]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
17 years agoDocument that jot(1), rs(1), and lam(1) first appeared in 4.2BSD.
cperciva [Fri, 20 Oct 2006 04:31:01 +0000 (04:31 +0000)]
Document that jot(1), rs(1), and lam(1) first appeared in 4.2BSD.

MFC after: 1 month

17 years agoUse TARGET_BIG_ENDIAN instead of ARM_BIG_ENDIAN
cognet [Thu, 19 Oct 2006 22:51:26 +0000 (22:51 +0000)]
Use TARGET_BIG_ENDIAN instead of ARM_BIG_ENDIAN

17 years agoMy previous commit made it impossible to export / over NFS; this small
rink [Thu, 19 Oct 2006 19:46:49 +0000 (19:46 +0000)]
My previous commit made it impossible to export / over NFS; this small
change fixes this.

Reported by: Lin Jui-Nan Eric < ericlin dot jnlin at gmail dot com >
Submitted by: Erik Trulsson < ertr1013 at student dot uu dot se > (commit is inspired by a patch from Erik)
Pointyhat to: me
Reviewed by: stable@
Approved by: imp (mentor)
MFC after: 3 days

17 years ago- Define time_t in sys/sem.h so it doesn't require sys/types.h.
jhb [Thu, 19 Oct 2006 14:36:41 +0000 (14:36 +0000)]
- Define time_t in sys/sem.h so it doesn't require sys/types.h.
- Move the pid_t, size_t, and time_t definitions earlier in the file, sort
  them, and fix whitespace.
- Drop redundant includes of sys/cdefs.h and sys/_types.h as sys/ipc.h
  already includes them.

PR: kern/104436
Reviewed by: bde
Reported by: clsung
MFC after: 3 days

17 years agoAdd missing "const" qualifiers.
ru [Thu, 19 Oct 2006 11:03:44 +0000 (11:03 +0000)]
Add missing "const" qualifiers.

Submitted by: bde

17 years agostyle(9)
obrien [Thu, 19 Oct 2006 10:04:46 +0000 (10:04 +0000)]
style(9)

17 years agostyle(9) cleanup.
obrien [Thu, 19 Oct 2006 10:01:26 +0000 (10:01 +0000)]
style(9) cleanup.

17 years agoThere is no reason to have a PAGE_SIZE alignment enforced on all DMA by the
scottl [Thu, 19 Oct 2006 08:03:22 +0000 (08:03 +0000)]
There is no reason to have a PAGE_SIZE alignment enforced on all DMA by the
parent tag.

17 years agoI can't find any reason why an 8 byte alignment should be enforced on rx
scottl [Thu, 19 Oct 2006 08:01:43 +0000 (08:01 +0000)]
I can't find any reason why an 8 byte alignment should be enforced on rx
and tx buffers.  Fix it there, and also don't have it be overridden by the
parent tag.

17 years ago- convert to the current style of section 4 man pages
danger [Thu, 19 Oct 2006 07:41:47 +0000 (07:41 +0000)]
- convert to the current style of section 4 man pages
- s/ip.forwarding/net.inet.ip.forwarding/ to match reality [1]

Approved by: keramida (mentor), trhodes (mentor)
MFC after: 5 days
Submitted by: Tom McLaughlin via #bsddocs

17 years agoRemove some spurious debugging, and use more complete error handling for
scottl [Thu, 19 Oct 2006 06:58:01 +0000 (06:58 +0000)]
Remove some spurious debugging, and use more complete error handling for
tx load and fragmentation problems.

17 years agoAdd missing acpi_wakecode.o: assym.s dependency, so that if assym.s
ru [Thu, 19 Oct 2006 05:55:09 +0000 (05:55 +0000)]
Add missing acpi_wakecode.o: assym.s dependency, so that if assym.s
is newer than acpi_wakecode.h, the latter is rebuilt.

Reported by: bde

17 years agotweak comments about OLDCARD workarounds
imp [Thu, 19 Oct 2006 05:25:53 +0000 (05:25 +0000)]
tweak comments about OLDCARD workarounds

17 years agotweak comment
imp [Thu, 19 Oct 2006 05:25:29 +0000 (05:25 +0000)]
tweak comment

17 years agoRemove references to pccard.conf, it is no longer used.
imp [Thu, 19 Oct 2006 05:19:00 +0000 (05:19 +0000)]
Remove references to pccard.conf, it is no longer used.

