]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
19 years ago - The VI_DOOMED flag now signals the end of a vnode's relationship with
jeff [Sun, 13 Mar 2005 12:03:14 +0000 (12:03 +0000)]
 - The VI_DOOMED flag now signals the end of a vnode's relationship with
   the filesystem.  Check that rather than VI_XLOCK.
 - Shorten ffs_reload by one step.  The old check for an inactive vnode
   was slightly racey, and the code which deals with still active vnodes
   is not much more expensive.

Sponsored by: Isilon Systems, Inc.

19 years ago - The VI_DOOMED flag now signals the end of a vnode's relationship with
jeff [Sun, 13 Mar 2005 12:01:50 +0000 (12:01 +0000)]
 - The VI_DOOMED flag now signals the end of a vnode's relationship with
   the filesystem.  Check that rather than VI_XLOCK.

Sponsored by: Isilon Systems, Inc.

19 years ago - Fix an assert now that the XLOCK no longer exists.
jeff [Sun, 13 Mar 2005 12:00:41 +0000 (12:00 +0000)]
 - Fix an assert now that the XLOCK no longer exists.

Sponsored by: Isilon Systems, Inc.

19 years ago - In ufs_mknod(), hold the lock across the call to vgone() as that is now
jeff [Sun, 13 Mar 2005 11:59:14 +0000 (11:59 +0000)]
 - In ufs_mknod(), hold the lock across the call to vgone() as that is now
   required.
 - In ufs_close(), don't do the EAGAIN vrele hack, the top layer now calls
   vn_start_write before the lock is acquired as it should.

Sponsored by: Isilon Systems, Inc.

19 years ago - Don't drop the lock in ufs_inactive().
jeff [Sun, 13 Mar 2005 11:57:39 +0000 (11:57 +0000)]
 - Don't drop the lock in ufs_inactive().
 - Also in ufs_inactive, don't acquire the vnode interlock where it isn't
   strictly needed.  Also owning the vnode interlock while calling vprint()
   will cause locking assertions to trip.

Sponsored by: Isilon Systems, Inc.

19 years ago - Do a vn_start_write in vn_close, we may write if this is the last ref
jeff [Sun, 13 Mar 2005 11:56:28 +0000 (11:56 +0000)]
 - Do a vn_start_write in vn_close, we may write if this is the last ref
   on an unlinked file.  We can't know if this is the case until after we
   have the lock.
 - Lock the vnode in vn_close, many filesystems had code which was unsafe
   without the lock held, and holding it greatly simplifies vgone().
 - Adjust vn_lock() to check for the VI_DOOMED flag where appropriate.

Sponsored by: Isilon Systems, Inc.

19 years ago - Remove vx_lock, vx_unlock, vx_wait, etc.
jeff [Sun, 13 Mar 2005 11:54:28 +0000 (11:54 +0000)]
 - Remove vx_lock, vx_unlock, vx_wait, etc.
 - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do
   writing ops without suspending.  This could suspend the vlruproc which
   should not be a problem under normal circumstances.
 - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance
   where it was used.
 - Acquire a lock before calling vgone() as it now requires it.
 - Move the acquisition of the vnode interlock from vtryrecycle() to
   getnewvnode() so that if it fails we don't drop and reacquire the
   vnode_free_list_mtx.
 - Check for a usecount or holdcount at the end of vtryrecycle() in case
   someone grabbed a ref while we were recycling.  Abort the recycle, and
   on the final ref drop this vnode will be placed on the head of the free
   list.
 - Move the redundant VOP_INACTIVE protection code into the local
   vinactive() routine to avoid code bloat.
 - Keep the vnode lock held across calls to vgone() in several places.
 - vgonel() no longer uses XLOCK, instead callers must hold an exclusive
   vnode lock.  The VI_DOOMED flag is set to allow other threads to detect
   a vnode which is no longer valid.  This flag is set until the last
   reference is gone, and there are no chances for a new ref.  vgonel()
   holds this lock across the entire function, which greatly simplifies
   logic.
 _ Only vfree() in one place in vgone() not three.
 - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock
   in the LK_NOWAIT case.  In other cases, check after we have slept and
   acquired an exlusive lock.  This will simulate the old vx_wait()
   behavior.

Sponsored by: Isilon Systems, Inc.

