. Most jobs from
PM are taken care of immediately by the main thread, but some jobs require a
worker thread context (to be able to sleep) and/or serialization with normal
work. Therefore, each process may have a PM request queued for execution, also
in the fproc table. Managing proper queuing, addition, and execution of both
normal and PM work is the responsibility of the worker thread infrastructure.
There are several special tasks that require a worker thread, and these are
implemented as normal work associated with a certain special process that does
not make regular VFS calls anyway. For example, the initial ramdisk mount
procedure uses a thread associated with the VFS process. Some of these special
tasks require protection against being started multiple times at once, as this
is not only undesirable but also disallowed. The full list of worker thread
task types and subtypes is shown in Table 1.
{{{
-------------------------------------------------------------------------
| Worker thread task | Type | Association | May use spare? |
+---------------------------+--------+-----------------+----------------+
| system call from process | normal | calling process | if system proc |
+---------------------------+--------+-----------------+----------------+
| resumed pipe operation | normal | calling process | no |
+---------------------------+--------+-----------------+----------------+
| postponed PM request | PM | target process | no |
+---------------------------+--------+-----------------+----------------+
| DS event notification | normal | DS | yes |
+---------------------------+--------+-----------------+----------------+
| initial ramdisk mounting | normal | VFS | no |
+---------------------------+--------+-----------------+----------------+
| reboot sequence | normal | PM | no |
-------------------------------------------------------------------------
}}}
Table 1: worker thread work types and subtypes
Communication with block drivers is asynchronous, but at this time, access to
these drivers is serialized on a per-driver basis. File Servers are treated
differently. VFS was designed to be able to send requests concurrently to File
Servers, although at the time of writing there are no File Servers that can
actually make use of that functionality. To identify which reply from an FS
belongs to which worker thread, all requests have an embedded transaction
identification number (a magic number + thread id encoded in the mtype field of
a message) which the FS has to echo upon reply. Because the range of valid
transaction IDs is isolated from valid system call numbers, VFS can use that ID
to differentiate between replies from File Servers and actual new system calls
from FSes. Using this mechanism VFS is able to support FUSE and ProcFS.
== Locking ==
## 4 Locking
To ensure correct execution of system calls, worker threads sometimes need
certain objects within VFS to remain unchanged during thread suspension
and resumption (i.e., when they need to communicate with a driver or File
Server). Threads keep most state on the stack, but there are a few global
variables that require protection: the fproc table, vmnt table, vnode table,
and filp table. Other tables such as lock table, select table, and dmap table
don't require protection by means of exclusive access. There it's required
and enough to simply mark an entry in use.
=== Locking requirements ===
## 4.1 Locking requirements
VFS implements the locking model described in [2]. For completeness of this
document we'll describe it here, too. The requirements are based on a threading
package that is non-preemptive. VFS must guarantee correct functioning with
several, semi-concurrently executing threads in any arbitrary order. The
latter requirement follows from the fact that threads need service from
other components like File Servers and drivers, and they may take any time
to complete requests.
1. Consistency of replicated values. Several system calls rely on VFS keeping a replicated representation of data in File Servers (e.g., file sizes, file modes, etc.).
1. Isolation of system calls. Many system calls involve multiple requests to FSes. Concurrent requests from other processes must not lead to otherwise impossible results (e.g., a chmod operation on a file cannot fail halfway through because it's suddenly unlinked or moved).
1. Integrity of objects. From the point of view of threads, obtaining mutual exclusion is a potentially blocking operation. The integrity of any objects used across blocking calls must be guaranteed (e.g., the file mode in a vnode must remain intact not only when talking to other components, but also when obtaining a lock on a filp).