17 years agoRemove references to pccard.conf
imp [Thu, 19 Oct 2006 05:17:55 +0000 (05:17 +0000)]
Remove references to pccard.conf

17 years agoUse a different task queue for host controller and peripheral driver
iedowse [Thu, 19 Oct 2006 01:15:58 +0000 (01:15 +0000)]
Use a different task queue for host controller and peripheral driver
tasks. Since the host controllers rely on tasks to process transfer
timeouts, if a synchronous transfer from a driver was invoked from
a task and timed out, it would never complete because the single
task thread was stuck performing the synchronous transfer so couldn't
process the timeout.

This affected the axe, udav and ural drivers.

Problem hardware provided by: guido

17 years agoFix previous revision:
marcel [Thu, 19 Oct 2006 00:53:35 +0000 (00:53 +0000)]
Fix previous revision:
o  day and mday are the same. No need to subtract 1 from mday.
o  Set dow to -1 as clock_ct_to_ts() checks this field and
   returns EINVAL on any day of the week but Sunday.

17 years agoPackage the dumpcisfile functionality into a new dumpcis command.
imp [Wed, 18 Oct 2006 21:09:43 +0000 (21:09 +0000)]
Package the dumpcisfile functionality into a new dumpcis command.
Also, don't build pccardc now.  I've left it in place for people that
wish to build it to test experimental support for oldcard
functionality that may have been lost in newcard.

17 years agoDelete now-unused portion of pccardd. OLDCARD is gone from the kernel.
imp [Wed, 18 Oct 2006 20:48:05 +0000 (20:48 +0000)]
Delete now-unused portion of pccardd.  OLDCARD is gone from the kernel.

17 years agoAlways build pccardc.
imp [Wed, 18 Oct 2006 19:56:41 +0000 (19:56 +0000)]
Always build pccardc.

# I should remove the oldcard vestages from it...

17 years agoFix remaining compile error.
grehan [Wed, 18 Oct 2006 19:56:20 +0000 (19:56 +0000)]
Fix remaining compile error.

17 years agoAdd 'reload' to the list of available commands for the amd rc.d script.
flz [Wed, 18 Oct 2006 15:56:11 +0000 (15:56 +0000)]
Add 'reload' to the list of available commands for the amd rc.d script.

PR: conf/104507
Submitted by: Douglas K. Rand <rand@meridian-enviro.com>
MFC after: 3 days

17 years agoo Backout rev. 1.55. Don't waste cpu cycles for bzero(), do not
maxim [Wed, 18 Oct 2006 13:16:06 +0000 (13:16 +0000)]
o Backout rev. 1.55.  Don't waste cpu cycles for bzero(), do not
call chflags() for whiteouted files.

Prodded by: ru

17 years agoNormalize MAN.
ru [Wed, 18 Oct 2006 12:35:24 +0000 (12:35 +0000)]
Normalize MAN.

17 years agoProperly lock the vnode around vgone() calls.
kib [Wed, 18 Oct 2006 11:17:14 +0000 (11:17 +0000)]
Properly lock the vnode around vgone() calls.

Unlock the vnode in devfs_close() while calling into the driver d_close()
routine.

devfs_revoke() changes by: ups
Reviewed and bugfixes by: tegge
Tested by: mbr, Peter Holm
Approved by: pjd (mentor)
MFC after: 1 week

17 years agoAvoid a spurious warning for each whiteout found during "ls -lW".
ru [Wed, 18 Oct 2006 10:58:27 +0000 (10:58 +0000)]
Avoid a spurious warning for each whiteout found during "ls -lW".

# ls -lW
total 2
-rw-r--r--  1 root  wheel  6 Oct 18 14:46 file1
ls: ./file2: No such file or directory
w---------  0 root  wheel  0 Jan  1  1970 file2

17 years agoUpdate for newsletter V-12: Inclusion of the new entries for
ru [Wed, 18 Oct 2006 09:02:11 +0000 (09:02 +0000)]
Update for newsletter V-12: Inclusion of the new entries for
"Serbia" and "Montenegro".