19 years ago - A lock is required before calling VOP_REVOKE. Our reference protects us
jeff [Sun, 13 Mar 2005 11:47:04 +0000 (11:47 +0000)]
 - A lock is required before calling VOP_REVOKE.  Our reference protects us
   from accessing another vnode so a naked VOP_LOCK is sufficient.

Sponsored by: Isilon Systems, Inc.

19 years ago - Don't VOP_UNLOCK prior to VOP_REVOKE. The lock is required now.
jeff [Sun, 13 Mar 2005 11:45:51 +0000 (11:45 +0000)]
 - Don't VOP_UNLOCK prior to VOP_REVOKE.  The lock is required now.

Sponsored by: Isilon Systems, Inc.

19 years ago - Don't drop the lock in the default inactive handler anymore, VOP_NULL
jeff [Sun, 13 Mar 2005 11:45:01 +0000 (11:45 +0000)]
 - Don't drop the lock in the default inactive handler anymore, VOP_NULL
   will do for vop_stdinactive now.

Sponsored by: Isilon Systems, Inc.

19 years ago - Get rid of VXLOCK, VXWANT, and vx_*. The vnode lock now protects us
jeff [Sun, 13 Mar 2005 11:44:02 +0000 (11:44 +0000)]
 - Get rid of VXLOCK, VXWANT, and vx_*.  The vnode lock now protects us
   against recycling.
 - Modify VSHOULDFREE, VCANRECYCLE, etc. now that certain flags are no
   longer important.  Remove VMIGHTFREE as it is only used in one place.

Sponsored by: Isilon Systems, Inc.

19 years ago - CLOSE, REVOKE, INACTIVE, and RECLAIM are not L L L, that's a locked vnode
jeff [Sun, 13 Mar 2005 11:42:16 +0000 (11:42 +0000)]
 - CLOSE, REVOKE, INACTIVE, and RECLAIM are not L L L, that's a locked vnode
   on enter, exit, error.  This allows for the removal of the XLOCK.

Sponsored by: Isilon Systems, Inc.

19 years agoEmbrace with #ifdef DEV_CARP carp-related code.
glebius [Sun, 13 Mar 2005 11:23:22 +0000 (11:23 +0000)]
Embrace with #ifdef DEV_CARP carp-related code.

19 years agoUnhook the recently departed lomac file from the build.
dougb [Sun, 13 Mar 2005 08:07:11 +0000 (08:07 +0000)]
Unhook the recently departed lomac file from the build.

Forgotten by: trhodes (the real one)

19 years agoFix a null pointer de-ref when passthrough ioctls are issued. This
scottl [Sun, 13 Mar 2005 06:25:53 +0000 (06:25 +0000)]
Fix a null pointer de-ref when passthrough ioctls are issued.  This
involves some code shuffle to avoid locking problems.

MFC After: 3 days

19 years agoMove configuration file note from kernel to userland, no content
bmah [Sun, 13 Mar 2005 04:24:51 +0000 (04:24 +0000)]
Move configuration file note from kernel to userland, no content
changes.

Minor markup/grammar fixes:  jail(8) security.jail.chflags_allowed,
cpufreq(4), ng_ipfw(4), ips(4) crash dumps, resolver query order,
manpage rewrites.

19 years agoAdd --strip-components option, per bin/77666.
kientzle [Sun, 13 Mar 2005 04:12:30 +0000 (04:12 +0000)]
Add --strip-components option, per bin/77666.

Thanks to: Sangwoo Shim

19 years agoFix the arrangement of periodic QH tree to give the correct interval
iedowse [Sun, 13 Mar 2005 04:07:40 +0000 (04:07 +0000)]
Fix the arrangement of periodic QH tree to give the correct interval
between passes over a QH. Previously the accesses to a QH were
bunched together in time, so the interval was often much longer
than intended. This now appears to match the diagrams in the EHCI
spec, so remove the XXX comment.

19 years agoSupport extracting entries with pathnames longer than PATH_MAX. In
kientzle [Sun, 13 Mar 2005 02:53:42 +0000 (02:53 +0000)]
Support extracting entries with pathnames longer than PATH_MAX.  In
testing, I've archived and restored dir trees with ~1MB pathnames.
Most formats, of course, have much smaller limits.

19 years agoMFC noted: libgpib.
bmah [Sun, 13 Mar 2005 02:43:01 +0000 (02:43 +0000)]
MFC noted:  libgpib.

