SOCK_DGRAM, we can send many pages at once without hitting any bug
- when creating a file, it is open for FUSE, but not for the kernel.
If the kernel does not do a subsequent open, we have a leak. We fight
against this by trying to close such file that the kernel left unopen
for some time.
- some code refactoring to make message exchange debug easier (more to come)
assume datagram semantics: when using SOCK_STREAM, if perfused sends
frames faster than the filesystem consumes them, it will grab multiple
frames at once and discard anything beyond the first one. For now the
code can work both with SOCK_DGRAM and SOCK_STREAM, but SOCK_STREAM
support will probably have to be removed for the sake of readability.
- Remeber to sync parent directories when moving a node
- In debug output, display the requeue type (readdir, write, etc...)
if EBADF is returned due to a read from a write-only stream, also
set ferror()
found by a selftest of Python-2.6.6, thanks to obache for pointing
out the details
- directories can be open R/W (for FSYNCDIR)
- do not skip calls to FSYNC or FSYNCDIR if the filesystem returned ENOSYS:
it may change its mind, and it may also actually do something when retunring
ENOSYS
- When FSYNC and FSYNCDIR return ENOSYS, do not report it to kernel (silent
failure)
therefore runs with filesystem privileges
- shut up warnings and debug messages when perfused is autostarted
- make perfused patch modifiable with CFLAGS for easier pkgsrc integration
- Fix build warnings
- after a node is deleted, some operations should return ENOENT, some
should be ignored. Fixed it for ACCESS, SETATTR and GETATTR. Other
operation may also need a fix.
- At reclaim time, there is no need to wait for READDIR and READ
completion, since the caller will never close a file before getting
readir() and read() replies. Waiting for WRITE completion is still
mandatory, but we must ensure that no queued WRITE is awaiting to
be scheduled. Once the queue is drained, we must check that the
reclaim operation was not canceled by a new file LOOKUP.
- At reclaim time, fixed a mix up between read and write fh to close
- Fixed permission checks for RENAME (it tested the node itself
instead of the source)
- When seting file mode, only MKNOD needs the filetype (S_* fields).
It is probably a bug to set it for other operations.
tre will be compiled without approx and wchar/mulibyte support to
only match the minimum requirement to replace our spencer regex.
This needs a lot of testing.
Only enabled when USE_LIBTRE is set to `yes'.
setattr(mtime, ctime) after close, while FUSE expects the file
to be open for these operations
- remove unused argument to node_mk_common()
- remove requeued requests when they are executed, not when they
are tagged for schedule
- try to make filehandle management simplier, by keeping track of only
one read and one write filehandle (the latter being really read/write).
- when CREATE is not available, we use the MKNOD/OPEN path. Fix a
bug here where we opened the parent directory instead of the node:
add the missing lookup of the mknod'ed node.
- lookup file we just created: glusterfs does not really see them
otherwise.
- open file when doing setattr(mtime, ctime) on non open files, as
some filesystems seems to require it.
- Do not flush pagecache for removed nodes
- Keep track of read/write operations in progress, and at reclaim
time, make sure they are over before closing and forgeting the file.
FUSE filesystems will attempt to start it on their own, and will
communicate using a socketpair
- do not advertise NULL file handle as being valid when sending themback to the FUSE filesystem.
- unmount if we cannot talk to the FUSE process anymore
- set calling process gid properly
- debug message cleanup
- Automatically call fsync on close for files. If we just close, fsync
will come later and we will have to reopen
- Add a PND_DIRTY flag to keep track of files that really need a sync.
perfuse_node_fsync only calls the FUSE fsync method if there are data
to push.
- Add a PND_OPEN flag to keep track of open files. Checking non NULL
fh is not enough, as some filesystems will always set fh to 0.
- Add a sync diagnostic flag, to watch fsync and dirty flag activity.
Make the fh diagnostic flag more verbose
- Send the fh in setattr (it was hardcoded to 0)
I am now able to build libperfuse in a glusterfs mounted filesystem. Yeah!
FUSE filesystem must be patched to #include <perfuse.h> in the source
files that open /dev/fuse and perform the mount(2) system call. The
FUSE filesystem must be linked with -lperfuse.
libperfuse(3) implements the FUSE kernel interface, on which libfuse or
any FUSE filesystem that opens /dev/fuse directly can be used.
For now, an external daemon called perfused(8) is used. This may change
in the future.
-mk/bsd.lib.mk picks up a .S asm file behind our back (did it do so
always?). s_modf.S is incorrect; I'm undecided whether it makes sense
to fix it, so add a stupid rule to enforce the .c file to be used.
