changelog

UNIX/LINUX下面的用户文件系统

	* kernel: detect newer UML kernels

2005-07-26  Miklos Szeredi <[e-mail]>

	* Make the installation path of fuse.ko and mount.fuse
	configurable through INSTALL_MOD_PATH and MOUNT_FUSE_PATH
	environment variables.  Requirement and help from Csaba Henk.

2005-07-22  Miklos Szeredi <[e-mail]>

	* Fix bug, that causes filesystem requests to hang when unique
	request counter becomes negative.  This happens after
	2,147,483,648 operations, so most people won't care.  Thanks to
	Franco Broi for the report and testing.

2005-07-21  Miklos Szeredi <[e-mail]>

	* Don't change mtime/ctime/atime to local time on read/write.
	Bug reported by Ben Grimm

	* Install fuse_common.h and fuse_lowlevel.h.  Report by Christian
	Magnusson

	* fusermount: use getopt_long() for option parsing.  It allows the
	use of '--' to stop argument scanning, so fusermount can now
	operate on directories whose names begin with a '-'.  Patch by
	Adam Connell

2005-07-15  Miklos Szeredi <[e-mail]>

	* fusermount: add '-v', '--version' and '--help' options

	* add inode based API

2005-07-12  Miklos Szeredi <[e-mail]>

	* lib: don't block signals in worker threads.  Problem noticed by
	Usarin Heininga

2005-07-07  Miklos Szeredi <[e-mail]>

	* lib: don't allow both 'allow_other' and 'allow_root' options to
	be given

2005-07-06  Miklos Szeredi <[e-mail]>

	* fusermount: check if mountpoint is empty (only '.' and '..' for
	directories, and size = 0 for regular files).  If "nonempty"
	option is given, omit this check.  This is useful, so users don't
	accidentally hide data (e.g. from backup programs).  Thanks to
	Frank van Maarseveen for pointing this out.

	* kernel: check if mandatory mount options ('fd', 'rootmode',
	'user_id', 'group_id') are all given

	* lib: simplify 'readdir_ino' handling

	* lib: add mount options 'umask=M', 'uid=N', 'gid=N'

2005-07-03  Miklos Szeredi <[e-mail]>

	* kernel: clean up 'direct_io' code

2005-06-28  Miklos Szeredi <[e-mail]>

	* Add 'mount.fuse' written by Petr Klima

	* '/dev/fuse' is created by 'make install' if does not yet exist

2005-06-20  Miklos Szeredi <[e-mail]>

	* Fix UCLIBC compile error.  Patch by Christian Magnusson

2005-06-08  Miklos Szeredi <[e-mail]>

	* Enable the auto-loading of the module via access to the
	corresponding device file.  Patch by Takashi Iwai.

	* Allow mounting a regular file (over a regular file) for
	unprivleged users.

	* Do not create temporary device file.  Require "/dev/fuse" to
	exist, and be readable/writable by the mounting user.

2005-06-02  Miklos Szeredi <[e-mail]>

	* Released 2.3.0

2005-06-02  Miklos Szeredi <[e-mail]>

	* Fix serious information leak: if the filesystem returns a short
	byte count to a read request, and there are non-zero number of
	pages which are not filled at all, these pages will not be zeroed.
	Hence the user can read out previous memory contents.  Found by
	Sven Tantau.

2005-05-27  Miklos Szeredi <[e-mail]>

	* Add "readdir_ino" mount option, which tries to fill in the d_ino
	field in struct dirent.  This mount option is ignored if "use_ino"
	is used.  It helps some programs (e.g. 'pwd' used over NFS from a
	non-Linux OS).  Patch by David Shaw.

2005-05-12  Miklos Szeredi <[e-mail]>

	* Released 2.3-rc1

2005-05-12  Miklos Szeredi <[e-mail]>

	* File save in krusader and other editors doesn't work with sshfs,
	because open() is interrupted by a periodic signal, and open()
	restarts forever, without any progress.  This could just be fixed
	in open(), but the problem is more generic: if signals are
	received more often than the filesystem can get the request to
	userspace, it will never finish.  This is probably only a
	theoretical problem, nevertheless I'm removing the possibility to
	interrupt requests with anything other than SIGKILL, even before
	being sent to userspace.  Bugreport by Eduard Czimbalmos.

2005-05-09  Miklos Szeredi <[e-mail]>

	* libfuse: add "tree_lock" rwlock, that is locked for write in
	rename, unlink and rmdir, and locked for read in all other
	operations.  This should fix the rename/release race reported by
	Valient Gough and others.  The solution is very coarse, a finer
	grained locking scheme could be implemented, but it would be much
	more complex.  Let's see whether this is good enough.

2005-05-09  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre7

2005-05-08  Miklos Szeredi <[e-mail]>

	* Better fix for out of order FORGET messages.  Now the
	LOOKUP/FORGET messages are balanced exactly (one FORGET can
	balance many lookups), so the order no longer matters.  This
	changes the kernel ABI slightly, but the library remains backward
	compatible.

