16 years ago- WARNS 6
Liam J. Foy [Fri, 25 Mar 2005 12:56:48 +0000 (12:56 +0000)]

Submitted by: H}kon Schad Bergsaker <hakonsb@student.matnat.uio.no>

16 years ago${CC}'s here were supposed be used as preprocessor, add -E.
YONETANI Tomokazu [Fri, 25 Mar 2005 11:42:36 +0000 (11:42 +0000)]
${CC}'s here were supposed be used as preprocessor, add -E.

Reported-by: Martin P. Hellwig <mhellwig (a) xs4all.nl>
16 years agoSync with the kernel to make ncptrace work again. CINV_PARENT no longer
Matthew Dillon [Thu, 24 Mar 2005 20:15:11 +0000 (20:15 +0000)]
Sync with the kernel to make ncptrace work again.  CINV_PARENT no longer

16 years agoDisplay additional information about a vnode's object.
Matthew Dillon [Thu, 24 Mar 2005 20:14:42 +0000 (20:14 +0000)]
Display additional information about a vnode's object.

16 years agoFix a server-side TCP NFS mount stream sequencing problem which could result
Matthew Dillon [Thu, 24 Mar 2005 19:58:19 +0000 (19:58 +0000)]
Fix a server-side TCP NFS mount stream sequencing problem which could result
in portions of the TCP stream being parsed out of order, corrupting it and
causing the NFS server to drop the connection.

The TCP protocol stack upcalls into nfsrv_rcv().  The various nfsd threads
also call nfsrv_rcv().  A short blocking condition in the parsing code plus
the blocking lock on the sockbuf (when called via an nfsd) could result in
data from the socket being incorporated into the nfs socket's mbuf chain
out of order.

Reported-by: Andreas Hauser <andy@splashground.de>
16 years agoLicense typo.
Joerg Sonnenberger [Thu, 24 Mar 2005 17:24:04 +0000 (17:24 +0000)]
License typo.

16 years agoAdjust man page sections.
Sascha Wildner [Thu, 24 Mar 2005 12:48:04 +0000 (12:48 +0000)]
Adjust man page sections.

16 years agotcb is now managed by rtld, caching it is not correct,
David Xu [Thu, 24 Mar 2005 12:38:39 +0000 (12:38 +0000)]
tcb is now managed by rtld, caching it is not correct,
otherwise we may get stale copy of tls data from previous
dead thread, caching should be done in rtld if needed.

16 years agoHave the server complain if the NFS rpc stream from a client gets corrupted,
Matthew Dillon [Thu, 24 Mar 2005 06:44:27 +0000 (06:44 +0000)]
Have the server complain if the NFS rpc stream from a client gets corrupted,
so we at least know why the tcp link is reconnecting.  The client already
complains if the NFS rpc stream from a server gets corrupted).

16 years agoImport man pages for pf(4), pflog(4) and pfsync(4) devices.
Simon Schubert [Thu, 24 Mar 2005 01:15:59 +0000 (01:15 +0000)]
Import man pages for pf(4), pflog(4) and pfsync(4) devices.

These are not the latest versions, but those that seem to suit
best with our version of pf.

Reminded-By: "Alexander 'alxl' Lobachov" <alxl@alxl.info>
Obtained-From: OpenBSD

16 years agoUndo last commit. This was not my hisenbug. :-P
Max Okumoto [Wed, 23 Mar 2005 20:58:35 +0000 (20:58 +0000)]
Undo last commit.  This was not my hisenbug. :-P

16 years agoRemove an assertion in bundirty() that requires the buffer to not be on
Matthew Dillon [Wed, 23 Mar 2005 20:37:03 +0000 (20:37 +0000)]
Remove an assertion in bundirty() that requires the buffer to not be on
a queue.  There is a code path in brelse() where the buffer may be put on
a queue prior to calling bundirty().

Reported-by: David Rhodus <sdrhodus@gmail.com>
16 years agoAdditional note to last commit. GCC-3.4 improperly generates a warning
Matthew Dillon [Wed, 23 Mar 2005 19:09:33 +0000 (19:09 +0000)]
Additional note to last commit.  GCC-3.4 improperly generates a warning
and an int $5 for unpromoted conversions (somebody really scewed up in GCC
land if they thought that that was a reasonable solution!)... It really
should generate a fatal error for this case since the traditional varadic
macros were able to handle the case.