-on i386, s_tanh.c gets miscompiled. It works with -O0, so add
COPTS for exactly that file.
between all platforms and a small assembler stub in crt0.S is used for
setting up the proper stack and whatever else MD needs.
Extract crti.S and crtn. from the old dot_init.h files. Prepare other
platforms that use the 6-argument form of __(_)start for this.
Rewrite the crtbegin and crtend modules in assembler to make them
compiler independent. Document the interface in README.
behavior. This unbreaks GCC 4.4's libgfortran build with the old
signal.h logic, because GCC decided to put the body for the sigsetop
functions in multiple objects.
functions in NetBSD, which does not support additional condition attributes.
Note that these do not conform to SUSv4, where two additional attributes are
mandated.
- pthread_barrier_{init,destroy,wait} into pthread_barrier(3).
- pthread_cond_* into pthread_cond(3).
- pthread_mutex_* into pthread_mutex(3).
- pthread_rwlock_* into pthread_rwlock(3).
- pthread_spin_* into pthread_spin(3).
Cross-link amongst groups.
OK jruoho@
document pthread_attr_getstack() and pthread_attr_setstack(). Try to also
document these a little better (in particular, note some items in the long
list of caveats and questions related to application-controlled thread stack).
we're ELF now, and there are many missing checks against OBJECT_FMT.
if we ever consider switching, the we can figure out what new ones
we need but for now it's just clutter.
this doesn't remove any of the support for exec_aout or any actually
required-for-boot a.out support, only the ability to build a netbsd
release in a.out format. ie, most of this code has been dead for
over a decade.
i've tested builds on vax, amd64, i386, mac68k, macppc, sparc, atari,
amiga, shark, cats, dreamcast, landisk, mmeye and x68k. this covers
the 5 MACHINE_ARCH's affected, and all the other arch code touched.
it also includes some actual run-time testing of sparc, i386 and
shark, and i performed binary comparison upon amiga and x68k as well.
some minor details relevant:
- move shlib.[ch] from ld.aout_so into ldconfig proper, and cut them
down to only the parts ldconfig needs
- remove various unused source files
- switch amiga bootblocks to using elf2bb.h instead of aout2bb.h
reference the latter as a return of main() results in a call to exit(3),
so this ensures that the libc constructors are run for statically linked
programs. Fixes PR 37454.
and remove renameing _Rune* -> _NBRune* namespace protection.
FreeBSD traditionaly exposes struct _Rune* in runetype.h
which included by ctype.h. it may cause conflicting type error
in our cross build process, former we use renaming namespace
to avoid this problem, now i reworked more resonable way.
2. merge rune_local.h to runetype_local.h, and remove it.
3. split bsdctype.h -> bsdctype_{file,local}.h
the crash went away -- it's not clear that it went away for any reason
besides by accident.
The change is: use calloc instead of malloc for allocating SCREENs.
2. unfortunately hppa's MB_LEN_MAX is defined incorrectly 6 instead of 32
so we have to add more setlocale(3) __RENAME func, __setlocale50.
3. move setlocale1.c and setlocale32.c to lib/libc/compat/locale/*
prepareing for next libc major crunk.
4. bump libc minor version.
This makes the size the same on 64bit archs. Don't bother bumping
any version, since you'd have explicitly had to jump through some
hoops to use pathconf before.
little finetuning (not that everyone is jumping right to using it,
but fixing will have to wait until tomorrow, and who knows what
i'll forget during the night and how long it will eventually take).
the signal number now, as required by POSIX. The latter should just
return 0 on success. Fixes a mysql problem reported by Kurt Schreiner.
-document the new behavior of sigwaitinfo and sigtimedwait
-retain non-POSIX behavior for NetBSD-5 binary compatibility -- the fix
would likely not be polled into the .0-branches, and having 5.0.x and
5.x behave differently would be confusing. Besides that, it was
documented in the manpage all the time.
2. do not use _CTYPE_NUM_CHARS macro to read data from LC_CTYPE(old BSDCTYPE style) database.
because 1<<CHAR_BIT is MD, so i added MI macro _CTYPE_CACHE_SIZE(1<<8).
3. remove _NB_CACHED_RUNE macro, use _CTYPE_CACHE_SIZE instead.
interlock. This is applicable in cases where the actual interlock
is the CPU the currently running thread is scheduled on. Borrowing
the scheduler lock as the mutex mandated by pthread_cond_wait()
does away with need to have an additional mutex. This both optimizes
runtime execution and simplifies code, as the extra lock typically
lead to quite some trickeries to avoid the dungeon collapsing due
to zaps from the wand of deadlock.