1. No deadlock. Not one call may cause another call to never complete. Deadlock situations are typically the result of two or more threads that each hold exclusive access to one resource and want exclusive access to the resource held by the other thread. These resources are a) data (global variables) and b) worker threads.
a. Conflicts between locking of different types of objects can be avoided by keeping a locking order: objects of different type must always be locked in the same order. If multiple objects of the same type are to be locked, then first a "common denominator" higher up in the locking order must be locked.
a. Some threads can only run to completion when another thread does work on their behalf. Examples of this are drivers and file servers that do system calls on their own (e.g., ProcFS, PFS/UNIX Domain Sockets, FUSE) or crashing components (e.g., a driver for a character special file that crashes during a request; a second thread is required to handle resource clean up or driver restart before the first thread can abort or retry the request).
1. No starvation. VFS must guarantee that every system call completes in finite time (e.g., an infinite stream of reads must never completely block writes). Furthermore, we want to maximize parallelism to improve performance. This leads to:
1. A request to one File Server must not block access to other FS processes. This means that most forms of locking cannot take place at a global level, and must at most take place on the file system level.
1. No read-only operation on a regular file must block an independent read call to that file. In particular, (read-only) open and close operations may not block such reads, and multiple independent reads on the same file must be able to take place concurrently (i.e., reads that do not share a file position between their file descriptors).
=== Three-level Lock ===
## 4.2 Three-level Lock
From the requirements it follows that we need at least two locking types: read
and write locks. Concurrent reads are allowed, but writes are exclusive both
from reads and from each other. However, in a lot of cases it possible to use
a third locking type that is in between read and write lock: the serialize
lock. This is implemented in the three-level lock [2]. The three-level
lock provides:
TLL_READ: allows an unlimited number of threads to hold the lock with the
same type (both the thread itself and other threads); N * concurrent.
TLL_READSER: also allows an unlimited number of threads with type TLL_READ,
but only one thread can obtain serial access to the lock; N * concurrent +
1 * serial.
TLL_WRITE: provides full mutual exclusion; 1 * exclusive + 0 * concurrent +
0 * serial.
In absence of TLL_READ locks, a TLL_READSER is identical to TLL_WRITE. However,
TLL_READSER never blocks concurrent TLL_READ access. TLL_READSER can be
upgraded to TLL_WRITE; the thread will block until the last TLL_READ lock
leaves and new TLL_READ locks are blocked. Locks can be downgraded to a
lower type. The three-level lock is implemented using two FIFO queues with
write-bias. This guarantees no starvation.
=== Data structures subject to locking ===
## 4.3 Data structures subject to locking
VFS has a number of global data structures. See Table 2.
{{{
--------------------------------------------------------------------
| Structure | Object description |
+------------+-----------------------------------------------------|
| fproc | Process (includes process's file descriptors) |
+------------+-----------------------------------------------------|
| vmnt | Virtual mount; a mounted file system |
+------------+-----------------------------------------------------|
| vnode | Virtual node; an open file |
+------------+-----------------------------------------------------|
| filp | File position into an open file |
+------------+-----------------------------------------------------|
| lock | File region locking state for an open file |
+------------+-----------------------------------------------------|
| select | State for an in-progress select(2) call |
+------------+-----------------------------------------------------|
| dmap | Mapping from major device number to a device driver |
--------------------------------------------------------------------
}}}
Table 2: VFS object types.
An fproc object is a process. An fproc object is created by fork(2)
and destroyed by exit(2) (which may, or may not, be instantiated from the
process itself). It is identified by its endpoint number ('fp_endpoint')
and process id ('fp_pid'). Both are unique although in general the endpoint
number is used throughout the system.
A vmnt object is a mounted file system. It is created by mount(2) and destroyed
by umount(2). It is identified by a device number ('m_dev') and FS endpoint
number ('m_fs_e'); both are unique to each vmnt object. There is always a
single process that handles a file system on a device and a device cannot
be mounted twice.
A vnode object is the VFS representation of an open inode on the file
system. A vnode object is created when a first process opens or creates the
corresponding file and is destroyed when the last process, which has that
file open, closes it. It is identified by a combination of FS endpoint number
('v_fs_e') and inode number of that file system ('v_inode_nr'). A vnode
might be mapped to another file system; the actual reading and writing is
handled by a different endpoint. This has no effect on locking.
