]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/log
FreeBSD/FreeBSD.git
21 years agoThis commit was generated by cvs2svn to compensate for changes in r117401,
Alexander Kabaev [Fri, 11 Jul 2003 03:42:19 +0000 (03:42 +0000)]
This commit was generated by cvs2svn to compensate for changes in r117401,
which included commits to RCS files with non-trunk default branches.

21 years agoGcc 3.3.1-pre 2003-07-11 libf2c bits.
Alexander Kabaev [Fri, 11 Jul 2003 03:42:19 +0000 (03:42 +0000)]
Gcc 3.3.1-pre 2003-07-11 libf2c bits.

21 years agoThis commit was generated by cvs2svn to compensate for changes in r117399,
Alexander Kabaev [Fri, 11 Jul 2003 03:42:15 +0000 (03:42 +0000)]
This commit was generated by cvs2svn to compensate for changes in r117399,
which included commits to RCS files with non-trunk default branches.

21 years agoGcc 3.3.1-pre 2003-07-11 Objective C support bits.
Alexander Kabaev [Fri, 11 Jul 2003 03:42:15 +0000 (03:42 +0000)]
Gcc 3.3.1-pre 2003-07-11 Objective C support bits.

21 years agoThis commit was generated by cvs2svn to compensate for changes in r117397,
Alexander Kabaev [Fri, 11 Jul 2003 03:42:04 +0000 (03:42 +0000)]
This commit was generated by cvs2svn to compensate for changes in r117397,
which included commits to RCS files with non-trunk default branches.

21 years agoGcc 3.3.1-pre 2003-07-11 C++ support bits.
Alexander Kabaev [Fri, 11 Jul 2003 03:42:04 +0000 (03:42 +0000)]
Gcc 3.3.1-pre 2003-07-11 C++ support bits.

21 years agoThis commit was generated by cvs2svn to compensate for changes in r117395,
Alexander Kabaev [Fri, 11 Jul 2003 03:40:53 +0000 (03:40 +0000)]
This commit was generated by cvs2svn to compensate for changes in r117395,
which included commits to RCS files with non-trunk default branches.

21 years agoGcc 3.3.1-pre as of 2003-07-11.
Alexander Kabaev [Fri, 11 Jul 2003 03:40:53 +0000 (03:40 +0000)]
Gcc 3.3.1-pre as of 2003-07-11.

21 years agoIf we must yell at the user than a file cannot be linked into place when
Brian Feldman [Fri, 11 Jul 2003 02:19:19 +0000 (02:19 +0000)]
If we must yell at the user than a file cannot be linked into place when
requested (not required) to do it, reassure them that cpio is still
intelligent enough that it will perform a full copy instead.

21 years agoThis commit was generated by cvs2svn to compensate for changes in r117392,
Marcel Moolenaar [Fri, 11 Jul 2003 01:03:55 +0000 (01:03 +0000)]
This commit was generated by cvs2svn to compensate for changes in r117392,
which included commits to RCS files with non-trunk default branches.

21 years agoRename DISABLE_TRACE to UWX_TRACE_ENABLE and change the conditional
Marcel Moolenaar [Fri, 11 Jul 2003 01:03:55 +0000 (01:03 +0000)]
Rename DISABLE_TRACE to UWX_TRACE_ENABLE and change the conditional
compilation accordingly. The net effect is that tracing is not by
default present and that it can easily be compiled-in.

The tracer uses getenv() and printf(), which cannot be used in all
cases (ie from the debugger) and with this change we improved the
applicability of the unwinder.

This change is made on the vendor branch and given back to the
vendor for possible inclusion in future versions.

21 years agoAdd init_param3() to subr_param. This function is called
Mike Silbersack [Fri, 11 Jul 2003 00:01:03 +0000 (00:01 +0000)]
Add init_param3() to subr_param.  This function is called
immediately after the kernel map has been sized, and is
the optimal place for the autosizing of memory allocations
which occur within the kernel map to occur.

Suggested by: bde

21 years agoLock down the IOMMU bus_dma implementation to make it safe to use
Thomas Moestl [Thu, 10 Jul 2003 23:27:35 +0000 (23:27 +0000)]
Lock down the IOMMU bus_dma implementation to make it safe to use
without Giant held.

A quick outline of the locking strategy:
Since all IOMMUs are synchronized, there is a single lock, iommu_mtx,
which protects the hardware registers (where needed) and the global and
per-IOMMU software states. As soon as the IOMMUs are divorced, each struct
iommu_state will have its own mutex (and the remaining global state
will be moved into the struct).
The dvma rman has its own internal mutex; the TSB slots may only be
accessed by the owner of the corresponding resource, so neither needs
extra protection.
Since there is a second access path to maps via LRU queues, the consumer-
provided locking is not sufficient; therefore, each map which is on a
queue is additionally protected by iommu_mtx (in part, there is one
member which only the map owner may access). Each map on a queue may
be accessed and removed from or repositioned in a queue in any context as
long as the lock is held; only the owner may insert a map.
To reduce lock contention, some bus_dma functions remove the map from
the queue temporarily (on behalf of the map owner) for some operations and
reinsert it when they are done. Shorter operations and operations which are
not done on behalf of the lock owner are completely covered by the lock.