Reported by: mnag

17 years agoo Zero out struct stat before usage. lstat(2) can fail and
maxim [Wed, 18 Oct 2006 08:22:33 +0000 (08:22 +0000)]
o Zero out struct stat before usage.  lstat(2) can fail and
leave garbage there which will break -W code path.

PR: bin/84569
Submitted by: Igor
MFC after: 2 weeks

17 years agoCheck pointer before dereferencing.
glebius [Wed, 18 Oct 2006 07:47:07 +0000 (07:47 +0000)]
Check pointer before dereferencing.

Reported by: Coverity
CID: 1556

17 years agokern_intr.c:
bde [Wed, 18 Oct 2006 04:48:09 +0000 (04:48 +0000)]
kern_intr.c:
- Count (scheduling of) software interrupts (SWIs) as SWIs, not as
  hardware interrupts.
- Don't count (scheduling of) delayed SWIs as interrupts at all, since
  in the delayed case it is expected that there are many more scheduling
  calls than handling calls.  Perhaps all interrupts should be counted
  only when they are handled, but it is only counts of delayed SWIs that
  shouldn never be combined with the other counts.

subr_trap.c:
- Count (handling of) Asynchronous System Traps (ASTs) as traps, not as
  software interrupts.

Before these changes, the counter for SWIs only counted ASTs, and SWIs
weren't counted separately, but a subcounter for ASTs alone is less
needed than for most other exception sources.

4.4BSD-Lite uses the counters for similar things (actually matching
their names) on its main arches (hp300, ..., !i386) where more of the
exceptions are in hardware.

17 years agoAttempt to fix compiling problem.
davidxu [Wed, 18 Oct 2006 02:09:46 +0000 (02:09 +0000)]
Attempt to fix compiling problem.

Noticed by: tinderbox

17 years agoFixed some style bugs (especially ones involving long lines and use
bde [Tue, 17 Oct 2006 22:07:07 +0000 (22:07 +0000)]
Fixed some style bugs (especially ones involving long lines and use
of __P(())).  There are many more.

17 years agoCollapse the two identical emu10k1-alsa%diked.h creation rules for
imp [Tue, 17 Oct 2006 18:08:04 +0000 (18:08 +0000)]
Collapse the two identical emu10k1-alsa%diked.h creation rules for
snd_emu10k1 and snd_emu10kx into one line.  The 'pci' dependency here
adds no value, so I eliminted it (we don't have a snd.all file that
might make it mildly useful, and even then it wouldn't be that
useful).  With the pci optional component eliminated, I could use the
'|' operator.  I could have also include pci on both sides of the |
operator, but since it isn't a value add at all, it was better to
eliminate it.

17 years agoInclude sys/socket.h to know about sa_family_t.
glebius [Tue, 17 Oct 2006 16:56:29 +0000 (16:56 +0000)]
Include sys/socket.h to know about sa_family_t.

17 years agoSome clenaup of ngs_rcvmsg():
glebius [Tue, 17 Oct 2006 16:52:09 +0000 (16:52 +0000)]
Some clenaup of ngs_rcvmsg():
  - Inline ship_msg() into ngs_rcvmsg().
  - Plug memory leak in case if no control socket present.
  - Remove malloc() and allocate the sockaddr on stack.
  - style(9).

17 years agoFix a driver bug which could result in frames MHLEN or (MHLEN - 1) bytes
gallatin [Tue, 17 Oct 2006 14:39:19 +0000 (14:39 +0000)]
Fix a driver bug which could result in frames MHLEN or (MHLEN - 1) bytes
long being DMA'ed 2 (or 1) bytes past the end of the mbuf and corrupting
random kernel memory.  I had forgotten about the 2 bytes of implict
padding the firmware assumes.

Sponsored by: Myricom Inc.

17 years agoRemove security/freebsd-update from package split; it's now in the
bmah [Tue, 17 Oct 2006 13:50:22 +0000 (13:50 +0000)]
Remove security/freebsd-update from package split; it's now in the
base system.