A filp object contains a file position within a file. It is created when a file
is opened or anonymous pipe created and destroyed when the last user (i.e.,
process) closes it. A file descriptor always points to a single filp. A filp
always point to a single vnode, although not all vnodes are pointed to by a
filp. A filp has a reference count ('filp_count') which is identical to the
number of file descriptors pointing to it. It can be increased by a dup(2)
or fork(2). A filp can therefore be shared by multiple processes.
A lock object keeps information about locking of file regions. This has
nothing to do with the threading type of locking. The lock objects require
no locking protection and won't be discussed further.
A select object keeps information on a select(2) operation that cannot
be fulfilled immediately (waiting for timeout or file descriptors not
ready). They are identified by their owner ('requestor'); a pointer to the
fproc table. A null pointer means not in use. A select object can be used by
only one process and a process can do only one select(2) at a time. Select(2)
operates on filps and is organized in such a way that it is sufficient to
apply locking on individual filps and not on select objects themselves. They
won't be discussed further.
A dmap object is a mapping from a device number to a device driver. A device
driver can have multiple device numbers associated (e.g., TTY). Access to
a driver is exclusive when it uses the synchronous driver protocol.
=== Locking order ===
## 4.4 Locking order
Based on the description in the previous section, we need protection for
fproc, vmnt, vnode, and filp objects. To prevent deadlocks as a result of
object locking, we need to define a strict locking order. In VFS we use the
following order:
{{{
fproc > [exec] > vmnt > vnode > filp > [block special file] > [dmap]
}}}
That is, no thread may lock an fproc object while holding a vmnt lock,
and no thread may lock a vmnt object while holding an (associated) vnode, etc.
Fproc needs protection because processes themselves can initiate system
calls, but also PM can cause system calls that have to be executed in their
name. For example, a process might be busy reading from a character device
and another process sends a termination signal. The exit(2) that follows is
sent by PM and is to be executed by the to-be-killed process itself. At this
point there is contention for the fproc object that belongs to the process,
hence the need for protection. This problem is solved in a simple way. Recall
that all worker threads are bound to a process. This also forms the basis of
fproc locking: each worker thread acquires and holds the fproc lock for its
associated process for as long as it is processing work for that process.
There are two cases where a worker thread may hold the lock to more than one
process. First, as mentioned, the reboot procedure is executed from a worker
thread set in the context of the PM process, thus with the PM process entry
lock held. The procedure itself then acquires a temporary lock on every other
process in turn, in order to clean it up without interference. Thus, the PM
process entry is higher up in the locking order than all other process entries.
Second, the exec(2) call is protected by a lock, and this exec lock is
currently implemented as a lock on the VM process entry. The exec lock is
acquired by a worker thread for the process performing the exec(2) call, and
thus, the VM process entry is below all other process entries in the locking
order. The exec(2) call is protected by a lock for the following reason. VFS
uses a number of variables on the heap to read ELF headers. They are on the
heap due to their size; putting them on the stack would increase stack size
demands for worker threads. The exec call does blocking read calls and thus
needs exclusive access to these variables. However, only the exec(2) syscall
needs this lock.
Access to block special files needs to be exclusive. File Servers are
responsible for handling reads from and writes to block special files; if
a block special file is on a device that is mounted, the FS responsible for
that mount point takes care of it, otherwise the FS that handles the root of
the file system is responsible. Due to mounting and unmounting file systems,
the FS handling a block special file may change. Locking the vnode is not
enough since the inode can be on an entirely different File Server. Therefore,
access to block special files must be mutually exclusive from concurrent
mount(2)/umount(2) operations. However, when we're not accessing a block
special file, we don't need this lock.
=== Vmnt (file system) locking ===
## 4.5 Vmnt (file system) locking
Vmnt locking cannot be seen completely separately from vnode locking. For
example, umount(2) fails if there are still in-use vnodes, which means that
FS requests [0] only involving in-use inodes do not have to acquire a vmnt
lock. On the other hand, all other request do need a vmnt lock. Extrapolating
this to system calls this means that all system calls involving a file
descriptor don't need a vmnt lock and all other system calls (that make FS
requests) do need a vmnt lock.