16 years agoUse the correct type in va_arg call, char is promoted to int before calling
Joerg Sonnenberger [Wed, 23 Mar 2005 17:22:21 +0000 (17:22 +0000)]
Use the correct type in va_arg call, char is promoted to int before calling
and has to be fetched as such.

16 years agoPass exact number of threads to thr_umtx_wake.
David Xu [Wed, 23 Mar 2005 11:21:02 +0000 (11:21 +0000)]
Pass exact number of threads to thr_umtx_wake.

16 years agoEliminate a redundant variable assignment.
Jeffrey Hsu [Wed, 23 Mar 2005 08:02:46 +0000 (08:02 +0000)]
Eliminate a redundant variable assignment.

Noticed by: Noritoshi Demizu <demizu@dd.iij4u.or.jp>

16 years agoWe can only do upper-layer protocol length checks on the first fragment.
Jeffrey Hsu [Wed, 23 Mar 2005 07:47:40 +0000 (07:47 +0000)]
We can only do upper-layer protocol length checks on the first fragment.

16 years agoCorrect the th_off check against ip_len. The check in ip_demux occurs
Matthew Dillon [Wed, 23 Mar 2005 07:27:56 +0000 (07:27 +0000)]
Correct the th_off check against ip_len.  The check in ip_demux occurs
before ip_len is adjusted so we have to add iphlen into the equation.  We
forgot to do this when the code was originally moved from tcp_input to
ip_demux (in tcp_input the check occurs after ip_len is adjusted).

This fixes a panic assertion in tcp_input when a mangled packet is received.

Reported-by: Joe Talbott <josepht@cstone.net>
16 years agoFixed reference of memory after free.
Max Okumoto [Wed, 23 Mar 2005 05:02:37 +0000 (05:02 +0000)]
Fixed reference of memory after free.

16 years agoBecause destroy_all_dev() checks the mask/match against the device's si_udev,
Matthew Dillon [Wed, 23 Mar 2005 02:50:53 +0000 (02:50 +0000)]
Because destroy_all_dev() checks the mask/match against the device's si_udev,
which is a field combining both major and minor numbers, we must mask off the
major bits (to retain only the minor bits) from the 'mask' variable in order
to allow a generic -1 to be passed as the mask.  Otherwise we will not match

Revamp a good chunk of the documentation to try to make the major/minor
number masking issues clear.

Bug-found-by: Chuck Tuffli <chuck_tuffli@agilent.com>
16 years agoAllow an info.size field of -1 to indicate that the descriptor should map
Matthew Dillon [Wed, 23 Mar 2005 01:13:20 +0000 (01:13 +0000)]
Allow an info.size field of -1 to indicate that the descriptor should map
the entire address space, requires for direct-tls %gs:OFFSET support (where
the OFFSET is negative).  Otherwise the descriptor only maps the specified
range (typically just the TCB area created by the thread library) and %gs:0
must be used to obtain a pointer from which negative relative memory accesses
can be made.

16 years agoReverse the order of dragonfly and FreeBSD since it says right after
Matthew Dillon [Wed, 23 Mar 2005 00:25:18 +0000 (00:25 +0000)]
Reverse the order of dragonfly and FreeBSD since it says right after
it is in alphabetical order.

Submitted-by: Craig Dooley <xlnxminusx@gmail.com>
16 years agoDon't activate -fstrict-aliasing by default, not even with -O2+.
Joerg Sonnenberger [Tue, 22 Mar 2005 23:58:32 +0000 (23:58 +0000)]
Don't activate -fstrict-aliasing by default, not even with -O2+.

16 years agoFix handling of deallocation of dynamic TLS, the previous code could
Joerg Sonnenberger [Tue, 22 Mar 2005 23:54:18 +0000 (23:54 +0000)]
Fix handling of deallocation of dynamic TLS, the previous code could
free random memory after a DSO with TLS segment was unloaded.