Pointed out by: kris

17 years agoSome cleanup and small changes:
glebius [Tue, 17 Oct 2006 12:21:48 +0000 (12:21 +0000)]
Some cleanup and small changes:
  - Use malloc() and free() instead of MALLOC() and FREE() macros.
  - Do not check malloc results if M_WAITOK was used.
  - Remove linked list of all netgraph sockets. It isn't needed.
  - Use ng_findhook() instead of searching the list ourselves.
  - Use NG_WAITOK in syscalls.
  - Remove unneeded includes.
  - style(9)

17 years agoMake the sg_len and sg_family members of the sockaddr_ng the same type
glebius [Tue, 17 Oct 2006 11:03:55 +0000 (11:03 +0000)]
Make the sg_len and sg_family members of the sockaddr_ng the same type
as the corresponding values in sockaddr are.

17 years agoMake ng_ID_t fixed size, so that its maximum value is platform independent.
glebius [Tue, 17 Oct 2006 11:01:20 +0000 (11:01 +0000)]
Make ng_ID_t fixed size, so that its maximum value is platform independent.
This will be important in future.

17 years ago- ng_address_ID() has already freed the message, don't do double free.
glebius [Tue, 17 Oct 2006 10:59:39 +0000 (10:59 +0000)]
- ng_address_ID() has already freed the message, don't do double free.
- Get error from ng_address_ID().

Reported by: Coverity via pjd

17 years agoOlder incarnations of the device used non-standard BARs.
ru [Tue, 17 Oct 2006 10:26:11 +0000 (10:26 +0000)]
Older incarnations of the device used non-standard BARs.

Reported by: Andriy Gapon
Confirmed by: many (including lm-sensors-2.10.1)
MFC after: 3 days

17 years agoConvert types 'Elf{32,64}_Nhdr' into aliases of the generic 'Elf_Note'
jkoshy [Tue, 17 Oct 2006 05:43:30 +0000 (05:43 +0000)]
Convert types 'Elf{32,64}_Nhdr' into aliases of the generic 'Elf_Note'
type.

Reviewed by: jb

17 years agoComment out a debug entry which doesn't compile. Needed to fix LINT.
jb [Tue, 17 Oct 2006 03:53:38 +0000 (03:53 +0000)]
Comment out a debug entry which doesn't compile. Needed to fix LINT.

17 years agorename casuptr to casuword.
davidxu [Tue, 17 Oct 2006 03:05:17 +0000 (03:05 +0000)]
rename casuptr to casuword.

17 years agoRegenerate.
davidxu [Tue, 17 Oct 2006 02:28:58 +0000 (02:28 +0000)]
Regenerate.

17 years agoSync with master.
davidxu [Tue, 17 Oct 2006 02:28:26 +0000 (02:28 +0000)]
Sync with master.

17 years agoo Add keyword volatile for user mutex owner field.
davidxu [Tue, 17 Oct 2006 02:24:47 +0000 (02:24 +0000)]
o Add keyword volatile for user mutex owner field.
o Fix type consistent problem by using type long for old
  umtx and wait channel.
o Rename casuptr to casuword.

17 years agoAdd sun4v to the 'universe'.
jb [Mon, 16 Oct 2006 22:18:59 +0000 (22:18 +0000)]
Add sun4v to the 'universe'.

17 years agoAdd sun4v to the TARGET_ARCH when just TARGET is set (like pc98 is done).
jb [Mon, 16 Oct 2006 22:18:13 +0000 (22:18 +0000)]
Add sun4v to the TARGET_ARCH when just TARGET is set (like pc98 is done).

This is required for 'make universe'.

17 years agoIn sun4v, use the sparc64 version. We haven't used the serial port on
jb [Mon, 16 Oct 2006 22:11:53 +0000 (22:11 +0000)]
In sun4v, use the sparc64 version. We haven't used the serial port on
sun4v yet, so this is a 'best-guess'.

17 years agoAdd a .PATH entry to search for sources in the ${MACHINE}/${MACHINE}
jb [Mon, 16 Oct 2006 22:09:48 +0000 (22:09 +0000)]
Add a .PATH entry to search for sources in the ${MACHINE}/${MACHINE}
directory before the ${MACHINE_ARCH}/${MACHINE_ARCH} directory so that
machine-specific files take precedence of architecture-specific ones.

This fixes the build on sun4v which doesn't use the sparc64 version
of mem.c.

Tested by: make universe

17 years agoComment out 'device isa'.
jb [Mon, 16 Oct 2006 22:06:59 +0000 (22:06 +0000)]
Comment out 'device isa'.

Add a lot of nodevice entries for things that depend on isa, kbd and
other PC-centric things.