To facilitate the locking, reorganize the streaming buffer handling;
while being there, fix an old oversight which would cause the streaming
buffer to always be flushed, regardless of whether streaming was enabled
in the TSB entry. The streaming buffer is still disabled for now, since
there are a number of drivers which lack critical bus_dmamp_sync() calls.

Additional testing by: jake

21 years ago-l prints all permissions, not just owner and group
Christian Brueffer [Thu, 10 Jul 2003 20:53:56 +0000 (20:53 +0000)]
-l prints all permissions, not just owner and group

PR: 54294
Submitted by: Per Hedeland <per@hedeland.org>
MFC after: 3 days

21 years agoAdd support for the 8139C+ chipset. Unlike the other chips in the 8139
Bill Paul [Thu, 10 Jul 2003 20:38:48 +0000 (20:38 +0000)]
Add support for the 8139C+ chipset. Unlike the other chips in the 8139
series, the 8139C+ has a descriptor-based DMA mechanism, and its
performance is actually pretty respectable. Note: the 8139D chip does
not support C+ mode. Only the 8139C+ and 8169 gigE chips support C+ mode.

Supported features:

- RX and TX checksum offload
- hardware VLAN tag insertion/extraction
- TX interrupt moderation using the 8139's on-board timer

Everything should be properly busdma'ed and endian-independent, so
things should work ok on non-x86 platforms. Unfortunately, my call
for testers on this code was met with deafening silence, and I don't
have access to any non-x86 FreeBSD boxes at the moment, so this is
speculation.

The device detection code has been cleaned up a little as well
(thanks to Michal Mertl) for the patches.

There are also updates to the rl(4) man page (which I accidentally
checked in before when I updated the dc(4) man page. Oops.)

Todo: finish support for the 8169 gigabit ethernet chip. This
mainly requires writing an rlgphy driver to handle the 8169's built-in
PHY. This will have to wait until I actually get my hands on an 8169
card for testing though. (I still can't find a source for one in the
U.S. Suggestions/pointers welcome.)

21 years agoRegenerate.
Bill Paul [Thu, 10 Jul 2003 18:43:39 +0000 (18:43 +0000)]
Regenerate.

21 years agoAdd support for a bunch of Microsoft networking products:
Bill Paul [Thu, 10 Jul 2003 18:43:17 +0000 (18:43 +0000)]
Add support for a bunch of Microsoft networking products:

- MN-110 10/100 USB ethernet (ADMtek Pegasus II, if_aue)
- MN-120 10/100 cardbus (ADMtek Centaur-C, if_dc)
- MN-130 10/100 PCI (ADMtek Centaur-P, if_dc)

Also update dc(4) man page to mention support for MN-120 and MN-130.

21 years agoProtect lint(1) from a #error.
Mark Murray [Thu, 10 Jul 2003 18:05:02 +0000 (18:05 +0000)]
Protect lint(1) from a #error.

21 years agoRewrite much of the embedded controller driver.
Nate Lawson [Thu, 10 Jul 2003 17:22:46 +0000 (17:22 +0000)]
Rewrite much of the embedded controller driver.
* Always use polled mode.  The intr approach did not work for many
  controllers and required the hw.acpi.ec.event_driven workaround.
* Only use an edge (not level) triggered GPE handler
* Add sc->ec_mtx for locking operations to a single EC.  There were
  many race conditions earlier between an SCI event and EcRead/Write.
* Use 1 ms as the global lock timeout
* Only acquire global lock if _GLK != 0
* Update EcWaitEvent to use an incremental backoff delay in its
  poll loop.  Wait 50 ms max instead of 10.  Most ECs respond
  in < 5 us (50 us when heavily loaded).  However, some time out
  occasionally even with a 10 ms timeout.  For delays past 1 ms, use
  msleep instead of DELAY to give SCI interrupts a chance to occur.
* Add EcCommand to send a command and wait for the appropriate event.
* The hw.acpi.ec.event_driven tunable is no longer applicable and
  has been removed.

Ideas from: Linux

21 years agoUse the default arguments for lockfunc and lockfuncarg in
Hartmut Brandt [Thu, 10 Jul 2003 13:55:09 +0000 (13:55 +0000)]
Use the default arguments for lockfunc and lockfuncarg in
bus_dma_tag_create. We need to be sure that our packets are
kept in-sequence (that's how ATM is supposed to work) and
therefor use BUS_DMA_NOWAIT in all calls to bus_dmamap_load.

For memory allocated with bus_dmamem_alloc the use of anything
other than NULL arguments for the locking is anyway bogus because
this memory never should need bouncing and hence the load should never
be defered.

Allow the receipt of OAM and RM cells on raw connections. Caveat: it seems
that RM cells are still processed by the hardware even when we open the
connection as UBR.

21 years agoCorrect makefile syntax error in r1.375.
Kris Kennaway [Thu, 10 Jul 2003 05:29:56 +0000 (05:29 +0000)]
Correct makefile syntax error in r1.375.

