]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
22 years agoAdd modules of scsi_low based SCSI drivers ncv, nsp, stg, and common
Noriaki Mitsunaga [Sat, 15 Dec 2001 12:44:38 +0000 (12:44 +0000)]
Add modules of scsi_low based SCSI drivers ncv, nsp, stg, and common
part scsi_low.

Submitted-by: takawata
22 years agoAdd module directories and makefiles for ncv, nsp, stg and scsi_low
Noriaki Mitsunaga [Sat, 15 Dec 2001 12:37:36 +0000 (12:37 +0000)]
Add module directories and makefiles for ncv, nsp, stg and scsi_low
KLD modules.

Submitted-by: takawata
22 years agoo Add KLD support for scsi_low.
Noriaki Mitsunaga [Sat, 15 Dec 2001 12:32:23 +0000 (12:32 +0000)]
o Add KLD support for scsi_low.
o Add KLD dependency of ncv, nsp and stg drivers to scsi_low.

Submitted by: takawata

22 years agoAdd a missing .. to get out of smbfs/print.
John Hay [Sat, 15 Dec 2001 08:31:13 +0000 (08:31 +0000)]
Add a missing .. to get out of smbfs/print.

22 years agoNew release notes: root filesystem on a SCSI CDROM, filesystem
Bruce A. Mah [Sat, 15 Dec 2001 06:48:18 +0000 (06:48 +0000)]
New release notes:  root filesystem on a SCSI CDROM, filesystem
bugs fixed.

MFCs noted:  em(4), faith(4) loadable/unloadable/clonable,
sis(4) VLAN support, interface capabilities, IP address hash
table, syn cache, newfs(8) defaults.

22 years agoAdd some granularity to the WARNS levels.
David E. O'Brien [Sat, 15 Dec 2001 06:02:15 +0000 (06:02 +0000)]
Add some granularity to the WARNS levels.
1:  add -Werror
2:  -Wall [only], as this is the most used warnings setting by developers
3:  our old `1'
4:  our old `2'

22 years agoAdd support for suspending/resuming CardBus bridges.
Warner Losh [Sat, 15 Dec 2001 05:58:28 +0000 (05:58 +0000)]
Add support for suspending/resuming CardBus bridges.

We really should have and use power state information, but none exists
today.

Submitted by: YAMAMOTO Shigeru-san <shigeru@iij.ad.jp>

22 years agoautomatic creation of faith0 and stf0 for backward compatibility.
Hajimu UMEMOTO [Sat, 15 Dec 2001 03:59:47 +0000 (03:59 +0000)]
automatic creation of faith0 and stf0 for backward compatibility.

MFC after: 2 days

22 years agoPatch up some existing style bugs and some that crept in with the
Peter Wemm [Sat, 15 Dec 2001 02:51:21 +0000 (02:51 +0000)]
Patch up some existing style bugs and some that crept in with the
DEVICE_POLLING stuff.

22 years agoFix (again) some blatent style bugs in DEVICE_POLLING code.
Peter Wemm [Sat, 15 Dec 2001 02:41:52 +0000 (02:41 +0000)]
Fix (again) some blatent style bugs in DEVICE_POLLING code.

22 years agoAdd support for 56 bit MPPE encryption.
Archie Cobbs [Sat, 15 Dec 2001 02:07:32 +0000 (02:07 +0000)]
Add support for 56 bit MPPE encryption.

MFC after: 3 days

22 years agoclnt_bcast.c:420:33: warning: extra tokens at end of #endif directive
Peter Wemm [Sat, 15 Dec 2001 00:07:37 +0000 (00:07 +0000)]
clnt_bcast.c:420:33: warning: extra tokens at end of #endif directive

22 years agoFix some nits in fork_exit() so it more properly duplicates the backend
John Baldwin [Fri, 14 Dec 2001 23:37:35 +0000 (23:37 +0000)]
Fix some nits in fork_exit() so it more properly duplicates the backend
of mi_switch:
- Set the oncpu value for the current thread.
- Always set switchticks, not just in the SMP case.
- Add a KTR entry for fork_exit that is the same as the "new proc"
  entry in mi_switch().
- Release sched_lock a bit later like we do with mi_switch().

22 years agoAdd description of DEVICE_POLLING option.
Luigi Rizzo [Fri, 14 Dec 2001 23:35:33 +0000 (23:35 +0000)]
Add description of DEVICE_POLLING option.