{{{
-------------------------------------------------------------------------------
| Category | System calls |
+-------------------+---------------------------------------------------------+
| System calls with | access, chdir, chmod, chown, chroot, creat, dumpcore+, |
| a path name | exec, link, lstat, mkdir, mknod, mount, open, readlink, |
| argument | rename, rmdir, stat, statvfs, symlink, truncate, umount,|
| | unlink, utime |
+-------------------+---------------------------------------------------------+
| System calls with | close, fchdir, fcntl, fstat, fstatvfs, ftruncate, |
| a file descriptor | getdents, ioctl, llseek, pipe, read, select, write |
| argument | |
+-------------------+---------------------------------------------------------+
| System calls with | fsync++, getvfsstat, sync, umask |
| other or no | |
| arguments | |
-------------------------------------------------------------------------------
}}}
Table 3: System call categories.
+ path name argument is implicit, the path name is "core.<pid>"
++ although fsync actually provides a file descriptor argument, it's only
used to find the vmnt and not to do any actual operations on
Before we describe what kind of vmnt locks VFS applies to system calls with a
path name or other arguments, we need to make some notes on path lookup. Path
lookups take arbitrary paths as input (relative and absolute). They can start
at any vmnt (based on root directory and working directory of the process doing
the lookup) and visit any file system in arbitrary order, possibly visiting
the same file system more than once. As such, VFS can never tell in advance
at which File Server a lookup will end. This has the following consequences:
* In the lookup procedure, only one vmnt must be locked at a time. When
moving from one vmnt to another, the first vmnt has to be unlocked before
acquiring the next lock to prevent deadlocks.
* The lookup procedure must lock each visited file system with TLL_READSER
and downgrade or upgrade to the lock type desired by the caller for the
destination file system (as VFS cannot know which file system is final). This
is to prevent deadlocks when a thread acquires a TLL_READSER on a vmnt and
another thread TLL_READ on the same vmnt. If the second thread is blocked
on the first thread due to it acquiring a lock on a vnode, the first thread
will be unable to upgrade a TLL_READSER lock to TLL_WRITE.
We use the following mapping for vmnt locks onto three-level lock types:
{{{
-------------------------------------------------------------------------------
| Lock type | Mapped to | Used for |
+------------+-------------+--------------------------------------------------+
| VMNT_READ | TLL_READ | Read-only operations and fully independent write |
| | | operations |
+------------+-------------+--------------------------------------------------+
| VMNT_WRITE | TLL_READSER | Independent create and modify operations |
+------------+-------------+--------------------------------------------------+
| VMNT_EXCL | TLL_WRITE | Delete and dependent write operations |
-------------------------------------------------------------------------------
}}}
Table 4: vmnt to tll lock mapping
The following table shows a sub-categorization of system calls without a
file descriptor argument, together with their locking types and motivation
as used by VFS.