21 years agoSupport for large frames for VLANs was added by tweaking the packet size
Bill Paul [Thu, 10 Jul 2003 05:24:33 +0000 (05:24 +0000)]
Support for large frames for VLANs was added by tweaking the packet size
register, present only on 3c90xB and later NICs. This meant that you could
not use a 1500 byte MTU with VLANs on original 3c905/3c900 cards (boomerang
chipset). The boomerang chip does support large frames though, just not
in the same way: you can set the 'allow large frames' bit in the MAC
control register to receive frames up to 4K in size.

Changes:

- Set the 'allow large frames' bit for boomerang chips and increase
  the packet size register for cyclone and later chips. This allows
  us to use IFCAP_VLAN_MTU on all supported xl(4) NICs.
- Actually set the IFCAP_VLAN_MTU flag in the capabilities word
  in xl_attach().
- Change the method used to detect older boomerang chips. My 3c575C
  cardbus NIC was being incorrectly identified as 3c90x chip instead
  of 3c90xB because the capabilities word in its EEPROM reports
  a bizzare value. In addition to checking for the supportsNoTxLength
  bit, also check for the absence of the supportsLargePackets bit.
  Both of these cases denote a 3c90xB chip.
- Make RX and TX checksums configurable via the SIOCSIFCAP ioctl.
- Avoid an unecessary le32toh() in xl_rxeof(): we already have the
  received frame size in the lower 16 bits of rxstat, no need to
  read it again.

Tested with 3c905-TX, 3c900-TPO, 3c980C and 3c575C NICs.

21 years agotdkphy.c is missing from this module's Makefile. Add it.
Bill Paul [Thu, 10 Jul 2003 05:12:10 +0000 (05:12 +0000)]
tdkphy.c is missing from this module's Makefile. Add it.

21 years agoContinue trying to build SENDMAIL_CF if the user uses SENDMAIL_CF
Gregory Neil Shapiro [Thu, 10 Jul 2003 04:53:32 +0000 (04:53 +0000)]
Continue trying to build SENDMAIL_CF if the user uses SENDMAIL_CF
instead of SENDMAIL_MC but don't remove on it 'make clean' as the
user may not have the original .mc file and removing it could be
dangerous (e.g., make SENDMAIL_CF=/etc/mail/sendmail.cf clean).

Noticed by: peter
MFC after: 3 days

21 years agounifdef -DLAZY_SWITCH and start to tidy up the associated glue.
Peter Wemm [Thu, 10 Jul 2003 01:02:59 +0000 (01:02 +0000)]
unifdef -DLAZY_SWITCH and start to tidy up the associated glue.

21 years agoAdd to the list of AMR cards.
David E. O'Brien [Thu, 10 Jul 2003 00:25:51 +0000 (00:25 +0000)]
Add to the list of AMR cards.

21 years agoFix the VADDR() macros to use either KVADDR() or UVADDR(), depending
Peter Wemm [Wed, 9 Jul 2003 23:04:23 +0000 (23:04 +0000)]
Fix the VADDR() macros to use either KVADDR() or UVADDR(), depending
on the implied sign extension.  The single unified VADDR() macro was
not able to avoid sign extending the VM_MAXUSER_ADDRESS/USRSTACK values.
Be explicit about UVADDR() (positive address space) and KVADDR()
(kernel negative address space) to make mistakes show up more
spectacularly.

Increase user VM space from 1/2TB (512GB) to 128TB.

21 years agoFix up bogus index/offset/mask calculations in the allocpte and the
Peter Wemm [Wed, 9 Jul 2003 22:59:45 +0000 (22:59 +0000)]
Fix up bogus index/offset/mask calculations in the allocpte and the
corresponding release code.  This was preventing the use of more than
1/2TB of user VM.  I also spent a week staring at this code only to
eventually find that I'd mistakenly typed a P as an R.

21 years agoTurn the 2MB page mappings that cover the kernel text+data+bss area back
Peter Wemm [Wed, 9 Jul 2003 22:55:00 +0000 (22:55 +0000)]
Turn the 2MB page mappings that cover the kernel text+data+bss area back
on now that pmap_pte() can handle it.  I never actually ran into anything
that broke that I know of, but this was turned off as a precaution.

21 years agoHave pmap_pte() on a 2MB mapped address return the 2MB pde itself
Peter Wemm [Wed, 9 Jul 2003 22:53:45 +0000 (22:53 +0000)]
Have pmap_pte() on a 2MB mapped address return the 2MB pde itself
rather than a non-existing pte.  There is code elsewhere in i386/amd64
pmap that neglects to handle the large page cases because it knows that
it will see PG_PS in the returned "pte".

21 years agoDon't resume sigwait thread If signal is masked.
David Xu [Wed, 9 Jul 2003 22:30:55 +0000 (22:30 +0000)]
Don't resume sigwait thread If signal is masked.

21 years agoReplace custom field offset macro with the system __offsetof() macro.
Jeffrey Hsu [Wed, 9 Jul 2003 22:23:10 +0000 (22:23 +0000)]
Replace custom field offset macro with the system __offsetof() macro.

Reviewed by: bde

21 years agoA few minor changes:
Mike Silbersack [Wed, 9 Jul 2003 21:59:48 +0000 (21:59 +0000)]
A few minor changes:

- Use atomic ops to update the bigpipe count
- Make the bigpipe count sysctl readable
- Remove a duplicate comparison in an if statement
- Comment two SYSCTLs.