22 years agoClarify the comments related to DUMMYNET and HZ
Luigi Rizzo [Fri, 14 Dec 2001 23:25:58 +0000 (23:25 +0000)]
Clarify the comments related to DUMMYNET and HZ

MFC after: 3 days

22 years agoPOSIX strto*() functions MAY return EINVAL, so don't assume that only one
Andrey A. Chernov [Fri, 14 Dec 2001 23:20:54 +0000 (23:20 +0000)]
POSIX strto*() functions MAY return EINVAL, so don't assume that only one
errno from them allowed and it is always ERANGE

22 years agoKernel support for smbfs is only built on the i386 at the moment, so
Sheldon Hearn [Fri, 14 Dec 2001 23:11:45 +0000 (23:11 +0000)]
Kernel support for smbfs is only built on the i386 at the moment, so
limit the building and installation of the userland utilities to that
architecture for now.

Reported by: bmah

22 years agoAdd share/examples/smbfs/print for userland smbfs import.
Sheldon Hearn [Fri, 14 Dec 2001 22:56:12 +0000 (22:56 +0000)]
Add share/examples/smbfs/print for userland smbfs import.

The addition of the smbfs examples should have included this delta.
Failure to include this delta broken world.

Reported by: jhay

22 years agoBetter document gif interface cloning intrastructure.
Brooks Davis [Fri, 14 Dec 2001 22:55:10 +0000 (22:55 +0000)]
Better document gif interface cloning intrastructure.

MFC After: 2 days

22 years agoFix two typos, TR6() and ITR6() need to use CTR6(), not CTR5().
John Baldwin [Fri, 14 Dec 2001 22:28:58 +0000 (22:28 +0000)]
Fix two typos, TR6() and ITR6() need to use CTR6(), not CTR5().

22 years agoFix a stray character that found its way into a filename.
Crist J. Clark [Fri, 14 Dec 2001 22:25:04 +0000 (22:25 +0000)]
Fix a stray character that found its way into a filename.

22 years agoMake bpf's read timeout feature work more correctly with
John Polstra [Fri, 14 Dec 2001 22:17:54 +0000 (22:17 +0000)]
Make bpf's read timeout feature work more correctly with
select/poll, and therefore with pthreads.  I doubt there is any way
to make this 100% semantically identical to the way it behaves in
unthreaded programs with blocking reads, but the solution here
should do the right thing for all reasonable usage patterns.

The basic idea is to schedule a callout for the read timeout when a
select/poll is done.  When the callout fires, it ends the select if
it is still in progress, or marks the state as "timed out" if the
select has already ended for some other reason.  Additional logic in
bpfread then does the right thing in the case where the timeout has
fired.

Note, I co-opted the bd_state member of the bpf_d structure.  It has
been present in the structure since the initial import of 4.4-lite,
but as far as I can tell it has never been used.

PR: kern/22063 and bin/31649
MFC after: 3 days

22 years agoDisabled input hardware checksum due to it being calculated incorrected
David Greenman [Fri, 14 Dec 2001 22:04:58 +0000 (22:04 +0000)]
Disabled input hardware checksum due to it being calculated incorrected
for some packets, in particular small (0 byte payload) packets. May also
be related to TCP options.

Approved by: wpaul
MFC after: 3 days

22 years agoAt least once mention the long names of WF2Q+ (Worst-case Fair Weighted
Ralf S. Engelschall [Fri, 14 Dec 2001 21:51:28 +0000 (21:51 +0000)]
At least once mention the long names of WF2Q+ (Worst-case Fair Weighted
Fair Queueing) and RED (Random Early Detection) to both give the reader
a hint what they are and to make it easier to find out more information
about them.

22 years agoAdjust the timings for the SiS chips a bit, also add the SiS 645.
Søren Schmidt [Fri, 14 Dec 2001 21:28:49 +0000 (21:28 +0000)]
Adjust the timings for the SiS chips a bit, also add the SiS 645.

Cosmetics on the Acer chips (print right modes)

22 years agoDo be so anal on detach, properly free interrupt even if no devices.
Søren Schmidt [Fri, 14 Dec 2001 21:23:19 +0000 (21:23 +0000)]
Do be so anal on detach, properly free interrupt even if no devices.

22 years agoAdd some wisdom to the jail setup instructions.
Poul-Henning Kamp [Fri, 14 Dec 2001 20:20:50 +0000 (20:20 +0000)]
Add some wisdom to the jail setup instructions.