Updated release note:  GNOME 2.10

Added trademark notices.

19 years agoWhen rejecting rediculously large pax attributes (such as pathnames
kientzle [Sun, 13 Mar 2005 02:35:52 +0000 (02:35 +0000)]
When rejecting rediculously large pax attributes (such as pathnames
over 1MB), issue a warning instead of forcing an internal assertion
failure.

19 years agoUnbreak build with POLLING. I should really listen and test with NOTES
mlaier [Sun, 13 Mar 2005 01:54:41 +0000 (01:54 +0000)]
Unbreak build with POLLING.  I should really listen and test with NOTES
instead of the module build.

19 years agoCorrectly pass low-level I/O errors back up to the caller when
kientzle [Sun, 13 Mar 2005 01:52:35 +0000 (01:52 +0000)]
Correctly pass low-level I/O errors back up to the caller when
reading cpio format.

19 years agoBetter error messages on read and file-open errors.
kientzle [Sun, 13 Mar 2005 01:51:16 +0000 (01:51 +0000)]
Better error messages on read and file-open errors.

19 years agoInclude detailed error message from zlib after a decompression error.
kientzle [Sun, 13 Mar 2005 01:48:33 +0000 (01:48 +0000)]
Include detailed error message from zlib after a decompression error.

19 years agoIf the output is not a regular file, it's okay to add it to the archive.
kientzle [Sun, 13 Mar 2005 01:47:31 +0000 (01:47 +0000)]
If the output is not a regular file, it's okay to add it to the archive.
In particular, /dev/st0 can be added to an archive being written to /dev/st0.

Thanks to: Jaakko Heinonen

19 years agoAllow kernels loaded by pxeboot, which was compiled with LOADER_TFTP_SUPPORT,
kan [Sat, 12 Mar 2005 21:26:53 +0000 (21:26 +0000)]
Allow kernels loaded by pxeboot, which was  compiled with LOADER_TFTP_SUPPORT,
to stll be able to mount NFS root as prescribed by DCHP configuration. Since
pxeboot is using TFTP to get to the files, pxeboot can not rely on NFS to
provide it a root directory hande as a side effect. pxeboot has to  make RPC
mount call itself.

19 years agoRemove mention of mac_lomac(4) policy from here too.
trhodes [Sat, 12 Mar 2005 21:10:42 +0000 (21:10 +0000)]
Remove mention of mac_lomac(4) policy from here too.

19 years agoRemove mac_lomac(4) functionality. The proper way is to use loader.conf
trhodes [Sat, 12 Mar 2005 21:09:15 +0000 (21:09 +0000)]
Remove mac_lomac(4) functionality.  The proper way is to use loader.conf
or build the policy into a kernel.

Approved by: rwatson

19 years agoAdd NO_GPIB
phk [Sat, 12 Mar 2005 20:42:18 +0000 (20:42 +0000)]
Add NO_GPIB

19 years agoDTRT with /var so that any packages installed are correctly recorded.
phk [Sat, 12 Mar 2005 20:41:26 +0000 (20:41 +0000)]
DTRT with /var so that any packages installed are correctly recorded.

19 years agoEnable packet mode in boot0, this makes it unimportant if the
phk [Sat, 12 Mar 2005 20:40:33 +0000 (20:40 +0000)]
Enable packet mode in boot0, this makes it unimportant if the
geometry is 100% correct on most hardware.

19 years agoNew release notes:
hrs [Sat, 12 Mar 2005 19:04:03 +0000 (19:04 +0000)]
New release notes:
jail(8) security.jail.chflags_allowed sysctl,
zs driver removed,
ng_ipfw(4) added,
ips(4) crash dump support,
getaddrinfo(3) DNS query order changed,
powerd(8) added, and
RFC derived manual pages replaced.

Update release notes:
acpi_throttle and acpi_perf merged into acpi and
can be disabled via device hints, and
some rewording due to zs driver removal in the entry which
explains uart(4) has been enabled by default on FreeBSD/sparc64.

19 years agoALTQ support for re(4).
mlaier [Sat, 12 Mar 2005 17:35:37 +0000 (17:35 +0000)]
ALTQ support for re(4).