21 years agoHandle the EINPROGRESS case of bus_dmamap_load() for data buffers.
Scott Long [Wed, 9 Jul 2003 21:16:23 +0000 (21:16 +0000)]
Handle the EINPROGRESS case of bus_dmamap_load() for data buffers.

21 years agoFix a missing } that got dropped from the last commit.
Scott Long [Wed, 9 Jul 2003 19:30:30 +0000 (19:30 +0000)]
Fix a missing } that got dropped from the last commit.

21 years agoAdd a new quirk for cards that incorrectly interpret the amount of memory
Scott Long [Wed, 9 Jul 2003 19:19:16 +0000 (19:19 +0000)]
Add a new quirk for cards that incorrectly interpret the amount of memory
in the system.  This might also have a small performance gain.

21 years agoDon't do session management in su.
Dag-Erling Smørgrav [Wed, 9 Jul 2003 18:40:49 +0000 (18:40 +0000)]
Don't do session management in su.

PR: misc/53293
Submitted by: ru

21 years ago- Fix a typo in the call to acpi_disabled() in probe() by removing an
John Baldwin [Wed, 9 Jul 2003 18:28:53 +0000 (18:28 +0000)]
- Fix a typo in the call to acpi_disabled() in probe() by removing an
  extra trailing space.
- Don't bother probing a generic ISA bus device if isab0 already exists.
  Some BIOSes place an ACPI psuedo-device with the HID of a generic ISA bus
  device under the PCI-ISA bridge device.  This is not the best solution
  but will work for now.  The isa bus driver only allows for one ISA bus
  anyways.

21 years agoAdd ID for UMAX Astra 2100U
Andrey A. Chernov [Wed, 9 Jul 2003 17:05:59 +0000 (17:05 +0000)]
Add ID for UMAX Astra 2100U

21 years ago- Fix the kernel build for configurations that include pci but not isa by
John Baldwin [Wed, 9 Jul 2003 16:14:10 +0000 (16:14 +0000)]
- Fix the kernel build for configurations that include pci but not isa by
  having the PCI-ISA bridge driver depend on both pci and isa.
- Have the PCI-EISA bridge driver depend on both pci and eisa as well.
- Make acpi_isab.c depend on acpi and isa.

Submitted by: Marius Strobl <marius@alchemy.franken.de> (1,2)

21 years agoUncomment the dc(4) driver, it should work just fine now.
Maxime Henrion [Wed, 9 Jul 2003 15:04:27 +0000 (15:04 +0000)]
Uncomment the dc(4) driver, it should work just fine now.

21 years agoMake the dc(4) driver endian-clean, so to that it works on sparc64.
Maxime Henrion [Wed, 9 Jul 2003 15:03:10 +0000 (15:03 +0000)]
Make the dc(4) driver endian-clean, so to that it works on sparc64.
There are such cards in Netra X1 boxes, which should thus be fully
supported now.

Tested by: jake

21 years agoPOSIX says if a thread is in sigwait state, although a signal may not in
David Xu [Wed, 9 Jul 2003 14:30:51 +0000 (14:30 +0000)]
POSIX says if a thread is in sigwait state, although a signal may not in
its waitset, but if the signal is not masked by the thread, the signal
can interrupt the thread and signal action can be invoked by the thread,
sigwait should return with errno set to EINTR.
Also save and restore thread internal state(timeout and interrupted)
around signal handler invoking.

21 years agoBlock SIGURG while reading from the control channel.
Yaroslav Tykhiy [Wed, 9 Jul 2003 13:54:33 +0000 (13:54 +0000)]
Block SIGURG while reading from the control channel.

Rationale:

SIGURG is configured by ftpd to interrupt system calls, which is useful
during data transfers.  However, SIGURG could interrupt I/O on the
control channel as well, which was mistaken for the end of the session.

A practical example could be aborting the download of a tiny file,
when the abort sequence reached ftpd after ftpd had passed the file
data to the system and returned to its command loop.

Reported by: ceri
MFC after: 1 week

21 years agoImprove error handling in getline():
Yaroslav Tykhiy [Wed, 9 Jul 2003 13:15:32 +0000 (13:15 +0000)]
Improve error handling in getline():
- always check the return value from getc(3) for EOF;
- if the attempt to read the TELNET command byte has
  returned EOF, exit from the loop instead of using
  the EOF value as a normal character.

MFC after: 1 week

21 years agoConfiguraiton ROM fix:
Hidetoshi Shimokawa [Wed, 9 Jul 2003 13:07:35 +0000 (13:07 +0000)]
Configuraiton ROM fix:
- Don't bump the generation if ROM has not changed and keep it
between 0x2 and 0xf.
- Refetch the ROM if CRC of the businfo block has changed.

21 years agoMake a malloced copy of "chrootdir" even if it points to an absolute
Yaroslav Tykhiy [Wed, 9 Jul 2003 12:46:24 +0000 (12:46 +0000)]
Make a malloced copy of "chrootdir" even if it points to an absolute
pathname inside "residue" so "chrootdir" can be simply freed later.