16 years agoDon't restart a timeout wait, one can periodically send signal to
David Xu [Tue, 22 Mar 2005 23:42:53 +0000 (23:42 +0000)]
Don't restart a timeout wait, one can periodically send signal to
thread and interval is less than the timeout, then the thread will
never be timeout.

16 years agoUse rtld's TLS interface to allocate tcb.
David Xu [Tue, 22 Mar 2005 23:07:24 +0000 (23:07 +0000)]
Use rtld's TLS interface to allocate tcb.

16 years agoAdd support for TLS.
David Xu [Tue, 22 Mar 2005 22:56:36 +0000 (22:56 +0000)]
Add support for TLS.

Obtained from : FreeBSD

16 years agoStart working on the full-duplex journaling feature, where the target can
Matthew Dillon [Tue, 22 Mar 2005 22:13:33 +0000 (22:13 +0000)]
Start working on the full-duplex journaling feature, where the target can
acknowledge the sequence space to prevent information loss if a journaling
stream is interrupted.  Implement a skeleton for the receiver thread.

Delete journals associated with a mount point that is undergoing an unmount.
(reported-by: Fabian <fabian.duelli@bluewin.ch>)

16 years agoDo not allow the journaling descriptor to be a regular file on the same
Matthew Dillon [Tue, 22 Mar 2005 21:42:39 +0000 (21:42 +0000)]
Do not allow the journaling descriptor to be a regular file on the same
filesystem that is being journaled.  This doesn't catch loops, but it
should catch the most common foot shooting case.

Reported-by: Fabian <fabian.duelli@bluewin.ch>
16 years agoFix a minor bug in pass*) generation. 'units' was not properly optioned.
Matthew Dillon [Tue, 22 Mar 2005 18:36:50 +0000 (18:36 +0000)]
Fix a minor bug in pass*) generation.  'units' was not properly optioned.

Submitted-by: Joerg Anslik <joerg@anslik.de>
16 years agoFixed indent for structure
Max Okumoto [Tue, 22 Mar 2005 11:29:49 +0000 (11:29 +0000)]
Fixed indent for structure

FreeBSD-Date: 2005/03/22 08:00:02
Author: harti

16 years agoRewrite a good chunk of MAKEDEV, commonizing paths, cleaning up device
Matthew Dillon [Tue, 22 Mar 2005 00:40:54 +0000 (00:40 +0000)]
Rewrite a good chunk of MAKEDEV, commonizing paths, cleaning up device
generation, and adding support for a /etc/devices.conf file to allow
custom owner/group/perms to be set.

Submitted-by: Joerg Anslik <joerg@anslik.de>
16 years agoInclude sys/types.h to make it self-contained.
Joerg Sonnenberger [Mon, 21 Mar 2005 23:08:55 +0000 (23:08 +0000)]
Include sys/types.h to make it self-contained.

16 years agoPick up changes in rev 1.8 of src/sys/dev/ic/mpt_netbsd.c from NetBSD.
Matthew Dillon [Mon, 21 Mar 2005 22:08:41 +0000 (22:08 +0000)]
Pick up changes in rev 1.8 of src/sys/dev/ic/mpt_netbsd.c from NetBSD.
Set the DMA SGL length correctly if the DMA request must be chained because
it is too large to fit in one SGL.

This should fix this driver for some Dell Precision systems.

Obtained-from: FreeBSD
Originally-from: NetBSD

16 years agoFix a crash in the firewire / DCONS code that occurs when the crom data
Matthew Dillon [Mon, 21 Mar 2005 22:07:24 +0000 (22:07 +0000)]
Fix a crash in the firewire / DCONS code that occurs when the crom data
block is empty.  This case isn't supposed to happen but apparently
it can happen when two machines are connected together via a firewire
cable and one machine is rebooting.  Either the data block is cleared before
the reboot occurs, or the other machines manages to poll the first one
while it is booting and the data block is empty.

The actual problem causing the empty data block has not been tracked down.