Submitted by: Chris Dionissopoulos, Theo Schlossnagle
PR: kern/78681
MFC after: 2 weeks

19 years agoIn uart_cpu_getdev_console() when determinig whether we should use
marius [Sat, 12 Mar 2005 17:06:03 +0000 (17:06 +0000)]
In uart_cpu_getdev_console() when determinig whether we should use
a serial console anyway because input-device is set to keyboard and
output-device is set to screen but no keyboard is plugged in don't
assume that a device node for the input-device alias exists. While
this is true for RS232 keyboards (the node of the SCC and UART
respectively which controls the keyboard doesn't disappear when no
keyboard is plugged in) this assumption breaks for USB keyboards.
It's most likely also not true for PS/2 keyboards but OFW doesn't
reliably switch to a serial console when the potential keyboard is
a PS/2 one which isn't plugged in so this couldn't be verified
properly.

Reported by: Will Andrews <will@csociety.org>, obrien
MFC after: 1 week

19 years agoFunction jailed() looks into ucred strcture, so be sure ucred is not NULL.
pjd [Sat, 12 Mar 2005 14:31:04 +0000 (14:31 +0000)]
Function jailed() looks into ucred strcture, so be sure ucred is not NULL.

Reviewed by: rwatson
MFC after: 1 week

19 years agoClean up a bit.
pjd [Sat, 12 Mar 2005 14:28:34 +0000 (14:28 +0000)]
Clean up a bit.

Reviewed by: rwatson
MFC after: 1 week

19 years agoExtend the coverage of the accept and socket mutexes in soisconnected()
rwatson [Sat, 12 Mar 2005 13:39:39 +0000 (13:39 +0000)]
Extend the coverage of the accept and socket mutexes in soisconnected()
so that the socket lock is held over the test-and-set removal of the
accept filter option during connect, and the two socket mutex regions
(transition to connected, perform accept filter) are combined.

19 years agoAdd a sysctl net.link.log_link_state_change, which allows to
glebius [Sat, 12 Mar 2005 12:58:03 +0000 (12:58 +0000)]
Add a sysctl net.link.log_link_state_change, which allows to
suppress logging of interface link state changes.

Requested by: sam, kan

19 years agoMove the logic implementing retrieval of the SO_ACCEPTFILTER socket option
rwatson [Sat, 12 Mar 2005 12:57:18 +0000 (12:57 +0000)]
Move the logic implementing retrieval of the SO_ACCEPTFILTER socket option
from uipc_socket.c to uipc_accf.c in do_getopt_accept_filter(), so that it
now matches do_setopt_accept_filter().  Slightly reformulate the logic to
match the optimistic allocation of storage for the argument in advance,
and slightly expand the coverage of the socket lock.

19 years agoFix a terrible braino in last commit. Put kern.debug back to /var/log/messages
glebius [Sat, 12 Mar 2005 12:31:16 +0000 (12:31 +0000)]
Fix a terrible braino in last commit. Put kern.debug back to /var/log/messages
and do exactly what last commit message described.

19 years agoPart two of post-SMPng cleanup of accept filter registration: perform all
rwatson [Sat, 12 Mar 2005 12:27:47 +0000 (12:27 +0000)]
Part two of post-SMPng cleanup of accept filter registration: perform all
allocation up front before grabbing the socket mutex and doing the
registration work.  The result is a lot cleaner.

19 years ago- Fix device names for snd_vibes(4) and snd_via82c686(4).
simon [Sat, 12 Mar 2005 10:54:08 +0000 (10:54 +0000)]
- Fix device names for snd_vibes(4) and snd_via82c686(4).
- Bump copyright year.

MFC after: 3 days

19 years agoDon't build the nve on pc98.
nyan [Sat, 12 Mar 2005 10:41:58 +0000 (10:41 +0000)]
Don't build the nve on pc98.

19 years agoMFi386: revision 1.217.
nyan [Sat, 12 Mar 2005 10:30:06 +0000 (10:30 +0000)]
MFi386: revision 1.217.

19 years agoDue to a CVS misfire, I ended up committing the wrong version of this.
obrien [Sat, 12 Mar 2005 08:02:06 +0000 (08:02 +0000)]
Due to a CVS misfire, I ended up committing the wrong version of this.

19 years agoMFCi386: Prevent integer underflow that could result in all memory being
scottl [Sat, 12 Mar 2005 07:05:59 +0000 (07:05 +0000)]
MFCi386: Prevent integer underflow that could result in all memory being
consumed.