22 years agoo Add IPOPT_ESO for the 'Extended Security' IP option (RFC1108)
Robert Watson [Fri, 14 Dec 2001 19:37:32 +0000 (19:37 +0000)]
o Add IPOPT_ESO for the 'Extended Security' IP option (RFC1108)

Obtained from: TrustedBSD Project

22 years agoo Add definition for IPOPT_CIPSO, the commercial security IP option
Robert Watson [Fri, 14 Dec 2001 19:34:42 +0000 (19:34 +0000)]
o Add definition for IPOPT_CIPSO, the commercial security IP option
  number.

Submitted by: Ilmar S. Habibulin <ilmar@watson.org>
Obtained from: TrustedBSD Project

22 years agowhitespace and style fixes recovered from -stable.
Jonathan Lemon [Fri, 14 Dec 2001 19:34:11 +0000 (19:34 +0000)]
whitespace and style fixes recovered from -stable.

22 years agominor style and whitespace fixes.
Jonathan Lemon [Fri, 14 Dec 2001 19:33:29 +0000 (19:33 +0000)]
minor style and whitespace fixes.

22 years agowhitespace fixes.
Jonathan Lemon [Fri, 14 Dec 2001 19:32:47 +0000 (19:32 +0000)]
whitespace fixes.

22 years agominor whitespace fixes.
Jonathan Lemon [Fri, 14 Dec 2001 19:32:00 +0000 (19:32 +0000)]
minor whitespace fixes.

22 years agominor style fix.
Jonathan Lemon [Fri, 14 Dec 2001 19:28:06 +0000 (19:28 +0000)]
minor style fix.

22 years agoMake the man page reflec t the code a bit better.
Julian Elischer [Fri, 14 Dec 2001 19:19:31 +0000 (19:19 +0000)]
Make the man page reflec t the code a bit better.
Specifically, note the condition of the memory on initial
and subsequent allocations is different.

22 years agoWhen removing kqueue descriptors from the descriptor table during a fork,
Jonathan Lemon [Fri, 14 Dec 2001 19:02:57 +0000 (19:02 +0000)]
When removing kqueue descriptors from the descriptor table during a fork,
update fd_freefile and fd_lastfile as well, to keep things in sync.

Pointed out by: Debbie Chu <dchu@juniper.net>

22 years agoReduce the local network slowstart flightsize from infinity to 4 packets.
Mike Silbersack [Fri, 14 Dec 2001 18:26:52 +0000 (18:26 +0000)]
Reduce the local network slowstart flightsize from infinity to 4 packets.

Now that we've increased the size of our send / receive buffers, bursting
an entire window onto the network may cause congestion.  As a result,
we will slow start beginning with a flightsize of 4 packets.

Problem reported by: Thomas Zenker <thz@Lennartz-electronic.de>

MFC after: 3 days

22 years agoDevice Polling code for -current.
Luigi Rizzo [Fri, 14 Dec 2001 17:56:12 +0000 (17:56 +0000)]
Device Polling code for -current.

Non-SMP, i386-only, no polling in the idle loop at the moment.

To use this code you must compile a kernel with

        options DEVICE_POLLING

and at runtime enable polling with

        sysctl kern.polling.enable=1

The percentage of CPU reserved to userland can be set with

        sysctl kern.polling.user_frac=NN (default is 50)

while the remainder is used by polling device drivers and netisr's.
These are the only two variables that you should need to touch. There
are a few more parameters in kern.polling but the default values
are adequate for all purposes. See the code in kern_poll.c for
more details on them.

Polling in the idle loop will be implemented shortly by introducing
a kernel thread which does the job. Until then, the amount of CPU
dedicated to polling will never exceed (100-user_frac).
The equivalent (actually, better) code for -stable is at

http://info.iet.unipi.it/~luigi/polling/

and also supports polling in the idle loop.

NOTE to Alpha developers:
There is really nothing in this code that is i386-specific.
If you move the 2 lines supporting the new option from
sys/conf/{files,options}.i386 to sys/conf/{files,options} I am
pretty sure that this should work on the Alpha as well, just that
I do not have a suitable test box to try it. If someone feels like
trying it, I would appreciate it.

NOTE to other developers:
sure some things could be done better, and as always I am open to
constructive criticism, which a few of you have already given and
I greatly appreciated.
However, before proposing radical architectural changes, please
take some time to possibly try out this code, or at the very least
read the comments in kern_poll.c, especially re. the reason why I
am using a soft netisr and cannot (I believe) replace it with a
simple timeout.