PR: bin/53435
Submitted by: Yutaka Ishihara <yutaka at fandc.co.jp>
MFC after: 1 week

21 years agoCatch up with the English version:
Hiroki Sato [Wed, 9 Jul 2003 08:58:08 +0000 (08:58 +0000)]
Catch up with the English version:

1.582 -> 1.583 relnotes/common/new.sgml

21 years agoBe specific about which domainname is not set. It seems the man page
Mike Makonnen [Wed, 9 Jul 2003 03:21:03 +0000 (03:21 +0000)]
Be specific about which domainname is not set. It seems the man page
reference is not enough.

21 years agoRestore signal mask correctly after fork().
David Xu [Wed, 9 Jul 2003 01:39:24 +0000 (01:39 +0000)]
Restore signal mask correctly after fork().

21 years agoSave and restore thread's error code around signal handling.
David Xu [Wed, 9 Jul 2003 01:06:12 +0000 (01:06 +0000)]
Save and restore thread's error code around signal handling.

Reviewed by: deischen

21 years ago- In vlan_input(), always mask off all but the VLID bits from tags
Bill Paul [Tue, 8 Jul 2003 21:54:20 +0000 (21:54 +0000)]
- In vlan_input(), always mask off all but the VLID bits from tags
  extracted from received frames, both in the IFCAP_VLAN_HWTAGGING case
  and not. (Some drivers may already do this masking internally, but
  doing it here doesn't hurt and insures consistency.)

- In vlan_ioctl(), don't let the user set a VLAN ID value with anything
  besides the VLID bits set, otherwise we will have trouble matching
  an interface in vlan_input() later.

PR: kern/46405

21 years agoHandle geoms which are withering away specially in the dump functions.
Poul-Henning Kamp [Tue, 8 Jul 2003 21:12:40 +0000 (21:12 +0000)]
Handle geoms which are withering away specially in the dump functions.

21 years agoGrammar fix
Christian Brueffer [Tue, 8 Jul 2003 21:10:32 +0000 (21:10 +0000)]
Grammar fix

PR: 54225
Submitted by: Christopher Nehren <apeiron@comcast.net>
MFC after: 3 days

21 years agoIn pmap_object_init_pt(), the pmap_invalidate_all() should be performed on
Alan Cox [Tue, 8 Jul 2003 19:40:35 +0000 (19:40 +0000)]
In pmap_object_init_pt(), the pmap_invalidate_all() should be performed on
the caller-provided pmap, not the kernel_pmap.  Using the kernel_pmap
results in an unnecessary IPI for TLB shootdown on SMPs.

Reviewed by: jake, peter

21 years agoAdd an ACPI to ISA psuedo bridge driver. It attaches an isab(4) device to
John Baldwin [Tue, 8 Jul 2003 18:59:32 +0000 (18:59 +0000)]
Add an ACPI to ISA psuedo bridge driver.  It attaches an isab(4) device to
ACPI nodes with the plug and play ID's defined for a "Generic ISA Bus
Device" as defined in section 10.7 of the ACPI 2.0 specification.  This
gives machines like the Libretto that contain a fake ISA bus that is not
connected via a PCI-ISA bridge an ISA bus for ISA devices to attach to.

Tested by: markm

21 years agoPull in the entire kmem_map size calculation from kern_malloc, rather
Mike Silbersack [Tue, 8 Jul 2003 18:59:21 +0000 (18:59 +0000)]
Pull in the entire kmem_map size calculation from kern_malloc, rather
than the shortcircuited version I had been using, which only worked
properly on i386 & amd64.

Also, change an autoscale constant to account for the more correct
kmem_map size.

Problem noticed by:     mux

21 years ago- Make the isab devclass global to allow for multiple ISA bridge drivers.
John Baldwin [Tue, 8 Jul 2003 18:56:58 +0000 (18:56 +0000)]
- Make the isab devclass global to allow for multiple ISA bridge drivers.
- Factor out code common to all ISA bridge drivers attach methods into a
  isab_attach() function.
- Rename the PCI-ISA bridge driver's attach function to pci_isab_attach()
  and have it call isab_attach().

21 years agoAdd my birthday to the calendar.
Marc Fonvieille [Tue, 8 Jul 2003 18:31:49 +0000 (18:31 +0000)]
Add my birthday to the calendar.

21 years agoBring build(7) in sync with src/Makefile by removing documentation
Sean Kelly [Tue, 8 Jul 2003 16:39:10 +0000 (16:39 +0000)]
Bring build(7) in sync with src/Makefile by removing documentation
for targets that have been unsupported since April:
   - upgrade
   - aout-to-elf
   - aout-to-elf-build
   - aout-to-elf-install
   - move-aout-libs

Approved by: imp

21 years agoCorrect to match reality regarding interface names.
Daniel Harris [Tue, 8 Jul 2003 13:24:42 +0000 (13:24 +0000)]
Correct to match reality regarding interface names.

PR: 51006
Submitted by: "Dmitry Pryanishnikov" <dmitry@atlantis.dp.ua>
mdoc clue by: "Simon L. Nielsen" <simon@nitro.dk>
MFC after: 10 days

21 years agoRemove utmp references, no longer done by login(1) in 5.x.
Daniel Harris [Tue, 8 Jul 2003 13:01:28 +0000 (13:01 +0000)]
Remove utmp references, no longer done by login(1) in 5.x.