19 years agoGuard against an integer underflow that could cause busdma to eat up all
scottl [Sat, 12 Mar 2005 07:01:53 +0000 (07:01 +0000)]
Guard against an integer underflow that could cause busdma to eat up all
available RAM.  This also results in the global bounce page limit being
applied to zones instead of globally.

Submitted by: Petr Lampa (in part)

19 years agoAdding new device ID for BCM5751M support.
avatar [Sat, 12 Mar 2005 06:51:25 +0000 (06:51 +0000)]
Adding new device ID for BCM5751M support.

Submitted by: FUJITA Kazutoshi <fujita at soum dot co dot jp>
Reviewed by: sam (mentor)

19 years agoFix bin/pax doesn't wait for compress/gzip before exiting.
obrien [Sat, 12 Mar 2005 06:38:01 +0000 (06:38 +0000)]
Fix bin/pax doesn't wait for compress/gzip before exiting.

Submitted by: Joerg Sonnenberger <joerg@britannica.bec.de>
Obtained from: OpenBSD (commited fix to PR 3455)

19 years agoUsed unsigned version.
obrien [Sat, 12 Mar 2005 06:06:04 +0000 (06:06 +0000)]
Used unsigned version.

Submitted by: jmallett

19 years agoFix kernel build on 64-bit machines.
obrien [Sat, 12 Mar 2005 03:50:39 +0000 (03:50 +0000)]
Fix kernel build on 64-bit machines.

19 years agoFix a long-standing bug in alpha's implementation of busdma. We need to
mux [Sat, 12 Mar 2005 02:43:50 +0000 (02:43 +0000)]
Fix a long-standing bug in alpha's implementation of busdma.  We need to
OR the physical address with alpha_XXX_dmamap_or to get the DMA address,
like the name of the variable suggests.  However, while we were doing
this correctly in the alpha_XXX_dmamap() macro, the busdma code added
the variable to the physical address instead of or'ing it.  Fortunately
and if my math is not entirely wrong, you would need more than 128GB of
RAM and a device able to do DMA in 64bits to experience the bug.

Spotted by: cognet

19 years agoFreeBSD consumer bits of the nForce MCP NIC binary blob.
obrien [Sat, 12 Mar 2005 00:29:30 +0000 (00:29 +0000)]
FreeBSD consumer bits of the nForce MCP NIC binary blob.

Demanded by: DES
Encouraged by: scottl
Obtained from: q@onthenet.com.au (partially)
KNF'ed by: obrien

19 years agoThis commit was generated by cvs2svn to compensate for changes in r143439,
obrien [Sat, 12 Mar 2005 00:17:37 +0000 (00:17 +0000)]
This commit was generated by cvs2svn to compensate for changes in r143439,
which included commits to RCS files with non-trunk default branches.

19 years agoImport the nForce MCP NIC bits. This is version 1.0-0301.
obrien [Sat, 12 Mar 2005 00:17:37 +0000 (00:17 +0000)]
Import the nForce MCP NIC bits.  This is version 1.0-0301.

Demanded by: DES
Encouraged by: scottl

19 years agoReplace my previous change for 32 bit systems with hz > 169 with Bruce's
peter [Sat, 12 Mar 2005 00:13:45 +0000 (00:13 +0000)]
Replace my previous change for 32 bit systems with hz > 169 with Bruce's
simpler one.

19 years agoMake the tty vmin/vtime timeouts work for hz > 169 on 32 bit machines.
peter [Sat, 12 Mar 2005 00:10:23 +0000 (00:10 +0000)]
Make the tty vmin/vtime timeouts work for hz > 169 on 32 bit machines.

19 years agoCorrect a last-minute thinko. Instead of copying the nul with the string,
njl [Fri, 11 Mar 2005 23:35:23 +0000 (23:35 +0000)]
Correct a last-minute thinko.  Instead of copying the nul with the string,
nul-terminate the dp->d_name directly and only copy the string.

19 years agoThe mbnambuf routines combine multiple substrings into a single
njl [Fri, 11 Mar 2005 23:27:45 +0000 (23:27 +0000)]
The mbnambuf routines combine multiple substrings into a single
long filename.  Each substring is indexed by the windows ID, a
sequential one-based value.  The previous code was extremely slow,
doing a malloc/strcpy/free for each substring.

This code optimizes these routines with this in mind, using the ID
to index into a single array and concatenating each WIN_CHARS chunk
at once.  (The last chunk is variable-length.)