Quick description of files touched by this commit:

sys/conf/files.i386
        new file kern/kern_poll.c
sys/conf/options.i386
        new option
sys/i386/i386/trap.c
        poll in trap (disabled by default)
sys/kern/kern_clock.c
        initialization and hardclock hooks.
sys/kern/kern_intr.c
        minor swi_net changes
sys/kern/kern_poll.c
        the bulk of the code.
sys/net/if.h
        new flag
sys/net/if_var.h
        declaration for functions used in device drivers.
sys/net/netisr.h
        NETISR_POLL
sys/dev/fxp/if_fxp.c
sys/dev/fxp/if_fxpvar.h
sys/pci/if_dc.c
sys/pci/if_dcreg.h
sys/pci/if_sis.c
sys/pci/if_sisreg.h
        device driver modifications

22 years agoLet M_LEADINGSPACE write into non-shared mbufs.
Luigi Rizzo [Fri, 14 Dec 2001 17:31:58 +0000 (17:31 +0000)]
Let M_LEADINGSPACE write into non-shared mbufs.
A similar thing has been in -stable for weeks and is completely safe.

This has very good performance implications as it saves some data
copying, and sometimes avoids triggering performance bugs in devices
(such as the "dc" and other Tulip clones) which do not like scattered
data.

22 years agoEnable UFS_DIRHASH in the GENERIC kernel.
Ian Dowse [Fri, 14 Dec 2001 16:27:11 +0000 (16:27 +0000)]
Enable UFS_DIRHASH in the GENERIC kernel.

Suggested by: silby
Reviewed by: dillon
MFC after: 5 days

22 years agoAdd prototypes for main() so that these programs compile with -Werror
Luigi Rizzo [Fri, 14 Dec 2001 16:22:41 +0000 (16:22 +0000)]
Add prototypes for main() so that these programs compile with -Werror
(which somehow now seems to be the default for compiling -current).
This error popped up while doing a PicoBSD cross-compile on a 4.3-ish system,
it may well be that there are other apps which have similar problems,
but I did not spot them as they are not included in my picobsd config.

Whether adding prototypes for main() is the correct solution or not
I have no idea, a request to -current on the matter went basically
unanswered. Those who have better ideas are welcome to back this out
and replace it with the correct fix.

22 years agoftp(1) was not the only user of FTP_PASSIVE_MODE, libfetch uses it
Mike Heffner [Fri, 14 Dec 2001 15:48:55 +0000 (15:48 +0000)]
ftp(1) was not the only user of FTP_PASSIVE_MODE, libfetch uses it
too, so add it back

22 years agoFixed to draw mouse cursor. The syscons driver for PC98 uses different
Yoshihiro Takahashi [Fri, 14 Dec 2001 15:27:15 +0000 (15:27 +0000)]
Fixed to draw mouse cursor. The syscons driver for PC98 uses different
attributes from i386.

Submitted by: chi@bd.mbn.or.jp (Chiharu Shibata)
MFC after: 3 days

22 years agomdoc(7) police: add -p to the synopsis line as well.
Ruslan Ermilov [Fri, 14 Dec 2001 15:03:32 +0000 (15:03 +0000)]
mdoc(7) police: add -p to the synopsis line as well.

22 years agoMFCrypto: Remove -r, -s, sort -p.
Ruslan Ermilov [Fri, 14 Dec 2001 14:46:51 +0000 (14:46 +0000)]
MFCrypto: Remove -r, -s, sort -p.

22 years agomdoc(7) police: remove -r from SYNOPSIS, sort -p in DESCRIPTION.
Ruslan Ermilov [Fri, 14 Dec 2001 14:41:07 +0000 (14:41 +0000)]
mdoc(7) police: remove -r from SYNOPSIS, sort -p in DESCRIPTION.

22 years agomdoc(7) police: bump document date.
Ruslan Ermilov [Fri, 14 Dec 2001 13:49:28 +0000 (13:49 +0000)]
mdoc(7) police: bump document date.

22 years agoBump __FreeBSD_version for userland smbfs import.
Sheldon Hearn [Fri, 14 Dec 2001 12:41:56 +0000 (12:41 +0000)]
Bump __FreeBSD_version for userland smbfs import.

22 years agoArrange for the smbfs examples to be installed.
Sheldon Hearn [Fri, 14 Dec 2001 12:17:03 +0000 (12:17 +0000)]
Arrange for the smbfs examples to be installed.

We don't install dot.nsmbrc or smbfs.sh.sample, since we already install
the former as /etc/nsmb.conf and the latter is unnecessary, since
boot-time mounts can be arranged directly within /etc/fstab without fear
of breaking the boot when the smbfs port (now unnecessary is removed).

The MFC reminder below is subject to <re@FreeBSD.org> approval
priod to 4.5-RELEASE.