2005-05-06  Miklos Szeredi <[e-mail]>

	* Fix abort for out of order FORGET messages.  Again.  Spotted by
	Franco Broi again.  Sorry :)

2005-04-29  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre6

2005-04-29  Miklos Szeredi <[e-mail]>

	* Make fusermount work with fuse kernel modules not yet supporting
	the "group_id" option (added for the purpose of stricter
	permission checking).

2005-04-28  Miklos Szeredi <[e-mail]>

	* Check for hard-linked directories in lookup.  This could cause
	problems in the VFS, which assumes that such objects never exist.

	* Make checking of permission for other users more strict.  Now
	the same privilege is required for the mount owner as for ptrace
	on the process performing the filesystem operation.

2005-04-23  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre5

2005-04-22  Miklos Szeredi <[e-mail]>

	* Add -msoft-float to kernel module compile flags for 2.4.X.  This
	is needed on certain architectures.  Report from Chris Kirby

	* Fix buggy behavior of open(..., O_CREAT|O_EXCL) if interrupted.
	Reported by David Shaw

	* Remove "allow_root" option from kernel module, and implement
	it's functionality in the library

	* Fix Oops caused by premature release of fuse_conn.  Clean up
	related code, to be more readable

	* Sendfile should not use page cache if "direct_io" mount option
	is given

2005-04-08  Miklos Szeredi <[e-mail]>

	* Fix Oops in case of nfs export.  Spotted by David Shaw

	* Fix another Oops in case of write over nfs with direct_io turned
	on.  Again spotted by David Shaw

2005-04-07  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre4

2005-04-07  Miklos Szeredi <[e-mail]>

	* lib: finalized new readdir() interface, which now supersedes the
	getdir() method.

2005-04-03  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre3

2005-04-03  Miklos Szeredi <[e-mail]>

	* Implement backward compatibility with version 5 kernel ABI

2005-04-01  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre2

2005-04-01  Miklos Szeredi <[e-mail]>

	* kernel: fix dirent offset handling

	* lib: add readdir and releasedir methods

	* lib: use fh field of fuse_file_info in opendir, readdir,
	releasedir and fsyncdir methods

	* lib: check kernel API version and bail out of it's old.  This
	will be properly fixed in the next release

2005-03-31  Miklos Szeredi <[e-mail]>

	* Released 2.3-pre1

2005-03-31  Miklos Szeredi <[e-mail]>

	* kernel API: add padding to structures, so 64bit and 32bit
	compiler will return the same size

	* kernel API: add offset field to fuse_dirent.  This will allow
	more sophisticated readdir interface for userspace

	* kernel API: change major number to 6

	* kernel: fix warnings on 64bit archs

	* kernel: in case of API version mismatch, return ECONNREFUSED

2005-03-24  Miklos Szeredi <[e-mail]>

	* kernel: trivial cleanups

2005-03-21  Miklos Szeredi <[e-mail]>

	* Add fsyncdir() operation

2005-03-19  Miklos Szeredi <[e-mail]>

	* kernel: add locking to background list (fixes previous fix)

2005-03-18  Miklos Szeredi <[e-mail]>

	* kernel: fix bug which could cause leave busy inodes after
	unmount, and Oops.

2005-03-08  Miklos Szeredi <[e-mail]>

	* examples: add -lpthread to link flags to work around valgrind
	quirk

	* lib: don't exit threads, so cancelation doesn't cause segfault

2005-03-04  Miklos Szeredi <[e-mail]>

	* kernel: fix nasty bug which could cause an Oops under certain
	situations.  Found by Magnus Johansson

2005-02-28  Miklos Szeredi <[e-mail]>

	* libfuse: added opendir() method.  This can be used in case
	permission checking in getdir() is too late.  Thanks to Usarin
	Heininga for pointing out this deficiency

	* libfuse: added init() and destroy() methods to fuse_operations

	* kernel: llseek() method for files and directories made explicit

	* kernel: fixed inode leak in NFS export in case of nodeid
	wrapping

2005-02-15  Miklos Szeredi <[e-mail]>

	* libfuse: clean up some unitialized memory found with valgrind

	* Add -lpthread to Libs in fuse.pc.  Valgrind seems to need an
	explicitly linked libpthread for applications

2005-02-10  Miklos Szeredi <[e-mail]>

	* fusermount: set umask, otherwise /etc/mtab will have
	unpredictable permission.  Spotted by Jindrich Kolorenc

	* fusermount: set owner and group of /etc/mtab to original values
	on unmount

	* libfuse: add 'use_ino' option to help.  Patch by Valient Gough

2005-02-07  Miklos Szeredi <[e-mail]>

	* Cleaned up directory reading (temporary file is not used)