16 years ago- Remove unused variable
Liam J. Foy [Mon, 21 Mar 2005 19:26:14 +0000 (19:26 +0000)]
- Remove unused variable

Good old: Samuel J. Greear <sjg@evilcode.net>

16 years agoCheck for the signal handlers another time, they might not disrupt
Joerg Sonnenberger [Mon, 21 Mar 2005 19:20:57 +0000 (19:20 +0000)]
Check for the signal handlers another time, they might not disrupt
the recvmsg.

16 years agoDefer work from the signal handlers into the main loop.
Joerg Sonnenberger [Mon, 21 Mar 2005 19:18:51 +0000 (19:18 +0000)]
Defer work from the signal handlers into the main loop.

Submitted-by: Samuel J. Greear <sjg@evilcode.net>
16 years agoNo more Alpha support for RTLD.
Joerg Sonnenberger [Mon, 21 Mar 2005 18:25:22 +0000 (18:25 +0000)]
No more Alpha support for RTLD.

16 years agoFix reboot -k, it didn't truncate /boot/nextboot.conf.
Joerg Sonnenberger [Mon, 21 Mar 2005 17:57:54 +0000 (17:57 +0000)]
Fix reboot -k, it didn't truncate /boot/nextboot.conf.

Submitted-by: Gary Allan <dragonfly@gallan.plus.com>
16 years agoint size --> size_t size
Joerg Sonnenberger [Mon, 21 Mar 2005 17:11:38 +0000 (17:11 +0000)]
int size --> size_t size

16 years agoAdjust userland prototypes as well.
Joerg Sonnenberger [Mon, 21 Mar 2005 17:09:52 +0000 (17:09 +0000)]
Adjust userland prototypes as well.

16 years ago- sysctl(3), sethostname(3) and malloc(3) set errno, use it!
Liam J. Foy [Mon, 21 Mar 2005 16:59:31 +0000 (16:59 +0000)]
- sysctl(3), sethostname(3) and malloc(3) set errno, use it!
- Don't bother duplicating strings, just use optarg
- Use herror(3) when reporting an error for gethostbyaddr(3)
- Remove unnecessary header

16 years agoregen
Joerg Sonnenberger [Mon, 21 Mar 2005 16:47:49 +0000 (16:47 +0000)]

16 years agoChange prototype of sys_set_tls_area and sys_get_tls_area to take
Joerg Sonnenberger [Mon, 21 Mar 2005 16:45:08 +0000 (16:45 +0000)]
Change prototype of sys_set_tls_area and sys_get_tls_area to take
the size argument as size_t.

16 years agoAdd Windoz autorun support to the CD to automatically pop up a browser with a
Matthew Dillon [Mon, 21 Mar 2005 05:46:40 +0000 (05:46 +0000)]
Add Windoz autorun support to the CD to automatically pop up a browser with a
nice DragonFly information page and hotlinks to various DFly resources.

Submitted-by: Eduardo Tongson <propolice@gmail.com>
16 years agoPunctuation, formatting, grammar, and spelling nits.
Chris Pressey [Mon, 21 Mar 2005 04:52:34 +0000 (04:52 +0000)]
Punctuation, formatting, grammar, and spelling nits.

16 years agoUpdate the information regarding the installer and the pfi
Chris Pressey [Mon, 21 Mar 2005 04:07:13 +0000 (04:07 +0000)]
Update the information regarding the installer and the pfi
boot-configuration system.

16 years agoUpdate installer to version 1.1.5. Highlights include:
Chris Pressey [Mon, 21 Mar 2005 03:58:04 +0000 (03:58 +0000)]
Update installer to version 1.1.5.  Highlights include:

- ESC key can be used to return to previous form or menu.
- Better error dialog when network can't be configured as specified.
- Fixed a case where "Configure an Installed System" could fail due
  to the device node not existing (added an appropriate MAKEDEV)
- A friendly 'splash screen' (index.html) installed for CGI frontend.

16 years agoWhen doing Limited Transmit, don't retract snd_nxt if it was previously
Jeffrey Hsu [Mon, 21 Mar 2005 01:56:22 +0000 (01:56 +0000)]
When doing Limited Transmit, don't retract snd_nxt if it was previously
the same as snd_max.