PR: 54201
Submitted by: mdg <mdg@secureworks.net>

21 years agopam_lastlog, not login, in 5.x.
Daniel Harris [Tue, 8 Jul 2003 12:46:25 +0000 (12:46 +0000)]
pam_lastlog, not login, in 5.x.

PR: 54200
Submitted by: Matthew George <mdg@secureworks.net>

21 years agoWhen _PTHREADSINVARIANTS is defined SIGABRT is not included
Mike Makonnen [Tue, 8 Jul 2003 09:58:23 +0000 (09:58 +0000)]
When _PTHREADSINVARIANTS is defined SIGABRT is not included
in the set of signals to block.
Also, make the PANIC macro call abort() instead of simply
exiting.

21 years ago* introduce a section on SYNTAX to document the handling
Luigi Rizzo [Tue, 8 Jul 2003 08:07:03 +0000 (08:07 +0000)]
* introduce a section on SYNTAX to document the handling
  spaces and comma-separated lists of arguments;

* reword the description of address specifications, to include
  previous and current changes for address sets and lists;

* document the new '-n' flag.

* update the section on differences between ipfw1 and ipfw2
  (this is becoming boring!)

MFC after: 3 days

21 years agoA bunch of changes (mostly syntactic sugar, all backward compatible):
Luigi Rizzo [Tue, 8 Jul 2003 07:52:47 +0000 (07:52 +0000)]
A bunch of changes (mostly syntactic sugar, all backward compatible):

  * Make the addr-set size optional (defaults to /24)
    You can now write 1.2.3.0/24{56-80} or  1.2.3.0{56-80}
    Also make the parser more strict.

  * Support a new format for the list of addresses:
        1.2.3.4,5.6.7.8/30,9.10.11.12/22,12.12.12.13, ...
    which exploits the new capabilities of O_IP_SRC_MASK/O_IP_DST_MASK

  * Allow spaces after commas to make lists of addresses more readable.
        1.2.3.4, 5.6.7.8/30, 9.10.11.12/22, 12.12.12.13, ...

  * ipfw will now accept full commands as a single argument and strip
    extra leading/trailing whitespace as below:
        ipfw "-q add allow ip from 1.2.3.4 to 5.6.7.8, 9.10.11.23 "
    This should help in moving the body of ipfw into a library
    that user programs can invoke.

  * Cleanup some comments and data structures.

  * Do not print rule counters for dynamic rules with ipfw -d list
    (PR 51182)

  * Improve 'ipfw -h' output (PR 46785)

  * Add a '-n' flag to test the syntax of commands without actually
    calling [gs]etsockopt() (PR 44238)

  * Support the '-n' flag also with the preprocessors;

Manpage commit to follow.

MFC after: 3 days

21 years agoMerge the handlers of O_IP_SRC_MASK and O_IP_DST_MASK opcodes, and
Luigi Rizzo [Tue, 8 Jul 2003 07:44:42 +0000 (07:44 +0000)]
Merge the handlers of O_IP_SRC_MASK and O_IP_DST_MASK opcodes, and
support matching a list of addr/mask pairs so one can write
more efficient rulesets which were not possible before e.g.

    add 100 skipto 1000 not src-ip 10.0.0.0/8,127.0.0.1/8,192.168.0.0/16

The change is fully backward compatible.
ipfw2 and manpage commit to follow.

MFC after: 3 days

21 years ago - When stealing a kse in kseq_move() ignore the current kseq's min nice
Jeff Roberson [Tue, 8 Jul 2003 06:19:40 +0000 (06:19 +0000)]
 - When stealing a kse in kseq_move() ignore the current kseq's min nice
   value.  We want to steal any thread, even one that is not given a slice
   on its current queue.

21 years agoPut some concrete limits on pipe memory consumption:
Mike Silbersack [Tue, 8 Jul 2003 04:02:31 +0000 (04:02 +0000)]
Put some concrete limits on pipe memory consumption:

- Limit the total number of pipes so that we do not
  exhaust all vm objects in the kernel map.  When
  this limit is reached, a ratelimited message will
  be printed to the console.

- Put a soft limit on the amount of memory consumable
  by pipes.  Once the limit has been reached, all new
  pipes will be limited to 4K in size, rather than the
  default of 16K.

- Put a limit on the number of pages that may be used
  for high speed page flipping in order to reduce the
  amount of wired memory.  Pipe writes that occur
  while this limit is exceeded will fall back to
  non-page flipping mode.

The above values are auto-tuned in subr_param.c and
are scaled to take into account both the size of
physical memory and the size of the kernel map.

These limits help to reduce the "kernel resources exhausted"
panics that could be caused by opening a large
number of pipes.  (Pipes alone are no longer able
to exhaust all resources, but other kernel memory hogs
in league with pipes may still be able to do so.)

PR: 53627
Ideas / comments from: hsu, tjr, dillon@apollo.backplane.com
MFC after: 1 week

21 years agoRemove last vestiges of the old rc.
Mike Makonnen [Tue, 8 Jul 2003 02:52:14 +0000 (02:52 +0000)]
Remove last vestiges of the old rc.