{{{
-------------------------------------------------------------------------------
| Group | System calls | Lock type | Motivation |
+-------------+--------------+------------+-----------------------------------+
| File open | chdir, | VMNT_READ | These operations do not interfere |
| ops. | chroot, exec,| | with each other, as vnodes can be |
| (non-create)| open | | opened concurrently, and open |
| | | | operations do not affect |
| | | | replicated state. |
+-------------+--------------+------------+-----------------------------------+
| File create-| creat, | VMNT_EXCL | File create ops. require mutual |
| and-open | open(O_CREAT)| for create | exclusion from concurrent file |
| ops | | VMNT_WRITE | open ops. If the file already |
| | | for open | existed, the VMNT_WRITE lock that |
| | | | is necessary for the lookup is |
| | | | not upgraded |
+-------------+--------------+------------+-----------------------------------+
| File create-| pipe | VMNT_READ | These create nameless inodes |
| unique-and- | | | which cannot be opened by means |
| open ops. | | | of a path. Their creation |
| | | | therefore does not interfere with |
| | | | anything else |
+-------------+--------------+------------+-----------------------------------+
| File create-| mkdir, mknod,| VMNT_WRITE | These operations do not affect |
| only ops. | slink | | any VFS state, and can therefore |
| | | | take place concurrently with open |
| | | | operations |
+-------------+--------------+------------+-----------------------------------+
| File info | access, lstat| VMNT_READ | These operations do not interfere |
| retrieval or| readlink,stat| | with each other and do not modify |
| modification| utime | | replicated state |
+-------------+--------------+------------+-----------------------------------+
| File | chmod, chown,| VMNT_READ | These operations do not interfere |
| modification| truncate | | with each other. They do need |
| | | | exclusive access on the vnode |
| | | | level |
+-------------+--------------+------------+-----------------------------------+
| File link | link | VMNT_WRITE | Identical to file create-only |
| ops. | | | operations |
+-------------+--------------+------------+-----------------------------------+
| File unlink | rmdir, unlink| VMNT_EXCL | These must not interfere with |
| ops. | | | file create operations, to avoid |
| | | | the scenario where inodes are |
| | | | reused immediately. However, due |
| | | | to necessary path checks, the |
| | | | vmnt is first locked VMNT_WRITE |
| | | | and then upgraded |
+-------------+--------------+------------+-----------------------------------+
| File rename | rename | VMNT_EXCL | Identical to file unlink |
| ops. | | | operations |
+-------------+--------------+------------+-----------------------------------+
| Non-file | sync, umask, | VMNT_READ | umask does not involve the file |
| ops. | getvfsstat | or none | system, so it does not need |
| | | | locks. sync does not alter state |
| | | | in VFS and is atomic at the FS |
| | | | level. getvfsstat caches stats |
| | | | only and requires no exclusion. |
-------------------------------------------------------------------------------
}}}
Table 5: System call without file descriptor argument sub-categorization
=== Vnode (open file) locking ===
## 4.6 Vnode (open file) locking
Compared to vmnt locking, vnode locking is relatively straightforward. All
read-only accesses to vnodes that merely read the vnode object's fields are
allowed to be concurrent. Consequently, all accesses that change fields
of a vnode object must be exclusive. This leaves us with creation and
destruction of vnode objects (and related to that, their reference counts);
it's sufficient to serialize these accesses. This follows from the fact
that a vnode is only created when the first user opens it, and destroyed
when the last user closes it. A open file in process A cannot be be closed
by process B. Note that this also relies on the fact that a process can do
only one system call at a time. Kernel threads would violate this assumption.
We use the following mapping for vnode locks onto three-level lock types:
{{{
-------------------------------------------------------------------------------
| Lock type | Mapped to | Used for |
+------------+-------------+--------------------------------------------------+
| VNODE_READ | TLL_READ | Read access to previously opened vnodes |
+------------+-------------+--------------------------------------------------+
| VNODE_OPCL | TLL_READSER | Creation, opening, closing, and destruction of |
| | | vnodes |
+------------+-------------+--------------------------------------------------+
| VNODE_WRITE| TLL_WRITE | Write access to previously opened vnodes |
-------------------------------------------------------------------------------
}}}
Table 6: vnode to tll lock mapping
When vnodes are destroyed, they are initially locked with VNODE_OPCL. After
all, we're going to alter the reference count, so this must be serialized. If
the reference count then reaches zero we obtain exclusive access. This should
always be immediately possible unless there is a consistency problem. See
section 4.8 for an exhaustive listing of locking methods for all operations on
vnodes.
=== Filp (file position) locking ===
## 4.7 Filp (file position) locking
The main fields of a filp object that are shared between various processes
(and by extension threads), and that can change after object creation,
are filp_count and filp_pos. Writes to and reads from filp object must be
mutually exclusive, as all system calls have to use the latest version. For
example, a read(2) call changes the file position (i.e., filp_pos), so two
concurrent reads must obtain exclusive access. Consequently, as even read
operations require exclusive access, filp object don't use three-level locks,
but only mutexes.