Submitted by: Noritoshi Demizu <demizu@dd.iij4u.or.jp>

16 years agoEnsure that Limited Transmit always sends new data, even after a
Jeffrey Hsu [Mon, 21 Mar 2005 01:49:43 +0000 (01:49 +0000)]
Ensure that Limited Transmit always sends new data, even after a
retransmission timeout.

Reported by: Noritoshi Demizu <demizu@dd.iij4u.or.jp>

16 years agoOverride _kevent, not kevent. This should fix the DNS issue, since
Joerg Sonnenberger [Sun, 20 Mar 2005 18:45:15 +0000 (18:45 +0000)]
Override _kevent, not kevent. This should fix the DNS issue, since
the libc resolver was calling _kevent directly, which prevented
libc_r from switching to another thread.

Suggested-by: Jonathan Lemon
Problem-reports: Brock Johnson, Laurent Sartran and others
Special-thanks-to: Laurent Sartran for tracking down the timeframe

16 years agoAdd kqueue overwrite for libc_r. We have to trace the opened descriptor.
Joerg Sonnenberger [Sun, 20 Mar 2005 17:46:31 +0000 (17:46 +0000)]
Add kqueue overwrite for libc_r. We have to trace the opened descriptor.

Obtained-from: FreeBSD

16 years agoMinor style changes.
Joerg Sonnenberger [Sun, 20 Mar 2005 14:57:28 +0000 (14:57 +0000)]
Minor style changes.

16 years ago- Style(9)
Liam J. Foy [Sun, 20 Mar 2005 14:07:43 +0000 (14:07 +0000)]
- Style(9)
- Bump WARNS to 6

Submitted by: Sepherosa Ziehau <sepherosa@softhome.net>

16 years agopatch-7.163
Max Okumoto [Sun, 20 Mar 2005 12:25:40 +0000 (12:25 +0000)]
 Use VarParse() instead of Var_Parse() wapper within var.c

16 years agopatch-7.162
Max Okumoto [Sun, 20 Mar 2005 12:25:19 +0000 (12:25 +0000)]
  Use the same names for variables passed to Var_Parse().

16 years agopatch-7.161
Max Okumoto [Sun, 20 Mar 2005 12:24:40 +0000 (12:24 +0000)]
 Use same code structure for code using VarFind().

16 years agotch-7.160
Max Okumoto [Sun, 20 Mar 2005 12:24:13 +0000 (12:24 +0000)]
 Removed some extra parans.

16 years agopatch-7.158
Max Okumoto [Sun, 20 Mar 2005 12:23:53 +0000 (12:23 +0000)]
- Change VarGetPattern() to return a Buffer object.
- Change struct VarPattern to store Buffer objects instead of
  a 'char *' and len.

16 years agopatch-7.157
Max Okumoto [Sun, 20 Mar 2005 12:23:25 +0000 (12:23 +0000)]
  Use lhs field in VarPattern object instead of variable re.

16 years agopatch-7.156
Max Okumoto [Sun, 20 Mar 2005 12:22:46 +0000 (12:22 +0000)]
  Merge struct VarPattern and struct VarREPattern.  This will help
  factor out more common code.

16 years agoAdd the pfi_curses_escdelay variable. When using the curses frontend,
Chris Pressey [Sun, 20 Mar 2005 03:54:01 +0000 (03:54 +0000)]
Add the pfi_curses_escdelay variable.  When using the curses frontend,
this variable determines the amount of time, in milliseconds, that
must pass after ESC is pressed, in order for it to be recognized as a
plain ESC keystroke and not as the start of an escape code.  See the
ESCDELAY entry in the ncurses manual page for more information.

16 years agoClean up Var_SetEnv().
Max Okumoto [Sun, 20 Mar 2005 00:44:45 +0000 (00:44 +0000)]
Clean up Var_SetEnv().

16 years agoRemove unused version.
YONETANI Tomokazu [Sat, 19 Mar 2005 19:49:05 +0000 (19:49 +0000)]
Remove unused version.