MFC after: 1 week

22 years agoAdd bmake glue for src/contrib/smbfs and connect userland smbfs
Sheldon Hearn [Fri, 14 Dec 2001 11:41:22 +0000 (11:41 +0000)]
Add bmake glue for src/contrib/smbfs and connect userland smbfs
support to the build.

The MFC reminder below is subject to <re@FreeBSD.org> approval
prior to 4.5-RELEASE.

Reviewed by: bp, fjoe
MFC: 1 week

22 years ago* cleanup comments and defines
Alexey Zelkin [Fri, 14 Dec 2001 11:36:37 +0000 (11:36 +0000)]
* cleanup comments and defines

Reviewed by: bde

22 years agoo Clarify the comments on AIO to note that yes, AIO really is unsuitable
Robert Watson [Fri, 14 Dec 2001 11:21:16 +0000 (11:21 +0000)]
o Clarify the comments on AIO to note that yes, AIO really is unsuitable
  for use on machines with untrusted local users, for security as well
  as stability reasons.
o Lack of clarity pointed out by: David Rufino <dr@soniq.net> via bugtraq.

22 years agoDetail the smbfs-1.4.1 import.
Sheldon Hearn [Fri, 14 Dec 2001 11:11:31 +0000 (11:11 +0000)]
Detail the smbfs-1.4.1 import.

22 years agoThis commit was generated by cvs2svn to compensate for changes in r87866,
Sheldon Hearn [Fri, 14 Dec 2001 11:06:03 +0000 (11:06 +0000)]
This commit was generated by cvs2svn to compensate for changes in r87866,
which included commits to RCS files with non-trunk default branches.

22 years agoImport smbfs-1.4.1.
Sheldon Hearn [Fri, 14 Dec 2001 11:06:03 +0000 (11:06 +0000)]
Import smbfs-1.4.1.

This is Boris Popov's SMB/CIFS file system implementation for FreeBSD.

Obtained from: Boris Popov via ftp://ftp.butya.kz/pub/smbfs/

22 years agoAdd disk I/O scheduling for positively niced processes.
Kirk McKusick [Fri, 14 Dec 2001 10:49:15 +0000 (10:49 +0000)]
Add disk I/O scheduling for positively niced processes.
When a positively niced process requests a disk I/O, make
it wait for its nice value of ticks before scheduling its
I/O request if there are any other processes with I/O
requests in the disk queue. For all the gory details, see
the ``Running fsck in the Background'' paper in the Usenix
BSDCon 2002 Conference Proceedings, pages 55-64.

22 years agomdoc(7) police overhaul.
Ruslan Ermilov [Fri, 14 Dec 2001 10:18:15 +0000 (10:18 +0000)]
mdoc(7) police overhaul.

22 years agomdoc(7) police overhaul.
Ruslan Ermilov [Fri, 14 Dec 2001 09:48:03 +0000 (09:48 +0000)]
mdoc(7) police overhaul.

22 years agoBump configvers for different handling of maxusers.
Peter Wemm [Fri, 14 Dec 2001 09:40:01 +0000 (09:40 +0000)]
Bump configvers for different handling of maxusers.

22 years agoProper fix for old config setting maxusers to 8.
Peter Wemm [Fri, 14 Dec 2001 09:39:29 +0000 (09:39 +0000)]
Proper fix for old config setting maxusers to 8.

22 years agoFixed prototype.
Ruslan Ermilov [Fri, 14 Dec 2001 09:25:11 +0000 (09:25 +0000)]
Fixed prototype.

Reviewed by: davidc

22 years agomdoc(7) police: don't xref to itself.
Ruslan Ermilov [Fri, 14 Dec 2001 09:22:34 +0000 (09:22 +0000)]
mdoc(7) police: don't xref to itself.

22 years agomdoc(7) police: markup, grammar and spelling nits.
Ruslan Ermilov [Fri, 14 Dec 2001 09:11:13 +0000 (09:11 +0000)]
mdoc(7) police: markup, grammar and spelling nits.

Reviewed by: davidc

22 years agomdoc(7) police: kill the undead whitespace at EOL.
Ruslan Ermilov [Fri, 14 Dec 2001 09:09:21 +0000 (09:09 +0000)]
mdoc(7) police: kill the undead whitespace at EOL.

22 years agomdoc(7) police: remove hard sentence break which I didn't
Ruslan Ermilov [Fri, 14 Dec 2001 09:06:11 +0000 (09:06 +0000)]
mdoc(7) police: remove hard sentence break which I didn't
submit to Jonathan Mini.