17 years agoAdd one more include to fix the case of !DDB and !atpic.
jhb [Mon, 16 Oct 2006 21:40:46 +0000 (21:40 +0000)]
Add one more include to fix the case of !DDB and !atpic.

17 years agoCorrect the WOL magic value.
obrien [Mon, 16 Oct 2006 20:33:23 +0000 (20:33 +0000)]
Correct the WOL magic value.

Submitted by: Peer Chen at nVIDIA

17 years agoTrim trailing whitespace.
jhb [Mon, 16 Oct 2006 20:07:23 +0000 (20:07 +0000)]
Trim trailing whitespace.

17 years agoRestore the 'proc' and 'tid' commands which allow one to switch to the
jhb [Mon, 16 Oct 2006 20:06:32 +0000 (20:06 +0000)]
Restore the 'proc' and 'tid' commands which allow one to switch to the
first thread in a process or to a specific thread via PIDs and TIDs,
respectively.

Submitted by: kan
Approved by: marcel

17 years agoTry to be more precise in identifying the various nForce networking adapters.
obrien [Mon, 16 Oct 2006 16:36:29 +0000 (16:36 +0000)]
Try to be more precise in identifying the various nForce networking adapters.

17 years ago- Inverted EAPD quirk for ASUS A8Jc.
ariff [Mon, 16 Oct 2006 14:43:22 +0000 (14:43 +0000)]
- Inverted EAPD quirk for ASUS A8Jc.
- Take the literal meaning of eapdbtl, but do the inversion during
  write.

Tested by: Chung-liang Wei <idreamer@gmail.com>

17 years agoBring manpage up-to-date with respect to the scheduling priority
ru [Mon, 16 Oct 2006 14:35:31 +0000 (14:35 +0000)]
Bring manpage up-to-date with respect to the scheduling priority
range, and mention the SCHED_OTHER scheduling policy.

Submitted by: davidxu
PR: docs/37843
MFC after: 3 days

17 years agofile == NULL:
ache [Mon, 16 Oct 2006 14:31:56 +0000 (14:31 +0000)]
file == NULL:
Issue __sflush() before possible setting O_APPEND mode or ftruncate(),
write to wrong place may occurse oserwise.
Use simplified _sseek() to the start, if no O_APPEND is set, instead
of _fseeko() (_sseek() to the end, if O_APPEND, occurse later, as for
file != NULL).
Don't check seek error return, as original fopen() and freopen() never
does.

file != NULL:
Add missing _sseek() to the end.

17 years agoImprove cleartmp in a number of aspects:
yar [Mon, 16 Oct 2006 13:01:45 +0000 (13:01 +0000)]
Improve cleartmp in a number of aspects:

+ Use rc.subr(8) features properly.
+ Do the whole job of obliterating /tmp contents in find(1).
+ Leave lost+found and quota.{user,group} in /tmp only if root-owned.
+ Make the overall structure clearer by first removing the X dirs
  (perhaps along with the rest of /tmp) and then re-creating them.
+ Use "find -exec rm -rf {} +" for efficiency: each rm instance gets
  a chance to kill as much files in /tmp as ARG_MAX permits.

PR: bin/104044
Submitted by: Andrey Simonenko <see PR for email>
Hacked by: yar
MFC after: 1 month

17 years agoHonor errno obtained from __sflags
ache [Mon, 16 Oct 2006 09:57:15 +0000 (09:57 +0000)]
Honor errno obtained from __sflags
Move errno restoring after FUNLOCKFILE in one case

17 years agoBack out ftruncate error handling. There can some file tipes
ache [Mon, 16 Oct 2006 09:39:15 +0000 (09:39 +0000)]
Back out ftruncate error handling. There can some file tipes
f.e. sockets when ftruncate normally fail.

17 years agoExpand "don't".
danger [Mon, 16 Oct 2006 07:59:05 +0000 (07:59 +0000)]
Expand "don't".

Approved by: trhodes (mentor), keramida (mentor)
MFC after: 3 days

17 years agoUse the loader.conf example.
danger [Mon, 16 Oct 2006 07:56:36 +0000 (07:56 +0000)]
Use the loader.conf example.