16 years agoUpdate ACPI build wrappers to use new ACPICA(20050309) code.
YONETANI Tomokazu [Sat, 19 Mar 2005 19:38:31 +0000 (19:38 +0000)]
Update ACPI build wrappers to use new ACPICA(20050309) code.
The previously imported version(20050211) still had
AE_AML_BUFFER_LIMIT issue, which was fixed in the release
made just after the import.

16 years agoImport acpica-20050309 from Intel.
YONETANI Tomokazu [Sat, 19 Mar 2005 19:24:31 +0000 (19:24 +0000)]
Import acpica-20050309 from Intel.


16 years agoFix order for SI_SUB_TUNABLES.
Sascha Wildner [Sat, 19 Mar 2005 18:57:34 +0000 (18:57 +0000)]
Fix order for SI_SUB_TUNABLES.

16 years ago- Make sure we call exit() after clnt_pcreateerror()
Liam J. Foy [Sat, 19 Mar 2005 17:43:18 +0000 (17:43 +0000)]
- Make sure we call exit() after clnt_pcreateerror()
- Better parsing of passed options
- Fix (xdrproc_t) warnings
- Bump WARNS 6

16 years agoAdd /etc/firmware, remove /etc/gnats.
Joerg Sonnenberger [Sat, 19 Mar 2005 14:56:08 +0000 (14:56 +0000)]
Add /etc/firmware, remove /etc/gnats.

16 years agoGeneric firmware support. Currently implemented is loading from
Joerg Sonnenberger [Sat, 19 Mar 2005 14:54:50 +0000 (14:54 +0000)]
Generic firmware support. Currently implemented is loading from
/etc/firmware, support kernel builtin images and kernel modules
will follow later.

Written-by: Johannes Hofmann and Joerg Sonnenberger
16 years agoReindent stripvarname() to use 8 space tabs.
Max Okumoto [Sat, 19 Mar 2005 10:44:07 +0000 (10:44 +0000)]
Reindent stripvarname() to use 8 space tabs.

16 years agopatch-7.155
Max Okumoto [Sat, 19 Mar 2005 00:21:19 +0000 (00:21 +0000)]
Sort function proto types

16 years agopatch-7.154
Max Okumoto [Sat, 19 Mar 2005 00:20:55 +0000 (00:20 +0000)]
Fix indent VarPattern in var.h

16 years agopatch-7.151
Max Okumoto [Sat, 19 Mar 2005 00:19:55 +0000 (00:19 +0000)]
- Use Buf_Data() to simplify code.
- Use Buf_Peel() to simplify code.

16 years agopatch-7.150
Max Okumoto [Sat, 19 Mar 2005 00:19:13 +0000 (00:19 +0000)]
- Use Buf_Data() to simplify code.
- Use Buf_Peel() to simplify code.

16 years agopatch-7.149
Max Okumoto [Sat, 19 Mar 2005 00:18:50 +0000 (00:18 +0000)]
       - Use Buf_Data() to simplify code.
       - Use Buf_Peel() to simplify code.

16 years agopatch-7.148
Max Okumoto [Sat, 19 Mar 2005 00:18:28 +0000 (00:18 +0000)]
 Removed warning by using Buf_Peel() to simplify code.

16 years agopatch-7.147
Max Okumoto [Sat, 19 Mar 2005 00:18:01 +0000 (00:18 +0000)]
 Sort function declarations in buf.h

16 years agopatch-7.146
Max Okumoto [Sat, 19 Mar 2005 00:17:40 +0000 (00:17 +0000)]
 Add new functions:
       Buf_Data() - return a reference to the data contained in the buffer.
       Buf_AppendBuf(a, b) - append a copy of thecontent of b into a.

16 years agopatch-7.145
Max Okumoto [Sat, 19 Mar 2005 00:17:07 +0000 (00:17 +0000)]
 Replace redundent code with calls to VarGetPattern().

16 years agopatch-7.144
Max Okumoto [Sat, 19 Mar 2005 00:16:19 +0000 (00:16 +0000)]
       Removed unreachable code, VAR_NOSUBST is never set.

16 years agopatch-7.143
Max Okumoto [Sat, 19 Mar 2005 00:15:44 +0000 (00:15 +0000)]
       - Update vp->ptr instead of using temp variable.
       - Renamed variable pattern to patt.