22 years agoWork around the bugfeature of test(1).
Ruslan Ermilov [Fri, 14 Dec 2001 08:58:21 +0000 (08:58 +0000)]
Work around the bugfeature of test(1).

PR: bin/32822

22 years agoA slightly different version of the vlrureclaim fix.
Matthew Dillon [Fri, 14 Dec 2001 07:18:31 +0000 (07:18 +0000)]
A slightly different version of the vlrureclaim fix.

Reported by: peter, ps

22 years agoRemove printf's on mbuf/cluster allocation failures. There are now
Luigi Rizzo [Fri, 14 Dec 2001 05:56:35 +0000 (05:56 +0000)]
Remove printf's on mbuf/cluster allocation failures. There are now
equivalent and less dangerous (rate limited) messages in
the mbuf allocation code.

MFC after: 3 days

22 years agoAdd disk I/O scheduling for positively niced processes.
Kirk McKusick [Fri, 14 Dec 2001 05:50:44 +0000 (05:50 +0000)]
Add disk I/O scheduling for positively niced processes.
When a positively niced process requests a disk I/O, make
it wait for its nice value of ticks before scheduling its
I/O request if there are any other processes with I/O
requests in the disk queue. For all the gory details, see
the ``Running fsck in the Background'' paper in the Usenix
BSDCon 2002 Conference Proceedings, pages 55-64.

22 years agoMoved the updating of if_ibytes from ether_demux() to ether_input() to fix
David Greenman [Fri, 14 Dec 2001 04:41:07 +0000 (04:41 +0000)]
Moved the updating of if_ibytes from ether_demux() to ether_input() to fix
a bug where the interface input bytes count wasn't updated when bridging
is enabled.

MFC after: 3 days

22 years agoWhitespace and capitalization fixes to sync with RELENG_4.
Bruce A. Mah [Fri, 14 Dec 2001 04:29:57 +0000 (04:29 +0000)]
Whitespace and capitalization fixes to sync with RELENG_4.

22 years agoToo many people are compiling kernels with maxusers set to 0 without the new
Matthew Dillon [Fri, 14 Dec 2001 04:01:08 +0000 (04:01 +0000)]
Too many people are compiling kernels with maxusers set to 0 without the new
config.  Hack the kernel to force auto-sizing if the old config is used.

22 years agoftp(1) uses passive mode by default now, therefore remove
Mike Heffner [Fri, 14 Dec 2001 02:08:23 +0000 (02:08 +0000)]
ftp(1) uses passive mode by default now, therefore remove
FTP_PASSIVE_MODE=YES.

Pointed out by: billf

22 years agoThis fixes a large number of bugs in our NFS client side code. A recent
Matthew Dillon [Fri, 14 Dec 2001 01:16:57 +0000 (01:16 +0000)]
This fixes a large number of bugs in our NFS client side code.  A recent
commit by Kirk also fixed a softupdates bug that could easily be triggered
by server side NFS.

* An edge case with shared R+W mmap()'s and truncate whereby
  the system would inappropriately clear the dirty bits on
  still-dirty data.  (applicable to all filesystems)

  THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING.
  see vm/vm_page.c line 1641

* The straddle case for VM pages and buffer cache buffers when
  truncating.  (applicable to NFS client side)

* Possible SMP database corruption due to vm_pager_unmap_page()
  not clearing the TLB for the other cpu's.  (applicable to NFS
  client side but could effect all filesystems).  Note: not
  considered serious since the corruption occurs beyond the file
  EOF.

* When flusing a dirty buffer due to B_CACHE getting cleared,
  we were accidently setting B_CACHE again (that is, bwrite() sets
  B_CACHE), when we really want it to stay clear after the write
  is complete.  This resulted in a corrupt buffer.  (applicable
  to all filesystems but probably only triggered by NFS)

* We have to call vtruncbuf() when ftruncate()ing to remove
  any buffer cache buffers.  This is still tentitive, I may
  be able to remove it due to the second bug fix.  (applicable
  to NFS client side)

* vnode_pager_setsize() race against nfs_vinvalbuf()... we have
  to set n_size before calling nfs_vinvalbuf or the NFS code
  may recursively vnode_pager_setsize() to the original value
  before the truncate.  This is what was causing the user mmap
  bus faults in the nfs tester program.  (applicable to NFS
  client side)

* Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made
  by Kirk).

Testing program written by: Avadis Tevanian, Jr.
Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step')
MFC after: 1 week

22 years agoNew release note: lukemftp.
Bruce A. Mah [Fri, 14 Dec 2001 01:14:41 +0000 (01:14 +0000)]
New release note:  lukemftp.