System calls that involve a file descriptor often access both the filp and
the corresponding vnode. The locking order requires us to first lock the
vnode and then the filp. This is taken care of at the filp level. Whenever
a filp is locked, a lock on the vnode is acquired first. Conversely, when
a filp is unlocked, the corresponding vnode is also unlocked. A convenient
consequence is that whenever a vnode is locked exclusively (VNODE_WRITE),
all corresponding filps are implicitly locked. This is of particular use
when multiple filps must be locked at the same time:
* When opening a named pipe, VFS must make sure that there is at most one filp for the reader end and one filp for the writer end.
* Pipe readers and writers must be suspended in the absence of (respectively) writers and readers.
Because both filps are linked to the same vnode object (they are for the same
pipe), it suffices to exclusively lock that vnode instead of both filp objects.
In some cases it can happen that a function that operates on a locked filp,
calls another function that triggers another lock on a different filp for
the same vnode. For example, close_filp. At some point, close_filp() calls
release() which in turn will loop through the filp table looking for pipes
being select(2)ed on. If there are, the select code will lock the filp and do
operations on it. This works fine when doing a select(2) call, but conflicts
with close(2) or exit(2). Lock_filp() makes an exception for this situation;
if you've already locked a vnode with VNODE_OPCL or VNODE_WRITE when locking
a filp, you obtain a "soft lock" on the vnode for this filp. This means
that lock_filp won't actually try to lock the vnode (which wouldn't work),
but flags the vnode as "skip unlock_vnode upon unlock_filp." Upon unlocking
the filp, the vnode remains locked, the soft lock is removed, and the filp
mutex is released. Note that this scheme does not violate the locking order;
the vnode is (already) locked before the filp.
A similar problem arises with create_pipe. In this case we obtain a new vnode
object, lock it, and obtain two new, locked, filp objects. If everything works
out and the filp objects are linked to the same vnode, we run into trouble
when unlocking both filps. The first filp being unlocked would work; the
second filp doesn't have an associated vnode that's locked anymore. Therefore
we introduced a plural unlock_filps(filp1, filp2) that can unlock two filps
that both point to the same vnode.
=== Lock characteristics per request type ===
## 4.8 Lock characteristics per request type
For File Servers that support concurrent requests, it's useful to know which
locking guarantees VFS provides for vmnts and vnodes, so it can take that
into account when protecting internal data structures. READ = TLL_READ,
READSER = TLL_READSER, WRITE = TLL_WRITE. The vnode locks applies to the
REQ_INODE_NR field in requests, unless the notes say otherwise.
{{{
------------------------------------------------------------------------------
| request | vmnt | vnode | notes |
+--------------+---------+---------+-----------------------------------------+
| REQ_BREAD | | READ | VFS serializes reads from and writes to |
| | | | block special files |
+--------------+---------+---------+-----------------------------------------+
| REQ_BWRITE | | WRITE | VFS serializes reads from and writes to |
| | | | block special files |
+--------------+---------+---------+-----------------------------------------+
| REQ_CHMOD | READ | WRITE | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_CHOWN | READ | WRITE | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_CREATE | WRITE | WRITE | The directory in which the file is |
| | | | created is write locked |
+--------------+---------+---------+-----------------------------------------+
| REQ_FLUSH | | | Mutually exclusive to REQ_BREAD and |
| | | | REQ_BWRITE |
+--------------+---------+---------+-----------------------------------------+
| REQ_FTRUNC | READ | WRITE | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_GETDENTS | READ | READ | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_INHIBREAD| | READ | |
+--------------+---------+---------+-----------------------------------------+
| REQ_LINK | READSER | WRITE | REQ_INODE_NR is locked READ |
| | | | REQ_DIR_INO is locked WRITE |
+--------------+---------+---------+-----------------------------------------+
| REQ_LOOKUP | READSER | | |
+--------------+---------+---------+-----------------------------------------+
| REQ_MKDIR | READSER | WRITE | |
+--------------+---------+---------+-----------------------------------------+
| REQ_MKNOD | READSER | WRITE | |