This code has been tested as working on an FS with difficult filename
sizes (255, 13, 26, etc.)  It gives a 77.1% decrease in profiled
time (total across all functions) and a 73.7% decrease in wall time.
Test was "ls -laR > /dev/null".

Per-function time savings:
mbnambuf_init:  -90.7%
mbnambuf_write: -18.7%
mbnambuf_flush: -67.1%

MFC after: 1 month

19 years agoRemove diffs to i386 version that came in via the compiler support ifdefs.
peter [Fri, 11 Mar 2005 22:16:09 +0000 (22:16 +0000)]
Remove diffs to i386 version that came in via the compiler support ifdefs.
This changes things like whitespace, inconsistent use of #ifndef vs
#if !defined(), different macro argument orders, mismatched comments, etc.

19 years agoMFi386: reduce apic clock interrupt rate
peter [Fri, 11 Mar 2005 22:12:38 +0000 (22:12 +0000)]
MFi386: reduce apic clock interrupt rate

19 years agoWhitespace sync with amd64. (Rather than re-add the extra blank lines
peter [Fri, 11 Mar 2005 22:10:25 +0000 (22:10 +0000)]
Whitespace sync with amd64.  (Rather than re-add the extra blank lines
on amd64, I'm removing them here)

19 years agogbde(8) is also rejndael user.
ume [Fri, 11 Mar 2005 22:07:04 +0000 (22:07 +0000)]
gbde(8) is also rejndael user.

Reported by: phk

19 years agoFix a mismerge of i386 rev 1.209
peter [Fri, 11 Mar 2005 21:57:38 +0000 (21:57 +0000)]
Fix a mismerge of i386 rev 1.209

19 years agoMatch i386 rev 1.38 with __cplusplus support
peter [Fri, 11 Mar 2005 21:46:01 +0000 (21:46 +0000)]
Match i386 rev 1.38 with __cplusplus support

19 years agoRemove an OBE set of comments, fix a minor whitespace nit while here.
peter [Fri, 11 Mar 2005 21:42:11 +0000 (21:42 +0000)]
Remove an OBE set of comments, fix a minor whitespace nit while here.

19 years agoFirst step in simplifying accept filter socket option logic in the
rwatson [Fri, 11 Mar 2005 21:37:45 +0000 (21:37 +0000)]
First step in simplifying accept filter socket option logic in the
post-SMPng world order.  Centralize handling of the socket option
clear case in do_setopt_accept_filter().

19 years agoreorder ath_rate_onoe to after ath_rate_sample so it gets used as the
sam [Fri, 11 Mar 2005 19:40:34 +0000 (19:40 +0000)]
reorder ath_rate_onoe to after ath_rate_sample so it gets used as the
default rate control algorithm; this should be done differently but for
now use this simple solution

19 years agoRemove an additional commented out reference to a possible future sx
rwatson [Fri, 11 Mar 2005 19:16:02 +0000 (19:16 +0000)]
Remove an additional commented out reference to a possible future sx
lock.

19 years agojust use crypto/rijndael
ume [Fri, 11 Mar 2005 17:58:24 +0000 (17:58 +0000)]
just use crypto/rijndael
(I forgot to commit this in my previous commit)

19 years agojust use crypto/rijndael, and nuke opencrypto/rindael.[ch].
ume [Fri, 11 Mar 2005 17:24:46 +0000 (17:24 +0000)]
just use crypto/rijndael, and nuke opencrypto/rindael.[ch].
the two became almost identical since latest KAME merge.

Discussed with: sam

19 years agoWhen setting up a socket in socreate(), there's no need to lock the
rwatson [Fri, 11 Mar 2005 16:30:02 +0000 (16:30 +0000)]
When setting up a socket in socreate(), there's no need to lock the
socket lock around knlist_init(), so don't.

Hard code the setting of the socket reference count to 1 rather than
using soref() to avoid asserting the socket lock, since we've not yet
exposed the socket to other threads.

This removes two mutex operations from each socket allocation.

19 years agoRemove suggestive sx_init() comment in soalloc(). We will have something
rwatson [Fri, 11 Mar 2005 16:26:33 +0000 (16:26 +0000)]
Remove suggestive sx_init() comment in soalloc().  We will have something
like this at some point, but for now it clutters the source.