22 years agoAdd a '-l' flag to show CPU load for the various states, similar to what
Luigi Rizzo [Fri, 14 Dec 2001 01:10:50 +0000 (01:10 +0000)]
Add a '-l' flag to show CPU load for the various states, similar to what
"top" does (thinking of it, i could have as well used the same format line!)

This only makes sense when "-w" option is also specified, because the
load is computed as the difference between subsequent samples.

I think this (and the "-d" feature which shows differences in the
network statistics counts) would also make sense in the standard
vmstat and netstat.

22 years agoAdd maxusers auto-sizing description to NOTES file for -current
Matthew Dillon [Fri, 14 Dec 2001 01:01:20 +0000 (01:01 +0000)]
Add maxusers auto-sizing description to NOTES file for -current

22 years agoo Wording fix in comment.
Robert Watson [Fri, 14 Dec 2001 00:38:01 +0000 (00:38 +0000)]
o Wording fix in comment.

Submitted by: tanimura via p4

22 years agoMinimize the time necessary to suspend operations on a filesystem
Kirk McKusick [Fri, 14 Dec 2001 00:15:06 +0000 (00:15 +0000)]
Minimize the time necessary to suspend operations on a filesystem
when taking a snapshot. The two time consuming operations are
scanning all the filesystem bitmaps to determine which blocks
are in use and scanning all the other snapshots so as to be able
to expunge their blocks from the view of the current snapshot.
The bitmap scanning is broken into two passes. Before suspending
the filesystem all bitmaps are scanned. After the suspension,
those bitmaps that changed after being scanned the first time
are rescanned. Typically there are few bitmaps that need to be
rescanned. The expunging of other snapshots is now done after
the suspension is released by observing that we can easily
identify any blocks that were allocated to them after the
suspension (they will be maked as `not needing to be copied'
in the just created snapshot). For all the gory details, see
the ``Running fsck in the Background'' paper in the Usenix
BSDCon 2002 Conference Proceedings, pages 55-64.

22 years agoConvert C++ style comments to proper C ones.
David E. O'Brien [Thu, 13 Dec 2001 23:48:34 +0000 (23:48 +0000)]
Convert C++ style comments to proper C ones.
Clean up C comments just a tad.
Fix ID's.

22 years agoConnect lukemftp to the build as the default ftp client. Lukemftp
Mike Heffner [Thu, 13 Dec 2001 23:46:44 +0000 (23:46 +0000)]
Connect lukemftp to the build as the default ftp client. Lukemftp
supports most of the previous features of FreeBSD ftp, but has been
better maintained and includes new features.

22 years agoIf we were called to allocate a vnode that is not associated with a
Peter Wemm [Thu, 13 Dec 2001 23:46:01 +0000 (23:46 +0000)]
If we were called to allocate a vnode that is not associated with a
mount point, do not dereference the NULL mp argument.

22 years agoRemove stale prototype for sonewconn3().
Brian Feldman [Thu, 13 Dec 2001 22:16:54 +0000 (22:16 +0000)]
Remove stale prototype for sonewconn3().

22 years agoo Back out portions of 1.50 and 1.47, eliminating sonewconn3() and
Robert Watson [Thu, 13 Dec 2001 22:09:37 +0000 (22:09 +0000)]
o Back out portions of 1.50 and 1.47, eliminating sonewconn3() and
  always deriving the credential for a newly accepted connection from
  the listen socket.  Previously, the selection of the credential
  depended on the protocol: UNIX domain sockets would use the
  connecting process's credential, and protocols supporting a creation
  of the socket before the receiving end called accept() would use
  the listening socket.  After this change, it is always the listening
  credential.

Reviewed by: green

22 years agoNew release notes: boot loader -p for i386, maxusers autosizing,
Bruce A. Mah [Thu, 13 Dec 2001 21:18:56 +0000 (21:18 +0000)]
New release notes:  boot loader -p for i386, maxusers autosizing,
burncd(8) for VCDs/SVCDs, newfs(8) new 16K/2K block/frag defaults,
watch(8) -f.

Updated release note: sis(4) VAN and note its MFC done a long time
ago.

22 years agoAlso fix cases when thousands separator should be put before number. For
Alexey Zelkin [Thu, 13 Dec 2001 21:05:27 +0000 (21:05 +0000)]
Also fix cases when thousands separator should be put before number. For
example before for grouping sequence "\003\003" number 123456 was formated
as ",123,456", now "123,456".