+--------------+---------+---------+-----------------------------------------+
|REQ_MOUNTPOINT| WRITE | WRITE | |
+--------------+---------+---------+-----------------------------------------+
|REQ_NEW_DRIVER| | | |
+--------------+---------+---------+-----------------------------------------+
| REQ_NEWNODE | | | Only sent to PFS |
+--------------+---------+---------+-----------------------------------------+
| REQ_PUTNODE | | READSER | READSER when dropping all but one |
| | | or WRITE| references. WRITE when final reference |
| | | | is dropped (i.e., no longer in use) |
+--------------+---------+---------+-----------------------------------------+
| REQ_RDLINK | READ | READ | In some circumstances stricter locking |
| | | | might be applied, but not guaranteed |
+--------------+---------+---------+-----------------------------------------+
| REQ_READ | | READ | |
+--------------+---------+---------+-----------------------------------------+
|REQ_READSUPER | WRITE | | |
+--------------+---------+---------+-----------------------------------------+
| REQ_RENAME | WRITE | WRITE | |
+--------------+---------+---------+-----------------------------------------+
| REQ_RMDIR | WRITE | WRITE | |
+--------------+---------+---------+-----------------------------------------+
| REQ_SLINK | READSER | READ | |
+--------------+---------+---------+-----------------------------------------+
| REQ_STAT | READ | READ | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_STATVFS | READ | READ | vmnt is only locked if file is not |
| | | | already opened |
+--------------+---------+---------+-----------------------------------------+
| REQ_SYNC | READ | | |
+--------------+---------+---------+-----------------------------------------+
| REQ_UNLINK | WRITE | WRITE | |
+--------------+---------+---------+-----------------------------------------+
| REQ_UNMOUNT | WRITE | | |
+--------------+---------+---------+-----------------------------------------+
| REQ_UTIME | READ | READ | |
+--------------+---------+---------+-----------------------------------------+
| REQ_WRITE | | WRITE | |
-----------------------------------------------------------------------------+
}}}
Table 7: VFS-FS requests locking guarantees
== Recovery from driver crashes ==
## 5 Recovery from driver crashes
VFS can recover from block special file and character special file driver
crashes. It can recover to some degree from a crashed File Server (which we
can regard as a driver).
=== Recovery from block drivers crashes ===
## 5.1 Recovery from block drivers crashes
When reading or writing, VFS doesn't communicate with block drivers directly,
but always through a File Server (the root File Server being default). If the
block driver crashes, the File Server does most of the work of the recovery
procedure. VFS loops through all open files for block special files that
were handled by this driver and reopens them. After that it sends the new
endpoint to the File Server so it can finish the recover procedure. Finally,
the File Server will retry pending requests if possible. However, reopening
files can cause the block driver to crash again. When that happens, VFS will
stop the recovery. A driver can return ERESTART to VFS to tell it to retry
a request. VFS does this with an arbitrary maximum of 5 attempts.
=== Recovery from character driver crashes ===
## 5.2 Recovery from character driver crashes
While VFS used to support minimal recovery from character driver crashes, the
added complexity has so far proven to outweigh the benefits, especially since
such crash recovery can never be fully transparent: it depends entirely on the
character device as to whether repeating an I/O request makes sense at all.
Currently, all operations except close(2) on a file descriptor that identifies
a device on a crashed character driver, will result in an EIO error. It is up
to the application to reopen the character device and retry whatever it was
doing in the appropriate manner. In the future, automatic reopen and I/O
restart may be reintroduced for a limited subset of character drivers.
=== Recovery from File Server crashes ===
## 5.3 Recovery from File Server crashes
At the time of writing we cannot recover from crashed File Servers. When
VFS detects it has to clean up the remnants of a File Server process (i.e.,
through an exit(2)), it marks all associated file descriptors as invalid
and cancels ongoing and pending requests to that File Server. Resources that
were in use by the File Server are cleaned up.
[0] http://wiki.minix3.org/en/DevelopersGuide/VfsFsProtocol
[1] http://www.cs.vu.nl/~dcvmoole/minix/blockchar.txt
[2] http://www.minix3.org/theses/moolenbroek-multimedia-support.pdf