16 years agoForced commit to supply a proper message for the previous change.
Sascha Wildner [Fri, 18 Mar 2005 23:20:34 +0000 (23:20 +0000)]
Forced commit to supply a proper message for the previous change.

1) Clean up:
* Traditional function declarations -> prototype form
* Remove unnecessary void casts
* Use '\0' where '\0' is meant
* Add some extra spaces
* include <stdlib.h> for exit()
* Add $DragonFly$ to Makefile

2) Raise WARNS to 6

16 years agoFreeBSD-Date: 2005/03/18 15:23:49
Max Okumoto [Fri, 18 Mar 2005 22:12:47 +0000 (22:12 +0000)]
FreeBSD-Date: 2005/03/18 15:23:49
FreeBSD-Date: 2005/03/18 15:25:23
Author: harti

o Replace a bogus use of Lst_Find with explicite LST_FOREACH calls
  and remove the unneeded second argument to ReadMakefile.

o Replace Lst_Find calls with LST_FOREACH loops. This helps in
  constification and simplifies the code because the one-liner
  predicates can be inlined into the code.

16 years ago- Correct usage of gethostname(3)
Liam J. Foy [Fri, 18 Mar 2005 22:08:08 +0000 (22:08 +0000)]
- Correct usage of gethostname(3)
- strlcpy

16 years agoFix dangling ref to auto variable.
Max Okumoto [Fri, 18 Mar 2005 18:39:07 +0000 (18:39 +0000)]
Fix dangling ref to auto variable.

16 years agoAdd a sysctl "net.inet.tcp.aggregate_acks" which controls the tcp ack
Matthew Dillon [Fri, 18 Mar 2005 18:29:05 +0000 (18:29 +0000)]
Add a sysctl "net.inet.tcp.aggregate_acks" which controls the tcp ack
aggregation feature, default to on.  This feature was added a while
back and most systems will want to leave it on.  The feature only effects
situations where TCP packets are aggregated by e.g. a GigE interface or a
router talking to a GigE interface and allows the tcp stack to send just
one ack after receiving a large burst (~8-10) of tcp packets rather then
sending four or five back-to-back acks.  The single ack actually gets the
sequence space updated more quickly on the other end as well as greatly
reduces the PPS rate and bandwidth used by the return channel.  Protocol
stacks on both ends are also made more efficient due to the fewer acks that
need to be sent and received.

The feature has one side-effect which warrents having the sysctl.  The
slow-start exponential ramp-up takes longer to ramp-up due to the lower
ack rate, reducing tcp bandwidth at the beginning of the connection (but
not once it gets going).  This ONLY effects links which huge bandwidth-delay
products (like 100ms x 100MBits), and then only for a short time, and
even in such cases the difference is not usually noticeable except when
explicitly tested for.  LANs and most WAN links are NOT effected.   The
positive aspects greatly outweight the negative ones in most cases and
sysops will almost always want to leave the feature on.

Delayed-Ack-Analysis-by: Noritoshi Demizu <demizu@dd.iij4u.or.jp>
16 years agoPatch originally from Matt, then modified by harti.
Max Okumoto [Fri, 18 Mar 2005 18:21:10 +0000 (18:21 +0000)]
Patch originally from Matt, then modified by harti.

FreeBSD-Date: 2005/03/18 15:16:09
Author: matt, harti

Fix a bug in matching suffixes. Under certain circumstances the code
would access memory before the beginning of the string to match (the
suffix match starts at the end of both the string and the suffix and
proceedes to the begin until either the start of the suffix is hit
or the character does not match). This could lead to a memcpy copying
into random memory. Fix this by checking the length of the string to
match too and replacing the Lst_Find calls with LST_FOREACH loops
(last part by harti).

16 years ago../../../ppt.msg
Sascha Wildner [Fri, 18 Mar 2005 17:42:46 +0000 (17:42 +0000)]

16 years agoRaise WARNS to 6:
Chris Pressey [Fri, 18 Mar 2005 01:57:58 +0000 (01:57 +0000)]
Raise WARNS to 6:
- Constify some char *'s.
- Mark parameters as unused in callback function.