21 years ago"towards" -> "toward". According to dictionary.com, the use of "towards"
Joseph Koshy [Tue, 8 Jul 2003 01:39:02 +0000 (01:39 +0000)]
"towards" -> "toward".  According to dictionary.com, the use of "towards"
is common in British English, while "toward" is the preferred form in
American English.  Use the American form for consistency.

Correct the date on the manual page.

Submitted by: Tom Rhodes <trhodes@freebsd.org>,
underway@comcast.net (Gary W. Swearingen)

21 years agoStyle nit.
David E. O'Brien [Tue, 8 Jul 2003 01:24:21 +0000 (01:24 +0000)]
Style nit.

21 years agoFix LOR between scheduler lock and SMP rendezvous lock.
David Xu [Tue, 8 Jul 2003 00:14:30 +0000 (00:14 +0000)]
Fix LOR between scheduler lock and SMP rendezvous lock.

Reviewed by: jhb

21 years agoMFC noted: chroot(8) -u/-g/-G.
Bruce A. Mah [Mon, 7 Jul 2003 22:44:38 +0000 (22:44 +0000)]
MFC noted:  chroot(8) -u/-g/-G.

21 years agoAdd support for a -n argument which displays user and group IDs
Brooks Davis [Mon, 7 Jul 2003 21:41:23 +0000 (21:41 +0000)]
Add support for a -n argument which displays user and group IDs
numerically rather than converting to a user or group name.

MFC After: 1 week

21 years agoFreeBSD 2.0.5 is old hat. Also cross-reference GEOM_VOL from the discussion
Garrett Wollman [Mon, 7 Jul 2003 21:19:04 +0000 (21:19 +0000)]
FreeBSD 2.0.5 is old hat.  Also cross-reference GEOM_VOL from the discussion
of wiring SCSI devices, since it provides a non-SCSI-specific way of
accomplishing a similar task.

21 years agoFix wording: `irregardless' is a solecism.
Garrett Wollman [Mon, 7 Jul 2003 21:15:01 +0000 (21:15 +0000)]
Fix wording: `irregardless' is a solecism.

21 years agoPut NFSSERVER in the right list of filesystem stuff. Building a kernel
Garrett Wollman [Mon, 7 Jul 2003 21:13:21 +0000 (21:13 +0000)]
Put NFSSERVER in the right list of filesystem stuff.  Building a kernel
with only NFSSERVER won't get you anywhere.

21 years agoSingle-character style fix.
Garrett Wollman [Mon, 7 Jul 2003 21:12:34 +0000 (21:12 +0000)]
Single-character style fix.

21 years ago - Clean up an unused variable.
Jeff Roberson [Mon, 7 Jul 2003 21:08:28 +0000 (21:08 +0000)]
 - Clean up an unused variable.

Submitted by: Steve Kargl <skg@routmask.apl.washington.edu>

21 years agoAdd "blame" as an "annotate" alias for interoperability with subversion.
David E. O'Brien [Mon, 7 Jul 2003 19:15:36 +0000 (19:15 +0000)]
Add "blame" as an "annotate" alias for interoperability with subversion.

21 years agoOnly define a default editor if one isn't already defined.
David E. O'Brien [Mon, 7 Jul 2003 19:13:39 +0000 (19:13 +0000)]
Only define a default editor if one isn't already defined.

21 years agoAdd a workaround for the USB_PRODUCT_MCT_SITECOM_USB232 device: limit
Bill Paul [Mon, 7 Jul 2003 18:01:23 +0000 (18:01 +0000)]
Add a workaround for the USB_PRODUCT_MCT_SITECOM_USB232 device: limit
the bulk out buffer size to 16 bytes. The bulk out endpoint descriptor
reports 32 bytes, but if you use this value, data will get dropped.

Reviewed/approved by: scottl

21 years ago* add a function to display a complete feeder chain on the console, for
Cameron Grant [Mon, 7 Jul 2003 17:40:02 +0000 (17:40 +0000)]
* add a function to display a complete feeder chain on the console, for
debugging.

* set the parent of non-format feeders, so that sndstat doesn't miss out
things like feeder_rate.

MFC: 1 week

21 years agoCorrectly print signal mask, the bug was introduced by cut and paste
David Xu [Mon, 7 Jul 2003 12:12:33 +0000 (12:12 +0000)]
Correctly print signal mask, the bug was introduced by cut and paste
in last commit.

21 years agoAdd corega PCCL-11 to match entry added to pccarddevs a while ago.
Warner Losh [Mon, 7 Jul 2003 07:57:35 +0000 (07:57 +0000)]
Add corega PCCL-11 to match entry added to pccarddevs a while ago.

Noticed going into openbsd...

21 years ago - Complete the vm object locking in vm_pageout_object_deactivate_pages().
Alan Cox [Mon, 7 Jul 2003 07:16:29 +0000 (07:16 +0000)]
 - Complete the vm object locking in vm_pageout_object_deactivate_pages().
 - Change vm_pageout_object_deactivate_pages()'s first parameter from a
   vm_map_t to a pmap_t.
 - Change vm_pageout_object_deactivate_pages()'s and
   vm_pageout_map_deactivate_pages()'s last parameter from a vm_pindex_t
   to a long.  Since the number of pages in an address space doesn't
   require 64 bits on an i386, vm_pindex_t is overkill.