Approved by: trhodes (mentor), keramida (mentor)
MFC after: 3 days

17 years agoRemove some debugging code that accidentally crept in.
scottl [Mon, 16 Oct 2006 04:30:09 +0000 (04:30 +0000)]
Remove some debugging code that accidentally crept in.

17 years ago- Add a command validator for use in debugging.
scottl [Mon, 16 Oct 2006 04:18:38 +0000 (04:18 +0000)]
- Add a command validator for use in debugging.
- Fix the locking protocol to eliminate races between normal I/O and AENs.
- Various small improvements and usability tweaks.

Sponsored by: IronPort
Portions Submitted by: Doug Ambrisko

17 years agoMFi386: revision 1.567 (add linux_aio.c).
nyan [Mon, 16 Oct 2006 02:26:06 +0000 (02:26 +0000)]
MFi386: revision 1.567 (add linux_aio.c).

17 years agoOverhaul the transmit and dma paths:
scottl [Sun, 15 Oct 2006 23:42:56 +0000 (23:42 +0000)]
Overhaul the transmit and dma paths:
- Use bus_dmamap_load_mbuf_sg() to eliminate the need for the callback and
all of the extra bookkeeping associated with it.
- Eliminate the bce_dmamap_arg structure and streamline the memory allocation
routines to not need it.  This does change some of the debugging messages.
- Refactor the loop that fills the buffer descriptor so that it can be done
with a single set of logic in a single loop instead of two sets of logic.
- Eliminate the need to cache and pass descriptor indexes between the start
loop and the encap function.
- Change the start loop to always check the ifnet sendq for more work.

This significantly helps the driver withstand large UDP workloads, though
it's still not perfect.  I suspect the remaining work lies with handling
the OACTIVE flag, and also in possibly streamlining the interrupt handler
some.  It is, however, nearly on par with the other popular gigabit drivers
in terms of stability now.

17 years agoMake it so that the synopsis and usage message almost agree.
dwmalone [Sun, 15 Oct 2006 17:44:49 +0000 (17:44 +0000)]
Make it so that the synopsis and usage message almost agree.

MFC after: 3 weeks

17 years agoAdd a -D option to traceroute that prints the differences between
dwmalone [Sun, 15 Oct 2006 17:34:51 +0000 (17:34 +0000)]
Add a -D option to traceroute that prints the differences between
the probe packet we sent and the packet quoted by the ICMP response.
Can be useful for spotting hops that change the packet in-flight
or have problems generating correct ICMP responses.

MFC after: 3 weeks

17 years agoAdd a newline to the printf().
hrs [Sun, 15 Oct 2006 16:52:59 +0000 (16:52 +0000)]
Add a newline to the printf().

Spotted by: Peter Carah <pete@altadena.net>
MFC after: 3 days

17 years agoDo real seek
ache [Sun, 15 Oct 2006 16:24:31 +0000 (16:24 +0000)]
Do real seek
Catch ftruncate errors

PR:             104295
Submitted by:   ru (seek)

17 years agoAdd idmapd_flags to defaults/rc.conf.
ceri [Sun, 15 Oct 2006 15:55:00 +0000 (15:55 +0000)]
Add idmapd_flags to defaults/rc.conf.
Document it and idmapd_enable.

17 years agoFix compile (use the right variable name).
netchild [Sun, 15 Oct 2006 14:34:03 +0000 (14:34 +0000)]
Fix compile (use the right variable name).

17 years agoregen (linux AIO stuff)
netchild [Sun, 15 Oct 2006 14:24:10 +0000 (14:24 +0000)]
regen (linux AIO stuff)

17 years agoMFP4 (with some minor changes):
netchild [Sun, 15 Oct 2006 14:22:14 +0000 (14:22 +0000)]
MFP4 (with some minor changes):

Implement the linux_io_* syscalls (AIO). They are only enabled if the native
AIO code is available (either compiled in to the kernel or as a module) at
the time the functions are used. If the AIO stuff is not available there
will be a ENOSYS.

From the submitter:
---snip---
DESIGN NOTES:

1. Linux permits a process to own multiple AIO queues (distinguished by
   "context"), but FreeBSD creates only one single AIO queue per process.
   My code maintains a request queue (STAILQ of queue(3)) per "context",
   and throws all AIO requests of all contexts owned by a process into
   the single FreeBSD per-process AIO queue.