2005-02-02  Miklos Szeredi <[e-mail]>

	* Released 2.2

2005-02-02  Miklos Szeredi <[e-mail]>

	* Fix possible race when operation is interrupted

2005-01-28  Miklos Szeredi <[e-mail]>

	* Fix compilation on 2.6.7

2005-01-26  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre6

2005-01-26  Miklos Szeredi <[e-mail]>

	* Fix bug in link() operation which caused the wrong path to be
	passed as the first argument.  Found by Anton Altaparmakov

2005-01-21  Miklos Szeredi <[e-mail]>

	* LIB: fix double reply in readdir operation

	* fusermount: fix uid checking bug.  Patch by Adam Connell

	* KERNEL: fix compile on various RedHat patched 2.4 kernels.
	Patch by Keshava Gowda

2005-01-20  Miklos Szeredi <[e-mail]>

	* KERNEL: provide correct llseek semantics for fuse device (fixes
	a bug on Progeny 2.4.20 kernel).  Reported by Valient Gough

2005-01-20  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre5 (matches kernel 2.6.11-rc1-mm2)

2005-01-18  Miklos Szeredi <[e-mail]>

	* KERNEL ABI: remove GETDIR operation, and add OPENDIR, READDIR
	and RELEASEDIR.  This ends the ugly hack of passing a file
	descriptor to the kernel, and actually makes the code simpler.

2005-01-17  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre4

2005-01-17  Miklos Szeredi <[e-mail]>

	* fusermount: remove capability setting, which was the cause of
	problems for some users.  It seems that FS related capabilities
	are removed by setfsuid(), so this isn't even needed.

2005-01-15  Miklos Szeredi <[e-mail]>

	* fix compilation on 2.4 kernels (reported by Valient Gough)

	* fix failure to unmount bug (found by David Shaw)

	* fusermount: improve parsing of /etc/fuse.conf

2005-01-13  Miklos Szeredi <[e-mail]>

	* Remove 'mount_max' and 'user_allow_other' module options.  These
	are now checked by fusermount, and can be set in /etc/fuse.conf

	* KERNEL: change check for fsid == 0 to capable(CAP_DAC_OVERRIDE)

2005-01-11  Miklos Szeredi <[e-mail]>

	* KERNEL: fix possible inode allocation problem, where
	sizeof(struct inode) is not aligned (found by Mike Waychison)

	* KERNEL: use new follow_link/put_link methods

	* KERNEL: cosmetic fixes

2005-01-10  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre3

2005-01-10  Miklos Szeredi <[e-mail]>

	* Add missing code that was accidently left out

2005-01-09  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre2

2005-01-09  Miklos Szeredi <[e-mail]>

	* Change "uid" mount option to "user_id" to avoid confusion with a
	mount option "uid" commonly used by many filesystems

2005-01-09  Miklos Szeredi <[e-mail]>

	* Released 2.2-pre1

2005-01-09  Miklos Szeredi <[e-mail]>

	* If FUSE is configured in the kernel, don't build it by default

2005-01-07  Miklos Szeredi <[e-mail]>

	* Compile fix by Christian Magnusson

2005-01-05  Miklos Szeredi <[e-mail]>

	* Fix compilation for 2.6.{0-5} kernels

2005-01-04  Miklos Szeredi <[e-mail]>

	* KERNEL: if request is interrupted, still keep reference to used
	inode(s) and file, so that FORGET and RELEASE are not sent until
	userspace finishes the request.

	* remove /{sys,proc}/fs/fuse/version, and instead add an INIT
	request with the same information, which is more flexible,
	simpler, works on embedded systems.

2004-12-16  Miklos Szeredi <[e-mail]>

	* KERNEL ABI: update interface to make it independent of type
	sizes.  This will help on 64 bit architectures which can run
	legacy 32 bit applications.

	* KERNEL ABI: add "len" field to request headers.  This will allow
	sending/receiving requests in multiple chunks.

	* KERNEL: handle file type change more intelligently

	* LIB: "-o debug" option should disable backgrounding (fix by
	Fabien Reygrobellet)

2004-12-13  Miklos Szeredi <[e-mail]>

	* KERNEL: invalidate dentry/attributes if interrupted request
	could leave filesystem in an unknown state.

2004-12-12  Miklos Szeredi <[e-mail]>

	* KERNEL: lots of cleanups related to avoiding possible deadlocks.
	These will cause some regressions, but stability is considered
	more important.  If any of these features turns out to be
	important, it can be readded with the deadlock problems addressed.

	* Make all requests interruptible (only with SIGKILL currently).
	This can be used to break any deadlock produced by the userspace
	filesystem accessing it's own exported files.  The RELEASE request
	is special, because if it's interrupted before sending it to
	userspace it is still sent, but the reply is not awaited.

	* If request is interrupted before being sent to userspace, and if
	it hasn't yet got any side effects, it is always restarted,
	regardless of the SA_RESTART flag.  This makes these interruptions
	transparent to the process.

	* Remove shared-writable mmap support, which was prone to an
	out-of-memory deadlock situation

	* Remove INVALIDATE userspace initiated request