22 years agoLimit maxprocperuid to 9/10 maxproc, and limit maxfilesperproc to 9/10
Mike Silbersack [Thu, 13 Dec 2001 20:00:45 +0000 (20:00 +0000)]
Limit maxprocperuid to 9/10 maxproc, and limit maxfilesperproc to 9/10
maxfiles.  This should make local resource exhaustion attacks easier
to handle with a non-tweaked setup.

MFC after: 3 days

22 years agoUse C comments instead of C++ comments.
John Baldwin [Thu, 13 Dec 2001 19:54:12 +0000 (19:54 +0000)]
Use C comments instead of C++ comments.

Reported by: gcc30

22 years agoRespect locale while handling of \' flag.
Alexey Zelkin [Thu, 13 Dec 2001 19:45:41 +0000 (19:45 +0000)]
Respect locale while handling of \' flag.

In original version grouping was hardcoded. It assumed that thousands
separator should be inserted to separate each 3 numbers. I.e. grouping
string "\003" was assumed for all cases. In correct case (per POSIX)
vfprintf should respect locale defined non-monetary (LC_NUMERIC
category) grouping sequence.

Also simplify thousands_sep handling.

22 years agoUUCP is gone from base.
Mark Murray [Thu, 13 Dec 2001 18:27:08 +0000 (18:27 +0000)]
UUCP is gone from base.

22 years agoFix missing semi-colons on some entity references.
Bruce A. Mah [Thu, 13 Dec 2001 18:12:12 +0000 (18:12 +0000)]
Fix missing semi-colons on some entity references.

22 years agoFix style bugs (mostly remove 'extern' from function prototypes)
Alexey Zelkin [Thu, 13 Dec 2001 17:22:17 +0000 (17:22 +0000)]
Fix style bugs (mostly remove 'extern' from function prototypes)

Inspired by: conversation with bde

22 years ago* remove reference to m68k-dependent sources
Alexey Zelkin [Thu, 13 Dec 2001 17:18:26 +0000 (17:18 +0000)]
* remove reference to m68k-dependent sources
* fix comment

22 years agoMove NO_WERROR from CROSSENV to BMAKE. In CROSSENV it disables it for
Warner Losh [Thu, 13 Dec 2001 17:00:59 +0000 (17:00 +0000)]
Move NO_WERROR from CROSSENV to BMAKE.  In CROSSENV it disables it for
the entire build.  We only want it for the bootstrap process.

Submitted by: ru

22 years agoTone down and remove some obnoxious warnings that are slightly overkill.
Jonathan Lemon [Thu, 13 Dec 2001 16:13:31 +0000 (16:13 +0000)]
Tone down and remove some obnoxious warnings that are slightly overkill.

22 years agoAdd module dependency on libmchain.
Sheldon Hearn [Thu, 13 Dec 2001 13:08:34 +0000 (13:08 +0000)]
Add module dependency on libmchain.

With this change, mounting an smb share (using mount_smb, which is not
yet included in the tree) without any of smbfs, libiconv or libmchain
compiled into the kernel or loaded works.

22 years agoUse ANSI C string contatenation instead of a multi-line string literal.
John Baldwin [Thu, 13 Dec 2001 11:14:28 +0000 (11:14 +0000)]
Use ANSI C string contatenation instead of a multi-line string literal.

Reported by: gcc30

22 years agoComment tokens after #undef <macroname> and #endif.
John Baldwin [Thu, 13 Dec 2001 11:12:30 +0000 (11:12 +0000)]
Comment tokens after #undef <macroname> and #endif.

Reported by: gcc30

22 years agoIf the OID is missing, sysctl(3) returns ENOENT, not EOPNOTSUPP.
Ruslan Ermilov [Thu, 13 Dec 2001 10:59:00 +0000 (10:59 +0000)]
If the OID is missing, sysctl(3) returns ENOENT, not EOPNOTSUPP.

PR: docs/32793
Submitted by: Hartmut Brandt <brandt@fokus.gmd.de>

22 years agoUse a per-thread variable for keeping state when a thread is processing
John Baldwin [Thu, 13 Dec 2001 10:33:20 +0000 (10:33 +0000)]
Use a per-thread variable for keeping state when a thread is processing
a KTR log entry.  Any KTR requests made while working on an entry are
ignored/discarded to prevent recursion.  This is a better fix for the
hack to futz with the CPU mask and call getnanotime() if KTR_LOCK or
KTR_WITNESS was on.  It also covers the actual formatting of the log entry
including dumping it to the display which the earlier hacks did not.