   When the process calls io_destroy(2), io_getevents(2), io_submit(2) and
   io_cancel(2), my code can pick out requests owned by the specified context
   from the single FreeBSD per-process AIO queue according to the per-context
   request queues maintained by my code.

2. The request queue maintained by my code stores contrast information between
   Linux IO control blocks (struct linux_iocb) and FreeBSD IO control blocks
   (struct aiocb). FreeBSD IO control block actually exists in userland memory
   space, required by FreeBSD native aio_XXXXXX(2).

3. It is quite troubling that the function io_getevents() of libaio-0.3.105
   needs to use Linux-specific "struct aio_ring", which is a partial mirror
   of context in user space. I would rather take the address of context in
   kernel as the context ID, but the io_getevents() of libaio forces me to
   take the address of the "ring" in user space as the context ID.

   To my surprise, one comment line in the file "io_getevents.c" of
   libaio-0.3.105 reads:

             Ben will hate me for this

REFERENCE:

1. Linux kernel source code:   http://www.kernel.org/pub/linux/kernel/v2.6/
   (include/linux/aio_abi.h, fs/aio.c)

2. Linux manual pages:         http://www.kernel.org/pub/linux/docs/manpages/
   (io_setup(2), io_destroy(2), io_getevents(2), io_submit(2), io_cancel(2))

3. Linux Scalability Effort:   http://lse.sourceforge.net/io/aio.html
   The design notes:           http://lse.sourceforge.net/io/aionotes.txt

4. The package libaio, both source and binary:
       http://rpmfind.net/linux/rpm2html/search.php?query=libaio
   Simple transparent interface to Linux AIO system calls.

5. Libaio-oracle:              http://oss.oracle.com/projects/libaio-oracle/
   POSIX AIO implementation based on Linux AIO system calls (depending on
   libaio).
---snip---

Submitted by: Li, Xiao <intron@intron.ac>

17 years agoRC script for idmapd(8), defaulting to off.
ceri [Sun, 15 Oct 2006 14:19:06 +0000 (14:19 +0000)]
RC script for idmapd(8), defaulting to off.

17 years agoHook up idmapd to the build, as the NFS4 client is not much use without
ceri [Sun, 15 Oct 2006 14:04:28 +0000 (14:04 +0000)]
Hook up idmapd to the build, as the NFS4 client is not much use without
it.

17 years ago.PATH isn't necessary here.
ceri [Sun, 15 Oct 2006 13:51:49 +0000 (13:51 +0000)]
.PATH isn't necessary here.
Remove hardcoded path to /usr/src/sys.

17 years agoTry to clarify that this daemon should run on the client machine.
ceri [Sun, 15 Oct 2006 13:50:16 +0000 (13:50 +0000)]
Try to clarify that this daemon should run on the client machine.
Fix a .Xr.

17 years agoMFP4 (106538 + 106541):
netchild [Sun, 15 Oct 2006 13:39:40 +0000 (13:39 +0000)]
MFP4 (106538 + 106541):
Implement CLONE_VFORK. This fixes the clone05 LTP test.

Submitted by: rdivacky

17 years agoRevert my previous commit, I mismerged this to the wrong place.
netchild [Sun, 15 Oct 2006 13:30:45 +0000 (13:30 +0000)]
Revert my previous commit, I mismerged this to the wrong place.

Pointy hat to: netchild

17 years agoMFP4 (106541): Fix the clone05 test in the LTP.
netchild [Sun, 15 Oct 2006 13:25:23 +0000 (13:25 +0000)]
MFP4 (106541): Fix the clone05 test in the LTP.

Submitted by: rdivacky

17 years agoMFP4 (107144[1]): Implement CLONE_FS on i386[1] and amd64.
netchild [Sun, 15 Oct 2006 13:22:14 +0000 (13:22 +0000)]
MFP4 (107144[1]): Implement CLONE_FS on i386[1] and amd64.

Submitted by: rdivacky [1]

17 years agoMFP4 (107868 - 107870):
netchild [Sun, 15 Oct 2006 12:51:43 +0000 (12:51 +0000)]
MFP4 (107868 - 107870):
Use a macro to test for a valid signal instead of doing it my hand everywhere.