21 years agoReset the PSM aux device twice to help pierce through some KVM's to
Mike Heffner [Mon, 7 Jul 2003 05:40:13 +0000 (05:40 +0000)]
Reset the PSM aux device twice to help pierce through some KVM's to
get the correct data from the attached mouse. Multiple resets should
be harmless, but just in case, the second one is non-fatal and is just
ignored.

21 years agoAdd a newline to debug message.
David Xu [Mon, 7 Jul 2003 04:32:17 +0000 (04:32 +0000)]
Add a newline to debug message.

21 years agoAvoid accessing user provided parameters in critical region.
David Xu [Mon, 7 Jul 2003 04:28:23 +0000 (04:28 +0000)]
Avoid accessing user provided parameters in critical region.

Reviewed by: deischen

21 years agoRemove MAINTAINER= lines from individual Makefiles in favor of the
Gregory Neil Shapiro [Mon, 7 Jul 2003 03:54:04 +0000 (03:54 +0000)]
Remove MAINTAINER= lines from individual Makefiles in favor of the
MAINTAINER file (which already had entries for sendmail).

21 years agoPut back install/distribution dependency on the buildable .cf
Gregory Neil Shapiro [Mon, 7 Jul 2003 03:19:46 +0000 (03:19 +0000)]
Put back install/distribution dependency on the buildable .cf
files to get mergemaster working again.

Noticed by: Shin-ichi YOSHIMOTO <yosimoto@waishi.jp>

21 years agoPrint thread's scope, also print signal mask for every thread and print
David Xu [Mon, 7 Jul 2003 03:08:11 +0000 (03:08 +0000)]
Print thread's scope, also print signal mask for every thread and print
it in one line.

21 years ago* support ich5
Cameron Grant [Sun, 6 Jul 2003 23:54:00 +0000 (23:54 +0000)]
* support ich5

PR: kern/53242
Submitted by: Shin-ichi Yoshimoto <yosimoto@waishi.jp>  (partly)
Tested by: Dominic Marks <dom@cus.org.uk> (version in PR)
MFC after: 1 week

21 years agoConvert the dc(4) driver to the busdma API. This is a necessary step
Maxime Henrion [Sun, 6 Jul 2003 21:45:31 +0000 (21:45 +0000)]
Convert the dc(4) driver to the busdma API.  This is a necessary step
to have this driver working on sparc64.  It still needs to be made
endian-clean before it can work there.

Special thanks to dragonk@evilcode.net for sending me a dc(4) card so
that I was able to do this work.

Many cheers to all the people that tested this change, thanks to them,
this change shouldn't break anything :-).

Tested by: marcel (i386 and ia64), ru (i386), wilko (alpha),
mbr (i386), wpaul (i386) and
Will Saxon <WillS@housing.ufl.edu> (i386)

21 years agoMFi386
Alan Cox [Sun, 6 Jul 2003 20:32:42 +0000 (20:32 +0000)]
MFi386
 Updates to cnt.v_wire_count, the global count of wired pages, should be
 performed using atomic ops.

21 years agoUse 'id' instead of 'grep' to detect the presence of the smmsp user/group.
Gregory Neil Shapiro [Sun, 6 Jul 2003 19:37:40 +0000 (19:37 +0000)]
Use 'id' instead of 'grep' to detect the presence of the smmsp user/group.
This fixes the check for users with smmsp in NIS instead of their local
files.

Suggested by: peter
MFC after: 5 days

21 years agoDo not install sendmail example/infrastructure files if NO_SENDMAIL is
Gregory Neil Shapiro [Sun, 6 Jul 2003 19:23:31 +0000 (19:23 +0000)]
Do not install sendmail example/infrastructure files if NO_SENDMAIL is
defined.  The only two files installed in this case are aliases (which
I believe other MTAs may use) and mailer.conf (which isn't sendmail,
it belongs to mailwrapper).

PR: 50477
MFC after: 5 days

21 years agoBow to popular demand and default SENDMAIL_SUBMIT_MC to
Gregory Neil Shapiro [Sun, 6 Jul 2003 19:17:04 +0000 (19:17 +0000)]
Bow to popular demand and default SENDMAIL_SUBMIT_MC to
`hostname`.submit.mc which is templated from freebsd.submit.mc if the
default file does not exist.  This makes the building of the submit.cf
behavior identical to that of the the sendmail.cf.

PR: 44256
Submitted by: Matt Emmerton <matt@gsicomp.on.ca>
MFC after: 5 days

21 years agoThanks to Ruslan's src/etc/sendmail/Makefile patch, it is now safe
Gregory Neil Shapiro [Sun, 6 Jul 2003 19:01:41 +0000 (19:01 +0000)]
Thanks to Ruslan's src/etc/sendmail/Makefile patch, it is now safe
(though probably not a good idea in general) to set the various
SENDMAIL_*_MC variables to /etc/mail/sendmail.mc or /etc/mail/submit.mc.

MFC after: 5 days