19 years agointegrate rijndael-alg-fst.h into rijndael.h.
ume [Fri, 11 Mar 2005 16:26:10 +0000 (16:26 +0000)]
integrate rijndael-alg-fst.h into rijndael.h.

19 years agoAdding missing module dependency. This should fix the undefined symbol
avatar [Fri, 11 Mar 2005 16:15:43 +0000 (16:15 +0000)]
Adding missing module dependency. This should fix the undefined symbol
error(ath_hal_computetxtime) during module loading.

Reviewed by: sam (mentor)

19 years agostop including rijndael-api-fst.h from rijndael.h.
ume [Fri, 11 Mar 2005 15:42:51 +0000 (15:42 +0000)]
stop including rijndael-api-fst.h from rijndael.h.
this is required to integrate opencrypto into crypto.

19 years agoThe SO_NOSIGPIPE socket option allows a user process to mark a socket
rwatson [Fri, 11 Mar 2005 15:06:16 +0000 (15:06 +0000)]
The SO_NOSIGPIPE socket option allows a user process to mark a socket
so that the socket does not generate SIGPIPE, only EPIPE, when a write
is attempted after socket shutdown.  When the option was introduced in
2002, this required the logic for determining whether SIGPIPE was
generated to be pushed down from dofilewrite() to the socket layer so
that the socket options could be considered.  However, the change in
2002 omitted modification to soo_write() required to add that logic,
resulting in SIGPIPE not being generated even without SO_NOSIGPIPE when
the socket was written to using write() or related generic system calls.

This change adds the EPIPE logic to soo_write(), generating a SIGPIPE
signal to the process associated with the passed uio in the event that
the SO_NOSIGPIPE option is not set.

Notes:

- The are upsides and downsides to placing this logic in the socket
  layer as opposed to the file descriptor layer.  This is really fd
  layer logic, but because we need so_options, we have a choice of
  layering violations and pick this one.

- SIGPIPE possibly should be delivered to the thread performing the
  write, not the process performing the write.

- uio->uio_td and the td argument to soo_write() might potentially
  differ; we use the thread in the uio argument.

- The "sigpipe" regression test in src/tools/regression/sockets/sigpipe
  tests for the bug.

Submitted by: Mikko Tyolajarvi <mbsd at pacbell dot net>
Talked with: glebius, alfred
PR: 78478
MFC after: 1 week

19 years agoFix typos in comments.
stefanf [Fri, 11 Mar 2005 14:20:09 +0000 (14:20 +0000)]
Fix typos in comments.

19 years agoUse socklen_t where appropriate.
stefanf [Fri, 11 Mar 2005 14:17:12 +0000 (14:17 +0000)]
Use socklen_t where appropriate.

19 years agoSimplify the print routines by using LST_FOREACH instead of Lst_ForEach
harti [Fri, 11 Mar 2005 13:24:08 +0000 (13:24 +0000)]
Simplify the print routines by using LST_FOREACH instead of Lst_ForEach
and inlining the small printing utility functions.
Create a function that can be used to produce printable representations
of flag words.

19 years agoCheck the return value of shutdown().
rwatson [Fri, 11 Mar 2005 13:05:18 +0000 (13:05 +0000)]
Check the return value of shutdown().

19 years agoRemove leading underscores from the pathname defines. All identifiers
harti [Fri, 11 Mar 2005 13:02:38 +0000 (13:02 +0000)]
Remove leading underscores from the pathname defines. All identifiers
with leading underscore followed by an uppercase letter are in the
implementation namespace.

19 years agoRemove the leading underscore from structure tags. All identifiers
harti [Fri, 11 Mar 2005 12:57:25 +0000 (12:57 +0000)]
Remove the leading underscore from structure tags. All identifiers
with a leading underscore are in the implementation namespace.

19 years agosys/crypto/md5.[ch] is used from nowhere. So, just nuke them.
ume [Fri, 11 Mar 2005 12:56:15 +0000 (12:56 +0000)]
sys/crypto/md5.[ch] is used from nowhere.  So, just nuke them.

19 years agoAdd sigpipe, a simple UNIX domain socket and TCP regression test that is
rwatson [Fri, 11 Mar 2005 12:47:14 +0000 (12:47 +0000)]
Add sigpipe, a simple UNIX domain socket and TCP regression test that is
intended to verify that SIGPIPE is delivered to a process writing or
sending on a socket that has been shut down for write.  If available,
SO_NOSIGPIPE is also tested.