Submitted by: rdivacky

17 years agoExplain that using 64bit arguments on 32bit architectures with CTRx()
pjd [Sun, 15 Oct 2006 12:40:27 +0000 (12:40 +0000)]
Explain that using 64bit arguments on 32bit architectures with CTRx()
macros is a bad idea.

17 years agoUpdate the FILES section which has been wrong for some time.
ceri [Sun, 15 Oct 2006 11:50:11 +0000 (11:50 +0000)]
Update the FILES section which has been wrong for some time.

MFC after: 3 days

17 years agoAdd the sun4v category.
ceri [Sun, 15 Oct 2006 11:43:42 +0000 (11:43 +0000)]
Add the sun4v category.

Reminded by: ru
MFC After: 1 re approval

17 years agoWrap debug printfs into SEQ_DEBUG() to avoid kernel messages clutter
ariff [Sun, 15 Oct 2006 08:23:48 +0000 (08:23 +0000)]
Wrap debug printfs into SEQ_DEBUG() to avoid kernel messages clutter
during module unload.

17 years agoFix the wraparound of memsize >=2GB.
tanimura [Sun, 15 Oct 2006 05:04:06 +0000 (05:04 +0000)]
Fix the wraparound of memsize >=2GB.

17 years agoLong ago, revision 1.22 of vm/vm_pager.h introduced a bug. Specifically,
alc [Sat, 14 Oct 2006 23:21:48 +0000 (23:21 +0000)]
Long ago, revision 1.22 of vm/vm_pager.h introduced a bug.  Specifically,
it introduced a check after the call to file system's get pages method
that assumes that the get pages method does not change the array of pages
that is passed to it.  In the case of vnode_pager_generic_getpages(),
this assumption has been incorrect.  The contents of the array of pages
may be shifted by vnode_pager_generic_getpages().  Likely, the problem
has been hidden by vnode_pager_haspage() limiting the set of pages that
are passed to vnode_pager_generic_getpages() such that a shift never
occurs.

The fix implemented herein is to adjust the pointer to the array of pages
rather than shifting the pages within the array.

MFC after: 3 weeks
Fix suggested by: tegge

17 years agoChange vnode_pager_addr() such that on returning it distinguishes between
alc [Sat, 14 Oct 2006 22:09:03 +0000 (22:09 +0000)]
Change vnode_pager_addr() such that on returning it distinguishes between
an error returned by VOP_BMAP() and a hole in the file.

Change the callers to vnode_pager_addr() such that they return
VM_PAGER_ERROR when VOP_BMAP fails instead of a zero-filled page.

Reviewed by: tegge
MFC after: 3 weeks

17 years agoMoved __BEGIN_DECLS up a little so that it covers __test_sse() and C++
bde [Sat, 14 Oct 2006 20:35:56 +0000 (20:35 +0000)]
Moved __BEGIN_DECLS up a little so that it covers __test_sse() and C++
isn't broken,

PR: 104425

17 years agoThis commit was generated by cvs2svn to compensate for changes in r163356,
ru [Sat, 14 Oct 2006 19:50:57 +0000 (19:50 +0000)]
This commit was generated by cvs2svn to compensate for changes in r163356,
which included commits to RCS files with non-trunk default branches.

17 years agoFix from upstream: unbreak generation of the terminfo.5 manpage.
ru [Sat, 14 Oct 2006 19:50:57 +0000 (19:50 +0000)]
Fix from upstream: unbreak generation of the terminfo.5 manpage.

PR: docs/46709, docs/56981, docs/80871
MFC after: 3 days
Obtained from: ftp://invisible-island.net/ncurses/ncurses-5.5.tar.gz

17 years agoPrevent IOC_IN with zero size argument (this is only supported
ru [Sat, 14 Oct 2006 19:01:55 +0000 (19:01 +0000)]
Prevent IOC_IN with zero size argument (this is only supported
if backward copatibility options are present) from attempting
to free memory that wasn't allocated.  This is an old bug, and
previously it would attempt to free a null pointer.  I noticed
this bug when working on the previous revision, but forgot to
fix it.

Security: local DoS
Reported by: Peter Holm
MFC after: 3 days

17 years agoReplace duplicate and not quite accurate capabilities
ru [Sat, 14 Oct 2006 16:39:03 +0000 (16:39 +0000)]
Replace duplicate and not quite accurate capabilities
description with a reference to the disktab(5) manpage.

PR: doc/48105