16 years agoWARNS=6.
Joerg Sonnenberger [Fri, 18 Mar 2005 00:57:10 +0000 (00:57 +0000)]

Partly-submitted-by: Samuel J. Greear <sjg@evilcode.net>
16 years agoWARNS=6.
Joerg Sonnenberger [Fri, 18 Mar 2005 00:52:49 +0000 (00:52 +0000)]

Submitted-by: Samuel J. Greear <sjg@evilcode.net>
16 years agoAdd RCNG support for setting the negative attribute cache timeout, make
Matthew Dillon [Thu, 17 Mar 2005 22:25:46 +0000 (22:25 +0000)]
Add RCNG support for setting the negative attribute cache timeout, make
the boot-time display of nfs client parameters more readable, and properly
document the nfs_access_cache and nfs_neg_cache RCNG configuration variables.

Increase the default NFS attribute cache timeout from 2 to 5 seconds.

16 years agoIgnore additional stty control characters like ^Y which interfere with
Matthew Dillon [Thu, 17 Mar 2005 21:41:16 +0000 (21:41 +0000)]
Ignore additional stty control characters like ^Y which interfere with
the editor's operation.

Reported-by: Jasse Jansson <jasse@hornet.ac>
16 years agoTesting a new commits monitoring script.
David Rhodus [Thu, 17 Mar 2005 20:59:34 +0000 (20:59 +0000)]
Testing a new commits monitoring script.

16 years agoUnbreak the pkg_add -r which the last commit broke.
David Rhodus [Thu, 17 Mar 2005 20:53:55 +0000 (20:53 +0000)]
Unbreak the pkg_add -r which the last commit broke.

16 years agoClean up a number of caching edge cases in NFS, rework the code to be
Matthew Dillon [Thu, 17 Mar 2005 17:28:46 +0000 (17:28 +0000)]
Clean up a number of caching edge cases in NFS, rework the code to be
a bit more readable, document some bits, and fix some cache coherency
detection issues.  The caching cleanups should allow the NFS client to
retain more of the NFS cache when doing complex operations on a file.

* Properly check and update the mtime using WCC records in the NFS response.
  This record gives us the 'before' and 'after' mtime.  The 'before' mtime
  must match our existing idea of the mtime, if it doesn't we flag the
  nfsnode as having been modified by the server.  Our notion of the mtime
  is then set to the 'after time.  This was not being done properly for
  several edge cases.

  This required extending the nfsm macros a bit in order to be able to tell
  loadattrcache how to handle the mtime data.  This also required rearranging
  (really fixing) the sequence in nfs_open(), nfs_write(), etc.

* Rearrange the flags a bit.  NSIZECHANGED -> NRMODIFIED (nfsnode modified
  by server), NMODIFIED -> NLMODIFIED (nfsnode modified by client).  Do
  not clear NRMODIFIED until we have actually invalidated the cache (this
  fixes a problem where programs using mmap() were not properly clearing
  the cache after a file was modified on the server).

* Don't code NRMODIFIED as an exception to NLMODIFIED.  Recode the flags so
  they (mostly) operate in tandem.

* When appending to a file, use nfs_flush() instead of nfs_vinvalbuf().
  There is no need to destroy our data cache for the file.  This makes
  appends considerably more efficient.

* Hopefully fix the last problem associated with attribute timeouts.

* Clear the attribute cache when a file is opened for write in nfs_open()
  BEFORE doing other checks rather then after.

* Document some of the nastier cache coherency hacks.

16 years agoFix warnings. Raise WARNS to 6.
Joerg Sonnenberger [Thu, 17 Mar 2005 15:39:34 +0000 (15:39 +0000)]
Fix warnings. Raise WARNS to 6.

Submitted-by: Samuel J. Greear <sjg@evilcode.net>
Add some explicit XXX comments about the signal handlers,
they have to be rewritten.

16 years agoUse a common source for the string to integer conversion. Add
Joerg Sonnenberger [Thu, 17 Mar 2005 15:09:01 +0000 (15:09 +0000)]
Use a common source for the string to integer conversion. Add
strtoimax and strtoumax.

Inspired-by: wide char versions.