This regression test is currently passed by RELENG_4, but not by HEAD or
RELENG_5, due to a bug in the write() code for sockets.  SO_NOSIGPIPE is not
present in RELENG_4, however, so is not tested there.

Reported by: Mikko Tyolajarvi <mbsd at pacbell dot net>
PR: 78478

19 years ago- use 1/2 space for rijndael context in ipsec
ume [Fri, 11 Mar 2005 12:45:09 +0000 (12:45 +0000)]
- use 1/2 space for rijndael context in ipsec
- rijndael_set_key() always sets up full context
- rijndaelKeySetupDec() gets back original protoype

Reviewed by: sam
Obtained from: OpenBSD

19 years agoStyle: Fix indentation.
harti [Fri, 11 Mar 2005 12:40:55 +0000 (12:40 +0000)]
Style: Fix indentation.

19 years agorefer opencrypto/cast.h directly.
ume [Fri, 11 Mar 2005 12:37:07 +0000 (12:37 +0000)]
refer opencrypto/cast.h directly.

19 years agoCall ParseFinishLine() also for the last line in a file. This
harti [Fri, 11 Mar 2005 11:29:39 +0000 (11:29 +0000)]
Call ParseFinishLine() also for the last line in a file. This
patch differs from the previous one in that it calls the function
only when a real file hits EOF. The bodies of .for loops are also
handled as files, but for these we don't want to end a dependency block
on the 'EOF' as in:

foo:
do-this
.for ...
do-something
.endfor
do-more

19 years agoFix getting stats from many links with index > 0.
glebius [Fri, 11 Mar 2005 10:29:38 +0000 (10:29 +0000)]
Fix getting stats from many links with index > 0.

Submitted by: Richard Kojedzinszky
MFC after: 3 days

19 years agoMove common credential save and restore code into a separate file.
dds [Fri, 11 Mar 2005 08:39:58 +0000 (08:39 +0000)]
Move common credential save and restore code into a separate file.
Improve credential handling in pt_file.c

19 years agoBack out the last commit. It turns out that this breaks more than
harti [Fri, 11 Mar 2005 07:53:32 +0000 (07:53 +0000)]
Back out the last commit. It turns out that this breaks more than
it fixes. This should fix the buildworld breakage.

19 years agoEnsure the configuration file is given with an absolute file path.
dds [Fri, 11 Mar 2005 07:50:09 +0000 (07:50 +0000)]
Ensure the configuration file is given with an absolute file path.

19 years agoAdd examples for tcplisten, data scattering, and file views.
dds [Fri, 11 Mar 2005 07:40:17 +0000 (07:40 +0000)]
Add examples for tcplisten, data scattering, and file views.
Improve formatting of existing examples.

19 years agoMake the pps interrupt register as MPSAFE and FAST. Use a spin lock
imp [Fri, 11 Mar 2005 07:03:46 +0000 (07:03 +0000)]
Make the pps interrupt register as MPSAFE and FAST.  Use a spin lock
to syncrhonize access to the data as a result.  This makes the pps
less likely to miss the 1ms pulse that I'm feeding it, but not
entirely reliable yet on my 133MHz P5.

Reviewed by: phk

19 years agoFix minor grammar and punctuation nits in a few recently-added
bmah [Fri, 11 Mar 2005 05:37:18 +0000 (05:37 +0000)]
Fix minor grammar and punctuation nits in a few recently-added
entries.

19 years agoDoh! silly typo precludes compiling
imp [Fri, 11 Mar 2005 05:30:59 +0000 (05:30 +0000)]
Doh!  silly typo precludes compiling

19 years agoRevert changes of 1.49. Lots-a-people broke with it, for reasons
imp [Fri, 11 Mar 2005 05:27:05 +0000 (05:27 +0000)]
Revert changes of 1.49.  Lots-a-people broke with it, for reasons
unknown (since my sony vaio didn't :-(.

Instead, fix the problem described by 1.49 in a different way: just
add the two calls I'd hoped I'd avoid in 1.49 by doing the (wrong)
gymnastics there.  While 1.49 is a good direction to go in, each step
of the way should work :-(.

19 years agoSampleRate rate control algorithm for the ath driver
sam [Fri, 11 Mar 2005 01:39:57 +0000 (01:39 +0000)]
SampleRate rate control algorithm for the ath driver

Submitted by: John Bicket